Datasets:

wckwan

/

M4LE

like 3

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the earliest extant correspondence between pierre rmond de montmort and a member of the bernoulli family is a letter from montmort to johann bernoulli dated february 27 , 1703 , concerning a paper on calculus that the latter had written for the acadmie royale des sciences in paris @xcite . they corresponded sporadically over the next few years . on april 29 , 1709 , montmort sent bernoulli a copy of his book on probability , _ essay danalyse sur les jeux de hazard _ , that he recently had published @xcite . the book is the first in a series of books in probability published by several others over the years 1708 to 1718 in what hald [ ( @xcite ) , page 191 ] calls the `` great leap forward '' in probability . bernoulli replied with a gift of a copy of his nephew s doctoral dissertation @xcite , the second book in hald s `` great leap forward '' ; nicolaus bernoulli s book dealt with applications of probability . once johann bernoulli received his copy of _ essay danalyse _ , he sent , on march 17 , 1710 , a detailed set of comments on the book . in the letter bernoulli included another set of comments on _ essay danalyse _ , this one by his nephew nicolaus ( @xcite , pages 283303 ) . thus began a series of correspondence between montmort and nicolaus bernoulli on problems in probability . montmort included much of this correspondence in part v of the second edition of _ essay danalyse _ @xcite . the correspondence between montmort and nicolaus bernoulli after 1713 , left unpublished and largely ignored by historians , contains scientific news and further discussion of problems in probability . the major topic is a continuing discussion of issues related to the card game le her . next , in terms of ink spilt on probability , are discussions of the `` problem of the pool , '' or waldegrave s problem , generalized to more than three players , and of the game les trennes ( which may be translated as `` the gifts '' ) . the correspondence also contains discussions of various problems in algebra , geometry , differential equations and infinite series . as an aristocrat , montmort s network included both political and scientific connections . his letters to bernoulli contain some references to his political activities that sometimes kept him from replying promptly . his brother , nicolas rmond , was chef de conseil for phillipe duc dorlans , who became regent of france after his uncle louis xiv died in 1715 ( @xcite , page 599 ) . among the mathematicians of the era , montmort corresponded with isaac newton , gottfried leibniz , brook taylor and abraham de moivre , in addition to the bernoullis as well as several others . as a talented amateur mathematician , his work was well regarded by the mathematicians of his day . he was generous to his scientific friends . he received as guests to the chteau de montmort nicolaus bernoulli , brook taylor and one of the sons of johann bernoulli . he also sent gifts of cases of wine and champagne to both newton and taylor . le her is a game of strategy and chance played with a standard deck of fifty - two playing cards . the simplest situation is when two players play the game , and the solution is not simply determined even in that situation . montmort calls the two players pierre and paul . pierre deals a card from the deck to paul and then one to himself . paul has the option of switching his card for pierre s card . pierre can only refuse the switch if he holds a king ( the highest valued card ) . after paul makes his decision to hold or switch , pierre now has the option to hold whatever card he now has or to switch it with a card drawn from the deck . however , if he draws a king , he must retain his original card . the player with the highest card wins the pot , with ties going to the dealer pierre . the game can be expanded to more than two players . montmort [ ( @xcite ) , pages 186187 ] originally described the problem for four players and posed the question : what are the chances of each player relative to the order in which they make their play ? because of the winning conditions , it is obvious that one would want to switch low cards and keep high ones . the key is to find what to do with the middle cards , such as seven and eight , when two players are playing the game . in other cases , cards are clearly too low to keep or too high to switch , being much below or above the average in a random draw . naturally , the threshold would be lower with more than two players . in part essay danalyse _ , only the game with two players is considered . initially , montmort and nicolaus bernoulli wrote back and forth about the problem and came to the same solution . however , two of montmort s friends contended that this solution was incorrect . these were an english gentleman named waldegrave and an abbot whose abbey was only a league and a half ( about 5.8 kilometers ) from chteau de montmort ( @xcite , page 338 ) . montmort identified waldegrave only as the brother of the lord waldegrave who married the natural daughter of king james ii of england . lord waldegrave is henry waldegrave , 1st baron waldegrave , and his wife is henrietta fitzjames , daughter of james ii , and his mistress arabella churchill . the abbot is the abb dorbais ; montmort also refers to him as the abb de monsoury . the reason for the two appellations for the abbot is that his full name is pierre cuvier de montsoury , abb dorbais . he has been described as `` un prodige de bon coeur , durbanit et de science '' ( @xcite ) . for the spelling choice between montsoury and monsoury , it should be noted that montmort often spelled his name `` monmort . '' two other problems were discussed extensively in the correspondence . the first is the problem of the pool , a problem that waldegrave suggested to montmort and solved himself ( @xcite , page 318 ) . in _ essay danalyse _ the problem is solved for three players . it is often called waldegrave s problem @xcite . the `` pool '' is a way of getting three or more players to gamble against one another , when the game put into play is for two players only . in the situation for three players ( montmort uses the names pierre , paul and jacques ) , all three begin by putting an ante into the pot . then pierre and paul play a game against each other . the winner plays against jacques and the loser puts money into the pot . the game continues until one player has beaten the other two in a row . that player takes the pot . the game can be expanded to more than three players , but that situation was not fully treated in _ essay danalyse_. the second is the problem of solving the game les trennes ( or `` estreine , '' an alternative old french spelling ) . as described by montmort [ ( @xcite ) , pages 406407 ] , this is a strategic game between a father and his son . the father holds an odd or even number of tokens in his hand , which his son can not see . when the son guesses even , he receives a gift of four cus ( silver coins ) if he is correct and nothing if wrong . when the son guesses odd , he receives one cu if he is correct and nothing if wrong . the discussion of this game in the correspondence is only brought in to enlighten le her whose strategic nature is in some important respects essentially similar . montmort concludes the last letter ( to bernoulli ) that appears in _ essay danalyse _ with a remark that waldegrave had volunteered to take care of getting the book printed in paris . montmort s letter was dated november 15 , 1713 , and was written from paris . what is also of concern to us are the letters after this date and how these letters relate to earlier discussions . the unpublished correspondence begins with a letter from montmort to bernoulli dated january 25 , 1714 , in which he says that he has sent bernoulli two copies of the second edition of _ essay danalyse_. montmort was still in paris , where he claimed to have been for three months . he was staying at a hotel in rue des bernardins , which in modern paris is only a walk of 350 meters to the printer , jacques quillau in rue galande . presumably , waldegrave s help consisted mainly in dealing with the printer and the proof sheets as they came off the press , thus relieving montmort of some tedious work . to understand the discussion of le her after 1713 , it is necessary to describe the treatment of the game as it appears in the second edition of _ essay danalyse_. hald [ ( @xcite ) , pages 314322 ] provides a detailed description of the mathematical calculations involved in assessing the game . yet he devotes little space to elucidating the discussions among bernoulli , montmort and waldegrave , as well as the abb dorbais , concerning the issues surrounding these mathematical calculations . it is the substance of these discussions that are of interest to us . hald s only comment on the discussion over le her concerns a comment made by waldegrave and abb dorbais to the effect that bernoulli s reasoning in obtaining his mathematical solution is faulty . after pointing out their observation that bernoulli s solution fails to account for a player s probability of playing in a certain way , hald [ ( @xcite ) , page 315 ] claims : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is no wonder that bernoulli does not the implications of this remark , since the writers themselves have not grasped the full implication of their point of view . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is indeed true that there was some confusion on bernoulli s side which he deftly tried to hide . henny [ ( @xcite ) , page 502 ] comments that he is amazed to find expressed in the letters many concepts and ideas that appear in modern game theory . at the same time , he is surprised to find waldegrave defending his position so strongly against bernoulli who was the superior mathematician . henny states further that waldegrave did not have the necessary mathematical skills to provide a mathematical proof of his results . as we will show in a review of the treatment of le her in _ essay danalyse _ and the subsequent unpublished correspondence , both hald s and henny s insights fall short of the mark . , though obviously correct for the limited aspect in which he viewed the problem , is unsatisfactory to common sense , which suggests that in all circumstances there must be , according to the degree of our knowledge , at least one rule of conduct which shall be not less satisfactory than any other ; and this his discussion fails to provide . '' our discussion below will show that this assessment is misinformed . ] one reason they fall short is that they do not consider the full range of the various events that were under discussion and their associated probabilities . two events are natural for a probabilist to consider . the first is the distribution of the cards to pierre and paul . the second is the randomizing device used to come up with the mixed strategy prescribing when the players should hold and when they should switch . the randomizing device considered in _ essay danalyse _ is a bag containing black and white counters or tokens ( the old french word used is `` jetton '' ) . the third event that montmort , waldegrave and bernoulli consider ( but not hald or henny ) is difficult , or perhaps impossible , to quantify . this is the possibility that paul , say , is a poor player and does not follow a strategy that is mathematically optimal , or the possibility that paul , say , is a very good player who tries to trick pierre into making a poor choice . this kind of event unfolds regularly in modern poker games . another reason for which hald and henny see some confusion in the discussions among bernoulli , montmort and waldegrave is that what we are seeing in the correspondence is the complete unfolding of a problem from its initial statement , and discussions around it , to a complete solution . this is different from a `` textbook '' statement of a problem followed by a solution . in the latter case , the problem and solution are both well laid out . in the former case , there is some grappling with the problem until it becomes clear how to proceed . we begin with the correspondence in _ essay _ _ danalyse _ where le her is first mentioned . in johann bernoulli s 1710 letter to montmort in _ essay danalyse _ , he suggests more efficient methods to reach montmort s conclusions for a variety of problems and in some cases generalizes montmort s results . there is only one reference to the problem of le her , which is the second of four problems proposed in montmort [ ( @xcite ) , pages 185187 ] : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the second and the third [ problem ] seem to me amenable , but not without much difficulty and work , that i prefer to defer to you and learn the solution , than to work long at the expense of my ordinary occupations that leave me scarcely any time to apply myself to other things . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ in his reply to this letter , which is dated november 15 , 1710 ( @xcite , pages 303307 ) , montmort makes no reference to this passage . nicolaus bernoulli s first letter to montmort , dated february 26 , 1711 , makes no reference to the game le her . it is a note in montmort s reply to nicolaus bernoulli , dated april 10 , 1711 ( @xcite , pages 315323 ) that initiates the discussion of montmort s second problem : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ i started some time ago to work on the solution of problems that i propose at the end of my book ; i find that in le her , when there are only two players left , pierre and paul , paul s advantage is greater than 1 in 85 , and less than 1 in 84 . this problem has difficulties of a singular nature . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ in a postscript to this letter , montmort makes an additional remark : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ as there are few copies of my book left , there will soon be a new edition . when i have decided , i will ask you permission , and your uncle , to insert your beautiful letters which will make the principal embellishment . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is this announcement that may have motivated nicolaus bernoulli to continue his correspondence with montmort and to send him much interesting material . publishing mathematical material outside a scientific society or without a patron to cover the costs was an expensive proposition , one that montmort could afford . because of the specialized type that was used and the accompanying necessary skill of the typesetter , the cost of a mathematical publication was well above the norm for less technical books . bernoulli could get his results in print at no cost to himself . bernoulli responded with a long letter , dated november 10 , 1711 ( @xcite , pages 323337 ) . in this letter , he announces , among many other things , that he has also solved the two - person case for le her ( @xcite , page 334 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ i also solved the problem on le her in the simplest case ; here is what i found . if we suppose that each player observes the conduct that is most advantageous to him , paul must only hold to a card that is higher than a seven and pierre to one that is higher than an eight , and we find under this supposition that the lot of pierre will be to that of paul as 2697 is to 2828 . supposing that paul also holds to a seven , then pierre must hold to an eight , and their lots will still be as 2697 to 2828 . nevertheless it is more advantageous for him not to hold to a seven than to hold to it , which is a puzzle that i leave you to develop . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this passage is carefully worded , yet it will be misinterpreted by montmort and waldegrave . as we will see , a key aspect that is neglected by montmort and waldegrave is the antecedent of bernoulli s conditional statement starting with `` if we suppose that each player observes the conduct '' montmort s reply , dated march 1 , 1712 ( @xcite , pages 337347 ) , highly praises bernoulli s prior letter . he complains that , being in paris , he has had no time and peace to think on his own and , as a consequence , the main object of his letter is to report progress made by his two friends , the abb dorbais and waldegrave , on a problem proposed by bernoulli , and on the problem of le her . on the latter , montmort reports that `` they dare however not submit to your decisions '' ( @xcite , page 338 ) . however , as he says in a passage that is key to understanding the forthcoming controversy , the abb dorbais also previously disagreed with montmort : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ when i worked on le her a few years ago , i told m. labb de monsoury what i had found , but neither my calculations nor my arguments could convince him . he always maintained that it was impossible to determine the lot of pierre and paul , because we could not determine which card pierre must hold to , and vice versa , which results in a circle , and makes in his opinion the solution impossible . he added a quantity of subtle reasonings which made me doubt a little that i had caught the truth . that is where i was when i proposed that you examine this problem ; my goal was to make sure from you of the goodness of my solution , without having the trouble of recalling my ideas on this which were completely erased . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ montmort then claims that bernoulli s solution confirms what he had found , a decision that prompts a reply from waldegrave objecting to bernoulli s solution , quoted at length in montmort ( @xcite ) , pages 339340 . according to waldegrave and the abb dorbais , it is not true that paul must hold only to an eight and pierre to a nine . rather , that paul should be indifferent to hold to a seven or to switch , and that pierre should be indifferent to hold to an eight or to switch . waldegrave wrote the following to montmort ( @xcite , page 339 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we argue that it is indifferent to paul to switch or hold with a seven , and to pierre to switch or hold with an eight . to prove this , i must first explain their lot in all cases . that of paul having a seven , is @xmath0 when he switches , and when he holds on to it his lot is @xmath1 if pierre holds on to an eight , and @xmath2 if pierre switches with an eight . the lot of pierre having an eight is @xmath3 if he holds on to it , and @xmath4 if he switches in the case that paul only holds on to a seven ; and @xmath5 by holding on to it , and @xmath6 by switching in the case that paul holds on to a seven , so here they are . the lots of paul @xmath7 , those of pierre @xmath8 or @xmath9 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ based on the numbers he obtains , waldegrave observes that `` 720 being more below 780 than 816 is above , it appears that paul must have a reason to switch with 7 '' ( @xcite , page 339 ) . the differences , @xmath10 and @xmath11 , are in the ratio @xmath12 , or @xmath13 , a ratio which later enters the discussion . in the rest of his argument , waldegrave talks of a weight instead of a reason . he first lets the weight that leads paul to switch be @xmath14 , and the weight that leads pierre to switch be @xmath15 . and he argues that the same weights lead paul and pierre to both strategies . @xmath14 leads paul to switch with 7 and , as a consequence , it also leads pierre to switch his 8 ; but what leads pierre to switch his 8 must lead paul to hold with 7 . so , @xmath14 leads paul to both switch with a 7 and hold on to it . the same goes for pierre . therefore , `` it is false that paul must only hold on to an 8 , and pierre to a 9 , '' which was bernoulli s claimed solution . the word `` probability '' comes up only once in this discussion , in the conclusion of the excerpt from waldegrave s letter to montmort . waldegrave writes ( @xcite , page 340 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ apparently mr . bernoulli was simply looking at the fractions that express the different lots of pierre and paul , without paying attention to the probability of what the other will do . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ montmort leaves the discussion there without further comment . 240pt@lcc@ + & * switch the 8 * & * hold the 8 * + & * ( and under)*&*(and over ) * + switch the 7 & & + ( and under ) & & + hold the 7 & & + ( and over ) & & + upon receiving montmort s letter , bernoulli agrees with these figures , saying that `` the lots they found for pierre and paul are very right '' ( @xcite , page 348 ) . and yet , when bernoulli proposes his solution , and when montmort eventually publishes a table of probabilities as an appendix to _ essay danalyse _ ( @xcite , page 413 ) , the numbers are different . the bernoulli montmort probabilities are shown in table [ table1 ] , which appears in hald ( @xcite ) , page 318 . none of the parties in this debate actually explain their calculations . waldegrave s probabilities are justified in todhunter ( @xcite ) , pages 107110 ; the bernoulli montmort probabilities are in hald ( @xcite ) , pages 315318 . the difference in the probabilities is that waldegrave s probabilities are conditional on paul having a seven in his hand and the bernoulli montmort probabilities are the marginal probabilities for all cards that paul may hold . in a letter dated june 2 , 1712 , bernoulli replies to waldegrave s argument by accusing him of committing a fallacy . he argues that if we suppose that @xmath14 leads paul to switch with a seven , and so leads pierre to switch with an eight ( if pierre knows paul switches with seven ) , then it also leads paul to hold on to a seven . therefore , @xmath14 both leads paul to switch with a seven and to hold on to a seven . his conclusion is that ( @xcite , page 348 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we are supposing two contradictory things at the same time ; that is , that paul knows and ignores at the same time what pierre will do , and pierre what paul will do . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ bernoulli explains that if we do not commit this fallacy regarding what paul and pierre know about the other s intent , we are led to reasoning in a circle , which shows that waldegrave s argument can not show anything . this argument is peculiar , and seems to suggest that bernoulli does not understand waldegrave s point . it might , however , be simply a misinterpretation of waldegrave s argument , for it is expressed in terms of weight rather than in terms of probability . the word `` weight '' or `` poids '' in french offers more opportunity for misinterpretation . moreover , bernoulli admits having written his letter hastily , as he was preparing for a long trip through the netherlands and england . as a result of this travel , some subsequent letters are delayed , and the arguments they contain do not follow the chronological order of when the letters were written . a letter to bernoulli , dated september 5 , 1712 ( @xcite , pages 361370 ) , announces that waldegrave and the abb dorbais have seen bernoulli s reply in which he accuses them of committing a fallacy . montmort includes a note from the abb dorbais in which he claims that waldegrave has written a beautiful and precise reply to bernoulli s objection ; the rebuttal , however , is not included . in this note , the abb dorbais also enjoins montmort to take a side in this dispute between them . this suggests that , even if montmort thanked bernoulli for his solution , which he claimed agreed with his own , montmort has not yet made up his mind as to whether bernoulli really solved the problem . the next letter concerning le her is from bernoulli to montmort , dated december 30 , 1712 ( @xcite , pages 375394 ) . adding important pieces to the puzzle , it contains a three - page discussion of le her ( bernoulli mentions having just received the june 2 letter , since it was sent from switzerland to holland , then to england , and finally back to switzerland ) . bernoulli insists that , despite waldegrave s arguments , paul does not do as well by abiding to the maxim of holding to a seven , than that of switching with a seven . bernoulli then says ( @xcite , page 376 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ if it were impossible to decide this problem , paul having a seven would not know what to do ; and to rid himself [ from deciding ] , he would subject himself to chance , for example , he would put in a bag an equal number of white tokens and black tokens , with the intent of holding to a seven if he draws a white one , & to switch with a seven if he draws a black one ; because if he put an unequal number he would be lead more to one party than to the other , which is against the assumption . pierre with an eight would do the same thing to see whether he must switch or not . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this comment introduces with clarity the idea of chance by `` the way of tokens '' ( as they will say later ) . what bernoulli says here seems to confirm that , at first , when he accused waldegrave of committing a fallacy , he did not interpret waldegrave s weights as probabilities . nonetheless , he suggests that the only probability allocation compatible with the supposed state of ignorance of the players is that each player chooses a strategy with probability @xmath16 . under these choices , he computes the lot of paul ( which is then @xmath17 ) and concludes that it would be a bad thing for paul to randomize in this way , since he could guarantee himself a lot of @xmath18 . therefore , bernoulli concludes paul must always switch with a seven . as bernoulli says ( @xcite , page 376 ) , `` it is better to make the choice where we risk less . '' he then explains the reasoning that he had left out of his hastily written letter from june 2 . in contemporary terms , he calculated the unconditional probability of winning under each pure strategy profile ( without assuming that any card has been dealt yet ) . he displays a refined version of the reasoning that led to accusing waldegrave of a fallacy , yet it does not do full justice to waldegrave s idea . eight months later , on august 20 , 1713 , montmort [ ( @xcite ) , pages 395400 ] finally replies to bernoulli , complaining that he has , despite his philosophical inclinations , been involved in political activities , and so he did not have the leisure for intellectual work . thus , his letter only contains scientific news . there is only one brief mention of le her ; he tells bernoulli that , despite his last effort to provide a thorough and precise argument , waldegrave and the abb dorbais are still unconvinced by his claimed solution . shortly after , in a letter dated september 9 , 1713 , bernoulli also asks montmort to explain his own views on the dispute . montmort obliges him in his letter dated november 15 , 1713 . this is the last letter published in the second edition of _ essay danalyse _ ( @xcite , pages 403413 ) . the letter also contains an excerpt of a letter from waldegrave and a table of the lots of paul and pierre for the four crucial combinations of strategies , which are summarized in table [ table1 ] . here , then , is montmort s understanding of the situation . to begin with , he agrees with bernoulli that it is not indifferent to paul to switch or hold with a seven , and to pierre to switch or hold with an eight , because of bernoulli s calculations of the unequal chances for each strategy . ( this shows that bernoulli and montmort use `` indifferent '' in the sense of having the same probability of winning . for waldegrave and dorbais , however , `` indifferent '' seems to mean , perhaps more awkwardly , that no strategy dominates the other in probability . ) this being said , montmort nonetheless disagrees with bernoulli that this establishes the strategy as a maxim , that is , as a rule of conduct that must be obeyed invariably to obtain the best results . rather , he thinks that it is impossible to establish such a maxim ( @xcite , page 403 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ [ t]he solution of the problem is impossible , that is , we can not prescribe to paul the conduct that he must adopt when he has a seven , and to pierre the conduct he must adopt when he has an eight . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ he grants that , if one is to choose a fixed and determined maxim , then switching on seven , for paul , will be better than any other , yet paul can hope to make his lot better . why , then , would a solution be impossible ? would the solution not be the optimum that one can reach in paul s hope of making his lot better ? montmort claims that , whereas he used to think that the use of black and white tokens to randomize strategies could avoid the `` circle , '' he does not think that anymore . he gives a general formula to find the probability of winning with a certain probability allocation for what we call a mixed strategy : @xmath19 where @xmath20 is paul s probability of switching with seven , @xmath21 is paul s probability of holding the seven , @xmath22 is pierre s probability of switching with an eight , and @xmath23 is pierre s probability of holding on to an eight . but how should the probabilities be chosen ? montmort claims that any argument will only inform us of what paul must do conditionally to what pierre does and vice versa , which leads us into a circle once again . he concludes that bernoulli s arguments to show that a circle does not occur are wrong , and instead formulates this thesis ( @xcite , page 404 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ [ w]e must suppose that both players are equally subtle , and that they will choose their conduct only based on their knowledge of the conduct of the other player . however , since there is here no fixed point , the maxim of a player depends on the yet unknown maxim of the other , so that if we establish one , we draw from this supposition a contradiction that shows that we must not have established it . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ montmort also disagrees with bernoulli that , under pain of contradiction , if we are to use white and black tokens to randomize , we must use an equal number of tokens . instead , he thinks that the probability of winning calculated for the fixed and determined maxims shows that paul must switch more often with a seven than hold on to it . yet , he maintains ( @xcite , page 405 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ but how much more often must he switch rather than hold , and in particular what he must do ( here and now ) is the principal question : the calculation does not teach us anything about that , and i take this decision to be impossible . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ thus , montmort believes , it seems , that there is no optimal probability allocation . but he has another reason for believing that the solution of the game is impossible . he has in mind the game les trennes ( @xcite , pages 406407 ) . montmort also believes that it is impossible to prescribe any strategy of play in les trennes because the players might always try , and indeed good players will try , to deceive other players into thinking that they will play something they are not playing , thus trying to outsmart each other ( `` jouer au plus fin '' is the phrase used in french ) . as he was finishing his letter , montmort received one from waldegrave and quoted extensively from it to bernoulli . _ essay danalyse _ essentially concludes with waldegrave s letter . waldegrave refers to a formula , which is not included by montmort ; it presumably is the formula displayed above . he explains that , if @xmath24 and @xmath25 ( so that the probability of paul switching with a seven is @xmath26 ) , then the lot of pierre is going to be @xmath27 no matter what @xmath22 and @xmath23 are . this shows that @xmath27 is paul s minimum lot . he can only adopt another conduct in the hope of making his lot better . this shows , he claims , that both bernoulli and ( formerly ) himself were wrong to claim that the lots of paul was to that of pierre as 2828 is to 2697 ; if both players play in the most advantageous way , paul s lot is @xmath28 . waldegrave is convinced that this is something that both bernoulli and montmort will agree to , now that it is agreed that one can use a randomized strategy . he also explains that , if pierre uses @xmath29 and @xmath30 , then @xmath27 will also be paul s maximum lot . waldegrave also asserts that it is impossible to establish a maxim ; he grants , however , that it is impossible for him to show this with the same level of evidence . this is often taken incorrectly as evidence of a lack of waldegrave s mathematical abilities . waldegrave is instead referring to the situation in which players may try to outsmart each other . waldegrave agrees that if paul does not use @xmath24 and @xmath25 , then it is possible for paul to do better than @xmath31 provided that pierre does not play in the best way . on the other hand , it would be worse if pierre plays correctly . furthermore , waldegrave remarks ( @xcite , page 411 ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ what means are there to discover the ratio of the probability that pierre will play correctly to the probability that he will not ? this appears to me to be absolutely impossible , and thus leads us into a circle . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ as with montmort , his main concern is that it is always possible for the players to try to outsmart each other ( `` jouer au plus fin '' ) . examining the detailed arguments provided by montmort , bernoulli and waldegrave reveals a picture that contrasts with the judgment that they were essentially confused on the fundamental concepts and methods required to solve a strategic game such as le her . in fact , we maintain that they understood most of the aspects of the problem with clarity . there are , however , a number of important outstanding issues left unresolved in the correspondence on le her as it appears in _ essay danalyse_. let us review them briefly . it is true that the letters reveal a certain type of misunderstanding ; however , it is not conceptual confusion , but rather mutual misinterpretation due to using terms differently . an instance of this is whether it is indifferent to paul to switch or hold to a seven . on the one hand , both montmort and bernoulli claim that it is not indifferent to paul because the chances of winning are not identical . on the other hand , waldegrave claims that it is indifferent , and the reason for that seems to be that neither pure strategy dominates the other in probability . another instance of this is the disagreement they appear to have on the existence of a circularity in the analysis of the game . montmort and waldegrave assert that there is a vicious circle that prevents one from establishing a maxim ; the circle they discuss , however , is really a regression _ ad infinitum _ , that is , to establish a maxim , we always need to go one step further in the `` @xmath14 must know what @xmath15 does '' loop ( bernoulli agrees with this point ) . however , bernoulli claims that there is a circle in waldegrave s argument , in the sense that either his argument is contradictory or a _ petitio principii _ ( but bernoulli is not considering randomizing strategies at this point ) . again , they are only contradicting each other in the wording , not in the idea . finally , a third instance is that montmort and waldegrave claim that the solution of the game is impossible , whereas bernoulli does not . here again , they disagree on what it means to `` solve '' the game le her . bernoulli claims that the solution is the strategy that guarantees the best minimal gain what we would call a minimax solution and that as such there is a solution . however , despite understanding this `` solution concept , '' montmort and waldegrave refuse to affirm that it `` solves '' the game , since there are situations in which it might not be the best rule to follow , namely , if a player is weak and can be taken advantage of . clearly , the concept of solution they have in mind differs from the minimax concept of solution . this latter concept , in addition to the probability of gain with a pure strategy and the probability allocation required to form mixed strategies , requires that we know the probability that a player will play an inferior strategy . but , they assert , this can not be analyzed by calculations , so the game can not be solved . this being said , there are a number of things that are said that suggest a certain level of confusion at a conceptual level . the two most important are these . first , bernoulli appears to have some difficulty with the relation between the knowledge of the players and the probabilities involved in mixing strategies . his circularity objection to waldegrave is awkward and somewhat mystifying . moreover , his argument that , if we allow randomized strategies with black and white tokens , it must be because neither player knows what the other player will do , and that as a result the only acceptable probability allocation of @xmath16 is problematic . this kind of mistaken argument has been repeated over the centuries by some of the greatest minds in probability , statistics and game theory . second , montmort understands very well the idea of randomizing strategies , but he nonetheless claims that there is no optimal probability allocation that can be calculated . this claim , however , was made before consulting waldegrave s letter in which he reveals the optimal probability . referring to a letter from bernoulli to montmort dated february 20 , 1714 , @xcite , in his treatment of le her , mentions only that bernoulli accepted waldegrave s solution to the problem . however , bernoulli had other things to say about le her in that same letter . henny also refers to a letter of january 9 , 1715 , from waldegrave to bernoulli in which waldegrave seemingly admits to bernoulli that he does not have the mathematical skills to actually prove his results . what henny leaves out is that the letter was written in reply to a detailed criticism of the solutions to le her that bernoulli had sent earlier to montmort . after some personal news and apologies for not writing sooner , in his letter of february 20 , 1714 , to , bernoulli initially thanks montmort for correcting , editing and making clearer his letters that montmort had printed in _ essay danalyse_. then follows the discussion of le her that @xcite only briefly mentions . initially , bernoulli suggests that the controversy is essentially over : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ concerning le her , i seem to have foreseen that in the end we would all be right . however , i congratulate mr . de waldegrave who has the final decision on this question , and i willfully grant him the honor of closing this affair _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ despite this , bernoulli still claims that he disagrees on a few minor points , and these point directly to the outstanding issues we mentioned above . the main concern is the relation between `` establishing a maxim '' and solving the problem of le her posed by montmort in his book . bernoulli states : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ one can establish a maxim and propose a rule to conduct one s game , without following it all the time . we sometimes play badly on purpose , to deceive the opponent , and that is what can not be decided in such questions , when one should make a mistake on purpose . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this point was raised before by montmort and waldegrave , but they do not consider that such a play would be necessarily a mistake . whether or not this kind of play is a mistake , we saw that from the same consideration , montmort and waldegrave conclude that solving the problem is impossible . however , bernoulli now phrases things more carefully : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ mr . de waldegrave wrongs me on p. 410 by claiming that i once said that the lot of paul is to that of pierre as @xmath32 . if you carefully read my letter from oct 10 , 1711 , you will find that i did not say it absolute and without restriction . i beg you to consider those words : once we have determined or rather supposed what are the cards to which the players will hold , etc . and the following words . you will see that i there supposed that the players want to hold to a fixed and determined card , and indeed i had not thought about the way of tokens , which , as mr . de waldegrave said , is not among the ordinary rules of the game . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ bernoulli essentially says that he was misinterpreted and that he only computed the best odds of winning with a pure strategy , not that he established what a player should do in an actual game . moreover , if we grant his supposition , then he has found the most advantageous maxim . after this correction , bernoulli thinks the discussion is over , saying , `` we are thus all agreeing , and we have made peace ; _ canamus receptui _ [ sing retreat ] . '' in his response to bernoulli , dated march 24 , 1714 , montmort concurs by writing , `` i am quite pleased that we are all together by and large agreeing . '' in this letter montmort claims that he disagreed with bernoulli on some aspect of the corrected interpretation of his position , but he leaves it to a later letter to explain . however , in his next letter to bernoulli , november 21 , 1714 , montmort does little to clarify . he says , `` if it is ever permitted to say to two persons maintaining contradictory claims that they are both right , it is assuredly at this occasion in our dispute . '' montmort emphasizes that what he seeks is the correct advice that should be given to the players , but the discussion does not go much further . on august 15 , 1714 , montmort sent a letter to bernoulli containing a two - page `` supplement '' that reignites the debate . he makes six points . first , he claims that telling paul always to switch with a seven is bad advice , since his minimum lot is then 2828 . second , that it would be better advice to tell him to do whatever he pleases with a seven , so that he can look at both options indifferently . third , we can not say that this would be the best advice either , for knowing that , pierre would switch with an eight , in which case paul should certainly have held on to a seven . this leads to a vicious circle . fourth , if we admit the way of tokens , the best advice that he knows is to tell paul to have the ratio @xmath33 for switching with a seven . but even then , he does not think that we can demonstrate that it is the best advice . fifth , he claims that it is impossible at this game to determine the lot of paul , because one can not determine what manner of playing is the most advantageous to each player , even when we admit a randomized strategy . this point makes explicit for the first time montmort s ( and presumably waldegrave s ) idea that you can only claim that you have found the lot of a player ( which is what montmort s problem in _ essay danalyse _ demanded ) if we can determine what is the best way to play . moreover , determining the best way to play demands knowing more than the optimal token ratio for the randomized strategy . he adds that , of course , some methods of playing are better than others , as informed by the chances that have previously been calculated . he concludes , sixth , that he would not know what advice to give paul if he had to . this letter sharpens the debate , in that it makes explicit the connection between `` solving '' a game and giving advice for play in actual situations . in a long letter to montmort dated august 28 , 1714 , along with a `` supplement '' dated november 1 , 1714 , bernoulli replies to montmort point by point . he asks montmort a question that is meant to dismiss his argument : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ if , admitting the way of tokens , the option of 3 to 5 for paul to switch with a seven is the best you know , why do you want to give paul other advice in article 6 ? it suffices for paul to follow the best maxim that he could know . it is not enough to claim that there is still a circle despite my reasons , one must fight my reasons . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ and he continues : `` it is not impossible at this game to determine the lot of paul . '' to counter montmort s previous argument , he once again insists that either paul knows what pierre will do , in which case his maxim is clear , or he does not , in which case paul should use the probability @xmath16 in the randomized strategy to determine what to do . as he admits , this is the exact same position he had at the beginning of the discussion , supported by the exact same argument . thus , it seems that bernoulli has missed the point montmort made explicit in his august 15 letter . it is at this point that waldegrave reenters the debate at montmort s request . in a letter dated january 9 , 1715 , waldegrave reiterates the six points that montmort had laid out for bernoulli in his letter of august 15 . for each of the six points , waldegrave s arguments are longer and more detailed than what montmort had previously given . it is not until march 22 , 1715 , that montmort replies to bernoulli on this dispute . it is part of a very long letter that also contains the main topic for their further correspondence , infinite series . in this letter , montmort writes once again about his views . they are the same as what we have seen already . however , montmort stresses that a lot of what remains under discussion is based on inconsistent terminology and . in essence , he believes that the outstanding disagreements are only apparent contradictions . nonetheless , he introduces one more element to clearly his view . he distinguishes between the advice that he would put in print , or give to paul publicly , and the advice he would give so that only paul hears it . montmort claims that , for the former , he would choose the mixed strategy with @xmath24 and @xmath25 , since it is the one that demonstrably brings about the lesser prejudice . however , he explains that , in practice , if paul is playing against an ordinary player and not a mathematician , he would quietly give different advice that could allow paul to take advantage of his opponent s weakness . as he explains , the objective of this sort of analysis is not only to provide a maxim to otherwise ignorant players , but also to warn them about the potential advantages of using finesse . however , this latter part is not possible to establish , and it is in this sense that there is no possible solution to this problem . the next letter , sent by bernoulli to montmort on may 4 , 1715 , disregards montmort s nuance . to begin , bernoulli `` is forced to admit that he does not precisely know on what point [ they ] contradict each other . '' nonetheless , bernoulli explains that , in his view , the distinction between public and private advice , the possibility of using finesse , or something similar , does not alter the fact that @xmath24 and @xmath25 is the best solution , and that it determines the lot of paul ( so that not only is the game solvable , but it is indeed solved ) . despite bernoulli s explanation , montmort s next letter , dated june 8 , 1715 , once again reiterates that `` you have badly solved the proposed question , or you have not solved it at all . '' he makes explicit what he takes the proposed question to be : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the question is and has always been to know whether we can establish the lots and as a result the advantage of playing first under the supposition not that pierre and paul follow this or that maxim ( this would have no utility , no difficulty ) , but that both of them having the same skills , each follow the conduct that is the most advantageous . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ montmort then says that this dispute is beginning to bore him . he considers that furthering it will not make them learn anything new and that in the end the dispute must be about some other thing . our presentation of the correspondence makes it clear that they are using different concepts of solution ; bernoulli s in essence is the concept of the minimax solution , whereas montmort s further depends on the probability of imperfect play ( i.e. , on the skill level of the players ) . around this time , montmort s interest shifts from probability and its applications to infinite series . in fact , most of the remaining correspondence with bernoulli turns to that topic . at the same time , montmort began an extensive correspondence with brook taylor , also mainly on infinite series ( st . john s college library , cambridge , taylorb / e4 ) . although the dispute with bernoulli seems to have petered out , montmort was not yet done with it . in a letter dated july 4 , 1716 , montmort asked taylor to examine his dispute with bernoulli about le her and to express his opinion on who was right . he referred taylor only to the correspondence that appears in _ essay danalyse_. taylor apparently wrote back but with the wrong impression about what montmort wanted . montmort replied to taylor on august 4 , 1716 , that he did not want any new research into the problem but only to examine , at his leisure , which of bernoulli or montmort was right . in a letter to taylor dated november 10 , 1717 , montmort thanked taylor for his opinion on the dispute and concluded the letter by saying that waldegrave would write him about le her as well as another game . unfortunately , neither taylor s reply expressing his opinion nor waldegrave s letter to taylor are extant . compared to the discussion of le her , the remaining discussion in the post-1713 correspondence with regard to probability problems is relatively minor . for example , after the remarks on le her that bernoulli made in his letter of february 20 , 1714 , to montmort , bernoulli comments that he thinks there is an error in montmort s solution to a problem related to the jeu du petit palet in _ essay danalyse_. he asks montmort to check his solution . the problem appears to be problme iv in montmort ( @xcite ) , page 254 . the jeu du petit palet is a game in which players toss coins or flat stones ( the `` palets '' ) toward a target set on the ground or a table . the player with the most coins or stones on the target wins . the english equivalent game is called chuck - farthing or chuck - penny . what takes up much of the discussion , other than le her , is news about abraham de moivre s work . de moivre corresponded with both montmort and bernoulli until about 1715 when he ceased corresponding with either of them . prior to this discussion , bernoulli had sent de moivre a general solution to the problem of the pool on december 30 , 1713 ( @xcite , pages 106107 ) . a report on de moivre s activities in probability takes up part of a letter from bernoulli to montmort dated april 4 , 1714 . bernoulli mentions that de moivre has sent him a long letter with reports of new solutions that will appear in a much expanded version of his treatise _ de mensura sortis _ ( de moivre , @xcite ) . de moivre s new work , which was entitled _ the doctrine of chances _ , did not appear until 1718 ( de moivre , @xcite ) . no details are given to montmort other than that de moivre has made inroads in three areas . first , de moivre used his own method for the solution of the problem of the pool to generalize it to more than three players . second , he developed a new kind of algebra to solve probability problems . finally , bernoulli reports that de moivre considered that nearly all problems in probability can be reduced to series summations . not only did de moivre report that he had generalized the problem of the pool , but he also sent bernoulli his solution to the problem . at the time of his writing to montmort , bernoulli had not read the solution and did not pass the solution on to montmort . the new algebra is probably the one that de moivre developed for finding probabilities of compound events . see , for example , hald [ ( @xcite ) , pages 336338 ] for a modern discussion of this topic . this part of the letter ends with what might be interpreted as a nasty comment about de moivre : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ i will share here in confidence what he wrote to me concerning you . here is what he told me about your comments that i had sent him . ` i can not stop myself etc . our society etc . i just received etc . kind [ regards ] . ' after the letter i find written there these words : ` in a sense , ' that made me laugh . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is difficult to know what exactly bernoulli is saying here . it appears that he sent de moivre montmort s severe criticism of _ de mensura sortis _ that montmort published in _ essay danalyse _ ( @xcite , pages 363369 ) . later that month , montmort reported back to bernoulli that he received a very polite and fair letter from de moivre in which de moivre announced that he had found a new solution to the problem of the duration of play . see bellhouse [ ( @xcite ) , pages 111114 ] for a discussion of the publication of this solution . de moivre sent reports about more of his results in probability to montmort and montmort sent on a pr ' ecis of these results to bernoulli in a letter dated august 15 , 1714 . many of the results that montmort mentions found their way into _ the doctrine of chances _ , including what is called woodcock s problem discussed in bellhouse [ ( @xcite ) , pages 125126 ] . on august 28 , 1714 , bernoulli finally wrote to montmort enclosing a copy of de moivre s general solution to the problem of the pool . in the letter , bernoulli asks montmort to tell him what he thinks of the solution . he further states that it appears that de moivre is using bernoulli s approach to the solution for three and four players that appears in _ essay danalyse _ ( @xcite , pages 380387 ) . at the same time he is using an analytical approach rather than infinite series ( de moivre actually used a recursive method for his general solution ) . montmort replied on march 22 , 1715 , that he agrees with bernoulli s assessment . on returning from a trip to england , montmort reported to bernoulli in a letter dated june 8 , 1715 , that one of bernoulli s solutions to the problem of the pool had just been printed in the _ philosophical transactions _ ( bernoulli , @xcite ) . bernoulli had sent de moivre two solutions ; de moivre claimed he had found an error in the first solution . many in the past have tried unsuccessfully to identify the waldegrave who solved the problem of le her and who suggested the problem of the pool , often called waldegrave s problem . @xcite reviewed these attempts at identification and narrowed the field down to charles , edward or francis waldegrave , the three brothers of henry waldegrave , 1st baron waldegrave . bellhouse argued for charles waldegrave , but in view of new information his choice was incorrect . key to the proper identification is that several waldegraves siblings , cousins and at least one uncle of henry followed king james ii into exile in france after james was deposed in 1688 . the proper identification of the waldegrave of interest may be found in legal papers in the archive nationales de france in conjunction with a letter from waldegrave to nicolaus bernoulli ; the letter to bernoulli is signed only `` waldegrave '' and is the only known letter in waldegrave s hand that is extant ( universittsbibliothek basel l ia 22 , nr . other waldegrave signatures to compare to the one on bernoulli s letter can be found on various legal documents , two in france ( archives nationales de france mc / et / xvii/486 and 514 ) and one in england ( house of lords record office hl / po / jo/10/1/439/481 ) . see figure [ table2 ] . from the signatures , it is obvious that francis is the waldegrave of interest . from these records , it is also apparent that charles waldegrave handled the family s affairs in england while francis waldegrave took charge of them in france . what little is known of the life of francis waldegrave comes mostly from montmort s correspondence with brook taylor and nicolaus bernoulli . montmort reported to taylor one of waldegrave s political activities . waldegrave was planning to take part in the jacobite uprising in england in 1715 . he was to be part of an invasion force led by the son of james ii , james stuart . the uprising in england fizzled out , james stuart remained in france and waldegrave fell ill just prior to the time when the planned invasion was to occur . montmort called waldegrave s illness apoplexy ; it was probably a stroke . from time to time , montmort commented to taylor and bernoulli about waldegrave s illness , recovery and setbacks . at one point , for a cure or a rest , waldegrave took the waters at a spa in france . he also spent time at montmort s chateau . though ill , he was alive in france in 1719 when montmort died so that the flow of information to bernoulli and taylor about waldegrave stopped . presumably , waldegrave died in france . how waldegrave obtained his mathematical training is unknown . in whatever way he was educated , he was an adept amateur mathematician . this is contrary to henny s interpretation of waldegrave s skills . for example , henny [ ( @xcite ) , page 502 ] claims that waldegrave did not have the mathematical skills to work out a general method of calculation in le her . on the contrary , there is a hint of the fairly high level of waldegrave s mathematical abilities in a letter from montmort to bernoulli dated march 24 , 1714 . there montmort says that he is getting waldegrave to read lhpital s ( @xcite ) calculus book _ analyse des infiniment petits _ and that waldegrave has a natural aptitude for mathematics . at the time that montmort was sending _ essay danalyse _ to his publisher , francis waldegrave was living in rue princesse near glise saint - sulpice in paris . in modern paris , it is a @xmath34 kilometer walk to montmort s publisher in rue galande . in section [ sec1 ] it was mentioned that montmort was staying only @xmath35 meters from his publisher . this was not the only time that montmort enlisted a colleague to take on some of the tedious parts of getting results to print . after montmort sent brook taylor a number of theorems about infinite series , they decided that the results should be published in the royal society s _ philosophical transactions _ in a letter dated june 15 , 1717 , montmort gave taylor complete editorial control over the paper that included having taylor translate the results from french into latin ( st . john s college library , cambridge taylorb / e4 ) . taylor replied august 9 , 1717 , saying that he had made many changes and corrections to the paper ( st . john s college library , cambridge , taylorb / e5 ) . the unpublished letters between bernoulli and montmort reveal a much more complex story than either @xcite or @xcite have described . the entire group bernoulli , montmort and waldegrave were for the most part clear about the issues at the conceptual level . in the end it came down to a disagreement about what it meant to solve a problem . further , henny recognized many modern game theory concepts , but we show that the group s understanding of the modern notions is deeper than what henny realized . apart from the technical and conceptual aspects of le her and other probability problems , we also get a glimpse into the social side of a rich amateur mathematician at work . montmort was a good mathematician , but mathematics was his hobby and at times he did not have time to pursue his hobby . there is a bit of quid pro quo in his relationships with bernoulli , taylor and waldegrave . montmort acquires some status through his connections to artists , philosophers and scientists . he can impose on his scientific friends to do some of the more menial work for him in getting his research to print . on the other side , his scientific friends enjoy his hospitality , his gifts and the benefits of any political and scientific connections that he may have . traditionally , the mixed strategy solution with @xmath24 and @xmath25 for le her has been attributed to waldegrave . it certainly appears to be the correct attribution based on the correspondence in the second edition of _ essay danalyse_. however , in the long letter from montmort to bernoulli dated march 22 , 1715 , that covers discussions of le her , de moivre and other topics , montmort appears to claim priority of solution . as part of the discussion of le her , he says , `` although i first found the determination of the numbers @xmath20 and @xmath21 , @xmath22 and @xmath23 '' montmort s suggestion of priority could have come about as a result of a conversation between waldegrave and montmort , with waldegrave putting pen to paper . this illustrates fasolt s ( @xcite ) claims about the limits of history . our data from the past is what has been written , not what has been spoken . further , we can never know the tone behind what was written , such as bernoulli s apparently nasty comments to montmort about de moivre in his letter of april 14 , 1714 . instead of coming up in conservation , montmort may be claiming priority because he found the general formula in @xmath20 , @xmath21 , @xmath22 and @xmath23 ; the numbers were only a special case . or it could be something else . like le her itself , depending on how the problem is approached , the assignment of priority is a problem with no solution . we would like to thank dr . fritz nagel of universitt basel for giving us access to the correspondence between nicolaus bernoulli and pierre rmond de montmort . we also thank kathryn mckee and jonathan harrison of st . john s college library , cambridge , for providing us with copies of the correspondence between brook taylor and montmort . the originals of the letters of bernoulli , montmort and waldgrave are in universittsbibliothek basel . the letters from montmort to bernoulli are catalogued handschriften l ia 22:2 nr.187206 and from bernoulli to montmort are l ia 21:2 bl.209275 . the letter from bernoulli to waldegrave is catalogued l ia 21:2 bl.229v232r and the letter from waldegrave to bernoulli is l ia 22 , nr . 261 . when referencing these letters , we have to do so by the date , writer and recipient , rather than the catalog numbers Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: a growing body of evidence suggests that the stellar population of the galactic nucleus is distinct from that of the bulge . surface brightness measurements from the nir to the fir show that in contrast to the bulge , the central few hundred parsecs of the galaxy are dominated by a flat , disk - like distribution of stars , gas , and dust @xcite . this difference is also reflected in the kinematics : oh / ir stars in the central hundred parsecs show higher rotational velocities than expected for an inner single - component , bulge population @xcite . evidence for ongoing star formation also distinguishes the nucleus from the bulge . the galactic bulge is composed primarily of an old starburst - like population formed early , some 710 gyr ago @xcite . in contrast , the galactic nucleus harbors substantial young and intermediate - age stellar populations . intermediate - age populations have been most commonly inferred from broad - band photometry @xcite . star formation tracers have also been used to study intermediate - age populations , including young supergiants and luminous agbs , which trace 10 myr - 1 gyr populations @xcite , and oh / ir stars , which trace 1 - 3 gyr populations @xcite . the young stellar populations in the galactic nucleus have been the subject of much recent work . these populations are predominantly concentrated in three young , massive clusters : the arches and quintuplet clusters ( figer et al . 1999 , 2002 ; stolte et al . 2005 , and references therein ) , approximately 30 - 50 pc in projected distance from the galactic center , and the central cluster , located within the central parsec ( krabbe et al . 1991 , 1995 ; paumard et al . 2006 , and references therein ) . star formation in the nucleus is thought to be ultimately linked to the inward transport of gas induced by the galactic bar @xcite . however , precisely how gas is funneled from the outer galaxy to the nucleus and how this process affects the resultant star formation , remains poorly known . at present , there is a concentration of molecular gas 2 - 8 pc from the galctic center termed the circumnuclear disk @xcite . if such a structure routinely fuels star formation in the central parsec , the stars formed in this region may represent an entirely distinct population from the arches and quintpulet clusters or the intermediate age populations found throughout the @xmath5200 pc nucleus . dynamical effects due to the supermassive black hole ( genzel et al . 2003 , ghez et al . 2005 , and references therein ) and the large stellar and remnant density in the central parsec may also lead to population differences between the central cluster and that in the larger nucleus , either prior to star formation or afterwards . a number of investigators have attempted to address the nature of the central cluster and its relationship to the larger 100 - 300 pc nucleus . most recently , @xcite investigated the properties of the young stellar population in the central parsec , spectroscopically identifying nearly 100 ob and wolf - rayet stars . they find that the majority of the young stars reside in two , inclined and counter - rotating disks , suggesting in situ star formation in dense gas accretion disks . studies of the older stellar population in the central parsec have been primarily limited to broadband photometry . they have suggested that the fraction of low mass stars in the population increases with distance from the center , perhaps due to dynamical mass segregation @xcite . a detailed analysis , however , of the late - type giant population requires spectroscopy , owing to the large scatter in extinction near the galactic center and the intrinsic variations in giant star colors . @xcite pioneered work in this area , using spectroscopic and photometric observations of the most luminous giants and supergiants ( 50% complete at @xmath610 mag . ) to construct an h - r diagram for stars in the inner 5 pc . they report that the gc star formation rate in the central few parsecs is largely similar to that of the bulge . specifically , they find that the majority of stars formed more than 5 gyr ago , though they also find evidence that significant star formation also occurred during the past 100 myr . however , their conclusions are limited by their bright magnitude limit , which samples only short - lived evolutionary stages , for which theoretical models are uncertain . in this paper , we build on the work of @xcite in an effort to better characterize the late - type giant population within the central parsec . we report deep photometric and spectroscopic observations of 329 late - type giants in the gc , complete to 50% at @xmath615.5 mag . our observations for the first time include the helium - burning red clump , as well as the red giant branch and asymptotic giant branch . this improved magnitude limit allows for the most robust picture of the galactic center star formation history to date , as these stars are much better understood than the supergiants and luminous agb stars studied by @xcite . in 2 , we present our observations and the techniques used to construct the hertzsprung - russell ( h - r ) diagram . section 3 discusses the resulting h - r diagram and implications for the gc star - formation history . we discuss our results in 4 and conclude with 5 . we observed eight galactc center fields between march and september 2006 using the integral field spectrometer spiffi @xcite in conjunction with the macao adaptive optics module @xcite mounted on the sinfoni eso vlt facility . we observed these fields for the dual purpose of identifiying main sequence b stars outside the central parsec ( martins et al . 2007b , in prep ) and gathering spectroscopic observations for a large number of late - type giants . we chose fields with a radial distance from sgr a@xmath8 of less than @xmath9 arcsec , selecting regions outside the minispiral to avoid nebular contamination in brackett @xmath10 , and choosing fields north of sgr a@xmath8 to minimize separation from the ao guide star . we also avoided regions with extremely bright stars ( @xmath11 ) . as stars this bright are relatively rare , the selection bias in avoiding these stars is negligible . if the fields had instead been chosen at random , we expect no more than a few @xmath11 stars would be selected . we observed each selected sinfoni field in 50 @xmath12 100 mas pixel mode , resulting in 0.2 arcsec resolution and a @xmath13 field of view . we simultaneously observed all fields in @xmath2 and @xmath14 band , leading to a spectral resolution of r @xmath15 1500 . figure [ mos ] displays the location of the observed fields . the total exposure time for all fields was 4200 s , except for the fields at ( 17@xmath16 , 17@xmath16 ) , ( 5@xmath16 , 11@xmath16 ) , and ( -8@xmath16 , 8@xmath16 ) , which had total exposure times of 6600 s , 7200 s , and 2400 s , respectively . the fields span a range in projected distance from sgr a * of 4 arcsec to 26 arcsec . a histogram of the observed stellar positions is shown in figure [ dist_hist ] . we reduced the raw data using a standard procedure to perform flat - fielding , sky subtraction , and wavelength calibration @xcite . following extraction , we removed the stellar continua by dividing by a second degree polynomial fitted to the line - free regions of the stellar spectra . the normalized spectra allow us to directly compare the vlt spectra to the normalized catalog spectra in @xcite and @xcite . a typical resultant gc @xmath14 band spectrum following normalization is shown in figure [ spectrum ] . three template spectra from the @xcite spectral library are also shown for comparison . to classify the stars in our sample , we used the @xmath17co 2.2935 @xmath18 m , @xmath19 rovibrational band head , the strongest feature in our spectra and a well - known @xmath7 indicator , for a given luminosity class @xcite . we use only this feature , as it is largely free of nebular and telluric lines , making it more reliable than other @xmath20 and @xmath21band signatures in our spectra . we note that in contrast to the work of @xcite , we do not need the h+k band h@xmath22o feature to reliably predict effective temperatures , as our sample consists only of giants with @xmath2310.3 ( see 2.2 ) . due to their rarity , stars sufficiently bright to be supergiant or long period variable candidates are not contained in our sample . m co index ( see text for definition ) as a function of temperature for the comparison stars taken from the @xcite and @xcite catalogs . the solid line is a least - squares fit to both data sets for stars with 17 @xmath24 co @xmath25 3 . ] to define a @xmath7 - @xmath17co index relation , we computed the co index defined by @xcite . although not as well - known as other co indices ( e.g. equivalent width , photometric indices of @xcite ) , @xcite showed that this index correlates well with other co indices . furthermore , it has the advantage of being insensitive to small variations in the nearby continuum because it does not require a fit to the continuum . the index is defined as co % = [ ( 1 @xmath26 f@xmath27 f@xmath28 ) @xmath12 100 ) ] , where both bands are 0.015 @xmath18 m wide and the continuum and co band centers are defined as 2.284 @xmath18 m and 2.302 @xmath18 m , respectively . we note that although we adopt the same definition for the co index as @xcite , our spectra are normalized while theirs are not , and thus , our computed indices are not directly comparable to theirs . we calculated uncertainties in the computed indices , assuming the noise is dominated by photon statistics of the source and background and that the uncertainty in the band is approximately equal to that of the nearby continuum . following @xcite , we used co strength to estimate effective temperatures for the cool giants in our gc sample . to define a co index versus @xmath7 relation , we used a set of archival comparison star spectra taken from @xcite and @xcite with well - determined effective temperatures given in the literature . in an effort to avoid systematic errors in our derived index@xmath26@xmath7 relation , we used a large number of references , assuming temperatures derived from a wide variety of methods . a summary of the comparison star data is given in table [ comp_table ] , and the resulting relationship between co index and @xmath7 is shown in figure [ comp_relation ] . the @xcite and @xcite relationships are in very good agreement with each other ( for stars in common , @xmath29co @xmath30 ) , suggesting that the co index is largely independent of resolution or observing system . the relation is tightest for co @xmath31 . as no stars in our gc sample have computed indicies exceding this value , we did not include comparison stars with co @xmath32 in our fit . the resulting index-@xmath7 relation using both empirical data sets is : @xmath33 @xmath7 = 3.7351 @xmath26 0.0060 @xmath12 co @xmath26 0.00040 @xmath12 co@xmath34 , for all stars with measured indices @xmath35 co @xmath25 3 . the errors in the derived coefficients are 0.0092 , 0.0022 , and 1.2@xmath36 , respectively . llllllll hr 7001 & a0 va & wh & 1.760 & 0.004 & 9420 & 60 & @xcite + hr 4295 & a1 v & wh & 1.434 & 0.003 & 9600 & 25 & @xcite + hr 5054 & a1 vp & wh & 1.823 & 0.004 & 8760 & 260 & @xcite + hr 6378 & a2 v & wh & 1.953 & 0.004 & 8690 & 78 & @xcite + hr 4534 & a3 v & wh & 1.695 & 0.005 & 8857 & 185 & @xcite + hr 4357 & a4 v & wh & 1.516 & 0.007 & 8243 & 150 & @xcite + hr 7557 & a7 v & wh & 1.147 & 0.004 & 7500 & 200 & @xcite + hr 2943 & f5 iv - v & wh & 1.176 & 0.014 & 6532 & 39 & @xcite + hr 6927 & f7 v & wh & 1.219 & 0.014 & 6087 & 22 & @xcite + hr 4375 & f8.5 v & wh & 1.540 & 0.015 & 5676 & 66 & @xcite + hr 21 & f2 iii & wh & 1.563 & 0.011 & 6847 & 137 & @xcite + hr 403 & a5 iii - iv & wh & 1.828 & 0.008 & 8420 & 360 & @xcite + hr 1412 & a7 iii & wh & 1.554 & 0.007 & 7690 & 320 & @xcite + hr 1457 & k5@xmath37 iii & wh & 12.633 & 0.023 & 3866 & 35 & @xcite + hr 2985 & g8 iii & wh & 5.540 & 0.018 & 5001 & 56 & @xcite + hr 3003 & k4 iii & wh & 13.944 & 0.023 & 3961 & 99 & @xcite + hr 3323 & g5 iiia & wh & 3.045 & 0.017 & 5136 & 88 & @xcite + & & kh & 3.039 & 0.007 & & + hr 4031 & f0 iii & wh & 2.145 & 0.011 & 6880 & 150 & @xcite + hr 4069 & m0 iii & wh & 14.143 & 0.021 & 3730 & 149 & @xcite + hr 4517 & m1 iii & wh & 13.894 & 0.019 & 3828 & 53 & @xcite + hr 4883 & g0 iiip & wh & 1.819 & 0.014 & 5747 & 54 & @xcite + & & kh & 1.377 & 0.007 & & + hr 5017 & f3 iii & wh & 1.395 & 0.017 & 7141 & 181 & @xcite + hr 5340 & k1.5 iiip & wh & 10.997 & 0.021 & 4233 & 55 & @xcite + hr 6299 & k2 iii & wh & 8.653 & 0.020 & 4571 & 100 & @xcite + & & kh & 8.530 & 0.009 & & + hr 6703 & g8.5 iii & wh & 4.426 & 0.020 & 5011 & 35 & @xcite + hr 6705 & k5 iii & wh & 12.859 & 0.023 & 3934 & 42 & @xcite + & & kh & 12.569 & 0.010 & & + hr 7635 & m0@xmath26 iii & wh & 13.035 & 0.025 & 3867 & 50 & @xcite + & & kh & 12.869 & 0.010 & & + hr 7806 & k2.5 iii & wh & 9.993 & 0.022 & 4286 & 100 & @xcite + & & kh & 9.785 & 0.010 & & + hr 7886 & m6 iii & wh & 16.722 & 0.025 & 3243 & 79 & @xcite + hr 8317 & k0.5 iii & wh & 7.319 & 0.024 & 4658 & 19 & @xcite + hr 8694 & k0@xmath26 iii & wh & 6.691 & 0.021 & 4830 & 100 & @xcite + & & kh & 6.463 & 0.010 & & + hr 8905 & f8 iii & wh & 1.265 & 0.014 & 5942 & 42 & @xcite + & & kh & 0.797 & 0.007 & & + swvir & m7 iii & wh & 18.350 & 0.024 & 2966 & 36 & @xcite + & & kh & 18.033 & 0.008 & & + bkvir & m7@xmath26 iii & wh & 20.449 & 0.022 & 2944 & 34 & @xcite + & & kh & 20.730 & 0.006 & & + rxboo & m7.5 - 8 iii & wh & 18.262 & 0.024 & 2786 & 46 & @xcite + [ comp_table ] in order to separate cool giants from warmer giants and main sequence stars , warm stars with co @xmath38 3 are removed from the sample . of the 355 detected stars in our spectroscopy , 329 were categorized as co - absorbers , and we assigned effective temperatures to these stars using the above index@xmath26@xmath7 relation . prior to temperature calculation , we cross - correlated all co - absorbing stars with a co - template in order to remove radial velocities . errors in temperatures were estimated based on the noise in each spectrum and the intrinsic dispersion in our index@xmath26@xmath7 relation ( @xmath39 = 0.0050 ) . as noted by @xcite , errors determined in this way are strictly only lower limits . the spectral classification and analysis of the remaining early - type ( non - co - absorbing ) stars in our sample is less straight - forward @xcite , and we defer analysis of these stars to a later paper ( martins et al . 2007b , in prep ) . we made photometric observations using the imaging system naos / conica ( naco ) , consisting of the adaptive optics system naos @xcite and the nir camera conica @xcite at the 8.2-m ut4 ( yepun ) of the eso vlt . in april 2006 , we collected several ao - corrected images in each of @xmath2 and @xmath3 band with a pixel size of @xmath40 27 mas . the total exposure time was 64 s for each band , broken into thirty - two dithered images , in which every fourth image was separated from the previous exposure by 16 arcsec . we used our standard reduction pipeline to perform sky subtraction , bad pixel corrections , flat field corrections , and stacking of images to create final mosaics . in figure [ mos ] , we display a sub - section of the final @xmath3 band mosaic , containing all stars analyzed in this paper . photometry was extracted for only this region to help ensure psf constancy across the image . we used the crowded field photometry package starfinder @xcite to establish relative photometry and source detection . an empirical psf for the central core ( fwhm @xmath41 0.10 arcsec ) was extracted from each image using seven bright , isolated stars . the full radial extent of the extracted psf was r @xmath41 0.19 arcsec . to derive the photometric curve of growth needed to place these results on an absolute scale , we adopted the mtf - fitting technique of @xcite . this technique fits the power spectrum of the image using a combination of the source spatial distribution function determined by starfinder and a parameterized description of the modulation transfer functions ( mtf ) of the atmosphere , telescope , ao system , and science camera . the advantage of this technique is that it derives the psf encircled energy curve of growth , including the extended , seeing - limited halo , and thus provides the aperture correction required for absolute photometric calibration . in deriving the curve of growth from the data themselves , we avoid systematic calibration errors introduced from using a psf standard acquired at a different time and under different observing conditions than the target data . given the rapid variations in ao performance @xcite , this technique is crucial for deriving accurate absolute photometry . the mtf - fitting technique has so far only been applied to data obtained for the keck observatory lgsao system @xcite . to apply this technique to the vlt images , we used the @xcite software with appropriate input parameters for the vlt telescope pupil size and geometry , the camera platescale and pixel size , and the deformable mirror actuator spacing . the software uses the idl procedure mpfit to perform a levenberg - marquardt least - squares fit of the parameterized model mtf to the data . the best fits to the @xmath2 and @xmath3 band image power spectra are shown in figure [ pspec ] . the spatial frequency , @xmath42 , is normalized relative to the telescope cutoff frequency in the image plane ( d@xmath43 ) . band data are aliased , and we therefore , fit only the power spectrum for @xmath44 ( see 2.2 ) . the spatial frequency , @xmath42 , is normalized relative to the cutoff frequency in the image plane ( d@xmath43 ) . the strehl ratios implied by these fits are 10@xmath45% and 19@xmath46% , for @xmath2 and @xmath3 band , respectively . ] all images , even seeing - limited images , contain power up to the spatial frequency cutoff , which is @xmath47 for a circular pupil . to faithfully record all spatial information allowed by the telescope aperture , the pixel sampling frequency must therefore satisfy @xmath48 ( see @xcite for a detailed discussion ) . the @xmath3 band data are nyquist - sampled : @xmath49 = 1 . however , the h - band data are undersampled : @xmath50 . this aliasing causes power at frequencies @xmath51 to corrupt frequencies @xmath52 . the contamination of frequencies between 0.52 and @xmath50.7 is negligible , as the power aliased to these frequencies consists only of detector and background noise . this power is orders of magnitude less than the non - aliased power at these spatial frequencies , and thus , has very little effect on the total power spectrum . on the other hand , the aliased power for contaminated spatial frequencies @xmath53 is of the same order - of - magnitude as the non - aliased power , so the power spectrum at these frequencies is heavily influenced by the aliasing . we , therefore , chose only to fit the power spectrum for all spatial frequencies @xmath54 . as only detector and background noise contribute to the power spectrum for spatial frequencies @xmath55 , we are excluding very little information about the psf . the relatively minor discrepency in figure [ pspec ] between the model and data power spectra in @xmath3 band is likely due to uncertainties in the deformable mirror influence function , as an explicit description of the vlt cilas mirror influence function was not available . we , therefore , used the approximate influence function for the keck xinetics mirror from @xcite . the overall power spectra fits are satisfactory , and we do not expect that the photometric accuracy of our technique is significantly affected by this choice . the strehl ratios implied by the psfs reconstructed from the fits in figure [ pspec ] are 10% and 19% , for @xmath2 and @xmath3 band , respectively . to test the integrity of these fits , we divided the images into nine subimages and fit the power spectrum of each subimage . this process yielded fairly consistent results for all subimages , implying an error in the @xmath2 and @xmath3 band strehl ratios of 0.5% and 1.0% , respectively . this agreement between subimages suggests that psf variations due to variable distance from the guide star and variable exposure time in the final mosaic are only present at a low level . we determined the photometric zero point from observations of the near - ir standard star 9178 @xcite , obtained on the same night . the measured atmospheric extinction coefficients used to correct for the difference in airmass between our standard stars and science targets were 0.06 mag . airmass@xmath56 for @xmath2 and 0.07 mag . airmass@xmath56 for @xmath3 . we derived photometric errors and completeness using the standard technique of inserting and recovering artificial stars of specified magnitude . completeness is not uniform , so we calculated errors and completeness separately for each spectral field . to avoid artificially crowding our images , we inserted only fifty stars at a time to each spectral subimage and repeated this procedure five times for each magnitude bin . we computed the photometric error at each magnitude using a gaussian fit to the difference between the input and recovered magnitudes . figure [ comp_err ] shows the average photometric errors and completeness at each magnitude bin . also shown is the spectroscopic completeness , estimated by comparing the number of stars detected in the @xmath2 and @xmath57 band photometry to the number detected in the sinfoni data . the incompleteness in the sinfoni data at @xmath613 - 15 mag . is caused by crowding and confusion with nearby bright stars in the field of view . and @xmath3 band photometric errors and completeness based on the artificial star tests described in 2 . the spectroscopic completeness is estimated based on the number of co - stars for which we could extract spectra , compared to the total number of stars detected in the photometry . ] using the derived photometry and effective temperatures presented in the preceding section , we are able to place our gc sample on the hertzsprung - russell ( h - r ) diagram . to convert the measured photometric magnitudes to luminosity , we use models that assume the @xmath14 band magnitudes are measured in the johnson - cousins - glass system @xcite . therefore , we first transformed the measured @xmath3 magnitudes to the johnson - cousins - glass system , using the approximate transformations given in @xcite , assuming the naco and 2mass @xmath58 color systems are identical , as in @xcite . we next adopted empirical bolometric corrections of @xcite for m stars and theoretical corrections of girardi 2005 ( http://pleiadi / pd.astro.it ) for hotter stars . for the theoretical corrections , we assumed a solar metallicity and a surface gravity of @xmath59 , though the results are largely independent of these assumptions . for all temperatures in the galactic center sample , we found systematic differences @xmath60 mag . in bc@xmath61 for @xmath62=1.0 $ ] dex and @xmath63 dex , leading to systematic differences in the luminosity of @xmath64 . additionally , we assumed a distance to the galactic center of 8.0 kpc @xcite . there has been much recent debate concerning the precise distance to the galactic center . however , the choice of distance has only a small systematic effect on our derived luminosities ; a change in distance of @xmath65 kpc induces a change in luminosity of @xmath66 . finally , we corrected each star s luminosity individually for extinction using the photometric color , the derived @xmath7 , and the interstellar extinction law of @xcite , derived from nicmos observations of galactic center stars . all results are tabulated in table [ gc_table ] . lccccccccccccc 1 & 6.26 & 6.24 & 15.275 & 0.042 & 17.419 & 0.072 & 8.28 & 3.658 & 0.006 & 2.99 & 0.15 & 1.88 & 0.06 + 2 & 6.32 & 6.05 & 15.259 & 0.042 & 17.316 & 0.072 & 8.30 & 3.658 & 0.007 & 2.86 & 0.15 & 1.83 & 0.07 + 3 & 5.67 & 5.91 & 14.898 & 0.031 & 17.051 & 0.056 & 6.93 & 3.674 & 0.004 & 3.03 & 0.11 & 2.09 & 0.05 + 4 & 5.67 & 6.13 & 14.239 & 0.013 & 16.344 & 0.049 & 8.61 & 3.654 & 0.006 & 2.93 & 0.09 & 2.25 & 0.04 + 5 & 5.51 & 6.24 & 15.304 & 0.042 & 17.438 & 0.072 & 7.01 & 3.673 & 0.005 & 3.00 & 0.15 & 1.91 & 0.06 + 6 & 5.02 & 6.48 & 15.531 & 0.042 & 17.826 & 0.204 & 7.15 & 3.672 & 0.008 & 3.23 & 0.37 & 1.91 & 0.15 + 7 & 4.86 & 6.29 & 15.131 & 0.031 & 17.274 & 0.072 & 7.01 & 3.673 & 0.007 & 3.01 & 0.14 & 1.98 & 0.06 + 8 & 4.99 & 6.16 & 15.327 & 0.042 & 17.406 & 0.072 & 7.34 & 3.669 & 0.005 & 2.91 & 0.15 & 1.86 & 0.06 + 9 & 4.75 & 6.13 & 15.503 & 0.042 & 17.678 & 0.072 & 8.76 & 3.652 & 0.006 & 3.03 & 0.15 & 1.78 & 0.06 + 10 & 4.35 & 6.34 & 15.474 & 0.042 & 17.135 & 0.056 & 6.82 & 3.675 & 0.004 & 2.30 & 0.12 & 1.57 & 0.05 + 11 & 4.35 & 6.51 & 15.240 & 0.031 & 17.387 & 0.072 & 6.69 & 3.677 & 0.004 & 3.02 & 0.14 & 1.96 & 0.06 + 12 & 3.54 & 6.56 & 14.318 & 0.017 & 16.427 & 0.049 & 7.73 & 3.665 & 0.005 & 2.95 & 0.09 & 2.26 & 0.04 + 13 & 3.56 & 6.40 & 13.803 & 0.013 & 15.707 & 0.012 & 9.40 & 3.643 & 0.005 & 2.62 & 0.03 & 2.27 & 0.02 + 14 & 3.48 & 6.13 & 14.969 & 0.031 & 16.899 & 0.056 & 6.90 & 3.675 & 0.005 & 2.70 & 0.11 & 1.93 & 0.05 + 15 & 3.08 & 6.13 & 12.132 & 0.003 & 14.305 & 0.006 & 11.62 & 3.611 & 0.005 & 2.96 & 0.02 & 2.99 & 0.01 + 16 & 3.05 & 5.83 & 14.123 & 0.013 & 16.226 & 0.042 & 8.65 & 3.653 & 0.006 & 2.93 & 0.08 & 2.30 & 0.04 + 17 & 2.81 & 5.86 & 15.867 & 0.078 & 17.802 & 0.204 & 8.77 & 3.652 & 0.005 & 2.67 & 0.39 & 1.49 & 0.16 + 18 & 3.27 & 5.62 & 15.596 & 0.042 & 17.751 & 0.204 & 9.25 & 3.645 & 0.007 & 2.99 & 0.37 & 1.71 & 0.15 + 19 & 2.78 & 5.32 & 15.115 & 0.031 & 17.234 & 0.056 & 9.20 & 3.646 & 0.007 & 2.94 & 0.11 & 1.88 & 0.05 + 20 & 4.35 & 5.18 & 11.399 & 0.002 & 13.519 & 0.003 & 12.10 & 3.604 & 0.005 & 2.87 & 0.01 & 3.23 & 0.01 + 21 & 4.78 & 5.37 & 14.438 & 0.017 & 16.735 & 0.049 & 9.71 & 3.639 & 0.004 & 3.19 & 0.09 & 2.24 & 0.04 + 22 & 5.13 & 5.51 & 15.436 & 0.042 & 17.641 & 0.072 & 8.55 & 3.655 & 0.004 & 3.08 & 0.15 & 1.84 & 0.06 + 23 & 4.97 & 5.10 & 15.341 & 0.042 & 17.335 & 0.072 & 10.09 & 3.634 & 0.005 & 2.74 & 0.15 & 1.68 & 0.06 + 24 & 5.53 & 4.54 & 14.770 & 0.031 & 17.054 & 0.056 & 6.77 & 3.676 & 0.006 & 3.22 & 0.11 & 2.22 & 0.05 + 25 & 5.10 & 3.46 & 14.078 & 0.013 & 15.666 & 0.012 & 8.31 & 3.658 & 0.005 & 2.17 & 0.03 & 2.02 & 0.02 + 26 & 5.16 & 3.67 & 15.087 & 0.031 & 17.025 & 0.056 & 9.92 & 3.636 & 0.011 & 2.66 & 0.12 & 1.76 & 0.06 + 27 & 5.45 & 3.91 & 15.316 & 0.042 & 17.369 & 0.072 & 10.07 & 3.634 & 0.007 & 2.82 & 0.15 & 1.73 & 0.06 + 28 & 5.08 & 4.18 & 13.152 & 0.005 & 15.121 & 0.006 & 10.32 & 3.631 & 0.006 & 2.69 & 0.02 & 2.53 & 0.02 + 29 & 5.13 & 4.51 & 13.238 & 0.005 & 15.345 & 0.012 & 10.91 & 3.622 & 0.005 & 2.88 & 0.02 & 2.55 & 0.02 + 30 & 5.05 & 4.70 & 13.980 & 0.013 & 16.298 & 0.049 & 12.37 & 3.600 & 0.007 & 3.16 & 0.09 & 2.30 & 0.04 + 31 & 4.56 & 4.56 & 13.382 & 0.009 & 15.542 & 0.012 & 8.63 & 3.653 & 0.004 & 3.01 & 0.03 & 2.63 & 0.02 + 32 & 4.29 & 4.29 & 15.536 & 0.042 & 17.606 & 0.072 & 7.99 & 3.662 & 0.004 & 2.89 & 0.15 & 1.74 & 0.06 + 33 & 4.59 & 4.13 & 16.137 & 0.078 & 18.223 & 0.204 & 5.27 & 3.692 & 0.003 & 2.95 & 0.39 & 1.61 & 0.16 + 34 & 4.40 & 3.48 & 13.845 & 0.013 & 15.739 & 0.012 & 8.65 & 3.653 & 0.006 & 2.62 & 0.03 & 2.28 & 0.02 + 35 & 4.24 & 3.70 & 13.676 & 0.009 & 15.619 & 0.012 & 7.61 & 3.666 & 0.005 & 2.71 & 0.03 & 2.42 & 0.02 + 36 & 4.32 & 3.02 & 13.194 & 0.005 & 15.295 & 0.012 & 9.46 & 3.642 & 0.005 & 2.91 & 0.02 & 2.63 & 0.02 + 37 & 3.00 & 3.32 & 12.562 & 0.003 & 14.834 & 0.006 & 3.70 & 3.707 & 0.004 & 3.24 & 0.01 & 3.20 & 0.01 + 38 & 3.11 & 3.83 & 15.976 & 0.078 & 18.227 & 0.204 & 9.31 & 3.645 & 0.008 & 3.13 & 0.39 & 1.61 & 0.16 + 39 & 3.62 & 3.56 & 15.106 & 0.031 & 17.234 & 0.056 & 5.89 & 3.686 & 0.004 & 3.00 & 0.11 & 2.03 & 0.05 + 40 & 4.02 & 3.97 & 15.532 & 0.042 & 17.603 & 0.072 & 7.97 & 3.662 & 0.006 & 2.89 & 0.15 & 1.74 & 0.06 + 41 & 3.62 & 4.05 & 15.160 & 0.031 & 17.431 & 0.072 & 9.04 & 3.648 & 0.005 & 3.17 & 0.14 & 1.96 & 0.06 + 42 & 2.92 & 4.13 & 12.738 & 0.003 & 15.117 & 0.006 & 12.73 & 3.594 & 0.005 & 3.23 & 0.02 & 2.82 & 0.02 + 43 & -6.97 & 10.15 & 15.030 & 0.026 & 16.988 & 0.058 & 8.98 & 3.649 & 0.008 & 2.71 & 0.11 & 1.83 & 0.05 + 44 & -7.24 & 10.31 & 14.751 & 0.026 & 16.765 & 0.058 & 7.28 & 3.670 & 0.006 & 2.82 & 0.11 & 2.05 & 0.05 + 45 & -7.10 & 9.85 & 15.112 & 0.026 & 17.048 & 0.058 & 5.96 & 3.685 & 0.006 & 2.72 & 0.11 & 1.91 & 0.05 + 46 & -7.32 & 9.88 & 15.075 & 0.026 & 17.022 & 0.058 & 6.46 & 3.680 & 0.006 & 2.73 & 0.11 & 1.91 & 0.05 + 47 & -7.94 & 9.88 & 15.494 & 0.040 & 17.295 & 0.060 & 4.70 & 3.698 & 0.006 & 2.53 & 0.13 & 1.72 & 0.06 + 48 & -8.99 & 10.12 & 13.124 & 0.005 & 15.144 & 0.008 & 10.30 & 3.631 & 0.008 & 2.77 & 0.02 & 2.57 & 0.02 + 49 & -8.80 & 9.67 & 15.374 & 0.040 & 17.408 & 0.060 & 7.61 & 3.666 & 0.009 & 2.84 & 0.13 & 1.80 & 0.06 + 50 & -9.29 & 9.37 & 15.199 & 0.026 & 17.210 & 0.058 & 8.72 & 3.652 & 0.007 & 2.79 & 0.11 & 1.81 & 0.05 + 51 & -8.69 & 9.21 & 15.156 & 0.026 & 17.093 & 0.058 & 6.25 & 3.682 & 0.006 & 2.71 & 0.11 & 1.88 & 0.05 + 52 & -7.45 & 9.37 & 11.937 & 0.003 & 13.686 & 0.002 & 5.88 & 3.686 & 0.004 & 2.44 & 0.01 & 3.07 & 0.01 + 53 & -5.59 & 9.21 & 10.319 & 0.002 & 12.222 & 0.002 & 11.08 & 3.619 & 0.008 & 2.58 & 0.02 & 3.59 & 0.02 + 54 & -6.75 & 8.80 & 15.448 & 0.040 & 17.392 & 0.060 & 6.35 & 3.681 & 0.005 & 2.72 & 0.13 & 1.76 & 0.06 + 55 & -6.75 & 8.56 & 15.948 & 0.090 & 18.090 & 0.245 & 6.47 & 3.679 & 0.008 & 3.02 & 0.46 & 1.68 & 0.19 + 56 & -6.18 & 8.18 & 15.273 & 0.040 & 17.416 & 0.060 & 8.85 & 3.651 & 0.009 & 2.98 & 0.13 & 1.85 & 0.06 + 57 & -6.29 & 8.07 & 15.706 & 0.040 & 17.964 & 0.245 & 8.91 & 3.650 & 0.007 & 3.15 & 0.44 & 1.74 & 0.18 + 58 & -6.10 & 7.86 & 15.506 & 0.040 & 17.688 & 0.060 & 4.64 & 3.699 & 0.005 & 3.10 & 0.13 & 1.94 & 0.06 + 59 & -5.97 & 7.29 & 15.037 & 0.026 & 17.188 & 0.058 & 4.32 & 3.702 & 0.008 & 3.05 & 0.11 & 2.12 & 0.05 + 60 & -6.16 & 6.43 & 13.182 & 0.005 & 15.300 & 0.009 & 7.56 & 3.667 & 0.008 & 2.97 & 0.02 & 2.73 & 0.02 + 61 & -6.51 & 6.64 & 15.628 & 0.040 & 17.772 & 0.245 & 9.21 & 3.646 & 0.008 & 2.98 & 0.44 & 1.69 & 0.18 + 62 & -6.80 & 6.86 & 14.604 & 0.022 & 16.766 & 0.058 & 7.46 & 3.668 & 0.006 & 3.03 & 0.11 & 2.19 & 0.05 + 63 & -7.32 & 7.78 & 13.190 & 0.005 & 15.195 & 0.008 & 8.64 & 3.653 & 0.006 & 2.78 & 0.02 & 2.61 & 0.02 + 64 & -7.45 & 8.29 & 15.413 & 0.040 & 17.374 & 0.060 & 6.44 & 3.680 & 0.005 & 2.75 & 0.13 & 1.79 & 0.06 + 65 & -7.64 & 8.56 & 15.467 & 0.040 & 17.477 & 0.060 & 6.86 & 3.675 & 0.006 & 2.82 & 0.13 & 1.78 & 0.06 + 66 & -7.48 & 8.99 & 12.976 & 0.005 & 14.980 & 0.008 & 12.14 & 3.603 & 0.009 & 2.70 & 0.03 & 2.53 & 0.02 + 67 & -8.23 & 8.88 & 15.498 & 0.040 & 17.511 & 0.060 & 5.75 & 3.687 & 0.008 & 2.83 & 0.13 & 1.81 & 0.06 + 68 & -8.91 & 8.64 & 15.444 & 0.040 & 17.651 & 0.060 & 9.59 & 3.641 & 0.008 & 3.06 & 0.13 & 1.79 & 0.06 + 69 & -8.59 & 8.18 & 12.292 & 0.004 & 14.300 & 0.005 & 8.80 & 3.651 & 0.005 & 2.78 & 0.01 & 2.97 & 0.02 + 70 & -8.13 & 8.13 & 14.233 & 0.010 & 16.282 & 0.047 & 8.98 & 3.649 & 0.009 & 2.84 & 0.09 & 2.21 & 0.04 + 71 & -8.18 & 8.32 & 15.497 & 0.040 & 17.501 & 0.060 & 7.83 & 3.664 & 0.007 & 2.79 & 0.13 & 1.72 & 0.06 + 72 & -7.94 & 8.45 & 15.375 & 0.040 & 17.420 & 0.060 & 9.34 & 3.644 & 0.007 & 2.83 & 0.13 & 1.73 & 0.06 + 73 & -7.75 & 8.26 & 15.616 & 0.040 & 17.814 & 0.245 & 6.29 & 3.682 & 0.006 & 3.10 & 0.44 & 1.85 & 0.18 + 74 & -7.97 & 7.51 & 12.569 & 0.004 & 14.867 & 0.008 & 10.02 & 3.635 & 0.009 & 3.19 & 0.02 & 2.97 & 0.02 + 75 & -7.61 & 7.21 & 14.774 & 0.026 & 16.954 & 0.058 & 5.67 & 3.688 & 0.007 & 3.08 & 0.11 & 2.20 & 0.05 + 76 & -8.45 & 6.99 & 10.843 & 0.002 & 13.278 & 0.002 & 12.81 & 3.593 & 0.013 & 3.32 & 0.03 & 3.60 & 0.03 + 77 & -9.21 & 6.53 & 12.543 & 0.004 & 14.852 & 0.008 & 10.86 & 3.623 & 0.008 & 3.18 & 0.02 & 2.95 & 0.02 + 78 & -8.64 & 7.34 & 15.402 & 0.040 & 17.464 & 0.060 & 8.56 & 3.654 & 0.007 & 2.87 & 0.13 & 1.76 & 0.06 + 79 & 2.94 & 12.93 & 15.314 & 0.049 & 16.938 & 0.055 & 5.66 & 3.688 & 0.004 & 2.26 & 0.13 & 1.65 & 0.06 + 80 & 2.67 & 12.74 & 14.953 & 0.027 & 16.828 & 0.055 & 7.95 & 3.662 & 0.005 & 2.60 & 0.11 & 1.86 & 0.05 + 81 & 2.54 & 12.45 & 15.140 & 0.027 & 16.947 & 0.055 & 8.00 & 3.661 & 0.006 & 2.50 & 0.11 & 1.74 & 0.05 + 82 & 2.59 & 12.04 & 15.189 & 0.027 & 16.632 & 0.046 & 4.90 & 3.696 & 0.005 & 2.00 & 0.09 & 1.62 & 0.04 + 83 & 3.21 & 12.69 & 16.003 & 0.075 & 17.874 & 0.210 & 7.56 & 3.667 & 0.007 & 2.60 & 0.40 & 1.45 & 0.16 + 84 & 3.00 & 12.64 & 15.868 & 0.075 & 17.752 & 0.210 & 6.33 & 3.681 & 0.006 & 2.64 & 0.40 & 1.56 & 0.16 + 85 & 5.62 & 11.88 & 15.501 & 0.049 & 17.546 & 0.080 & 9.55 & 3.641 & 0.006 & 2.82 & 0.17 & 1.67 & 0.07 + 86 & 5.83 & 11.42 & 11.834 & 0.003 & 14.160 & 0.003 & 13.30 & 3.585 & 0.007 & 3.14 & 0.02 & 3.11 & 0.02 + 87 & 5.94 & 11.12 & 13.811 & 0.011 & 16.328 & 0.046 & 10.99 & 3.621 & 0.008 & 3.49 & 0.09 & 2.56 & 0.04 + 88 & 5.48 & 11.45 & 13.479 & 0.008 & 16.289 & 0.046 & 11.75 & 3.609 & 0.007 & 3.90 & 0.08 & 2.83 & 0.04 + 89 & 5.18 & 11.53 & 15.624 & 0.049 & 17.880 & 0.210 & 6.90 & 3.675 & 0.005 & 3.18 & 0.38 & 1.86 & 0.16 + 90 & 4.83 & 11.74 & 15.531 & 0.049 & 17.641 & 0.080 & 7.38 & 3.669 & 0.005 & 2.96 & 0.17 & 1.79 & 0.07 + 91 & 4.78 & 11.53 & 15.510 & 0.049 & 17.643 & 0.080 & 4.91 & 3.696 & 0.006 & 3.02 & 0.17 & 1.90 & 0.07 + 92 & 4.35 & 11.69 & 14.863 & 0.027 & 16.784 & 0.055 & 8.03 & 3.661 & 0.005 & 2.67 & 0.11 & 1.92 & 0.05 + 93 & 3.70 & 11.42 & 12.593 & 0.003 & 14.495 & 0.006 & 5.95 & 3.685 & 0.003 & 2.67 & 0.01 & 2.90 & 0.01 + 94 & 3.43 & 11.39 & 13.861 & 0.011 & 15.757 & 0.042 & 8.13 & 3.660 & 0.005 & 2.63 & 0.08 & 2.30 & 0.03 + 95 & 2.38 & 10.96 & 12.391 & 0.003 & 14.350 & 0.006 & 11.31 & 3.616 & 0.004 & 2.66 & 0.01 & 2.78 & 0.01 + 96 & 2.59 & 10.77 & 13.164 & 0.005 & 14.799 & 0.006 & 6.53 & 3.679 & 0.004 & 2.26 & 0.01 & 2.49 & 0.01 + 97 & 3.16 & 10.69 & 14.694 & 0.019 & 16.701 & 0.046 & 5.54 & 3.690 & 0.006 & 2.83 & 0.09 & 2.13 & 0.04 + 98 & 3.73 & 10.53 & 12.842 & 0.005 & 14.355 & 0.006 & 8.50 & 3.655 & 0.005 & 2.06 & 0.01 & 2.47 & 0.01 + 99 & 4.51 & 10.85 & 11.905 & 0.003 & 13.794 & 0.003 & 5.57 & 3.689 & 0.003 & 2.65 & 0.01 & 3.18 & 0.01 + 100 & 5.10 & 10.66 & 13.593 & 0.008 & 15.708 & 0.008 & 9.17 & 3.646 & 0.007 & 2.93 & 0.02 & 2.49 & 0.02 + 101 & 5.67 & 10.80 & 15.151 & 0.027 & 17.216 & 0.055 & 8.89 & 3.650 & 0.005 & 2.86 & 0.11 & 1.85 & 0.05 + 102 & 5.40 & 9.96 & 15.114 & 0.027 & 17.121 & 0.055 & 6.77 & 3.676 & 0.006 & 2.81 & 0.11 & 1.92 & 0.05 + 103 & 5.16 & 9.88 & 15.134 & 0.027 & 17.133 & 0.055 & 8.75 & 3.652 & 0.007 & 2.77 & 0.11 & 1.83 & 0.05 + 104 & 5.91 & 9.45 & 15.730 & 0.049 & 17.717 & 0.080 & 6.20 & 3.682 & 0.009 & 2.79 & 0.17 & 1.68 & 0.07 + 105 & 5.08 & 10.18 & 13.307 & 0.008 & 15.413 & 0.008 & 11.02 & 3.620 & 0.007 & 2.88 & 0.02 & 2.51 & 0.02 + 106 & 3.59 & 9.45 & 13.281 & 0.008 & 15.303 & 0.008 & 11.49 & 3.613 & 0.006 & 2.74 & 0.02 & 2.45 & 0.02 + 107 & 3.29 & 9.40 & 15.212 & 0.027 & 17.153 & 0.055 & 7.64 & 3.666 & 0.009 & 2.70 & 0.11 & 1.81 & 0.05 + 108 & 3.13 & 9.53 & 15.129 & 0.027 & 16.961 & 0.055 & 6.68 & 3.677 & 0.005 & 2.55 & 0.11 & 1.81 & 0.05 + 109 & 3.08 & 9.83 & 13.819 & 0.011 & 15.715 & 0.008 & 9.31 & 3.644 & 0.006 & 2.61 & 0.03 & 2.27 & 0.02 + 110 & 3.46 & 10.12 & 13.302 & 0.008 & 15.255 & 0.008 & 14.05 & 3.572 & 0.007 & 2.56 & 0.03 & 2.27 & 0.02 + 111 & 2.94 & 10.07 & 14.905 & 0.027 & 16.849 & 0.055 & 8.69 & 3.653 & 0.005 & 2.69 & 0.11 & 1.89 & 0.05 + 112 & 2.51 & 10.23 & 14.381 & 0.019 & 15.823 & 0.042 & 5.62 & 3.689 & 0.005 & 1.99 & 0.08 & 1.92 & 0.04 + 113 & 2.67 & 9.53 & 15.518 & 0.049 & 17.613 & 0.080 & 7.98 & 3.662 & 0.007 & 2.92 & 0.17 & 1.76 & 0.07 + 114 & -4.86 & 12.72 & 15.647 & 0.042 & 17.899 & 0.224 & 7.68 & 3.665 & 0.007 & 3.16 & 0.41 & 1.82 & 0.16 + 115 & -5.40 & 12.39 & 14.498 & 0.019 & 16.366 & 0.048 & 7.14 & 3.672 & 0.005 & 2.60 & 0.09 & 2.07 & 0.04 + 116 & -6.18 & 12.69 & 15.679 & 0.042 & 17.406 & 0.072 & 6.76 & 3.676 & 0.005 & 2.40 & 0.15 & 1.53 & 0.06 + 117 & -6.59 & 12.77 & 16.205 & 0.068 & 18.072 & 0.224 & 5.31 & 3.692 & 0.005 & 2.62 & 0.42 & 1.45 & 0.17 + 118 & -6.78 & 12.66 & 16.404 & 0.119 & 18.180 & 0.224 & 6.30 & 3.681 & 0.008 & 2.48 & 0.45 & 1.28 & 0.19 + 119 & -7.48 & 12.26 & 15.019 & 0.033 & 16.737 & 0.048 & 7.69 & 3.665 & 0.005 & 2.37 & 0.10 & 1.75 & 0.05 + 120 & -7.70 & 12.45 & 15.224 & 0.033 & 16.988 & 0.053 & 6.94 & 3.674 & 0.005 & 2.45 & 0.11 & 1.73 & 0.05 + 121 & -7.07 & 11.93 & 15.166 & 0.033 & 16.674 & 0.048 & 5.40 & 3.691 & 0.004 & 2.09 & 0.10 & 1.65 & 0.04 + 122 & -6.86 & 11.85 & 15.556 & 0.042 & 17.281 & 0.072 & 6.11 & 3.683 & 0.004 & 2.40 & 0.15 & 1.60 & 0.06 + 123 & -6.56 & 11.99 & 15.669 & 0.042 & 17.310 & 0.072 & 5.47 & 3.690 & 0.004 & 2.29 & 0.15 & 1.53 & 0.06 + 124 & -6.21 & 12.10 & 15.638 & 0.042 & 17.358 & 0.072 & 6.14 & 3.683 & 0.006 & 2.40 & 0.15 & 1.56 & 0.06 + 125 & -5.16 & 12.20 & 17.135 & 0.131 & 19.005 & 0.252 & 6.87 & 3.675 & 0.005 & 2.61 & 0.51 & 1.03 & 0.21 + 126 & -4.89 & 12.12 & 15.325 & 0.042 & 17.257 & 0.072 & 5.98 & 3.685 & 0.005 & 2.71 & 0.15 & 1.82 & 0.06 + 127 & -5.18 & 11.85 & 15.313 & 0.042 & 17.172 & 0.053 & 5.02 & 3.695 & 0.004 & 2.62 & 0.12 & 1.82 & 0.05 + 128 & -5.83 & 11.56 & 14.756 & 0.033 & 16.508 & 0.048 & 7.64 & 3.666 & 0.004 & 2.42 & 0.10 & 1.88 & 0.05 + 129 & -6.56 & 11.45 & 14.853 & 0.033 & 16.636 & 0.048 & 8.90 & 3.650 & 0.005 & 2.45 & 0.10 & 1.80 & 0.05 + 130 & -7.21 & 11.53 & 13.517 & 0.008 & 15.305 & 0.010 & 9.99 & 3.635 & 0.005 & 2.43 & 0.02 & 2.29 & 0.02 + 131 & -7.51 & 11.74 & 14.319 & 0.019 & 16.096 & 0.045 & 7.41 & 3.669 & 0.004 & 2.46 & 0.09 & 2.08 & 0.04 + 132 & -7.75 & 11.58 & 15.081 & 0.033 & 16.952 & 0.053 & 8.21 & 3.659 & 0.007 & 2.59 & 0.11 & 1.80 & 0.05 + 133 & -8.23 & 11.45 & 15.838 & 0.068 & 17.605 & 0.072 & 7.01 & 3.673 & 0.004 & 2.46 & 0.18 & 1.48 & 0.08 + 134 & -8.29 & 11.12 & 15.889 & 0.068 & 17.677 & 0.072 & 6.84 & 3.675 & 0.004 & 2.49 & 0.18 & 1.48 & 0.08 + 135 & -7.80 & 11.18 & 14.080 & 0.011 & 15.992 & 0.045 & 10.11 & 3.634 & 0.005 & 2.62 & 0.08 & 2.14 & 0.04 + 136 & -7.70 & 10.83 & 15.377 & 0.042 & 17.235 & 0.053 & 6.32 & 3.681 & 0.005 & 2.60 & 0.12 & 1.74 & 0.05 + 137 & -7.32 & 11.29 & 14.858 & 0.033 & 16.667 & 0.048 & 7.69 & 3.665 & 0.004 & 2.51 & 0.10 & 1.87 & 0.05 + 138 & -7.07 & 10.85 & 15.836 & 0.068 & 17.186 & 0.053 & 5.97 & 3.685 & 0.006 & 1.85 & 0.15 & 1.27 & 0.07 + 139 & -6.83 & 10.88 & 14.494 & 0.019 & 16.366 & 0.048 & 10.31 & 3.631 & 0.006 & 2.55 & 0.09 & 1.94 & 0.04 + 140 & -7.02 & 10.53 & 14.002 & 0.011 & 16.085 & 0.045 & 9.61 & 3.641 & 0.005 & 2.88 & 0.08 & 2.29 & 0.04 + 141 & -7.26 & 10.50 & 14.588 & 0.019 & 16.339 & 0.048 & 7.36 & 3.669 & 0.005 & 2.43 & 0.09 & 1.96 & 0.04 + 142 & -5.99 & 11.18 & 11.861 & 0.003 & 13.400 & 0.002 & 10.47 & 3.628 & 0.005 & 2.06 & 0.01 & 2.79 & 0.01 + 143 & -5.89 & 10.66 & 13.946 & 0.011 & 15.939 & 0.045 & 8.36 & 3.657 & 0.005 & 2.77 & 0.08 & 2.31 & 0.04 + 144 & -5.45 & 10.31 & 15.461 & 0.042 & 17.386 & 0.072 & 5.11 & 3.694 & 0.004 & 2.71 & 0.15 & 1.79 & 0.06 + 145 & -5.26 & 10.18 & 14.684 & 0.019 & 16.372 & 0.048 & 7.67 & 3.666 & 0.004 & 2.33 & 0.09 & 1.87 & 0.04 + 146 & -4.86 & 9.85 & 12.872 & 0.004 & 14.806 & 0.007 & 10.37 & 3.630 & 0.005 & 2.64 & 0.02 & 2.62 & 0.01 + 147 & 4.94 & 16.85 & 15.552 & 0.033 & 17.771 & 0.193 & 8.39 & 3.657 & 0.007 & 3.10 & 0.35 & 1.80 & 0.14 + 148 & 4.18 & 16.82 & 14.313 & 0.014 & 16.510 & 0.043 & 9.01 & 3.649 & 0.006 & 3.06 & 0.08 & 2.26 & 0.04 + 149 & 3.51 & 16.69 & 15.608 & 0.033 & 17.420 & 0.058 & 7.22 & 3.671 & 0.005 & 2.52 & 0.12 & 1.59 & 0.05 + 150 & 5.94 & 16.58 & 15.878 & 0.073 & 18.001 & 0.193 & 5.31 & 3.692 & 0.005 & 3.00 & 0.37 & 1.74 & 0.15 + 151 & 5.83 & 16.28 & 15.617 & 0.033 & 17.657 & 0.058 & 7.14 & 3.672 & 0.006 & 2.86 & 0.12 & 1.72 & 0.05 + 152 & 5.59 & 16.09 & 15.837 & 0.073 & 17.933 & 0.193 & 6.11 & 3.683 & 0.005 & 2.95 & 0.37 & 1.71 & 0.15 + 153 & 4.97 & 16.42 & 12.691 & 0.003 & 15.117 & 0.005 & 14.50 & 3.564 & 0.006 & 3.24 & 0.02 & 2.76 & 0.02 + 154 & 4.32 & 16.28 & 16.030 & 0.073 & 18.088 & 0.193 & 6.94 & 3.674 & 0.008 & 2.89 & 0.37 & 1.58 & 0.15 + 155 & 2.73 & 16.60 & 15.914 & 0.073 & 17.716 & 0.058 & 5.83 & 3.686 & 0.008 & 2.52 & 0.17 & 1.51 & 0.08 + 156 & 2.84 & 16.25 & 13.465 & 0.007 & 15.255 & 0.007 & 8.70 & 3.653 & 0.006 & 2.46 & 0.02 & 2.37 & 0.02 + 157 & 3.00 & 15.98 & 15.508 & 0.033 & 17.399 & 0.058 & 4.92 & 3.696 & 0.006 & 2.66 & 0.12 & 1.76 & 0.05 + 158 & 3.35 & 15.90 & 15.638 & 0.033 & 17.629 & 0.058 & 7.04 & 3.673 & 0.005 & 2.79 & 0.12 & 1.69 & 0.05 + 159 & 3.73 & 15.96 & 15.501 & 0.033 & 17.434 & 0.058 & 6.61 & 3.678 & 0.006 & 2.71 & 0.12 & 1.73 & 0.05 + 160 & 4.67 & 15.93 & 15.194 & 0.026 & 17.162 & 0.046 & 6.63 & 3.678 & 0.004 & 2.76 & 0.09 & 1.87 & 0.04 + 161 & 5.32 & 15.77 & 15.439 & 0.033 & 17.512 & 0.058 & 7.90 & 3.663 & 0.005 & 2.89 & 0.12 & 1.78 & 0.05 + 162 & 5.43 & 15.88 & 15.235 & 0.026 & 17.402 & 0.058 & 9.64 & 3.640 & 0.005 & 3.00 & 0.11 & 1.85 & 0.05 + 163 & 5.78 & 15.80 & 15.236 & 0.026 & 17.399 & 0.058 & 8.57 & 3.654 & 0.007 & 3.01 & 0.11 & 1.89 & 0.05 + 164 & 5.83 & 15.52 & 15.698 & 0.033 & 18.004 & 0.193 & 8.79 & 3.651 & 0.006 & 3.22 & 0.35 & 1.78 & 0.14 + 165 & 5.99 & 15.58 & 16.165 & 0.073 & 18.239 & 0.193 & 6.62 & 3.678 & 0.008 & 2.91 & 0.37 & 1.54 & 0.15 + 166 & 5.08 & 15.50 & 16.363 & 0.076 & 18.561 & 0.216 & 8.51 & 3.655 & 0.006 & 3.07 & 0.41 & 1.46 & 0.17 + 167 & 4.99 & 15.15 & 15.308 & 0.033 & 17.408 & 0.058 & 10.85 & 3.623 & 0.007 & 2.88 & 0.12 & 1.72 & 0.05 + 168 & 3.81 & 15.31 & 12.615 & 0.003 & 14.638 & 0.005 & 13.49 & 3.581 & 0.007 & 2.68 & 0.02 & 2.61 & 0.02 + 169 & 3.38 & 15.31 & 15.383 & 0.033 & 17.319 & 0.058 & 8.35 & 3.657 & 0.006 & 2.68 & 0.12 & 1.71 & 0.05 + 170 & 3.16 & 15.15 & 14.040 & 0.008 & 16.061 & 0.043 & 9.34 & 3.644 & 0.006 & 2.79 & 0.08 & 2.25 & 0.04 + 171 & 2.92 & 15.66 & 15.052 & 0.026 & 16.976 & 0.046 & 8.92 & 3.650 & 0.008 & 2.66 & 0.09 & 1.81 & 0.05 + 172 & 2.78 & 15.34 & 14.778 & 0.026 & 16.934 & 0.046 & 9.07 & 3.648 & 0.005 & 3.00 & 0.09 & 2.05 & 0.04 + 173 & 2.54 & 14.80 & 13.166 & 0.004 & 14.934 & 0.005 & 5.00 & 3.695 & 0.005 & 2.48 & 0.01 & 2.62 & 0.01 + 174 & 3.21 & 14.42 & 15.317 & 0.033 & 17.176 & 0.046 & 6.99 & 3.674 & 0.006 & 2.59 & 0.10 & 1.74 & 0.05 + 175 & 3.24 & 14.66 & 15.590 & 0.033 & 17.545 & 0.058 & 7.46 & 3.668 & 0.006 & 2.73 & 0.12 & 1.67 & 0.05 + 176 & 3.62 & 14.80 & 15.449 & 0.033 & 17.362 & 0.058 & 9.52 & 3.642 & 0.005 & 2.63 & 0.12 & 1.61 & 0.05 + 177 & 4.10 & 14.88 & 15.905 & 0.073 & 18.090 & 0.193 & 7.66 & 3.666 & 0.005 & 3.06 & 0.37 & 1.67 & 0.15 + 178 & 6.10 & 15.07 & 15.197 & 0.026 & 17.898 & 0.193 & 10.05 & 3.634 & 0.006 & 3.78 & 0.35 & 2.16 & 0.14 + 179 & 5.75 & 14.50 & 10.803 & 0.002 & 13.327 & 0.002 & 16.04 & 3.536 & 0.007 & 3.31 & 0.02 & 3.49 & 0.02 + 180 & 4.56 & 14.23 & 12.404 & 0.003 & 15.485 & 0.007 & 14.27 & 3.568 & 0.007 & 4.22 & 0.02 & 3.28 & 0.02 + 181 & 4.00 & 14.28 & 15.637 & 0.033 & 17.484 & 0.058 & 4.80 & 3.697 & 0.003 & 2.60 & 0.12 & 1.69 & 0.05 + 182 & 3.89 & 13.77 & 12.332 & 0.003 & 14.286 & 0.005 & 12.22 & 3.602 & 0.007 & 2.62 & 0.02 & 2.75 & 0.02 + 183 & 3.67 & 14.09 & 15.154 & 0.026 & 17.022 & 0.046 & 9.60 & 3.641 & 0.005 & 2.56 & 0.09 & 1.70 & 0.04 + 184 & 3.21 & 13.47 & 13.180 & 0.004 & 15.232 & 0.005 & 11.63 & 3.611 & 0.007 & 2.78 & 0.02 & 2.50 & 0.02 + 185 & 3.00 & 14.09 & 15.071 & 0.026 & 17.010 & 0.046 & 7.33 & 3.670 & 0.005 & 2.70 & 0.09 & 1.88 & 0.04 + 186 & 2.46 & 13.45 & 15.667 & 0.033 & 17.564 & 0.058 & 7.33 & 3.670 & 0.005 & 2.64 & 0.12 & 1.61 & 0.05 + 187 & 4.75 & 13.82 & 15.838 & 0.073 & 17.896 & 0.193 & 7.32 & 3.670 & 0.004 & 2.88 & 0.37 & 1.64 & 0.15 + 188 & 5.35 & 13.53 & 15.493 & 0.033 & 18.000 & 0.193 & 9.48 & 3.642 & 0.009 & 3.51 & 0.35 & 1.95 & 0.14 + 189 & 5.21 & 13.01 & 13.400 & 0.007 & 15.575 & 0.007 & 11.59 & 3.612 & 0.006 & 2.97 & 0.02 & 2.49 & 0.02 + 190 & 4.86 & 13.04 & 15.419 & 0.033 & 17.675 & 0.058 & 4.92 & 3.696 & 0.005 & 3.20 & 0.12 & 2.01 & 0.05 + 191 & 4.62 & 13.28 & 15.571 & 0.033 & 17.613 & 0.058 & 7.31 & 3.670 & 0.005 & 2.86 & 0.12 & 1.74 & 0.05 + 192 & 4.29 & 13.42 & 15.229 & 0.026 & 17.173 & 0.046 & 7.96 & 3.662 & 0.007 & 2.70 & 0.09 & 1.79 & 0.04 + 193 & 4.13 & 13.18 & 15.177 & 0.026 & 17.135 & 0.046 & 5.54 & 3.690 & 0.006 & 2.76 & 0.09 & 1.91 & 0.04 + 194 & 3.81 & 12.96 & 14.855 & 0.026 & 17.095 & 0.046 & 7.21 & 3.671 & 0.005 & 3.15 & 0.09 & 2.14 & 0.04 + 195 & 3.75 & 12.82 & 14.958 & 0.026 & 16.816 & 0.046 & 6.62 & 3.678 & 0.006 & 2.59 & 0.09 & 1.90 & 0.04 + 196 & 3.46 & 12.18 & 14.986 & 0.026 & 16.925 & 0.046 & 7.58 & 3.667 & 0.007 & 2.70 & 0.09 & 1.90 & 0.04 + 197 & 4.62 & 12.31 & 14.028 & 0.008 & 16.714 & 0.043 & 11.82 & 3.608 & 0.007 & 3.72 & 0.08 & 2.53 & 0.04 + 198 & 4.67 & 11.96 & 15.444 & 0.033 & 17.517 & 0.058 & 8.54 & 3.655 & 0.011 & 2.88 & 0.12 & 1.76 & 0.06 + 199 & 5.48 & 12.34 & 14.066 & 0.008 & 16.048 & 0.043 & 9.55 & 3.641 & 0.009 & 2.73 & 0.08 & 2.21 & 0.04 + 200 & 5.75 & 11.96 & 15.084 & 0.026 & 18.396 & 0.216 & 10.30 & 3.631 & 0.010 & 4.68 & 0.39 & 2.55 & 0.16 + 201 & 5.83 & 12.50 & 15.007 & 0.026 & 16.824 & 0.046 & 7.83 & 3.664 & 0.007 & 2.52 & 0.09 & 1.81 & 0.04 + 202 & 5.56 & 12.80 & 15.206 & 0.026 & 17.276 & 0.058 & 7.35 & 3.669 & 0.009 & 2.90 & 0.11 & 1.90 & 0.05 + 203 & 5.32 & 12.64 & 15.004 & 0.026 & 16.793 & 0.046 & 7.93 & 3.662 & 0.006 & 2.47 & 0.09 & 1.79 & 0.04 + 204 & 17.98 & 18.79 & 13.875 & 0.008 & 15.987 & 0.041 & 12.13 & 3.603 & 0.006 & 2.86 & 0.08 & 2.23 & 0.03 + 205 & 17.42 & 18.66 & 12.761 & 0.005 & 14.750 & 0.005 & 7.67 & 3.665 & 0.004 & 2.77 & 0.01 & 2.82 & 0.01 + 206 & 16.96 & 18.36 & 15.226 & 0.021 & 17.257 & 0.058 & 8.52 & 3.655 & 0.005 & 2.82 & 0.11 & 1.82 & 0.05 + 207 & 16.63 & 18.68 & 14.970 & 0.021 & 16.876 & 0.048 & 8.24 & 3.658 & 0.005 & 2.64 & 0.09 & 1.86 & 0.04 + 208 & 16.42 & 18.63 & 15.582 & 0.027 & 17.474 & 0.058 & 6.17 & 3.683 & 0.004 & 2.65 & 0.11 & 1.69 & 0.05 + 209 & 16.42 & 18.39 & 15.578 & 0.027 & 17.470 & 0.058 & 8.39 & 3.657 & 0.007 & 2.62 & 0.11 & 1.60 & 0.05 + 210 & 15.55 & 18.36 & 15.736 & 0.027 & 17.868 & 0.202 & 6.30 & 3.681 & 0.006 & 3.00 & 0.36 & 1.76 & 0.15 + 211 & 15.47 & 18.68 & 16.367 & 0.083 & 18.105 & 0.202 & 5.70 & 3.688 & 0.004 & 2.43 & 0.39 & 1.30 & 0.16 + 212 & 15.39 & 19.06 & 16.023 & 0.061 & 17.959 & 0.202 & 3.62 & 3.708 & 0.004 & 2.74 & 0.38 & 1.62 & 0.15 + 213 & 15.15 & 18.58 & 14.538 & 0.015 & 16.357 & 0.047 & 7.22 & 3.671 & 0.004 & 2.53 & 0.09 & 2.02 & 0.04 + 214 & 14.61 & 18.50 & 15.400 & 0.027 & 17.426 & 0.058 & 7.51 & 3.667 & 0.008 & 2.83 & 0.11 & 1.79 & 0.05 + 215 & 14.61 & 18.04 & 12.841 & 0.005 & 14.809 & 0.006 & 5.63 & 3.689 & 0.003 & 2.77 & 0.01 & 2.85 & 0.01 + 216 & 15.07 & 18.09 & 15.299 & 0.027 & 17.248 & 0.048 & 6.82 & 3.675 & 0.006 & 2.73 & 0.10 & 1.81 & 0.04 + 217 & 15.61 & 17.90 & 15.055 & 0.021 & 16.851 & 0.048 & 4.29 & 3.702 & 0.003 & 2.53 & 0.09 & 1.90 & 0.04 + 218 & 15.82 & 17.93 & 15.329 & 0.027 & 17.293 & 0.058 & 5.46 & 3.690 & 0.004 & 2.76 & 0.11 & 1.86 & 0.05 + 219 & 16.09 & 18.01 & 15.249 & 0.021 & 17.087 & 0.048 & 7.03 & 3.673 & 0.004 & 2.56 & 0.09 & 1.76 & 0.04 + 220 & 16.36 & 17.82 & 16.300 & 0.083 & 18.281 & 0.204 & 8.44 & 3.656 & 0.006 & 2.75 & 0.39 & 1.36 & 0.16 + 221 & 16.96 & 17.98 & 13.942 & 0.008 & 15.841 & 0.041 & 9.37 & 3.644 & 0.004 & 2.61 & 0.07 & 2.22 & 0.03 + 222 & 17.60 & 18.09 & 15.513 & 0.027 & 17.565 & 0.058 & 9.45 & 3.643 & 0.006 & 2.83 & 0.11 & 1.67 & 0.05 + 223 & 17.12 & 17.74 & 14.996 & 0.021 & 16.761 & 0.048 & 8.00 & 3.661 & 0.004 & 2.44 & 0.09 & 1.78 & 0.04 + 224 & 16.98 & 17.58 & 13.668 & 0.005 & 15.518 & 0.007 & 9.04 & 3.648 & 0.004 & 2.54 & 0.02 & 2.31 & 0.01 + 225 & 16.52 & 17.55 & 15.812 & 0.061 & 18.124 & 0.202 & 5.87 & 3.686 & 0.004 & 3.28 & 0.38 & 1.85 & 0.15 + 226 & 16.31 & 17.28 & 14.495 & 0.015 & 16.480 & 0.047 & 9.03 & 3.648 & 0.004 & 2.74 & 0.09 & 2.06 & 0.04 + 227 & 15.74 & 17.39 & 15.275 & 0.027 & 17.219 & 0.048 & 6.77 & 3.676 & 0.006 & 2.72 & 0.10 & 1.82 & 0.04 + 228 & 15.34 & 17.39 & 15.470 & 0.027 & 17.529 & 0.058 & 6.90 & 3.675 & 0.005 & 2.89 & 0.11 & 1.80 & 0.05 + 229 & 15.42 & 16.98 & 16.262 & 0.083 & 18.066 & 0.202 & 6.45 & 3.680 & 0.007 & 2.52 & 0.39 & 1.35 & 0.16 + 230 & 15.34 & 16.71 & 15.123 & 0.021 & 17.461 & 0.058 & 6.69 & 3.677 & 0.005 & 3.30 & 0.11 & 2.12 & 0.05 + 231 & 15.93 & 16.71 & 13.527 & 0.005 & 15.422 & 0.007 & 10.90 & 3.622 & 0.006 & 2.57 & 0.02 & 2.31 & 0.02 + 232 & 16.44 & 17.15 & 13.639 & 0.005 & 15.636 & 0.007 & 10.57 & 3.627 & 0.005 & 2.73 & 0.02 & 2.34 & 0.02 + 233 & 16.98 & 17.17 & 14.927 & 0.021 & 16.440 & 0.047 & 7.26 & 3.670 & 0.004 & 2.07 & 0.09 & 1.68 & 0.04 + 234 & 17.25 & 17.04 & 15.273 & 0.027 & 17.212 & 0.048 & 8.20 & 3.659 & 0.006 & 2.69 & 0.10 & 1.76 & 0.04 + 235 & 17.39 & 17.20 & 14.481 & 0.015 & 16.331 & 0.047 & 8.79 & 3.651 & 0.005 & 2.55 & 0.09 & 2.00 & 0.04 + 236 & 17.77 & 16.69 & 14.440 & 0.015 & 16.309 & 0.047 & 7.90 & 3.663 & 0.007 & 2.59 & 0.09 & 2.06 & 0.04 + 237 & 17.31 & 16.69 & 15.226 & 0.021 & 16.971 & 0.048 & 4.79 & 3.697 & 0.004 & 2.45 & 0.09 & 1.79 & 0.04 + 238 & 16.98 & 16.77 & 16.383 & 0.083 & 18.381 & 0.204 & 6.51 & 3.679 & 0.008 & 2.80 & 0.39 & 1.42 & 0.16 + 239 & 17.52 & 16.07 & 15.026 & 0.021 & 16.916 & 0.048 & 10.57 & 3.627 & 0.009 & 2.57 & 0.09 & 1.72 & 0.05 + 240 & 17.31 & 15.71 & 13.073 & 0.005 & 15.095 & 0.006 & 10.61 & 3.626 & 0.005 & 2.77 & 0.02 & 2.58 & 0.02 + 241 & 16.39 & 15.63 & 15.333 & 0.027 & 17.192 & 0.048 & 6.66 & 3.677 & 0.005 & 2.59 & 0.10 & 1.75 & 0.04 + 242 & 16.12 & 15.42 & 14.285 & 0.015 & 16.163 & 0.041 & 8.51 & 3.655 & 0.008 & 2.59 & 0.08 & 2.10 & 0.04 + 243 & 16.04 & 15.77 & 15.011 & 0.021 & 16.470 & 0.047 & 4.92 & 3.696 & 0.004 & 2.02 & 0.09 & 1.70 & 0.04 + 244 & 15.28 & 15.47 & 15.066 & 0.021 & 16.950 & 0.048 & 6.97 & 3.674 & 0.005 & 2.63 & 0.09 & 1.86 & 0.04 + 245 & 15.31 & 16.09 & 15.352 & 0.027 & 17.243 & 0.048 & 6.55 & 3.679 & 0.007 & 2.64 & 0.10 & 1.76 & 0.05 + 246 & 15.50 & 16.42 & 15.382 & 0.027 & 17.339 & 0.058 & 7.03 & 3.673 & 0.004 & 2.73 & 0.11 & 1.77 & 0.05 + 247 & 15.12 & 16.36 & 14.726 & 0.015 & 16.606 & 0.047 & 10.30 & 3.631 & 0.006 & 2.56 & 0.09 & 1.85 & 0.04 + 248 & 18.71 & 13.90 & 14.621 & 0.016 & 17.520 & 0.065 & 11.38 & 3.615 & 0.005 & 4.04 & 0.12 & 2.44 & 0.05 + 249 & 18.58 & 13.80 & 13.958 & 0.009 & 15.403 & 0.009 & 6.33 & 3.681 & 0.004 & 1.99 & 0.02 & 2.07 & 0.02 + 250 & 17.87 & 13.93 & 14.910 & 0.027 & 16.616 & 0.044 & 4.50 & 3.700 & 0.004 & 2.39 & 0.09 & 1.90 & 0.04 + 251 & 16.98 & 13.61 & 14.856 & 0.027 & 17.116 & 0.055 & 8.54 & 3.655 & 0.004 & 3.16 & 0.11 & 2.10 & 0.05 + 252 & 16.17 & 13.66 & 16.296 & 0.100 & 18.025 & 0.198 & 5.75 & 3.687 & 0.006 & 2.41 & 0.39 & 1.32 & 0.16 + 253 & 15.93 & 13.45 & 15.774 & 0.064 & 17.581 & 0.065 & 5.53 & 3.690 & 0.007 & 2.53 & 0.16 & 1.58 & 0.07 + 254 & 16.33 & 13.23 & 14.934 & 0.027 & 16.765 & 0.055 & 6.30 & 3.681 & 0.004 & 2.56 & 0.11 & 1.90 & 0.05 + 255 & 16.60 & 13.45 & 16.083 & 0.064 & 18.269 & 0.233 & 4.92 & 3.696 & 0.003 & 3.10 & 0.43 & 1.70 & 0.17 + 256 & 16.90 & 13.45 & 15.023 & 0.027 & 16.766 & 0.055 & 6.39 & 3.680 & 0.004 & 2.43 & 0.11 & 1.81 & 0.05 + 257 & 17.09 & 13.18 & 15.868 & 0.064 & 17.650 & 0.065 & 5.28 & 3.692 & 0.006 & 2.50 & 0.16 & 1.54 & 0.07 + 258 & 17.28 & 13.18 & 14.923 & 0.027 & 16.713 & 0.044 & 6.56 & 3.678 & 0.004 & 2.49 & 0.09 & 1.88 & 0.04 + 259 & 17.55 & 13.31 & 15.131 & 0.027 & 16.904 & 0.055 & 7.08 & 3.672 & 0.005 & 2.46 & 0.11 & 1.76 & 0.05 + 260 & 18.14 & 13.58 & 15.066 & 0.027 & 16.834 & 0.055 & 7.71 & 3.665 & 0.004 & 2.45 & 0.11 & 1.76 & 0.05 + 261 & 18.31 & 13.50 & 15.182 & 0.027 & 16.984 & 0.055 & 7.14 & 3.672 & 0.004 & 2.50 & 0.11 & 1.76 & 0.05 + 262 & 18.90 & 13.20 & 11.054 & 0.002 & 12.618 & 0.002 & 11.69 & 3.610 & 0.005 & 2.06 & 0.01 & 3.06 & 0.01 + 263 & 18.50 & 12.58 & 14.975 & 0.027 & 16.649 & 0.044 & 7.40 & 3.669 & 0.005 & 2.31 & 0.09 & 1.75 & 0.04 + 264 & 18.31 & 12.28 & 12.522 & 0.003 & 14.414 & 0.005 & 10.74 & 3.624 & 0.004 & 2.57 & 0.01 & 2.72 & 0.01 + 265 & 18.71 & 12.26 & 15.774 & 0.064 & 17.527 & 0.065 & 5.80 & 3.687 & 0.005 & 2.45 & 0.16 & 1.54 & 0.07 + 266 & 19.17 & 12.18 & 12.940 & 0.005 & 14.683 & 0.005 & 11.04 & 3.620 & 0.007 & 2.34 & 0.02 & 2.45 & 0.02 + 267 & 17.87 & 12.64 & 15.189 & 0.027 & 17.110 & 0.055 & 7.94 & 3.662 & 0.005 & 2.67 & 0.11 & 1.79 & 0.05 + 268 & 17.74 & 12.26 & 13.417 & 0.008 & 15.210 & 0.006 & 9.23 & 3.646 & 0.004 & 2.46 & 0.02 & 2.37 & 0.01 + 269 & 17.15 & 12.18 & 16.087 & 0.064 & 18.519 & 0.233 & 6.83 & 3.675 & 0.006 & 3.44 & 0.43 & 1.78 & 0.17 + 270 & 16.98 & 12.31 & 15.043 & 0.027 & 16.845 & 0.055 & 8.59 & 3.654 & 0.003 & 2.48 & 0.11 & 1.75 & 0.05 + 271 & 16.69 & 12.18 & 12.248 & 0.003 & 14.120 & 0.003 & 12.93 & 3.591 & 0.006 & 2.48 & 0.02 & 2.70 & 0.02 + 272 & 16.77 & 12.99 & 14.951 & 0.027 & 16.704 & 0.044 & 6.96 & 3.674 & 0.003 & 2.43 & 0.09 & 1.83 & 0.04 + 273 & 16.39 & 12.61 & 12.552 & 0.003 & 14.315 & 0.005 & 4.08 & 3.704 & 0.002 & 2.48 & 0.01 & 2.89 & 0.01 + 274 & 15.55 & 13.10 & 15.324 & 0.039 & 17.308 & 0.065 & 4.81 & 3.697 & 0.003 & 2.80 & 0.13 & 1.89 & 0.06 + 275 & 15.42 & 12.50 & 14.718 & 0.016 & 16.613 & 0.044 & 6.34 & 3.681 & 0.003 & 2.65 & 0.08 & 2.03 & 0.03 + 276 & 15.96 & 12.23 & 13.624 & 0.008 & 15.550 & 0.009 & 8.26 & 3.658 & 0.004 & 2.67 & 0.02 & 2.41 & 0.01 + 277 & 15.47 & 11.77 & 15.327 & 0.039 & 17.333 & 0.065 & 4.94 & 3.696 & 0.005 & 2.83 & 0.13 & 1.90 & 0.06 + 278 & 15.50 & 11.56 & 15.479 & 0.039 & 17.545 & 0.065 & 6.01 & 3.684 & 0.008 & 2.91 & 0.14 & 1.84 & 0.06 + 279 & 15.85 & 11.34 & 13.934 & 0.009 & 15.885 & 0.045 & 7.63 & 3.666 & 0.006 & 2.72 & 0.08 & 2.32 & 0.04 + 280 & 16.01 & 11.15 & 15.087 & 0.027 & 17.174 & 0.055 & 9.49 & 3.642 & 0.008 & 2.89 & 0.11 & 1.86 & 0.05 + 281 & 15.82 & 11.10 & 15.398 & 0.039 & 17.480 & 0.065 & 6.14 & 3.683 & 0.005 & 2.93 & 0.13 & 1.87 & 0.06 + 282 & 16.50 & 11.58 & 15.047 & 0.027 & 16.877 & 0.055 & 8.00 & 3.661 & 0.010 & 2.53 & 0.11 & 1.79 & 0.05 + 283 & 16.74 & 11.58 & 14.931 & 0.027 & 16.922 & 0.055 & 8.07 & 3.661 & 0.009 & 2.77 & 0.11 & 1.93 & 0.05 + 284 & 16.71 & 11.31 & 15.040 & 0.027 & 17.024 & 0.055 & 9.80 & 3.638 & 0.005 & 2.73 & 0.11 & 1.81 & 0.05 + 285 & 16.71 & 10.99 & 15.251 & 0.039 & 17.147 & 0.055 & 10.42 & 3.629 & 0.009 & 2.58 & 0.12 & 1.64 & 0.06 + 286 & 16.96 & 10.66 & 16.019 & 0.064 & 17.658 & 0.065 & 5.89 & 3.686 & 0.006 & 2.28 & 0.16 & 1.37 & 0.07 + 287 & 17.17 & 11.20 & 15.337 & 0.039 & 17.459 & 0.065 & 8.96 & 3.649 & 0.007 & 2.95 & 0.14 & 1.81 & 0.06 + 288 & 17.20 & 11.58 & 15.334 & 0.039 & 17.229 & 0.055 & 5.43 & 3.691 & 0.005 & 2.66 & 0.12 & 1.81 & 0.05 + 289 & 17.58 & 11.29 & 15.161 & 0.027 & 17.013 & 0.055 & 4.87 & 3.696 & 0.002 & 2.61 & 0.11 & 1.88 & 0.05 + 290 & 18.28 & 11.50 & 15.082 & 0.027 & 16.835 & 0.055 & 6.09 & 3.684 & 0.005 & 2.44 & 0.11 & 1.81 & 0.05 + 291 & 18.33 & 11.69 & 15.143 & 0.027 & 16.849 & 0.055 & 5.51 & 3.690 & 0.004 & 2.38 & 0.11 & 1.78 & 0.05 + 292 & 18.06 & 10.77 & 14.916 & 0.027 & 16.814 & 0.055 & 10.69 & 3.625 & 0.008 & 2.58 & 0.11 & 1.76 & 0.05 + 293 & 17.77 & 10.75 & 14.598 & 0.016 & 16.558 & 0.044 & 11.05 & 3.620 & 0.005 & 2.66 & 0.08 & 1.91 & 0.04 + 294 & 17.63 & 10.34 & 15.301 & 0.039 & 17.319 & 0.065 & 8.23 & 3.659 & 0.007 & 2.81 & 0.14 & 1.79 & 0.06 + 295 & -6.67 & 18.85 & 14.814 & 0.014 & 16.883 & 0.042 & 10.09 & 3.634 & 0.007 & 2.85 & 0.08 & 1.93 & 0.04 + 296 & -7.51 & 19.31 & 15.600 & 0.029 & 17.609 & 0.051 & 4.45 & 3.700 & 0.004 & 2.84 & 0.11 & 1.81 & 0.05 + 297 & -7.70 & 19.12 & 15.219 & 0.014 & 17.213 & 0.042 & 5.71 & 3.688 & 0.004 & 2.81 & 0.08 & 1.91 & 0.03 + 298 & -8.29 & 19.01 & 14.783 & 0.014 & 16.753 & 0.042 & 8.22 & 3.659 & 0.005 & 2.74 & 0.08 & 1.97 & 0.03 + 299 & -8.37 & 18.82 & 15.067 & 0.014 & 16.961 & 0.042 & 5.94 & 3.685 & 0.003 & 2.66 & 0.08 & 1.90 & 0.03 + 300 & -9.15 & 18.74 & 11.437 & 0.002 & 13.493 & 0.002 & 8.46 & 3.656 & 0.003 & 2.86 & 0.01 & 3.35 & 0.01 + 301 & -8.26 & 18.23 & 14.468 & 0.011 & 16.434 & 0.044 & 9.28 & 3.645 & 0.004 & 2.71 & 0.08 & 2.05 & 0.03 + 302 & -6.72 & 18.14 & 13.651 & 0.005 & 15.524 & 0.007 & 9.17 & 3.646 & 0.004 & 2.58 & 0.02 & 2.33 & 0.01 + 303 & -7.07 & 17.77 & 15.142 & 0.014 & 17.064 & 0.042 & 7.86 & 3.663 & 0.007 & 2.67 & 0.08 & 1.82 & 0.04 + 304 & -8.13 & 17.63 & 15.598 & 0.029 & 17.594 & 0.051 & 5.03 & 3.695 & 0.004 & 2.82 & 0.11 & 1.78 & 0.05 + 305 & -8.48 & 17.74 & 15.391 & 0.029 & 17.322 & 0.051 & 7.87 & 3.663 & 0.005 & 2.68 & 0.11 & 1.72 & 0.05 + 306 & -8.72 & 17.63 & 15.467 & 0.029 & 17.360 & 0.051 & 7.25 & 3.670 & 0.004 & 2.64 & 0.11 & 1.69 & 0.05 + 307 & -9.94 & 17.90 & 15.508 & 0.029 & 17.631 & 0.051 & 10.16 & 3.633 & 0.006 & 2.93 & 0.11 & 1.69 & 0.05 + 308 & -10.07 & 17.28 & 15.566 & 0.029 & 17.617 & 0.051 & 7.43 & 3.668 & 0.006 & 2.87 & 0.11 & 1.74 & 0.05 + 309 & -10.12 & 17.50 & 15.794 & 0.060 & 17.879 & 0.196 & 5.97 & 3.685 & 0.007 & 2.94 & 0.36 & 1.72 & 0.15 + 310 & -8.88 & 17.28 & 15.160 & 0.014 & 17.095 & 0.042 & 8.64 & 3.653 & 0.005 & 2.68 & 0.08 & 1.78 & 0.03 + 311 & -8.80 & 17.15 & 15.482 & 0.029 & 17.506 & 0.051 & 9.00 & 3.649 & 0.007 & 2.80 & 0.11 & 1.69 & 0.05 + 312 & -8.56 & 17.31 & 15.606 & 0.029 & 17.457 & 0.051 & 8.29 & 3.658 & 0.004 & 2.56 & 0.11 & 1.57 & 0.05 + 313 & -8.05 & 16.98 & 15.477 & 0.029 & 17.505 & 0.051 & 7.95 & 3.662 & 0.006 & 2.83 & 0.11 & 1.74 & 0.05 + 314 & -6.94 & 17.12 & 15.234 & 0.014 & 17.284 & 0.051 & 9.04 & 3.648 & 0.007 & 2.84 & 0.10 & 1.80 & 0.04 + 315 & -6.59 & 16.98 & 15.524 & 0.029 & 17.516 & 0.051 & 5.68 & 3.688 & 0.005 & 2.80 & 0.11 & 1.79 & 0.05 + 316 & -6.86 & 16.71 & 15.647 & 0.029 & 17.546 & 0.051 & 6.88 & 3.675 & 0.004 & 2.65 & 0.11 & 1.64 & 0.05 + 317 & -6.94 & 16.58 & 15.254 & 0.029 & 17.181 & 0.042 & 6.51 & 3.679 & 0.003 & 2.70 & 0.09 & 1.83 & 0.04 + 318 & -7.64 & 16.39 & 15.395 & 0.029 & 17.375 & 0.051 & 8.66 & 3.653 & 0.007 & 2.74 & 0.11 & 1.72 & 0.05 + 319 & -6.61 & 15.77 & 14.757 & 0.014 & 16.307 & 0.044 & 10.12 & 3.633 & 0.006 & 2.08 & 0.08 & 1.65 & 0.04 + 320 & -7.48 & 15.36 & 15.453 & 0.029 & 17.392 & 0.051 & 8.26 & 3.658 & 0.005 & 2.69 & 0.11 & 1.68 & 0.05 + 321 & -7.75 & 15.55 & 13.625 & 0.005 & 15.630 & 0.007 & 10.90 & 3.622 & 0.007 & 2.73 & 0.02 & 2.33 & 0.02 + 322 & -8.34 & 16.52 & 13.477 & 0.005 & 15.438 & 0.007 & 8.02 & 3.661 & 0.004 & 2.73 & 0.02 & 2.50 & 0.01 + 323 & -8.96 & 16.96 & 14.418 & 0.011 & 16.562 & 0.044 & 9.48 & 3.642 & 0.005 & 2.97 & 0.08 & 2.17 & 0.04 + 324 & -9.10 & 16.79 & 13.835 & 0.008 & 15.898 & 0.039 & 10.18 & 3.633 & 0.005 & 2.84 & 0.07 & 2.32 & 0.03 + 325 & -8.53 & 15.66 & 12.615 & 0.003 & 14.755 & 0.006 & 9.82 & 3.638 & 0.005 & 2.96 & 0.01 & 2.87 & 0.02 + 326 & -8.59 & 15.34 & 13.698 & 0.005 & 15.683 & 0.007 & 10.55 & 3.627 & 0.005 & 2.71 & 0.02 & 2.31 & 0.01 + 327 & -9.05 & 15.28 & 12.764 & 0.004 & 14.750 & 0.005 & 9.86 & 3.637 & 0.006 & 2.73 & 0.02 & 2.72 & 0.02 + 328 & -9.18 & 15.85 & 15.145 & 0.014 & 17.151 & 0.042 & 6.83 & 3.675 & 0.004 & 2.81 & 0.08 & 1.90 & 0.03 + 329 & -9.99 & 15.98 & 14.581 & 0.011 & 16.522 & 0.044 & 5.01 & 3.695 & 0.004 & 2.73 & 0.08 & 2.16 & 0.03 + [ gc_table ] the gc h - r diagram is shown in figure [ hr ] . our data clearly show the red clump at @xmath67 , as well as the upper red giant branch / early asymptotic giant branch . there is an indication of the agb bump at @xmath68 , and we also detect some lower red giant branch stars at lower luminosities , though the observations are highly incomplete in this region . the data set in figure [ hr ] allows for a more robust analysis of the galactic center star formation history than any other previously published data set . photometric studies are limited to the modeling of luminosities alone @xcite , due to intrinsic variations in late - type giant colors and the large variation in galactic center extinction . this point is illustrated in figure [ cmd ] , which compares the observed h - r diagram to a color - magnitude diagram ( cmd ) for the same gc stars . the rgb / agb and rc populations are more clearly distinguished in the h - r diagram than in the cmd , and there is less scatter in @xmath7 than @xmath69 , due to variations in gc extinction . as a result , only the k - band luminosity function can reliably be modeled with broadband photometry alone . the observed h - r diagram in this study is also an improvement with respect to previous spectroscopic work @xcite , due to our improved magnitude limit ( @xmath55 mag . deeper ) , which allows for the analysis of well - populated regions in the h - r diagram , in which the evolutionary models are fairly well understood . than @xmath70 , due to variations in gc extinction . ] the errors in temperature and luminosity are shown with the data in grey in figure [ hr ] . average errors in temperature and luminosity for stars with luminosities @xmath71 ( the red clump region ) are shown at right . the temperature limit for the minimum co index and the 50% completeness limit for the average extinction correction ( @xmath72 mag . ) are also shown . solar metallicity ( in red , left ) and metal - poor ( z=0.008 , in blue , right ) isochrones from @xcite are overplotted . the isochrones show the wide range of ages spanned by the galactic center population . figure [ hr ] also demonstrates the age - metallicity degeneracy inherent in this part of the diagram . stars in the same part of the h - r diagram may represent a metal - rich , intermediate - age ( @xmath73 a few gyr ) population or a metal - poor , old ( @xmath74 5 gyr ) population . for stars younger than @xmath55 gyr , the galactic center stellar population is known to be approximately solar @xcite . for older stars , the galactic center metallicity distribution is not well - known . we will address this uncertainty later in the paper . to investigate the star formation history implied by our sample , we begin with a qualitative discussion , followed by quantitative analysis in 3.4 - 3.5 . we start by considering three models based on candidate star formation histories presented in the literature . the first scenario consists of an ancient burst of star formation 7.5 - 8.5 gyr ago , similar to the single stellar population of the bulge . @xcite compared this scenario to the k band luminosity function ( klf ) of the inner parsec . the second model consists of constant star formation between 10 myr and 10 gyr ago . this model is based on the best - fit model of @xcite , who considered the galactic center klf within 40 pc of the center . the third model corresponds to the best - fit star formation history of @xcite , who fit the h - r diagram of asymptotic giant branch and cool supergiant stars within the central 5 pc , using four specified age bins . all models are summarized in table [ sf_models ] . llccccccccc 1 & bulge - like & 7.50 - 8.50 & 0.015 & 2.35 & 0.7 & 120 & 3228.8 & 0.00 & 1.00 & - + 2 & continuous & 0.01 - 10.0 & 0.019 & 2.35 & 0.7 & 120 & 404.9 & 3.76 & 1.00 & - + 3 & blum et al . & 0.01 - 0.10 & 0.019 & 2.35 & 0.7 & 120 & 678.8 & 0.01 & 0.65 & - + & & 0.10 - 1.00 & 0.019 & 2.35 & 0.7 & 120 & & & 0.06 & - + & & 1.00 - 5.00 & 0.019 & 2.35 & 0.7 & 120 & & & 0.09 & - + & & 5.00 - 12.0 & 0.019 & 2.35 & 0.7 & 120 & & & 0.21 & - + 4 & two - bin , solar & 0.01 - 5.00 & 0.019 & 2.35 & 0.7 & 120 & 384.5 & 3.03 & 0.82 & 0.04 + & & 5.00 - 12.0 & 0.019 & 2.35 & 0.7 & 120 & & & 0.18 & 0.07 + 5 & two - bin , solar & 0.01 - 7.00 & 0.019 & 2.35 & 0.7 & 120 & 270.0 & 16.66 & 0.90 & 0.04 + & & 7.00 - 12.0 & 0.019 & 2.35 & 0.7 & 120 & & & 0.10 & 0.09 + 6 & two - bin , poor & 0.01 - 5.00 & 0.019 & 2.35 & 0.7 & 120 & 450.0 & 0.68 & 0.91 & 0.05 + & & 5.00 - 12.0 & 0.008 & 2.35 & 0.7 & 120 & & & 0.09 & 0.08 + 7 & closed - box & 0.01 - 12.0 & 0.004 - 0.019 & 2.35 & 0.7 & 120 & 1035.6 & 0.00 & 1.00 & - + 8 & flat imf & 0.01 - 12.0 & 0.019 & 0.85 & 0.7 & 120 & 242.8 & 39.70 & 1.00 & - + 9 & high @xmath75 & 0.01 - 12.0 & 0.019 & 2.35 & 2.5 & 120 & 4201.8 & 0.00 & 1.00 & - + 10 & high @xmath75 & 0.01 - 12.0 & 0.019 & 2.35 & 1.5 & 120 & 551.4 & 0.00 & 1.00 & - + 11 & model 5 + 8 hybrid & 0.01 - 7.00 & 0.019 & 0.85 & 0.7 & 120 & 257.0 & 26.3 & 0.48 & 0.02 + & & 7.00 - 12.0 & 0.019 & 0.85 & 0.7 & 120 & & & 0.51 & 0.20 + [ sf_models ] for each scenario , we generated model h - r diagrams using the synthetic color - magnitude diagram computation algorithm iac - star @xcite . the algorithm uses a monte carlo approach to compute composite stellar populations on a star - by - star basis , using a specified set of evolutionary tracks . the code accommodates several additional inputs , including the initial mass function , star formation rate function , and chemical enrichment law . to compare the generated models to the data , we added gaussian noise based on the average observed errors in luminosity and temperature . we also randomly removed stars from the models according to our estimated spectroscopic completeness ( fig [ comp_err ] ) . the models are shown against our galactic center sample in figure [ sf_models_lit ] . a qualitative comparison suggests that the ancient burst model is insufficient to fully describe the observed data set . the continuous star formation model and the @xcite model span the full range in temperature and luminosity of the observed data set but seem to have relatively too many stars at cool temperatures . it is unlikely this observed effect is caused by systematic errors or a selection bias . a systematic underestimate of the gc co indices would shift the pattern to hotter temperatures . however , the red edge of the hertzsprung gap ( log @xmath7 @xmath76 ) is well matched by the models , making such a systematic effect unlikely . in addition , some cool rgb / agb stars are present in the data , while a systematic shift to warmer temperatures would allow none . finally , the brightness and co strength of cool rgb / agb stars compared to warmer , dimmer red clump stars means that cool rgb / agb stars are relatively easy to detect and classify . there is no obvious way of selectively removing these stars from the sample . motivated by the comparison in figure [ sf_models_lit ] , we consider three alternative scenarios that could potentially explain the relative paucity of stars at low temperatures : 1 . the first and simplest possibility is that the star formation rate was low at early times ( @xmath77 gyr ) . to test this hypothesis , we generated two models consisting of linear combinations of constant star formation in two specified age bins ( models 4 and 5 in table [ sf_models ] ) . model 4 assumes bins of 0.01 - 5 gyr and 5 - 12 gyr . model 5 assumes bins of 0.01 - 7 gyr and 7 - 12 gyr . a second possibility is that the oldest stars are metal poor . the previously described models assume a solar metallicity for all times . this assumption is motivated by the work of @xcite , who found the stellar [ fe / h ] abundance to be approximately solar for stars younger than @xmath55 gyr . for older stars , the galactic center metallicity distribution has not been determined . however , the bulge , which formed 7 - 12 gyr ago , is known to have a nearly solar distribution @xcite . an extremely metal poor ancient population would , therefore , represent a completely separate population from the bulge . given the current observed differences between the bulge and the nucleus discussed in 1 , we can not exclude this possibility . we , therefore , generated two models to represent this scenario . the first model , model 6 , consists of a linear combination of constant star formation in two age bins with solar metallicity for 0.01 - 5 gyr and a metallicity of 0.008 for 5 - 12 gyr . the second model , model 7 , is a simple closed box model assuming constant star formation for 0.01 - 12 gyr , a starting metallicity of 0.004 and an ending metallicity of solar . 3 . a final way to explain the relative lack of stars at cool temperatures invokes a non - standard initial mass function ( imf ) . in all above models , we assume a salpeter imf with an upper mass limit of 120 @xmath78 and a lower mass limit of 0.7 @xmath78 . as stars less massive than @xmath79 have main sequence lifetimes comparable to the age of the universe , we are not sensitive to stars below this mass limit . recent results indicate that the current initial mass function of the galactic center is flatter than a salpeter function or has an unusally high lower mass cut - off @xcite . such a mass function could explain the relative lack of low mass stars in the observed old stellar population . to test this hypothesis , we consider three models , all assuming continuous star formation at solar metallicity for 0.01 - 12 gyr . the first model , model 8 , assumes a flat single power - law slope ( @xmath80 ) , chosen to match the results of @xcite . models 9 and 10 both assume a standard salpeter slope ( @xmath81 ) , but with lower mass limits of 2.5 @xmath78 and 1.5 @xmath78 , respectively . results stemming from these three hypotheses are tested in 3.5 . to quantitatively compare the models described above to the data , we adopted the numerical techniques described in @xcite . we first binned the observed and model h - r diagrams in temperature and luminosity for stars above our 50% completeness threshold , using uniform bins with size three times our average errors ( @xmath82 , @xmath83 ) . the choice of bins in this technique is somewhat subjective . however , tests using different bins sizes showed that while the binning scheme does change the fit quality , it does not significantly affect the derived star formation rates . this finding is in agreement with @xcite . for models with two age bins , we used the numerical recipes procedure amoeba @xcite to search for the linear combination of models that minimized the poisson maximum likelihood parameter : @xmath84 @xcite . here , @xmath85 is the number of observed stars and @xmath86 is the number of model stars in bin @xmath87 . for models with fixed relative star formation rates , we scaled the model distribution to minimize @xmath88 . to estimate the errors in the derived star formation history , we implemented the technique described in @xcite . we built a set of 100 h - r diagrams consisting of a random sampling of 329 stars drawn from the observed h - r diagram , allowing each observed star to be selected any number of times . we then re - derived the star formation history for each h - r diagram and given model . the resulting standard deviation in the derived star formation rate was taken as the 1@xmath39 uncertainty . we measured the fit quality through a second set of monte carlo simulations . in this set , we generated 10,000 synthetic data sets drawn from the fitted models , selecting 329 stars randomly for each trial and re - deriving the star formation history for the selected subset . for a model that is an accurate representation of the true star formation history , @xmath88 derived from the actual data set should be comparable to @xmath88 derived from a typical synthetic data set . therefore , to establish the goodness of fit , we calculated the percentage of trials , @xmath89 , in which @xmath88 was larger when fitting the synthetic data sets to the models than when fitting the true data sample to the models . if the model is a good representation of the true star formation history , @xmath89 should be approximately 50% . the results of all model fits are listed in table [ sf_models ] . hess diagrams showing the difference between the observed data histogram and the best - fit model histograms are shown in figure [ hess ] . white indicates regions where the model produces too many stars relative to the observed data set , and black indicates regions where the model produces too few stars relative to the observed data set . examination of the @xmath89 values in table [ sf_models ] confirms the qualitative result discussed in 3.3 that no literature model ( models 1 - 3 ) is a likely description of the observed data set . this result is also reflected in the first three panels in figure [ hess ] , which show white diagonal streaks , corresponding to the overdensity of cool stars in the models relative to the data . table [ sf_models ] also suggests that the early - history low star formation rate models ( models 4 and 5 ) are unlikely representations of the data , though model 5 returns a considerably better fit , and we are unable to completely exclude this model . however , examination of the fourth and fifth panels in figure [ hess ] show that both models systematically overpredict the number of cool stars , as in models 1 - 3 . the early - history low metallicity models ( models 6 and 7 ) also show systematic deficiencies when compared to the data . both models predict a red clump / horizontal branch morphology that is too blue compared to the observed distribution . similar discrepencies between the models and data are present in the high imf mass limit models ( models 9 and 10 ) . each predicts a hotter and more luminous red clump than is observed . however , while models 9 and 10 predict no old , cool stars , the flat imf model ( model 8) predicts few old , cool stars , in good agreement with the observations . model 8 also gives a satisfactory goodness of fit ( @xmath89 = 40% ) suggesting that this model is a likely representation of the observed distribution . using the models listed in table [ sf_models ] , we tested for population differences as a function of distance from the cluster center . such differences within the central 1 - 2 pc are expected due to mass segregation , and several broadband photometry studies have found indications of such an effect @xcite . to test for this effect in our data , we separated the data into two sets containing the four innermost ( @xmath90 , @xmath91 median ) and outermost ( @xmath92 , @xmath93 median ) spectral regions shown in figure [ mos ] . we then refit the two best - fit models ( models 5 and 8) to each subset , applying the average errors and completeness specific to that subset . this process returned fits for each subset that are consistent with those obtained for the entire data set . in particular , the subset @xmath89 values were within 10% of that obtained for the entire data set , and the relative star formation rates returned by both subsets for model 5 were within 1@xmath39 of the rates returned by the entire data set . we therefore , find no significant evidence in our data for a population gradient . this result is not that surprising , as the broadband photometry results of @xcite suggest that variations in the cool , low - mass stellar population within the central parsec should be most evident 3 - 7 arcsec from sgr a@xmath0 . our sample is outside of the region expected to exhibit the largest population differences ( see figure [ mos ] ) . in the future , the technique we describe could be applied to regions closer to sgr a@xmath0 , although limited spectroscopic completeness due to increased stellar density would complicate such a study . still , a thorough spectroscopic study of the red - clump and rgb / agb populations as a function of distance from sgr a@xmath0 could provide the first definitive test of mass segregation within the central parsec @xcite . with the restriction to simple models , the top - heavy imf model appears to be superior to the remaining three model families we consider ( low star formation rate ( sfr ) at early times , low metallicity at early times , imf with high lower mass limit ) . the low - sfr , low - z , and high @xmath75 families all show consistent systematic deficiencies when compared to the data . it is , therefore , unlikely one could slightly change the chosen parameters ( i.e. precise metallicity , age bin cut - offs , mass ranges , etc . ) within these families to produce a model that is an adequate description of the data . of the models considered in table [ sf_models ] , model 8 ( continuous star formation at solar metallicity with a top - heavy imf ) fits the observations best , and our monte carlo tests show that it is a reasonable description of the data . based on the goodness of fit , we can not completely exclude model 5 . however , as discussed in the previous section , this model shows systematic deviations from the data ; we , therefore , favor model 8 . we note that model 8 is probably not the only possible description of the data with this degree of likelihood . all models so far discussed assume a simplistic description of the gc star formation history , and there are likely more complicated scenarios that also adequately describe the data . for instance , an additional model with age bins identical to model 5 and an imf slope identical to model 8 yields a reasonable fit ( @xmath94 , @xmath95 ) . however , we do not feel that increasing the number of free parameters is justified , given the limits , uncertainties , and size of our data set . in the context of our current knowledge about the galactic center , continuous star formation with a top - heavy imf is reasonable . at present , there is substantial evidence for stars in the galactic nucleus spanning a wide range of ages , based on results from broadband photometry , and several distinct age tracers , including supergiants , agb stars , and oh / ir stars ( see 1 ) . the present distribution and kinematics of gas in the inner galaxy is also consistent with continuous star formation @xcite . in addition , the most recent epoch of star formation in the galactic center is likely represented by a top - heavy initial mass function @xcite . if this recent epoch of star formation is not anomolous and periodic bursts of gc star formation have occurred throughout the history of the galaxy , there is no a priori reason to believe the gc imf would change with time . the present data set appears to support this picture . we note that the mass traced by giants in the central cluster is primarily set by the initial mass formed , rather than dynamical effects , since mass segregation is expected to only strongly affect the central @xmath50.01 pc of the galactic center @xcite . on the 1 - 2 pc scale studied in this paper , the efficiency of mass segregation is thought to be much lower , in agreement with observational photometry results @xcite . though our findings are broadly consistent with several stellar population studies of the galactic center and the nucleus at large , there remains some disagreement . in particular , our results are somewhat different from those presented in @xcite , who probed the central 5 pc and used supergiants and bright agb stars to quantify the star formation history in a method very similar to that employed here . using their observed h - r diagram , @xcite argued for variable star formation over a wide range of ages , with the majority of stars formed more than 5 gyr ago at solar metallicity . we find a much smaller fraction of our sample is represented by solar - metallicity low - mass stars with ages @xmath74 5 gyr . the present work and the work of @xcite study different regions in the h - r diagram . each regime has relative advantages and disadvantages for studying the gc star formation history , and we discuss each in turn : 1 . _ field of view _ : due to the rarity of supergiants and bright agb stars in the galactic center , @xcite probed all late - type stars above their magnitude limit for the entire central cluster ( r @xmath96 pc ) . they , therefore , derived absolute star formation rates and were able to compare the cluster mass implied to estimates from dynamical studies . in contrast , our study probes a relatively small region on the sky ( @xmath50.2 pc@xmath34 within r @xmath97 pc ) . as such , our results are somewhat susceptible to population inhomogeneities in the central cluster , and we are able to derive only relative star formation rates . stellar crowding _ : spectral extraction from our sinfoni fields is complicated by stellar crowding . while we took care to ensure that our derived co indices are largely independent of exact pixel and background selection , contamination by neighboring stellar spectra remains a source of uncertainty for many of the dim stars . the stars studied by @xcite , on the other hand , are well separated from each other . in addition , the gc agbs and supergiants are much brighter than neighboring stars , so there is negligible uncertainty in the @xcite spectra due to crowding . spectral classification _ : the spectral classification method presented by @xcite is very similar to that presented here . our depper study has the advantage that only giants are observed , and thus , a separate luminosity class determination is not required . additionally , the stars we study are warmer than those studied by @xcite , and thus , our co-@xmath7 relationship is tighter than theirs , due to reduced uncertainties in model atmosphere spectra at warmer temperatures @xcite . we further note that the derived temperatures for the red supergiants in the work of @xcite are likely systematically too cool , as @xcite recently showed that red supergiants are @xmath5400 k warmer than previously thought . adequacy of stellar evolution models _ : the evolutionary models in the part of the h - r diagram probed by our study ( red clump , red giant branch , early agb ) are much less uncertain than the part of the h - r diagram studied by @xcite ( supergiants , tp - agbs ) . @xcite review the adequacy of agb stellar evolution models for deriving star formation histories , concluding that the observed bright agb populations are often more sensitive to poorly known modeling parameters than the star formation history . the input physics for the late agb stages is not well determined ( e.g. mass loss , convection , efficiency of the third dredge - up ) , and currently , only the padova libraries include these stages @xcite . while uncertainties in mass loss and convection are also present for the red clump , rgb , and early agb , the input physics for these stages is much better understood , and a number of stellar evolution models including these phases have been compared and tested against observations . for this reason , we believe our star formation history fit is more reliable than that presented in @xcite . finally , our data set also has the benefit that it contains several evolutionary features ( red clump , agb bump ) that are cleraly distinguishable from surrounding regions in the h - r diagram . the diagram morphology assures us that no large systematic effects are present in the data and allows us to construct physically motivated models . the region of the h - r diagram studied by @xcite has no such morphological features . while we believe that our results are more robust than those presented in @xcite , we also note that the findings are not necessarily in conflict . the region probed by @xcite extends to @xmath52.5 pc from the center , whereas we probe only the central parsec . furthermore , we note that there are some similarities in our findings . both studies find evidence for star formation throughout the history of the galaxy , and both studies suggest that purely solar metallicity models are needed to produce the observed data . still , further work is needed to resolve the remaining discrepencies ( i.e. , variable versus continuous star formation , salpeter versus flat imf ) . specifically , a sample tracing a large region on the sky and containing thousands of stars would represent a significant step forward in this field . the planned flamingos-2 gc survey on gemini @xcite will obtain 4000 late - type giant spectra out to one degree in galacto - centric radius ( @xmath5140 pc ) and with a spectral resolution of r@xmath520,000 . it will , therefore , provide unprecedented information on the galactic center chemical enrichment and star formation history . we observed 329 late - type giants @xmath98 north of sgr a@xmath0 with the integral field spectrometer sinfoni on the vlt . combining spectral classifications with naco photometry , we derived luminosities and effective temperatures for these stars . due to the improved magnitude limit of our sample relative to previous work , our derived h - r diagram clearly shows the red clump , as well as the red giant branch and asymptotic giant branch . using a maximum likelihood analysis , we compared the observed distribution to models representing ten possible star formation histories . the best - fit model corresponds to continuous star formation over the last 12 gyr with a top - heavy imf . the similarity of this result to the imf observed for the most recent epoch of star formation is intriguing and perhaps suggests a connection between recent star formation and the stars formed throughout the history of the galactic center . the upcoming flamingos-2 gc survey on gemini will provide important information needed to further understand this suggestive result . the authors thank amiel sternberg for his helpful discussions and insights . is grateful to the graduate research fellowship program at nsf and the graduate opportunity program fellowship at uc berkeley for funding this research . acknowleges support from the alexander von humboldt foundation . t.a . is supported by minerva grant 8563 and a new faculty grant by sir h. djangoly , cbe , of london , uk . this work has been supported by the national science foundation science through ast 0205999 and the center for adaptive optics , managed by the university of california at santa cruz under cooperative agreement no . ast 9876783 . this work has made use of the iac - star synthetic cmd computation code . iac - star is suported and maintained by the computer division of the instituto de astrofisica de canarias . adelman , s. j. , pintado , o. i. , nieva , f. , rayle , k. e. , & sanders , s. e. 2002 , , 392 , 1031 alexander , t. , & sternberg , a. 1999 , , 520 , 137 alexander , t. 2005 , , 419 , 65 alonso , a. , arribas , s. , & martnez - roger , c. 1999 , , 139 , 335 aparicio , a. , & gallart , c. 2004 , , 128 , 1465 ballero , s. k. , matteucci , f. , origlia , l. , & rich , r. m. 2007 , , 467 , 123 bell , r. a. , & gustafsson , b. 1989 , , 236 , 653 bessell , m. s. , & brett , j. m. 1988 , , 100 , 1134 blackwell , d. e. , & lynas - gray , a. e. 1994 , , 282 , 899 blackwell , d. e. , & lynas - gray , a. e. 1998 , , 129 , 505 blum , r. d. , sellgren , k. , & depoy , d. l. 1996a , , 470 , 864 blum , r. d. , sellgren , k. , & depoy , d. l. 1996b , , 112 , 1988 blum , r. d. , ramrez , s. v. , sellgren , k. , & olsen , k. 2003 , , 597 , 323 bonnet , h. , et al . 2003 , , 4839 , 329 carpenter , j. m. 2001 , , 121 , 2851 carr , j. s. , sellgren , k. , & balachandran , s. c. 2000 , , 530 , 307 davidge , t. j. , simons , d. a. , rigaut , f. , doyon , r. , & crampton , d. 1997 , , 114 , 2586 diolaiti , e. , bendinelli , o. , bonaccini , d. , close , l. , currie , d. , & parmeggiani , g. 2000 , , 147 , 335 dolphin , a. 1997 , new astronomy , 2 , 397 dolphin , a. e. 2002 , , 332 , 91 dyck , h. m. , benson , j. a. , van belle , g. t. , & ridgway , s. t. 1996 , , 111 , 1705 eikenberry , s. , et al . 2006 , , 6269 , eisenhauer , f. , et al . 2003a , the messenger , 113 , 17 eisenhauer , f. , et al . 2003b , , 4841 , 1548 eisenhauer , f. , schdel , r. , genzel , r. , ott , t. , tecza , m. , abuter , r. , eckart , a. , & alexander , t. 2003c , , 597 , l121 engelke , c. w. 1992 , , 104 , 1248 feast , m. w. 1996 , , 278 , 11 figer , d. f. , mclean , i. s. , & morris , m. 1999 , , 514 , 202 figer , d. f. , et al . 2002 , , 581 , 258 figer , d. f. , rich , r. m. , kim , s. s. , morris , m. , & serabyn , e. 2004 , , 601 , 319 freitag , m. , amaro - seoane , p. , & kalogera , v. 2006 , , 649 , 91 frogel , j. a. , persson , s. e. , matthews , k. , & aaronson , m. 1978 , , 220 , 75 fitzgerald , m. p. , & graham , j. r. 2006 , , 637 , 541 fluks , m. a. , plez , b. , the , p. s. , de winter , d. , westerlund , b. e. , & steenman , h. c. 1994 , , 105 , 311 gallart , c. , zoccali , m. , & aparicio , a. 2005 , , 43 , 387 genzel , r. , et al . 2003 , , 594 , 812 ghez , a. m. , salim , s. , hornstein , s. d. , tanner , a. , lu , j. r. , morris , m. , becklin , e. e. , & duchne , g. 2005 , , 620 , 744 girardi , l. , bressan , a. , bertelli , g. , & chiosi , c. 2000 , , 141 , 371 gray , d. f. , & brown , k. 2001 , , 113 , 723 guesten , r. , genzel , r. , wright , m. c. h. , jaffe , d. t. , stutzki , j. , & harris , a. i. 1987 , , 318 , 124 hartung , m. , et al.2003 , , 4841 , 425 hopman , c. , & alexander , t. 2006 , , 645 , l133 jackson , j. m. , geis , n. , genzel , r. , harris , a. i. , madden , s. , poglitsch , a. , stacey , g. j. , & townes , c. h. 1993 , , 402 , 173 kent , s. m. 1992 , , 387 , 181 kleinmann , s. g. , & hall , d. n. b. 1986 , , 62 , 501 kormendy , j. , & kennicutt , r. c. , jr . 2004 , , 42 , 603 krabbe , a. , genzel , r. , drapatz , s. , & rotaciuc , v. 1991 , , 382 , l19 krabbe , a. , et al . 1995 , , 447 , l95 launhardt , r. , zylka , r. , & mezger , p. g. 2002 , , 384 , 112 levesque , e. m. , massey , p. , olsen , k. a. g. , plez , b. , josselin , e. , maeder , a. , & meynet , g. 2005 , , 628 , 973 lindqvist , m. , habing , h. j. , & winnberg , a. 1992 , , 259 , 118 malagnini , m. l. , & morossi , c. 1990 , , 85 , 1015 malagnini , m. l. , & morossi , c. 1997 , , 326 , 736 martins , f. , genzel , r. , hillier , d. j. , eisenhauer , f. , paumard , t. , gillessen , s. , ott , t. , & trippe , s. 2007 , , 468 , 233 mccaughrean , m. j. , close , l. m. , scholz , r .- d . , lenzen , r. , biller , b. , brandner , w. , hartung , m. , & lodieu , n. 2004 , , 413 , 1029 morris , m. , & serabyn , e. 1996 , , 34 , 645 narayanan , v. k. , gould , a. , & depoy , d. l. 1996 , , 472 , 183 nayakshin , s. , & sunyaev , r. 2005 , , 364 , l23 paumard , t. , et al . 2006 , , 643 , 1011 perrin , g. , coude du foresto , v. , ridgway , s. t. , mariotti , j .- m . , traub , w. a. , carleton , n. p. , & lacasse , m. g. 1998 , , 331 , 619 persson , s. e. , murphy , d. c. , krzeminski , w. , roth , m. , & rieke , m. j. 1998 , , 116 , 2475 philipp , s. , zylka , r. , mezger , p. g. , duschl , w. j. , herbst , t. , & tuffs , r. j. 1999 , , 348 , 768 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , cambridge : university press , |c1992 , 2nd ed . , ramirez , s. v. , depoy , d. l. , frogel , j. a. , sellgren , k. , & blum , r. d. 1997 , , 113 , 1411 ramrez , s. v. , sellgren , k. , carr , j. s. , balachandran , s. c. , blum , r. , terndrup , d. m. , & steed , a. 2000 , , 537 , 205 ramrez , i. , & melndez , j. 2005 , , 626 , 446 reid , m. j. 1993 , , 31 , 345 reid , n. & hawley , s. l. 2000 , new light on dark stars : red dwarfs , low mass stars , brown dwarfs / neill reid and suzanne l. hawley . new york : springer , 2000 . ( springer - praxis series in astronomy and astrophysics ) rieke , m. j. 1987 , nearly normal galaxies . from the planck time to the present , 90 rieke , m. j. 1999 , asp conf . ser . 186 : the central parsecs of the galaxy , 186 , 32 rousset , g. , et al . 2003 , , 4839 , 140 sadler , e. m. , rich , r. m. , & terndrup , d. m. 1996 , , 112 , 171 schoedel , r. , et al . 2007 , arxiv astrophysics e - prints , arxiv : astro - ph/0703178 schreiber , j. , thatte , n. , eisenhauer , f. , tecza , m. , abuter , r. , & horrobin , m. 2004 , asp conf . ser . 314 : astronomical data analysis software and systems ( adass ) xiii , 314 , 380 sellgren , k. , hall , d. n. b. , kleinmann , s. g. , & scoville , n. z. 1987 , , 317 , 881 serabyn , e. , keene , j. , lis , d. c. , & phillips , t. g. 1994 , , 424 , l95 sheehy , c. d. , mccrady , n. , & graham , j. r. 2006 , , 647 , 1517 sjouwerman , l. o. , habing , h. j. , lindqvist , m. , van langevelde , h. j. , & winnberg , a. 1999 , asp conf . ser . 186 : the central parsecs of the galaxy , 186 , 379 smalley , b. 1993 , , 265 , 1035 sokolov , n. a. 1998 , , 130 , 215 stolte , a. , brandner , w. , grebel , e. k. , lenzen , r. , & lagrange , a .- m . 2005 , , 628 , l113 taylor , b. j. 2003 , , 398 , 721 theodossiou , e. , & danezis , e. 1991 , , 183 , 91 vacca , w. d. , sheehy , c. d. , & graham , j. r. 2007 , arxiv astrophysics e - prints , arxiv : astro - ph/0701628 van dam , m. a. , le mignant , d. , & maccintosh , b. a. 2004 , appl opt . , 43 , 5458 van loon , j. t. , et al . 2003 , , 338 , 857 wallace , l. , & hinkle , k. 1997 , , 111 , 445 wood , p. r. , habing , h. j. , & mcgregor , p. j. 1998 , , 336 , 925 zoccali , m. , et al . 2003 , , 399 , 931 zoccali , m. , et al . 2006 , , 457 , l1 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: event shape variables are perhaps the most popular observables for testing qcd and for improving our understanding of its dynamics . event shape studies began in earnest towards the late seventies as a simple quantitative method to understand the nature of gluon bremsstrahlung @xcite . for instance , it was on the basis of comparisons to event shape data that one could first deduce that gluons were vector particles , since theoretical predictions employing scalar gluons did not agree with experiment @xcite . quite generally , event shapes parametrise the geometrical properties of the energy - momentum flow of an event and their values are therefore directly related to the appearance of an event in the detector . in other words the value of a given event shape encodes in a continuous fashion , for example , the transition from pencil - like two - jet events with hadron flow prominently distributed along some axis , to planar three - jet events or events with a spherical distribution of hadron momenta . thus they provide more detailed information on the final state geometry than say a jet finding algorithm which would always classify an event as having a certain finite number of jets , even if the actual energy flow is uniformly distributed in the detector and there is no prominent jet structure present in the first place . event shapes are well suited to testing qcd mainly because , by construction , they are collinear and infrared safe observables . this means that one can safely compute them in perturbation theory and use the predictions as a means of extracting the strong coupling @xmath1 . they are also well suited for determinations of other parameters of the theory such as constraining the quark and gluon colour factors as well as the qcd beta function . additionally they have also been used in other studies for characterising the final state , such as investigations of jet and heavy quark multiplicities as functions of event shape variables @xcite . aside from testing the basic properties of qcd , event shape distributions are a powerful probe of our more detailed knowledge of qcd dynamics . all the commonly studied event shape variables have the property that in the region where the event shape value is small , one is sensitive primarily to gluon emission that is soft compared to the hard scale of the event and/or collinear to one of the hard partons . such small - transverse - momentum emissions have relatively large emission probabilities ( compared to their high transverse momentum counterparts ) due to logarithmic soft - collinear dynamical enhancements as well as the larger value of their coupling to the hard partons . predictions for event shape distributions therefore typically contain large logarithms in the region where the event - shape is small , which are a reflection of the importance of multiple soft and/or collinear emission . a successful prediction for an event shape in this region requires all - order resummed perturbative predictions or the use of a monte - carlo event generator , which contains correctly the appropriate dynamics governing multiple particle production . comparisons of these predictions to experimental data are therefore a stringent test of the understanding qcd dynamics that has been reached so far . one other feature of event shapes , that at first appeared as an obstacle to their use in extracting the fundamental parameters of qcd , is the presence of significant non - perturbative effects in the form of power corrections that vary as an inverse power of the hard scale , @xmath2 . for most event shapes , phenomenologically it was found that @xmath3 ( as discussed in @xcite ) , and the resulting non - perturbative effects can be of comparable size to next - to - leading order perturbative predictions , as we shall discuss presently . the problem however , can be handled to a large extent by hadronisation models embedded in monte - carlo event generators @xcite , which model the conversion of partons into hadrons at the cost of introducing several parameters that need to be tuned to the data . once this is done , application of such hadronisation models leads to very successful comparisons of perturbative predictions with experimental data with , for example , values of @xmath1 consistent to those obtained from other methods and with relatively small errors . however since the mid - nineties attempts have also been made to obtain a better insight into the physics of hadronisation and power corrections in particular . theoretical models such as those based on renormalons have been developed to probe the non - perturbative domain that gives rise to power corrections ( for a review , see @xcite ) . since the power corrections are relatively large effects for event shape variables , scaling typically as @xmath4 ( no other class of observable shows such large effects ) , event shapes have become the most widely used means of investigating the validity of these ideas . in fact an entire phenomenology of event shapes has developed which is based on accounting for non - perturbative effects via such theoretically inspired models , and including them in fits to event shape data alongside the extraction of other standard qcd parameters . in this way event shapes also serve as a tool for understanding more quantitatively the role of confinement effects . the layout of this article is as follows . in the following section we list the definitions of several commonly studied event shapes in @xmath5 annihilation and deep inelastic scattering ( dis ) and discuss briefly some of their properties . then in section [ sec : perturbative ] we review the state of the art for perturbative predictions of event - shape mean values and distributions . in particular , we examine the need for resummed predictions for distributions and clarify the nomenclature and notation used in that context , as well as the problem of matching these predictions to fixed order computations . we then turn , in sections [ sec : exp ] and [ sec : distributions ] , to comparisons with experimental data and the issue of non - perturbative corrections , required in order to be able to apply parton level calculations to hadronic final state data . we discuss the various approaches used to estimate non - perturbative corrections , ranging from phenomenological hadronisation models to analytical approaches based on renormalons , and shape - functions . we also discuss methods where the standard perturbative results are modified by use of renormalisation group improvements or the use of ` dressed gluons ' . we study fits to event shape data , from different sources , in both @xmath5 annihilation and dis and compare the various approaches that are adopted to make theoretical predictions . in addition we display some of the results obtained by fitting data for various event shape variables for parameters such as the strong coupling , the qcd colour factors and the qcd beta function . lastly , in section [ sec : outlook ] , we present an outlook on possible future developments in the field . we list here the most widely studied event shapes , concentrating on those that have received significant experimental and theoretical attention . _ the _ canonical event shape is the thrust @xcite , @xmath6 : @xmath7 where the numerator is maximised over directions of the unit vector @xmath8 and the sum is over all final - state hadron momenta @xmath9 ( whose three - vectors are @xmath10 and energies @xmath11 ) . the resulting @xmath8 is known as the thrust axis . in the limit of two narrow back - to - back jets @xmath12 , while its minimum value of @xmath13 corresponds to events with a uniform distribution of momentum flow in all directions . the infrared and collinear safety of the thrust ( and other event shapes ) is an essential consequence of its linearity in momenta . often it is @xmath14 ( also called @xmath15 ) that is referred to insofar as it is this that vanishes in the @xmath16-jet limit . a number of other commonly studied event shapes are constructed employing the thrust axis . amongst these are the invariant squared jet - mass @xcite and the jet - broadening variables @xcite , defined respectively as @xmath17 where the plane perpendicular to the thrust axis is used to separate the event into left and right hemisphere , @xmath18 and @xmath19 . given these definitions one can study the heavy - jet mass , @xmath20 and wide - jet broadening , @xmath21 ; analogously one defines the light - jet mass ( @xmath22 ) and narrow - jet broadening @xmath23 ; finally one defines also the sum of jet masses @xmath24 and the total jet broadening @xmath25 and their differences @xmath26 , and @xmath27 . like @xmath14 , all these variables ( and those that follow ) vanish in the two - jet limit ; @xmath22 and @xmath23 are special in that they also vanish in the limit of three narrow jets ( three final - state partons ; or in general for events with any number of jets in the heavy hemisphere , but only one in the light hemisphere ) . another set of observables @xcite making use of the thrust axis starts with the thrust major @xmath28 , @xmath29 where the maximisation is performed over all directions of the unit vector @xmath30 , such that @xmath31 . the thrust minor , @xmath32 , is given by @xmath33 and is sometimes also known @xcite as the ( normalised ) out - of - plane momentum @xmath34 . like @xmath22 and @xmath23 , it vanishes in the three - jet limit ( and , in general , for planar events ) . finally from @xmath28 and @xmath32 one constructs the oblateness , @xmath35 . an alternative axis is used for the spherocity @xmath36 @xcite , @xmath37 while there exist reliable ( though algorithmically slow ) methods of determining the thrust and thrust minor axes , the general properties of the spherocity axis are less well understood and consequently the spherocity has received less theoretical attention . an observable similar to the thrust minor ( in that it measures out - of - plane momentum ) , but defined in a manner analogous to the spherocity is the acoplanarity @xcite , which minimises a ( squared ) projection perpendicular to a plane . it is also possible to define event shapes without reference to an explicit axis . the best known examples are the @xmath38 and @xmath39 parameters @xcite which are obtained from the momentum tensor @xcite . because of the quadratic dependence on particle momenta , these observable are collinear unsafe and so no longer widely studied . ] @xmath40 where @xmath41 is the @xmath42 component of the three vector @xmath43 . in terms of the eigenvalues @xmath44 , @xmath45 and @xmath46 of @xmath47 , the @xmath38 and @xmath39 parameters are given by @xmath48 the @xmath38-parameter is related also to one of a series of fox - wolfram observables @xcite , @xmath49 , and is sometimes equivalently written as @xmath50 the @xmath39-parameter , like the thrust minor , vanishes for all final - states with up to 3 particles , and in general , for planar events . the @xmath38-parameter can actually be considered ( in the limit of all particles being massless ) as the integral of a more differential observable , the energy - energy correlation ( eec ) @xcite , @xmath51 where the average in eq . ( [ eq : sigmachi ] ) is carried out over all events . another set of variables that characterise the shape of the final state are the @xmath52-jet resolution parameters that are generated by jet finding algorithms . examples of these are the jade @xcite and durham @xcite jet clustering algorithms . one introduces distance measures @xmath53 @xmath54 for each pair of particles @xmath55 and @xmath56 . the pair with the smallest @xmath53 is clustered ( by adding the four - momenta the @xmath57 recombination scheme ) and replaced with a single pseudo - particle ; the @xmath53 are recalculated and the combination procedure repeated until all remaining @xmath53 are larger than some value @xmath58 . the event - shapes based on these jet algorithms are @xmath59 , defined as the maximum value of @xmath58 for which the event is clustered to 3 jets ; analogously one can define @xmath60 , and so forth . other clustering jet algorithms exist . most differ essentially in the definition of the distance measure @xmath53 and the recombination procedure . for example , there are e0 , p and p0 variants @xcite of the jade algorithm , which differ in the details of the treatment of the difference between energy and the modulus of the 3-momentum . the geneva algorithm @xcite is like the jade algorithm except that in the definition of the @xmath53 it is @xmath61 that appears in the denominator instead of the total energy . an algorithm that has been developed and adopted recently is the cambridge algorithm @xcite , which uses the same distance measure as the durham algorithm , but with a different clustering sequence . we note that a number of variants of the above observables have recently been introduced @xcite , which differ in their treatment of massive particles . these include the @xmath62-scheme where all occurrences of @xmath11 are replaced by @xmath63 and the @xmath57-scheme where each 3-momentum @xmath43 is rescaled by @xmath64 . the former leads to a difference between the total energy in the initial and final states , while the latter leads to a final - state with potentially non - zero overall @xmath65-momentum . while such schemes do have these small drawbacks , for certain observables , notably the jet masses , which are quite sensitive to the masses of the hadrons ( seldom identified in experimental event - shape studies ) , they tend to be both theoretically and experimentally cleaner . ref . @xcite also introduced a ` decay ' scheme ( an alternative is given in @xcite ) where all hadrons are artificially decayed to massless particles . since a decay is by definition a stochastic process , this does not give a unique result on an event - by - event basis , but should rather be understood as providing a correction factor which is to be averaged over a large ensemble of events . it is to be kept in mind that these different schemes all lead to identical perturbative predictions ( with massless quarks ) and differ only at the non - perturbative level . a final question relating to event - shape definitions concerns the hadron level at which measurements are made . since shorter lived hadrons decay during the time of flight , one has to specify whether measurements were made at a stage before or after a given species of hadron decays . it is important therefore when experimental results are quoted , that they should specify which particles have been taken to be stable and which have not . as well as the @xmath66 variables discussed above , it is also possible to define , by analogy , event shapes in dis . the frame in which dis event shapes can be made to most closely resemble those of @xmath66 is the breit frame @xcite . this is the frame in which @xmath67 , where @xmath68 is the incoming proton momentum and @xmath69 the virtual photon momentum . one defines two hemispheres , separated by the plane normal to the photon direction : the remnant hemisphere ( @xmath70 , containing the proton remnant ) , and the current hemisphere ( @xmath71 ) . at the level of the quark - parton model , @xmath71 is like one hemisphere of @xmath0 and it is therefore natural to define event shapes using only the momenta in this hemisphere . in contrast , any observable involving momenta in the remnant hemisphere must take care to limit its sensitivity to the proton remnant , whose fragmentation can not be reliably handled within perturbation theory . a possible alternative to studying just the current hemisphere , is to take all particles except those in a small cone around the proton direction @xcite . a feature that arises in dis is that there are two natural choices of axis . for example , for the thrust @xmath72 one can either choose the unit vector @xmath73 to be the photon ( @xmath74 ) axis , @xmath75 , or one can choose it to be the true thrust axis , that which maximises the sum , giving @xmath76 . similarly one defines two variants of the jet broadening , @xmath77 for the jet - mass and @xmath38-parameter the choice of axis does not enter into the definitions and we have @xmath78 all the above observables can also be defined with an alternative normalisation , @xmath79 replacing @xmath80 , in which case they are named @xmath81 and so on . we note that @xmath81 was originally proposed in @xcite . for a reader used to @xmath0 event shapes the two normalisations might at first sight seem equivalent however when considering a single hemisphere , as in dis , the equivalence is lost , and indeed there are even events in which the current hemisphere is empty . this is a problem for observables normalised to the sum of energies in @xmath71 , to the extent that to ensure infrared safety it is necessary to exclude all events in which the energy present in the current hemisphere is smaller than some not too small fraction of @xmath82 ( see for example @xcite ) . additionally there are studies of variables that vanish in the 2 + 1-parton limit for dis , in particular the @xmath34 defined in analogy with the thrust minor of @xmath0 , with the thrust axis replaced by the photon axis @xcite , or an azimuthal correlation observable @xcite . rather than being examined just in @xmath71 , these observables use particles also in the remnant hemisphere ( for @xmath34 , all except those in a small cone around the proton ) . as in @xmath0 , jet rates are studied also in dis . unlike the event - shapes described above , the jet - shapes make use of the momenta in both hemispheres of the breit frame . their definitions are quite similar to those of @xmath0 except that a clustering to the proton remnant ( beam jet ) is also included @xcite . though only @xmath0 and dis are within the scope of this review , we take the opportunity here to note that related observables are being considered also for other processes . notably for drell - yan production an out - of - plane momentum measurement has been proposed in @xcite and various thrust and thrust - minor type observables have been considered in hadron - hadron dijet production in @xcite . the observables discussed above are all infrared and collinear safe they do not change their value when an extra soft gluon is added or if a parton is split into two collinear partons . as emerges from the original discussion of sterman and weinberg @xcite , this is a necessary condition for the cancellation of real and virtual divergences associated with such emissions , and therefore for making finite perturbative predictions . for an event shape that vanishes in the @xmath52-jet limit ( which we shall generically refer as an @xmath52-jet observable ) , the leading perturbative contribution is of order @xmath83 . for example the thrust distribution in @xmath0 is given by @xmath84 + { { \cal o}\left({\alpha_s}^2\right ) } \nonumber\end{aligned}\ ] ] ( see e.g. @xcite ) . when calculating perturbative predictions for mean values of event shapes , as well as higher moments of their distributions , one has integrals of the form @xmath85 we note that in general , fixed - order event - shape distributions diverge in the limit as @xmath86 goes to @xmath87 , cf . ( [ eq : thrustdist ] ) for @xmath88 . in eq . ( [ eq : meanv ] ) , the weighting with a power of @xmath86 is sufficient to render the singularity integrable , and the integral is dominated by large @xmath86 ( and so large transverse momenta of order @xmath82 , the hard scale ) . this dominance of a single scale ensures that the coefficients in the perturbative expansion for @xmath89 are well - behaved ( for example they are free of any enhancements associated with logarithms of ratios of disparate scales ) . beyond leading order , perturbative calculations involve complex cancellations between soft and collinear real and virtual contributions and are nearly always left to general - purpose `` fixed - order monte carlo '' programs . the current state of the art is next - to - leading order ( nlo ) , and available programs include : for @xmath0 to 3 jets , event @xcite , eerad @xcite and event2 @xcite ; for @xmath0 to 4 jets , menlo parc @xcite , mercutio @xcite and eerad2 @xcite ; for dis to @xmath90 jets , mepjet @xcite , disent @xcite and disaster++ @xcite ; and for all the above processes and additionally dis to @xmath91 jets and various hadron - hadron and photo - production processes , nlojet++ @xcite ( for photo - production , there is additionally jetvip @xcite ) . in recent years much progress has been made towards nnlo calculations , though complete results remain to be obtained ( for a review of the current situation , see @xcite ) . for the event shape distributions themselves , however , fixed order perturbative estimates are only of use away from the @xmath92 region , due to the singular behaviour of the fixed order coefficients in the @xmath93 limit . in this limit , at higher orders each power of @xmath94 is accompanied by a coefficient which grows as @xmath95 . these problems arise because when @xmath86 is small one places a restriction on real emissions without any corresponding restriction on virtual contributions the resulting incompleteness of cancellations between logarithmically divergent real and virtual contributions is the origin of the order - by - order logarithmic enhancement of the perturbative contributions . to obtain a meaningful answer it is therefore necessary to perform an all - orders _ resummation _ of logarithmically enhanced terms . when discussing resummations it is convenient to refer to the integrated distribution , @xmath96 quite generally @xmath97 has a perturbative expansion of the form @xmath98 one convention is to refer to all terms @xmath99 as leading logarithms ( ll ) , terms @xmath100 as next - to - leading logarithms ( nll ) , etc . , and within this hierarchy a resummation may account for all ll terms , or all ll and nll terms and so forth . such a resummation gives a convergent answer up to values of @xmath101 , beyond which terms that are formally subleading can become as important as the leading terms ( for example if @xmath102 , then the nnll term @xmath103 is of the same order as the ll term @xmath104 ) . at its limit of validity , @xmath105 , a n@xmath106ll resummation neglects terms of relative accuracy @xmath107 . an important point though is that for nearly all observables that have been resummed there exists a property of exponentiation : @xmath108 in some cases @xmath109 is written in terms of a sum of such exponentiated contributions ; in certain other cases the exponentiation holds only for a suitable integral ( e.g. fourier ) transform of the observable . the fundamental point of exponentiation is that the inner sum , over @xmath110 , now runs only up to @xmath111 , instead of @xmath112 as was the case in eq . ( [ eq : dumbsum ] ) . with the exponentiated form for the resummation , the nomenclature `` ( next - to)-leading - logarithmic '' acquires a different meaning n@xmath106ll now refers to all terms in the exponent @xmath113 . to distinguish the two classification schemes we refer to them as n@xmath106ll@xmath114 and n@xmath106ll@xmath115 . the crucial difference between n@xmath106ll@xmath114 and n@xmath106ll@xmath115 resummations lies in the range of validity and their accuracy . a n@xmath106ll@xmath115 resummation remains convergent considerably further in @xmath116 , up to @xmath102 ( corresponding usually to the peak of the distribution @xmath117 ) ; at this limit , neglected terms are of relative order @xmath118 . consequently a n@xmath106ll@xmath115 resummation includes not only all n@xmath106ll@xmath114 terms , but also considerably more . from now when we use the term n@xmath106ll it should be taken to mean n@xmath106ll@xmath119 . two main ingredients are involved in the resummations as shown in eq . ( [ eq : expresum ] ) . firstly one needs to find a way of writing the observable as a factorised expression of terms for individual emissions . this is often achieved with the aid of one or more integral ( mellin , fourier ) transforms . secondly one approximates the multi - emission matrix element as a product of individual independent emission factors . it is to be kept in mind that not all observables exponentiate . the jade jet resolution is the best known such example @xcite ( and it has not yet been resummed ) . the general criteria for exponentiation are discussed in @xcite . resummed results to nll accuracy exist for a number of @xmath16-jet observables in @xmath120 jets : the thrust @xcite , the heavy - jet mass @xcite and the single - jet and light - jet masses @xcite , jet broadenings @xcite , @xmath38-parameter @xcite , durham and cambridge jet resolutions @xcite , the thrust major and oblateness @xcite , eec @xcite and event - shape / energy - flow correlations @xcite ; for @xmath65-jet observables in @xmath121 jet events we have the thrust minor @xcite and the @xmath39-parameter @xcite . for dis to @xmath122 jets the current - hemisphere observables are resummed in refs . @xcite and the jet rates in @xcite ( though only to nll@xmath114 accuracy ) while the @xmath90 jet observables , @xmath34 and azimuthal correlations , have been resummed in @xcite . methods for automated resummation of arbitrary observables are currently in development for a range of processes@xcite , and techniques have also being developed for dealing with arbitrary processes @xcite . we also note that for the @xmath0 thrust and heavy - jet mass there have been investigations of certain classes of corrections beyond nll accuracy @xcite . the above resummations all apply to events with only light quarks . investigations to nll@xmath123 accuracy have also been performed for @xmath0 jet rates with heavy quarks @xcite . the above resummations are all for @xmath52-jet observables in the @xmath52-jet limit . for some @xmath65-jet observables it is useful to have resummations in the @xmath16-jet limit ( because it is here that most of the data lie ) , though currently little attention has been devoted to such resummations , with just nll@xmath123 results for jet - resolution parameters @xcite and nll@xmath115 results for the light - jet mass and narrow - jet broadening @xcite . furthermore in addition to the resummation of the @xmath38-parameter for @xmath124 there exists a ll resummation @xcite ( that could straightforward be extended to nll ) of a ` shoulder ' structure at @xmath125 ( related to a step function at lo ) , which corresponds to symmetric three jet events . some of the observables ( for example many single - jet observables ) have the property that they are sensitive to emissions only in some of the phase space . these are referred to as non - global observables and nll resummed predictions for them require that one account for coherent ensembles of energy - ordered large - angle gluons . this has so far been done only in the large-@xmath126 limit @xcite ( for reviews see @xcite ) , though progress is being made in extending this to finite @xmath126 @xcite . at @xmath127 . the resummed distribution has been matched ( @xmath128 scheme , see below ) so as to include the full lo and nlo contributions . the inset differs from the main plot only in the scales.,scaledwidth=60.0% ] to illustrate the impact of resummations we show in figure [ fig : thrlonlo ] the lo , nlo and resummed results for the thrust distribution . not only is the divergence in the lo distribution clearly visible , but one also observes a marked difference in the qualitative behaviour of the lo and nlo results , both of which show large differences relative to the nll resummed prediction at small @xmath14 . for the practical use of resummed results an important step is that of matching with the fixed - order prediction . at the simplest level one may think of matching as simply adding the fixed - order and resummed predictions while subtracting out the doubly counted ( logarithmic ) terms . such a procedure is however too naive in that the fixed - order contribution generally contains terms that are subleading with respect to the terms included in the resummation , but which still diverge . for example when matching nll and nlo calculations , one finds that the nlo result has a term @xmath129 , which in the distribution diverges as @xmath130 . this is unphysical and the matching procedure must be sufficiently sophisticated so as to avoid this problem , and in fact it should nearly always ensure that the matched distribution goes to zero for @xmath131 . several procedures exist , such as @xmath128 and @xmath109 matching @xcite and multiplicative matching @xcite and they differ from one - another only in terms that are nnlo and nnll , i.e. formally beyond the state - of - the - art accuracy . we note the matching generally ensures that the resummation , in addition to being correct at nll accuracy ( our shorthand for nll@xmath115 ) , is also correct at nnll@xmath123 accuracy . matching also involves certain other subtleties , for example the ` modification of the logarithm ' , whereby @xmath132 is replaced with @xmath133 , where @xmath134 is the largest kinematically allowed value for the observable . this ensures that the logarithms go to zero at @xmath134 , rather than at some arbitrary point ( usually @xmath135 ) , which is necessary in order for the resulting matched distribution to vanish at @xmath134 . [ sec : means ] [ sec : meanpc ] much of the interest in event shapes stems from the wealth of data that is available , covering a large range of centre of mass energies ( @xmath0 ) or photon virtualities ( dis ) . the @xmath0 data comes from the pre - lep experiments @xcite , the four lep experiments @xcite @xcite @xcite and sld @xcite . in addition , because many observables have been proposed only in the last ten or so years and owing to the particular interest ( see below ) in data at moderate centre of mass energies , the jade data have been re - analysed @xcite . have also been obtained from lep 1 @xcite by considering events with an isolated hard final - state photon and treating them as if they were pure qcd events whose centre of mass energy is that of the recoiling hadronic system . such a procedure is untrustworthy because it assumes that one can factorise the gluon production from the photon production . this is only the case when there is strong ordering in transverse momenta between the photon and gluon , and is not therefore applicable when both the photon and the gluon are hard . tests with monte carlo event generators which may indicate that any ` non - factorisation ' is small @xcite are not reliable , because the event generator used almost certainly does not contain the full 1-photon , 1-gluon matrix element . it is also to be noted that the isolation cuts on the photon will bias the distribution of the event shapes , in some cases @xcite similarly to an event - shape energy - flow correlation @xcite . therefore we would argue that before relying on these data , one should at the very least compare the factorisation approximation with exact lo calculations , which can be straightforwardly obtained using packages such as grace @xcite , comphep @xcite or amegic++ @xcite . ] results in dis come from both h1 @xcite ( mean values and distributions ) and zeus @xcite ( means ) , @xcite ( means and distributions in rapidity gap events ) . thrust , compared to lo , nlo and nlo+@xmath136 ; right : data for the durham jet resolution @xmath137 compared to the nlo prediction ( figure taken from @xcite).,title="fig:",scaledwidth=48.0% ] thrust , compared to lo , nlo and nlo+@xmath136 ; right : data for the durham jet resolution @xmath137 compared to the nlo prediction ( figure taken from @xcite).,title="fig:",scaledwidth=48.0% ] let us start our discussion of the data by examining mean values ( distributions are left until section [ sec : distributions ] ) . as we have already mentioned , one of the most appealing properties of event shapes , in terms of testing qcd , is the fact that they are calculable in perturbation theory and so provide a direct method for the extraction of @xmath1 as well as testing other parameters of the theory such as the colour factors @xmath138 and @xmath139 by using a wealth of available experimental data . however one obstruction to the clean extraction of these fundamental parameters is the presence in many cases , of significant non - perturbative effects that typically fall as inverse powers @xmath140 of @xmath82 , the hard scale in the reaction ( the centre - of mass energy in @xmath5 annihilation and the momentum transfer in dis ) . the importance of such _ power corrections _ varies from observable to observable . for example , it can be seen from fig . [ fig : meandata ] ( left ) that the data for the mean value of the thrust variable need a significant component @xmath141 in addition to the lo and nlo fixed order perturbative estimates , in order to be described . on the other hand the comparison ( right ) for the mean value of the durham jet resolution parameter @xmath137 with the nlo prediction alone is satisfactory , without the need for any substantial power correction term . the problem of non - perturbative corrections is so fundamental in event - shape studies that it is worthwhile giving a brief overview of the main approaches that are used . no statement can be made with standard perturbative methods about the size of these power corrections and hence the earliest methods adopted to quantify them were phenomenological hadronisation models embedded in monte carlo event generators @xcite . one of the main issues involved in using such hadronisation models is the existence of several adjustable parameters . this essentially means that a satisfactory description of data can be obtained by tuning the parameters , which does not allow much insight into the physical origin of such power behaved terms which are themselves of intrinsic theoretical interest ( for a more detailed critique of hadronisation models see @xcite ) . however since the mid - nineties there have been developments that have had a significant impact on the theoretical understanding of power corrections . perhaps the most popular method of estimating power corrections is based on the renormalon model . in this approach one examines high - order terms @xmath142 of the perturbative series that are enhanced by coefficients @xmath143 , where @xmath144 is the first coefficient of the @xmath145 function . the factorial divergence leads to an ambiguity in the sum of the series of order @xmath2 , where the value of @xmath62 depends on the speed with which the high - order terms diverge . this approach has seen far more applications than can possibly be described here and for a full discussion and further references on renormalons the reader is referred to @xcite . the most extensive phenomenological applications of renormalon based ideas have been in the estimation of power corrections for event - shape variables . the reason is that event shapes have much larger ( and so experimentally more visible ) power corrections ( typically @xmath146 ) than most other observables ( typically @xmath147 or smaller ) . for event shape variables in @xmath5 annihilation in the two - jet limit , renormalon - inspired ( or related ) studies of power corrections have been carried out in refs . for the @xmath122 jet limit of dis , results on power corrections based on the renormalon approach can be found in @xcite . studies in the @xmath65-jet limit have been given in @xcite . the approach that has been most commonly used in comparing theoretical predictions with experimental data is that initiated by dokshitzer and webber @xcite . here the full result for the mean value , @xmath148 , is given by @xmath149 where @xmath150 is the perturbative prediction for the mean value . the hadronisation correction is included through the term @xmath151 , where @xmath62 is related to the speed of divergence of the renormalon series ( as discussed above ) , @xmath152 is an observable - dependent ( calculable ) coefficient and @xmath153 is a non - perturbative factor , scaling as @xmath154 which is hypothesised to be common across a whole class of observables with the same value of @xmath62 ( strictly speaking the full story is a little more complicated , see e.g. @xcite ) . most event shapes have @xmath3 , implying a leading power correction scaling as @xmath146 , essentially a consequence of the observables linearity on soft momenta . for these observables with @xmath3 the form for @xmath155 that has become standard is @xcite , @xmath156 where @xmath157 , @xmath158 and @xmath159 . in eq . ( [ eq : cpfin ] ) an arbitrary _ infrared matching scale _ @xmath160 has been introduced , intended to separate the perturbative and non - perturbative regions . it is usually taken to be @xmath16 gev ( and for systematic error estimates it is varied between @xmath161 and @xmath65 gev ) . the only truly non - perturbative ingredient in eq . ( [ eq : cpfin ] ) is @xmath162 , which can be interpreted as the average value of an infrared finite strong coupling for scales below @xmath160 ( such a concept was first applied phenomenologically in @xcite ) . though one could imagine estimating it from lattice studies of the coupling ( such as @xcite ) , one should keep in mind that the coupling in the infrared is not a uniquely defined quantity . in practice the phenomenological test of the renormalon approach will be that @xmath163 has a consistent value across all observables ( in the @xmath3 class ) . the terms with negative sign in the parenthesis in eq . ( [ eq : cpfin ] ) are a consequence of merging standard nlo perturbative results , which include the small spurious contributions from the infrared region ( scales up to @xmath160 ) , with the non - perturbative power correction that accounts correctly for scales from zero to @xmath160 . carrying out the subtraction of the perturbative terms that arise from the infrared region , below @xmath160 , amounts to inclusion of the negative sign terms in parenthesis above . note that this subtraction procedure must be carried out to the level of accuracy of the corresponding perturbative estimate . in other words if a perturbative estimate becomes available at @xmath164 then an additional subtraction term proportional to @xmath165 is required . with the subtraction procedure carried out to @xmath166 , as above , one expects a residual @xmath160 dependence @xmath167 . the scale @xmath168 is the renormalisation scale , which should be taken of the same order as the hard scale @xmath82 . the ` milan factor ' , @xmath169 in eq . ( [ eq : cpfin ] ) , is @xcite @xmath170 where the numerical result is given for @xmath171 , since only light flavours will be active at the relevant ( low ) scales . it accounts for the fact that the usual , ` naive ' , calculation for @xmath152 is carried out on the basis of a decaying virtual gluon ( i.e. cutting a bubble in the chain of vacuum polarisation insertions that lead to the running of the coupling ) , but without fully accounting for the non - inclusiveness of that decay when dealing with the observable . it was pointed out in @xcite that this is inconsistent and full ( two - loop ) calculations revealed @xcite that if the ` naive ' @xmath152 coefficient is calculated in an appropriate scheme , then the factor @xmath169 comes out to be universal . as discussed in @xcite this is essentially a consequence of the fact that regardless of whether one accounts for the virtual gluon decay , the distribution of ` non - perturbative ' transverse momentum is independent of rapidity ( the ` tube - model ' of @xcite , based essentially on boost invariance ) . while the factor @xmath169 corrects the @xmath152 at two - loop level , the question of even higher order corrections is still open , although one can argue that such effects will be suppressed by a factor @xmath172 relative to the leading power correction . this argument relies on the hope that the strong coupling remains moderate even in the infrared region . .coefficients of @xmath4 power corrections for @xmath5 event shapes ; @xmath173 . [ cols="^,^,^,^,^,^,^",options="header " , ] tables 1 and 2 display the values of the @xmath152 coefficients obtained in the 2-jet @xmath5 and and @xmath122-jet dis cases respectively . note the different behaviour , proportional to @xmath174 , that arises in the case of the jet broadenings in both processes , a consequence of the fact that the broadening is more sensitive to recoil induced by perturbative radiation @xcite . the improved theoretical understanding that led to this prediction was strongly stimulated by experimental analysis @xcite of features of earlier predictions @xcite that were inconsistent with the data . since the origin of the @xmath174 enhancement is a perturbative , non - perturbative interplay best explained with reference to the power correction for the distribution , we delay its discussion to section [ sec : anldist ] . another consequence of this interplay of the broadening power correction with perturbative radiation is the fact that the dis broadening @xmath175 has a bjorken @xmath176 dependent term corresponding to dglap evolution of the parton densities , as indicated by the term proportional to @xmath177 @xcite , where @xmath178 denotes a parton density function . the coefficients for the other dis variables are @xmath176 independent since in those cases the power correction arises from soft emission alone . we also mention the result for the power correction to another interesting variable , the energy - energy correlation ( eec ) , in the back - to - back region . like the broadening this variable exhibits the impact of perturbative - non perturbative interplay . the results are fractional power corrections that vary as @xmath179 ( for the quark - gluon correlation ) and @xmath180 for the hard quark - antiquark correlation @xcite , instead of @xmath4 and @xmath181 contributions respectively , that would be obtained by considering np emission without the presence of harder perturbative emissions . observables for which the power corrections have yet to be understood include the jet resolution parameters . renormalon based predictions were given in @xcite suggesting a @xmath146 correction for @xmath182 and @xmath147 for @xmath183 . however it seems that in both cases there could be significant perturbative non - perturbative interplay which will complicate the picture . nevertheless , as we have seen in fig . [ fig : meandata ] ( right ) any correction to @xmath184 is certainly small compared to that for other observables . a general point to be kept in mind is that the universality of @xmath163 can be broken by contributions of order @xmath185 , where @xmath186 is the mass scale for hadrons @xcite . whether this happens or not depends on the hadron mass scheme in which the observable is defined ( cf . section [ sec : definitions ] ) . most observables are implicitly in the @xmath62-scheme , which generally involves a small negative breaking of universality ( which is almost observable - independent , so that an illusion of universality will persist ) . the jet masses are an exception and in their usual definitions they have a significant ( positive ) universality breaking component . the @xmath57-scheme is free of universality breaking terms , as is the decay scheme ( since hadron masses are zero ) . we also note that the power correction associated with hadron mass effects is enhanced by an anomalous dimension , @xmath187 with @xmath188 . the normal ` renormalon ' power correction would also be expected to have an anomalous dimension , but it has yet to be calculated . as well as studies of mean values ( first moments ) of observables there has also been work on higher - moments , @xmath189 with @xmath190 . simple renormalon - inspired arguments suggest that the @xmath191 moments of event shapes should have as their leading power correction at most a @xmath192 contribution ( essentially since @xmath193 vanishes at @xmath194 for small @xmath195 ) . however as was pointed out by webber @xcite , the fact ( to be discussed in section [ sec : anldist ] ) that the @xmath146 power correction essentially corresponds to a shift of the distribution ( for @xmath6 , @xmath38 and @xmath196 , but not for the broadenings ) , means that to a first approximation we will have @xmath197 given that @xmath198 is of order @xmath94 , all higher moments of event shapes will receive power corrections of order @xmath199 , which is parametrically larger than @xmath192 . strictly speaking it is not possible to guarantee the coefficient of @xmath199 given in eq . ( [ eq : momentpc ] ) , since the shift approximation on which it is based holds only in the @xmath16-jet region , whereas the dominant contribution to the relevant integral comes from the @xmath65-jet region . for the @xmath38-parameter an alternative coefficient for the @xmath199 power correction has been proposed in @xcite , however since it also is not based on a full calculation of the power correction in the three - jet region ( which does not yet exist ) , it is subject to precisely the same reservations as eq . ( [ eq : momentpc ] ) . we note also an interesting result by gardi @xcite regarding the exact renormalon analysis of @xmath200 . while it is clear that the physical answer @xmath199 will not appear in a leading renormalon analysis ( the extra factor of @xmath94 means that it is associated with a subleading renormalon ) the calculation @xcite has the surprising result that the leading renormalon contribution is not @xmath147 as naively expected but rather @xmath201 . considerable progress has been made in recent years in the calculation of power corrections for three - jet observables ( those that vanish in the two and the three - jet limits ) . in the three - jet limit , there exist explicit results for the thrust minor @xcite and the @xmath39-parameter @xcite in @xmath202 , the out of plane momentum in drell - yan plus jet @xcite and in @xmath90-jet dis @xcite , as well as azimuthal correlations in dis @xcite . except for the @xmath39-parameter , all these calculations involve perturbative , non - perturbative interplay in a manner similar to that of the broadenings and back - to - back eec . the explicit forms for the results are rather complicated and so we refer the reader to the original publications . simpler ( though numerical ) results have been obtained @xcite in the case of the @xmath39-parameter ( confirmed also by @xcite ) integrated over all 3-jet configurations , where one finds @xmath203 the power of @xmath94 comes from the matrix - element weighting of the @xmath65-jet configurations and the fact that the @xmath39-parameter vanishes in the @xmath16-jet limit . a final @xmath65-jet result of interest @xcite is that for the @xmath38-parameter just above the sudakov shoulder , at @xmath125 . this is the only case where a proper calculation exists for the power correction to a two - jet observable in the three jet limit : @xmath204 this is slightly less than half the correction that appears in the two - jet limit ( @xmath205 ) . a point that emerges clearly from eqs . ( [ eq : dpower ] ) and ( [ eq : c34power ] ) , but that is relevant for all @xmath65-jet studies is that the power correction acquires a dependence on @xmath139 , i.e. there is sensitivity to hadronisation from a gluon . in situations where one selects only three - jet events there is additionally non trivial dependence on the geometry , due to the coherence between the three jets . comparisons to data for such observables would therefore allow a powerful test of the renormalon - inspired picture . in particular , other models , such as the flux tube model @xcite ( based essentially on boost invariance along the @xmath206 axis ) , which in the @xmath16-jet limit give the same predictions as renormalon - inspired approaches , can not as naturally be extended to the @xmath65-jet case . one of the most widespread ways of testing the dokshitzer - webber approach to hadronisation is to carry out simultaneous fits to the mean value data for @xmath1 and @xmath163 . figure [ fig : meandata ] , where the term labelled @xmath4 is actually of the form @xmath207 , shows the good agreement that is obtained with the thrust data . the quality of agreement is similar for other observables ( see for example fig . 11 of @xcite ) . the true test of the approach however lies in a verification of the universality hypothesis , namely that @xmath163 is the same for all observables and processes . , one may wonder why one also fits for @xmath1 . the two principal reasons are ( a ) that one is in any case interested in the value of @xmath94 and ( b ) that the data and perturbative prediction differ also by higher - order terms in @xmath94 and fixing @xmath94 would mean trying to fit these higher - order terms with a power correction , which would be misleading . ] while there are strong general reasons for expecting universality within @xmath0 ( boost invariance and the flux - tube model ) , universality across difference processes is less trivial for example one could imagine dis hadronisation being modified by interactions between outgoing low - momentum gluons ( or hadrons ) and the ` cloud ' of partons that make up the proton remnant ( present even for a fast - moving proton , since the longitudinal size of the cloud is always of order @xmath208 ) . figure [ fig : meanfits ] shows @xmath161-@xmath209 contours for fits of @xmath1 and @xmath163 in @xmath0 @xcite ( left , all experiments combined ) and dis ( right , h1 @xcite and zeus @xcite merged into one plot ) . aside from the different scales used , some care is required in reading the figures because of different treatments of the errors in the different plots . for example h1 include experimental systematic errors whereas zeus do not . in the @xmath0 fit , the systematics are included but treated as uncorrelated , which may underestimate the final error . in neither figure are theoretical uncertainties included : over the available range of @xmath0 energies , renormalisation scale dependence gives an uncertainty of about @xmath210 on @xmath94 , while a ` canonical ' variation of the milan factor @xmath211 by @xmath212 ( to allow for higher - order terms ) leads to an uncertainty of @xmath213 on @xmath163 @xcite . in dis the corresponding uncertainties are larger for @xmath94 ( @xmath214 , essentially because the fits are dominated by lower @xmath82 values , and they vary substantially for @xmath163 ( larger for photon - axis observables , smaller for the others ) . also to be kept in mind is that the dis figure includes the jet mass , @xmath215 in the default scheme , and in @xmath0 the origins of the arrows indicate the default - scheme results for @xmath216 and @xmath196 . since the default scheme for jet masses breaks universality ( cf . section [ sec : dwdetails ] ) these results should not be compared directly to those for other observables . , @xmath163 fit to @xmath0 data on mean values of event shapes , with the jet masses converted to the @xmath62-scheme ( the arrow indicates the change in going from default to @xmath62-scheme ) @xcite . right : fits to dis mean event - shape data by h1 @xcite and zeus @xcite ( curves taken from figures in @xcite ) . contours indicate 1-@xmath209 uncertainties ( statistical and experimental systematic , except for the zeus results which are just statistical).,title="fig:",scaledwidth=48.0% ] , @xmath163 fit to @xmath0 data on mean values of event shapes , with the jet masses converted to the @xmath62-scheme ( the arrow indicates the change in going from default to @xmath62-scheme ) @xcite . right : fits to dis mean event - shape data by h1 @xcite and zeus @xcite ( curves taken from figures in @xcite ) . contours indicate 1-@xmath209 uncertainties ( statistical and experimental systematic , except for the zeus results which are just statistical).,title="fig:",scaledwidth=48.0% ] taking into account the uncertainties , which are not necessarily correlated from one observable to another , fig . [ fig : meanfits ] indicates remarkable success for the renormalon - inspired picture . the results are in general consistent with a universal value for @xmath163 in the range @xmath217 to @xmath218 . furthermore the @xmath0 results for @xmath94 are in good agreement with the world average . on the other hand the dis results for @xmath94 seem to be somewhat larger than the world average . the discrepancy is ( just ) within the theoretical systematic errors , however it remains a little disturbing and one wonders whether it could be indicative of large higher - order corrections , or some other problem ( known issues include the fact that the low-@xmath82 data can bias the fits , for example because of heavy - quark effects and @xmath219 corrections ; and also at high-@xmath82 there is @xmath220 interference which is not usually accounted for in the fixed - order calculations ) . there is potentially also some worry about the @xmath221 result which has an anomalously low @xmath163 and large @xmath94 ( and additionally shows unexpectedly substantial @xmath176 dependence @xcite ) . another point relates to the default scheme jet masses though the universality - breaking effects should be purely non - perturbative , they have an effect also on the fit results for @xmath1 . this is a consequence of the anomalous dimension that accompanies @xmath146 hadron mass effects . similar variations of @xmath94 are seen when varying the set of particles that are considered stable @xcite . though as discussed above some issues remain , they should not however be seen as detracting significantly from the overall success of the approach and the general consistency between @xmath0 and dis . we also note that the first moment of the coupling as extracted from studies of heavy - quark fragmentation @xcite is quite similar to the value found for event shapes . there have also been experimental measurements of higher moments of event shapes , notably in @xcite . the parameters @xmath1 and @xmath163 are fixed from fits to mean values and then inserted into eq . ( [ eq : momentpc ] ) to get a prediction for the second moments . for @xmath222 this gives very good agreement with the data ( though the fact that @xmath196 is in the default scheme perhaps complicates the situation ) , while @xmath223 and @xmath224 show a need for a substantial extra correction . in @xcite this extra contribution is shown to be compatible with a @xmath219 term , though it would be interesting to see if it is also compatible with a @xmath225 term with a modified coefficient . currently ( to our knowledge ) no fits have been performed for three - jet event shapes , though we understand that such fits are in progress @xcite for the mean value of the @xmath39-parameter . we look forward eagerly to the results . in the meantime it is possible to verify standard parameters for @xmath94 and @xmath163 against a single published point at @xmath127 @xcite and one finds reasonable agreement within the uncertainties . two other points of view that have also been used in analysing data on @xmath5 event shapes are the approach of _ dressed gluons _ employed by gardi and grunberg @xcite and the use of _ renormalisation group improved _ perturbation theory @xcite as carried out by the delphi collaboration @xcite . in the gardi - grunberg approach one uses the basic concept of renormalons ( on which the dw model is also based ) , but one choses to treat the renormalon integral differently to the dw model . this renormalon integral is ill - defined due to the landau singularity in the running coupling and instead of assuming an infrared finite coupling below some matching scale @xmath160 as was the case in the dw model , gardi and grunberg define the renormalon integral by its principal value ( a discussion of the theoretical merits of different approaches has been give in @xcite ) . in doing this they explicitly include higher order renormalon contributions in their perturbative result , rather than including them via power behaved corrections below scale @xmath160 that result from assuming an infrared finite coupling ( though they do also compare to something similar , which they call a ` cutoff ' approach ) . however since one is dealing with an ambiguous integral ( a prescription other than a principal value one would give a result differing by an amount proportional to a power correction ) one must still allow for a power correction term . gardi and grunberg studied the mean thrust in @xmath5 annihilation and used the following form for fitting to the data : @xmath226+\frac{\lambda}{q}\ ] ] where the subscript pv denotes the principal value of the renormalon integral @xmath227 for the thrust ( which includes the full lo contribution and parts of the higher - order contributions ) and @xmath228 is a piece that accounts for the difference between the true nlo coefficient for @xmath14 and the contribution included in @xmath227 . the best fit values obtained for @xmath229 and @xmath230 are respectively @xmath231 and @xmath232 . we note that @xmath233 in this approach corresponds to a smaller power correction than is required in the dw model . this is probably a consequence of the inclusion of pieces of higher perturbative orders via a principal value prescription , although it leads to a somewhat small value ( compared to the world average ) of the coupling at scale @xmath234 . next we turn to the renormalisation group improved ( rgi ) perturbative estimates @xcite that have also been used to compare with event shape data for the mean value of different event shapes in @xmath5 annihilation . the basic idea behind this approach is to consider the dependence of the observable on the scale @xmath82 , which can be expressed using renormalisation group invariance as @xmath235 where in the above formula @xmath236 , @xmath237 , with @xmath238 being the mean value of a given event shape and @xmath239 being the coefficient of @xmath240 in its perturbative expansion . from comparisons to @xmath5 event - shape data using the rgi approach . the band shows the mean value of @xmath94 and uncertainty , as obtained from the observables below the dashed line . figure taken from ref . @xcite.,scaledwidth=40.0% ] thus the first term in the perturbative expansion of @xmath109 is simply @xmath241 . the @xmath242 are renormalisation scheme independent quantities . in particular the quantity @xmath243 is simply the ratio of the first two coefficients of the qcd @xmath145 function while @xmath244 additionally depends on the first three ( up to nnlo ) perturbative coefficients in the expansion of @xmath245 . current studies using this method are therefore restricted to an nlo analysis involving @xmath246 alone . with this simplification , the solution of eq . ( [ eq : rgi ] ) simply corresponds to the introduction of an observable - specific renormalisation scheme and associated scale parameter @xmath247 which is such that it sets the nlo perturbative term to be zero ( and neglects higher perturbative orders ) . its relation to the standard @xmath248 is easily obtained : @xmath249 where @xmath250 and @xmath251 , with @xmath252 being the nlo coefficient in the @xmath253 expansion for the observable . the terms in the bracket account for different definitions of the coupling in terms of @xmath254 as used in the @xmath109 and @xmath253 schemes . in the specific case of fits to @xmath5 event shape mean values , the above approach has met with a considerable success @xcite . in particular the introduction of @xmath247 seems to remove any need for a significant power correction . the delphi collaboration @xcite have performed a combined fit of the parameter @xmath247 and a parameter @xmath255 that quantifies the power correction @xcite . the value of @xmath255 was found to be consistent with zero in most cases and rather small in all cases which indicates that there is no real need for a power behaved correction once the perturbative expansion is fixed through the rgi technique . the agreement between @xmath1 values extracted from different observables is impressive ( see figure [ fig : rgi ] ) with a spread that is only about half as large as that obtained using the standard perturbative result supplemented by a power correction term , using the dw initiated model . while it is clear that the size of the power correction piece inevitably depends on how one choses to define the perturbative expansion ( in fact all the methods discussed thus far , including the dw model and the gardi - grunberg approach , account for this effect ) , it is nevertheless very interesting that for several different observables , with significantly different perturbative coefficients , one observes after the introduction of rgi , hardly any need for a power correction term . certainly it is not clear on the basis of any physical arguments , why the genuine non - perturbative power correction should be vanishingly small and that a perturbative result defined in a certain way should lead to a complete description of the data . the clarification of this issue is still awaited and the above findings are worth further attention and study . a final point to be kept in mind about the rgi approach in the above form , is that it is valid only for inclusive observables that depend on just a single relevant scale parameter @xmath82 . this makes its applicability somewhat limited and for instance it is not currently clear how to extend the procedure so as to allow a meaningful study of event shape distributions in @xmath5 annihilation or dis event shape mean values and distributions ( which involve additional scales ) . so far our discussion of comparisons between theory and experiment has been limited to a study of mean values of the event shapes . as we have seen , there is some ambiguity in the interpretation of these comparisons , with a range of different approaches being able to fit the same data . this is in many respects an unavoidable limitation of studies of mean values , since the principal characteristic of the different models that is being tested is their @xmath82-dependence , which can also be influenced by a range of ( neglected ) higher - order contributions . in contrast , the full distributions of event shapes contain considerably more information and therefore have the potential to be more discriminatory . in section [ sec : perturbative ] we discussed the perturbative calculation of distributions . as for mean values though , the comparison to data is complicated by the need for non - perturbative corrections . these corrections however involve many more degrees of freedom than for mean values , and there are a variety of ways of including them . accordingly we separate our discussion into two parts : in section [ sec : evgen ] we shall consider studies in which hadronisation corrections are taken from monte carlo event generators , and where the main object of study is the perturbative distribution , with for example fits of the strong coupling . in section [ sec : anldist ] we shall then consider studies which involve analytical models for the non - perturbative corrections , and where it is as much the non - perturbative models , as the perturbative calculations that are under study . compared to a variety of event generators @xcite . the shaded band in the lower plot indicates the experimental statistical and systematic errors added in quadrature . figure taken from @xcite.,title="fig:",scaledwidth=48.0% ] compared to a variety of event generators @xcite . the shaded band in the lower plot indicates the experimental statistical and systematic errors added in quadrature . figure taken from @xcite.,title="fig:",scaledwidth=48.0% ] before considering the purely perturbative studies mentioned above , let us recall that one major use of event shape distributions has been in the testing and tuning @xcite of monte carlo event generators such as herwig @xcite , jetset / pythia @xcite and ariadne @xcite . figure [ fig : tuning ] shows comparisons for two event shapes , the thrust and thrust major , and the agreement is remarkable testimony to the ability of the event generators to reproduce the data . the use of event generators to probe the details of qcd is unfortunately rather difficult , essentially because they contain a number of parameters affecting both the non - perturbative modelling and in some cases the treatment of the perturbative shower . furthermore though considerable progress is being made in matching to fixed order calculations ( see for example @xcite ) , event generators are in general able to guarantee neither the nll accuracy nor the nlo accuracy of full matched nll - nlo resummed calculations . event - shapes ( global @xcite two - jet event shapes ) herwig @xcite is expected to be correct to nll accuracy . ] nevertheless , the good agreement of the event generators with the data suggests that the bulk of the dynamics is correct and in particular that a good model for the hadronisation corrections can be had by comparing parton and hadron ` levels ' of the generator . there has been a very widespread use of event generators in this way to complement the nll+nlo perturbative calculations . such a method has both advantages and drawbacks and it is worth devoting some space to them here . on one hand , the parton level of an event generator is not a theoretically well - defined concept it is regularised with some effective parton mass or transverse momentum cutoff , which already embodies some amount of non - perturbative correction . in contrast the nlo+nll partonic prediction integrates down to zero momenta without any form ( or need ) of regularisation . this too implies some amount of non - perturbative contribution , but of a rather different nature from that included via a cutoff . the resulting difference between the event - generator and the purely perturbative nlo+nll parton levels means that the ` hadronisation ' that must be applied to correct them to hadron level is different in the two cases . there are nevertheless reasons why event - generators are still used for determining the hadronisation corrections . the simplest is perhaps that they give a very good description of the data ( cf . [ fig : tuning ] ) , which suggests that they make a reasonable job of approximating the underlying dynamics . furthermore the good description is obtained with a single set of parameters for all event shapes , whereas as we shall see , other approaches with a single ( common ) parameter are currently able to give equally good descriptions only for a limited number of event shapes at a time . additionally , the objection that the monte carlo parton level is ill - defined can , partially , be addressed by including a systematic error on the hadronisation : it is possible for example to change the internal cutoff on the monte carlo parton shower and at the same time retune the hadronisation parameters in such a way that the monte carlo description of the hadron - level data remains reasonable . in this way one allows for the fact that the connection between monte carlo and nll+nlo ` parton - levels ' is not understood . this is the procedure that has been used for example in @xcite , and the results are illustrated in figure [ fig : sldhadr ] as a hadronisation correction factor @xmath256 with an uncertainty corresponding to the shaded band . , with uncertainties , as discussed in the text ) obtained from the ratio of jetset parton to hadron - level distributions . it is compared to nlo and nll+nlo predictions . also shown is the detector to hadron - level correction factor @xmath257 . figure taken from @xcite . ] two further points are to be made regarding monte carlo hadronisation corrections . firstly , as is starting now to be well known , it is strongly recommended not to correct the data to parton - level , but rather to correct the theoretical perturbative prediction to hadron level . this is because the correction to parton level entails the addition of extra assumptions , which a few years later may be very difficult to deconvolve from the parton - level ` data ' . it is straightforward on the other hand to apply a new hadronisation model to a perturbative calculation . secondly it is best not to treat the hadronisation correction as a simple multiplicative factor . to understand why , one should look again at figure [ fig : sldhadr ] ( which , dating from several years ago , corrects data to parton level ) . the multiplicative factor @xmath256 ( parton@xmath258hadron ) varies very rapidly close to the peak of the distribution and goes from above @xmath161 to below @xmath161 . this is because the distribution itself varies very rapidly in this region and the effect of hadronisation is ( to a first approximation , see below ) to shift the peak to larger values of the observable . but a multiplicative factor , rather than shifting the peak , suppresses it in one place and recreates it ( from non - peak - like structure ) in another . instead of applying a multiplicative correction , the best way to include a monte carlo hadronisation correction is to determine a transfer matrix @xmath259 which describes the fraction of events in bin @xmath56 at parton level that end up in in bin @xmath55 at hadron level . then for a binned perturbative distribution @xmath260 , the binned hadronised distribution @xmath261 is obtained by matrix multiplication , @xmath262 . this has been used for example in @xcite . having considered the basis and methods for including hadronisation corrections from event generators , let us now examine some of the perturbative studies made possible by this approach . the majority of them use nll+nlo perturbative calculations to fit for the strong coupling ( as discussed below some , e.g. @xcite , just use nlo results , which exist for a wider range of observables ) and such fits can be said to be one of the principal results from event - shape studies . in general there is rather good agreement between the data and the resummed predictions ( cf . [ fig : sldhadr ] ) . obtained by fitting nll+nlo event - shape distributions to l3 data at a range of centre of mass energies ( figure taken from @xcite ) . only experimental uncertainties are shown.,scaledwidth=60.0% ] an example of such a study is given in figure [ fig : l3asevln ] , which shows results for fits of @xmath1 to l3 data at a range of energies ( including points below @xmath263 from radiative events which , as discussed earlier , are to be taken with some caution ) . one important result from such studies is an overall value for @xmath1 , which in this case is @xmath264 @xcite , where the first error is experimental and the second is theoretical . similar results have been given by other collaborations , e.g. @xmath265 from aleph @xcite . there is also good evidence for the evolution of the coupling , though leaving out the potentially doubtful radiative points leaves the situation somewhat less clearcut , due to the limited statistics at high energies ( another resummed study which gives good evidence for the running of the coupling uses jade and opal data @xcite , however it is limited to jet rates ) . these and similar results from the other lep collaborations and jade highlight two important points . firstly at higher energies there is a need to combine results from all the lep experiments in order to reduce the statistical errors and so improve the evidence for the running of the coupling in the region above @xmath127 . work in this direction is currently in progress @xcite . so far only preliminary results are available @xcite , and as well as improving the picture of the high - energy running of the coupling , they suggest a slightly smaller value of @xmath1 than that quoted above , more in accord with world averages @xcite . secondly , in the overall result for @xmath266 , by far the most important contribution to the error is that from theoretical uncertainties . however theoretical uncertainties are notoriously difficult to estimate , since they relate to unknown higher orders . systematic investigations of the various sources of uncertainty have been carried out in @xcite and in particular @xcite proposes a standard for the set of sources of uncertainty that ought to be considered , together with an approach for combining the different sources into a single overall uncertainty on @xmath94 . one of the main principles behind the method is that while one may have numerous estimates of sources of theoretical uncertainty , it is not advisable to combine them in quadrature , as one might be tempted to do , because this is likely to lead to a double - counting of uncertainties . rather , one should examine the different sources of uncertainty across the whole range of the distribution and at each point , take the maximum of all sources to build up an uncertainty envelope or band , represented by the shaded area in figure [ fig : lepqcdbands ] ( shown relative to a reference prediction for a ` standard ' theory ) . the overall uncertainty on the coupling ( rather than the distribution ) is given by the range of variation of @xmath94 such that the prediction remains within the band . it is to be noted that this kind of approach is of relevance not just to event - shape studies but also quite generally to any resummed matched calculation , which is inherently subject to many sources of arbitrariness . allowed such that the prediction stays within the uncertainty band.,scaledwidth=63.0% ] one of the drawbacks of nll+nlo studies is that the nll resummed predictions exist for only a fraction of observables . recently there has been an extensive study by the delphi collaboration @xcite of 18 event - shape distributions compared to nlo calculations . since in most cases nlo calculations with a renormalisation scale @xmath267 tend to describe the data rather poorly , they have examined various options for estimating higher orders so as to improve the agreement . their main approach involves a simultaneous fit of the coupling and the renormalisation scale , i.e. the renormalisation scale dependence is taken as a way of parameterising possible higher orders . in such a procedure the choice of fit - range for the observable is rather critical , since the true higher - orders may involve have quite a different structure from that parameterised by the @xmath168 dependence . the delphi analysis restricts the fit range to be that in which a good fit can be obtained . with these conditions they find relatively consistent values of @xmath94 across their whole range of observables ( finding a scatter of a few percent ) , with a relatively modest theoretical uncertainty , as estimated from the further variation of @xmath168 by a factor of @xmath13 to @xmath16 around the optimal scale . their overall result from all observables treated in this way is @xmath268 ( including both experimental and theoretical errors ) . they also compare this approach to a range of other ways of ` estimating ' higher orders . they find for example that some theoretically based methods for setting the scale ( principle of minimal sensitivity @xcite , effective charge @xcite ) lead to scales that are quite correlated to the experimentally fitted ones , while another method ( blm @xcite ) is rather uncorrelated . they compare also to resummation approaches , though there is no equivalent renormalisation scale choice that can be made . instead the comparison involves rather the quality of fit to the distribution and the final spread of final @xmath1 values as a measure of neglected higher - order uncertainties . not surprisingly , they find that while the nll results work well in the two - jet region , in the three - jet region the combined nll+nlo predictions fare not that much better than the pure ( @xmath267 ) nlo results . they go on to argue that nll+nlo is somewhat disfavoured compared to nlo with an ` optimal ' scale . this statement is however to be treated with some caution , since the fit range was chosen specifically so as to obtain a good fit for nlo with the ` optimal ' scale one could equally have chosen a fit range tuned to the nll+nlo prediction and one wonders whether the nlo with optimal scale choice would then have fared so well . regardless of this issue of fit range , it is interesting to note that over the set of observables which can be treated by both methods , the total spread in @xmath1 values is quite similar , being @xmath269 in the case of nlo with optimal scale choice and @xmath270 for nll+nlo . this is similar to the spread seen for fits to mean values with analytical hadronisation models ( fig . [ fig : meanfits ] ) , though the rgi fits of @xcite have a somewhat smaller error . the jet rates ( especially the durham and cambridge algorithms and certain members of the jade family ) are among the few observables for which the pure nlo calculation gives a reasonable description of the distribution ( cf . table 3 of @xcite ) . one particularly interesting set of nlo studies makes use of the 3-jet rate as a function of @xmath271 ( this is just the integral of the distribution of @xmath137 ) in events with primary @xmath144-quarks as compared to light - quark events @xcite ( some of the analyses use other observables , such as the 4-jet rate or the thrust ) . using nlo calculations which account for massive quarks @xcite makes it possible , in such studies , to extract a value for the @xmath144 mass at a renormalisation scale of @xmath263 , giving first evidence of the ( expected ) running of the @xmath144-quark mass , since all the analyses find @xmath272 in the range @xmath273 to @xmath274 gev ( with rather variable estimates of the theoretical error ) . we have already seen , in section [ sec : meanpc ] , that analytical models for hadronisation corrections , combined with normal perturbative predictions , can give a rather good description of mean values . there the inclusion of a non - perturbative ( hadronisation ) correction was a rather simple affair , eq . ( [ eq : meancorrections ] ) , since it was an additive procedure . in contrast the general relation between the full distribution @xmath275 for an observable @xmath276 and the perturbatively calculated distribution @xmath277 is more complicated , @xmath278 where @xmath279 encodes all the non - perturbative information . ( [ eq : shapegenconvoluion ] ) can be seen as a generalised convolution , where the shape of the convolution function depends on the value @xmath86 of the variable as well as the coupling and hard scale . such a general form contains far more information however than can currently be predicted , or even conveniently parameterised . accordingly analytical hadronisation approaches generally make a number of simplifying assumptions , intended to be valid for some restricted range of the observable . the most radical simplification of eq . ( [ eq : shapegenconvoluion ] ) that can be made is to replace @xmath280 with a @xmath281-function , @xmath282 , leading to @xmath283 this was proposed and investigated in @xcite , and the combination @xmath284 is the same that appears in the power correction for the mean value @xcite , discussed in section [ sec : meanpc ] . ( [ eq : npshift ] ) holds for observables with an @xmath1 independent power correction in tables [ tab : cvs ] and [ tab : cvsdis ] , in the region @xmath285 , where the lower limit ensures that one can neglect the width of the convolution function @xmath286 , while the upper limit is the restriction that one be in the born limit ( in which the @xmath152 were originally calculated ) . for certain 3-jet observables a similar picture holds , but with a shift that depends on the kinematics of the 3-jet configuration @xcite . for the broadenings , the situation is more complicated because the power correction is enhanced by the rapidity over which the quark and thrust axes can be considered to coincide , @xmath287 , @xmath288 being the angle between thrust and quark axes this angle is strongly correlated with the value of the broadening ( determined by perturbative radiation ) and accordingly the extent of the shift becomes @xmath252-dependent @xcite . the simplest case is the @xmath0 wide - jet broadening , for which one has @xmath289 where the shift is written for the integrated distribution @xmath290 in order to simplify the expressions . the @xmath291 enhancement for the power correction to the mean broadening ( tables [ tab : cvs ] and [ tab : cvsdis ] ) comes about simply because the average value of @xmath292 , after integration over @xmath86 with the resummed distribution , is of order @xmath293 . the full form for @xmath294 and analogous results for the total @xmath0 and dis broadenings have been given in @xcite , with results existing also for the thrust minor @xcite , and the dis and drell - yan out - of plane momenta @xcite . yet subtler instances of perturbative , non - perturbative interplay arise for observables like the eec @xcite and dis azimuthal correlation @xcite , with the appearance of _ fractional _ powers of @xmath82 in the power correction , as mentioned before . -dependent non - perturbative shift in the jet broadening distribution for @xmath5 annihilation . the dashed curve is the perturbative prediction ( nll+nlo ) , while the solid curve includes a non - perturbative ( @xmath252-dependent ) shift . figure taken from @xcite . right : nll+nlo distribution with a @xmath4 shift for @xmath295 in dis at a range of @xmath82 values ( round dots indicate points used in fits ) @xcite . the @xmath82 values range from 15 gev ( bottom ) to 81 gev ( top).,title="fig:",scaledwidth=44.0%,scaledwidth=48.0% ] -dependent non - perturbative shift in the jet broadening distribution for @xmath5 annihilation . the dashed curve is the perturbative prediction ( nll+nlo ) , while the solid curve includes a non - perturbative ( @xmath252-dependent ) shift . figure taken from @xcite . right : nll+nlo distribution with a @xmath4 shift for @xmath295 in dis at a range of @xmath82 values ( round dots indicate points used in fits ) @xcite . the @xmath82 values range from 15 gev ( bottom ) to 81 gev ( top).,title="fig:",scaledwidth=48.0% ] one might a priori have thought that the formal domain of validity of the shift approximation , @xmath296 , would be somewhat limited . figure [ fig : broadpc ] ( left ) illustrates what happens in the case of @xmath297 quite remarkably the shift describes the data well over a large range of @xmath297 , with slight problems only in the extreme two - jet region ( peak of the distribution ) and in the four - jet region ( @xmath298 ) . similar features are seen in the right - hand plot of fig . [ fig : broadpc ] for @xmath299 in dis at a range of @xmath82 values : a large part of the distribution is described for all @xmath82 values , and problems appear only in the @xmath91 jet region ( @xmath300 ) , and in the @xmath122 jet region for low @xmath82 . this success is reproduced for quite a range of observables in @xmath0 and dis . -@xmath209 contours for simultaneous fits of @xmath94 and @xmath163 to @xmath0 event - shape distributions . the error contours account for statistic , systematical and theoretical uncertainties . the shaded band indicates the result for @xmath1 when using monte carlo hadronisation models . figure taken from @xcite . right : @xmath161-@xmath209 contours from fits to h1 dis event shape distributions @xcite with statistical and systematic errors added in quadrature . figure taken from @xcite.,title="fig:",scaledwidth=48.0% ] -@xmath209 contours for simultaneous fits of @xmath94 and @xmath163 to @xmath0 event - shape distributions . the error contours account for statistic , systematical and theoretical uncertainties . the shaded band indicates the result for @xmath1 when using monte carlo hadronisation models . figure taken from @xcite . right : @xmath161-@xmath209 contours from fits to h1 dis event shape distributions @xcite with statistical and systematic errors added in quadrature . figure taken from @xcite.,title="fig:",scaledwidth=48.0% ] as was the case for mean values , a more systematic test involves carrying out a simultaneous fit of @xmath94 and @xmath163 for each observable and then checking for consistency between observables and with the corresponding results for mean values . results from @xmath0 and dis are shown in figure [ fig : distcont ] , taken from @xcite and @xcite . other fits that have been carried out in recent years include @xcite .- parameter in @xmath65-jet events . ] just as for the mean values , one notes that ( with exception of @xmath301 and @xmath302 in @xmath0 , to be discussed shortly ) there is good consistency between observables , both for @xmath0 and dis . this statement holds holds also for the eec , fitted in @xcite , not shown in fig . [ fig : distcont ] . there is also good agreement with the results for the mean values , fig . [ fig : meanfits ] , except marginally for dis @xmath94 results , which for distributions are in better accord with the world average . the good agreement between distributions and mean values is important in the light of alternative approaches to fitting mean values , such as the rgi which , we recall , seems to suggest that the power correction for mean values can just as well be interpreted as higher order contributions . were this really the case then one would expect to see no relation between @xmath163 results for mean values and distributions . despite the generally good agreement , some problems do persist . examining different groups results for the @xmath1-@xmath163 fits one finds that for some observables there is a large spread in the results . for example for the @xmath0 @xmath297 , aleph @xcite find @xmath303 while the jade result @xcite shown in fig . [ fig : distcont ] corresponds to @xmath304 ( and this differs from a slightly earlier jade result @xmath305 @xcite ) . while the results all agree to within errors , those errors are largely theoretical and would be expected to be a common systematic on all results . in other words we would expect the results to be much closer together than @xmath161-@xmath209 . that this is not the case suggests that the fits might be unstable with respect to small changes in details of the fit , for example the fit range ( as has been observed when including data in the range @xmath306 in the fit corresponding to the right hand plot of figure [ fig : distcont ] @xcite ) . these differences can lead to contradictory conclusions regarding the success of the description of different observables and it would be of value for the different groups to work together to understand the origin of the differences , perhaps in a context such as the lep qcd working group . there do remain two cases however where there seems to be clear incompatibility between the data and theoretical prediction : the heavy - jet mass , @xmath196 ( also referred to as @xmath302 ) and the wide - jet broadening , @xmath301 . the heavy - jet mass is subject to @xmath146 hadron mass effects , as discussed already in the context of mean values , however even after accounting for this by considering @xmath62 or @xmath57-scheme data , as has been done by delphi @xcite , the results appear to be rather inconsistent with other observables : @xmath1 is too small and @xmath163 too large , as for @xmath301 , with a fixed @xmath307 lead to a consistent value for @xmath308 , though the overall fit quality is poor . ] though less extreme . the common point between these two observables is that they both select the heavier / wider hemisphere and so are less inclusive than other observables . studies with event generators @xcite suggest that whereas for more inclusive observables the 2-jet limit of the shift works well even into the 3-jet region , for heavy - wide observables the 3-jet limit behaves very differently from the 2-jet limit and these differences becomes relevant even for relatively low values of the observable ( some partially related discussion of the problem has been given also in @xcite ) . we will come back to this and other potential explanations below when we discuss shape functions . beyond the ` usual ' observables discussed above , there are some recent results also for the @xmath38-parameter shoulder at @xmath125 , where @xcite the shoulder position is seen to be consistent with a shift with the usual 2-jet coefficient @xmath205 this is in contradiction with the ( smaller ) result eq . ( [ eq : c34power ] ) for the shift to the shoulder ; however the analysis @xcite does nt take into account resummation effects at the shoulder which may also contribute to an effective shift . a first analysis of corrections to the @xmath309 distribution has been given in @xcite which shows evidence for @xmath310 contributions in the @xmath309 distribution at the two lowest jade centre of mass energies ( 14 and 22 gev ) . one last interesting point to discuss regarding the shift approximation for hadronisation is an application to the measurement of the qcd colour factors . the fact that the hadronisation is encoded via just a single parameter , with all the dependence on colour factors available explicitly for the perturbative and non - perturbative contributions , means that it is feasible to carry out a simultaneous fit not only for @xmath311 and @xmath163 but also for one or more of the qcd colour factors @xmath139 , @xmath138 and @xmath312 @xcite . this would have been much more difficult using monte carlo hadronisation because all the hadronisation parameters would have had to have been retuned for each change of the colour factors . the results for this fit are summarised and compared to other approaches in figure [ fig : colfact ] @xcite , which shows @xmath16-@xmath209 contours for @xmath138 and @xmath139 . while other approaches constrain the ratio @xmath313 , the event - shape approach gives by far the best independent measurement of @xmath139 and @xmath138 . as a result the combination of the results gives rather tight limits on the colour factors , which are in good agreement with the qcd expectation . ( shift ) power corrections and a variety of other methods . figure taken from @xcite . the ellipses show correlated measurements from studies of the angular distributions of 4-jet events @xcite and event shapes @xcite , while the lines are obtained from studies of the gluon fragmentation function @xcite and from multiplicities in gluon jets @xcite.,scaledwidth=51.0% ] we have already mentioned that the approximation of a simple shift is valid only in the region @xmath296 . to better understand the origin of the lower limit , let us expand eq . ( [ eq : npshift ] ) , @xmath314 where @xmath315 is the @xmath316 derivative of @xmath317 with respect to @xmath86 . since @xmath315 goes as @xmath318 one sees that when the shift @xmath319 all terms in the series are of the same order . while the @xmath320 term can be related to the power correction to the mean value of @xmath276 , the shift merely provides an ansatz for the higher terms . the breakdown for small values of @xmath321 is clearly visible for the ( dis ) thrust in fig . [ fig : broadpc ] ( right ) . a physically transparent way of dealing with the @xmath322 terms has been developed by korchemsky , sterman and collaborators @xcite . one approximates @xmath323 in eq . ( [ eq : shapegenconvoluion ] ) by a _ shape function _ @xcite @xmath324 ( rather than the @xmath281-function which corresponds to a pure shift ) , @xmath325 in eq . ( [ eq : expandedshift ] ) , @xmath326 is then replaced by @xmath327 where @xmath328 is the @xmath316 moment of @xmath329 . there is considerable freedom in one s choice of the form for the shape function . accordingly , success in fitting a given event - shape distribution for any single value of @xmath82 can not , alone , be considered strong evidence in favour of the shape - function picture . the true test of the approach comes by determining the shape function at one value of @xmath82 and then establishing whether it applies for all values of @xmath82 . this is illustrated for the @xmath0 thrust in fig . [ fig : thrustshape ] ( left ) , where the following shape - function has been used @xcite , @xmath330 with two free parameters , @xmath331 and @xmath254 ( it has been argued that the behaviour of this function for small @xmath176 and large @xmath176 is similar to that expected from a simple physical model for the dynamics at the origin of the shape function @xcite ) . one notes the remarkable agreement , far into the two - jet region , for the whole range of @xmath82 values . ) . shown for @xmath332 and @xmath333 gev . figure taken from @xcite . right : the thrust distribution at the perturbative ( dge ) level , with a non - perturbative shift , and with a shape function . figure taken from @xcite.,title="fig:",scaledwidth=48.0%,scaledwidth=48.0% ] ) . shown for @xmath332 and @xmath333 gev . figure taken from @xcite . right : the thrust distribution at the perturbative ( dge ) level , with a non - perturbative shift , and with a shape function . figure taken from @xcite.,title="fig:",scaledwidth=48.0%,scaledwidth=48.0% ] this is a clear success for the shape - function approach as compared to the simple shift discussed above . however the drawback of shape - functions is that a priori one loses all predictivity ( save for the first moment ) when going from one observable to another . this predictivity was one of the main strengths of the shift approach . considerable work has therefore been devoted to understanding both the general properties of shape functions and the relations between shape functions of different observables . one line of investigation has been to relate the shape functions for the thrust and heavy - jet mass ( @xmath196 ) this can be done with the help of the observation @xcite that in the two - jet limit the thrust is just the sum of the light and heavy - jet ( squared normalised ) masses . in @xcite korchemsky and tafat introduced a shape function @xmath334 describing the distribution of non - perturbative radiation in each of the two hemispheres . the shape function for the thrust is then obtained by @xmath335 instead for the heavy - jet mass one calculates the full ( np ) double differential distribution for the two jet masses and from that derives the heavy - jet mass distribution ( this is not quite equivalent to a simple shape - function for @xmath196 ) . this approach gave a good fit to the thrust and heavy - jet mass data , on the condition that there was significant correlation between the non - perturbative contribution to the two hemispheres .- parameter ( with an overall rescaling to ensure the correct first moment ) , however the data were less well described than for the thrust . this is perhaps not surprising : owing to different sensitivities to large - angle radiation , the thrust and @xmath38-parameter shape functions are expected to have a different structure of higher moments @xcite . ] it is difficult to interpret this result however , because as has already been mentioned , the default measurement scheme for the jet masses suffers from extra ( universality - breaking ) @xmath146 corrections , and it is possible that in @xcite these are being mimicked by inter - hemisphere correlations . the thrust and heavy - jet mass have been investigated in great detail also by gardi and rathsman @xcite . in analogy with the approach for the mean thrust @xcite , they attempt not only to account for nll+nlo and power correction contributions , but also for all ( perturbative ) enhanced effects associated with the running of the coupling , within a formalism that they refer to as dressed gluon exponentiation ( dge ) . among the interesting theoretical results ( see also @xcite ) is that the renormalon analysis suggests that the second moment of the shape function should be precisely that expected from a pure shift ( the same is true for the @xmath38-parameter @xcite ) . this observation seems to be consistent with fits to the data , and would explain why the shift approximation works so well , at least for the relevant observables . the right - hand plot of fig . [ fig : thrustshape ] shows the result of their analysis for the thrust , with both a shift and a shape - function . while the shape - function is clearly needed in the immediate vicinity of the peak and below , it is striking to see , above the peak , how well it is approximated by the shift . one of the other interesting results in @xcite relates to the analysis of the heavy - jet mass . in contrast to @xcite , a common hadron - mass scheme ( decay - scheme ) is used for the thrust and the heavy - jet mass , and there are no inter - hemisphere correlations in their shape function . using the same parameters for the thrust and @xmath196 leads to reasonable fits , suggesting that the true inter - hemisphere correlations may actually be small . ( a fit for @xmath196 alone does however prefer a lower value of @xmath94 than for the thrust , as is seen also with the usual shifted nll+nlo approach @xcite ) . a puzzle that emerges in the work of @xcite is that all the fits lead to quite small values of the coupling , @xmath336 with a theoretical error of about @xmath210 . a similarly low result was obtained in the analogous analysis of the mean thrust @xcite . in contrast a study of @xmath144-fragmentation using dressed gluon exponentiation finds more standard values , @xmath337 @xcite . let us close our discussion of shape functions with a recent result by berger & sterman @xcite . given that there are no simple relations between the shape functions for most common event shapes , they instead consider a new class of event shapes @xcite whose definition involves a parameter @xmath331 which allows the nature of the event shape to be continuously varied from a broadening ( @xmath338 ) to a thrust - type observable ( @xmath339 ) and beyond . their observation is that for all variants of the observable with @xmath340 , the logarithm of any moment of the shape functions scales as @xmath341 . in practice this scaling is found to hold in monte carlo studies of this class of observables it would therefore be interesting to have data for these observables with which to make a comparison . conceptually , event shapes are rather simple observables , yet they are sensitive to a range of characteristics of qcd radiation . this combination of theoretical simplicity and experimental sensitivity is perhaps one of the main reasons why event shapes have found so many applications . as we have seen , they provide vital inputs in studies of the ingredients of the qcd lagrangian , such as the coupling and the colour factors ; they play an important role in the tuning and testing of event generators ; and they are at the heart of recent investigations into analytical approaches to understanding the dynamics of hadronisation . there remain several directions in which progress may be expected ( or at the very least hoped for ) in coming years . all the experimental comparisons discussed here have been for event - shapes that vanish in the @xmath16-jet limit , yet there have been significant theoretical developments in recent years for observables that vanish also in the @xmath65-jet limit , examples being the @xmath39-parameter and thrust minor @xcite . studies with jet rates @xcite suggest that going to such ` three - jet ' observables could reduce theoretical errors , though as discussed in @xcite this needs to be investigated more systematically . additionally three - jet observables will allow much more stringent tests of analytical hadronisation approaches since for the first time they introduce sensitivity to predictions about hadronisation from a gluon @xcite , not just from a quark . one of the frontiers of event shape studies is their extension to new processes . while @xmath0 is the traditional domain for event - shape studies , we have seen that studies in dis are in many respects competitive with @xmath0 results . additionally they provide important input on the question of universality of analytical hadronisation models , ruling out for example significant modifications of the hadronisation picture stemming from interactions with the proton remnant . though so far the question has received only limited study , one interesting application of dis studies would be in the use of event shapes for obtaining information on parton distribution functions , as suggested in @xcite . currently an option that is attracting growing experimental @xcite and theoretical @xcite interest is the development of event shapes in hadron - hadron collisions . though much work remains to be done on this subject , it seems that the sensitivity to parton distributions will be somewhat stronger than in most dis observables . furthermore one expects a rich hadronisation structure associated both with the four - jet structure of hadronic dijet events and with the properties of the underlying event . it is to be hoped that the availability of automated resummation methods @xcite will make the theoretical aspects of the study of new observables and processes more straightforward than has been the case in the past . another frontier on which progress is expected in near future is with respect to the accuracy of theoretical predictions . nnlo perturbative calculations are making rapid progress ( for a recent review see @xcite ) with the main outstanding problem being that of a full subtraction procedure for combining zero , one and two - loop contributions and its implementation in a monte carlo integrator . progress on nnll resummations is also being made , albeit so far only in inclusive cases @xcite . though technically more involved , an extension to event shape resummations is certainly conceivable in the near future . aside from the expected gains in accuracy that are the main motivation for these calculations , it will also be interesting to compare the exact higher - order calculations with predictions from approaches such as @xcite which aim to identify the physically dominant higher - order contributions and in some cases @xcite claim to significantly reduce the theoretical uncertainties on @xmath94 . less predictable is what development can be expected on the subject of hadronisation corrections . as we have seen , much has been learnt from event - shape studies . yet a variety of open questions remain . at a practical level it would for example be of interest to see an extension of the shape - function approach to a wider range of variables , including the quite subtle case of the broadenings and also to dis event shapes . the wealth of dis data at low @xmath82 , where shape functions are most important , makes the latter especially interesting . there remain also important issues that are poorly understood even at the conceptual level . for example predictions are currently limited to the domain in which the event - shape is close to vanishing ( e.g. @xmath14 in the @xmath16-jet limit ) . it is tempting to suggest that problems that persist for a couple of ` recalcitrant ' observables ( specifically , @xmath196 , @xmath301 ) may actually be due to large differences between the power correction in the three - jet region ( a significant part of the fit - range ) and that in the well - understood two - jet limit ( see e.g. fig . 3 of @xcite ) . but techniques that would allow the calculation of the power correction to , say , the thrust or the broadening in the @xmath65-jet limit do not yet exist . another ( partially related ) issue is that of anomalous dimensions these have so far only been calculated for @xmath146 effects that are associated with hadron masses @xcite . yet , as has been pointed out also in @xcite , they are bound to be present for all classes of power suppressed contributions . one clear physical origin for them is that soft - gluons ( or hadrons ) with transverse momenta of order @xmath254 are radiated not just from the @xmath206 system but also from all soft perturbative gluons with transverse momenta between @xmath254 and @xmath82 . accounting for radiation from the latter will clearly lead to enhancements of power corrections by terms @xmath342 . and a final point , not to be forgotten , is that most of the analytical hadronisation approaches are essentially perturbative methods ` in disguise ' . an understanding of how they really relate to _ _ hadron__isation is very much lacking . let us close this review by remarking that one of the characteristics of event shape studies is that progress is made not merely through a better understanding of existing observables and data , but also through critical experimental studies of new observables with ( supposedly ! ) better theoretical properties . the dialogue between experimenters and theorists is crucial in this respect . we look forward to its being as fruitful in coming years as it has been up to now . it goes without saying that we are grateful to our collaborators and numerous other colleagues , including members of the lep qcd working group , for the many stimulating discussions that we have had about event shapes . for discussions on issues that arose specifically while writing this review , we thank in particular g. heinrich , f. krauss , c. j. maxwell , g. rodrigo , m. h. seymour , h. stenzel and g. zanderighi . furthermore we are indebted to a. banfi , yu . l. dokshitzer and g. zanderighi for their careful reading of and helpful comments on the manuscript . one of us ( gps ) is grateful also to the cern th division for hospitality during the course of this work . 999 j. r. ellis , m. k. gaillard and g. g. ross , nucl . b * 111 * ( 1976 ) 253 [ erratum - ibid . b * 130 * ( 1977 ) 516 ] . e. farhi , phys . * 39 * ( 1977 ) 1587 . h. georgi and m. machacek , phys . * 39 * ( 1977 ) 1237 . g. c. fox and s. wolfram , phys . * 41 * ( 1978 ) 1581 ; ibid . b * 149 * ( 1979 ) 413 [ erratum - ibid . b * 157 * ( 1979 ) 543 ] . c. l. basham , l. s. brown , s. d. ellis and s. t. love , phys . * 41 * ( 1978 ) 1585 ; c. l. basham , l. s. brown , s. d. ellis and s. t. love , phys . d * 19 * ( 1979 ) 2018 . r. k. ellis , w. j. stirling and b. r. webber , _ qcd and collider physics _ , cambridge university press , 1996 . g. marchesini , b. r. webber , g. abbiendi , i. g. knowles , m. h. seymour and l. stanco , comput . commun . * 67 * ( 1992 ) 465 . g. corcella _ et al . _ , jhep * 0101 * ( 2001 ) 010 [ hep - ph/0011363 ] . t. sjstrand , comput . commun . * 82 * ( 1994 ) 74 ; t. sjstrand , p. eden , c. friberg , l. lnnblad , g. miu , s. mrenna and e. norrbin , hep - ph/0010017 . l. lnnblad , comput . * 71 * ( 1992 ) 15 . m. beneke , phys . * 317 * ( 1999 ) 1 [ hep - ph/9807443 ] . s. brandt , c. peyrou , r. sosnowski and a. wroblewski , phys . * 12 * ( 1964 ) 57 . l. clavelli , phys . b * 85 * ( 1979 ) 111 ; t. chandramohan and l. clavelli , nucl . b * 184 * ( 1981 ) 365 ; l. clavelli and d. wyler , phys . b * 103 * ( 1981 ) 383 . s. catani , g. turnock and b. r. webber , phys . b * 295 * ( 1992 ) 269 . d. p. barber _ et al . _ [ mark - j collaboration collaboration ] , phys . * 63 * ( 1980 ) 337 . a. banfi , g. marchesini , y. l. dokshitzer and g. zanderighi , jhep * 0007 * ( 2000 ) 002 [ hep - ph/0004027 ] . a. banfi , y. l. dokshitzer , g. marchesini and g. zanderighi , phys . b * 508 * ( 2001 ) 269 [ hep - ph/0010267 ] . a. de rujula , j. r. ellis , e. g. floratos and m. k. gaillard , nucl . b * 138 * ( 1978 ) 387 . r. k. ellis , d. a. ross and a. e. terrano , nucl . b * 178 * ( 1981 ) 421 . g. parisi , phys . b * 74 * ( 1978 ) 65 ; j. f. donoghue , f. e. low and s. y. pi , phys . rev . d * 20 * ( 1979 ) 2759 . j. d. bjorken and s. j. brodsky , phys . d * 1 * ( 1970 ) 1416 ; g. hanson _ et al . _ , phys . * 35 * ( 1975 ) 1609 . c. berger _ [ pluto collaboration ] , phys . b * 82 * ( 1979 ) 449 . w. bartel _ et al . _ [ jade collaboration ] , z. phys . c * 33 * ( 1986 ) 23 . s. catani , yu . l. dokshitzer , m. olsson , g. turnock and b. r. webber , phys . b * 269 * ( 1991 ) 432 . s. bethke , z. kunszt , d. e. soper and w. j. stirling , nucl . b * 370 * ( 1992 ) 310 [ erratum - ibid . b * 523 * ( 1998 ) 681 ] . y. l. dokshitzer , g. d. leder , s. moretti and b. r. webber , jhep * 9708 * ( 1997 ) 001 [ hep - ph/9707323 ] . g. p. salam and d. wicke , jhep * 0105 * ( 2001 ) 061 [ hep - ph/0102343 ] . e. gardi and j. rathsman , nucl . b * 638 * ( 2002 ) 243 [ hep - ph/0201019 ] . r. p. feynman , _ photon hadron interactions _ , w. a. benjamin , new york , 1972 ; r. d. peccei and r. ruckl , nucl . b * 162 * ( 1980 ) 125 ; l. v. gribov , y. l. dokshitzer , s. i. troian and v. a. khoze , sov . jetp * 67 * ( 1988 ) 1303 [ zh . fiz . * 94 * ( 1988 ) 12 ] . k. h. streng , t. f. walsh and p. m. zerwas , z. phys . c * 2 * ( 1979 ) 237 . a. banfi , g. marchesini , g. smye and g. zanderighi , jhep * 0111 * ( 2001 ) 066 [ hep - ph/0111157 ] . m. dasgupta and g. p. salam , jhep * 0208 * ( 2002 ) 032 [ hep - ph/0208073 ] . a. banfi , g. marchesini and g. smye , jhep * 0204 * ( 2002 ) 024 [ hep - ph/0203150 ] . s. catani , y. l. dokshitzer and b. r. webber , phys . b * 285 * ( 1992 ) 291 . a. banfi , g. marchesini , g. smye and g. zanderighi , jhep * 0108 * ( 2001 ) 047 [ hep - ph/0106278 ] . i. a. bertram [ d0 collaboration ] , acta phys . b * 33 * ( 2002 ) 3141 . a. banfi , g. p. salam and g. zanderighi , hep - ph/0304148 ; _ ibid . _ in preparation . a. banfi , hep - ph/0309161 . g. sterman and s. weinberg , phys . * 39 * ( 1977 ) 1436 . w. t. giele and e. w. glover , phys . d * 46 * ( 1992 ) 1980 . s. catani and m. h. seymour , phys . b * 378 * ( 1996 ) 287 [ hep - ph/9602277 ] ; s. catani and m. h. seymour , nucl . b * 485 * ( 1997 ) 291 [ erratum - imbid . b * 510 * ( 1997 ) 503 ] [ hep - ph/9605323 ] . l. j. dixon and a. signer , phys . d * 56 * ( 1997 ) 4031 [ hep - ph/9706285 ] ; a. signer , comput . commun . * 106 * ( 1997 ) 125 . s. weinzierl and d. a. kosower , phys . d * 60 * ( 1999 ) 054028 [ hep - ph/9901277 ] . j. m. campbell , m. a. cullen and e. w. glover , eur . j. c * 9 * ( 1999 ) 245 [ hep - ph/9809429 ] . z. nagy and z. trocsanyi , nucl . b * 486 * ( 1997 ) 189 [ hep - ph/9610498 ] ; z. nagy and z. trocsanyi , phys . * 79 * ( 1997 ) 3604 [ hep - ph/9707309 ] ; z. nagy and z. trocsanyi , phys . d * 59 * ( 1999 ) 014020 [ erratum - ibid . d * 62 * ( 2000 ) 099902 ] [ hep - ph/9806317 ] ; z. nagy and z. trocsanyi , phys . lett . * 87 * ( 2001 ) 082001 [ hep - ph/0104315 ] ; z. nagy , phys . * 88 * ( 2002 ) 122003 [ hep - ph/0110315 ] ; z. nagy , hep - ph/0307268 . b. potter , comput . phys . commun . * 133 * ( 2000 ) 105 [ hep - ph/9911221 ] ; b. potter , comput . commun . * 119 * ( 1999 ) 45 [ hep - ph/9806437 ] . t. gehrmann , hep - ph/0310178 . m. dasgupta and g. p. salam , phys . b * 512 * ( 2001 ) 323 [ hep - ph/0104277 ] . n. brown and w. j. stirling , phys . b * 252 * ( 1990 ) 657 . g. leder , nucl . b * 497 * ( 1997 ) 334 [ hep - ph/9610552 ] . s. catani , g. turnock , b. r. webber and l. trentadue , phys . b * 263 * ( 1991 ) 491 . s. catani , l. trentadue , g. turnock and b. r. webber , nucl . b * 407 * ( 1993 ) 3 . s. catani , g. turnock and b. r. webber , phys . b * 272 * ( 1991 ) 368 . s. j. burby and e. w. glover , jhep * 0104 * ( 2001 ) 029 [ hep - ph/0101226 ] . y. l. dokshitzer , a. lucenti , g. marchesini and g. p. salam , jhep * 9801 * ( 1998 ) 011 [ hep - ph/9801324 ] . s. catani and b. r. webber , phys . b * 427 * ( 1998 ) 377 [ hep - ph/9801350 ] . g. dissertori and m. schmelling , phys . b * 361 * ( 1995 ) 167 . a. banfi , g. p. salam and g. zanderighi , jhep * 0201 * ( 2002 ) 018 [ hep - ph/0112156 ] . y. l. dokshitzer and d. diakonov , phys . b * 84 * ( 1979 ) 234 . g. parisi and r. petronzio , nucl . b * 154 * ( 1979 ) 427 . j. kodaira and l. trentadue , phys . b * 123 * ( 1983 ) 335 . j. c. collins and d. e. soper , phys . * 48 * ( 1982 ) 655 . p. e. rakow and b. r. webber , nucl . b * 187 * ( 1981 ) 254 . y. l. dokshitzer , g. marchesini and b. r. webber , jhep * 9907 * ( 1999 ) 012 [ hep - ph/9905339 ] . c. f. berger , t. kucs and g. sterman , phys . d * 68 * ( 2003 ) 014012 [ hep - ph/0303051 ] . a. banfi , y. l. dokshitzer , g. marchesini and g. zanderighi , jhep * 0105 * ( 2001 ) 040 [ hep - ph/0104162 ] . v. antonelli , m. dasgupta and g. p. salam , jhep * 0002 * ( 2000 ) 001 [ hep - ph/9912488 ] . m. dasgupta and g. p. salam , eur . j. c * 24 * ( 2002 ) 213 [ hep - ph/0110213 ] . r. bonciani , s. catani , m. l. mangano and p. nason , phys . b * 575 * ( 2003 ) 268 [ hep - ph/0307035 ] . e. gardi and j. rathsman , nucl . b * 609 * ( 2001 ) 123 [ hep - ph/0103217 ] . e. gardi and l. magnea , `` the c parameter distribution in e+ e- annihilation , '' jhep * 0308 * ( 2003 ) 030 [ hep - ph/0306094 ] . f. krauss and g. rodrigo , phys . b * 576 * ( 2003 ) 135 [ hep - ph/0303038 ] . s. catani and b. r. webber , jhep * 9710 * ( 1997 ) 005 [ hep - ph/9710333 ] . m. dasgupta and g. p. salam , jhep * 0203 * ( 2002 ) 017 [ hep - ph/0203009 ] ; c. f. berger , t. kucs and g. sterman , phys . d * 65 * ( 2002 ) 094031 [ hep - ph/0110004 ] ; a. banfi , g. marchesini and g. smye , jhep * 0208 * ( 2002 ) 006 [ hep - ph/0206076 ] . r. b. appleby and m. h. seymour , jhep * 0212 * ( 2002 ) 063 [ hep - ph/0211426 ] ; r. b. appleby and m. h. seymour , jhep * 0309 * ( 2003 ) 056 [ hep - ph/0308086 ] . y. l. dokshitzer and g. marchesini , jhep * 0303 * ( 2003 ) 040 [ hep - ph/0303101 ] . m. dasgupta , hep - ph/0304086 ; r. b. appleby and g. p. salam , hep - ph/0305232 . h. weigert , hep - ph/0312050 . i. h. park _ et al . _ ( amy collaboration ) , phys . lett . * 62 * ( 1989 ) 1713 ; + y. k. li _ et al . _ ( amy collaboration ) , phys . * d 41 * ( 1990 ) 2675 ; + h. j. behrend _ et al . _ ( cello collaboration ) , z. phys . * c44 * ( 1989 ) 63 ; + d. bender _ et al . _ ( hrs collaboration ) , phys . * d 31 * ( 1985 ) 1 ; + d. p. barber _ et al . _ ( mark j collaboration ) , phys . * 43 * ( 1979 ) 830 ; + a. petersen _ et al . _ ( mark ii collaboration ) , phys . d 37 * ( 1988 ) 1 ; + s. bethke _ et al . _ ( mark ii collaboration ) , z. phys . * c43 * ( 1989 ) 325 ; + c. berger _ ( pluto collaboration ) , z. phys . * c12 * ( 1982 ) 297 ; + w. braunschweig _ et al . _ ( tasso collaboration ) , phys . * b214 * ( 1988 ) 286 ; + w. braunschweig _ et al . _ ( tasso collaboration ) , z. phys . * c45 * ( 1989 ) 11 ; + w. braunschweig _ et al . _ ( tasso collaboration ) , z. phys . * c47 * ( 1990 ) 187 ; + i. adachi _ et al . _ ( topaz collaboration ) , phys . * b227 * ( 1989 ) 495 ; + k. nagai _ et al . _ ( topaz collaboration ) , phys . * b278 * ( 1992 ) 506 ; + y. ohnishi _ et al . _ ( topaz collaboration ) , phys . * b313 * ( 1993 ) 475 . d. decamp _ et al . _ [ aleph collaboration ] , phys . b * 255 * , 623 ( 1991 ) . d. decamp _ et al . _ [ aleph collaboration ] , phys . b * 284 * , 163 ( 1992 ) . d. buskulic _ et al . _ [ aleph collaboration ] , z. phys . c * 73 * ( 1997 ) 409 . r. barate _ [ aleph collaboration ] , phys . * 294 * ( 1998 ) 1 . aleph collaboration , `` qcd measurements in @xmath0 annihilations at centre - of - mass energies between 189 and 202 gev '' , aleph 2000 - 044 conf 2000 - 027 . p. abreu _ et al . _ [ delphi collaboration ] , z. phys . c * 73 * , 229 ( 1997 ) . p. abreu _ et al . _ [ delphi collaboration ] , phys . b * 456 * , 322 ( 1999 ) . p. abreu _ et al . _ [ delphi collaboration ] , eur . j. c * 14 * , 557 ( 2000 ) [ hep - ex/0002026 ] . j. abdallah _ et al . _ [ delphi collaboration ] , eur . j. c * 29 * ( 2003 ) 285 [ hep - ex/0307048 ] . o. adriani _ et al . _ [ l3 collaboration ] , phys . b * 284 * , 471 ( 1992 ) . b. adeva _ et al . _ [ l3 collaboration ] , z. phys . c * 55 * ( 1992 ) 39 . o. adriani _ et al . _ [ l3 collaboration ] , phys . rept . * 236 * , 1 ( 1993 ) . m. acciarri _ et al . _ [ l3 collaboration ] , phys . b * 411 * , 339 ( 1997 ) . m. acciarri _ et al . _ [ l3 collaboration ] , phys . b * 371 * , 137 ( 1996 ) . m. acciarri _ et al . _ [ l3 collaboration ] , phys . b * 404 * , 390 ( 1997 ) . m. acciarri _ et al . _ [ l3 collaboration ] , phys . b * 444 * , 569 ( 1998 ) . m. acciarri _ et al . _ [ l3 collaboration ] , phys . b * 489 * , 65 ( 2000 ) [ hep - ex/0005045 ] . p. achard _ et al . _ [ l3 collaboration ] , phys . b * 536 * ( 2002 ) 217 [ hep - ex/0206052 ] . p. d. acton _ et al . _ [ opal collaboration ] , z. phys . c * 55 * ( 1992 ) 1 . d. acton _ et al . _ [ opal collaboration ] , z. phys . c * 59 * , 1 ( 1993 ) . g. alexander _ et al . _ [ opal collaboration ] , z. phys . c * 72 * , 191 ( 1996 ) . k. ackerstaff _ et al . _ [ opal collaboration ] , z. phys . c * 75 * , 193 ( 1997 ) . g. abbiendi _ et al . _ [ opal collaboration ] , eur . j. c * 16 * ( 2000 ) 185 [ hep - ex/0002012 ] . p. a. movilla fernandez , o. biebel , s. bethke , s. kluth and p. pfeifenschneider [ jade collaboration ] , eur . j. c * 1 * , 461 ( 1998 ) [ hep - ex/9708034 ] . o. biebel , p. a. movilla fernandez and s. bethke [ jade collaboration ] , phys . b * 459 * , 326 ( 1999 ) [ hep - ex/9903009 ] . p. pfeifenschneider _ et al . _ [ jade collaboration ] , eur . j. c * 17 * , 19 ( 2000 ) [ hep - ex/0001055 ] . p. a. movilla fernandez , s. bethke , o. biebel and s. kluth , eur . j. c * 22 * ( 2001 ) 1 [ hep - ex/0105059 ] . s. kluth , m. blumenstengel , p. a. movilla fernandez , s. bethke , o. biebel , c. pahl and j. schieck [ jade collaboration ] , hep - ex/0305023 . j. schieck [ for the opal collaboration ] , hep - ex/0312036 . f. yuasa _ et al . _ , prog . * 138 * ( 2000 ) 18 [ hep - ph/0007053 ] . a. pukhov _ et al . _ , hep - ph/9908288 . f. krauss , r. kuhn and g. soff , jhep * 0202 * ( 2002 ) 044 [ hep - ph/0109036 ] . c. adloff _ et al . _ [ h1 collaboration ] , phys . b * 406 * ( 1997 ) 256 [ hep - ex/9706002 ] . c. adloff _ et al . _ [ h1 collaboration ] , eur . j. c * 14 * ( 2000 ) 255 [ addendum - ibid . c * 18 * ( 2000 ) 417 ] [ hep - ex/9912052 ] . h. u. martyn [ h1 collaboration ] , hep - ex/0010046 . s. chekanov _ et al . _ [ zeus collaboration ] , eur . j. c * 27 * ( 2003 ) 531 [ hep - ex/0211040 ] . j. breitweg _ et al . _ [ zeus collaboration ] , phys . b * 421 * ( 1998 ) 368 [ hep - ex/9710027 ] . b. r. webber , nucl . * 71 * ( 1999 ) 66 [ hep - ph/9712236 ] . a. v. manohar and m. b. wise , phys . b * 344 * ( 1995 ) 407 [ hep - ph/9406392 ] . g. p. korchemsky and g. sterman , nucl . b * 437 * ( 1995 ) 415 [ hep - ph/9411211 ] . g. p. korchemsky and g. sterman , hep - ph/9505391 . g. p. korchemsky and g. sterman , nucl . b * 555 * ( 1999 ) 335 [ hep - ph/9902341 ] . g. p. korchemsky and s. tafat , jhep * 0010 * ( 2000 ) 010 [ hep - ph/0007005 ] . b. r. webber , phys . b * 339 * ( 1994 ) 148 [ hep - ph/9408222 ] . y. l. dokshitzer and b. r. webber , phys . b * 352 * ( 1995 ) 451 [ hep - ph/9504219 ] . y. l. dokshitzer , g. marchesini and b. r. webber , nucl . b * 469 * ( 1996 ) 93 [ hep - ph/9512336 ] . r. akhoury and v. i. zakharov , phys . b * 357 * ( 1995 ) 646 [ hep - ph/9504248 ] . y. l. dokshitzer , a. lucenti , g. marchesini and g. p. salam , nucl . b * 511 * ( 1998 ) 396 [ erratum - ibid . b * 593 * ( 2001 ) 729 ] [ hep - ph/9707532 ] . y. l. dokshitzer , g. marchesini and g. p. salam , eur . j. directc * 1 * ( 1999 ) 3 [ hep - ph/9812487 ] . m. dasgupta , l. magnea and g. smye , jhep * 9911 * ( 1999 ) 025 [ hep - ph/9911316 ] . g. e. smye , jhep * 0105 * ( 2001 ) 005 [ hep - ph/0101323 ] . e. gardi and g. grunberg , jhep * 9911 * ( 1999 ) 016 [ hep - ph/9908458 ] . e. gardi , jhep * 0004 * ( 2000 ) 030 [ hep - ph/0003179 ] . c. w. bauer , a. v. manohar and m. b. wise , phys . * 91 * ( 2003 ) 122001 [ hep - ph/0212255 ] . z. trocsanyi , jhep * 0001 * ( 2000 ) 014 [ hep - ph/9911353 ] . m. dasgupta and b. r. webber , eur . j. c * 1 * ( 1998 ) 539 [ hep - ph/9704297 ] . m. dasgupta and b. r. webber , jhep * 9810 * ( 1998 ) 001 [ hep - ph/9809247 ] . g. p. salam and z. trcsnyi , _ limited automated predictor of power suppressed effects ( lappse ) _ , unpublished work . b. r. webber , hep - ph/9510283 . y. l. dokshitzer , `` why it seems too early to report on status of qcd , '' _ given at international school of subnuclear physics : 31th course : from supersymmetry to the origin of space - time , erice , italy , 4 - 12 jul 1993 _ j. m. campbell , e. w. n. glover and c. j. maxwell , phys . * 81 * ( 1998 ) 1568 [ hep - ph/9803254 ] . d. t. barclay , c. j. maxwell and m. t. reader , phys . d * 49 * ( 1994 ) 3480 . m. beneke , phys . b * 307 * ( 1993 ) 154 . p. abreu _ et al . _ ( delphi collaboration ) , z. phys . * c73 * ( 1996 ) 11 . d. buskulic _ et al . _ ( aleph collaboration ) , z. phys . * c55 * ( 1992 ) 209 . i. g. knowles _ , `` qcd event generators , '' hep - ph/9601212 . s. frixione and b. r. webber , jhep * 0206 * ( 2002 ) 029 [ hep - ph/0204244 ] . s. frixione , p. nason and b. r. webber , jhep * 0308 * ( 2003 ) 007 [ hep - ph/0305252 ] . s. gieseke , a. ribon , m. h. seymour , p. stephens and b. webber , hep - ph/0311208 . lep qcd working group , work in progress . h. stenzel , hep - ex/0309061 . s. bethke , nucl . * 121 * ( 2003 ) 74 [ hep - ex/0211012 ] . k. hagiwara _ et al . _ [ particle data group collaboration ] , phys . rev . d * 66 * ( 2002 ) 010001 . r. w. l. jones , m. ford , g. p. salam , h. stenzel and d. wicke , jhep * 0312 * ( 2003 ) 007 [ hep - ph/0312016 ] . p. m. stevenson , phys . d * 23 * ( 1981 ) 2916 . g. grunberg , phys . d * 29 * ( 1984 ) 2315 . s. j. brodsky , g. p. lepage and p. b. mackenzie , phys . d * 28 * ( 1983 ) 228 . p. abreu _ et al . _ [ delphi collaboration ] , phys . b * 418 * ( 1998 ) 430 . g. abbiendi _ et al . _ [ opal collaboration ] , eur . j. c * 21 * ( 2001 ) 411 [ hep - ex/0105046 ] . a. brandenburg , p. n. burrows , d. muller , n. oishi and p. uwer , phys . b * 468 * ( 1999 ) 168 [ hep - ph/9905495 ] . r. barate _ [ aleph collaboration ] , eur . j. c * 18 * ( 2000 ) 1 [ hep - ex/0008013 ] . p. bambade , m. j. coster , j. fuster and p. tortosa , delphi conference note 2003 - 022-conf-642 ( unpublished ) . w. bernreuther , a. brandenburg and p. uwer , phys . * 79 * ( 1997 ) 189 [ hep - ph/9703305 ] . g. rodrigo , a. santamaria and m. s. bilenky , phys . * 79 * ( 1997 ) 193 [ hep - ph/9703358 ] . p. nason and c. oleari , nucl . b * 521 * ( 1998 ) 237 [ hep - ph/9709360 ] . p. a. movilla fernandez , hep - ex/0209022 . t. kluge , hep - ex/0310063 . g. p. salam and g. zanderighi , nucl . * 86 * ( 2000 ) 430 [ hep - ph/9909324 ] . a. heister _ et al . _ [ aleph collaboration ] , eur . j. c * 27 * ( 2003 ) 1 . g. abbiendi _ et al . _ [ opal collaboration ] , eur . j. c * 20 * ( 2001 ) 601 [ hep - ex/0101044 ] . p. abreu _ et al . _ [ delphi collaboration ] , eur . j. c * 13 * ( 2000 ) 573 . g. abbiendi _ et al . _ [ opal collaboration ] , eur . j. c * 23 * ( 2002 ) 597 [ hep - ex/0111013 ] . g. p. korchemsky , hep - ph/9806537 . g. sterman , nucl . b * 281 * ( 1987 ) 310 ; + s. catani and l. trentadue , nucl . b * 327 * ( 1989 ) 323 ; + s. catani and l. trentadue , nucl . b * 353 * ( 1991 ) 183 ; + g. p. korchemsky , mod . lett . a * 4 * ( 1989 ) 1257 ; + g. p. korchemsky and g. marchesini , phys . b * 313 * ( 1993 ) 433 ; + i. i. y. bigi , m. a. shifman , n. g. uraltsev and a. i. vainshtein , int . j. mod . phys . a * 9 * ( 1994 ) 2467 [ hep - ph/9312359 ] ; + r. d. dikeman , m. a. shifman and n. g. uraltsev , int . j. mod . a * 11 * ( 1996 ) 571 [ hep - ph/9505397 ] ; + m. neubert , phys . d * 49 * ( 1994 ) 3392 [ hep - ph/9311325 ] ; + m. neubert , phys . d * 49 * ( 1994 ) 4623 [ hep - ph/9312311 ] ; + g. p. korchemsky and g. sterman , phys . b * 340 * ( 1994 ) 96 [ hep - ph/9407344 ] ; + a. g. grozin and g. p. korchemsky , phys . d * 53 * ( 1996 ) 1378 [ hep - ph/9411323 ] . m. cacciari and e. gardi , nucl . b * 664 * ( 2003 ) 299 [ hep - ph/0301047 ] . g. bozzi , s. catani , d. de florian and m. grazzini , phys . b * 564 * ( 2003 ) 65 [ hep - ph/0302104 ] . l. magnea , hep - ph/0211013 . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the intuitive feeling that hadron matter should be maximally perturbed in the high energy extremely inelastic collisions was a main reason of our effort to consider such processes . it was a hope to observe new dynamical phenomena , or new degrees of freedom , unattainable in other ordinary hadron reactions . this paper presents a first attempt to describe the particularity of considered processes , to give a review of existing models prediction and , at the end , we will offer the field - theoretical formalism for hadron inelastic processes . thus , considering hadrons mean multiplicity @xmath3 as a natural scale of the produced hadrons multiplicity @xmath4 at given cm energy @xmath5 , we would assume that n>>. 0.1at the same time we wish to have n<<n_max=/m , 0.2where @xmath6 gev is the characteristic hadron mass . the last restriction is introduced to weaken the unphysical constraints from the finite , for given @xmath7 , phase space volume . we should assume therefore that @xmath7 is high enough . the multiple production is the process of colliding particles where kinetic energy is dissipated into the mass of produced particles . then one may validate that the entropy @xmath8 accedes its maximum in the domain ( ) since the multiplicity @xmath4 characterizes the rate of stochastization , i.e. the level of incident energy dissipation over existing ( free ) degrees of freedom . there is also another quantitative definition of our reactions . let @xmath9 be the energy of the fastest particle in the given frame and let @xmath10 be the total incident energy in the same frame . then the difference ( @xmath11 ) is the energy spent on production of the less energetic particles . it is useful to consider the inelasticity coefficient = = 1 - 1 . 0.2ait defines the portion of spent energy . therefore , we wish to consider processes with 1 - 1 . 0.2cfollowing ( ) we will assume that 1->>. 0.2dtherefore , the kinetic energy of produced particles in our processes can not be arbitrarily small . using thermodynamical terminology , we wish investigate the production and properties of comparatively ` cold ' multi - hadron ( mostly of @xmath12-mesons ) state . we would like to note from the very beginning that we have only a qualitative scenario of such states which may be produced and the review of the corresponding why s may be considered as the main purpose of this paper . although at the end of this paper ( appendix k ) we will describe principal features of possible field - theoretical solution of our problem . it should be noted the absence of any authentic experimental information concerning discussed processes . moreover , actually the hadron inelastic interactions with a set peculiar to them of unsolved theoretical problems will be considered . nevertheless we suggest to work in this field in spite of these difficulties because the system with extremal properties may be more transparent since the asymptotics always simplify a picture . we would demonstrate this idea and will try to put it in the basis of developed theoretical methods . absence of experimental information about so high inelastic hadron processes is the consequence of the smallness of corresponding cross sections . besides this it was unclear for what purpose the experimental efforts should be done . we would like to convince the reader that the discussed problem is interesting and important . for instance , we will discuss a possibility that asymptotics over @xmath4 may replace in definite sense the asymptotics over @xmath5 . a short address to experimentalists will be given in concluding sec.4.1.2 . we hope that the paper would be useful both for theorists and experimentalists . by this reason the main text of this paper will contain only the qualitative discussion of the problems and results . the quantitative proof , formulation of pure theoretical methods etc . are added in the appendices . considered extremal problem is a good theoretical laboratory and is described in the appendices and theoretical methods may be applied for other physical problems . we would like to point out that the special technique was built for the problem discussed , see sec.2 . _ having the very high multiplicity ( vhm ) state it is natural to use the thermodynamical methods . _ we will offer for this purpose the real - time @xmath0-matrix interpretation of thermodynamics . it can be shown in what quantitative conditions it will coincide with simpler canonical imaginary - time matsubara formalism . we will give also the generalization of the real - time finite - temperature perturbation theory in the case of local temperature @xmath13 distribution , when @xmath14 . this will allow to use the thermodynamical description if the system is far from equilibrium . _ the particle spectrum in the vhm region is soft . it is just a situation when the collective phenomena should be important . _ to describe this phenomena , the decomposition on correlators will be adopted . the origin of this decomposition lie in the mayer s ` group decomposition ' . in multiple production physics this decomposition is known also as the ` multi - component description ' . it based on the idea that the multiple production process may include various mechanisms . in sec.3 we will investigate model predictions for vhm region . we would like to note two main conclusions : existing multiperipheral - type models are unable to describe the vhm region . the infrared region of the pqcd becomes important even if the constraint ( ) is taken into account . in sec.2.1 we will try to formulate the phenomenology of our problem , i.e. the way the vhm processes may be described and what type of phenomena one may expect . the importance of thermodynamical methods will became evident and we will offer in sec.2.2 a general description of the corresponding formalism . it is important to note that we may classify the possible asymptotics over @xmath4 . we will find that there exist only three classes of asymptotics . this will simplify consideration definitely restricting the possibilities . in sec.2.3 we will use the thermodynamical language to give a physical interpretation of these classes of asymptotics . we will see in result that in our choice of the vhm final state this should lead to reorganization of multiple production dynamics : we will get out of the habitual multiperipheral picture in the vhm domain . moreover , one may assume that the semiclassical approximation becomes exact in the vhm domain . this naturally leads to the idea to search such a scheme of calculation which depends on the choice of final state . quantitative description of this idea may be realized as is described in appendix k. the vhm production phenomena includes two sub - problems . first of all it is the dynamical problem of incident energy degradation into the secondary particle energies and the second one is the description of the final state . we will start discussion in sec.2.1.1 from the second part of the problem to explain that the statistical methods are essential for us . in sec.2.1.2 we will try to outline at least qualitatively the main mechanisms of hadron production . the peculiarity of hadron production phenomena consists in the presence of hidden constraints , the consequence of local non - abelian gauge symmetry . the constraints may prevent thermalization and the incident energy dissipation is confused in this case . just the ` confusing ' effect is dominant in the hadron multiple production processes if @xmath15 . the expected change of dominant mechanism of hadron production is discussed sec.2.1.3 . it is important that , in spite of hidden constraints , the system may freely evolve to define the vhm state . such a vhm state should be in equilibrium . formal definition of the ` equilibrium ' notion will be given in sec.2.2.2 . the problem described contains small parameters @xmath16 and @xmath17 . to have the possibility of estimation of contributions in accordance with these parameters one should include them into the formalism . this becomes possible if and only if the integral quantities are calculated . so , the multiplicity @xmath4 is an index only if the multiple production amplitudes @xmath18 are considered . but the cross section @xmath19 is a nontrivial @xmath20 of @xmath4 . we will calculated by this logic mostly integrals of @xmath21 , living out the consideration the amplitudes , see also sec.2.2 , where the first realization of this idea is offered ( the @xmath22 attempt to realize this idea one may find in ) . this is a general methodological feature of our consideration . the multiple production cross section @xmath19 falls down rapidly in the discussed very high multiplicity ( vhm ) domain ( ) and for this reason the multiplicities @xmath23 are not accessible experimentally . at the lhc energy @xmath24 is valid and we will assume that @xmath25 is just the discussed vhm region ( @xmath26 at the lhc energy ) . we will explain later why n~^2 0.1is chosen for the definition of vhm region . generally speaking , having the state of a large number of particles , it is reasonable to depart from an exact definition of the number @xmath4 of created particles , their individual energies @xmath27 , momenta @xmath28 , etc . since they can not be defined exactly by experiment . indeed , for instance , full reconstruction of kinematics is a practically impossible task because of neutral particles , neutrinos , the more so as @xmath29 is considered . we suppose that nothing will happen if @xmath4 is measured with @xmath30 accuracy since @xmath31 is easily attainable in the vhm region . besides it is practically impossible to deal with theory which operates by the @xmath32 ( @xmath33 ! ) variables . artificial reduction of the set of the necessary variables may lead to a temporary success only . indeed , the last thirty years of multiple production physics development was based on the inclusive approach , when the measured quantities ( cross sections ) depend on a few dynamical parameters only . but later on the experiment and its fractal analyses shows that the situation is not so simple , also as , for instance , for the classical turbulence . so , the event - by - event experimental data shows that the particle density fluctuation is unexpectedly large and the fractal dimension @xmath34 is not equal to zero . we know that if the fractal dimension is non - trivial , then the system is extremely ` non - regular ' . so , @xmath350.3 for the perimeter of great britain and @xmath350.5 for norway . the discrepancy marks the fact that the shore of norway is much more broken then of great britain . it is noticeable that the fractal dimension @xmath34 crucially depends on the type of reaction , incident energy and so on . it is evident that one may choose from @xmath32 an arbitrary finite set of variables to characterize the multiple production process . but the fractal analyses shows that such approach would lead to the same effect as if one may hear , for example , only the first violin of mahler s music . so , it is important to understand when the restricted set of dynamical variables will admit to describe the process ( state ) @xmath36 . the same problem was solved in statistical physics , where the ` rough ' description by a restricted number of ( thermodynamical ) parameters is a basis of its success , see the discussion of rough variables description e.g. in the review of uhlenbeck . we will search the same solution desiring to build a complete theory of the vhm hadron reactions . we want to note that just vhm process may be in this sense ` simple ' : at all evidence , the system becomes ` quiet ' in the vhm region and by this reason its ` rough ' thermodynamical description is available . it seems natural , therefore , to start investigation of multiple production phenomena from the ( extremely rare ) vhm processes . the dominant inelastic hadron processes at @xmath15 are saturated by production of low transverse momentum hadrons . one of the approaches explains this phenomena by the nonperturbative effect of quarks created from vacuum . corresponding dynamics looks as follows . at the expanse of transverse kinetic motion color charges may separated at large distances . nevertheless the transverse motion is suppressed since separation leads to increasing polarization of vacuum , because of confinement phenomenon . then , as in qed , the vacuum becomes unstable in regard to the tunneling creation of real fermions . just on their creation , the transverse kinetic energy is spent and , as a result , particles can not have , with exponential accuracy , high transverse energies . this picture is attractive being simple and transparent , but despite numerous efforts , there is no quantitative description of this phenomenon till now . briefly , the problem is connected with the unknown mechanism of strong coloured electric fields formation among distant colored charges , see also . one may use other terms . the soft channel of multiple production means the long - range correlation among hadron coloured constituents . under this special correlation the non - abelian gauge field theory conservation laws constraint were implied . they are important in dynamics since each conservation law decreases the number of independent degrees of freedom at least on one unity ( this may explain why hadrons @xmath37 ) , i.e. it has the nonperturbative effect . moreover , in so - called integrable systems each independent integral of motion ( in involution ) reduces the number of degrees of freedom on two units . as a result there is no thermalization in such systems and the corresponding mean multiplicity @xmath38 should be equal to zero , see appendix k. existence of multiple production , @xmath39 , testifies the statement that the thermalization phenomena exist in hadron processes , i.e. the system of yang - mills fields is not completely integrable . but the most probable process with @xmath15 did not lead to the final state with maximal entropy since @xmath40 , i.e. the definite restrictions on the dissipation dynamics should be taken into account . such problems , being intermediate , are mostly complicated ones . the quantitative theory of this phenomena may lead to deep revision of the main notions of existing quantum field theory , see appendix k. so , the dynamical display of hidden conservation laws of the hadron system are probably unstable since we expect that the system is not completely integrable , solitary field configurations @xmath41 . then the quantum theory should be able to describe quantum excitations of these fields , i.e. to count the fluctuations of ` curved ' manifolds . the canonical perturbation theory methods , formulated in terms of creation and annihilation of particles in the external field @xmath41 , are too complicated , see and references cited therein . for this reason , existing calculations usually do not exceed the semiclassical approximation . we hope that , as described in appendix k , the quantization scheme would be able to solve this problem ( see also the example described there ) . another approach assumes that the special ` @xmath42-channel ' ladder type feynman diagrams are able to describe the @xmath15 region . this approach did not take into account the confinement nonperturbative effects introducing the hardly controllable supposition that the free quarks and gluons may form a complete set of states . formally this is right , but at all evidence , the decomposition on this fock basis is realized in the non - abelian gauge field theories on zero measure ( see appendix k ) . nevertheless one may reject this argument assuming that the process is happening on a sufficiently small distances . corresponding contribution came from the so called ` hard pomeron ' . but the intrinsic problems of the accuracy of chosen logarithmic approximations , the understanding of the so called non - logarithmic corrections , of the fate of the infrared divergences stay unsolved till now . so , the multiple production of soft hadron phenomena seems unsolvable on the day - to - day level of understanding and we may with pure conscience move it away . this is why the hadron inelastic reactions lost some popularity , migrating the last two decades to the class of ` non - interesting ' problems . yet , in a number of modern fundamental experiments , multiple production plays , at least , the role of background to the investigated phenomena and for this reason we should be ready for the quantitative estimation of it . our hope to describe such a complicated problem as the multiple production phenomenon in the vhm conditions is based on the following idea . at the very beginning of this century , a couple p. and t.euhrenfest , had offered a model to visualize boltzmann s interpretation of the irreversibility phenomena in statistics . the model is extremely simple and fruitful . it considers the two boxes with @xmath43 numerated balls . choosing the label of the balls @xmath44 one must take the ball with the corresponding label from one box and put it into another one . one may repeat this action an arbitrary number of times @xmath42 . starting from the highly ` nonequilibrium ' state with all balls in one box , @xmath45 , it is seen stationary with @xmath42 tendency to equalization of the number of balls in the boxes fig.1 . the stationarity means that the number of balls in the other box rises @xmath46 at least on an early stage of the process . this signifies presence of an @xmath47 flow ( of balls ) toward the preferable ( equilibrium ) state . one can hope that this model reflects a physical reality of nonequilibrium processes with initial state very far from equilibrium . a theory of such processes with ( irreversible ) flow toward a state with maximal entropy should be sufficiently simple being close to the stationary markovian . the vhm production process may be , at least on an early stage , stationary markovian . if this is so then one may neglect long - range effects , non - perturbative as well , since they are not markovian as follows from the experience described in sec.2.1.2 . this is possible if the vhm process is happening so fast ( being the short - range phenomenon ) that the confinement forces became ` frozen ' . it can be shown that the quantitative reason for this phenomena is a fast ( exponential ) reduction with @xmath4 of the soft channel contribution into the hadron production process . so , we expect a change of the multiple production dynamics in the vhm region . thus , the main input idea consists from two general propositions . the first of them is following : ( * i * ) _ the hadron vhm production processes should be close to the stationary markovian . _ ` freezing ' the confinement constraints , the entropy @xmath8 may exceed for given energy @xmath5 its available maximum in the vhm domain . then one can assume that the vhm final state is in ` equilibrium ' , or is close to it . so , ( * ii * ) _ the vhm final state should be close to equilibrium _ + is our second basic proposition . we would select and appreciate particle physics models in accordance with these propositions . the field - theoretical description of statistical systems at a finite temperature is based usually on the formal analogy between imaginary time and inverse temperature @xmath48 . this analogy is formulated by schwinger as the ` euclidean postulate ' and it assumes that ( i ) the system is in equilibrium , i.e. it should allow the arbitrary rearrangement of states of temporal sequence in the described process , and ( ii ) there is not special space - time long - range correlations among states of the process , i.e , for instance , the symmetry constraints should not play a crucial role . we do not know _ ad hoc _ whether or not to apply the ` euclidean postulate ' for given @xmath4 and @xmath7 , even if ( ) is satisfied . for this reason we are forced to formulate the theory in the natural real - time terms . the first important quantitative attempt to build the real - time finite - temperature field theory discovered the formal problem of the so called ` pinch - singularities ' . further investigation of the theory has allowed to demonstrate the cancellation mechanism of these unphysical singularities . this is attained by doubling the degrees of freedom : the green functions of the theory represent @xmath49 @xmath50 @xmath49 matrix . it surely makes the theory more complicated , but the operator formalism of the thermo - field dynamics shows the unavoidable character of this complication . notice that the canonical real - time finite temperature field - theoretical description of the statistical systems based on the kubo - martin - schwinger ( kms ) boundary condition for a field is : ( t)=(t - i ) . 22.1 it , without fail , leads to the @xmath51 fluctuation - dissipation conditions ( see also ) . by this reason it can not be applied in our case , where the dissipation problem is solved . the origin of this boundary condition is shown in appendix a. we will use a more natural microcanonical formalism for particle physics . the thermodynamical ` rough ' variables are introduced in these approach as the lagrange multipliers of corresponding conservation laws . the physical meaning of these ` rough ' variables are defined by the corresponding equations of state . we shall use the @xmath0-matrix approach which is natural for the description of the time evolution . ( the @xmath0-matrix description was used also in . ) for this purpose the amplitudes = a_nm(p , q ) 22.2 of the @xmath52- into @xmath4-particles transition will be introduced . the in- and out - states must be composed from mass - shell particles . using these amplitudes we will calculate & r_nm(p , q)=|a_nm(p , q)|^2= + & = < p_1,p_2, ... ,p_n|q_1,q_2, ... ,q_m > < q_1,q_2, ... ,p_n>. 22.3this will lead to the doubling of the degrees of freedom . the temperature description will be introduced ( see also ) noting that , for instance , & d_n=|a_nm(p , q)|^2d_n(p ) , + & d_n(p)=_1^n , ( p)= ( p^2 + m^2)^1/2 , 22.4 is the differential measure of the final state . it is a first example where the usefulness of the probability - like quantity @xmath53 is seen . the measure ( ) is defined on the energy - momentum shell : _ i=1^np_i = p . 22.4ait should be underlined that @xmath54 are the translationally invariant amplitudes and four equalities : _ i=1^np_i=_i=1^mq_i 22.4bare obeyed identically . so , ( ) are the constraints and to take them into account one may multiply @xmath55 on @xmath56 where @xmath57 is the time - like 4-vector . it is evident that integration over @xmath57 with factor @xmath58 gives the constraints ( ) . one may simplify the calculation assuming that all calculations are performed , for example , in the cm frame @xmath59 . then one may ignore the space components considering @xmath60 . this is the equivalent of the assumption that only the energy conservation law is important . the last step is the substitution @xmath61 , where @xmath62 is our lagrange multiplier . to define its physical meaning one should solve the equation of state : e =- d_n_k=1^ne^-(p_k)-d_n ( ) . 3such a definition of temperature as the lagrange multiplier of the energy conservation law is obvious for microcanonical description . the initial - state temperature will be introduced in the same way , taking into account ( ) . so , we will construct the two - temperature theory . it is impossible to use the kms boundary condition in such a two temperatures description ( the equation of state can be applied at the very end of the calculations ) . it should be notice that the ` density matrix ' @xmath63 , defined in ( [ 22.3 ] ) , describes the ` closed - path motion ' in the functional space . so , if = < n , out|e^is(_+)|m , in > o1and ^*= < q_1,q_2, ... ,p_n>= + = < m , in|e^-is(_-)|n , out > o2then , by definition , _+(_)=_-(_)= ( _ ) , o3with some ` turning - point ' fields @xmath64 , where @xmath65 is the remote hypersurface . the value of @xmath66 specifies the environment of the system . we will show that ( ) coincides with the kms boundary condition in some special cases . here consequences of the vacuum boundary condition : (_)=0 03aare analyzed . one should admit also that below boundary conditions are not unique : one can consider arbitrary organization of the environment of the considered system . the @xmath0-matrix interpretation is able to show the way as an arbitrary boundary condition may be adopted . this should extend the potentialities of the real - time finite - temperature field - theoretical methods . it seems useful to illustrate the above microcanonical approach by the simplest example , see also . by definition , the @xmath4 particles production cross section _ n(s)=d_n(p)(q_1+q_2-^n_i=1p_i)|a_n(p , q)|^2 , 18where @xmath67 is the ordinary @xmath4 particle production amplitude in accelerator experiments . considering the fourier transform of energy - momentum conservation @xmath68-function one can introduce the generating function @xmath69 , see and references cited therein . we may find in the result that @xmath70 is defined by the equality : _ n(e)=^+i_-ie^e_n ( ) , e=(q_1)+(q_2 ) , 3awhere _ n()= \{^n_i=1 } |a_n|^2=d_n ( ) . 19most probable value of @xmath62 in ( ) is defined by the equation of state ( ) . inserting ( ) into ( ) we find the expression ( ) if the momentum conservation shell is neglected . the last one is possible since the cross sections are always measured in the definite frame . let us consider the simplest example of noninteracting particles : @xmath71 where @xmath72 is the bessel function . inserting this expression into ( ) we can find that in the nonrelativistic case ( @xmath73 ) @xmath74 i.e. , we find the well known equality : e_kin = t , o4where @xmath75 is the mean kinetic energy and @xmath76 is the temperature ( boltzmann constant was taken equal to one ) . it is important to note that the equation ( ) has a unique real solution @xmath77 rising with @xmath4 and decreasing with @xmath7 . the expansion of the integral ( ) near @xmath77 unavoidably gives an asymptotic series with zero convergence radii since @xmath78 is the essentially nonlinear function of @xmath62 , see also sec.2.2.2 . this means that , generally speaking , fluctuations in the vicinity of @xmath77 may be arbitrarily high and in this case @xmath77 has no physical sense . but if the fluctuations are gaussian , then @xmath78 coincides with the partition function of the @xmath4 particle state and @xmath77 may be interpreted as the inverse temperature . we will put the observation of this important fact in the basis of our thermodynamical description of the vhm region . the notion of ` equilibrium ' over some parameter @xmath79 in our understanding is a requirement that the fluctuations in the vicinity of its mean value , @xmath80 , have a gaussian character . notice , in this case , one can use this variable for a ` rough ' description of the system . we would like to show now that the corresponding equilibrium condition would have the meaning of the correlations relaxation condition of bogolyubov , see also . let us define the conditions when the fluctuations in the vicinity of @xmath81 are gaussian . firstly , to estimate the integral ( ) in the vicinity of the extremum , @xmath81 , we should expand @xmath82 over @xmath62 : _ n(+_c)=_n(_c)-+^2 _ n(_c)-^3_n(_c)+ ... 21cand , secondly , expand the exponent in the integral ( ) over , for instance , @xmath83 etc . in the result , if higher terms in ( ) are neglected , @xmath84-th term of the perturbation series _ n , k~\{}^k ( ) . 22ctherefore , because of euler s @xmath85 function , the perturbation theory near @xmath81 leads to the asymptotic series . the supposition to define this series formally , for instance , in the borel sense is not interesting from physical point of view . indeed , such a formal solution assumes that the fluctuations near @xmath81 may be arbitrarily high . then , by this reason , the value of @xmath81 loses its significance : arbitrary values of @xmath86 are important in this case . nevertheless it is important to know that our asymptotic series exists in some definite sense , i.e. we can calculate the integral over @xmath62 by expanding it over @xmath86 . therefore , if the considered series is asymptotic , we may estimate it by first term if /^3_c < < ( ^2_n(_c)/^2_c)^3/2 . 23cone of the possible solutions of this condition is /^3_c0 . 24cif this condition is satisfied , then the fluctuations are gaussian with dispersion @xmath87 see ( ) . let us consider now ( ) carefully . we will find by computing derivatives that this condition means the following approximate equality : -3 + 2 0 , 25cwhere @xmath88 means the @xmath84-th derivative . for identical particles , ^(k)_n(_c)=n^k(-1)^kd_n(_c)^k_i=1(q_i ) , 26ctherefore , the left hand side of ( ) is the 3-point correlator @xmath89 since @xmath90 is a density of states for given @xmath62 : & k_3d_3(q)(<^3_i=1(q_i)>__c- 3<^2_i=1(q_i)>__c < ( q_3)>__c+ . + & + .2 ^ 3_i=1<(q_i)>__c ) , 27cwhere the index @xmath81 means that averaging is performed with the boltzmann factor @xmath91 . notice , in distinction with bogolyubov , @xmath89 is the @xmath92 correlation function . so , in our interpretation , one can introduce the notion of temperature @xmath93 if and only if the @xmath94 energy flows , measured by the corresponding correlation functions , are to die out . as a result , to have all the fluctuations in the vicinity of @xmath81 gaussian , we should have @xmath95 , @xmath96 . notice , as follows from ( ) , the set of minimal conditions actually looks as follows : k_m<<k_2 , m3 . 27dif the experiment confirms this conditions then , independently from the number of produced particles , the final state may be described with high enough accuracy by one parameter @xmath81 and the energy spectrum of particles is gaussian . in this conditions one may return to the statistical and the hydrodynamical models . considering @xmath81 as a physical ( measurable ) quantity , we are forced to assume that both the total energy of the system , @xmath97 , and the conjugate to it , variable @xmath81 , may be measured simultaneously with high accuracy . we would like to show now that the ordinary big partition function of the statistical system coincide with _ n , m_(_1,z_1;_2,z_2 ) r_nm(p , q)=(,z ) , densmwhere @xmath98 is defined by ( ) . the summation and integration are performed with constraints that the mean energy of particles in the initial(final ) state is @xmath99 . on may interpret @xmath100 in the first approximation as the temperature and @xmath101 as the activity for initial(final ) state . direct calculation , see appendix b , gives the following expression for generating functional : ( , z)=e^-i*n*(_i^*_j)r_0 ( ) , 22.22awhere the particle number operator ( @xmath102 ) & * n*(_i^*_j)=- dxdx(_+(x)d_+-(x - x,_2,z_2)_-(x)- + & -_-(x)d_-+(x - x,_1,z_1)_+(x ) ) p22and r_0()=z(_+)z^ * (-_-) , p23where @xmath103 is defined in ( ): @xmath104 and , for the vacuum boundary condition @xmath105 , d_+-(x - x,,z)=-id_1(q)e^iq(x - x)e^-(q)z(q ) 22.23a + d_-+(x - x,,z)=id_1(q)e^-iq(x - x)e^-(q)z(q ) 22.24aare respectively the positive and negative frequency correlation functions at @xmath106 . it is evident , r_nm(p , q)=. _ k=1^n\{e^_1(p_k ) } _ k=1^m\{e^_2(q_k ) } ( , z)|_z_i=0 . p2notice , defining @xmath98 through the generating functional we extract the boltzmann factors @xmath107 since the energy - momentum conservation @xmath68-functions were extracted from amplitudes @xmath54 . we suppose that @xmath103 may be computed perturbatively . as a result , ( @xmath108 is the variational derivative ) r(,z)= e^-iv(-i_+)+iv(-i_- ) e^ dx dxj_i ( x)d_ik(x - x;,z)j_k ( x ) , 22.29awhere @xmath109 is the matrix green function . these green functions are defined on the mills time contours @xmath110 in the complex time plane ( @xmath111 ) , see fig.2 . this definition of the time contours coincides with the keldysh time contour . the generating functional ( ) has the same structure as the generating functional of niemi and semenoff . the difference is only in the definition of green functions @xmath112 . this choice is a consequence of the boundary condition ( ) . so , if ( ) is used , then the green function is defined by eq . notice also that if @xmath113 then a new green function obey kms boundary condition , see ( ) . following niemi and semenoff one can write ( ) in the form : ( ) = d_nse^is_ns( ) , nisewhere the functional measure @xmath114 and the action @xmath115 are defined on the closed complex time contour @xmath116 , see fig.3 . the choice of initial time @xmath117 and @xmath118 is arbitrary . then one can perform shifts : @xmath119 and @xmath120 . in result , ( * i * ) if @xmath113 , ( * ii * ) if contributions from imaginary parts @xmath121 and @xmath122 of the contour @xmath116 have disappeared in this limit , ( * iii * ) if the integral ( ) may be calculated perturbatively then this integral is a compact form of the representation ( ) . notice that the requirements ( * i * ) - ( * iii * ) are the equivalent of the euclidean postulate of schwinger . in this frame one can consider another limit @xmath123 . then the @xmath116 contour reduces to the matsubara imaginary time contour , fig.4 . later on we will use this @xmath0-matrix interpretation of thermodynamics . but one should take in mind that corresponding results will hide assumptions ( * i * ) - ( * iii * ) . we would like to mention the ambivalent role of external particles in our @xmath0-matrix interpretation of thermodynamics . in the ordinary matsubara formalism the temperature is measured assuming that the system under consideration is in equilibrium with the thermostat , i.e. temperature is the energy characteristics of @xmath124 particles . in our definition the temperature is the mean energy of produced , i.e. non - interacting , particles . it can be shown that both definitions lead to the same result . explanation of this coincidence is the following . let us consider the point of particle production as the coordinate of fictitious ` particle ' . this ` particle ' interacts since the connected contributions into the amplitudes @xmath125 only are considered , and has the equal to produced particles momentum and so on . the set of these ` particles ' form a system . interaction among these ` particles ' may be described by corresponding correlation functions , see sec.2.3.3 . let us consider now the limit @xmath123 . in this limit ( ) reduces to ( ) = d_me^-s_m( ) , matswhere the imaginary time measure @xmath126 and action @xmath127 are defined on the matsubara time contour . the periodic boundary condition ( ) should be used calculating integral ( ) . the rules and corresponding problems in the integral ( ) can be calculated as described in many textbooks , see also . in the limit considered the time was eliminated in the formalism and the integral in ( ) performed over all states of the ` particles ' system with the weight @xmath128 . notice , the doubling of degrees of freedom has disappeared and our fictitious ` particles ' became real ones . on the other hand , the produced particles may be considered as the probes through which we measure the interacting fields . as was mentioned above , their mean energy defines the temperature , if the energy correlations are relaxed . if even one of the conditions ( * i * ) - ( * iii * ) is not satisfied then one can not reduce our @xmath0-matrix formalism to the imaginary time matsubara theory . then one can ask : is there any possibility , staying in the frame of @xmath0-matrix formalism , to conserve the statistics formalism . this question is discussed in the appendix c , where the wigner functions approach is applied . it may be shown that the formalism may be generalized to describe the kinetic phase of the nonequilibrium process , where the temperature should have the local meaning . the comparison with the ` local equilibrium hypothesis ' is discussed at the end of appendix c. our further consideration will based on the model independent ( formal ) classification of asymptotics . we will consider the generating function : t(s , z)=_n=1^n_max z^n_n(s ) , s=(p_1+p_2)^2>>m^2 , n_max=/m . 21.3this step is natural since the number of particles is not conserved in our problem . so , the total cross section and the averaged multiplicity will be : _ tot(s)=t(s,1)=_n_n(s ) , = _ n n ( _ n(s)/_tot)= .t(s , z)|_z=1 . 21.4 at the same time , the inverse mellin transform gives _ n=.t(s , z)|_z=0= t(s , z)= e^(-nz+t(s , z ) ) . 21.5the essential values of @xmath129 in this integral are defined by the equation ( of state ) : n = zt(z , s ) . 21.6taking into account the definition of the mean multiplicity @xmath3 , given in ( ) , we can conclude that the solution of ( ) @xmath130 is equal to one at @xmath131 . therefore , @xmath132 are essential in the vhm domain . the asymptotics over n ( @xmath133 is assumed ) are governed by the smallest solution @xmath130 of ( ) because of the asymptotic estimation of the integral ( ): _ n(s)e^-nz_c(n , s ) . 21.7 let us assume that in the vhm region and at high energies , @xmath134 , there exist such a value of @xmath135 that we can neglect in ( ) the dependence on the upper boundary @xmath136 . this formal trick with the thermodynamical limit allows to consider @xmath137 as the nontrivial function of @xmath129 for finite @xmath7 . then , it follows from ( ) that z_c(n , s)z_s at n , 21.8where @xmath138 is the leftmost singularity of @xmath137 in the right half plane of complex @xmath129 . one can say that the singularity of @xmath137 attracts @xmath135 if @xmath139vhm . we will put this observation in the basis of vhm processes phenomenology . we would like to underline once more that actually @xmath137 is regular for arbitrary finite @xmath129 if @xmath7 is finite . but @xmath135 behaves in the vhm domain as if it is attracted by the ( imaginary ) singularity @xmath138 . and just this @xmath135 defines @xmath70 in the vhm domain . we want to note that actually the energy @xmath5 should be high enough to use such an estimation . one can notice from the estimation ( ) that @xmath70 weakly depends on the character of the singularity . therefore it is enough to classify only the possible positions of @xmath138 . we may distinguish following possibilities : & ( * a * ) z_s= : _ n < o(e^-n ) + & ( * b * ) z_s=1 : _ n > o(e^-n ) + & ( * c * ) 1<z_s < : _ n = o(e^-n ) , 21.9i.e . , following this classification , the cross section may decrease faster ( * a * ) , slower ( * b * ) , or as ( * c * ) an arbitrary power of @xmath140 . it is evident , if all these possibilities may be realized in nature , then we should expect the asymptotics ( * b * ) . as was explained in sec.2.2.1 , @xmath70 has the meaning of the @xmath4 particle partition function in the energy representation . then @xmath137 should be the ` big partition function ' . taking this interpretation into account , as follows from lee - yang theorem , @xmath137 can not be singular at @xmath141 . at the same time , the direct calculations based on the physically acceptable interaction potentials give the following restriction from above : ( * d * ) _ n < o(1/n ) 21.2this means that @xmath70 should decrease faster than any power of @xmath142 . it should be noted that our classification predicts rough ( asymptotic ) behavior only and did not exclude local increase of the cross section @xmath70 . one may notice that -=z_c(n , s)+o(1/n ) . 21.10using thermodynamical terminology , the asymptotics of @xmath70 is governed by the physical value of the activity @xmath135 . one can introduce also the chemical potential @xmath143 . it defines the work needed for one particle creation , @xmath144 , where @xmath145 is the produced particles mean energy . so , one may introduce the chemical potential if and only if @xmath146 and @xmath135 may be used as the ` rough ' variables . then the above formulated classification has a natural explanation . so , ( * a * ) means that the system is stable with reference to particle production and the activity @xmath135 is the increasing function of @xmath4 , the asymptotics ( * b * ) may realized if and only if the system is unstable . in this case @xmath135 is the decreasing function . the asymptotics ( * c * ) is not realized in equilibrium thermodynamics . we will show that the asymptotics ( * a * ) reflects the multiperipheral processes kinematics : created particles form jets moving in the cm frame with different velocities along the incoming particles directions , i.e. with restricted transverse momentum , see sec.3.1.1 . the asymptotics ( * b * ) assumes the condensation - like phenomena , see sec.3.3 . the third type asymptotics ( * c * ) is predicted by stationary markovian processes with the pqcd jets kinematics , see sec.3.2.2 . the dis kinematics may be considered as the intermediate , see sec.3.2.1 . this interpretation of classes ( ) allows to conclude that we should expect reorganization of production dynamics in the vhm region : the soft channel ( * a * ) of particle production should yield a place to the hard dynamics ( * c * ) , if the ground state of the investigated system is stable with reference to the particle production . otherwise we will have asymptotics ( * b * ) . let us consider the system with several correlation scales . for example , in statistics one should distinguish correlation length among particles ( molecules ) and correlation length among droplets if the two - phase region is considered . in particle physics , one should distinguish in this sense correlation among particles produced in result of resonance decay and correlations among resonances . in pqcd one may distinguish correlations of particles in jet and correlation among jets . there exist many model description of this physical picture . in statistics mayer s group decomposition is well known . in particle physics one should note also the many - component formalism . we will consider the generating functions ( functionals ) formalism considering mostly jet correlations . in many respects it overlaps the above mentioned approaches . the generating function @xmath137 may be written in the form : t(z , s)=_k=1^c_k(s)= _ l=1^z^lb_l , g1where the coefficients @xmath147 are the moments of the multiplicity distribution p_n(s)=_n(s)/_tot(s ) . g2so , c_1(s)=_n np_n(s)= , c_2(s)=_n n(n-1)p_n(s)-^2 g3and so on . using the connection with the inclusive distribution functions @xmath148 : t(z , s)=_k=1^d_k(q ) f_k(q_1,q_2, ... ,q_k;s ) , g4it is easy to find that & c_1(s)=d_1(q ) f_1(q;s)=|f_1(s ) , + & c_2(s)=d_2(q ) \{f_2(q_1,q_2;s)-f_1(q_1;s)f_1(q_2;s)}= _ \{k}_l=0^(_i=1^lk_i - k ) _ i=1^l\ { } , g6where @xmath149 and @xmath150 one may invert formulae ( ): |f_k(s)= _ \{n_k}=0^(_k=1^kn_k - l ) _ k=1^()^n_k g7 the mayer s group coefficients @xmath151 in ( ) have the following connection with @xmath147 : b_l(s)=_k=0^c_k+l(s ) . g23 it seems useful to illustrate the effectiveness of the generating function method by the following example . we will consider the transformation ( multiplicity @xmath152 activity @xmath129 ) to show the origin of the koba - nielsen - olesen scaling ( kno - scaling ) . if @xmath153 , @xmath154 , then @xmath70 is described by the poisson formulae : _ k8it corresponds to the case of absence of correlations . let us consider more weak assumption : c_m(s)=_m(c_1(s))^m , k9where @xmath155 is the energy independent constant , see also , where a generalization of kno scaling on the semi - inclusive processes was offered . then t(z , s)=_m=1\{(z-1)|n(s)}^m . k10to find the consequences of this assumption , let us find the most probable values of @xmath129 . the equation of state @xmath156 has solution @xmath157 increasing with @xmath4 since @xmath137 is an increasing function of @xmath129 , if and only if , @xmath137 is nonsingular at finite @xmath129 . as was mentioned above , the last condition has deep physical meaning and practically assumes the absence of the first order phase transition . let us introduce a new variable : = ( z-1)|n(s ) . k12the corresponding equation of state looks as follows : = ( 1+)t ( ) . k13so , with @xmath158 accuracy , one can assume that _ c(n/|n(s ) ) . k14are essential . it follows from this estimation that such scaling dependence is rightful at least in the neighborhood of @xmath106 , i.e. in vicinity of main contributions into @xmath159 . this gives : |n(s)_n(s)=_tot(s)(n/|n(s ) ) , k15where ( n/|n(s))t(_c(n/|n(s)))\{n/|n(s)_c(n/|n(s))}o(e^-n ) k16is the unknown function . the asymptotic estimation follows from the fact that @xmath160 should be a nondecreasing function of @xmath4 , as follows from nonsingularity of @xmath137 . the estimation ( ) is right at least at @xmath161 . the range validity of @xmath4 , where solution of ( ) is acceptable , depends from exact form of @xmath137 . indeed , if @xmath162 , @xmath163 , then ( ) is right at all values of @xmath4 and it is enough to have the condition @xmath161 . but if @xmath164 , @xmath163 , then ( ) is acceptable if and only if @xmath165 . representation ( ) shows that just @xmath38 is the natural scale of multiplicity @xmath4 . this representation was offered first as a reaction on the so called feynman scaling for inclusive cross section : f_k(q_1,q_2, ... ,q_k)~^k_i=1 . k17 as follows from estimation ( ) , the limiting kno prediction assumes that @xmath166 . in this regime @xmath137 should be singular at @xmath167 . the normalization condition @xmath168 gives : @xmath169 , where @xmath170 is the constant . notice , such behavior of the big partition function @xmath137 is natural for stationary markovian processes described by logistic equations . in the field theory such equation describes the qcd jets . we wish to generalize expansion ( ) to take into account the possibility of many - component structure of the multiple production processes . let us consider particle production through the generation , for instance , of jets . in this case decay of a particle of high virtuality @xmath171 forms a jet of lower virtuality particles . it is evident that one should distinguish correlation among particles in the jet , and correlation among jets . let @xmath172 be the @xmath173 that @xmath174-th jet of mass @xmath175 includes @xmath176 particles , @xmath177 , where _ i=1^n_jn_i = n g8the jets are the result of particles decay . then let us assume that @xmath178 defines the mean number of jets of mass @xmath175 and momentum @xmath179 : & ^(1)_n()=_n_j_\{n}_n_j ( _ i=1^n_jn_i - n ) + & _ i=1^n_j \ { e^-(p_i)|n_1(m_i , p_i)_n_i(m_i ) } , g9where @xmath180 . notice the boltzmann factor @xmath107 , where @xmath181 is the jets energy , play the same role as the corresponding factor in ( ) and introduced to take into account the energy conservation law . we consider the vhm domain and for this reason @xmath182 is assumed . it is useful to avoid the particles number conservation law ( ) . for this purpose we will introduce ^(1)(,z)=_n^(1)_n()=\ { e^-(p)|n_1(m , p)(t(z , m)-1 ) } , g10where t(z , m)=_n z^n_n(m ) . g11comparing ( ) with ( ) we may conclude that @xmath183 plays the role of activity of jets . then the generalization is evident : & ( , z)=_k^(k)(,z)= + & = \{_k_i=1^k\{dm_id_1(p_i ) e^-(p_i)(t(z , m_i)-1)}. + & .|n_k(m_1,p_1, ... ,m_k , p_k ) } , g12where @xmath184 has the same meaning as @xmath147 , i.e. @xmath184 is the correlation function of @xmath84 jets . let us consider the following ` bootstrap ' regime when @xmath185 is defined by the equation : ( , z ) e^-(p)|n_1(m , p)t(z , m ) . g13inserting here the strict expression ( ) we find a nonlinear equation for @xmath183 . the solution of ( ) assumes that e^-(p)|n_1 t > > \{e^-(p ) |n_1t}^2 g14and e^-|n_1 t > > ( e^-t ) ( e^-t ) |n_2 g15 to solve eq . ( ) in the vhm region , where the leftmost singularity over z is important , let us consider the @xmath186 : t(z , m)= , _ 0>0 , g16where @xmath187 is the polynomial function of @xmath129 , @xmath188 . using the normalization condition : |n_j=.t(z , m)|_z=1 g17we can find : a(m)_0=|n_j-(1,m ) , (1,m).(z , m)|_z=1 . g18 the partition function of the jet @xmath183 defined by @xmath186 ( ) is singular at z_s(m)=1 + . g19this singularity would be significant in the vhm region if @xmath189 is decreasing function of @xmath52 . this means an assumption that @xmath190 so , in first approximation we will choose a(m)=|n_j(m ) . g20this choice may be confirmed by concrete model calculations . taking into account the energy conservation law , conditions ( ) and ( ) are satisfied if \{-n } < g21at @xmath139vhm . therefore , ( ) obey eq . ( ) with exponential accuracy in the vhm region , i.e. if n > > = . g22we assume here that one can find so large @xmath4 and @xmath7 that with exponential accuracy the factors @xmath191 did not play an important role . but at low energies the condition ( ) is important and the factors @xmath191 should be taken into account . notice now important consequence of our ` bootstrap ' solution : it means that we can leave production of the heavy jets only , if @xmath192 . on the other hand , let us choose @xmath193 , where @xmath194 is the function of @xmath7 , and consider @xmath195 . then the condition ( ) defines @xmath196 : if z_0 > > g22then we are able to obey the inequalities ( ) and ( ) . the jet mean multiplicity , see sec.3.2.2 , |n_j(s)~. g26then = \{e^/-1}^-1~<<|n_j(s ) . g50at @xmath195 . therefore , ( ) may be satisfied outside the vhm domain . let us compare now the solutions of the equation of state . inserting ( ) into ( ) we can find for a jet of mass @xmath5 that z^1_c = z^1_s-=1 + - . g24the two - jet contribution of the masses @xmath197 gives : z^2_c = z^2_s-=1 + - . g25at arbitrary finite energies @xmath198 and , as follows from ( ) , they decrease @xmath199 with energy . noting the normalization condition , @xmath200 , and assuming that the vacuum is stable , i.e. @xmath201 , we can conclude that if @xmath202 then @xmath138 attracts @xmath130 , i.e. @xmath203 , and if @xmath204 then this contributions should be significant in @xmath159 ; if @xmath195 , then @xmath204 , and if @xmath4 satisfy the inequality ( ) , or if @xmath196 satisfy the inequality ( ) , then considered contributions are significant in @xmath159 . + it is the ( energy - multiplicity ) asymptotics equivalence principle . one of the simplest consequences of this principle is the prediction that the mean transverse momentum of created particles should increase with multiplicity at sufficiently high energies . this principle is the consequence of independence of contributions in the vhm domain on the type of singularity in the complex @xmath129 plane and of the energy conservation law . just the last one shifts the two - jet singularity to the right and @xmath205 . we would like to notice also that this effect , when the mostly ` energetic population ' survives was described mathematically by v.volterra . intuitively evident is that one may find the ` energetic population ' searching the vhm one , or , it is the same , giving it a rich supply , i.e. to give the population enough energy . this is our ( energy - multiplicity ) equivalence ( ( @xmath206)-equivalence ) principle . notice , if the amount of supply is too high then few populations may grow . this is the case when the difference @xmath198 tends to zero at high energies . we would like to note that singular at finite @xmath129 partition functions was predicted in the @xmath207-theory , in qcd jets , in the generalized bose - einstein distribution model . in all of this models decay of the essentially nonequilibrium initial state ( highly virtual parton , heavy resonance , etc ) was described . one may distinguish the phases of the media by a characteristic correlation length . then the phase transition may be considered as the process of changing correlation length . our ` bootstrap ' solution predicts just such phenomena : at low multiplicities the long - range correlations among light jets is dominant . the ` bootstrap ' solution predicts that for vhm processes just the short - range correlations among particles of the heavy jet become dominant . the ( @xmath206)-equivalence means that this transition is a pure dynamical effect . multiple production phenomena was first observed more than seventy years ago . during this time a vast experimental information was accumulated concerning hadron inelastic interactions , see the review papers . now we know that at high energy @xmath208 : + ( i ) the total cross section @xmath209 of hadron interactions is enhanced almost completely by the inelastic channels ; + ( ii ) the mean multiplicity of produced hadrons @xmath3 slowly ( logarithmically ) grows with @xmath208 ; + ( iii ) the interaction radii of hadrons @xmath210 slowly ( logarithmically ) increase with energy . ; + ( iv ) the multiplicity distribution @xmath19 is wider than the poisson distribution . ; + ( v ) the mean value of transverse momentum @xmath84 of produced hadrons is restricted and is independent of the incident energy @xmath5 and produced particle multiplicity @xmath4 ; + ( vi ) the one - particle energy spectrum @xmath211 . first of all , ( i ) means that the high energy hadron interaction may be considered as the ordinary dissipation process . in this process the kinetic energy of incident particles is spent in produced particle mass formation . the vhm process takes place in the vacuum and then it was assumed on the early stages that the multiple production phenomena reflects a natural tendency of the excited hadron system to get to equilibrium with the environment . in this way one can introduce as a first approximation the model that the excited hadron system evolves without any restrictions . in this model we should have @xmath212 . the dissipation is maximal in this case and the entropy @xmath8 exceeds its maximum . this simple model has definite popularity up to 70-th . but the experimental data ( ii ) and ( iii ) prohibit this model and it was forgotten . choosing the model we would like to hope that the considered model takes into account experimental conditions ( i ) - ( vi ) in the @xmath15 domain ; has natural asymptotics over multiplicity to the vhm region . + it is necessary to remember also that new channels of hadron production may arise in the vhm region . + it is impossible to understand all possibilities without those offered in sec.2.3.2 in the classification of asymptotics . thus , we will observe predictions of _ multiperipheral models _ , + distinguishing the soft pomeron models , see . _ the dual - resonance model _ + predictions for the vhm region are described also . it can be shown that this models predict asymptotics ( a ) if @xmath213 . just this result explains why vhm domain is defined by the condition ( ) . we are forced since it allows to include pqcd , forbidden by the multiperipheral models , considering _ hard pomeron model _ + production of mini - jets . but we will find using monte carlo simulations that the pqcd pomeron is unable to adopt the hard channels of hadron production . then _ the deep inelastic processes _ + for vhm region will be considered to generalize the dglap kinematics in the case of heavy qcd jets production . the analysis shows that transition to the vhm leads to the necessity to include low-@xmath214 sub - processes . as a result we get out of the range of pqcd validity . _ multiple reduction of jets . _ + we will see , that at very high energies in the vhm region the heavy jets creation should be a dominant process if the vacuum is stable with reference to the particle production . we will consider also decay of the ` false vacuum ' to describe the consequence of _ phase transition _ + in the vhm domain . this channel is hardly seen for @xmath15 since the confinement constraints may prevent cooling of the system up to phase transitions condition . later on multiple production physics was developed on the basis of experimental observation ( iv ) . the regge pole model naturally explains this experimental data and , at the same time , absorbs all experimental information , ( i)-(vi ) . at the very beginning , adopting the regge poles notion without its microscopical explanation , this description was self - consistent . the efforts to extend the regge pole model to the relativistic hadron reactions was ended by the reggeon diagram technique , see and references cited therein , and it was used later to construct the perturbation theory for @xmath70 . it was shown that the multiplicity distribution is wider then the poissonian because of the multi - pomeron exchanges . the leading energy asymptotics pomeron contribution reflects the created particle kinematics described in appendix d , where the available kinematical scenario in the frame of pqcd are described . so , the longitudinal momentum of produced particles is large and is strictly ordered . at the same time , particles transverse momentum is restricted . let us consider the inelasticity coefficient introduced in ( ) @xmath215 , where @xmath9 is the energy of the fastest particle in the laboratory frame . then the strict ordering of particles in the pomeron kinematics means that @xmath216 is independent of the index of the particle . so , if the fastest particle has the energy @xmath217 , then the following particle should have the energy @xmath218 , and so on . following to this law , @xmath219-th particle would have the energy @xmath220 . in the laboratory frame the energy should degrade to @xmath221 . inserting here the above formulated estimation of @xmath222 we can find that if the number of produced particles is n_0 , n_0=-(1-)^2>0 , = ( s / m^2 ) , 1.1athen we may expect the total degradation of energy . this degradation is the necessary condition noting that the total cross section of slowly moving particles may depend only slightly on energy and is seems necessary for natural explanation of the weak dependence of the hadron cross sections on the energy . this consideration would be lorentz - covariant if one can find the slowly moving particle in an arbitrary frame . resulting estimation of mean multiplicity have good qualitative experimental confirmation . notice that it was assumed deriving ( ) that the energy degrades step by step . in other words , if we introduce a time of degradation , then the time @xmath223 is needed for complete degradation of energy . assuming the random walk in the normal to incident particle plane , we can conclude that the points of particle production are located on a disk ( in the moving frame ) of radii @xmath224 . this means that the interaction radii should grow with the energy of the colliding particles . if @xmath225 is the cross section to observe particle @xmath226 inclusively in the @xmath227 and @xmath228 particles collision , then it was found experimentally that the ratio = = 1.2is universal . this may be interpreted as the direct evidence of fact that the hadron interactions have a large - distance character , i.e. that the interaction radii should be large . this picture assumes that the probability to have total degradation of energy is ~e^-b^2/4 , 1.3where @xmath229 is some dimensional constant ( the slope of regge trajectory ) and @xmath230 is the 2-dimensional impact parameter . this formulae has also the explanation connected to the vacuum instability with reference to the real particle production in the strong colour electric field . the above picture has natural restrictions . we can assume that each of the produced particles may be the source of above described @xmath42-channel cascade of the energy degradation . this means that in the frame of the pomeron phenomenology , we are able to describe the production of n < ^2 1.4particles only . if @xmath231 then the density of particles in the diffraction disk becomes large and ( a ) one should introduce short - distance interactions , or ( b ) rise interaction radii . it will be shown that just ( a ) is preferable . we will build the perturbation theory in the phenomenological frames ( i ) - ( vi ) . considering the system with variable number of particles the generating function t(z , s)=_n z^n _ n(s ) i1would be useful . one can use also the decomposition : t(z , s)=_tot(s)\{(z-1)c_1(s)+(z-1)^2c_2(s)+ ... } , i2where , by definition , c_1(s)==.t(z , s)|_z=1 i3is the mean multiplicity , c_2(s)=.t(z , s)|_z=1 i4is the second binomial momentum , and so on . our idea is to assume that all @xmath232 , @xmath154 may be calculated perturbatively choosing p(0,s)=e^(z-1 ) i5as the born approximation ` superpropagator ' . it is evident that ( ) leads to poisson distribution . then , having in mind ( ii ) , ( iii ) and ( v ) we will use following @xmath186 : p(q , s)=e^(o)-q^2se^(z-1 ) , i6where the transverse momentum @xmath233 is conjugate to the impact parameter @xmath230 . so , the born term ( ) is a fourier transform of the simple product of ( ) and ( ) . it contains only one free parameter , the pomeron intercept @xmath234 . on the phenomenological level it is not important to know the dynamical ( microscopical ) origin of ( ) . for our purpose the laplace transform of @xmath235 would be useful . if @xmath236 , then ( , q^2)=_0^de^-p(q^2,s)= . i8it is the propagator of two - dimensional field theory with mass squared @xmath237 knowing the gribov s reggeon calculus completed by the abramovski - gribov - kancheli ( agk ) cutting rules one can investigate the consequences of this approach . the lla approximation of the pqcd gives = ( 0)-1=_s0.55 , _ s=0.2 . i10but radiative corrections give @xmath238 . we will call this solution as the bfkl model . the quantitative origin of the restriction ( ) is following . the contribution of the diagram with @xmath239 pomeron exchange gives , since the diffraction radii increase with @xmath7 , see ( ) , mean value of the impact parameter decreasing with @xmath239 : @xmath240 where @xmath241 . on the other hand , the number of necessary pomeron exchanges @xmath242 since one pomeron gives maximal contribution ( with factorial accuracy ) at @xmath243 . in result , |^2~a. i11therefore , if the transverse momentum of created particles is a restricted quantity , i.e. @xmath244 , where @xmath245 is a constant , then the mechanism of particle production is valid up to n~^2 . o8 following our general idea , it will be enough for us find the position of singularity over @xmath129 . analysis shows that ( ) predict the singularity at infinity . in appendix e the gribov s reggeon diagram technique with cut pomerons is described and ( ) is derived . it can be shown using this technique that the model with the critical pomeron , @xmath246 , is inconsistent from the physical point of view . as was mentioned above , the model with @xmath247 is natural for the pqcd . the concrete value of @xmath248 will not be important for us . we will assume only that 0<0@xmath249>0@xmath250|^2~b^24(+(n/))@xmath251z@xmath252n~n_max~s@xmath253e^-n@xmath254~s^-1/2@xmath255(m)@xmath256m@xmath257m@xmath258^r(m)@xmath259^r=1/2@xmath260^r_n(m)@xmath261m@xmath262n@xmath263|n^r(m)@xmath264^r_n(m)@xmath265|n^r(m)@xmath266m / m_0@xmath267n@xmath268n|n^2(s)= |n_0^r ( /m_0)@xmath269|n(s)<<n<|n^2(s)@xmath270q|q@xmath271n@xmath272d_ab(x , q^2;n)@xmath273d_ab(x , q^2)@xmath274b@xmath275q^2<0@xmath276a@xmath277~@xmath278>>@xmath279_s()<<1@xmath280q^2@xmath279x@xmath281(1/x)>>1@xmath282d_ab(x , q^2)@xmath283(1/x)@xmath284|q^2|@xmath285~(1/x)/|q^2|@xmath286@xmath287@xmath288 ^ 2<<k_i^2<<-q^2@xmath289k_i^2>0@xmath290~(1/)@xmath291k_i^2>>^2@xmath292|n_j@xmath293|k|@xmath294|n_j @xmath295@xmath296||k_i|@xmath297|k_i|@xmath298k_i@xmath299x@xmath300x0@xmath301t_ab(x , q^2;z)@xmath302n@xmath303z_c@xmath304n@xmath305z_c@xmath306z_c(x , q^2;n)@xmath307n@xmath308d_ab(x , q^2;n)@xmath309t_ab(x , q^2;z)@xmath310z=1@xmath311d_ab(x , q^2;n)@xmath312c_ab(x , q^2;n)@xmath313n@xmath314c_ab(x , q^2;n)~n@xmath315c_ab(x , q^2;n)=const.@xmath316n@xmath317nn_p@xmath318n@xmath319c_ab(x , q^2;n)@xmath320|_ab(x , q^2;n)@xmath321|n_j()@xmath322@xmath323@xmath324n>>|n_g@xmath325(,z)@xmath326x0@xmath327_s(-q^2/^2)~1@xmath279_s(1/x)~1@xmath328|q^2|la^2@xmath329@xmath330_s(^2)<<1@xmath331(^3)_6@xmath332|q^2|la^2@xmath333z_s-1~(1/|n)@xmath334a^(1j)(n , m)@xmath335n@xmath336|n_j(m)@xmath337m@xmath338c_j@xmath339n/|n@xmath340m@xmath341m_1@xmath279m_2@xmath342m_1=m_2=m/2@xmath343m_2<<m_1m@xmath344a^(1j)@xmath345n_1+n_2=n@xmath346@xmath347@xmath348|n_j(m)@xmath349m@xmath350a^1j@xmath336a^2j@xmath351z@xmath352s@xmath353t(z , s)@xmath354z=1@xmath355z_c@xmath356s@xmath279n@xmath357t@xmath358s@xmath359s@xmath360n@xmath361n@xmath362_n(s)@xmath363_n(s)@xmath364n@xmath365w_n(x_1,x_2, ... ,x_n)@xmath366x_k=(u , q)_k@xmath367s@xmath368s@xmath369(|n(s)/n)<<1@xmath370|n(s)@xmath371n@xmath372s@xmath373_max@xmath374(_max / s)0@xmath375s@xmath376n@xmath377n_max=/m@xmath378@xmath336t(z , s)@xmath379z@xmath380z>1@xmath381t(z , s)@xmath382n_max@xmath383t(z , s)@xmath384n@xmath385z@xmath386t(z)@xmath387z@xmath388t(z)@xmath389z=1@xmath390z@xmath391n@xmath392z@xmath393n@xmath394n=@xmath395z=1@xmath396t(z)@xmath397z=1@xmath398z=@xmath399z@xmath400_n@xmath401_n\{-cn^}@xmath402t(z)@xmath403z=1@xmath4041 , 1.3awhere @xmath130 is smallest solution of the equation of state n = zt(z ) . 1.2a therefore , to have the singularity at @xmath106 , we should consider @xmath405 as a decreasing function of @xmath4 . on the other hand , at constant temperature , @xmath406 is the chemical potential , i.e. is the work necessary for creation of one particle . so , the singularity at @xmath106 means that the system is unstable : the less work is necessary for creation of one more particle if @xmath407 is a decreasing function of @xmath4 . the physical explanation of this phenomena is the following , see also . the generating function @xmath408 has following expansion : t(z)=\{_lz^lb_l } , 1.4awhere @xmath151 are known as the mayer s group coefficients . they can be expressed through the inclusive correlation functions and may be used to describe formation of droplets of correlated particles , see sec.2.3.3 . so , if droplet consist of @xmath409 particles , then b_l~e^-l^(d-1)/d 1.5is the mean number of such droplets . here @xmath410 is the surface energy of @xmath411-dimensional droplet . inserting this estimation into ( ) , t(z)~_l e^(l - l^(d-1)/d ) , = z. 1.6the first term in the exponent @xmath412 is the volume energy of the droplet and being positive it tries to enlarge the droplet . the second surface term @xmath413 tries to shrink it . therefore , the singularity at @xmath106 is the consequence of instability : at @xmath132 the volume energy abundance leads to unlimited growth of the droplet . in conclusion we wish to formulate once more the main assumptions . it was assumed first of all that the system under consideration is in equilibrium . this condition may be naturally reached in the statistics , where one can wait the arbitrary time till the system becomes in equilibrium . note , in the critical domain , the time of relaxation @xmath414 , @xmath415 , @xmath416 , @xmath417 is the critical temperature . we can not give the guarantee that in the high energy hadron collisions the final state system is in equilibrium . the reason of this uncertainty is the finite time the inelastic processes and presence of hidden ( confinement ) constraints on the dynamics . but if the confinement forces are frozen in the vhm domain , i.e. the production process is ` fast ' , then the equilibrium may be reached . we may formulate the quantitative conditions , when the equilibrium is satisfied . one should have the gaussian energy spectra of created particles . if this condition is hardly investigated in the experiment , then one should consider the relaxation of ` long - range ' correlations . this excludes the usage of relaxation condition for the ` short - range ' ( i.e. resonance ) correlation ( ii ) . the second condition consists in the requirement that the system should be in the critical domain , where the ( equilibrium ) fluctuations of the system become high . having no theory of hadron interaction at high energies we can not define where lies the ` critical domain ' and even whether it exists or not . but , having the vhm ` cold ' final state , we can hope that the critical domain is achieved . the quantitative realization of this picture is given in appendix j. it is important to note that used there the semiclassical approximation is rightfully in the vhm domain . it seems useful to start the discussion of models outlined the main problems , from authors point of view . * a. * _ soft colour parton problems _ the infrared region of soft colour parton interactions is a very important problem of high energy hadron dynamics . such fundamental questions as the infrared divergences of pqcd , collective phenomena in the coloured particles system and symmetry breaking are the phenomena of the infrared domain . the standard ( most popular ) hadron theory considers pqcd at small distances ( in the scale of @xmath418gev ) as the exact theory . this statement is confirmed by a number of experiments , namely deep - inelastic scattering data , hard jets observation . but the pqcd predictions have a finite range of validity since the non - perturbative effects should be taken into account at distances larger then @xmath419 . it is natural to assume , building the complete theory , that at large distances the non - perturbative effects @xmath420 the perturbative ones . as a result pqcd loses its predictability @xmath421 by the non - perturbative effects . notice , the pqcd running coupling constant @xmath422 becomes infinite at @xmath423 and we do not know what happens with pqcd if @xmath424 . there are few possibilities . for instance , there is a suspicion that at @xmath425 the properties of theory changed so drastically ( being defined on a new vacuum ) that even the notions of pqcd @xmath426 . this means that pqcd should be truncated from below on the ` fundamental ' scale @xmath427 . it seems natural that this infrared cut - off would influence the soft hadrons emission . the new possibility is described in appendix k. this strict formalism allows to conclude that pure pqcd contributions are realized on zero measure , i.e. it is the phenomenological theory only . the successive approach shows that the yang - mills theory should be described in terms of ( action , angle)-like variables . the last one means that the self - consistent description excludes such notions as the ` gluon ' . as a result of this substitution new perturbation theory would be free from infrared divergences , i.e. there is not necessity to introduce the infrared cutoff parameter @xmath427 . ( moreover , in the sector of vector fields ( without quarks ) the theory is ultraviolet stable . ) it seems important for this reason to investigate experimentally just vhm events , where the soft colour partons production is dominant . it is important to try to raise the role of pqcd in the ` forbidden ' area of large distances . the vhm processes are at highly unusual condition , where the non - perturbative effects must be negligible . * b*. _ dissipation problems _ the highly nonequilibrium states decay ( thermalization ) which means in the pqcd terms that the process of vhm formation should be enhanced , at least in asymptotics over multiplicity and energy , by jets . it is the general conclusion of nonequilibrium thermodynamics and it means that the very nonequilibrium initial state tends to equilibrium ( thermalized ) as fast as possible . the entropy @xmath428 of a system is proportional to the number of created particles and , therefore , @xmath428 should tend to its maximum in the vhm region . but the maximum of entropy testifies also the equilibrium of the system . * c*. _ collective phenomena _ we should underline that the collective phenomena may take place if and only if the particles interaction energy is comparable to the kinetic one . the vhm system considered may be ` cold ' and ` equilibrium ' . for this reason the vhm state is mostly adopted for investigation of collective phenomena . one of possible states in which the collective effects , see a.g . , may be important is the ` could coloured plasma ' . the fundamental interest presents the problem of vacuum structure of yang - mills theory . for instance , if the process of cooling is ` fast ' , since the dissipation process of vhm final state formation should be as fast as possible , then one may consider formation of vacuum domains with various properties . then decay of these domains may lead to large fluctuations , for instance , of the isotopic spin . another important question is the collective phenomena in the vhm final state . the last one may be created ` perturbatively ' , for instance , by the formation of heavy jets . then the colour charges should be confined . there are various predictions about this process . one of them predicts that there should be a first order phase transition . now we can ask : what can models say concerning the above problems ? * a*. _ soft coloured partons production _ the multiperipheral models predict fast decreasing of topological cross section in the vhm domain @xmath429 , @xmath430 . at the same time the mean transverse momentum should @xmath431 in this domain since the interaction radii should ` increase ' with @xmath4 . the bfkl pomeron predicts the same asymptotics , @xmath430 , but the pqcd jet predicts @xmath166 . the @xmath22 attempt to insert into the bfkl pomeron the production of particles via ( mini)jets seems impossible . this ` insertion ' can be done into the dis ladder but investigation of the lla kinematics in the vhm domain allows the conclusion that the ` low - x ' contributions should be taken into account . all this experience allows the assumption that in the vhm domain no ` @xmath42-channel ladder ' diagrams play sufficient role . this means the existence of a transition to the processes with jet dominance . the pqcd is unable to predict the transition mechanism . * b*. _ transition into equilibrium _ if the @xmath42-channel ladders are ` destroyed ' in the vhm region , then jets , despite the small factor @xmath432 in the cross sections , are the only mechanism of particle production in the vhm domain . dominance of heavy jets in the vhm domain may naturally explain the tendency towards equilibrium . but the description of thermalization in terms of jets of massless gluons production destroy this hope : the jet contribution @xmath166 assumes ` bremsstrahlung ' of @xmath433 gluons . this prevents the equilibrium since ordering without fail introduce the non - relaxing correlations . * c*. _ collective phenomena _ considering the collective phenomena , we proposer to distinguish ( a ) the collective phenomena connected with the vacuum and ( b ) the collective phenomena produced in the vhm system . following the experience of sec.3.3 we can conclude that the signal of vacuum instability is inequality : @xmath434 . the case ( b ) will not effect the cross section @xmath70 . but if the system became equilibrium in the vhm domain then the collective phenomena may be investigated using ordinary thermodynamical methods . for instance , noting that @xmath435 is the free energy one can measure the thermal capacity ( _ c , z)=c(_c , z ) . o23then , comparing capacities of hadron and @xmath436-quanta systems we can say whether or not the phase transition happened . the connection of the equilibrium and relaxation of correlations is well known . continuing this idea , if the vhm system is in equilibrium one may assume that the colour charges in the pre - confinement phase of vhm event form the plasma . one should note here that the expected plasma is ` cold ' and ` dense ' . for this reason no long - range confinement forces would act among colour charges . then , being ` cold ' , in such a system various , collective phenomena may be important . the experimental possibilities in the vhm domain are not clear up to now . nevertheless first steps toward formulation of trigger system was done , see . by this reason we would like to restrict ourselves by following two general questions . it seems that these questions are mostly important being in the very beginning of vhm theory . 0.3 cm * i*. _ for what values of multiplicity at a given energy the vhm processes become hard _ ? + the answer to this question depends on the value of the incident energy . if we know the answer then it will appear possible to estimate the role of multiperipheral contributions , the jet production rate , the role of vacuum instabilities . it seems that the experimental answer to this question is absent since produced particles are soft , theirs mean transverse and longitudinal momenta have the same value . in our understanding this question means : the total transverse energy may be extremely high . it is interesting also to search the heavy jets , i.e. to observe the fluctuations of particle density in the event - by - event experiment , but this program seems vague since , for all evidence , fractal dimensions tend to zero with increasing multiplicities . 0.3 cm * ii*. _ for what values of multiplicity does the vhm final state become equilibrium _ ? + we hope that having answer on this question we would be able to investigate the status of pqcd , to observe the phase transition phenomena directly , to estimate the role of confinement constraints . the equilibrium means that the energy correlation functions mean values are small . it is interesting also , for example , the charge equilibrium , when the mean value of charge correlation functions are small . notice , the effect of the phase space boundary may lead to ` equilibrium ' . indeed , if @xmath437 and @xmath438 then one may neglect the momentum dependence of the amplitudes @xmath439 . in this case the momentum dependence is defined by the boltzmann exponent @xmath440 , @xmath441 , only and we get naturally to the gaussian law for momentum distribution . correlators should be small in this case since there is no interactions among particles ( @xmath439 are constants ) . but our question assumes that we investigate the possibility of equilibrium when @xmath133 and @xmath442 . there are various approaches to build the real - time finite - temperature field theories of schwinger - keldysh type ( e.g. ) . all of them use various tricks for analytical continuation of imaginary - time matsubara formalism to real time . the basis of the approaches is the introduction of the matsubara field operator _m ( * x * , ) = e^h_s ( * x * ) e^-h , 32where @xmath443 is the interaction - picture operator introduced instead of the habitual heisenberg operator @xmath444 eq . ( ) introduces the averaging over the gibbs ensemble instead of averaging over zero - temperature vacuum states . if the interaction switched on at the moment @xmath117 adiabatically and switched off at @xmath118 then there is the unitary transformation : (x , t)=u(t_i , t_f ) u(t_i , t)_s ( x)u(t , t_i ) . a32introducing the complex mills time contours to connect @xmath117 to @xmath42 , @xmath42 to @xmath118 and @xmath118 to @xmath117 we form a ` closed - time ' contour @xmath445 ( the end - points of the contours @xmath446 together ) this allows to write the last equality ( ) in the compact form : @xmath447 where @xmath448 is the time - ordering on the contour @xmath445 operator . the generating functional @xmath449 of correlation ( green ) functions has the form : @xmath450 where @xmath451 means averaging over the initial state . if the initial correlations have a little effect , we can perform averaging over the gibbs ensemble . this is the main assumption of the formalism : the generating functional of the green functions @xmath449 has the form in this case : @xmath452 with @xmath453 . in accordance with ( ) we have : @xmath454 and , as a result , z(j)=d e^i_c _ d^4 x \ { l(x ) + j(x)(x ) } 33where the path integration is performed with kms periodic boundary condition : @xmath455 in ( ) the contour @xmath456 connects @xmath117 to @xmath118 , @xmath118 to @xmath117 and @xmath117 to @xmath457 . therefore it contains an imaginary - time matsubara part @xmath117 to @xmath457 . a more symmetrical formulation uses the following realization : @xmath117 to @xmath118 , @xmath118 to @xmath458 , @xmath458 to @xmath459 and @xmath460 to @xmath457 ( e.g. ) . this case also contains the imaginary - time parts of the time contour . therefore , eq . ( ) presents the analytical continuation of the matsubara generating functional to real times . one can note that if this analytical continuation is possible for @xmath449 then representation ( ) gives good recipe of regularization of frequency integrals in the matsubara perturbation theory , e.g. . but it gives nothing new for our problem since the matsubara formalism is a formalism for equilibrium states only . the starting point of our calculations is the @xmath4- into @xmath52-particles transition amplitude @xmath125 , the derivation of which is well known procedure in the lehmann - symanzik - zimmermann ( lsz ) reduction formalism framework , see also . the @xmath461-point green function @xmath462 are introduced for this purpose through the generating functional @xmath463 : g_nm(x , y)=(-i)^n+m_k=1^n(x_k)_k=1^m ( y_k)z_j , 22.5where ( x)= , * and z_j = de^is_j ( ) . 22.6 the action s_j ( ) = s_0()-v()+dxj(x)(x ) , 22.7 where @xmath464 is the free part and @xmath465 describes the interactions . at the end of the calculations one can put @xmath466 . to provide the convergence of the integral ( [ 22.6 ] ) over the scalar field @xmath467 the action @xmath468 must contain a positive imaginary part . usually for this purpose feynman s @xmath469-prescription is used . but it is better for us to use the integral on the mills complex time contour @xmath470 . for example , c _ : tt+i , + 0 , -t+o5and after all the calculations return the time contour on the real axis putting @xmath471 . in eq . ( [ 22.6 ] ) the integration is performed over all field configurations with standard vacuum boundary condition : d^4 x _ ( ^)= _ _ d_^=0 , 22.8 which leads to zero contribution from the surface term . let us introduce now field @xmath472 through the equation : - = j(x ) 22.9 and perform the shift @xmath473 in integral ( [ 22.6 ] ) , conserving boundary condition ( [ 22.8 ] ) . considering @xmath472 as the probe field created by the source : ( x)=dy g_0 ( x - y)j(y ) , + ( ^2 + m^2)_x g_0 ( x - y)=(x - y ) , 22.10 only the connected green function @xmath474 will be interesting for us . therefore , g_nm^c(x , y)=(-i)^n+m_k=1^n(x_k ) _ k=1^m(y_k)z ( ) , 22.11 where z()=de^is()-iv(+ ) 22.12 is the new generating functional . to calculate the nontrivial elements of the @xmath0-matrix we must put the external particles on the mass shell . formally this procedure means amputation of the external legs of @xmath474 and further multiplication on the free particle wave functions . as a result the amplitude of @xmath4- into @xmath52-particles transition @xmath125 in the momentum representation has the form : a_nm(q , p)=(-i)^n+m_k=1^n(q_k ) _ k=1^m^ * ( p_k ) z ( ) . 22.13 here we introduce the particle distraction operator ( q)=dxe^-iqx(x ) , ( x)=. 22.14 supposing that the momentum of particles are insufficient for us the probability of @xmath4- into @xmath52-particles transition is defined by the integral : r_nm = d_n ( q ) d_m ( p ) ^(4)(_k=1^nq_k - _ k=1^mp_k ) |a_nm|^2 , 22.15 where d_n(q)=_k=1^nd(q_k)= _ k=1^n , 22.16is the lorentz - invariant phase space element . we assume that the energy - momentum conservation @xmath68-function was extracted from the amplitude . note that @xmath475 is the divergent quantity . to avoid this problem with trivial divergence , connected integration over reference frame , let us divide the energy - momentum fixing @xmath68-function into two parts : ^(4)(q_k -p_k)=d^4 p ^(4)(p - q_k)^(4)(p - p_k ) 22.17 and consider a new quantity : r(p)=_n , md_n ( q ) d_m ( p ) ^(4)(p-_k=1^n q_k ) ^(4)(p-_k=1^n p_k ) |a_nm|^2 22.18defined on the energy momentum shell ( ) . here we suppose that the numbers of particles are not fixed . it is not too hard to see that , up to a phase space volume , r = d^4p r(p ) 22.19 is the imaginary part of the amplitude @xmath476 . therefore , computing @xmath477 the standard renormalization procedure can be applied and the new divergences will not arise in our formalism . the fourier transformation of @xmath68-functions in ( [ 22.18 ] ) allows one to write @xmath478 in the form : r(p)= e^ip(_1+_2 ) ( ) , 22.20where ( ) = _ n , m _ k=1^m\{d(q_k)e^-i_1 q_k } _ k=1^n\{d(p_k)e^-i_2 p_k } |a_nm|^2 . 22.21 introducing the fourier - transformed probability @xmath479 we assume that the phase - space volume is not fixed exactly , i.e. it is proposed that the 4-vector @xmath235 is fixed with some accuracy if @xmath480 are fixed . the energy and momentum in our approach are still locally conserved quantities since an amplitude @xmath125 is translationally invariant . so , we can perform the transformation : _ 1 q_k = ( _ 1 -_1 ) q_k + _ 1 q_k ( _ 1 -_1 ) q_k + _ 1 p since 4-momenta are conserved . the choice of @xmath481 fixes the reference frame . this degree of freedom of the theory was considered in . inserting ( [ 22.13 ] ) into ( [ 22.21 ] ) we find that & ( ) = \{-idx dx ( _ + ( x)d_+-(x - x,_2)_-(x)- + & -_-(x)d_-+(x - x,_1)_+(x ) ) } z(_+)z^ * (-_-) , 22.22where @xmath482 and @xmath483 are the positive and negative frequency correlation functions correspondingly : d_+-(x - x,)=-id_1(q)e^iq(x - x- ) 22.23 describes the process of particles creation at the time @xmath484 and its absorption at @xmath485 , @xmath486 , and @xmath57 is the cm 4-coordinate . function d_-+(x - x,)=id_1(q)e^-iq(x - x+ ) 22.24 describes the opposite process , @xmath487 . these functions obey the homogeneous equations : ( ^2 + m^2)_x g_+-=(^2 + m^2)_x g_-+=0 22.25since the propagation of on mass - shell particles is described . we suppose that @xmath103 may be computed perturbatively . for this purpose the following transformations will be used ( @xmath488 at @xmath489 ) : & e^-iv()= e^-idx ( x)(x ) e^idx j(x)(x ) e^-iv()= + & = e^dx ( x)(x ) e^-iv()= + & = e^-iv(-i ) e^idx j(x)(x ) , 22.26 where @xmath490 was defined in ( [ * ] ) and @xmath491 in ( [ 22.14 ] ) . at the end of the calculations , the auxiliary variables @xmath492 , @xmath493 can be taken equal to zero . using the first equality in ( [ 22.26 ] ) we find that z()= e^-idx ( x)(x ) e^-iv(+ ) e^-dx dx j(x)d_++(x - x)j(x ) , 22.27where @xmath494 is the causal green function : ( ^2 + m^2)_x g_++ ( x - y)=(x - y ) . 22.28 inserting ( [ 22.27 ] ) into ( [ 22.22 ] ) after simple manipulations with differential operators , see ( [ 22.26 ] ) , we may find the expression : & ( ) = e^-iv(-i_+)+iv(-i_-)\ { dx dx + & ( j_+ ( x)d_+-(x - x,_1)j_- ( x)- j_- ( x)d_-+(x - x,_2)j_+ ( x)- + & - j_+ ( x)d_++(x - x)j_+ ( x)+ j_- ( x)d_(x - x)j_- ( x ) ) } , 22.29where d_=(d_++)^ * 22.30 is the anticausal green function . considering the system with a large number of particles , we can simplify the calculations choosing the cm frame @xmath495 . it is useful also to rotate the contours of integration over @xmath496 for the result , omitting the unnecessary constant , we will consider @xmath497 . external particles play a double role in the @xmath0-matrix approach : their interactions create and annihilate the system under consideration and , on the other hand , they are probes through which the measurement of a system is performed . since @xmath498 are the conjugate to the particles energies quantities we will interpret them as the inverse temperatures in the initial ( @xmath499 ) and final ( @xmath500 ) states of interacting fields . they are the ` good ' parameters if and only if the energy correlations are relaxed . 0.2 cm * kubo - martin - schwinger boundary condition * 0.2 cm the simplest ( minimal ) choice of @xmath501 assumes that the system under consideration is surrounded by black - body radiation . this interpretation restores niemi - semenoff s formulation of the real - time finite temperature field theory . indeed , as follows from ( ) , the generating functional @xmath479 is defined by corresponding generating functional & _ 0 ( _ ) = z(_+)z^ * (-_-) = d_+ d_- e^is_0(_+)-is_0(_- ) + & e^-iv(_+ + _ + ) + iv(__- ) , 22.35 see ( [ 22.22 ] ) . the fields @xmath502 and @xmath503 were defined on the time contours @xmath470 and @xmath504 . as was mentioned above , see ( ) , the path integral ( [ 22.35 ] ) describes the closed path motion in the space of fields @xmath467 . we want to use this fact and introduce a more general boundary condition which also guarantees the cancelation of the surface terms in the perturbation framework . we will introduce the equality : _ _ d__+^_+ = _ _ d__- ^_- . 22.36 the solution of eq.([22.36 ] ) requires that the fields @xmath505 and @xmath506 ( and their first derivatives @xmath507 ) coincide on the boundary hypersurface @xmath65 : _(_)=(_)0 , 22.37 where , by definition , @xmath508 is the arbitrary turning - point " field . in the absence of the surface terms , the existence of a nontrivial field @xmath509 has the influence only on the structure of the green functions g_++=<t_+_+ > , g_+-=<_+_- > , + g_-+=<_-_+ > , g_=<_-_- > , 22.38 where @xmath510 is the antitemporal time ordering operator . these green functions must obey the equations : & ( ^2 + m^2)_x g_+- ( x - y)= ( ^2 + m^2)_x g_-+ ( x - y)=0 , + & ( ^2 + m^2)_x g_++ ( x - y)= ( ^2 + m^2)_x^ * g_ ( x - y)=(x - y ) , 22.39 and the general solution of these equations : g_ii = d_ii+g_ii , + g_ij = g_ij , ij 22.40 contain the arbitrary terms @xmath511 which are the solutions of homogenous equations : ( ^2 + m^2)_x g_ij ( x - y)=0 , i , j=+,- . 22.41 the general solution of these equations ( they are distinguished by the choice of the time contours @xmath110 ) g_ij(x - x)=d_1(q ) e^iq(x - x ) n_ij ( q ) 22.42 are defined through the functions @xmath512 which are the functionals of the ` turning - point ' field @xmath508 : if @xmath513 we must have @xmath514 . our aim is to define @xmath512 . we can suppose that @xmath515 the simplest supposition gives : n_ij~<_i_j>~<^2(_)>. 22.43 we will find the exact definition of @xmath512 starting from the @xmath0-matrix interpretation of the theory . it was noted previously that the turning - point field @xmath509 may be arbitrary . we will suppose that on the remote @xmath65 there are only free , on the mass - shell , particles . formally it follows from ( [ 22.40 ] - [ 22.42 ] ) . this assumption is natural also in the @xmath0-matrix framework . in other respects the choice of boundary condition is arbitrary . therefore , we wish to describe the evolution of the system in a background field of mass - shell particles . the restrictions connected with energy - momentum conservation laws will be taken into account and in other respects background particles are free . then our derivation is the same as in . here we restrict ourselves mentioning only the main quantitative points . calculating the product @xmath516 we describe a time ordered processes of particle creation and absorption described by @xmath482 and @xmath483 . in presence of the background particles , this time - ordered picture is slurred over because of the possibility to absorb particles before their creation appears . the processes of creation and absorption are described in vacuum by the product of operators @xmath517 and @xmath518 . we can derive ( see also ) the generalizations of ( [ 22.22 ] ) . the presence of the background particles will lead to the following generating functional : r_cp = e^-i*n*(_i^*_j)r_0 ( _ ) , 22.44 where @xmath519 is the generating functional for the vacuum case , see ( [ 22.35 ] ) . the operator @xmath520 describes the external particles environment . the operator @xmath521 can be considered as the creation and @xmath522 as the annihilation operator and the product @xmath523 acts as the activity operator . so , in the expansion of @xmath524 we can leave only the first nontrivial term : ( _ i^*_j)= d(q ) ^*_i ( q ) n_ij _ j ( q ) , 22.45since no special correlation among background particles should be expected . if the external ( nondynamical ) correlations are present then the higher powers of @xmath525 will appear in expansion ( [ 22.45 ] ) . following the interpretation of @xmath525 , we conclude that @xmath512 is the mean multiplicity of background particles . computing @xmath526 we must conserve the translation invariance of amplitudes in the background field . then , to take into account the energy - momentum conservation laws one should adjust to each vertex of in - going @xmath125 particles the factor @xmath527 and for each out - going particle we have correspondingly @xmath528 . so , the product @xmath529 can be interpreted as the probability factor of the one - particle ( creation+annihilation ) process . the @xmath4-particles @xmath530 process probability is the simple product of these factors if there is no special correlations among background particles . this interpretation is evident in the cm frame @xmath531 . after these preliminaries , it is not too hard to find that in the cm frame we have : & n_++(q_0)=n_(q_0)= = + & = = ( |q_0| ) . 22.47 computing @xmath512 for @xmath532 we must take into account the presence of one more particle : n_+-(q_0)= ( q_0 ) + ( -q_0 ) = + = ( q_0)(1+(q_0 _ 1))+ ( -q_0)(-q_0 _ 1 ) 22.48 and n_-+(q_0)= ( q_0)(q_0 _ 2)+ ( -q_0)(1 + ( -q_0 _ 2 ) ) . 22.49 using ( [ 22.47 ] ) , ( [ 22.48 ] ) and ( [ 22.49 ] ) , and the definition ( [ 22.40 ] ) we find the green functions : g_i , j(x - x,())= e^iq(x - x ) _ ij ( q , ( ) ) 22.50 where i_ij ( q , ( ) ) = ( ) + + + + 2(q^2 -m^2 ) ( ) 22.51 and a_()=-e^(|q_0|q_0 ) . 22.52 the corresponding generating functional has the standard form : r_p(j_)=e^-iv(-i_+)+iv(-i_- ) e^dx dx j_i ( x)g_ij(x - x,())j_j(x ) 22.53where the summation over repeated indexes is assumed . inserting ( [ 22.53 ] ) in the equation of state ( [ 3 ] ) we can find that @xmath533 . if @xmath534 is a ` good ' parameter then @xmath535 coincides with the green functions of the real - time finite - temperature field theory and the kms boundary condition : g_+-(t - t)=g_-+(t - t-i ) , g_-+(t - t)=g_+-(t - t+i ) , 22.54 is restored . the eq.([22.54 ] ) can be deduced from ( [ 22.51 ] ) by direct calculations . we start this consideration from the assumption that the temperature fluctuations are large scale . we can assume that the temperature is a ` good ' parameter in a cell the dimension of which is much smaller then the fluctuation scale of temperature . ( the ` good ' parameter means that the corresponding fluctuations are gaussian . ) let us divide the remote hypersurface @xmath536 on a @xmath537 and let us propose that we can measure the energy and momentum of groups of in- and out - going particles in each cell . the 4-dimension of cells can not be arbitrary small because of the quantum uncertainty principle . to describe this situation we decompose the @xmath68-function of the initial state constraint ( ) on the product of @xmath538 @xmath68-functions : @xmath539 where @xmath540 are the momentum of @xmath84-th in - going particle in the @xmath239-th cell and @xmath541 is the total 4-momenta of @xmath542 in - going particles in this cell , @xmath543 . therefore , @xmath544 the same decomposition will be used for the second @xmath68-function of outgoing particle constraints . we must take into account the multinomial character of particle decomposition on @xmath545 groups . this will give the coefficient : @xmath546 where @xmath547 is the kronecker s symbol . the summation over @xmath548 is assumed . in result , the quantity & r_n_c(p , q)= _ \{n , m}_n_c |a_nm|^2 + & ^n_c_=1\ { ^(4)(q_-^m__k=1q_k , ) ^(4)(p_-^n__k=1p_k , ) } 34defines the probability to find in the @xmath239-th cell the fluxes of in - going particles with total 4-momentum @xmath541 and of out - going particles with the total 4-momentum @xmath549 . the sequence of these two measurements is not fixed . the fourier transformation of @xmath68-functions in ( [ 34 ] ) gives : @xmath550 where @xmath551 has the form : _ n_c()=_=1^n_c\{^m__k=1 e^-i_1,q_k , ^n__k=1 e^-i_2,p_k , } |a_nm|^2 . 35inserting @xmath552 into ( [ 35 ] ) we find : & _ n_c()= \ { i_=1^n_c dx dx [ _ + ( x)d_+-(x - x;_2,)_-(x)- . + & .-_-(x)d_-+(x - x;_1,)_+(x ) ] } _ 0 ( ) , 36 where @xmath553 , and @xmath554 are the positive and negative frequency correlation functions . we must integrate over sets @xmath555 and @xmath556 if the distribution of momenta over cells is not fixed . in the result , r(p)=d^4_1(p ) d^4_2 ( p)_n_c ( ) , 37where the differential measure @xmath557 takes into account the energy - momentum conservation laws : @xmath558 the explicit integration gives that @xmath559 where @xmath560 is 3-vector of the cm frame . choosing cm frame , @xmath561 , @xmath562 in this frame @xmath563 where @xmath564 and @xmath565 was defined in ( [ 36 ] ) with @xmath566 . we will calculate integrals over @xmath498 using the stationary phase method . the equations for the most probable values of @xmath498 : -k(e,\{}_n_c)= _ n_c ( ) , k=1,2 , 38always have unique positive solutions @xmath567 . we propose that the fluctuations of @xmath568 near @xmath569 are small , i.e. are gaussian . this is the basis of the local - equilibrium hypothesis . in this case @xmath570 is the temperature in the initial state in the measurement cell @xmath239 and @xmath571 is the temperature of the final state in the @xmath239-th measurement cell . the last formulation ( [ 37 ] ) implies that the 4-momenta @xmath555 and @xmath556 can not be measured . it is possible to consider another formulation also . for instance , we can suppose that the initial set @xmath555 is fixed ( measured ) but @xmath556 is not . in this case we will have a mixed experiment : @xmath572 is defined by the equation : @xmath573 and @xmath574 is defined by the second equation in ( [ 38 ] ) . considering the continuum limit , @xmath575 , the dimension of the cells tends to zero . in this case we are forced by quantum uncertainty principle to assume that the 4-momenta sets @xmath576 and @xmath577 are not fixed . this formulation becomes pure thermodynamical : we must assume that just @xmath578 and @xmath579 are measurable quantities . for instance , we can fix @xmath578 and try to find @xmath579 as a function of the total energy @xmath10 and the functional of @xmath578 . in this case eqs.([38 ] ) become the functional equations . in the considered microcanonical description , the finiteness of temperature does not touch the quantization mechanism . indeed , one can see from ( [ 36 ] ) that all thermodynamical information is confined in the operator exponent @xmath580 the expansion of which describes the environment , and the ` mechanical ' perturbations are described by the functional @xmath581 . this factorization was achieved by the introduction of the auxiliary field @xmath472 and is independent from the choice of boundary conditions , i.e. from the choice of the systems environment . 0.2 cm * wigner functions * 0.2 cm we will adopt the wigner functions formalism in the carruthers - zachariazen formulation . for the sake of generality , the @xmath52 into @xmath4 particles transition will be considered . this will allow the inclusion of the heavy ion - ion collisions . in the previous section , the generating functional @xmath565 was calculated by means of dividing the ` measuring device ' on the remote hypersurface @xmath536 into @xmath537 cells _ n_c()=e^-i*n*(;,z)_0 ( ) , 39where & * n*(;,z)= \{_=1^n_c dx dx. + & .(_+(x)d_+-(x - x;_2,,z_2)_-(x).- + & .-_-(x)d_-+(x - x;_1,,z_1)_+(x ) ) } 40is the particle number operator . the frequency correlation functions @xmath482 and @xmath482 are defined by equalities : d_+-(x - x;_2,,z_2)=-id_1(q)e^iq_2,(x - x ) e^-_2,(q_2 , ) z_2(q_2 , ) a3.5 + d_-+(x - x,_1,,z_1)=id_1(q)e^-iq_1,(x - x ) e^-_1,(q_1,)z_1(q_1 , ) b3.5it was assumed that the dimension of the device cells tends to zero ( @xmath575 ) . now we wish to specify the cells coordinates . in the result we will get to the wigner function formalism . let us introduce wigner variables : x - x=r , x+x=2y : x = y+r/2 , x=y - r/2 . 3.6then & * n*(;,z)=-i_=1^n_c d(q)dr + & ( _ + ( y+r/2)_-(y - r/2)z_2(q_2,)e^iq_2,r e^-_2,(q_2,).+ + & + ._-(y+r/2)_+(y - r/2)z_1(q_1 , ) e^-iq_1,re^-_1,(q_1,))dy 3.10the boltzmann factor , @xmath582 , can be interpreted as the probability to find a particle with the energy @xmath583 in the final ( @xmath584 ) or initial ( @xmath585 ) state . the total probability , i.e. the process of creation and further absorption of @xmath4 particles , is defined by multiplication of these factors . besides @xmath586 is the out - going particle momentum measured in the @xmath239-th cell . generally it is impossible to adjust the 4-index of cell @xmath239 with coordinate @xmath587 . for this reason the summation over @xmath239 and the integration over @xmath588 are performed in ( ) independently . but let us assume that the 4-dimension of the cell @xmath589 is higher then the scale of the characteristic quantum fluctuations @xmath590 , l>>l_q . o6one can divide the 4-dimensional @xmath587 space into the @xmath589 dimensional cells . then , because of ( ) , the quantum fluctuations can not take away particles from this cell . then we can adjust the index of the measurements cell with the index of the @xmath587 space cell . as a result , & * n*(;,z)=-idyd_i(q)dr + & ( _ + ( y+r/2)_-(y - r/2)z_2(q_2,y)e^iq_2r e^-_2(y)(q_2)+ . + & .+_-(y+r/2)_+(y - r/2)z_1(q_1,y ) ) e^-iq_1re^-_1(y)(q_1 ) , 3.10where dy=__c()dy , o7and @xmath591 is the dimension @xmath589 of the @xmath587 space cell with index @xmath239 . notice that the momentum @xmath592 did not carry the index @xmath239 ( or the index @xmath587 of the space cell ) . our formalism allows the introduction of more general ` closed - path ' boundary conditions . the presence of external black - body radiation will only reorganize the differential operator @xmath593 and a new generating functional @xmath526 has the same form : @xmath594 the calculation of operator @xmath595 is strictly the same as in appendix b. introducing the cells we will find that @xmath596 where the occupation number @xmath597 carries the cell index @xmath587 : @xmath598 and ( @xmath599 ) @xmath600 @xmath601 @xmath602 for simplicity the cm system was used . other calculations are the same as the constant temperature case . first of all , two light - like 4-momenta @xmath603 are introduced . here @xmath604 are momenta of colliding particles . the final state particles momenta have the following representation : & p_1=_1p_2+_1p_1+p_1 , p_2=_2p_2+_2p_1+p_2 , + & k_i=_ip_2+_ip_1+k_i . m1sudakov s parameters , @xmath605 , are not independent . the mass shell conditions and the energy - momentum conservation laws give : & s_1_1=m^2+(p_1)^2=e_1 ^ 2 , s_2_2=e_1 ^ 2 , s_i_i = e_i^2 , + & _1+_2+_i=1 , _1+_2+_i=1 , m2where @xmath606 is the transverse energy . we have for the multiperipheral kinematics : 1_1>>_1>> ... >>_n>>_2~ , + < < _1<<_1<< ... <<_n<<_2~1 m3and the transverse momenta are restricted : considered cm frame : _ i= , |_i|>>|_i| , m5if the particle moves along @xmath607 , and a similar expression exists for particles moving in the opposite direction , where @xmath608 . in the ` central region ' of the cm frame @xmath609 the angles of produced particles are large and energies are small . it should be underlined that all this excludes the ( mini)jets formation . the final - state particles phase space volume element is & d_22+n = c^n_v + & ( _ 2-_1 ) ^n+1_i=1()^2(q_i^2)= dz_n , where @xmath610 and the 4-momentum of produced particle @xmath611 the square of pairs invariant mass : @xmath612 the energy conservation law takes the form : @xmath613 the trajectory of reggeized gluon is @xmath614 where @xmath615 is the gluon ` mass ' . if this virtuality is large , @xmath616 , then the gluon decay creating a pqcd jet . but the constraint on the multiperipheral kinematics prevent this possibility . 0.2 cm * deep inelastic reactions * 0.2 cm for the pure deep inelastic case , when one of the initial hadrons is scattered at the angle @xmath617 have the energy @xmath618 in the cms of beams whereas the another is scattered at small angle and the large transfer momentum @xmath619 is distributed to the some number of the emitted particles due to evolution mechanism we have ( @xmath617 is small ) : & d_n^dis= dd_n ded , + & dd_n=()^n_m^2^q^2 _ m^2^k_n^2_m^2^k_2_x^1 d_n^(1)_n__n^1 d_n-1^(1)_n-1 ... + & _ _ 2 ^ 1d _ 1^(1)_1 p() ... p(_1 ) , p(z)=2 , where @xmath620 and the emission angle @xmath621 . 0.2 cm * large angle production * 0.2 cm for the large - angle particle production process the differential cross section ( as well as the total one ) fall down with cm energy @xmath208 . let us consider for definiteness the process of annihilation of electron - positron pairs to @xmath4 photons : & d_n^dl = df_n , + & df_n=()^n ^n_i=1d x_idy_i(x_i - y_i)(y_i - y_i-1 ) ( y_i)(x_i ) + & ( -x_n)(-y_n),y_i=,= , the similar formulae takes place for subprocess of quark - antiquark annihilation into @xmath4 large - angle moving gluons . at the end , one can consider the following possibilities : + a ) pomeron regime ( p ) ; + b ) evolution regime ( dis ) ; + c ) double logarithmic regime ( dl ) ; + d ) dis+p regime ; + e ) p+dl+p regime . + the description of every regime may be performed in terms of effective ladder - type feynman diagrams . this can be done using the blocks @xmath622 , @xmath623 and @xmath624 . we will consider , see ( ) ( q , ;z)=_0^de^-p(q,;z)= , = ( s / m^2 ) . s9as the ` propagator of the cut pomeron ' . it will be assumed also that @xmath625 so , the resonance short range correlations will be ignored in this definition or propagator . it was assumed also that the ` bare ' slope @xmath229 is @xmath129 independent . it should be underlined that the ` propagator ' ( ) is written phenomenologically . it absorbs the assumptions that ( i ) the diffraction cone shrinks with energy and ( ii ) the inclusive cross sections are universal , see ( ) the set of principal rules concerning multiperipheral kinematics of feynman diagrams is given in appendix d. the reggeon calculus supposes that the virtuality of each line of the feynman diagram is restricted . this ignores ` hard jets ' , later known as the pqcd jets . then the @xmath239 pomeron exchange eikonal diagram has only @xmath626 ways of being cut . if the cut line goes through @xmath627 pomerons , then the corresponding contributions are : ^_( , q)= d_(m^_(q_1, ... ,q_))^2 y^__l=1^(q_i , _i;z)_i=+1^(q_i , _i;z=1 ) , s1where @xmath628 is the ` vertex function ' , the combinatorial coefficient is @xmath629 and the phase space element is d_=_i=1^ ( -_l=1^_l)^2(q-_i=1^q_l ) . s2 as usual , the contribution ( ) leads to the following mean multiplicity of produced particles : ^_=.d_(s / m^2)^^_( , q=0)|_z=1 ~. s3therefore , ~s4is essential in the vhm region , where @xmath630 is assumed . the impact parameter representation : ^_(s , q)=d()^ e^i^_( , q ) s5would be useful also . the contribution ( ) describes interactions with impact parameter 4(s / m^2)/. s6notice @xmath631 is the number of cut pomerons independent of @xmath627 . but , remembering that @xmath632 and that the regge model is only able to describe large distance interactions , @xmath633 , one can conclude that the regge pole description is valid only for n^2 . s7thus is why the vhm region is defined by @xmath634 . then the cut pomeron propagator in the impact parameter representation ( , ; z)=g(,)e^(z-1 ) , l3where g(,)=e^e^-^2/4 l4is the uncut pomeron profile function . using this definition one can find that the contribution of the eikonal diagrams gives a contribution : _ 0 ( , ; z)= ( 1-e^-^2g(,))- ( 1-e^2 ^ 2g(,)(e^(z-1)-1 ) ) , l5where @xmath615 is a constant . first bracket is essential for @xmath636 . so , with exponential accuracy , the first term is equal to @xmath637 let us now consider the second bracket . for @xmath638 ^24^2\{1+(1-e^(z-1))}= 4^2(,z ) l6are essential . it is not hard to see that @xmath639 decreases if @xmath640 and @xmath639 is equal to zero for @xmath106 . in this case , by the definition of the generating function , the integral over @xmath230 of @xmath641 defines the contribution to the total cross section . so , the model predicts the production of particles in the ring : 4^2(,z)^24^2 . l7if @xmath638 , i.e. if @xmath642 . notice also that @xmath643 . then , _ 0 ( , ; z=0)=_0 ( , ; z=1 ) . l8so , the elastic part of the total cross section is half of the total cross section . this means , using the optical analogy , that the scattering on the absolutely black disk is well described . the last conclusion means that the interaction radii should increase with @xmath4 in the vhm region . indeed , as follows from ( ) , _ 0 ( , ; z)(e^2 ^ 2g(,)(e^(z-1)-1)-1 ) , l9at @xmath132 and , therefore , 0 ^ 2b^2=4(+(z-1 ) ) l10are essential . our purpose is to investigate the role of the exponential spectrum ( ) in the asymptotic region over multiplicity @xmath4 . in this case one can validate heavy resonance creation and such formulation of the problem have definite advantages . \(i ) if creation of heavy resonances at @xmath644 is expected , then one can neglect the dependence on the resonance momentum @xmath645 . so , the ` low - temperature ' expansion is valid in the vhm region . \(ii ) having the big parameter @xmath4 , one can construct the perturbations expanding over @xmath142 . \(iii ) we will be able to show at the end the range of applicability of these assumptions . + for this purpose , the following formal phenomena will be used . the grand partition function t(z , s)=_n z^n _ n(s ) , t(1,s)=_tot(s ) , n / m_0n_max(s ) , gpfwill be introduced , see ( ) . then the inverse mellin transformation , see ( ) _ n(s)=t(z , s ) . metrwill be performed expanding it in the vicinity of the solution @xmath646 of the equation of state , see ( ): n = zt(z , s ) . eqst1it is assumed , and this should be confirmed at the end , that the fluctuations in the vicinity of @xmath130 are gaussian . it is natural at first glance to consider @xmath647 as an increasing function of @xmath4 . indeed , this immediately follows from the positivity of @xmath19 and the finiteness of @xmath648 at finite @xmath7 . but one can consider the ` thermodynamical limit ' , see sec.2.3.1 , or the limit @xmath649 . theoretically , the last one is right because of the pcac hypotheses and nothing should happen if the pion mass @xmath649 . in this sense @xmath137 may be considered as the whole function of @xmath129 . then , @xmath647 would be an increasing function of @xmath4 if and only if @xmath137 is a regular function at @xmath106 . the proof of this statement is as follows . we should conclude , as follows from eq . ( ) , that z_c(n , s)z_s at n , and at s = const , asyi.e . the singularity point @xmath138 @xmath650 @xmath130 in asymptotics over @xmath4 . if @xmath651 , then @xmath652 , when @xmath4 tends to infinity . the concrete realization of this possibility is shown in sec.3.3 . but if @xmath653 , then @xmath654 in vhm region , see secs.3.1 , 3.5 . on may use the estimation , see also ( ): -= z_c(n , s)+o(1/n ) , est0where @xmath130 is the @xmath655 solution of ( ) . it should be underlined that this estimation is @xmath656 of the character of singularity , i.e. the position @xmath138 is only important with @xmath657 accuracy . 0.2 cm * partition function * 0.2 cm introducing the ` grand partition function ' ( ) the ` two - level ' description means that & = _ k^k_i=1\{d_k(q)dm_i(q_i , z)e^-_i } + & n_k(q_1,q_2, ... ,q_k;)-(z , s ) , vdwhere @xmath658 . this is our group decomposition . the quantity @xmath659 may be considered as the local activity . so , .|_=1~_totb_1(q ) r1cfif the resonance decay forms a group of particles with total 4-momentum @xmath592 , then @xmath660 is the mean number of such groups . the second derivative gives : .|_=1~_tot \{b_2(q_1,q_2)-b_1(q_1)b_1(q_2)}_totk_2(q_1,q_2 ) r2cfwhere @xmath661 is the two groups correlation function , and so on . one can consider @xmath662 as mayer s group coefficients , see sec.2.3.3 . the lagrange multiplier @xmath62 was introduced in ( ) to each resonance : the boltzmann exponent @xmath663 takes into account the energy conservation law @xmath664 where @xmath10 is the total energy of colliding particles , @xmath665 in the cm frame . this conservation law means that @xmath62 is defined by equation : = t(z , ) . eqst2so , to define the state one should solve two equations of state ( ) and ( ) . the solution @xmath81 of the eq . ( ) has the meaning of inverse temperature of the gas of resonances if and only if the fluctuations in vicinity of @xmath81 are gaussian , see sec.2.2.2 . on the second level , we should describe the resonance decay into hadrons . using ( ) we can write in the vicinity of @xmath106 : ( q , z)=_n z^n ^r_n(q)=g^r()e^(z-1)|n(m ) . rgpfthe assumptions * b * and * d * , see ( ) , were used here so , -(z , s)=_k^k_i=1\{dm_i(m_i , z)}_k(m ; ) , vd1where @xmath666 @xmath667 was defined in ( ) and _ k(m;)=_i=1^k\ { d_k(q)e^-_i(q_i ) } b_k(m;q ) . 1assuming now that @xmath668 are essential , _ k(m;)b_k(m ) _ i=1^k\{e^-m_i } est following the duality assumption , one may write : b_k(m)=|b_k(m ) _ i=1^k \{m_i^e^_0 m_i } stringand @xmath669 is a slowly varying function of @xmath666 : @xmath670 in the result , the low - temperature expansion looks as follows : -(z , s)=_k \{_m_0^dm m^+3/2 e^(z-1)|n^r(m)-(-_0)m}^k . fe1we should assume that @xmath671 . in this sense one may consider @xmath672 as the limiting temperature and the above mentioned constraint means that the resonance energies should be high enough . 0.2 cm * thermodynamical parameters * 0.2 cm remembering that the position of the singularity over @xmath129 is essential , let us assume that the resonance interactions can not renormalize it , i.e. that the sum ( ) is convergent . then , leaving the first term in the sum ( ) , -(z , s)= _ m_0^dm ( m / m_0)^+3/2 e^(z-1)|n^r(m)-(-_0)m . fe2we expect that this assumption is satisfied if & _ m_0^dm m^+3/2 e^(z-1)|n^r(m)-(-_0)m > > + & \{_m_0^dm m^+3/2 e^(z-1)|n^r(m)-(-_0)m}^2 res1for n , s , < < 1 restr so , we would solve our equations of state with the following free energy : -(z , s)= _ m_0^d ( ) ( ) ^-1 e^-(m / m_0 ) , fe3where , using ( ) , =+2(z-1)|n^r_0 + 5/2=2(z-1)|n^r_0 , = m_0(-_0)0 , = const . nw we have in terms of these new variables the following equation for @xmath129 , n = z . eq1cthe equation for @xmath62 takes the form : n_max= , eq2cwhere @xmath673 and @xmath674 is the incomplete @xmath675-function : @xmath676 0.2 cm * asymptotic solutions * 0.2 cm following physical intuition , one should expect the cooling of the system when @xmath644 , for fixed @xmath208 , and heating when @xmath677 , for fixed @xmath4 . but , as was mentioned above , since the solution of eq . ( ) @xmath81 is defined by the value of the total energy , one should expect that @xmath81 decreases in both cases . so , the solution _ c0 , < 0 at n , < 0 at ssol2is natural for our consideration . the physical meaning of @xmath129 is activity . it defines at @xmath678 the work needed for one particle creation . then , if the system is stable and @xmath137 may be singular at @xmath132 only , > 0 at n , < 0 at s. sol3 one should assume solving equations ( ) and ( ) that z_c^_c+1 < < ( _c+1,_c ) . ineq1this condition contains the physical requirement that @xmath133 . in the opposite case , the finiteness of the phase space for @xmath679 should be taken into account . as was mentioned above , the singularity @xmath138 attracts @xmath130 at @xmath644 . for this reason one may consider the following solutions . a. @xmath680 : @xmath681 , @xmath682 . in this case ^-(,)~e^(/ ) . a1this estimation gives the following equations : n = c_1(/)e^(/ ) , = c_2 ()<<1 , a2where @xmath683 are the unimportant constants . the inequality is a consequence of ( ) . these equations have the following solutions : _ c<<1 , _c~n>>1 . a3using them one can see from ( ) that it gives _ n < o(e^-n ) . a4 b. @xmath684 : @xmath685 , @xmath686 . one should estimate @xmath687 near the singularity at @xmath106 and in the vicinity of @xmath246 to consider the consequence of this solution . expanding @xmath687 over @xmath248 at @xmath688 , ( ,)=()-^e^-+o(^+1 ) + o(1 ) . b1this gives the following equations for @xmath689 : n = c_1e^(1/ ) . b2the equation for @xmath248 has the form : n_max = c_2e^(+1)(1/ ) . b3where @xmath690 are unimportant constants . at 0<(1/)-1<<1 , i.e. at ( 1/)<<n<<^2(1/ ) , b4we find : _c~. b5inserting this solution into ( ): _ c~. b6it is remarkable that @xmath691 in the leading approximation is @xmath4 independent . by this reason @xmath692 becomes @xmath4 independent also : _c~ : z_c=1 + . b7this means that _ n = o(e^-n ) b8and obeys the kno scaling with mean multiplicity @xmath693 . considering particle creation in the dis processes , one should distinguish correlation of particles in the ( mini)jets and the correlations between ( mini)jets . we will start from the description of the jet correlations . one should introduce the inclusive cross section for the @xmath239 jets creation @xmath694 where @xmath695 , @xmath696 are the jets 4-momentum in the dis kinetics , @xmath697 . having @xmath698 we can find the correlation functions @xmath699 where @xmath700 and @xmath701 defines the sort of created colour particle . it is useful to introduce the generating functional f^ab(q^2,x;w)=_nd_n(k)_i=1^n w^r_i(k_i ) the amplitude , @xmath702 is the phase space volume and @xmath703 are the arbitrary functions . it is evident that f^ab(q^2,x;w).|_w=1=d^ab(q^2,x ) . 11athe inclusive cross sections _^(r)(k_1,k_2, .... ,k_;q^2,x)=_i=1^f^ab(q^2,x;w).|_w=1 . 12the correlation function n_^(r)(k;q^2,x ) = _ 1=1^f^ab(q^2,x;w).|_w=1 . 13awe can find the partial structure functions @xmath704 , where @xmath4 is the number of produced ( time - like ) gluons , using there definitions . it will be useful to introduce the laplace transform over the variable @xmath705 : f^ab(q^2,x;w)=_j<0()^j f^ab(q^2,j;w ) 14a the expansion parameter of our problem @xmath706 . for this reason one should take into account all possible cuts of the ladder diagrams . so , calculating @xmath707 in the lla all possible cuts of the skeleton ladder diagrams are defined by the factor : \{^ab_rg_r_r^ab } , 15ai.e . the cut line may not only get through the exact green function @xmath708 but through the exact vertex functions @xmath709 also ( @xmath710 are negative ) . we have in the lla ( see appendix d ) @xmath711 and @xmath712 following our approximation , see previous section , we would not distinguish the way in which the cut line goes through the born amplitude @xmath713 we will simply associate @xmath714 to each rung of the ladder . considering the asymptotics over @xmath4 , the time - like partons virtuality @xmath715 should be maximal . here @xmath716 is the fraction of the longitudinal momentum of the jet . then , slightly limiting the jets phase space , k_i^2=|q_i+1|^2(1+o((1/x)/|q_i+1|^2 ) ) . 16 as a result , introducing @xmath717 , where @xmath718 , @xmath719 in the lla variable , we can find the following set of equations : f_ab(q^2,j;w)=_c , r^r_ac(j)w^r ( ) f_ab(q^2,x;w ) , 17awhere ^r_ac(j)=_ac(j)=^1_0x^jp_ac(x ) 18aand @xmath720 is the regular kernel of the bethe - salpeter equation . at @xmath721 this equation is the ordinary one for @xmath707 . we will search the correlation functions from eq . ( ) in terms of the laplace transform @xmath722 let us write : f_ab(q^2,j;w)=d_ab(q^2,j)\ { _ _ i=1^ ( ( w^r_i(_i)-1))n_ab^(r)_(k;q^2,j ) } 19ainserting ( ) into ( ) and expanding over @xmath723 we find the sequence of coupled equations . omitting the cumbersome calculations , we write in the lla that & _ ab^(r)_(_1,_2, .... ,_;q^2,j)= + & = d_ac_1(j,_1)^r_1_c_1c_2(j)d_c_2c_3(j,_2)^r__c_c_+1(j)d_c_+1b(j,_+1 ) . 20one should take into account the conservation laws : _ 1_2_+1= , _ 1<_2< <_+1<. 21computing the laplace transform of this expression we find @xmath724 the kernel @xmath725 was introduced in ( ) . let us write it in the form : d_ab(j,)=_=^_(j ) , 22where d^_qq=_s-_gg , d^_qg=_s-_qq , d^_qg=_gq , d^_gq=_qg 23and _ = \{_qq+_gg+^1/2}. 24if @xmath726 , then @xmath727 are essential . in this case , _ gg~_gq>>_qg~_qq = o(1 ) . 25this means the gluon jets dominance and n^g_gg=_gg+o(1 ) . 26one can find the following estimation of the two - jet correlation function : n_ab^r_1r_2(_1,_2;,)= o(max\{(_1/)^_gg,(_2/)^_gg , ( _ 1/_2)^_gg}}. 27this correlation function is small since in the lla @xmath728 . this means that the jet correlation becomes high if and only if the mass of the correlated jets are comparable . but this condition shrinks the range of integration over @xmath729 and for this reason one may neglect the ` short - range ' correlations among jets . therefore , as follows from ( ) , f_ab(q^2,j;w)=d_gg(,j)\ { _ gg^__0 w^g( ) } 28we will use this expression to find the multiplicity distribution in the dis domain . 0.2 cm * generating function * 0.2 cm to describe particle production , one should replace : @xmath730 where @xmath731 is the @xmath173 of @xmath4 particle production , _ nw^r_n=1 . 29having @xmath239 jets , one should take into account the conservation condition @xmath732 . for this reason , the generating functions formalism is useful . in the result , one can find that if we take ( ) w^g = w^g(,z ) , w^g(,z).|_z=1=1 , 30then @xmath733 defined by ( ) is the generating functional of the multiplicity distribution in the ` @xmath492 representation ' . in this expression @xmath734 is the generating function of the multiplicity distribution in the jet of mass @xmath735 . in the result , see ( ) , f^ab(q^2,x;w)_j<0(1/x)^j e^_gg(,z ) 31where (,z)=^__0w^g(,z ) . 32noting the normalization condition ( ) , (,z=1)=. 33 the integral ( ) may be calculated by the steepest descent method . it is not hard to see that jj_c=1+\{4n(,z)/(1/x)}^1/2 34is essential . notice that @xmath736 should be essential but we find , instead of the constraint ( ) , that (,z)<<(1/x ) . 35 in the frame of this constraint , f^ab(q^2,x;w)\{4}. 36 generally speaking , there exist such values of @xmath129 that @xmath737 . this is possible if @xmath738 is a regular function of @xmath129 at @xmath106 . then @xmath130 should be an increasing function of @xmath4 and consequently @xmath739 would be an increasing function of @xmath4 . therefore , one may expect that in the vhm domain @xmath737 . then @xmath740 would be essential in the integral ( ) . this leads to the following estimation : @xmath741 but this is impossible since @xmath742 should be an increasing function of @xmath129 . this shows that the estimation ( ) has a finite range of validity . solution of this problem with unitarity is evident . one should take into account correlations among jets considering the expansion ( ) . indeed , smallness of @xmath743 may be compensated by large values of @xmath744 in the vhm domain . one may neglect quark jets in the vhm region since the gluons mean multiplicity @xmath745 - quarks multiplicity and in the vhm region the leftmost singularities are important . then we can write : t_j(,z)=t_j(,z ) ^__0d(t_j(,z)-1 ) , j1where @xmath746 and @xmath747 is the generating function of the distribution over the number of gluons @xmath748 : t_j(,z)=_n z^n w_n ( ) , t_j(,z=1)=1 . j2we search a solution in the vhm region , where n>>n_j\ { } , a=12/11 . j3let us consider the following solution : w_n=()^e^-n/|n_j . j4it is useful to introduce _ k()=_n=1^n^k-1w_n ( ) j5for this solution . inserting ( ) in this expression , _ k()=|n^k()_k , j6where @xmath498 ( i ) should be positive and ( ii ) @xmath729 independent . these conditions are satisfied for the following values of @xmath84 . indeed , at @xmath749 , _ k=_n=1^ ( ) ^k+e^-n/|n_j^-(k+ ) . j7 the generating function @xmath750 has the following form in terms of @xmath751 : t_j(,z)=_n=1^(z)^k_k ( ) . j8inserting ( ) into ( ) and assuming that @xmath498 is a @xmath729 independent quantity , we find the following recurrent equation for @xmath498 : _ k=^k_k_1=1_k_1_k - k_1- 2_k . j9therefore , if k > > j10then we can neglect last term in the right hand side of ( ) and in this case @xmath498 are positive and @xmath729 independent . noting that in ( ) @xmath752 are essential the inequality ( ) means that the solution ( ) is correct if n>>|n_j|n_j , j11i.e . , only for this value of @xmath4 @xmath753 has the form ( ) and the corresponding generating functional has singularity at z_s=1 + . j12 it is important for the vhm experiment to have an upper restriction on the asymptotics . we wish to show that @xmath70 decreases faster than any power of @xmath142 : _ n < o(1/n ) . 2.1to prove this estimation , one should know the type of singularity at @xmath106 . one can imagine that the points , where the external particles are created , form the system . here we assume that this system is in equilibrium , i.e. there is not in this system macroscopical flows of energy , particles , charges and so on . the lattice gas approximation is used to describe such a system . this description is quite general and does not depend on details . motion of the gas particles leads to the necessity to sum over all distributions of the particles on cells . for simplicity we will assume that only one particle can occupy the cell . so , we will introduce the occupation number @xmath754 in the @xmath174-th cell : @xmath755 means that we have no particle in the cell and @xmath756 means that a particle exists in a cell . assuming that the system is in equilibrium , we may use the ergodic hypothesis and sum over all ` spin ' configurations of @xmath757 , with the restriction : @xmath758 . it is evident that this restriction introduces the interactions . the corresponding partition function in temperature representation ( , h)=de^-s ( ) 2.2awhere integration is performed over @xmath759 and , considering the continuum limit , @xmath760 . the action s_()=dx\{()^2 -^2+g^4- } 2.3where ~(1- ) , g~ , ~()^1/2h . 2.4and @xmath761 is the critical temperature . 0.2 cm * unstable vacuum * 0.2 cm we start this consideration from the case @xmath762 , i.e. assuming that @xmath763 . in this case the ground state is degenerate if @xmath764 . the extra term @xmath765 in ( ) can be interpreted as the interaction with external magnetic field @xmath766 . this term regulates the number of ` down ' spins with @xmath767 and is related to the activity : z^1/2=e^h , 2.5i.e . @xmath766 coincides with the chemical potential . the potential v()=-^2+g^4 , >0 , 2.6has two minima at @xmath768 if the dimension @xmath769 , no tunnelling phenomena exist . but choosing @xmath770 the system in the correct minimum ( it corresponds to the state without particles ) becomes unstable . the system tunneling into the state with an absolute minimum of energy . the partition function @xmath185 becomes singular at @xmath764 because of this instability . the square root branch point gives ( b , z)=e^-a_2()/h^2 , a_i>0 . 2.7note , @xmath771 at @xmath764 . deforming the contour in the mellin integral over @xmath129 on the branch line , _ n()=^_1 e^-4a_2 ^ 2/^2z . 2.8 in this integral z_c\{}^1/3 2.9is essential . this leads to the following estimation : _ ne^-3(a_2 ^ 2)^1/3n^2/3<o(1/n ) . 2.10 it is useful to note at the end of this section that + ( i ) the value of @xmath69 is defined by @xmath772 and the metastable states , the decay of which gives a contribution into @xmath773 , are not important . + ( ii ) it follows from ( ) that in the vhm domain h~h_c~z_c~(1/n)^1/30 . 2.11so , the calculations are performed for the ` weak ' external field case , when the degeneracy is weakly broken . it is evident that the life time of the unstable ( without particles ) state is large in this case and for this reason the semiclassical approximation is correct . this is an important consequence of ( ) . 0.2 cm * stable vacuum * 0.2 cm let us consider now @xmath774 , i.e. @xmath775 . the potential ( ) has only one minimum at @xmath776 in this case . the inclusion of an external field shifts the minimum to the point @xmath777 . in this case the expansion in the vicinity of @xmath778 should be useful . as a result , ( , z)=\{dx _ c - w(_c ) } , 2.12where @xmath779 can be expanded over @xmath778 : w(_c)=_l_k\{dx_k_c(x_k;h)}_l ( x_1, ... ,x_l ) . 2.13 in this expression , @xmath780 is the one - particle irreducible green function , i.e. @xmath781 is the virial coefficient . then @xmath778 can be considered as the effective activity of the correlated @xmath409-particle group . the sum in ( ) should be convergent and , therefore , @xmath782 if @xmath783 . but in this case the virial decomposition is equivalent to the expansion over the inverse density of particles . in the vhm region it is high and the mean field approximation becomes correct . in result is _ c-()^1/3 : |s_c| if to calculate @xmath69 . in this case z_ce^4gn^3 at n , 2.17is essential and in the vhm domain _ ne^-4gn^4<o(e^-n ) . 2.18this result is an evident consequence of vacuum stability . it should be noted once more that the conditions ( ) considerably simplify calculations . * @xmath0-matrix unitarity constraints * to explain our idea , let us consider the spectral representation of the one - particle amplitude : a_1(x_1 , x_2 ; e)= , + 0 , 3ait describes the transition of a particle with energy @xmath10 from point @xmath784 to @xmath785 . according to our general idea , see introduction to sec.2.1 , we will calculate r_1(e)=dx_1 dx_2 a_1(x_1 , x_2 ; e)a^*_1 ( x_1 , x_2 ; e ) . 4 the integration over the end points @xmath784 and @xmath785 is performed only for the sake of simplicity . inserting ( ) into ( ) and using ortho - normalizability of the wave functions @xmath786 we find that & r_1(e)=_n||^2 = _ n\{-}= + & = _ n=_n(e - e_n ) . 5on other hand , the closed - path amplitude , offered for calculation in , & c_1(e)=_ndx= _ n= + & = _ n\{p+ i(e - e_n)}=_np+ir_1(e ) . 4so , we wish to calculate only the imaginary part of the closed - path contribution : @xmath787 notice the extra factor @xmath788 in the left hand side . the reason for this choice is evident : the real part of @xmath789 is equal to zero at @xmath790 , i.e. did not contribute to the measurable . to calculate the bound states energy spectrum , it is enough to know the only imaginary part of the closed - path amplitude . this property is not accidental . it is known as the optical theorem and is the consequence of the total probability conservation principles . formal realization of it is the unitarity condition for the @xmath791-matrix : @xmath792 . in terms of the amplitudes @xmath793 , @xmath794 , the unitarity condition presents an infinite set of nonlinear operator equalities : i*a**a*^ * = * a * - * a*^*. 2notice that expressing the amplitude by the path integral one can see that the left hand side of this equality offers the double integral and , at the same time , the right hand side is the linear combination of integrals . thus , the continuum contributions into the amplitudes should be canceled to provide the conservation of total probability . in this sense it is a necessary condition . indeed , to see the integral form of our approach , let us use the proper - time representation : a_1(x_1 , x_2 ; e)=_n _ n ( x_1)^*_n ( x_2)i ^_0dte^i(e - e_n+i)t 6 and insert it into ( [ 4 ] ) : r_1(e)=_n ^_0 dt_+dt_- e^-(t_++t_- ) e^i(e - e_n)(t_+-t_- ) . 7 we will introduce new time variables instead of @xmath795 : t_=t , 8 where , as follows from the jacobian of transformation , @xmath796 . but we can put @xmath797 since @xmath798 is essential in the integral over @xmath799 . as a result , _ 1(e)=2_n^_0 dt e^-2 t ^+_- e^2i(e - e_n ) . 9 in the last integral , the continuum of contributions with @xmath800 are canceled . note that the product of amplitudes @xmath801 was ` linearized ' after the introduction of ` virtual ' time @xmath802 . we wish to calculate the density matrix @xmath185 including the consequence of the unitarity condition cancelation of unnecessary contributions . here we demonstrate the result and the intermediate steps we will formulate , without proof , as the statements offered in , where the formalities are described . the statement , see and references cited therein , 0.1 cm _ s1 . the unitarity condition unambiguously determines contributions in the path integrals for @xmath803 _ 0.1 cm looks like a tautology since @xmath804 , where @xmath805 is the action , is the unitary operator which shifts a system along the trajectory . so , it seems evident that the unitarity condition is already included in the path integrals . the rule as the path integrals should be calculated is weel known , see e.g. . nevertheless the general path - integral solution contains unnecessary degrees of freedom ( unobservable states with @xmath806 in the above example ) . we would define the path integrals in such a way that the condition of absence of unnecessary contributions in the final ( measurable ) result be loaded from the very beginning . just in this sense , the unitarity looks like the necessary and sufficient condition unambiguously determining the complete set of contributions . _ s2 . the @xmath52- into @xmath4-particles transition ( unnormalized ) @xmath173 @xmath807 would have on the dirac measure the following symmetrical form _ : & r_nm(p_1, ... ,p_n , q_1, ... ,q_m)= < ^m_k=1|(q_k;u)|^2 ^n_k=1|(p_k;u)|^2>_u= + & = e^-i(j , e)dm(u)e^is_o(u)-iu(u , e ) ^m_k=1|(q_k;u)|^2 ^n_k=1|(p_k;u)|^2 + & ( u)^m_k=1|(q_k;u)|^2 ^n_k=1|(p_k;u)|^2 . 1.1ehere @xmath808 are the in(out)-going particle momenta . it should be underlined that this representation is strict and is valid for arbitrary lagrange theory of arbitrary dimensions . the eikonal approximation for inelastic amplitudes was considered in the functionals @xmath813 and @xmath814 are defined by the equalities : s_o(u)=(s_0(u+e)-s_0(u - e ) ) + 2_c_+dx dte(x , t)(^2+m^2)u(x , t ) , u(u , e)= v(u+e)-v(u - e)-2_c_+ dx dte(x , t)v(u ) , 2.3awhere @xmath815 is the free part of the lagrangian and @xmath816 describes interactions . the quantity @xmath814 is not equal to zero if @xmath817 have nontrivial topological charge ( see also ) . according to s1 , considering motion in the phase space @xmath818 the measure @xmath819 has the dirac form : dm(u , p)= _ x , t du(x , t ) dp(x , t ) ( - ) ( + ) 2.4awith the total hamiltonian h_j(u , p)=dx \ { p^2 + ( u)^2 + v(u)-ju } . 2.5athis last one includes the energy @xmath820 of quantum fluctuations . * a. * only @xmath821 solutions of equations -=0 , + = 0 equwith @xmath466 should be taken into account . this ` rigidness ' of the formalism means the absence of pseudo - solutions ( similar to multi - instanton , or multi - kink ) contribution . the parameter @xmath826 is connected directly with @xmath827 particle energy , momentum , spin , polarization , charge , etc . , and is sensitive to the symmetry properties of the interacting fields system.@xmath828 , if @xmath829 is the magnetics order parameter , and the phase transition means that @xmath830 . @xmath831 has just the same meaning as @xmath832 . ] for the sake of simplicity , @xmath833 is the real scalar field . the generalization would be evident . as a consequence of ( ) , @xmath826 is the function of the external particle momentum @xmath592 and is a @xmath834 functional of @xmath833 : ( q;u)=-dx e^iqx = dx e^iqx(^2 + m^2)u(x ) , q^2=m^2 , 1.2efor the mass @xmath52 field . this parameter presents the momentum distribution of the interacting field @xmath833 on the remote hypersurface @xmath536 if @xmath833 is the regular function . notice , the operator @xmath835 cancels the mass - shell states of @xmath833 . the construction ( ) means , because of the klein - gordon operator and the external states being mass - shell by definition , the solution @xmath836 is possible for a particular topology ( compactness and analytic properties ) of @xmath837 field @xmath833 . so , @xmath826 carries the following remarkable properties : notice , the space - time topology of @xmath838 becomes important in calculating integral ( ) by parts . this procedure is available if and only if @xmath838 is the regular function . but the @xmath837 fields are always singular . therefore , the solution @xmath839 is valid if and only if the quasiclassical approximation is exact . just this situation is realized in the soliton sector of sin - gordon model . despite evident ambiguity @xmath826 carries the definite properties of the order parameter since the opposite solution @xmath836 can only be the dynamical display of an @xmath840 symmetry - matrix was introduced ` phenomenologically ' , see also the example considered in , postulating the lsz reduction formulae , see eq . so , the formal constraints , e.g. the haag theorem , would not be taken into account on the chosen level of accuracy . ] , i.e. of the nontrivial topology of interacting fields , as the consequence of unbroken symmetry . if ( ) have nontrivial solution @xmath41 , then this ` extended objects ' quantization problem arises . we solve it by introducing convenient dynamical variables . the main formal difficulty , see e.g. , of this program consists of transformation of the path - integral measure which was solved in . then + _ s3 . the measure ( ) admits the transformation : u_c : ( u , p)(,)w = g / g_c . o11and the transformed measure has the form : dm(u , p)= _ x , t__c _ _ d(t ) d(t ) ( - ) ( + ) , o12where @xmath841 is the transformed hamiltonian . _ 0.1 cm it is evident that @xmath842 are parameters of integration of eqs . ( ) and they form the factor space @xmath843 . for instance , if one particle dynamics is considered , then one may choose @xmath844 and @xmath845 . one may consider also the following possibility : @xmath846 and @xmath847 in this terms @xmath848 and new hamilton equations have the form : = 1-j , = j. o13so , we have at @xmath466 : @xmath849 and @xmath850 . by this reason the ( action , angle)-type variables are mostly useful_. according to ( ) there exists transformation of the perturbation generating operator : + _ s5 . the operator @xmath851 has following transformed form : 2=dt \{__+ _ _ } , o14 in the factor space , where @xmath852 , @xmath853 , are new auxiliary variables . _ as a result of mapping of the perturbation generating operator @xmath851 on the manifold @xmath854 the equations of motion became linearized : dm=_t(--j _ ) ( -j _ ) . o21then + _ s6 . if the feynman s @xmath469-prescription is adopted , then the green function of eq . ( ) _ g(t - t)=(t - t ) o22 later on we will consider the soliton sector of sin - gordon model . in this case @xmath855 is the coordinate and @xmath856 is the momentum of @xmath174-th soliton and @xmath545 is the number of solitons . expansion of @xmath857 gives the ` strong coupling ' perturbation series . its analysis shows that + _ s7 . action of the integro - differential operator @xmath809 leads to the following representation _ : & r_nm(p , q)=_w\ { d(0)r^_nm(p , q).+ + & . d(0)r^_nm(p , q)}. o20this means that the contributions into @xmath98 are accumulated strictly on the boundary ` bifurcation manifold ' @xmath858 , i.e. depends directly on the topology of @xmath854 . performing the shifts in ( ): _ i ( t ) _ i ( t ) + dt g(t - t ) j_,i(t ) _ i ( t ) + _i ( t ) , + _ i ( t ) _ i ( t ) + dt g(t - t ) j_,i(t ) _ i ( t ) + _i ( t ) , 4.16we can get the green function @xmath859 into the operator exponent : ( ej ) = dt dt ( t - t)\{(t)_(t ) + (t)_(t)}. note the lorentz noncovariantness of our perturbation theory with green function ( ) . using the definition : @xmath860 the functional integrals on the measure ( ) are reduced to the ordinary integrals over initial data @xmath861 . these integrals define zero modes volume . notice that the zero - modes measure was defined without the faddeev - popov @xmath186 . this strategy is necessary since it seems to be important to show the role of quantum corrections noting that for all physically acceptable field theories @xmath862 in the quasiclassical approximation . the @xmath545-soliton solution @xmath863 depends upon @xmath864 parameters . half of them , @xmath545 , can be considered as the position of solitons and the other @xmath545 as the solitons momentum . generally at @xmath865 the @xmath863 solution decomposed on the single solitons @xmath866 and on the double soliton bound states @xmath867 : @xmath868 note that this asymptotic is achieved if @xmath869 or / and @xmath870 . this last one defines the bifurcation line of our model . so , the one soliton @xmath866 and two - soliton bound state @xmath867 would be the main elements of our formalism . its @xmath842 parametrizations have the form : u_s(x;,)=-\{(mx- ) } , = so and u_b(x;,)= -\ { } . bo in the quasiclassical approximation @xmath871 we have : u_n = u_n ( x;_0 , _ 0 + (_0)t ) . 5.7notice that the surface term dx^_(e^iqxu_n)=0 . 5.8then d^2x e^iqx(^2 + m^2)u_n ( x , t ) = -(q^2-m^2)d^2x e^iqxu_n ( x , t ) = 0 5.9since @xmath872 belongs to the mass shell by definition . the condition ( ) is satisfied for all @xmath873 since @xmath863 belongs to schwarz space ( the periodic boundary condition for @xmath838 do not alter this conclusion ) . therefore , in the quasiclassical approximation @xmath874 . expanding the operator exponent in ( [ 5.1 ] ) we find that action of operators @xmath875 , @xmath876 creates terms ~d^2x e^iqx ( t - t ) ( ^2 + m^2)u_n ( x , t ) 0 . 5.12so , generally , if the quantum corrections are included , @xmath877 . now we will show that the quasiclassical approximation is exact in the soliton sector of sin - gordon model . the structure of the perturbation theory is readily seen in the ` normal - product ' form : r_2(q)=_n ^n_i=1 \{d_0 d_0}_i : e^-iu(u_n;/2i)e^is_0(u_n)|(q;u_n)|^2 : , 6.1where = _ - _ = _x 6.2and _ x = dt ( t - t)(t ) 6.3with the @xmath864-dimensional vector @xmath878 . in eq . ( ) @xmath879 is the ordinary symplectic matrix . the colons in ( ) mean that the operator @xmath880 should stay to the left of all functions in the perturbation theory expansion over it . the structure ( ) shows that each order over @xmath881 is proportional at least to the first order derivative of @xmath863 over the variable conjugate to @xmath882 . the expansion of ( ) over @xmath883 can be written in the form ( omitting the quasiclassical approximation ) : r_2(q)=_n ^n_i=1 \{d_0 d_0}_i \{^2n_i=1p_x_i(u_n ) } , 6.4where @xmath884 is the infinite sum of ` time - ordered ' polynomials ( see ) over @xmath863 and its derivatives . the explicit form of @xmath884 is complicated since the interaction potential is nonpolynomial . but it is enough to know , see ( ) , that p_x_i(u_n)~_ij . 6.5 authors would like to take the opportunity to thank j. allaby , a. m. baldin , j. a. budagov , r. cashmore , g. a. chelkov , m. della negra , d. denegri , a. t. filippov , d. froidevaux , f. gianotti , i. a. golutvin , p. jenni , v. g. kadyshevski , o. v. kancheli , z. krumstein , e. a. kuraev , e. m. levin , l. n. lipatov , l. maiani , v. a. matveev , v. a. nikitin , a. nikitenko , a. g. olchevski , g. s. pogosyan , n. a. russakovich , v. i. savrin , m. v. saveliev , ju . schukraft , n. b. skachkov , a. n. tavkhelidze , v. m. ter - antonyan , h. t. torosyan , n.e . tyurin , d. v. shirkov , e. sarkisian , s. tapprogge , o. i. zavialov , g. zinovyev for valuable discussions and important comments . the scientific communications with them on the various stages were important . we acknowledged also to members of the seminar ` symmetries and integrable systems ' of the n.n.bogolyubov laboratory of theoretical physics ( jinr ) , to the scientific seminars of the v. p. dzelepov laboratory of nuclear problems , d. v. scobeltsyn inp of the moscow state university , of the atlas experiment community for a number of important discussions . we would like also to thank m. gostkin and n. shubitidze for the help in the formulation of the monte carlo simulation programs and for the discussions . one of us ( j.m . ) was supported in part by georgian academy of sciences . a.n.sissakian and l.a.slepchenko , preprint jinr , p2 - 10651 , 1977 ; + a.n.sissakian and l.a.slepchenko , fizika , 10 ( 1978 ) 21 ; + s. ch . mavrodiev , v. k. mitryushkin , a. n. sissakian and h. t. torosyan , sov . phys . , 30 ( 1979 ) 245 r.feynman , phys . lett . , 23 ( 1969 ) 1415 ; + a.a.logunov and m.a.mestvirishvili , cern th/1659 ; + a.a.logunov , m.a.mestvirishvili , nguen van hieu , phys . , 25b ( 1967 ) 611 ; + a.a.logunov and m.a.mestvirishvili , cern , th-1707 , 1973 g.giacomelli , nucl . phys . 25b ( 1992 ) 30 ; + g.giacomelli , boll . fis . , nuova ser . 6 ( 1990 ) 20 , int a5 ( 1990 ) 223 ; + h.d.wahl , acta phys . , xxy ( 1983 ) 399 ; + a.b.kaidalov , survey in high energy physics , 7 ( 1994 ) 143 ; + p.g.ratcliff , proc . 8th int . symp . in high - en . spin phys . , minneapolis , vol.1 ( 1988 ) 145 ; + d.grosh et al . , rev . , d49 ( 1994 ) 3113 ; + proc . int . symp . in hadron interaction , eds . j.fischer , p.kolar , v.kundrat , czech , acad . ( 1988 ) 431 e.kuraev , l.lipatov and v.fadin , sov . jetp , 44 ( 1976 ) 443 ; zh . fiz . , 71 ( 1976 ) 840 ; + l.lipatov , sov.j . 20 ( 1975 ) 94 ; + v.gribov and l.lipatov , sov.j . phys . , 15 ( 1972 ) 438 , 675 ; + g.altarelli and g.parisi , nucl . phys . , b126 ( 1977 ) 298 ; + i.v.andreev , chromodynamics and hard processes at high energies ( nauka , moscow , 1981 ) i.p.kadanoff and g.baym , qunantum statistical mechanics ( benjamin , london , 1962 ) ; + g.niklasson and a.sjolander , ann.phys . , 49 ( 1968 ) 249 ; + p.danielevich , ann.phys . , 152 ( 1984 ) 239 + p.kadanoff and p.c.martin , ann.phys.(ny ) , 24 ( 1963 ) 419 e.wigner , phys . , 40 , ( 1932 ) 749 ; + k.hisimi , proc . jap . , 23 ( 1940 ) 264 ; + r.j.glauber , phys . 10 ( 1963 ) 84 ; + e.c.g.sudarshan , phys . rev . 10 ( 1963 ) 177 ; + r.e.cahill and r.g.glauber , phys . , 177 ( 1969 ) 1882 ; + s.mancini , v.i.manko and p.tombesi , quant . , 7 ( 1995 ) 615 ; + v.i.manko , l.rose and p.vitale , hep - th/9806164 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: strongly correlated fermionic systems , responsible for relevant many - body phenomena such as high - temperature superconductivity , the fractional quantum hall effect or metal - insulator transitions , represent one of the most important theoretical challenges in condensed matter physics . among the simplest possible models of interacting fermions in a 2d lattice is the hubbard model,@xcite which is believed to be one of the keys to understanding the theoretical riddle of high - temperature superconductivity,@xcite and which serves as a good example to illustrate the nature and scale of the difficulties encountered . in spite of a titanic effort by the condensed matter community spanning several decades , still today the phase diagram of the 2d hubbard model and its relation to high - temperature superconductors remain highly controversial . in the absence of exactly solvable models , accurate numerical simulations are essential in order to gain further insight into the physics of strongly correlated systems . while quantum monte carlo ( qmc ) techniques are very powerful in simulating bosonic systems , they suffer from the so - called _ negative sign problem _ in the case of fermionic and frustrated models . @xcite on the other hand , generic 1d lattice systems can be accurately addressed with the density matrix renormalization group ( dmrg ) method , @xcite but this approach scales inefficiently with the lattice size in 2d systems . recent progress in the simulation of 2d fermionic models has been made with a variety of methods . @xcite however , results obtained with different methods are often inconsistent , highlighting the need for further improvement and for alternative approaches . a promising new route to studying strongly correlated fermion systems in a 2d lattice , presently under intense investigation , @xcite is based on using a _ tensor network _ as ground state variational ansatz . for bosonic ( e.g. spin ) 2d lattice models , tensor network anstze include _ projected entangled - pair states _ ( peps ) for inhomogeneous@xcite and homogeneous systems , @xcite and the _ _ multi - scale entanglement renormalization ansatz__@xcite ( mera ) . [ homogeneous peps are also known with names such as ( vertex ) _ _ tensor product states__@xcite ] . the interest in these approaches resides in the fact that they manage to retain some of the useful features of dmrg and qmc , while avoiding their main shortcomings . indeed , peps and mera approaches are free from the negative sign problem that prevents the application of qmc to fermionic and frustrated models . at the same time , and unlike dmrg , both peps and mera can efficiently represent ground states of 2d lattice models . in addition , compared to other variational approaches , peps and mera are relatively unbiased towards specific ground states . still at an early stage of development , the major limitation of these methods is that the cost of simulations increases sharply with the amount of entanglement in the ground state . this limits the range of models that can be analyzed accurately at present . nevertheless , several systems of frustrated antiferromagnets beyond the reach of dmrg and qmc have already been addressed.@xcite in recent months , generalizations of tensor network algorithms to fermionic systems have been put forward independently by several groups.@xcite as a result , it is now possible to study interacting fermions in 2d lattices both within the context of the mera @xcite and peps . @xcite the fundamental new step , common in all the proposals , is to incorporate the fermionic character of the ground state wave function directly into the ansatz . this is accomplished by considering a _ network of fermionic operators _ , that is , a set of linear maps , made of anticommuting operators , that are connected according to a network pattern , as first proposed in refs . for the mera and in ref . for the peps . in actual computations , one is still forced to store and manipulate tensors ( i.e. multi - dimensional arrays of coefficients ) corresponding e.g. to the expansion coefficients of the fermionic maps . the process of ` _ _ fermionization _ _ ' of a tensor network algorithm , i.e. its extension to fermionic systems , can in practice be achieved and visualized in a variety of ( ultimately equivalent ) ways , depending on how the underlying network of fermionic operators is translated into a set of tensors and rules for their manipulation . examples include the use of a jordan - wigner transformation , @xcite or the introduction of additional bond indices between tensors.@xcite a particularly simple form of ` _ _ fermionization _ _ ' of tensor networks was introduced in ref . , where it was applied in the context of the mera . we emphasize that ref . is based on reformulating previous work by corboz , evenbly , verstraete and vidal on fermionic mera,@xcite which in turn had its origins in a key observation by verstraete.@xcite the fermionization procedure of ref . , which applies to any tensor network , is remarkedly simple . it does not require the introduction of a jordan - wigner transformation in the bond indices , or to have to explicitly keep track and dynamically modify a global fermionic order ; neither does it require the introduction of additional bond indices in the tensors . instead , the fermionic character of the tensor network is engraved in two simple rules : ( i ) use of parity invariant tensors and ( ii ) replacement of line crossings with so - called fermionic swap gates . the net result of applying these rules is a modified variational ansatz that can be manipulated using standard tensor network operations ( tensor multiplications , etc . ) , thus producing a straightforward fermionic version of existing tensor network algorithms . importantly , the computational cost of bosonic and fermionic algorithms scales in the same way with the amount of entanglement in the ground state . @xcite this remarkable result was also independently derived in ref . . this paper has two main goals . the first is to explain how to obtain peps algorithms for fermionic systems by applying the above ` _ _ fermionization _ _ ' rules to existing bosonic peps algorithms . fermionic peps were originally proposed by kraus , schuch , verstraete and cirac in ref . and have also been discussed by barthel , pineda and eisert in ref . . our formulation of fermionic peps must be , at some level , equivalent to those proposals . however , the present formulation , which is based on previous independent work,@xcite is remarkably straightforward and appears to be comparatively much simpler . in particular , it allowed us to numerically implement a fermionic peps algorithm for infinite systems by only introducing a small number of changes to existing code for bosonic systems . a second main goal of this paper is to demonstrate the usefulness of fermionic peps . in spite of the several existing formulations of fermionic peps , @xcite and with the exception of ref . , where some qualitative results are reported for the @xmath0 model , no evidence has been presented yet showing that fermionic peps are a good variational ansatz for interacting fermion systems . [ notice , however , that ref . shows that gaussian fermionic peps can represent states of non - interacting fermions ] . here we do present such evidence , in the form of ground state computations for several 2d models . specifically , we use a fermionic version of the infinite peps ( ipeps ) algorithm@xcite to address models on an infinite square lattice . first , results for free spinless fermions are compared with the corresponding exact solution , showing that a peps with small bond dimension is capable of reproducing the ground state energy with several digits of accuracy . then a model of interacting spinless fermions is addressed . qualitatively , the simulation reproduces the phase diagram predicted within hartree - fock , with metal and charge - density wave phases separated by a line of first order phase transitions . at a quantitative level , however , we obtain ground state energies that are lower than those obtained with hartree - fock , and this shifts the boundary between phases significantly . finally , for the @xmath0 model , we obtain ground state energies that are close to those of a specialized gutziller - projected ansatz . the rest of the paper is organized as follows : sec . [ sec : peps ] reviews the peps formalism for bosonic systems and the general fermionization procedure of tensor networks introduced in ref . , which is then applied to peps algorithms . [ sec : ctm ] considers in more detail the fermionic version of the ipeps algorithm for infinite 2d lattices , which was employed to obtain the benchmark results presented in this paper [ sec : results ] describes ground state calculations for systems of free and interacting fermions in an infinite 2d lattice . [ sec : conclusions ] contains some conclusions , while appendix [ app : genoperators ] defines generalized fermionic operators and appendix [ app : twosite ] describes in detail one step of the update in the fermionic ipeps algorithm . _ note on terminology. _ for the purposes of this paper , a _ tensor _ is simply a multi - dimensional array of complex coefficients , and a _ tensor network _ is a set of tensors some of whose indices are connected according to a network pattern , where being connected means that there is a sum or trace over that index , in the sense of tensor multiplication . accordingly , in this paper a bosonic / fermionic tensor network is a tensor network used in the context of simulating a bosonic / fermionic system . thus , in the present formulation a fermionic peps is simply a `` tensor network that serves as a variational ansatz for fermionic systems '' . it is different from a bosonic peps in the presence of special gates called fermionic swap gates ( and in that its tensors are necessarily parity preserving ) . in particular , even though the rules used to create a fermionic tensor network , as introduced in ref . and reviewed here , were obtained by studying how to mimic a network of fermionic operators ( that is , of operators that obey anticommuting relations ) , here a fermionic peps is _ not _ a network of fermionic operators . one of the merits of the present formulation is precisely that it replaces the considerable complexity involved in dealing with a network of fermionic operators with a simple set of rules . in particular , it avoids having to explicitly define , keep track and dynamically modify a fermionic order for the bond indices . the equivalence between our formulation of a fermionic tensor network and a network of fermionic operators was already established in ref . for the case of the mera . a general derivation of this equivalence would distract from the purpose of this paper and will be presented elsewhere . the goal of this section is to introduce a fermionic version of bosonic peps algorithms,@xcite so that they can be applied to simulate fermionic systems in a 2d lattice . we start by reviewing some key aspects of the peps formalism for bosonic systems . this allows us to introduce the notation and the diagrammatic representation of tensors used throughout secs . [ sec : peps ] and [ sec : ctm ] . then we describe the fermionization rules of ref . , which we also extensively review . we apply these rules to obtain a fermionic peps ansatz ( see also refs . ) , and provide a discussion of how fermionic peps algorithms can be obtained by modifying existing bosonic peps algorithms . let us consider a quantum many - body system in a lattice @xmath1 made of @xmath2 sites , labelled by an integer @xmath3 . each site @xmath4 is described by a complex vector space @xmath5 of finite dimension @xmath6 , with basis states @xmath7 . the vector space @xmath5 could represent , for instance , the possible states of a quantum spin sitting on that site of @xmath1 . the system is further characterized by a _ local _ ( _ bosonic _ ) hamiltonian . this is a hermitian operator @xmath8 that ( when expressed in terms of bosonic operators , i.e. operators that commute when acting on different sites ) decomposes as a sum of terms each involving only a small number of sites . let @xmath9 be a pure state , @xmath10 where index @xmath11 labels a basis on site @xmath4 . a task of interest is to compute a specific state @xmath12 somehow related to @xmath13 , e.g. its ground state , and to evaluate the expectation value @xmath14 of some local observable @xmath15 . however , representing a vector @xmath16 requires a number of complex coefficients @xmath17 that grows exponentially in @xmath2 . this poses a serious computational challenge . exact diagonalization techniques are only affordable for small systems ( e.g. , at most @xmath18 for @xmath19 ) , and alternative numerical strategies are required to analyze large systems . projected entangled - pair states@xcite ( peps ) were introduced as a means to obtain an efficient description for some states @xmath16 of a 2d lattice @xmath1 . for concreteness , in this work we consider the case of a square lattice , although all the discussions can be extended to other type of lattices . to each site @xmath20 there corresponds a vector of integers @xmath21 , and we also write @xmath22 to denote a site of lattice @xmath1 . a peps is made of a collection of @xmath2 tensors @xmath23}\}$ ] , one for each site @xmath24 , connected through _ bond indices _ that follow the pattern of links of the lattice @xmath1 . upon tracing over all bond indices , a peps yields a tensor @xmath25 with the @xmath26 complex coefficients @xmath17 of a state @xmath27 . with coefficients @xmath28 for a state @xmath29 of a @xmath30 square lattice this tensor is expressed in terms of a peps made of a set of 9 tensors @xmath31}\}$ ] , one for each site @xmath32 . ( b ) bulk tensor @xmath33}$ ] with components @xmath33}_{ulsdr}$ ] . notice that the legs corresponding to indices @xmath34 , @xmath35 , @xmath36 , @xmath6 and @xmath37 emerge from the circle in anti - clockwise order . ( c ) hermitian conjugate @xmath33\dagger}$ ] of the peps tensor @xmath33}$ ] , represented as its mirror image . notice that the legs corresponding to indices @xmath37 , @xmath6 , @xmath36 , @xmath35 and @xmath34 of @xmath38\dagger})_{rdslu}$ ] emerge again in anti - clockwise order . ( d ) hermitian conjugates of @xmath25 and its peps representation.,width=302 ] throughout this paper a diagrammatic representation of tensors and tensor networks is used , see fig . [ fig : bosonpeps ] . each tensor is depicted as a shape ( circle , square , diamond , etc ) and its indices as emerging lines . a line connecting two shapes ( or starting and ending at the same shape ) denotes an index over which a trace is taken . as an example , fig . [ fig : bosonpeps](a ) represents , in the case of a lattice @xmath1 made of @xmath30 sites , a tensor @xmath25 with 9 indices , corresponding to the coefficients @xmath39 of @xmath40 , followed by a peps made of 9 tensors @xmath31}\}$ ] . the number of indices in a tensor @xmath33}$ ] depends on the number of nearest neighbors of site @xmath22 , with tensors in the bulk having more indices than tensors at a boundary . specifically , a bulk tensor has components @xmath33}_{ulsdr}$ ] , with one _ physical _ index @xmath36 and four _ bond _ indices @xmath41 . the physical index @xmath36 labels the basis of the vector space @xmath5 for site @xmath24 and therefore takes @xmath6 different values , whereas each bond index connects the tensor with a tensor in a nearest neighbor site and ranges from @xmath42 to @xmath43 , where @xmath43 is the so - called _ bond dimension _ of the peps . correspondingly , a bulk peps tensor is represented by a circle with five legs , see fig . [ fig : bosonpeps](b ) . as in the rest of the paper , here we follow the prescription that the indices of a tensor are drawn in anti - clockwise order . notice also that the open indices of the peps in fig . [ fig : bosonpeps](a ) reach the exterior of the tensor network in exactly the same ( anti - clockwise ) order that they appear in @xmath44 . these notational and diagrammatical details are somewhat superfluous in the bosonic case ( since one can change the order of indices in a tensor by simply permuting its components ) but will become important in the extension to fermions . a peps on a @xmath45 lattice , where @xmath46 , contains @xmath47 bulk tensors , each depending on @xmath48 complex coefficients . therefore the peps is characterized by @xmath49 parameters . if @xmath43 has a fixed value independent of @xmath2 , then the peps is indeed an efficient encoding of some states @xmath16 , and it can be used e.g. as a variational ansatz for the ground state of @xmath13 . however , for the peps to be a useful ansatz , we also need to provide an efficient strategy to optimize its tensors and manipulate them in order to extract physically relevant information , as discussed next . two major tasks to be accomplished with a peps algorithm@xcite are : ( i ) optimization of its @xmath49 parameters so as to obtain a good approximation to e.g. the ground state of the local hamiltonian @xmath13 ; ( ii ) given a peps for a state @xmath12 , computation of expectation values @xmath14 of local observables @xmath15 , as given by @xmath50 these two tasks happen to be closely related . they involve taking the trace over all the bond indices of a composite tensor network , an operation referred to as _ contracting _ the tensor network . written in terms of tensors @xmath25 and @xmath51 . ( b ) the same scalar product , but written in terms of a peps together with its hermitian conjugate . ( c)-(d ) using the jump move of fig . [ fig : jump ] , this tensor network can be modified so that each tensor @xmath33}$ ] is drawn next to its hermitian conjugate @xmath33}$ ] . ( e ) tensor network @xmath52 in terms of reduced tensors @xmath53}$ ] . ( f ) @xmath53}$ ] defined in terms of @xmath33}$ ] and @xmath33\dagger}$ ] , eq . .,width=302 ] an emblematic example of tensor network contraction required in a peps algorithm concerns the computation of the scalar product @xmath54 . we need to introduce a few more definitions and notation . let @xmath55 denote the hermitian conjugate of a tensor @xmath56 , obtained by reversing the order of the indices of @xmath56 and taking the complex conjugate of each of its coefficients ( and diagrammatically represented as the mirror image of @xmath56 ) . for instance , tensor @xmath51 in fig . [ fig : bosonpeps](d ) , corresponding to @xmath57 , @xmath58 has coefficients @xmath59 fig . [ fig : bosonpeps](c ) represents the hermitian conjugate @xmath33\dagger}$ ] of a bulk tensor @xmath33}$ ] , which has coefficients @xmath60\dagger})_{rdslu } = { a^{[\vec{r}]}_{ulsdr}}^*.\ ] ] we build the hermitian conjugate of the peps for @xmath25 as the set of tensors @xmath31\dagger}\}$ ] connected according to the mirror image of the network of connections in the peps for @xmath25 , see fig . [ fig : bosonpeps](d ) . for each site @xmath22 , let us define the reduced tensor @xmath53}$ ] in terms of tensor @xmath33}$ ] and @xmath33\dagger}$ ] by tracing over the physical index @xmath36 . for instance , in the bulk , the reduced tensor @xmath53}$ ] has components @xmath61}_{u \bar{u } l \bar{l } \bar{d } d \bar{r } r } \equiv \sum_{s=1}^d a^{[\vec{r}]}_{ulsdr } ( a^{[\vec{r}]\dagger})_{\bar{r } \bar{d } s \bar{l } \bar{u } } , \label{eq : a}\ ] ] see fig . [ fig : bosonscalar](f ) . then the scalar product @xmath54 results from contracting a tensor network @xmath52 that consists of all reduced tensors @xmath53}$ ] connected according to the links in @xmath1 , see fig . [ fig : bosonscalar](a)-(e ) , where the ( trivial ) jump move of fig . [ fig : jump ] is used . contracting the tensor network @xmath52 to obtain the scalar product @xmath54 comes with a cost that grows exponentially in the linear size @xmath62 of lattice @xmath1 , and therefore can not be accomplished efficiently . a key ingredient of peps algorithms is precisely a strategy to _ efficiently _ but approximately contract the tensor network @xmath52 , thus producing an approximation to @xmath54 . this can be done in several ways , depending on the size and topology of lattice @xmath1 . in a finite lattice with open boundary conditions , one can use matrix product state ( mps ) techniques . @xcite in the case of a torus , coarse - graining techniques known as _ tensor entanglement renormalization group _ ( terg ) @xcite can be used . finally , in an infinite lattice , both infinite mps@xcite and _ corner transfer matrix _ ( ctm)@xcite techniques have been employed . in order to optimize a peps so as to approximate the ground state of @xmath13 , as well as to evaluate the expectation value of local observables , it is useful to contract certain class of tensor networks called _ environments_. the environment @xmath63}$ ] of a site @xmath22 is the tensor network obtained from @xmath52 by removing tensor @xmath53}$ ] , and can be used to compute the expectation value of a local observable acting on that site . , made of reduced tensors @xmath53}$ ] , corresponding to the scalar product @xmath54 in an @xmath45 lattice @xmath1 with open boundary conditions . ( b ) environment @xmath64}$ ] for two contiguous sites @xmath65 , obtained from @xmath52 by removing tensors @xmath66}$ ] and @xmath67}$ ] . ( c ) approximate environment @xmath68}$ ] consisting of six tensors @xmath69 . ( d ) tensor @xmath70 of coefficients for a two - site local operator @xmath15 , eq . . ( e ) tensor network made of the approximate environment @xmath68}$ ] and tensors @xmath71}$ ] , @xmath71\dagger}$ ] , @xmath72}$ ] , @xmath72\dagger}$ ] and @xmath70 . its contraction produces an approximation to @xmath73.,width=321 ] similarly , a two - site environment @xmath64}$ ] is the tensor network obtained by removing tensors @xmath66}$ ] and @xmath67}$ ] from @xmath52 , and can be used e.g. to compute the expectation value of a two - site observable @xmath74 acting on sites @xmath75 . figure [ fig : bosonenv](b ) shows the environment @xmath64}$ ] corresponding to two nearest neighbor sites @xmath76 . again , the exact contraction of the environment can not be performed efficiently , but efficient schemes , analogous to those employed to contract @xmath52 , can be used in order to approximately contract @xmath77}$ ] . the whole environment is in this way approximated by a smaller tensor network @xmath68}$ ] made of 6 tensors @xmath78 , see fig . [ fig : bosonenv](c ) . these tensors can then be connected to tensors @xmath71}$ ] , @xmath71\dagger}$ ] , @xmath72}$ ] , @xmath72\dagger}$ ] and @xmath70 to produce an approximation to @xmath73 , see fig . [ fig : bosonenv](e ) . in a system with a hamiltonian @xmath13 made of two - site interactions @xmath79}$ ] between nearest neighbors , the approximate environment @xmath68}$ ] is particularly useful . on the one hand , it is employed in the optimization of tensors @xmath71}$ ] and @xmath72}$ ] after a gate has been applied on the two sites , as part of simulating an _ imaginary - time evolution _ according to @xmath13 , @xmath80 which offers one way of obtaining a peps approximation to the ground state @xmath81 of @xmath13 . on the other hand , @xmath82}$ ] can also be used to compute the expectation value of the energy on that link , @xmath83 } { \mbox{$| \psi \rangle$}}$ ] , as part of an algorithm to optimize the peps by minimizing its energy , @xmath84}\ } } \frac { { \mbox{$\langle \psi_{\{a^{[\vec{r}]}\ } } |$}}~ \hat{h}~ { \mbox{$| \psi_{\{a^{[\vec{r}]}\ } } \rangle$ } } } { { \mbox{$\langle \psi_{\{a^{[\vec{r}]}\ } } | \psi_{\{a^{[\vec{r}]}\ } } \rangle$ } } } \label{eq : mine}\ ] ] which is another way of obtaining a peps approximation to the ground state @xmath81 of @xmath13 . we refer to refs . for more details . other peps algorithms@xcite bypass the computation of environments . this concludes our short review of peps algorithms for bosonic 2d lattice models . let us now consider a fermionic lattice system . for the sake of simplicity , we first assume that each site @xmath20 is described by a complex vector space @xmath5 of dimension @xmath19 that is associated to a fermionic annihilation operator @xmath85 , with anticommutation relations @xmath86 [ we will shortly extend the discussion to sites with vector space of finite dimension @xmath87 , see also appendix [ app : genoperators ] ] . a basis of the vector space @xmath88 of the lattice system is given by @xmath89 recall that fermionic operators can be expressed in terms of pauli matrices @xmath90 by means of a jordan - wigner transformation , @xmath91 the fermionic lattice system @xmath1 is further characterized by a local fermionic hamiltonian @xmath13 . this is a hamiltonian that , when expressed in terms of the fermionic operators @xmath92 , decomposes as the sum of terms involving only a small number of sites . as in eq . , let @xmath16 be a pure state of lattice @xmath1 , @xmath93 here we will assume that @xmath12 is somehow related to the fermionic hamiltonian @xmath13 , for instance it is its ground state . once more , we would like to find a variational ansatz depending on @xmath47 parameters to efficiently encode the tensor @xmath25 containing the @xmath26 coefficients @xmath17 of a pure state @xmath12 . one possibility would be to use a peps exactly as in the bosonic case . however , this might not be a good idea . remember that the label @xmath94 provides an order to the set of sites of @xmath1 , whose position in the lattice is given by @xmath95 . two nearest neighbor sites @xmath96 and @xmath97 on the square lattice might correspond to values @xmath98 and @xmath99 that are far apart . then , when expressed in terms of pauli matrices , the local fermionic hamiltonian @xmath13 will no longer look local . for instance , a nearest neighbor hopping term @xmath100 develops a string of @xmath101 s . this might be harmful in two ways . on the one hand , the presence of strings of @xmath101 s would require important modifications in the algorithms to approximate the ground state of @xmath13 with a peps , either through imaginary - time evolution or energy minimization . on the other hand , it is unclear that the peps itself , which was originally designed as an ansatz for ground states of local bosonic hamiltonians , will be as good an ansatz also for ground states of fermionic hamiltonians , given that the latter are non - local when expressed in bosonic variables . below we will explain how to modify the peps so that it is suitable to study fermionic systems ( see also refs . ) . before , however , we introduce the notation necessary to deal with local vector spaces @xmath5 of dimension @xmath87 . fermionic systems are governed by hamiltonians that preserve the _ parity of the fermionic particle number _ , to which we refer simply as ` _ _ parity _ _ ' . that is , fermions can only be created or annihilated in pairs , and parity is a constant of motion . as a result , we can assume that the pure state @xmath102 of lattice @xmath1 has a well - defined parity , and observables @xmath15 and reduced density matrices are block diagonal in parity . let us consider again the vector space @xmath5 of a single site , now with finite dimension @xmath87 . it is natural to decompose @xmath5 as the direct sum of an even parity subspace @xmath103 and an odd parity subspace @xmath104 , @xmath105 and to choose a basis of vectors with well - defined parity . accordingly , the physical index @xmath36 describing one such basis is decomposed as @xmath106 , where @xmath107 is the parity and @xmath108 ( denoted @xmath109 and @xmath110 ) enumerates the different basis states with parity @xmath111 . the parity operator @xmath112 then acts on this basis as @xmath113 in the case of spinless fermions , with a local dimension @xmath19 , the two possible states of a site are the local vacuum @xmath114 , signaling the absence of a fermion , and the state @xmath115 , corresponding to one fermion . in our notation these states read @xmath116 in the case of the @xmath0 model there are three possible states per site , @xmath117 , where @xmath118 and @xmath119 denote an electron with spin up and spin down , respectively . in our notation these states read @xmath120 finally , in the case of the hubbard model with local dimension @xmath121 there is an additional state , @xmath122 , corresponding to a doubly occupied site , which in our notation reads @xmath123 in analogy with the physical index @xmath106 , we introduce a parity operator also on the bond indices of a peps . accordingly , the tensor @xmath33}_{ulsdr}$ ] has bond indices @xmath124 , etc . this means that e.g. bond index @xmath34 can take values @xmath125 where the bond dimension @xmath43 is given by @xmath126 . the actual values of @xmath127 and @xmath128 can be chosen at convenience . ) ; ( b ) line crossings are replaced with a special gate , a fermionic swap gate @xmath129.,width=264 ] given a peps for bosonic systems , cf . [ fig : bosonpeps](a ) , in this work we obtain a peps for fermionic systems by applying the two rules used in ref . to fermionize the mera . these rules are applied both to the peps and to all related tensor networks that are involved e.g. in optimizing the ansatz or computing expectation values of local observables . * rule 1 : * each tensor @xmath56 in a tensor network is chosen to be _ parity preserving _ , i.e. @xmath130 where @xmath131 denotes the _ parity _ of the basis state labelled by @xmath132 , see fig . [ fig : rules](a ) . * rule 2 : * each crossing of lines in the tensor network is replaced with a _ fermionic swap gate _ @xmath133 , see fig . [ fig : rules](b ) . this gate implements a fermionic exchange and has the form @xmath134with @xmath135 given by @xmath136 of a @xmath30 lattice @xmath1 , obtained from the bosonic peps in fig . [ fig : bosonpeps](a ) by replacing line crossings with fermionic swap gates . the coefficients @xmath140 are obtained by contracting this fermionic peps using standard tensor multiplication techniques ( see text ) . the presence of fermionic swap gates introduces a complex structure of minus signs in @xmath140 . ( b ) bulk tensor @xmath33}$ ] , which is chosen to be parity preserving , eq . . ( c ) hermitian conjugate of @xmath33}$ ] . ( d ) scalar product @xmath54 written as a tensor network of reduced tensors @xmath53}$ ] defined in ( e ) . notice that , in a @xmath45 lattice , @xmath141 fermionic swap gates pervade the peps and yet , thanks to the jump move , fig . [ fig : jumpmove ] , it is possible to write the scalar product @xmath54 in terms of only @xmath142 fermionic swap gates and in such a way that all of them are near a pair @xmath38},a^{[\vec{r}]\dagger})$ ] and can therefore be absorbed into the definition of the reduced tensor @xmath53}$ ] . [ for a detailed derivation , replace line crossings with fermionic swap gates in fig . [ fig : bosonscalar ] ] . as a result , there are no fermionic swap gates left in @xmath52 , and therefore this tensor network can be contracted using the techniques employed for bosonic peps.,width=264 ] rule 1 is very convenient from a computational perspective : it ensures that the parity of the wave function is exactly preserved during ( otherwise approximate ) calculations , while the block diagonal structure of tensors can be exploited to reduce computational costs . in addition , rule 1 is important in order to account for the antisymmetric character of fermionic wavefunctions ( rule 2 ) in a simple way . however , we emphasize that tensor networks made of parity preserving tensors are also useful to describe bosonic systems ( e.g. , a @xmath143 invariant spin system , such as the quantum ising model ) and therefore rule 1 is not what turns a bosonic tensor network ansatz into a fermionic one . rule 2 accounts for the fermionic character of the tensor network , in the sense that it is employed to mimic the effect of anticommutators in a network of fermionic operators , as justified in ref . in the context of the mera ( see also the note on terminology in the introduction of this paper ) . several additional remarks concerning fermionic tensor networks and their manipulations are in order . we start with a number of comments on fermionic tensor network representations that are relevant to the present formulation of the fermionic peps ansatz . \(i ) _ fermionic order_. the label @xmath144 for the sites of @xmath1 establishes an order on these sites . this order has been used to define a local basis of the hilbert space in eq . , and can also be used to translate the local fermionic lattice model into a non - local bosonic one through the jordan - wigner transformation of eq . ( although this is not the strategy that we follow here ) . according to our prescription to graphically represent tensors and tensor networks , this order is also the order in which the open indices are drawn in fig . [ fig : fermipeps](a ) . notice that the structure of line crossings in the peps depends on the fermionic order and therefore the number and location of fermionic swap gates , see fig . [ fig : orders ] . notice also that , in contrast with refs . , we do not explicitly introduce fermionic operators ( and corresponding order ) on the bond indices of the peps , but use instead a simple graphical notation and two rules that already account for the complex pattern of fermionic - exchange minus signs in tensor @xmath145 . corresponds to a specific choice of fermionic order , which appears in the definition of the local basis of eq . and jordan - wigner transformation of eq . . ( b ) another possible fermionic peps , associated to another fermionic order . the two tensor networks are not equivalent in that they can not be mapped into each other by jump moves alone ( where one is not allowed to drag a line over the end of another open line ) , but the mapping is possible if we allow some additional fermionic exchanges to modify the order of the open lines . importantly , it can be seen that the tensors @xmath31}\}$ ] in both peps appear in the same way in any expectation value . in particular , they would be optimized using exactly the same figure of merits . this shows that the tensors @xmath31}\}$ ] are independent of the choice of fermionic order.,width=264 ] \(ii ) _ local operators_. a local fermionic operator @xmath15 is characterized by a tensor of components that describes the action of @xmath15 on a given basis of states . for instance , in the simple case where each site has a vector space @xmath5 of dimension @xmath19 , we expand a two - site operator @xmath15 as @xmath146 where @xmath147 and @xmath148 and @xmath149 are the annihilation operators acting on the two sites ( see appendix [ app : genoperators ] for the @xmath150 case ) . the coefficients @xmath151 are given by @xmath152 for instance , a hopping term @xmath153 reads @xmath154 since the only non - vanishing coefficient is @xmath155 which is obtained using the anticommutation relations . when the two - site operator @xmath15 acts on two sites @xmath156 , the state @xmath12 of the system is modified into some other state @xmath157 . this is implemented simply by connecting the tensor ( or tensor network ) that represents @xmath158 with the four - index tensor @xmath159 as indicated in fig . [ fig : operator](a ) . in particular , when the two sites are not contiguous in the fermionic order , @xmath160 , a number of line crossings appear . this is reflected in the computation of the expectation value @xmath14 , see fig . [ fig : operator](b)-(d ) . is described by a tensor @xmath161 of coefficients obtained by connecting tensors @xmath25 and @xmath70 . the example corresponds to a @xmath162 lattice and a choice of fermionic order such that the two sites on which @xmath15 acts , which are nearest neighbors in @xmath1 , are not contiguous in the fermionic order ( they occupy positions @xmath163 and @xmath164 ) . the line crossings ( fermionic swap gates ) that appear in connecting tensors @xmath25 and @xmath70 can be interpreted as changes in the fermionic order needed in order to bring the two sites on which @xmath15 acts together , then bringing them back to their original position . ( b ) expectation value @xmath73 . ( c ) the same expectation value in terms of a fermionic peps . ( d ) after some manipulation , an approximation of the expectation value is expressed in terms of an approximate two - site environment @xmath82}$ ] and tensors @xmath71}$ ] , @xmath71\dagger}$ ] , @xmath72}$ ] , @xmath72\dagger}$ ] and @xmath70 , as it was done for a bosonic system in fig . [ fig : bosonenv ] . the only difference here is the presence of a few fermionic swap gates.,width=302 ] \(iii ) _ parity changing tensors_. parity preserving tensors allow us to represent both states @xmath12 with even fermionic particle number ( i.e. parity @xmath165 ) and local operators @xmath15 that are parity preserving . but they also allow us to represent states with an odd fermionic particle number ( parity @xmath166 ) and parity changing operators . this is so because a parity changing tensor @xmath167 , @xmath168 can be represented as a parity preserving tensor @xmath56 , @xmath169 where the additional index @xmath170 only takes one value , @xmath171 . for instance , in order to represent a state @xmath102 with an odd number of particles by means of a peps , an additional index @xmath172 is attached to one of the peps tensors , see fig . [ fig : pchanging](a ) . similarly , a fermionic annihilation operator @xmath173 , which changes parity , can be represented as a tensor with three indices , one of which is fixed to @xmath171 , see fig . [ fig : pchanging](b ) . it is sometimes computationally convenient ( e.g. in the computation of two - point correlators , fig . [ fig : correlator ] ) to represent a parity preserving operator that is the product of two parity changing operators , such as the hopping operator @xmath174 in eq . , by two parity preserving tensors connected by a fixed index @xmath172 , see fig . [ fig : pchanging](b ) . , of a @xmath175 lattice , with odd parity @xmath166 . all peps tensors are parity preserving . one of the tensors has an additional index @xmath176 with fixed parity @xmath166 . ( b ) the tensor @xmath70 of coefficients @xmath177 for a two - site operator such as a hopping term @xmath178 in eq . , that is the product of two parity changing operators , @xmath179 and @xmath149 , can be decomposed as the product of two parity - preserving tensors @xmath180 and @xmath181 connected by an index @xmath176 with fixed parity @xmath166 . this decomposition is useful e.g. in the computation of a correlator involving two distant sites , see fig . [ fig : correlator ] . notice that parity preserving tensor @xmath180 can also be used independently of @xmath181 in order to represent the parity changing operator @xmath148 . , width=302 ] \(iv ) _ simplification_. in the particular case of a pair of crossing lines @xmath137 and @xmath170 where index @xmath170 only takes one possible value @xmath111 of the parity , the fermionic swap gate @xmath133 reduces to a product of two gates . namely , to a product of two identity operators @xmath182 if @xmath183 , and to a product of the parity @xmath112 on line @xmath137 and identity @xmath184 on line @xmath170 if @xmath166 . it follows , for instance , that the jump move applied to a line @xmath137 and a parity preserving tensor @xmath56 with an index @xmath185 ( used e.g. to represent a parity changing tensor @xmath167 in eqs . -[eq : tt ] ) allows us to ignore index @xmath170 in @xmath56 at the price of applying the parity operator @xmath112 to line @xmath137 , see fig . [ fig : jumpmove](b ) . this simplification appears to be useful e.g. in the calculation of the expectation value @xmath186 , see fig . [ fig : correlator ] . a fermionic tensor network can be manipulated simply by performing a sequence of tensor multiplications . it turns out , however , that by considering a special property ( jump move ) of fermionic tensor networks obeying rules 1 and 2 , the fermionic swap gates can be treated in a very special and advantageous way : they can be ignored until they correspond to a crossing of two indices connected to the same tensor , in which case they can be absorbed into the tensor using a low cost operation . as a result , in a fermionic tensor network algorithm we can follow the same sequence of tensor multiplications that we would have employed to manipulate its bosonic counterpart . in particular , we can use the same optimal sequence of contractions , with the same computational cost ( to leading order ) . let us discuss all these aspects in more detail . is represented with a parity preserving tensor @xmath56 with an extra index @xmath172 that remains open . in this case the jump rule introduces an additional fermionic swap gate @xmath129 , involving @xmath170 and the jumping line , that reduces to a parity operator @xmath187 on the jumping line ( see text ) . ( c ) a line can also be dragged over a set of open lines provided that the latter have even overall parity ( in other words , provided they could be connected to a parity preserving tensor),width=302 ] \(i ) _ jump move_. it follows from rules 1 and 2 that in a fermionic tensor network , as in the bosonic case , lines can be dragged over tensors , see fig . [ fig : jumpmove](a ) . this invariance under ` jump ' moves allows us to modify fermionic tensor networks in such a way that the fermionic swap gates do not increase the leading cost of manipulations . for a proof of this property , which exploits the fact that tensors are parity - preserving , we refer again to ref . . notice that the jump move does not include dragging a line over the open end of another line ( this would amount to a change in the local fermionic order ) . the latter transformation is only allowed when it involves a set of open lines with even overall parity , as explained next . \(ii ) _ open lines_. by construction , line ends only appear in the diagrams grouped in such a way that they could be connected to a parity preserving tensor ( that is , their combined parity is always even ) without introducing additional crossings to the network . for instance , the set of all open legs of the peps in figs . [ fig : fermipeps](a ) or [ fig : pchanging](a ) have even overall parity , since the peps as a whole has parity @xmath183 . another example is given by the open legs of an environment , fig . [ fig : corner](c ) . since such groups of open legs with even overall parity could be connected to a parity preserving tensor , the jump move extends naturally to them , as illustrated in fig . [ fig : jumpmove](c ) . \(iii ) _ absorption of a fermionic swap gate_. given a tensor @xmath56 , with coefficients @xmath188 , such that the two contiguous indices @xmath189 are connected to a fermionic swap gate , it is possible to absorb the latter into the former and produce a new tensor @xmath167 with coefficients @xmath190 the cost of the absorption is thus just proportional to the number of coefficients in @xmath56 and comparable to permuting two adjacent indices in a bosonic tensor network . this cost is , in particular , smaller than the tensor multiplication that may have produced @xmath56 or a subsequent tensor multiplication involving @xmath167 . therefore the absorption of fermionic swap gates only makes a sub - leading contribution to the total cost of contracting a fermionic tensor network ( in the same way that the permutation of indices only makes a sub - leading contribution to the total cost of contracting a bosonic tensor network ) . and @xmath191 connected by two lines . ( b ) by means of a jump move , all lines standing between @xmath192 and @xmath191 can be dragged away . ( c ) any fermionic swap gate involving the lines that connect @xmath192 and @xmath191 can be absorbed into e.g. tensor @xmath192 , which is mapped into some @xmath193 . ( d ) finally tensors @xmath194 and @xmath191 are multiplied together , producing tensor @xmath56 . the cost of all these manipulations is dominated by the last step.,width=302 ] \(iv ) _ same leading cost_. in a bosonic system , the contraction of a tensor network is implemented by a sequence of multiplications of tensors , where each multiplication involves two tensors and the sequence is chosen carefully to minimize computational costs . in the fermionic case , it is natural to ask whether the presence of fermionic swap gates will modify the optimal sequence of tensor multiplications and increase the computational cost of the contraction . it turns out that one can always follow the same sequence of multiplications as in the bosonic case . indeed , as illustrated in the example of fig . [ fig : contractjump ] , in order to multiply two tensors , one can use jump moves and fermionic swap gate absorptions to eliminate any fermionic swap gate between the two tensors , which can then be multiplied together exactly as in the bosonic case . the leading cost of these manipulations is given by tensor multiplications , and it is therefore the same for bosonic and fermionic systems . we have explained how to build a fermionic peps for the state @xmath16 of a fermionic lattice system . the next step is to consider algorithms to compute the expectation value @xmath195 of a local operator @xmath15 from a fermionic peps , as well as to optimize the coefficients in its tensors @xmath31}\}$ ] . it turns out that these algorithms can be obtained by just introducing simple modifications to existing bosonic peps algorithms . here we discuss these modifications broadly . let us consider first the computation of the scalar product @xmath54 . as we did in the bosonic case , see fig . [ fig : bosonscalar ] , we build this scalar product by connecting the open indices of a peps for @xmath12 with those for a peps for @xmath57 obtained by hermitian conjugation . the resulting tensor network may contain , in the fermionic case , a large number of fermionic swap gates . however , an important point is that , by using the jump move of fig . [ fig : jumpmove](a ) , the scalar product @xmath54 can again be re - expressed in terms of a tensor network @xmath52 made of reduced tensors @xmath196}\}$ ] , see fig . [ fig : fermipeps](d ) . as shown in fig . [ fig : fermipeps](e ) , the fermionic reduced tensor @xmath53}$ ] differs from its bosonic counterpart by the presence of four fermionic swap gates . importantly , in a finite or infinite lattice with open boundary conditions , _ all _ fermionic swap gates present in the scalar @xmath54 are absorbed into the reduced tensor @xmath53}$ ] . thus , the tensor network @xmath52 contains no fermionic swap gates . as a result , an approximation to the scalar product @xmath54 can be obtained by contracting @xmath52 with _ exactly _ the same approximate contraction techniques ( namely mps techniques for finite peps@xcite and either infinite mps or ctm techniques for infinite peps@xcite ) as in the bosonic case . [ in some implementations , it is possible to lower computational costs by considering the components @xmath33}$ ] and @xmath33\dagger}$ ] of @xmath53}$ ] separately . in this case the approximate contraction techniques have to be slightly modified so as to account for the four fermionic swap gates involved in the definition of @xmath53}$ ] ] . figure [ fig : cylinder ] considers a peps for a fermionic lattice system on a cylinder . its scalar product can again be expressed in terms of a tensor network @xmath52 , made of reduced tensors @xmath196}\}$ ] , that contains no line crossings . therefore its contraction can also be performed exactly as in the bosonic case . instead , in a fermionic lattice system on a torus , the analogous tensor network @xmath52 contains a number of fermionic swap gates , see fig . [ fig : torus ] . however , it can be seen that even in this case @xmath52 can still be coarse - grained using the same terg techniques of ref . for the bosonic case , by properly coarse - graining the swap fermionic gates at each step of the coarse - graining . of fermionic @xmath30 lattice with cylindrical boundary conditions . ( b ) tensor network for the scalar product @xmath54 . ( c ) tensor network @xmath52 of reduced tensors @xmath196}\}$ ] . as in a system with open boundary conditions , @xmath52 has no line crossings and can be contracted as in the bosonic case . , width=321 ] of fermionic @xmath30 lattice with toroidal boundary conditions . ( b ) tensor network for the scalar product @xmath54 . ( c ) tensor network @xmath52 of reduced tensors @xmath196}\}$ ] . on the torus , it is not possible to eliminate all line crossings in @xmath52 , which must contain fermionic swap gates and therefore differs from the bosonic case . however , terg techniques@xcite can still be applied by noting that the coarse - graining of the reduced tensors @xmath196}\}$ ] can take place independently of the fermionic swap gates , which are also coarse - grained in an obvious way . the ( coarse - grained ) fermionic swap gates only need to be absorbed into the rest of the tensor network at the latest coarse - graining step , when the torus has been reduced to only a few lattice sites . , width=321 ] as discussed in sec . [ sec : peps : optim ] , given an environment @xmath197}$ ] , an approximation @xmath198}$ ] can be used both for the computation of the expectation value @xmath199 of a local operator @xmath15 and for the optimization of the tensors @xmath200}\}$ ] defining the peps . the approximate contraction of @xmath197}$ ] leading to @xmath198}$ ] is very similar to the contraction of @xmath52 for the scalar product @xmath54 , and can again be accomplished in the fermionic case using the same techniques than in the bosonic case . from @xmath198}$ ] , obtaining an approximation to the expectation value @xmath73 involves contracting the tensor network of fig . [ fig : operator](d ) , which differs from its bosonic counterpart in the presence of 12 fermionic swap gates . an analogous tensor network is central to the update of the peps tensors @xmath31}\}$ ] during an imaginary - time evolution towards the ground state @xmath81 of a nearest neighbor fermionic hamiltonian @xmath13 , cf . more details of this update will be provided in sec . [ sec : ctm ] in the context of an infinite lattice system , where also the computation of two - point correlators between distant sites will be addressed . to summarize , the differences between bosonic and fermionic peps algorithms are in practice reduced to : ( i ) use of parity preserving tensors @xmath31}\}$ ] to efficiently encode the state @xmath9 ; and ( ii ) presence of fermionic swap gates in some of the tensor networks that need to be contracted , e.g. in order to compute the expected value of a local observable or optimize the tensors @xmath31}\}$ ] of the variational ansatz . however , thanks to the jump move , the optimal sequence of tensor multiplications involved in the contraction of a given fermionic tensor network is the same than in the bosonic case , as is the incurred computational cost . all in all , we see that fermionic peps algorithms can be obtained by introducing a few modifications to bosonic peps algorithms . this is further illustrated in the next section , where we provide more details on the specific peps algorithm used in sec . [ sec : results ] , namely the ipeps algorithm.@xcite in sec . [ sec : results ] the formulation of fermionic peps presented in this paper is tested by computing the ground state of a number of models on an infinite lattice . in this section we provide additional details on the fermionic ipeps ansatz and algorithm used for those computations . we start by reviewing the bosonic ipeps algorithm.@xcite then we describe the modifications required in order to address a fermionic system . } $ ] for two contiguous sites @xmath65 , in terms of the reduced tensors @xmath205 and @xmath206 corresponding to the peps tensors @xmath202 and @xmath203 . ( b ) approximate environment @xmath82}$ ] expressed in terms of 10 tensors , corresponding to the 10 shaded regions in ( a ) , as used in the ctm approach@xcite . some of the tensor are used in fig . [ fig : correlator](b ) . ( c ) approximate environment @xmath82}$ ] written in terms of 6 tensors @xmath69 , as used in the computation of the expectation value of a two - site observable @xmath15 ( fig . [ fig : correlator](a ) ) and the update of tensors @xmath202 and @xmath203 during an imaginary - time evolution ( fig . [ fig : revupdate3]).,width=340 ] let us first consider how to compute the expectation value of a two - site operator @xmath15 , see eq . . this requires computing both @xmath54 and @xmath73 . the scalar product @xmath54 is expressed in terms of an infinite 2d tensor network @xmath52 of reduced tensors @xmath205 and @xmath206 distributed according to a checkerboard pattern . as in a finite system , an environment @xmath207}$ ] for two nearest sites @xmath76 is then built from @xmath52 by removing two reduced tensors , see fig . [ fig : corner](a)-(b ) . contracting @xmath207}$ ] requires an approximation scheme , which produces an approximate environment @xmath82}$ ] consisting of six tensors @xmath69 connected through bond indices that take @xmath208 different values , see fig . [ fig : corner](d ) . here @xmath208 quantifies the degree of approximation in @xmath82}$ ] . one possible approximation scheme consists in using infinite mps techniques , as was discussed in the original ipeps algorithm in ref . . another possible approximation scheme consists in using ctm techniques,@xcite as discussed in the context of the ipeps algorithm in ref . . an approximation @xmath209 to the expectation value @xmath195 of a two - site observable @xmath15 is then obtained by computing @xmath210 where @xmath184 denotes the identity operator on the space @xmath211 of two sites of @xmath1 and we use the same approximate environment @xmath82}$ ] to compute @xmath212 and @xmath213 , where the latter is computed as an expectation value for @xmath214 . the approximate value @xmath215 will in general differ from the exact value @xmath195 . one expects , however , that in the limit of a large @xmath208 one recovers the exact value , @xmath216 in practice , we compute @xmath215 for increasingly large values of @xmath208 , e.g. @xmath217 , until the expectation value @xmath215 does no longer depend substantially in @xmath208 , and assume that this corresponds to @xmath195 . the cost of computing @xmath215 using ctm techniques scales with the peps bond dimension @xmath43 and the environment bond dimension @xmath208 as @xmath218 . notice that for any value of the bond dimension @xmath43 , a peps produces a variational energy @xmath219 , @xmath220 that is @xmath221 , where @xmath222 is the exact ground state energy . therefore , if we could compute @xmath219 exactly , we would obtain an upper bound to @xmath222 , which could e.g. be compared with another upper bound corresponding to another variational ansatz . however , the approximate value @xmath223 is not guaranteed to be an upper bound to @xmath222 . in this paper we assume that @xmath223 is an upper bound to @xmath222 once it has converged for large values of @xmath208 . this assumption is seen to be correct in the case of free fermions in sec . [ sec : results ] . in the ipeps algorithm of refs . an approximation to the ground state is obtained by simulating an evolution in imaginary time , eq . . the optimization of tensors @xmath202 and @xmath203 during the imaginary - time evolution is achieved by breaking the evolution into steps ( suzuki - trotter decomposition ) and then considering updates that involve a single link . there are four types of updates , corresponding to the four inequivalent links given by the bond indices @xmath41 of tensor @xmath202 . for each link , new tensors @xmath224 and @xmath225 are produced as a result of an optimization that aims to account for the action of a two - site gate @xmath226 } \delta t)$ ] acting on that link . we refer to refs . for a detailed explanation on how to use the approximate environment @xmath82}$ ] to obtain updated tensors @xmath224 and @xmath225 . the algorithm proceeds by iteratively updating tensors @xmath202 and @xmath203 until they converge to some pair of tensors that depend on @xmath208 and the time step @xmath227 . a finite time step @xmath227 introduces errors in the imaginary - time evolution . a better approximation to the ground state is obtained by gradually reducing @xmath227 until its value does no longer affect significantly quantities of interest ( e.g. the energy ) . for instance , this occurs for @xmath228 in the simulations of sec . [ sec : results ] . the cost of simulating an imaginary - time evolution is proportional to the number of time steps that are being simulated and scales as @xmath229 , since each update requires reconverging an approximate two - site environment @xmath68}$ ] . a simplified way of simulating an evolution in imaginary time was proposed in ref . . this simplified update does not involve the environment @xmath68}$ ] and has a much lower cost per iteration , namely @xmath230 when applied in a straightforward way and reducible to @xmath231 after more careful considerations . on the one hand , ignoring the environment ( that is , the rest of the wavefunction ) implies that the update may not be optimal , and indeed there are cases , such as the 2d quantum ising model near its critical point , where it produces a less accurate approximation to the ground state,@xcite although we also found many cases where it only performs marginally worse than when using the environment . on the other hand , its much lower cost accelerates the simulations considerably . notice , however , that the computation of expectation values with the resulting peps still requires computing @xmath68}$ ] , which has cost @xmath229 . defined in terms of peps tensors @xmath202 and @xmath232 and four fermionic swap gates . ( b ) reduced tensor @xmath206 defined in terms of peps tensors @xmath203 and @xmath233 and four fermionic swap gates.,width=340 ] to the expectation value @xmath14 of a local observable @xmath15 acting on two contiguous sites . this tensor network only differs from the one in fig . [ fig : bosonenv](e ) for a bosonic system in the presence of 12 fermionic swap gates . ( b ) computation of two - point correlators . for concreteness , we consider the expected value @xmath234 discussed in eq . . the figure shows the tensor network representing @xmath235 , where sites @xmath189 are in the same row of @xmath1 but separated by 4 sites . notice the line connecting the two tensors corresponding to operators @xmath236 and @xmath237 , which crosses a number of other lines introducing a number of fermionic swap gates . since this line corresponds to an index @xmath172 with well - defined parity @xmath166 , the fermionic swap gates simplify into @xmath238.,width=340 ] as in the bosonic case , the fermionic ipeps ansatz exploits translation invariance of a system on an infinite lattice @xmath1 to store the state @xmath12 using only a small number of peps tensors , which are repeated throughout the lattice . whether we are interested in approximating a ground state invariant under translations by one site or with checkerboard order , we consider again just two ( parity preserving ) peps tensors @xmath202 and @xmath203 as in eq . . it may not seem obvious that a fermionic ipeps , characterized by tensors @xmath202 and @xmath203 , represents a state @xmath12 of the infinite lattice @xmath1 that is invariant under diagonal shifts ( checkerboard order ) . indeed , the presence of the ubiquitous fermionic swap gates , which are not homogeneously distributed across the tensor network corresponding to the ipeps ( see fig . [ fig : fermipeps ] for an illustration in the case of a finite peps ) may seem incompatible with the checkerboard order . however , the checkerboard order becomes manifest during the computation of expectation values for local observables . given a fermionic ipeps , the computation of the expectation value of e.g. a two - site observable @xmath15 is very similar to the bosonic case . since there are no fermionic swap gates in the tensor network @xmath52 for the scalar product @xmath54 or the two site environment @xmath64}$ ] , these tensor networks can be contracted exactly in the same way as in a bosonic ipeps . [ if one wants to use the decomposition of the reduced tensors @xmath205 and @xmath206 in terms of @xmath202 and @xmath203 , fig . [ fig : reducedab ] , the presence of four fermionic swap gates needs to be taken into account ] . in the simulations of sec . [ sec : results ] , we have used the directional ctm approach discussed in ref . in order to produce an approximate environment @xmath68}$ ] . then an approximation @xmath209 to the expected value @xmath195 is computed using the tensor network of fig . [ fig : correlator](a ) . figure [ fig : correlator](b ) describes the tensor network that needs to be contracted in order to compute a two - point correlators . the simulation of time evolution proceeds in a very similar way as in the bosonic ipeps algorithm , with the difference that the tensor networks involved contain fermionic swap gates instead of simple line crossings . a detailed description of the two updates used in this paper is presented in appendix . [ app : twosite ] . before we move to presenting the results of ground state computations , we conclude this section with a summary of the fermionic ipeps algorithm : \(i ) _ ansatz . _ the state @xmath12 of the system on an infinite square lattice @xmath1 is encoded in two parity symmetric tensors @xmath202 and @xmath203 . these tensors depend on @xmath239 parameters , where @xmath6 is the dimension of the vector space @xmath5 of one site of @xmath1 and @xmath43 is the bond dimensions of the ipeps . \(ii ) _ computation of expectation values_. given tensors @xmath202 and @xmath203 , the reduced tensors @xmath205 and @xmath206 are computed according to fig . [ fig : reducedab ] . from the reduced tensors @xmath205 and @xmath206 , an approximation @xmath68}$ ] to a two - site environment @xmath64}$ ] is obtained using the ctm algorithm , see fig . [ fig : corner ] . the computational cost scales as @xmath229 , where @xmath208 is the bond dimension of the approximate environment . then expectation values are computed by contracting small tensor networks involving @xmath68}$ ] . for instance , an approximation @xmath215 to the expected value @xmath195 , eq . , for an operator @xmath15 on two nearest neighbor sites is obtained according to fig . [ fig : correlator](a ) , whereas an approximation to a two - point correlator is obtained according to fig . [ fig : correlator](b ) . \(iii ) _ approximation of the ground state_. starting from e.g. random tensors @xmath202 and @xmath203 , an imaginary - time evolution is used to find an approximation to the ground state of a local hamiltonian @xmath13 . the two updates of appendix [ app : twosite ] can be used : the _ standard _ update , which in general produces better ground state approximations and has cost @xmath229 ; and the _ simplified _ update , which has a significantly lower cost and is often only marginally less accurate than the standard update . in this section we test the fermionic ipeps algorithm , as summarized at the end of the last section , for several models of free and interacting fermions in an infinite square lattice @xmath1 . we start by considering an exactly solvable model of free spinless fermions , which allows us to compare the numerical results with the exact solution , and therefore assess the accuracy of the approach . it turns out that , similarly as in the 2d mera , @xcite the accuracy of the numerical results depends on the amount of entanglement in the system . we also compare the use of the standard and simplified updates discussed in appendix [ app : twosite ] . the second model , describing interacting spinless fermions on a square lattice , is no longer exactly solvable . we compare our results for the phase diagram with the hartree - fock ( hf ) solution from ref . , and show that our results for large @xmath240 are an improvement upon the hf result . the final example is the @xmath0 model , where we compare energies obtained with ipeps with previous variational monte carlo results based on gutzwiller - projected ansatz wave functions . in all these examples we present ipeps results for different bond dimensions @xmath43 , and study the convergence of the energy as a function of @xmath208 , the bond dimension of the environment . in all these examples we chose @xmath241 , see eq . . the first model under consideration is an exactly solvable model of free spinless fermions,@xcite given by the hamiltonian @xmath242 - 2\lambda \sum_i { \hat c}_{i}^{\dagger } { \hat c}_i , \label{eq : free}\ ] ] with @xmath243 denoting the sum over nearest - neighbor pairs , @xmath244 the chemical potential , and @xmath245 the pairing potential . figure [ fig : ff_pd ] shows the exact phase diagram of the model.@xcite for @xmath246 the model reduces to the usual tight - binding model of free fermions , with a metal phase for @xmath247 and a band insulator for @xmath248 . also the line @xmath249 , @xmath250 corresponds to a metal phase with a one dimensional fermi surface . for @xmath250 , @xmath251 the system is superconducting , with a critical phase for @xmath252 and a gapped phase for @xmath253 . as a function of chemical potential @xmath244 and pairing potential @xmath245 . for @xmath250 , @xmath251 the system is superconducting.,width=264 ] , for different values of @xmath245 and @xmath43 . the dimension @xmath208 is 20 and 40 for @xmath254 and @xmath255 , respectively . , width=302 ] fig . [ fig : e_ff ] shows the relative error in the ground state energy obtained by simulating an imaginary - time evolution with the standard update ( cf . appendix [ app : twosite ] ) . results for bond dimensions @xmath254 and @xmath255 and at different locations in the phase diagram are shown . in the gapped phase , @xmath253 , accurate results are already obtained for @xmath254 , and for @xmath255 these accuracies are increased by one to three orders of magnitude . [ a special case is the band insulator for @xmath246 , @xmath256 , ( not shown ) which corresponds to an unentangled or product ground state and can be reproduced exactly even with bond dimension @xmath257 . ] the accuracy in the critical phase , @xmath258 , is of the order of @xmath259 for @xmath254 . increasing the bond dimension to @xmath255 , an order of magnitude is gained for @xmath260 , but the gain is smaller for @xmath261 . finally , the free fermion regime with a 1d fermi surface ( @xmath246 , @xmath262 ) is the most challenging case . note that in this case , the entanglement entropy exhibits a logarithmic correction to the area law.@xcite this shows that the accuracy of the results depends on the amount of entanglement in the system , similarly to the findings with the 2d mera . @xcite as a function of distance r ( in x - direction ) , see eq . , for two different values of @xmath263 of the free - fermion model . middle panels : absolute value of the correlation function @xmath264 in semi - logarithmic scale . lower panels : the difference between the simulation result @xmath265 and the exact result @xmath266 for different values of @xmath43 . notice that better accuracies are obtained in the gapped phase , which corresponds to a less entangled ground state.,width=302 ] next we study the accuracy obtained for the two - point correlator @xmath267 where @xmath268 , i.e. site @xmath170 is in the same row as site @xmath170 but separated by @xmath269 columns , see fig . [ fig : correlator](b ) . the ipeps results shown in fig . [ fig : ff_corrs ] are seen to approach the exact values with increasing @xmath43 . in the critical phase ( left panels ) only the correlations at short distances are reproduced accurately . the precision of the correlator at longer distance is rather poor for small @xmath43 , but improves upon increasing @xmath43 . in the gapped phase ( right panels ) the accuracy is clearly better , already for small bond dimension . for @xmath240 accurate results are obtained up to distance @xmath270 where the magnitude of the correlator is of the order @xmath271 . these results are obtained from simulations with the simplified update , and we checked that they are similar to the ones obtained with the standard update ( for @xmath254 and @xmath255 ) . we compare and discuss the two updates further below . , the bond dimension of the environment , for @xmath255 . in these examples the energy decreases monotonously with @xmath208 , i.e. the energy from the simulation is always larger than the true ground state energy.,width=302 ] since the ipeps is a variational ansatz , the ground state energy @xmath219 of a state represented by an ipeps is an upper bound of the exact ground state energy @xmath222 . with increasing bond dimension @xmath43 the true ground state can be represented more accurately , and consequently the energy @xmath219 becomes lower , i.e. a better upper bound of @xmath222 . however , as discussed in the previous section , quantities such as the energy can only be extracted in an approximate way from the ipeps . the error of this approximation depends on the bond dimension of the environment @xmath208 . in fig . [ fig : chidep ] the dependence of the relative error of the energy as a function of @xmath208 is plotted for an ipeps with bond dimension @xmath255 . we observe that , with increasing @xmath208 , the energy @xmath223 converges to some value that is indeed an upper bound for @xmath222 . this is consistent with the assumption that @xmath223 has converged to the true energy @xmath219 of our @xmath255 ipeps . in the other models analyzed in this section we will also assume that once @xmath223 does no longer change with increasing @xmath208 , it has attained @xmath219 and therefore is an upper bound to @xmath222 . [ fig : chidep ] shows that , in the present free model , the approximate energy @xmath223 monotonically decreases with increasing @xmath208 , and therefore any value of @xmath223 is already an upper bound to @xmath222 . however , this behavior is not true in general , as we will see further below . it is therefore important to study the convergence in @xmath208 in each case separately . , and @xmath272 ( @xmath273 for @xmath274 ) . the environment in the gapped phase converges considerably faster than in the critical phase.,width=302 ] let us make a few comments about the convergence of the environment in the ipeps algorithm , for fixed sets of tensors @xmath202 and @xmath203 which approximate the ground state of different phases . figure [ fig : envdep ] illustrates how the energy converges with the number of renormalization steps in the directional ctm algorithm @xcite for different points in the phase diagram . the environment in the gapped phase clearly converges faster than in the critical phase . here , the tensors at the boundary of the ipeps have been chosen randomly at the beginning . depending on the initial conditions of these boundary tensors the number of steps needed to converge can vary significantly . in practice , when using the standard update , we use the environment from the previous imaginary time step as an initial condition , so that only a few renormalization steps are needed to re - converge the environment . note also that for different initial boundary tensors the energy might converge to slightly different values ( especially in the critical phase ) . it is therefore advisable to check results for different initial boundary tensors . for this model , we have also compared the precision of the ground state energy obtained with the standard and simplified updates , see fig . [ fig : ff_compare ] . the accuracies obtained with the standard update are typically slightly better than the ones obtained with simplified updates . however , the simulations with the simplified update are computationally considerably cheaper , because the environment has to be computed only once at the end of the simulation for the evaluation of observables , whereas in the standard update the environment it has to be recomputed at each step of the imaginary - time evolution . the leading cost of the two methods is the same , but the computational cost differs by a rather large factor , which depends on the total number of imaginary - time steps . for the remaining examples of this paper we will consider only simplified updates this allowed us to perform simulations with @xmath240 and @xmath275 on a standard computer in roughly one day . we checked for @xmath254 and @xmath255 that the results obtained with the two updates give similar accuracies . it is conceivable , however , that the difference in accuracy between the two update schemes become larger with increasing @xmath43 , depending on the model under consideration . .,width=302 ] the second model under consideration is a model of interacting spinless fermions , which is not exactly solvable . it is defined by the hamiltonian @xmath276 - \mu \sum_i { \hat c}_{i}^{\dagger } { \hat c}_i + v \sum_{\langle ij \rangle } { \hat c}_i^{\dagger}{\hat c}_i { \hat c}_j^{\dagger}{\hat c}_j , \label{eq : hint}\ ] ] with @xmath277 the nearest - neighbor hopping amplitude , the ( repulsive ) nearest - neighbor interaction strength , and @xmath278 the chemical potential . the hartree - fock ( hf ) phase diagram @xcite as a function of @xmath279 and particle density @xmath280 is given fig . [ fig : if_pd ] ( solid line ) . the hf calculation predicts a gapped charge - density - wave ( cdw ) phase at half filling ( @xmath281 ) and a translational invariant normal state ( metal phase ) far away from half filling . in between these two phases they find a thermodynamically unstable region , which we identify as a phase separation ( ps ) region , i.e. where the system splits into two parts , one in the metal phase and the other in the cdw phase . is the interaction strength , and @xmath280 is the particle density ( filling ) . the charge - density wave phase ( checkerboard pattern ) at exactly half - filling ( @xmath281 ) is separated from the metal phase by a first order phase transition , with an intermediate region corresponding to phase separation ( ps ) . with increasing @xmath43 , the boundary between the metal phase and the ps region moves away from the hartree - fock ( hf ) result from ref . . dotted and dashed lines are a guide to the eye . the uncertainty of the phase boundaries obtained with ipeps is smaller than the symbol size ( in x - direction ) . , width=302 ] for @xmath282 , obtained by ipeps and hartree - fock ( hf).@xcite the first order phase transition between the metal phase and the charge - density wave ( cdw ) phase occurs at a value @xmath283 where the two corresponding energies cross . lower panel : particle density @xmath280 as a function of chemical potential . at the first order phase transition point @xmath283 , @xmath280 jumps from a certain value @xmath284 in the metal phase to @xmath281 in the cdw phase . for densities in between @xmath284 and @xmath281 the system exhibits phase separation . the numbers in brackets indicate the uncertainty in the last digit . , width=302 ] our results for the phase diagram , obtained with the simplified update using an ipeps with bond dimension @xmath255 and @xmath240 , are given by the squares and triangles in fig . [ fig : if_pd ] , respectively . the phase diagram obtained with ipeps qualitatively agrees with the hf solution . however , with increasing @xmath43 the phase boundary to ps region moves away from the hf result . in the following we explain how we computed the phase boundary and discuss the origin of this deviation . we focus only on the left half of the phase diagram , @xmath285 , since it is mirror symmetric with the @xmath281 line . we determined the phase boundary between the metal phase and the ps region for several values of @xmath279 , for @xmath255 and @xmath240 . for each value of @xmath279 , we studied the first order phase transition , which occurs at a certain value @xmath286 . the ipeps algorithm known to be particularly suitable to study first order phase transitions , thanks to displaying some degree of hysteresis . @xcite specifically , if we start a simulation with an ipeps that represents a state in e.g. the metal phase , it will remain in the metal phase upon increasing @xmath278 , even for values ( slightly ) larger than @xmath283 ( where the ground state is no longer metallic ) . this allows us to compute the energy of the metal phase in the region @xmath287 even though the cdw - state has a lower energy , and vice versa . we can therefore compute the energies of the ground states of the two phases individually , and the phase transition occurs where the two energies cross , as shown in the upper panel of fig . [ fig : if_energy ] for @xmath282 . at the transition point @xmath283 , the density @xmath280 jumps from a certain value @xmath288 in the metal phase to the value @xmath281 in the cdw phase , as plotted in the lower panel of fig . [ fig : if_energy ] . for densities with values between @xmath284 and @xmath281 there exists no homogenous ground state , i.e. the system exhibits phase separation ( cf . [ fig : if_pd ] ) . let us now compare the energies obtained with ipeps with the hf solution in fig . [ fig : if_pd ] . while the results in the cdw phase coincide ( up to @xmath289 ) , we obtain lower energies for @xmath240 than the hf result ( of the order of @xmath290 ) . this leads to a shift of the transition point @xmath283 to a larger value of @xmath278 , and thus the phase boundary appears at a larger value of @xmath284 than in the hf case . the error of the energy due to the finite @xmath208 is smaller than the symbol size ( cf . [ fig : if_chidep ] ) . within this error , we regard the results as `` variational '' energies , i.e. an upper bound of the true ground state energy . accordingly , our @xmath240 results in the metal phase are closer to the true ground state energy , and the phase boundary for @xmath240 is an improvement upon the hf solution . note that the convergence of the energy with @xmath208 in fig . [ fig : if_chidep ] is not monotonous as in the previous examples in fig . [ fig : chidep ] . still , for large @xmath208 the energies do not seem to change significantly anymore . for some simulations we observe that the environment does not converge to a fixed point , but rather oscillates slightly . in such a case the energy fluctuates around a certain value . we take this error also into account in our study . the final uncertainty of the phase boundary obtained by ipeps is smaller than the symbol sizes ( in x - direction ) in fig . [ fig : if_pd ] . all simulation results for this model were obtained with the simplified update of appendix [ app : twosite ] . is not monotonous . for large @xmath208 the values do not seem to change significantly anymore . we associate an error bar to the values of the energy , depending on the convergence behavior in @xmath208 . this error is smaller than the symbol sizes in fig . [ fig : if_energy].,width=302 ] as a final example we consider the @xmath0 model , @xmath291 with @xmath292 the spin index , @xmath293 the electron density and @xmath294 the spin @xmath295 operator on site @xmath137 , and @xmath296 . the @xmath0 model is an effective model of the hubbard model in the limit of strong on - site repulsion . the local hilbert space of each site contains three basis states @xmath297 , i.e. two electrons with opposite spins can not occupy the same lattice site as in the hubbard model . the @xmath0 model is an important model in the context of high - t@xmath298 superconductivity and its phase diagram is still controversial . we focus here on the parameter @xmath299 which lies in the relevant parameter range of cuprate superconductors . at half filling , i.e. particle density @xmath300 , the model corresponds to the antiferromagnetic spin 1/2 heisenberg model , where the ground state has long - range antiferromagnetic ( nel ) order . far away from half filling the model is a metal . at small , but finite doping @xmath301 , several studies predict a @xmath302-wave superconducting phase , see e.g. refs . and references therein . in studies with dmrg also a phase with stripe order @xcite was found at low doping . some studies suggest that the formation of stripes is due to phase separation of the undoped antiferromagnet and the superconducting phase . @xcite here we focus on the variational monte carlo ( vmc ) results from ref . , based on gutzwiller - projected ansatz wave functions . this study was done for finite lattices of size @xmath303 , where the finite size corrections of the energy are estimated to be of the order of @xmath304 . the monte carlo sampling error is of the same order of magnitude . at low doping ( @xmath305 ) the best variational energies are obtained by an ansatz wave function including superconductivity and antiferromagnetic order . at larger doping @xmath306 a better variational energy is obtained with an ansatz wave function that is superconducting without antiferromagnetic order . in the following we compare our energies with the best values obtained in this previous study . figure [ fig : tj_energy ] shows the energy per site ( with the chemical potential term subtracted ) as a function of particle density . one can see that the ipeps results approach the vmc results from ref . with increasing @xmath43 . for @xmath240 the ipeps energies are roughly @xmath259 higher than the vmc energies , and for @xmath307 the deviation is of the order of @xmath304 , which is the same order of magnitude as the error bar of the vmc results . note also , that some of the energies for @xmath307 are lower than the vmc results . however , the @xmath307 results are not necessarily `` variational '' energies , as we discuss below . of the @xmath0 model as a function of filling @xmath280 , with @xmath299 . with increasing @xmath43 , the energies obtained by ipeps are approaching the values from the variational monte carlo ( vmc ) study in ref . note that the chemical potential term has been subtracted from the energy . the error bars of the vmc results are of the order @xmath308 . lower panel : relative deviation between the energies obtained by ipeps and vmc . full symbols indicate that ipeps energy is lower than the vmc energy.,width=302 ] these preliminary results are encouraging for the future study of the phase diagram of the @xmath0 model using peps algorithms , because the current energies with the largest dimension @xmath307 are compatible with previous variational studies , and we expect to be able to increase @xmath43 by properly exploiting the global symmetries of the model . we also would like to emphasize that ipeps is a general ansatz , i.e. the same ansatz is used for any model on a square lattice , whereas in other variational studies the ansatz wave function is typically based on the specific physics of the model . it is remarkable that ipeps , which starts from a random initial state , yields comparable energies as the ones obtained by specialized ansatz wave functions . finally , it will be interesting to verify whether the phases obtained by ipeps correspond to the ones predicted by the variational study , or whether a phase with different dominant correlations appears , e.g. stripe - ordered phase , as previously found in dmrg studies . @xcite figure [ fig : tj_chi ] shows that the energies as a function of @xmath208 are sufficiently converged , both for @xmath255 and @xmath240 , i.e. the uncertainty due to a finite @xmath208 is much smaller than the symbol sizes in fig.[fig : tj_energy ] . we can thus view the energies for @xmath309 as `` variational '' in the sense that they are an upper bound of the true ground state energy . however , for at finite doping the energy increases with increasing @xmath208 , and it is at present not clear what value it will reach for larger ( and presently unaffordable ) values @xmath208 . thus , in this case the energy is not believed to necessarily be an upper bound to the exact ground state energy . . the energies for @xmath309 do not seem to change significantly anymore for large @xmath208 , whereas for @xmath307 the energy is still increasing at the largest value of @xmath208 used.,width=302 ] in recent months several theoretical proposals have appeared describing fermionic versions of tensor network algorithms for 2d lattice systems , namely fermionic mera@xcite and fermionic peps@xcite . in this paper we have explained how to obtain fermionic peps algorithms by applying the general fermionization procedure of tensor networks introduced in ref . . a highlight of our formulation of fermionic peps is its simplicity : it replaces the complexity involved in dealing with a network of fermionic operators with a tensor network built by following two rules , namely the use of parity preserving tensors and the substitution of line crossings with fermionic swap gates . we have then used the fermionic ipeps algorithm to compute an approximation to the ground state of several fermionic models on an infinite lattice . by simulating an exactly solvable model of free fermions,@xcite we have been able to see that even a fermionic peps with small bond dimension @xmath43 ( @xmath310 ) is already capable of reproducing up to several digits of the exact ground state energy , as well as two - point correlators . the results also showed that , similarly to what had been observed with the fermionic mera,@xcite the accuracy of the approach depends on the amount of entanglement in the system . generally speaking , gapped systems are less entangled than critical ones and , accordingly , a fermionic ipeps with a given bond dimension @xmath43 produces better accuracies for the former . the simulation of model of interacting spinless fermions has provided us with a first clear indication of the usefulness of the fermionic peps as a variational ansatz for the ground state of systems of interacting fermions . the fermionic peps approach reproduces the hartree - fock phase diagram@xcite , with metal and charge - density wave phases ; but even with bond dimension @xmath255 and @xmath240 , it improves the ground state energies on the metallic phase and this results in a significant shift of the phase boundary . finally , results for the @xmath0 model in the relevant parameter range for cuprate superconductors are particularly encouraging , given that fermionic peps , still at an early stage of development , already produce ground state energies comparable with those of previous variational studies.@xcite the main limitation in present calculations is due to the scaling @xmath229 of simulation costs with the peps bond dimension @xmath43 and the environment bond dimension @xmath208 , which restricts @xmath43 and @xmath208 to relatively small values . there are , however , several ways in which larger values of @xmath43 and @xmath208 could become affordable , thereby leading to more accurate results . on the one hand , one can incorporate the internal global symmetries of a model ( e.g. particle number conservation ) into the tensors and exploit their block structure to reduce computational costs.@xcite this strategy is expected to be decisive for the characterization of the phase diagram of the @xmath0 and hubbard models . on the other hand , monte carlo sampling techniques could be used to reduce the formal dependence of the simulation costs in @xmath43 and @xmath208 . @xcite finally , the use parallel computing on a large cluster would also lead to improved fermionic peps simulations . _ acknowledgements.- _ the authors thank j. de woul , e. langmann , and d. ivanov for providing us their data from refs . and . we also acknowledge inspiring discussions with j. jordan , r. pfeifer , l. tagliacozzo and h .- q . support from the university of queensland ( ecr2007002059 ) and the australian research council ( ff0668731 , dp0878830 ) is acknowledged . in the case of a lattice where each site is described by a generic vector space @xmath5 of finite dimension @xmath6 , we decompose @xmath5 into even and odd parity sectors , @xmath311 and use an index @xmath106 to label a basis with well defined parity , @xmath312 see sec . [ sec : peps : parity ] . we can also introduce a set @xmath313 of generalized fermionic operators on each site , where @xmath314 is defined as @xmath315 then a local operator @xmath15 acting on just one site can be expanded as @xmath316 notice that @xmath317 is parity preserving if @xmath318 and parity changing if @xmath319 . fermionic operators @xmath320}_{s_1}$ ] and @xmath321}_{s_2}$ ] acting on two different sites @xmath322 fulfill @xmath323}_{s_1 } { \hat f}^{[\vec{r}_2]}_{s_2 } = s(s_1,s_2 ) { \hat f}^{[\vec{r}_2]}_{s_2 } { \hat f}^{[\vec{r}_1]}_{s_1},\ ] ] where @xmath324 a two - site operator @xmath15 acting on sites @xmath325 can then be written as @xmath326 where @xmath327\dagger } { \hat f}_{s'_2}^{[\vec{r}_2]\dagger } { \mbox{$| 0_10_2 \rangle$ } } { \mbox{$\langle 0_10_2 |$ } } { \hat f}_{s_2}^{[\vec{r}_2 ] } { \hat f}_{s_1}^{[\vec{r}_1]}.\ ] ] and @xmath328.,width=302 ] is obtained from tensor @xmath202 by crossing the legs @xmath36 and @xmath6 . ( b ) tensor @xmath329 is obtained from tensor @xmath328 by crossing the legs @xmath330 and @xmath331 . ( c ) tensors @xmath332 and @xmath333 are obtained by joining the @xmath334 indices and the @xmath335 indices of @xmath336 , and doing a singular value decomposition of the resultant matrix . the singular values are included in the definition of tensor @xmath333 . ( d ) tensors @xmath337 and @xmath338 are obtained by joining indices @xmath339 and indices @xmath340 of @xmath203 together , then performing a singular value decomposition of the resulting matrix , then splitting the composed indices back apart . the singular values are included in tensor @xmath337 . ( e ) tensors @xmath341 and @xmath342 are obtained by joining indices @xmath343 and indices @xmath344 of @xmath345 , then performing a singular value decomposition of the resulting matrix , then splitting the composed indices back apart . the singular values are included in tensor @xmath342 . ( f ) tensors @xmath346 and @xmath347 are obtained by joining indices @xmath348 and indices @xmath349 of @xmath329 , then performing a singular value decomposition of the resulting matrix , then splitting the composed indices back apart . the singular values are included in tensor @xmath346.,width=340 ] is obtained by contracting a tensor network that contains the approximate environment @xmath82 } = \{e_1 , \cdots , e_6\}$ ] , tensors @xmath332 , @xmath341 , @xmath350 and @xmath347 , and a few fermionic swap gates . ( b ) tensor @xmath351 is obtained from the contraction of @xmath352 and the gate @xmath353.,width=321 ] this appendix describes how to update the tensors @xmath202 and @xmath203 that define a fermionic ipeps . we consider the two strategies employed to obtain the results of sec . [ sec : results ] , namely ( i ) the _ standard _ update , used in refs . for bosonic systems , which requires the approximate environment @xmath82 } = \{e_1,\cdots , e_6 \}$ ] of fig . [ fig : corner](d ) ; and ( ii ) a _ simplified _ update , used in ref . , which does not involve an environment . in these two schemes , a gate @xmath354}\delta t)$ ] is applied to the two sites @xmath355 of a given link , with tensors @xmath202 and @xmath203 , and new tensors @xmath224 and @xmath225 are chosen in order to best account for the action of the gate . here we simply list the steps required in order to obtain the updated tensors @xmath224 and @xmath225 . for a justification of the schemes we refer to refs . . we emphasize that the only difference between these updates and the ones used in a bosonic system is the presence of fermionic swap gates . in particular , if the fermionic swap gates are eliminated ( by setting @xmath356 in eqs . -[eq : s ] ) we obtain update algorithms for bosonic peps . in some models , such as the quantum ising model near criticality , @xcite the standard update produces significantly more accurate results than the simplified update , but this did not seem to be the case in the gapless phases studied in this paper . in all models away from criticality that we could test , the simplified update produces only marginally worse accuracies . on the other hand , the much lower computational cost of the simplified update allows to consider larger bond dimension @xmath43 than with the standard update . for concreteness , in the following we assume that the gate @xmath353 is applied on a horizontal link where tensors @xmath202 and @xmath203 are at the left and right , respectively . similar derivations apply to the other three types of links . and @xmath357 are obtained by iteratively solving two linearized equations . specifically , we iterate the following two steps until convergence : ( a ) given @xmath337 and @xmath346 , we obtain @xmath358 by solving the linear equation @xmath359 , where @xmath360 and @xmath361 are obtained as indicated in the diagram . then , we set @xmath362 , and compute @xmath363 . ( b ) given @xmath333 and @xmath342 , we obtain @xmath357 by solving the linear equation @xmath364 , where @xmath360 and @xmath361 are again obtained as indicated in the diagram . then , we set @xmath365 , and compute @xmath366.,width=245 ] given tensors @xmath202 and @xmath203 , an approximate environment @xmath82 } = \{e_1,\cdots , e_6 \}$ ] is obtained as described in sec . [ sec : ctm ] : first build the reduced tensors @xmath205 and @xmath206 , fig . [ fig : reducedab ] , which are the building blocks of the exact environment @xmath207}$ ] ; then use e.g. ctm techniques@xcite to produce an approximation @xmath82}$ ] to @xmath207}$ ] . on the other hand , it is convenient to introduce a number of additional tensors . first compute tensors @xmath345 and @xmath328 according to fig . [ fig : revupdate1 ] , as well as tensors @xmath336 and @xmath329 according to fig . [ fig : revupdate2](a , b ) . then perform a singular value decomposition of tensors @xmath336 , @xmath203 , @xmath345 and @xmath329 as explained in the caption of fig . [ fig : revupdate2](c - f ) . here we use the notation @xmath367 for the singular value decomposition of a matrix @xmath56 , where @xmath368 and @xmath369 are isometries and @xmath370 is the diagonal matrix of singular values . notice that the singular value decompositions are performed by regarding a tensor as a matrix after grouping its indices into two sets . is obtained from the contraction of @xmath332 and @xmath358 . ( b ) the updated tensor @xmath371 is computed by crossing the @xmath36 and @xmath6 indices of @xmath372 . ( c ) the updated tensor @xmath373 is obtained from the contraction of @xmath357 and @xmath350.,width=340 ] tensor @xmath374 is then obtained by contracting the tensor network of fig . [ fig : revupdate3](a ) , which in turn is used to produce tensor @xmath351 in fig . [ fig : revupdate3](b ) . from tensors @xmath374 and @xmath351 , updated tensors @xmath358 and @xmath357 are obtained by iterating until convergence the process explained in the caption of fig . [ fig : revupdate4 ] . for the simplified update from ref . , the structure of the peps tensor network is slightly different to the one considered so far in this paper . here the infinite peps is specified by two tensors @xmath384 and @xmath375 , and four diagonal matrices @xmath376 with non - negative diagonal entries that assign weights to the indices of @xmath384 and @xmath375 , see fig . [ fig : revupdate6](a ) . notice that the usual expression in terms of tensors @xmath202 and @xmath203 can be recovered e.g. by multiplying the square root of the weight matrices @xmath376 with the tensors @xmath384 and @xmath375 , see fig . [ fig : revupdate6](b)-(c ) . the simplified update consists of the following steps : first , a tensor @xmath385 is computed in analogous way as tensor @xmath336 in fig . [ fig : revupdate2](a ) . then tensor @xmath380 is obtained by contracting the network in fig . [ fig : revupdate7](a ) , and subsequently decomposed through a singular value decomposition as shown in fig . [ fig : revupdate7](b ) . this results in tensors @xmath377 and @xmath378 and the matrix of weights @xmath379 , which corresponds to the singular values of @xmath380 . at this stage @xmath379 is truncated into @xmath386 , which keeps only the @xmath43 largest diagonal entries of @xmath379 . then tensors @xmath377 and @xmath378 are also truncated accordingly . next , tensors @xmath387 and @xmath383 are obtained from ( the truncated version of ) tensors @xmath377 and @xmath378 as shown in fig . [ fig : revupdate8](a)-(b ) . finally , tensor @xmath388 is obtained from @xmath387 again in an analogous way as @xmath224 in fig . [ fig : revupdate5](b ) . h. c. jiang , z. y. weng , and t. xiang , phys . lett . 101 , 090603 ( 2008 ) . z. y. xie , h. c. jiang , q. n. chen , z. y. weng , and t. xiang , phys . 103 , 160601 ( 2009 ) . chen , c .- y . lai , and m .- f . yang , j. stat . mech . : theory exp . ( 2009 ) p10001 . g. vidal , phys . 99 , 220405 ( 2007 ) . g. vidal , phys . . lett . 101 , 110501 ( 2008 ) . l. cincio , j. dziarmaga , and m. m. rams phys . 100 , 240603 ( 2008 ) . g. evenbly and g. vidal , phys . 102 , 180406 ( 2009 ) . g. evenbly and g. vidal , phys . rev . b 79 , 144108 ( 2009 ) . v. murg , f. verstraete , and j. i. cirac , phys . rev . b 79 , 195119 ( 2009 ) . g. evenbly and g. vidal , arxiv:0904.3383 . b. bauer , g. vidal , and m. troyer , j. stat . mech . : theory exp . ( 2009 ) p09006 . p. chen , c .- y . lai , and m .- f . yang , arxiv:0910.5081 . h. yokoyama and m. ogata . 65 , 3615 ( 1996 ) . d. a. ivanov . phys . rev . b 70 , 104503 ( 2004 ) . m. lugas , l. spanu , f. becca , and s. sorella . phys . rev . b 74 , 165122 ( 2006 ) . l. spanu , m. lugas , f. becca , and s. sorella . phys . rev . b 77 , 024510 ( 2008 ) . s. r. white and d. j. scalapino . 80 , 1272 ( 1998 ) . s. r. white and d. j. scalapino . 81 , 3227 ( 1998 ) . s. r. white and d. j. scalapino . phys . rev . b 61 , 6320 ( 2000 ) . v. j. emery , s. a. kivelson , and h. q. lin . 64 , 475 ( 1990 ) . c. s. hellberg and e. manousakis , phys . 78 , 4609 ( 1997 ) . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: brown dwarfs are self - gravitating objects with physical and atmospheric properties intermediate between stars and planets . with masses below @xmath10.075 m@xmath2 depending on age and metallicity . see @xcite for an extensive discussion of evolutionary models . ] @xcite , these objects can not sustain hydrogen fusion , and hence cool and dim as they age , radiating primarily at infrared wavelengths . the evolution of their spectra spans the spectral classes m , l , t , and y , with transitions demarcated by the appearance and disappearance of absorption lines and bands as molecules form and condense out of their atmospheres at different temperatures and pressures ( and references therein ) . despite having a basic understanding of their evolution , brown dwarf formation remains an open question . standard jeans collapse of molecular clouds requires high densities so that gravity can overcome thermal pressure . once the collapse has begun , halting the accretion becomes problematic . several mechanisms have been proposed to resolve this issue , including turbulent fragmentation of protostellar clouds @xcite , fragmentation of pre stellar disks @xcite , ejection by dynamical interactions with other protostars @xcite , and photoerosion of prestellar cores . in principle , these formation mechanisms should leave traces on the statistical properties of brown dwarfs , including the occurrence of multiple systems and distributions of their separation , relative masses and eccentricity . observationally , it has been shown that multiplicity increases with primary mass , even at the lower mass end of the main sequence , with the g dwarf binary fraction being 57@xmath3 higher than that for m dwarfs @xcite . current estimates of the binary fraction of very low mass ( vlm ) late - m to t dwarfs ( vlm @xmath4 ) are @xmath5 , with a peak in separation at @xmath6 4 au and a mass ratio distribution peaking at nearly equal masses @xcite . however , these multiplicity statistics have been largely determined from resolved imaging programs , sampling separations greater than 3 au . @xcite pointed out that the current peak in the binary angular separation distribution is coincident with the resolution limit of hst and ground - based adaptive optics ( ao ) imaging , indicating that tight ( @xmath7au ) vlm binaries could be undercounted . likewise , @xcite and @xcite report a higher unresolved binary fraction ( @xmath8 ) based on overluminous binary candidates in color - magnitude plots . conversely , spectroscopic radial velocity ( rv ) studies find binary fractions of 2.5@xmath3 in systems separated by @xmath7 au @xcite and @xmath9 up to 3 au . for the 0 - 6 au range , @xcite estimate a binary fraction of 26@xmath10 . however , the difficulty of obtaining high resolution spectra of faint vlm dwarfs results in small sample size . since total binary fractions for vlm stars and brown dwarfs could range between @xmath11 , it is imperative to constrain this statistic to make conclusions about brown dwarf formation . an alternative method for detecting tight unresolved binaries , developed by @xcite , involves identifying blended light pairs , or spectral binaries . we will refer as _ spectral _ binaries to those objects whose combined - light spectrum shows distinct peculiarities that come from the highly structured spectra of single m , l and t dwarfs when blended together , as opposed to _ spectroscopic _ binaries which are binaries that show rv variations . the first brown dwarf spectral binary , 2mass j05185995@xmath122828372 , was serendipitously identified by @xcite based on its hybrid characteristics containing features of both l and t dwarfs . the superposition of l plus t dwarf spectra proved to be the simplest model of its peculiar spectrum and it was later resolved as a binary using the hubble space telescope @xcite . the unusually blue l dwarf sdss j080531.84 + 481233.0 was next identified as a spectral binary with l4.5 and t5 components by @xcite , based on a peculiar methane absorption band starting at @xmath13 m , and was later confirmed as an astrometric variable by @xcite . a third system , 2mass j03202839@xmath120446358 , was concurrently identified as an unresolved m9+t5 spectral binary @xcite and a rv variable with an orbital period of eight months @xcite . these examples serve to illustrate how spectral binaries can encompass a broad range of system architectures . to date , 34 vlm spectral binaries and candidates have been reported ( see table [ tab : sb ] ) , and ten have been confirmed by direct imaging , rv or astrometric variability (; gelino et al . in prep ) . detecting binaries using the spectral binary method is particularly useful for multiplicity statistics , as the method is independent of separation within 0@xmath145 , which translates to @xmath1510 - 20 au for field brown dwarfs at distances of 20 - 40 pc . the closest separation pairs can be followed - up to measure orbits and component masses , as well as infer ages by comparison to evolutionary models @xcite . systems with independent age constraints can also be used to test the evolutionary models directly @xcite . additionally , unresolved binaries are strong contaminants in luminosity functions that later lead to uncertainties in mass functions and studies of formation history through stellar populations @xcite , so their identification is extremely important . finally , spectral binaries with late - m / early - l primaries and t dwarf secondaries can straddle the hydrogen burning limit , thus giving additional insight into brown dwarf evolution . in this paper we adapt the technique of @xcite to search for spectral binaries composed of late - m or early - l dwarf primaries with t dwarf secondaries . m dwarfs are the most common stars in the galaxy @xcite , and are the brightest vlm objects , enabling better statistics through larger magnitude - limited search volumes and sample sizes . m - dwarf spectra are also intrinsically distinct from t - dwarf spectra , but differ in brightness by several magnitudes , rendering peculiar features extremely subtle . in [ sec : obs ] we describe our spectral sample used to find spectral binaries , drawn from the spex prism libraries and new observations . in [ sec : mtbin ] we explain our two methods to identify spectral binary candidates : by visual examination ( [ sec : visual ] ) and through spectral indices ( [ sec : specind ] ) . in [ sec : specfit ] , we perform single and binary template fitting to identify fourteen binary candidates . in [ sec : candidates ] , we describe the properties of the candidates . in [ sec : blue ] and [ sec : separation ] , we discuss our major contaminant , blue l dwarfs , and show preliminary evidence that the separations of spectral binaries are tighter than the resolved population . our results are summarized in [ sec : summary ] . the spex prism library is composed of low resolution ( @xmath16 ) spectra acquired with the spex @xmath17 m spectrograph , mounted on the 3.0 m nasa infrared telescope facility ( irtf ) , located in mauna kea , hawaii @xcite . all spectra were obtained using the prism - dispersed spex mode , which continuously samples wavelengths between @xmath18 m at a dispersion of @xmath19 pixel@xmath20 . the library includes close to 2,000 sources , both previously published data and 530 new spectra acquired between november 2000 and december 2013 ( table [ tab : newobs ] ) . the new observations were obtained with the 0@xmath145 or 0@xmath147 slit , generally aligned with the parallactic angle . total integration times ranged between 360 s and 1200 s , depending on source brightness and atmospheric conditions , and were obtained in an abba dither patter along the slit . spectra of nearby a0 v stars were used to flux calibrate the raw spectra and correct for telluric absorption . internal flat fields and argon arc lamps were observed with each flux standard for pixel response and wavelength calibration . all data were reduced with the spextool package @xcite using standard settings . a detailed description of our reduction procedures is given in @xcite . the sources observed have optical and/or near - infrared spectral classifications reported in the literature . to obtain a self - consistent set of spectral types , we computed spex spectral types based on spectral indices , following the method described in @xcite . from these , we selected two samples : the `` candidate '' sample which has been purged of spectral types outside the m7-l7 range , optical subdwarfs , giants , and poor quality spectra , but keeping binaries , objects suspected of being binaries from previous studies , young objects and unusually red and blue dwarfs , and the `` template '' sample which has been purged of binaries , candidate binaries , giants and poor quality spectra ( as determined by visual inspection ) only . the `` candidate '' sample contains 815 spectra of 738 objects with spex spectral types between m7-l7 , as those would be the potential primaries for late - m / early - l plus t binaries . the `` template '' sample comprise 1110 spectra of 992 single sources whose spectral types range between m7-l7 for primaries and t1-t8 for secondaries used in spectral fitting . the distribution of spectral types for both samples is shown in figure [ fig : sample ] . in both samples , the number of spectra decreases toward later spectral types due to declining space densities for l dwarfs @xcite and sensitivity limits for late l and t dwarfs . since there are significantly more sources with late - m spectral types in our samples , it is more likely to find binaries with a late - m primary . the sources included were observed as part of several different programs , including our ongoing program to compile a magnitude - limited sample of l dwarfs ( burgasser et al . in prep . ) . as such , we do not claim the sample to be complete or unbiased . the spectral morphology of unresolved late - m / early - l plus t dwarf binary systems gives rise to a distinctive feature in blended - light spectra : a small `` dip '' centered at 1.63 @xmath21 m , which is the combination of ch@xmath22 absorption from the secondary and feh from the primary @xcite . methane does not exist in the spectra of late - m / early - l dwarfs , so its presence indicates a t dwarf companion . however , this feature is very weak in blended - light spectra since a t dwarf is significantly fainter than the m / l primary ( e.g. @xmath23 3.5 mag between an m8 and a t5 , which is the case for 2mass j03202839@xmath120446358 ) . moreover , variations in the spectral slope for a blue or red l dwarf , can make this feature ambiguous , as can poor correction of hydrogen lines in the a0v calibrators . alternative indicators such as a relatively higher flux around the 1.25 @xmath21 m peak and * an * inflated bump short ward of 2.2 @xmath21 m , may also reveal the presence of a t dwarf companion , or that the spectrum of the source is unusually blue . to facilitate our visual inspection , we fit the candidate sample to templates of single objects , following the same chi - squared minimization routine as in section [ sec : specfit ] , and then subtracted the median combination of the ten best fitting single sources from each spectrum . the objects with residuals consistent with a t dwarf spectrum were selected as visual candidates . to validate this procedure , we also performed the same template subtraction on four confirmed spectral binaries : sdss j000649.16@xmath12085246.3 , 2mass j03202839@xmath120446358 , sdss j080531.89 + 481233.0 , and 2mass j13153094@xmath122649513 ( see table [ tab : sb ] ) . the residuals from these subtractions clearly exhibited t dwarf - like morphologies . twelve sources were selected as visual candidates . in addition to visual inspection , we also used spectral indices to identify additional candidate binaries due to the subtlety of t dwarf features in combined - light spectra @xcite . we initially examined standard classification indices from @xcite , as well as the `` _ _ h__-dip '' index from @xcite , and further defined five new indices . the new indices were designed by comparing the residuals of the four known binary spectra after subtracting their best single template fits . as a control sample , we also examined single templates subtracted from each other , which showed no evidence for a t dwarf companion . the new spectral indices specifically designed in this paper are : * _ h_-bump : measures the peak in the continuum from the t dwarf in the _ h _ band relative to the dip centered around 1.63 @xmath21 m seen in m+t binaries , making this index complimentary to _ h_-dip . a higher value of _ h_-bump implies a larger flux at 1.55 @xmath21 m , possibly caused by the presence of a t dwarf . * _ j_-curve : designed to detect the flux coming from both the 1.05 @xmath21 m and 1.27 @xmath21 m peaks of a t dwarf , as compared to the deep methane absorption at 1.12 @xmath21 m . * _ j_-slope and _ k@xmath24_-slope : measure the slope of the peaks in the _ j _ and _ k@xmath24 _ bands at 1.27 @xmath21 m and 2.10 @xmath21 m . in both cases , the peaks in a single late - m / early - l should look somewhat flat , giving values close to one , whereas in a late - m / early - l plus t dwarf binary the slope of the _ j _ and _ k@xmath24 _ band peaks are slightly negative and positive , respectively . * h@xmath25o-_y _ : measures the prominence of the _ y_-band peak of the t dwarf at @xmath26 m compared to the water and methane absorption around @xmath27 m . m and l dwarfs do not present peaks in the _ y_-band . the thirteen indices examined are described in table [ tab : index ] . we also used _ j - k@xmath24 _ , _ j - h _ and _ h - k@xmath24 _ colors synthesized from the spectra themselves , and the source spectral type , for a total of seventeen parameters . comparing all seventeen parameters against each other yielded 136 pairings . after visual examination to determine which pairings best segregated the four known m / l+t binaries , twelve combinations were selected ( figure [ fig : paramspace ] ) . we then used two techniques to define regions of interest in each combination for candidate selection . if a trend among all sources was clear , we fit the points to a second order polynomial and defined a region demarcated in the y - axis by the + 1@xmath28 or -1@xmath28 curves from the fit function , and in the x - axis by the horizontal spread of the binary benchmarks . conversely , if the points did not indicate any trends , then the region was demarcated such that it included the four binary benchmarks . the limits to these regions are described in table [ tab : selectreg ] . objects falling in eight or more selection regions were considered strong index candidates , those falling in four to eight regions were considered weak index candidates ( figure [ fig : histogram ] ) . the number of selected sources rises steadily below four or five combinations , suggesting that sources selected fewer than four times could be spurious . three of our benchmarks were selected by all twelve combinations , while sdss j0006@xmath120852 missed only the spt / ch@xmath22-h cut , since it falls within one standard deviation from the fitting curve . in total , eight strong and twenty - two weak candidates were selected , including the previously identified spectral binaries 2mass j20261584@xmath122943124 @xcite and 2mass j13114227 + 3629235 @xcite . seven visual candidates overlapped with the index candidates : five as strong and two as weak . to statistically test the binary hypothesis for our visual and index - selected candidates , we compared each spectrum to templates of both single sources and binary systems , using the method described in @xcite . the candidates determined by visual inspection or spectral index selection were first rejected from the template pool . then , all spectra were interpolated onto a common wavelength scale from @xmath29 to @xmath30 m and normalized to the peak flux between @xmath31 m . each candidate spectrum @xmath32 $ ] was directly compared to all single templates @xmath33 $ ] and ranked by a weighted chi - squared statistic . @xmath34\left[\frac{c[\lambda]-\alpha t [ \lambda]}{\sigma_c [ \lambda]}\right]^2\ ] ] where @xmath35 $ ] is a vector of weights proportional to the wave band size of each pixel ( see @xcite ) , @xmath36 is a scaling factor minimizing @xmath37 and @xmath38 $ ] is the noise spectrum for each candidate . the statistic was computed over the wavelength range : @xmath39 m , @xmath40 m and @xmath41 m , avoiding regions of strong telluric absorption . binary templates were constructed by first scaling each template spectrum to absolute fluxes using the 2mass m@xmath42 versus spectral type relation of @xcite , and then combining all pairs of single templates , such that the spectral type of the primary was earlier than that of the secondary resulting in a total of @xmath43 binary templates . more specifically , the primary spectral type was fixed to lie between m7-l7 while the secondary spectral type ranged between t1-t8 , since types earlier than t1 do not evidence strong methane features yet . the best binary fits were ranked using a chi - squared minimization routine . we determined the true significance that a binary template is superior to a single template by comparing the @xmath37 distributions of the binary and single fits using the one - sided f - test statistic @xmath44 : @xmath45 here , dof is the degrees of freedom for each fit ( equation 2 in @xcite ) . candidates with an f - statistic falling under the @xmath46 confidence level were rejected , including five visual candidates . in particular , 2mass j14493784 + 2355378 and 2mass j14232186 + 6154005 ( also a weak index - selected candidate ) , two previously identified spectral binary candidates from @xcite and @xcite were rejected due to their low confidence level that the binary fit was statistically better than the single fit . since our template sample includes a wide range of objects such as young and unusually blue and red dwarfs , the peculiarities of these candidates may be better explained by factors other than unresolved binarity . one exception to the index selection was 2massi j1711457 + 223204 whose spex spectral type was too late to be included in the candidate sample , yet it was a visual candidate and passed the binary fit f - test . figures [ fig : strongfit ] and [ fig : weakfit ] show the best single ( left ) and binary ( right ) template fits to our strong and weak candidates . table [ tab : cand ] summarizes the results of these fits . upon further visual examination , some binary fits still proved unsatisfactory . this was the case for the blue l dwarfs 2mass j11181292@xmath120856106 , 2mass j14162409 + 1348267 , 2mass j15150083 + 4847416 , 2mass j17114558 + 40285779 , and the subdwarfs 2mass j03303847@xmath122348463 , 2mass j03301720 + 3505001 , 2mass j04024315 + 1730136 , 2mass j15412408 + 5425598 and 2mass j23311807 + 4607310 . section [ sec : blue ] discusses these issues in more detail . as a result , fourteen candidates have been recognized , of which eleven are newly identified . in an effort to balance the tradeoff between fidelity of binary candidates and completeness , we are leaning towards the former . our binary selection criteria are conservative and it is likely that other spectral binaries may be identified with slightly looser constraints . in summary , from the @xmath6800 sources compiled in the candidate sample , twelve were selected by visual inspection and thirty were selected by spectral indices . seven sources overlapped the results of these selection methods . after fitting all thirty - five candidates , seventeen were rejected due to their confidence level being lower than @xmath46 , and four more due to their unusually blue colors ( see section [ sec : blue ] ) , leading to a final count of fourteen : six strong , seven weak and one visual candidate not overlapping with the index - selected . labels of strong and weak candidates come from index selection . originally discovered by @xcite , 2mass j0236 + 0048 was classified as an l6 in the optical and l6.5 in the infrared by @xcite . in their study , @xcite comment that this object may belong to the pleiades moving group , given its proper motion of @xmath47 = [ 161.33\pm10.10 , 176.33\pm19.16]$ ] mas yr@xmath20 and agreement between photometric and moving group distances at d = 26 pc . however , @xcite reclassified this object as an l9 , reducing its spectroscopic distance to 18 pc while its strong feh band at 0.99@xmath21 m argues against low surface gravity @xcite . nevertheless , the spectrum of this source does not show any signs of youth @xcite . 2mass j0236 + 0048 is selected by eleven out of twelve spectral index combinations , and its binary fit is significantly better than its single fit , making this a strong binary candidate with l5.0@xmath480.6 and t1.9@xmath481.1 components . @xcite discovered sdss j0931 + 2802 in the sdss catalog and classified it as an l3 at a mean distance of 29@xmath489 pc . its spectrum shows excess flux in the _ j_-band at @xmath49 m and a noticeable dip in the _ h_-band in the vicinity of @xmath50 m , as expected for a t dwarf component . this source was selected as a visual candidate , and by eleven out of twelve spectral index combinations , and our spectral fitting predicts component types of l1.4@xmath480.1 and t2.6@xmath480.9 . identified as a brown dwarf candidate by , 2mass j1311 + 3629 is a peculiar l5 . while also classified as unusually blue in wavelengths longer than _ j _ band @xcite , it lacks evidence of low metallicity or h@xmath25 collision - induced absorption ( cia ) in _ h _ and _ k _ bands . @xcite identified the methane feature in the _ h _ band centered around 1.63 @xmath21 m suggesting unresolved binarity . in this study , it was selected by eleven spectral index combinations and also as a visual candidate due to its methane absorption band starting at 1.60 @xmath21 m . template fitting gives spectral types of l4.8@xmath480.6 and t4.1@xmath482.7 . 2mass j1341@xmath123052 was discovered by @xcite and classified as an l3 in the optical by @xcite , who also measured its parallax and distance ( @xmath51 pc ) . 2mass j1341@xmath123052 was selected by eight spectral indices , and its spectral fitting suggests component spectral types of l1.2@xmath480.3 and t6.3@xmath481.0 . 2mass j1453 + 1420 was classified as an l1 in both the infrared @xcite and the optical @xcite , where it clearly shows excess flux in the _ j _ band and a dip in the _ h _ band . it is selected by eleven out of twelve spectral index combinations , and it is slightly blue with a _ j - k@xmath24 _ color of 1.18@xmath480.05 as compared to the median for * the * l1 spectral type 1.34@xmath480.19 @xcite . it is best fit by l1.1@xmath480.0 and t6.0@xmath481.1 components . 2mass j2026@xmath122943 had already been identified as a spectral binary candidate by @xcite , but it failed to be resolved by keck ao , thus setting an upper limit in separation of 0@xmath1425 or a projected separation of 9 au at a distance of 36@xmath485 pc @xcite . this source clearly shows a dip in its spectrum centered at 1.63 @xmath21 m , and it is best fit by a combination of l1.0@xmath480.5 and t5.8@xmath481.0 components . 2mass j0206 + 2235 was discovered and classified as an l5.5 by @xcite , and characterized as a blue l dwarf by @xcite . it was selected by seven spectral index combinations and fit to l5.1@xmath480.5 and t3.2@xmath482.3 components . 2mass j0427@xmath121127 was discovered and classified as an m7 by . it was selected by five spectral index combinations and best fit by m7.4@xmath480.2 and t5.1@xmath481.5 components . 2mass j1036@xmath123441 was classified as an l6 @xcite at a distance of 21@xmath483 pc . it almost made the cut for a strong candidate , since it was selected by seven spectral index combinations . this source was best fit by components with l5.2@xmath480.4 and t1.4@xmath480.4 spectral types . despite not having a pronounced methane absorption feature centered at 1.63 @xmath21 m , the binary fit is significantly better than the single fit , especially at the _ j _ band peak . 2mass j1059@xmath122113 is an l1 @xcite at a distance of 32.1 @xmath482.2 pc . this source was selected by four spectral index combinations and its best binary fit yields components with l0.6@xmath480.4 and t3.4@xmath481.3 spectral types . its spectrum shows a strong absorption feature centered at 1.63 @xmath21 m , as well as a flux excess at 1.23@xmath21 m and 2.20@xmath21 m . sdss j1422 + 2215 was identified and classified as an l6 in the nir by @xcite and also as an unusually blue l dwarf , showing strong h@xmath25o and feh absorption bands , which may be due to subsolar metallicity and/or thinner condensate cloud decks . it was selected by four out of twelve spectral index combinations with most likely component spectral types of l4.2@xmath480.6 and t4.1@xmath482.3 . wise j1623@xmath120508 was classified as an l1 in the nir @xcite . this source was selected by four spectral index combinations and best fit by l0.6@xmath480.3 and t6.0@xmath480.3 components . 2mass j1707@xmath120138 was discovered and classified as an l2 by . selected by five spectral index combinations , its spectrum is best fit by components with l0.7@xmath480.5 and t4.3@xmath482.0 spectral types . its spectrum shows a strong absorption feature centered at 1.63@xmath21 m . 2mass j1711 + 2232 was first identified and classified as an l6.5 in the optical by @xcite . due to its feh and ch@xmath22 absorption features in the _ h _ band , @xcite suggested it could be a spectral binary with l5.0 and t5.5 components . we find slightly different component spectral types of l1.5@xmath480.6 and t2.5@xmath481.0 , yet this source was not selected by spectral indices because of its late spex spectral type of l8.8 . despite having been imaged with hst / wfpc , it remains unresolved @xcite . four of the candidates selected by spectral indices were rejected after spectral fitting due to their poor binary fits . when we investigated these sources in detail , we found they were classified as blue objects in the literature and/or showed an unusually blue spectrum . 2mass j11181292@xmath120856106 was classified as a metal - poor subdwarf by @xcite . sdss j141624.09 + 134826.7 is part of a resolved binary system with a t7.5 companion that is itself a blue outlier . @xcite rejected unresolved binarity for the primary based on a qualitative comparison to the unusually blue l dwarf 2mass j11263991@xmath125003550 . the l6 2mass j15150083 + 4847416 shows a stable rv of -29.97@xmath480.14 km s@xmath52 @xcite and no signs of binarity from its spectrum . finally , 2mass j17114558 + 40285779 was discovered by @xcite as an unusually blue wide companion to the k star g203 - 50 . they discuss the possibility that the object may be unusual due to unresolved binarity , but argue in favor of low metallicity . for all of these sources , the lack of single templates akin to blue objects resulted in statistically better binary fits , yet the match is still relatively poor around the 1.63 @xmath21 m methane absorption feature . a few more previously unidentified nir subdwarfs were also selected as weak candidates and subsequently rejected due to their poor binary fits . the best binary fits for 2mass j03303847@xmath122348463 , 2mass j03301720 + 3505001 , 2mass j04024315 + 1730136 , 2mass j15412408 + 5425598 , and 2mass j23311807 + 4607310 use another subdwarf as a primary , which again indicates that they are part of a rare blue population that has a short supply of examples in this sample . m+t binaries have slightly bluer spectra caused by the extra flux in the _ j _ band corresponding to the peak in the t dwarfs . particularly , some sources originally classified as unusually blue have been later identified as spectral binaries ( e.g. sdss j0805 + 4812 @xcite ) . in contrast , intrinsically blue l dwarfs have low metallicity , thin cloud coverage , large - grain clouds or a combination of these , causing a blue tilt to the nir spectrum @xcite . @xcite have defined red and blue photometric outliers as the objects whose _ j - k@xmath24 _ color placed them 2@xmath28 or 0.4 mag away from the average for their spectral type , while pointing out the difficulty to distinguish outliers beyond a spectral type of l9 due to the small sample of objects . figure [ fig : j - k ] shows the _ j - k _ colors for our sample as compared to their spectral types , including the median and @xmath53 lines as calculated from the sample ( solid lines ) and reported in the literature ( dashed lines ) by @xcite and @xcite for samples of m and l dwarfs , respectively . figure [ fig : j - k ] suggests that blue l dwarfs are a major contaminant in our sample since a significant fraction of both known binaries and candidates have similar colors and thus lie in the same region as blue sources . we conclude that the blue l dwarf contaminants can be recognized if rejected due to their poor fits to binary template spectra . true confirmation of our candidates requires observational follow - up to either resolve the systems or measure rv or astrometric variability . as noted in the introduction , spectral binaries can be used to devise an unbiased method to measure the vlm binary separation distribution . therefore , it is worth examining the separation distribution of vlm and brown dwarf spectral binaries confirmed to date , to see if there are any differences compared to the resolved population . figure [ fig : separation ] shows the distribution of projected separations from 122 confirmed vlm binaries . among the observational methods for detecting binaries , such as direct imaging , radial velocity variations , astrometric variations , and microlensing , direct imaging has proven to be the most successful so far ( 73@xmath3 of confirmed vlm binaries ) , but its biggest drawback is its limit in resolution . at minimum angular scales of 0@xmath141 - 0@xmath142 for ao and hst programs , and typical distances of field brown dwarfs of 20 - 30 pc , telescope sensitivity reaches its limit at separations of around 2 - 6 au . at 2.90 au , the mean projected separation of eight independently - confirmed spectral binaries plotted in figure [ fig : separation ] falls at the lower end of this sensitivity limit , at less than the mean of known vlm binaries excluding the spectral binaries ( 3.75 au ) , raising the possibility that there may be significantly more tightly bound systems . to assess whether this is a significant difference , we performed a two - sample kolmogorov - smirnov test comparing the projected separation distributions of all binary systems to the confirmed spectral binaries . specifically , the distributions were constrained in angular separation to 50 - 500 mas , where the lower limit corresponds to the smallest possible imaging resolution in good seeing , while the upper limit restricts the maximum size of the slit . in addition , the distance was constrained to less than 30 pc , since objects that are farther away would be more difficult to confirm as binaries . in this way , we intend to fairly compare the spectral binary method to the other available methods for binary detection . these constraints reduced the number of spectral binaries to six . the result was a d statistic of 0.41 and a probability of @xmath54 . while the low probability is indicative of a difference between the samples , the small sample size makes this statistic inconclusive . many more of the existing spectral binaries need to be characterized before a significant difference can be confirmed or ruled out . we have identified fourteen brown dwarf binary candidates with late - m / early - l plus t dwarf components based on visual inspection of low resolution data , and analysis with spectral indices and template fitting . we combined five new spectral indices , with previously defined ones , spectral type , and _ j - h _ , _ h - k@xmath24 _ , and _ j - k@xmath24 _ colors to define pairings that effectively select spectral binary candidates , and confirmed them by comparison to both single and binary template spectra from the spex prism library . unusually blue l dwarfs were the main contaminant of this analysis , with four candidates classified as unusually blue but nonetheless being poorly matched to binary spectra . exploring the separation distribution of binary systems we find suggestive evidence that spectral binaries are more closely separated than other binaries , but the confirmed sample is too small to be conclusive . we are now undertaking follow - up ao imaging and rv monitoring of these candidates to confirm them and measure orbital properties . the authors thank telescope operators for their assistance during our observations . dbg would like to thank the friends of the international center at ucsd for their generous scholarship as well as davy kirkpatrick and fellow graduate students alex mendez and david vidmar for their helpful discussion and coding tips . this publication makes use of data from the spex prism spectral libraries , maintained by adam burgasser at http://www.browndwarfs.org/spexprism ; the dwarf archives compendium , maintained by chris gelino at http://dwarfarchives.org ; and the vlm binaries archive maintained by nick siegler at http://vlmbinaries.org . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . + _ facilities : _ irtf ( spex ) . lccccccccccc sdss j000649.16 - 085246.3 & m9 & & m8.5@xmath480.5 & t5@xmath481 & 14.14@xmath480.04 & 1.01@xmath480.05 & 3.15@xmath480.31 & 0.29@xmath480.01 & rv & 2 ; 2 + ulas j004757.41 + 154641.4 & & t2.0@xmath482.0 & l8.0 & t7.0 & 17.83@xmath480.05 & 1.41@xmath480.07 & & @xmath55 & & 22 + 2mass j00521232 + 0012172 & l5 & & l4 & t3 & 16.36@xmath480.11 & 0.90@xmath480.19 & & @xmath56 & & 24 + sdss j011912.22 + 240331.6 & & t2 & t0@xmath480.7 & t4@xmath480.4 & 17.02@xmath480.18 & @xmath57 & -0.42@xmath480.19 & @xmath58 & & 21 + ulas j020529.62 + 142114.0 & & t1.0@xmath480.5 & t1.0 & t3.0 & 17.99@xmath480.04 & 1.06@xmath480.07 & & @xmath59 & & 22 + 2mass j02060879 + 22355930 & & l5.5 & l5.1@xmath480.5 & t3.2@xmath482.3 & 16.56@xmath480.11 & 1.39@xmath480.17 & 1.61@xmath480.89 & & & 26 + 2mass j02361794 + 0048548 & l6.5 & & l5.0@xmath480.6 & t1.9@xmath481.1 & 16.10@xmath480.08 & 1.43@xmath480.12 & 1.05@xmath480.48 & @xmath60 & & 26 + sdss j024749.90 - 163112.6 & & t2.0@xmath481.5 & t0@xmath480.2 & t7@xmath480.3 & 17.19@xmath480.18 & 1.57@xmath480.27 & 0.68@xmath480.10 & @xmath61 & & 21 + 2mass j03202839 - 0446358 & m8 : & l1 & m8.5@xmath480.3 & t5@xmath480.9 & 12.13@xmath480.03 & 1.13@xmath480.04 & 3.5@xmath480.2 & @xmath62 & rv & 5 ; 1 + sdss j035104.37 + 481046.8 & & t1.0@xmath481.5 & l6.5@xmath480.7 & t5@xmath480.7 & 16.47@xmath480.13 & 1.47@xmath480.18 & 0.31@xmath480.31 & @xmath63 & & 21 + denis - p j04272708 - 1127143 & & dm7 & m7.4@xmath480.2 & t5.1@xmath481.5 & 13.74 & 0.99@xmath480.14 & 4.13@xmath480.62 & & & 26 + 2mass j05185995 - 2828372 & & & l6 & t4 & 15.98@xmath480.10 & 1.82@xmath480.12 & 0.13@xmath480.19 & 1.80@xmath480.50 & di & 10 ; 7 + 2mass j07354882 + 2720167 & l1 & & l1 & l4 & 16.94@xmath480.13 & 1.28@xmath480.21 & & @xmath64 & & 23 + sdss j080531.84 + 481233.0 & l4 & l9.5 & l4.5 & t5 & 14.73@xmath480.04 & 1.46@xmath480.05 & 1.50@xmath480.09 & 0.9 - 2.3 & av & 6 ; 11 + sdss j090900.73 + 652527.2 & & t1.5 & t1.5@xmath480.5 & t2.5@xmath480.3 & 16.03@xmath480.09 & 0.86@xmath480.17 & -0.12@xmath480.10 & @xmath65 & & 21 + sdss j092615.38 + 584720.9 & t4.5 & & t3 & t6 & 16.77@xmath480.14 & @xmath66 & 0.4@xmath480.2 & 2.6@xmath480.5 & di & 16 ; 7 , 8 + sdss j093113.23 + 280227.1 & l3 & & l1.4@xmath480.1 & t2.6@xmath480.9 & 14.98@xmath480.04 & 1.25@xmath480.06 & 2.22@xmath480.23 & @xmath67 & & 26 + 2mass j09490860 - 1545485 & & t2 & t1@xmath480.2 & t2@xmath480.2 & 16.15@xmath480.12 & 0.92@xmath480.20 & -0.07@xmath480.05 & @xmath68 & & 21 + 2mass j10365305 - 3441380 & l6 & & l5.2@xmath480.4 & t1.4@xmath480.4 & 15.62@xmath480.05 & 1.82@xmath480.06 & 0.51@xmath480.32 & @xmath69 & & 26 + sdss j103931.35 + 325625.5 & & t1 & l7@xmath480.2 & t4@xmath480.2 & 16.41@xmath480.15 & 1.25@xmath480.22 & 0.26@xmath480.09 & @xmath63 & & 21 + 2mass j10595138 - 2113082 & l1 & & l0.6@xmath480.4 & t3.4@xmath481.3 & 14.56@xmath480.04 & 1.35@xmath480.06 & 2.58@xmath480.32 & @xmath1552.5 & & 26 + 2mass j11061197 + 2754225 & & t2.5 & t0@xmath480.2 & t4.5@xmath480.2 & 14.82@xmath480.04 & 1.02@xmath480.07 & -0.37@xmath480.06 & @xmath152.67 & ol & 4 , 14 ; 15 + sdss j120747.17 + 024424.8 & l8 & t0 & l6.5@xmath480.7 & t2.5@xmath480.5 & 15.58@xmath480.07 & 1.59@xmath480.09 & 0.48@xmath480.28 & @xmath1517 & & 21 + 2mass j12144089 + 6316434 & & t3.5 & t2 & t6 & 16.59@xmath480.12 & 0.71@xmath480.26 & & @xmath1524 & & 24 + 2mass j13114227 + 3629235 & l5pec & & l4.8@xmath480.6 & t4.1@xmath482.7 & 15.54@xmath480.05 & 1.40@xmath480.09 & 2.19@xmath481.02 & @xmath70 & & 27 + 2mass j13153094 - 2649513 & l5 & & l3.5@xmath482.5 & t7@xmath480.6 & 15.07@xmath480.05 & 1.63@xmath480.07 & 3.03@xmath480.03 & 6.60@xmath480.90 & di & 3 ; 3 + 2mass j13243559 + 6358284 & & t2 : & l8@xmath480.2 & t3.5@xmath480.2 & 15.60@xmath480.07 & 1.54@xmath480.09 & -0.05@xmath480.06 & @xmath71 & & 21,2 4 + 2mass j13411160 - 30525049 & l3 & & l1.2@xmath480.3 & t6.3@xmath481.0 & 14.61@xmath480.03 & 1.53@xmath480.04 & 3.28@xmath480.53 & @xmath72 & & 26 + sdss j141530.05 + 572428.7 & & t3@xmath481 & l8@xmath480.5 & t5@xmath480.3 & 16.73@xmath480.16 & @xmath73 & -0.13@xmath480.20 & @xmath1529 & & 21 + sdss j142227.20 + 221557.5 & l6.5 & & l4.2@xmath480.6 & t4.1@xmath482.3 & 16.87@xmath480.03 & 1.23@xmath480.04 & 2.36@xmath480.78 & @xmath58 & & 26 + 2mass j14232186 + 6154005 & l4 & & l2 & l5 & 16.63@xmath480.15 & 1.35@xmath480.2 & & & & & 23 + sdss j143553.25 + 112948.6 & & t2@xmath481 & l7.5@xmath480.4 & t6@xmath480.3 & 17.14@xmath480.23 & @xmath74 & 0.41@xmath480.12 & @xmath72 & & 21 + sdss j143945.86 + 304220.6 & & t2.5 & t1@xmath480.2 & t5@xmath480.6 & 17.22@xmath480.23 & @xmath75 & 0.06@xmath480.24 & @xmath76 & & 21 + 2mass j14532589 + 1420418 & l1 & & l1.1@xmath480.0 & t6.0@xmath481.1 & 15.07@xmath480.04 & 1.18@xmath480.05 & 3.27@xmath480.46 & @xmath77 & & 26 + sdss j151114.66 + 060742.9 & & t0@xmath482 & l5.5@xmath480.8 & t5@xmath480.4 & 16.02@xmath480.08 & 1.47@xmath480.13 & 0.54@xmath480.32 & @xmath1521 & ol & 4 ; 12 + sdss j151603.03 + 025928.9 & & t0 : & l7.5@xmath481.1 & t2.5@xmath482.2 & 17.23@xmath480.20 & 1.80@xmath480.27 & 0.30@xmath480.65 & @xmath1527 & & 21 , 24 + wise j16235970 - 0508114 & & l1 & l0.6@xmath480.3 & t6.0@xmath480.8 & 14.94@xmath480.04 & 1.39@xmath480.06 & 3.39@xmath480.40 & & & 26 + 2mass j17072529 - 0138093 & & & l0.7@xmath480.5 & t4.3@xmath482.0 & 14.29@xmath480.03 & 1.22@xmath480.04 & 2.87@xmath480.75 & @xmath78 & & 26 + 2massi j17114573 + 2232044 & l6.5 & & l1.5@xmath480.6 & t2.5@xmath481.0 & 17.09@xmath480.18 & 2.36@xmath480.20 & 1.20@xmath480.40 & @xmath79 & & 21 + 2mass j17310140 + 5310476 & l6 & & l5 & l8 & 16.37@xmath480.11 & 1.52@xmath480.18 & & @xmath1548 & & 23,24 + 2mass j17373467 + 5953434 & l9 & & l5 & t5 & 16.88@xmath480.16 & 1.16@xmath480.31 & & @xmath80 & & 23 + 2mass j20261584 - 2943124 & l1 : & & l0.1@xmath480.5 & t5.8@xmath481.0 & 14.80@xmath480.03 & 1.44@xmath480.05 & 3.42@xmath480.40 & @xmath159.00 & di & 13 + sdss j205235.31 - 160929.8 & & t1@xmath481 & l7.5@xmath480.6 & t2@xmath480.2 & 16.33@xmath480.12 & 1.21@xmath480.19 & 0.04@xmath480.18 & 3.20@xmath480.50 & di & 4 ; 17 + 2mass j21265916 + 7617440 & l7 & t0 & l7 & t3.5 & 14.34@xmath480.03 & 1.18@xmath480.05 & & & & 28 + 2mass j21392676 + 0220226 & & t1.5 & l8.5@xmath480.7 & t3.5@xmath481.0 & 15.26@xmath480.05 & 1.68@xmath480.07 & @xmath810.21 & @xmath82 & & 21 + ulas j222958.30 + 010217.2 & & t3.0@xmath480.5 & t1.0 & t5.0 & 17.89@xmath480.04 & 0.67@xmath480.10 & & @xmath83 & & 22 + ulas j223348.82 + 002214.0 & & t0.0@xmath480.5 & l5.0 & t4.0 & 18.07@xmath480.05 & 1.43@xmath480.07 & & @xmath84 & & 22 + ulas j235618.01 + 075420.4 & & t0.0@xmath481.0 & l7.0 & t7.0 & 18.09@xmath480.05 & 1.87@xmath480.06 & & @xmath85 & & 22 + ulas j232315.39 + 071931.0 & & t2.0@xmath480.5 & t0.0 & t2.0 & 17.30@xmath480.02 & 1.01@xmath480.04 & & @xmath56 & & 22 + kelu-1a & l3pec@xmath481.5 & l2.0@xmath481.0 & l0.5@xmath480.5 & t7.5@xmath481 & 13.88@xmath480.03 & 1.57@xmath480.04 & & @xmath86 & & 25 + lllllccccl 2mass j0000286 - 124515 & j00002867 - 1245153 & m8.5 & & m9.2 & 13.20 & 1.23 & 2013 aug 14 & 120 & 44 ; 44 + lehpm 1 - 162 & j00054768 - 2157176 & m8.5 & m8 & m8.7 & 13.27 & 1.07 & 2006 sep 03 & 120 & 32 ; 75 + 2massi j0006205 - 172051 & j00062050 - 1720506 & l2.5 & & l3.0 & 15.66 & 1.65 & 2008 sep 08 & 120 & 25 ; 25 + sdss j000632.60 + 140606.4 & j00063260 + 1406064 & l1 & & l0.4 & 15.85 & 0.79 & 2009 jun 30 & 120 & 48 ; 48 + sdss j000646.81 + 151225.8 & j00064681 + 1512258 & l4 : : & & l4.3 & 16.22 & 1.41 & 2013 sep 03 & 120 & 62 ; 62 + 2mass j0007078 - 245804 & j00070787 - 2458042 & m7 & & m7.7 & 13.12 & 1.05 & 2013 oct 23 & 120 & 44 ; 44 + 2mass j00100009 - 2031122 & j00100009 - 2031122 & l0 & & m8.8 : & 14.13 & 1.25 & 2008 jul 14 & 120 & 44 ; 44 + 2massi j0013578 - 223520 & j00135779 - 2235200 & l4 & & l5.7 & 15.78 & 1.74 & 2009 nov 04 & 120 & 25 ; 25 + 2massw j0015447 + 351603 & j00154476 + 3516026 & l2 & & l1.0 & 13.88 & 1.61 & 2008 sep 08 & 120 & 13 ; 13 + sdss j001608.44 - 004302.3 & j00160843 - 00430209 & & l5.5 & l4.3 & 16.33 & 1.78 & 2013 sep 03 & 120 & 34 ; 34 + sdss j001637.62 - 103911.2 & j00163762 - 1039112 & l0 & & m8.9 & 15.46 & 0.92 & 2009 nov 04 & 120 & 48 ; 48 + sdss j001911.65 + 003017.8 & j00191165 + 0030176 & l1 & & l0.4 & 14.92 & 1.35 & 2009 nov 07 & 120 & 20 ; 20 + denis - p j0020.9 - 4414 & j00205982 - 4414340 & m8v & & m9.4 & 14.90 & 1.19 & 2007 nov 10 & 120 & 6 ; 6 + lehpm 1 - 494b & j00210589 - 4244433 & m9.5 & & l0.6 : & 13.52 & 1.22 & 2006 aug 28 & 120 & 6 ; 15 + sdss j00220934 - 0110397 & j00220934 - 0110397 & l0 & & m8.9 & 15.82 & 1.12 & 2012 sep 27 & 120 & 60 ; 60 + bri 0021 - 0214 & j00242514 - 0158166 & m9.5 & & l0.6 & 11.99 & 1.45 & 2007 jul 04 & 120 & 80 ; 44 + & & & & & & & 2008 jul 14 & 120 & 80 ; 44 + lhs 1074 & j00255117 - 0748069 & sdm6 & & m7.9 & 14.68 & 0.81 & 2004 sep 08 & 120 & ; 39 + lehpm 1 - 606 & j00271049 - 1813083 & & m8 & m7.5 & 13.45 & 1.02 & 2006 sep 03 & 120 & 32 ; 75 + 2massi j0028394 + 150141 & j00283943 + 1501418 & l4.5 & l3 & l6.1 : & 16.51 & 1.95 & 2009 dec 07 & 120 & 13 ; 13 , 34 + 2mass j00285545 - 1927165 & j00285545 - 1927165 & l0 : & & l0.7 & 14.19 & 1.35 & 2009 nov 04 & 120 & 49 ; 49 + 2massw j0030438 + 313932 & j00304384 + 3139321 & l2 & & l3.2 & 15.48 & 1.45 & 2008 sep 08 & 120 & 11 ; 11 + sips j0031 - 3840 & j00311925 - 3840356 & l2.5 & l2 & l1.2 & 14.10 & 1.18 & 2006 aug 28 & 120 & 36 ; 59 , 36 + 2mass j00320509 + 0219017 & j00320509 + 0219017 & l1.5 & m9 & l0.4 & 14.32 & 1.52 & 2009 jun 30 & 120 & 49 ; 49 , 30 + & & & & & & & 2009 nov 07 & 120 & 49 ; 49 , 30 + 2massi j0032431 - 223727 & j00324308 - 2237272 & l1 & & m9.7 : & 15.39 & 1.43 & 2009 dec 07 & 120 & 25 ; 25 + eros - mp j0032 - 4405 & j00325584 - 4405058 & l0gamma & & l0.5 & 14.78 & 1.51 & 2008 sep 07 & 120 & 8 ; 56 + 2massi j0034568 - 070601 & j00345684 - 0706013 & l3 & & l4.3 & 15.53 & 1.59 & 2009 nov 07 & 120 & 25 ; 25 + sdss j003843.99 + 134339.5 & j00384397 + 13433950 & l1 & & l0.6 & 15.91 & 1.15 & 2012 sep 27 & 120 & 20 ; 20 + sdss j004154.54 + 134135.5 & j00415453 + 1341351 & l0 & & l0.6 & 14.45 & 1.22 & 2009 jan 24 & 120 & 20 ; 20 + & & & & & & & 2009 nov 08 & 120 & 20 ; 20 + lhs 1135 & j00433134 + 2054316 & sdm6.5 & & m7.0 & 13.95 & 0.81 & 2004 jul 24 & 120 & ; 39 + 2massw j0045214 + 163445 & j00452143 + 1634446 & l2beta & l3.5 & l1.3 : : & 13.06 & 1.69 & 2007 sep 16 & 120 & 30 ; 56 , 30 + 2mass j00464841 + 0715177 & j00464841 + 0715177 & l0 : : & m9 & m9.5 : & 13.89 & 1.34 & 2008 jul 13 & 120 & 30 ; 49 , 30 + sips j0050 - 1538 & j00502444 - 1538184 & l1 : & & l0.5 & 13.78 & 1.13 & 2008 sep 08 & 120 & 36 ; 44 + sdssp j005406.55 - 003101.8 & j00540654 - 00310180 & l1 & & l2.2 & 15.73 & 1.35 & 2012 sep 25 & 120 & 19 ; 20 + 2mass j00550460 - 3052000 & j00550460 - 3052000 & m8 : & & m7.2 & 13.03 & 1.12 & 2012 dec 27 & 120 & ; + lhs 1166 & j00554418 + 2506235 & sdm6.5 & & m7.2 & 14.26 & 0.69 & 2004 sep 07 & 120 & ; 39 + sdss j00570556 - 0846241 & j00570556 - 0846241 & l0 & & m9.3 & 15.70 & 1.34 & 2012 sep 25 & 120 & 60 ; 60 + 2massw j0058425 - 065123 & j00584253 - 0651239 & l0 & & l0.0 & 14.31 & 1.41 & 2008 jul 14 & 120 & 13 ; 13 + lhs 132 & j0102510 - 373743 & m8 & & m8.1 & 11.13 & 1.06 & 2008 sep 07 & 120 & ; + sdss j010311.51 - 004417.0 & j01031151 - 0044170 & m9.5 & & m8.5 & 14.29 & 1.12 & 2013 oct 29 & 120 & 62 ; 62 + 2massi j0103320 + 193536 & j01033203 + 1935361 & l6 & & l5.4 & 16.29 & 2.14 & 2006 sep 02 & 120 & 13 ; 13 + 2massi j0104075 - 005328 & j01040750 - 0053283 & l4.5 & & l5.8 : & 16.53 & 1.20 & 2009 nov 07 & 120 & 29 ; 81 + sdss j010637.33 + 151855.0 & j01063733 + 1518550 & m8.5 & m8 pec & m9.2 & 14.36 & 0.93 & 2013 sep 03 & 120 & 62 ; 62 , 66 + sdss j01071600 - 1517570 & j01071600 - 1517570 & m7 & & m7.2 & 13.34 & 1.06 & 2012 dec 27 & 120 & ; + sdssp j010752.33 + 004156.1 & j01075242 + 0041563 & l8 & l5.5 & l6.6 & 15.82 & 2.12 & 2005 oct 19 & 120 & 21 ; 20 , 34 + sdss j01084048 + 1347392 & j01084048 + 1347392 & l0 & & m9.3 & & & 2012 sep 27 & 120 & 60 ; 60 + 2mass j01165457 - 1357342 & j01165457 - 1357342 & m9 & & m8.8 & 14.21 & 1.24 & 2009 nov 08 & 120 & 49 ; 49 + 2mass j01170586 + 1752568 & j01170586 + 1752568 & m9 & m8 & m8.6 & 14.14 & 1.20 & 2009 nov 04 & 120 & 61 ; 61 + ssspm j0124 - 4240 & j01235905 - 4240073 & m8 & l2.5 & m8.4 & 13.15 & 1.12 & 2008 sep 07 & 120 & 37 ; 49 , 37 + 2massi j0125369 - 343505 & j01253689 - 3435049 & l2 & & l1.2 & 15.52 & 1.62 & 2008 sep 07 & 120 & 25 ; 25 + 2mass j01273917 + 2805536 & j01273917 + 2805536 & m8.5 & & m9.3 & 14.04 & 1.18 & 2013 aug 14 & 120 & ; + sdss j012743.51 + 135420.9 & j01274352 + 13542099 & l5 & l6+/-1 & l5.5 & 16.83 & 1.40 & 2012 sep 25 & 120 & 20 ; 20 , 34 + 2massw j0129122 + 351758 & j01291221 + 3517580 & l4 & & l4.7 & 16.78 & 2.08 & 2009 nov 07 & 120 & 11 ; 11 + 2mass j01335299 + 0033017 & j01335299 + 0033017 & m9 : & & m9.1 & 15.84 & 1.21 & 2009 jan 24 & 120 & 62 ; 62 + nltt 5282 & j0134592 - 090442 & esdm9.5 ? & & m8.8 & 15.25 & 0.49 & 2008 jul 14 & 120 & ; + 2massw j0135358 + 120522 & j01353586 + 1205216 & l1.5 & & l0.6 & 14.41 & 1.49 & 2009 nov 04 & 120 & 13 ; 13 + 2massw j0141032 + 180450 & j01410321 + 1804502 & l1 & l4.5 & l0.8 & 13.88 & 1.38 & 2008 sep 07 & 120 & 30 ; 44 , 30 + 2mass j01460119 - 4545263 & j01460119 - 4545263 & m9 & & m9.5 & 14.40 & 1.36 & 2008 oct 12 & 120 & 49 ; 49 + & & & & & & & 2009 dec 07 & 120 & 49 ; 49 + 2massw j0147334 + 345311 & j01473344 + 3453112 & l0.5 & & l1.0 & 14.95 & 1.37 & 2008 sep 08 & 120 & 11 ; 11 + 2mass j01490895 + 2956131 & j01490895 + 2956131 & m9.5 & & m9.7 & 13.45 & 1.47 & 2008 sep 07 & 120 & ; + sdss j01535423 + 1404528 & j01535423 + 1404528 & l0 & & m9.0 & 15.21 & 1.25 & 2012 sep 27 & 120 & 60 ; 60 + lehpm 2153 & j02042212 - 3632308 & m8 : & & m8.1 & 13.27 & 1.08 & 2013 dec 05 & 120 & 49 ; 49 + sdss j020608.97 + 223559.2 & j02060879 + 22355930 & & l5.5 & l6.9 : & 16.56 & 1.39 & 2013 dec 05 & 120 & 40 ; 40 + denis j02065660 - 0735190 & j02065660 - 0735190 & m8.5 & & m8.5 & 14.34 & 1.35 & 2013 aug 14 & 120 & 59 ; 59 + sdss j020735.60 + 135556.3 & j02073556 + 13555639 & l3 & l3+/-1.5 & l0.6 & 15.46 & 1.65 & 2012 sep 27 & 120 & 20 ; 20 , 34 + 2massw j0208549 + 250048 & j02085498 + 25004880 & l5 & & l5.9 & 16.21 & 1.80 & 2012 sep 25 & 120 & 13 ; 13 + 2massi j0218291 - 313322 & j02182913 - 3133230 & l3 & & l3.8 & 14.73 & 1.57 & 2009 nov 08 & 120 & 27 ; 27 + 2mass j02192196 + 0506306 & j02192196 + 0506306 & l1 : & l1 & l3.0 & 14.97 & 1.49 & 2008 sep 23 & 120 & ; 55 + ssspm j0219 - 1939 & j02192807 - 1938416 & l1 & l2.5 & l0.6 & 14.11 & 1.20 & 2008 sep 08 & 120 & 83 ; 37 + 2massi j0224367 + 253704 & j02243669 + 25370419 & l2 & & l2.3 & 16.58 & 1.89 & 2012 sep 27 & 120 & 13 ; 13 + 2mass j02284243 + 1639329 & j02284243 + 16393299 & l0 : & & m8.7 & 13.17 & 1.35 & 2012 oct 27 & 120 & 49 ; 49 + 2mass j02301551 + 2704061 & j02301551 + 2704061 & l0 : & & m9.4 & 14.29 & 1.31 & 2008 sep 07 & 120 & 44 ; 44 + denis j02304500 - 0953050 & j02304500 - 0953050 & l0 & & l0.9 & 14.68 & 1.69 & 2013 aug 14 & 120 & 59 ; 59 + sdss j023547.56 - 084919.8 & j02354755 - 08491980 & l2 & & l2.0 : & 15.57 & 1.38 & 2012 sep 27 & 120 & 20 ; 20 + sdssp j023617.93 + 004855.0 & j02361793 + 00485479 & l6 & l6.5 & l6.6 : & 16.10 & 1.43 & 2012 sep 25 & 120 & 21 ; 20 , 34 + 2massi j0239424 - 173547 & j02394245 - 1735471 & l0 & & m9.0 & 14.29 & 1.25 & 2009 nov 08 & 120 & 27 ; 27 + 2massi j0241536 - 124106 & j02415367 - 1241069 & l2 : & & l1.9 & 15.61 & 1.67 & 2008 sep 08 & 120 & 27 ; 27 + 2massw j0242435 + 160739 & j02424354 + 16073920 & l1.5 & & l1.7 & 15.78 & 1.43 & 2012 oct 27 & 120 & 11 ; 11 + sdss j02511323 + 00473631 & j02511323 + 00473631 & m8 & & m8.2 & 13.77 & 1.09 & 2013 sep 03 & 120 & 48 ; 48 + 2mass j02522628 + 0056220 & j02522628 + 0056220 & m8 & & m8.3 & 13.13 & 1.16 & 2012 dec 27 & 120 & ; + sdss j02540582 - 1934523 & j02540582 - 1934523 & m9 & m9 & m8.1 & 13.08 & 1.17 & 2009 jan 25 & 120 & 49 ; 85 + & & & & & & & 2012 dec 27 & 120 & 49 ; 85 + 2mass j02594471 + 2254443 & j02594471 + 2254443 & m9 & & m9.5 & 14.11 & 1.25 & 2009 dec 07 & 120 & 49 ; 49 + 2massi j0302012 + 135814 & j03020121 + 13581419 & l3 & & l2.7 : & 16.53 & 1.90 & 2005 oct 19 & 120 & 13 ; 13 + & & & & & & & 2012 sep 27 & 120 & 13 ; 13 + lehpm 1 - 3070 & j03061185 - 3647417 & m8 & l0 & m8.7 & 11.69 & 1.06 & 2006 sep 03 & 120 & 32 ; 37 + 2massw j0306268 + 154514 & j03062684 + 15451370 & l6 : & & l6.0 : & 17.11 & 1.97 & 2012 sep 25 & 120 & 13 ; 13 + sdss j03083243 - 08105138 & j03083243 - 08105138 & m8 & & m7.6 & 14.98 & 0.89 & 2013 sep 03 & 120 & 48 ; 48 + 2massw j0309088 - 194938 & j03090888 - 19493870 & l4.5 & & l4.9 & 15.75 & 1.69 & 2012 sep 25 & 120 & 13 ; 13 + 2mass j03101401 - 2756452 & j03101401 - 2756452 & l5 & & l6.5 & 15.80 & 1.84 & 2009 nov 08 & 120 & 44 ; 44 + 2mass j03140344 + 1603056 & j03140344 + 16030560 & l0 & & m9.4 & 12.53 & 1.29 & 2012 sep 25 & 120 & 49 ; 49 + 2massi j0316451 - 284852 & j03164511 - 28485209 & l0 : & & l1.0 & 14.57 & 1.46 & 2013 dec 05 & 120 & 27 ; 27 + 2mass j03201710 - 1026120 & j03201710 - 1026120 & m8 & & m7.6 & 13.87 & 1.15 & 2013 aug 14 & 120 & 44 ; 44 + 2mass j03250136 + 2253039 & j03250136 + 22530390 & l3 & & l3.3 & 15.43 & 1.65 & 2013 sep 03 & 120 & 44 ; 44 + 2massw j0326137 + 295015 & j03261367 + 2950152 & l3.5 & & l4.6 & 15.48 & 1.65 & 2009 dec 07 & 120 & 11 ; 11 + 2mass j03264453 + 1919309 & j03264453 + 1919309 & m8.5 & & m8.6 & 13.12 & & 2012 sep 25 & 120 & 49 ; 49 + sdssp j032817.38 + 003257.2 & j03281737 + 00325719 & l3 & & l2.5 : & 15.99 & 1.83 & 2012 sep 27 & 120 & 19 ; 20 + lspm j0330 + 3504b & j03301720 + 3505001 & & d / sdm7 & m8.6 & 16.00 & 0.68 & 2005 oct 17 & 120 & ; 75 + & & & & & & & 2005 dec 31 & 120 & ; 75 + sdssp j033035.13 - 002534.5 & j03303511 - 0025346 & l4 & & l6.5 & 15.31 & 1.47 & 2009 nov 04 & 120 & 12 ; 20 + lehpm 1 - 3365 & j03303847 - 2348463 & & esdm7 & m8.8 & 15.80 & 0.86 & 2004 sep 09 & 120 & 32 ; 75 + 2mass j03320043 - 2317496 & j03320043 - 2317496 & m8 & & m8.4 & 13.64 & 1.10 & 2013 oct 29 & 120 & 49 ; 49 + 2mass j03335134 + 0014068 & j03335134 + 0014068 & sdl & & m7.6 & 16.43 & -0.35 & 2013 dec 05 & 120 & 62 ; 62 + lehpm 1 - 3396 & j03340904 - 4953382 & & m8 & m9.4 & 11.38 & 0.98 & 2004 sep 06 & 120 & 32 ; 75 + 2mass j03350236 + 2342407 & j03350236 + 2342407 & m8.5 & & m7.4 & 12.25 & 0.99 & 2013 dec 05 & 120 & 14 ; 67 + 2mass j03354535 + 0658058 & j03354535 + 0658058 & m8 & & m8.0 & 13.41 & 1.15 & 2012 oct 27 & 120 & 49 ; 49 + 2massw j0337036 - 175807 & j03370359 - 1758079 & l4.5 & & l6.0 : & 15.62 & 2.04 & 2009 nov 08 & 120 & 13 ; 13 + 2mass j03395284 + 2457270 & j03395284 + 2457270 & m8 & & m7.7 & 12.84 & 1.10 & 2012 dec 27 & 120 & ; + sdss j03440891 + 0111249 & j03440891 + 0111249 & l1 & & l2.9 : & 14.74 & 1.22 & 2012 oct 27 & 120 & 60 ; 60 + sdss j03510002 - 00524587 & j03510002 - 00524587 & m8 & & m7.3 & 11.30 & 1.07 & 2012 oct 27 & 120 & 48 ; 48 + 2mass j03521086 + 0210479 & j03521086 + 02104797 & m9 & l0+/-1 & m8.6 & 13.08 & 1.12 & 2012 oct 27 & 120 & 47 ; 49 , 47 + sdss j035308.54 + 103056.0 & j03530854 + 1030560 & l1 & & l0.9 & 15.45 & 1.27 & 2013 oct 29 & 120 & 60 ; 60 + 2mass j03540135 + 2316330 & j03540135 + 2316330 & m8 & & m7.9 & 13.12 & 1.13 & 2012 dec 27 & 120 & ; + lehpm 2 - 471 & j03551067 - 1858173 & & m7 & m7.6 & 13.81 & 0.93 & 2006 sep 03 & 120 & 32 ; 75 + 2mass j0355201 + 143929 & j0355201 + 143929 & m8 & & m8.8 & 13.81 & 1.11 & 2013 oct 29 & 120 & 44 ; 44 + 2mass j03552337 + 1133437 & j03552337 + 1133437 & l5gamma & & l4.8 : : & 14.05 & 2.52 & 2009 dec 07 & 120 & 49 ; 56 + 2massw j0355419 + 225702 & j03554190 + 22570159 & l3 & & l2.9 : & 16.11 & 1.83 & 2012 sep 27 & 120 & 11 ; 11 + denis - p j035729.6 - 441730 & j03572695 - 4417305 & l0beta & & l0.6 : & 14.37 & 1.46 & 2007 sep 16 & 120 & 38 ; 56 + 2mass j04012977 - 4050448 & j04012977 - 40504488 & l0 : & & l0.0 & 14.53 & 1.36 & 2012 dec 27 & 120 & 49 ; 49 + lspm j0402 + 1730 & j04024315 + 1730136 & & sdm7 & m7.0 & 15.59 & 0.56 & 2005 oct 17 & 120 & ; 75 + & & & & & & & 2005 dec 31 & 120 & ; 75 + 2mass j04070752 + 1546457 & j04070752 + 1546457 & l3.5 & & l3.9 & 15.48 & 1.92 & 2009 dec 07 & 120 & 49 ; 49 + 2mass j0407089 - 234829 & j0407089 - 234829 & m8 : & & m8.5 & 13.77 & 1.15 & 2013 oct 29 & 120 & 44 ; 44 + 2mass j04081032 + 0742494 & j04081032 + 0742494 & m8 & & m8.1 & 13.59 & 1.17 & 2013 dec 05 & 120 & 49 ; 49 + 2massi j0408290 - 145033 & j04082905 - 1450334 & l2 & l4.5 & l1.3 & 14.22 & 1.40 & 2009 nov 04 & 120 & 30 ; 27 , 30 + 2massi j0409095 + 210439 & j04090950 + 2104393 & l3 & & l3.4 & 15.51 & 1.66 & 2009 dec 07 & 120 & 13 ; 13 + 2mass j0417474 - 212919 & j0417474 - 212919 & m8 & & m8.3 & 13.85 & 1.18 & 2013 oct 29 & 120 & 44 ; 44 + sdss j04270723 + 0859027 & j04270723 + 0859027 & m8 & & m8.8 & 12.92 & 1.19 & 2012 dec 27 & 120 & ; + denis j0427270 - 112713 & j04272708 - 1127143 & m7 & & m7.9 & 13.74 & 1.07 & 2013 dec 05 & 120 & 59 ; 59 + 2massi j0428510 - 225323 & j04285095 - 22532270 & l0.5 & & m9.9 & 13.51 & 1.39 & 2012 sep 25 & 120 & 25 ; 25 + lp 655 - 23b & j04305157 - 0849007 & m8 & & m8.5 & 12.90 & 1.12 & 2013 dec 05 & 120 & 28 ; 46 + lp 775 - 31 & j0435161 - 160657 & & m7 & m8.0 & 10.41 & 1.05 & 2004 sep 07 & 120 & ; + 2mass j04362054 - 4218523 & j04362054 - 42185236 & l0 : & & l0.4 & 14.49 & 1.35 & 2012 dec 27 & 120 & 49 ; 49 + 2massi j0443058 - 320209 & j04430581 - 32020899 & l5 & & l6.4 : & 15.27 & 1.40 & 2012 sep 27 & 120 & 25 ; 25 + sdss j044337.61 + 000205.1 & j04433761 + 0002051 & m9 & & m9.0 & 12.51 & 1.29 & 2013 dec 05 & 120 & 20 ; 44 + 2mass 04441479 + 0543573 & j04441479 + 0543573 & m8 & & m7.9 & 13.67 & 1.15 & 2013 dec 05 & 120 & 49 ; 49 + 2massi j0451009 - 340214 & j04510092 - 34021500 & l0.5 & & l0.6 & 13.54 & 1.25 & 2012 sep 27 & 120 & 27 ; 27 + 2massi j0453264 - 175154 & j04532647 - 1751543 & l3 : & & l3.1 & 15.14 & 1.68 & 2009 nov 04 & 120 & 27 ; 27 + 2mass j05002100 + 0330501 & j05002099 + 03305010 & l4 & & l4.1 & 13.67 & 1.61 & 2012 dec 27 & 120 & 49 ; 49 + 2massi j0502134 + 144236 & j05021345 + 1442367 & l0 & & m8.9 & 14.27 & 1.32 & 2013 dec 05 & 120 & 60 ; 55 + 2mass j05120636 - 2949540 & j05120636 - 29495400 & l4.5 & & l5.7 : : & 15.46 & 2.18 & 2013 oct 29 & 120 & 27 ; 52 + 2mass j05170548 - 4154413 & j05170548 - 4154413 & m9 & & m8.8 & 13.46 & 1.19 & 2012 dec 27 & 120 & ; + lehpm 2 - 183 & j05173729 - 3348593 & m8 & m9 & m9.4 & 12.00 & 1.17 & 2004 sep 06 & 120 & 32 ; 27 , 75 + 2mass j05184616 - 2756457 & j05184616 - 2756457 & l0 : & & l1.0 & 15.26 & 1.65 & 2009 jan 25 & 120 & 44 ; 44 + 2mass j05264348 - 4455455 & j05264348 - 4455455 & m9.5 & & m9.8 & 14.08 & 1.38 & 2013 dec 12 & 120 & 49 ; 49 + 2mass j05301261 + 6253254 & j05301261 + 6253254 & l1 & & l1.1 & 14.05 & 1.34 & 2008 jan 09 & 120 & 49 ; 49 + 2mass j0534584 - 151143 & j0534584 - 151143 & m9 & & m8.9 & 13.15 & 1.15 & 2013 oct 29 & 120 & 44 ; 44 + sdssp j053951.99 - 005902.0 & j05395199 - 00590189 & l5 & l5 & l5.0 & 14.03 & 1.51 & 2009 jan 24 & 120 & 12 ; 12 , 34 + 2mass j05431887 + 6422528 & j05431887 + 6422528 & l1 & & l1.2 & 13.57 & 1.52 & 2008 jan 09 & 120 & 49 ; 49 + 2mass j05441150 - 2433010 & j05441150 - 2433010 & m8 & & m7.6 & 12.53 & 1.07 & 2012 dec 27 & 120 & ; + 2mass j06022216 + 6336391 & j06022216 + 6336391 & l1 : & & l1.5 & 14.27 & 1.58 & 2008 jan 08 & 120 & 49 ; 49 + lsr 0602 + 3910 & j06023045 + 3910592 & l1 & & l0.8 & 12.30 & 1.44 & 2009 nov 04 & 120 & 88 ; 88 + 2mass j06050190 - 2342260 & j06050190 - 2342260 & l0 : & & m9.8 & 14.51 & 1.37 & 2012 dec 27 & 120 & 44 ; 44 + 2mass j06050196 - 2342270 & j06050190 - 2342260 & l0 : & & l1.1 : & 14.51 & 1.37 & 2009 jan 25 & 120 & 44 ; 44 + 2mass j06085283 - 2753583 & j06085283 - 2753583 & m8.5 low g & l0 lg & m9.6 & 3.600 & -8.78 & 2006 sep 02 & 120 & 27 ; 59 , 71 + sips j0614 - 2019 & j06141196 - 2019181 & & l4 & l1.3 & 14.78 & 1.41 & 2006 sep 01 & 120 & ; 72 + denis - p j0615493 - 010041 & j06154934 - 01004158 & l2+/-1 & & l1.0 & 13.75 & 1.21 & 2010 jan 02 & 120 & 53 ; 53 + & & & & & & & 2012 dec 27 & 120 & 53 ; 53 + 2mass j06244595 - 4521548 & j06244595 - 45215479 & l5 : & & l5.1 & 14.48 & 1.88 & 2013 oct 29 & 120 & 49 ; 49 + sdss j062621.22 + 002934.2 & j06262121 + 00293410 & l1 & & l0.3 & 15.93 & 1.07 & 2010 jan 02 & 120 & 20 ; 20 + 2mass j06411840 - 4322329 & j06411839 - 43223290 & l1.5 & & l2.4 : & 13.75 & 1.30 & 2012 dec 27 & 120 & 49 ; 49 + denis - p j0652197 - 253450 & j06521977 - 25345058 & l0 & & m9.2 & 12.76 & 1.24 & 2009 jan 25 & 120 & 53 ; 53 + & & & & & & & 2009 nov 04 & 120 & 53 ; 53 + denis - p j0652197 - 253450 & j06521977 - 2534505 & l0 & & m9.2 & 12.76 & 1.24 & 2006 sep 02 & 120 & 53 ; 53 + lehpm 2 - 461 & j06591011 - 4747002 & m6.5 & m7 & m7.4 & 13.64 & 0.92 & 2006 dec 20 & 120 & 32 ; 87 , 75 + lehpm 2 - 436 & j07075333 - 4900574 & m8.5 & m8 & m8.8 & 13.23 & 1.12 & 2006 dec 20 & 120 & 32 ; 75 + & & & & & & & 2006 dec 21 & 120 & 32 ; 37 , 75 + 2massw j0708213 + 295035 & j07082132 + 29503500 & l5 & & l5.4 : & 16.72 & 1.95 & 2013 dec 12 & 120 & 13 ; 13 + denis - p j0716478 - 063037 & j07164790 - 06303696 & l1+/-1 & & l1.1 & 13.90 & 1.33 & 2010 jan 02 & 120 & 53 ; 53 + & & & & & & & 2011 apr 04 & 120 & 53 ; 53 + 2mass j07200325 - 0846499 & j07200325 - 0846499 & m9 : & & m9.8 & 10.63 & 1.16 & 2013 dec 05 & 120 & 70 ; 70 + 2mass j07231462 + 5727081 & j07231462 + 5727081 & l1 & & l0.2 & 13.97 & 1.36 & 2008 jan 08 & 120 & 49 ; 49 + sdss j07342570 + 30065792 & j07342571 + 3006583 & m8 & & m7.6 & 15.13 & 1.05 & 2009 jan 25 & 120 & 48 ; 48 + sdss j07351959 + 4108503 & j07351959 + 4108503 & l0 & & m8.9 & 15.78 & 0.93 & 2010 jan 26 & 120 & 60 ; 60 + & & & & & & & 2011 mar 09 & 120 & 60 ; 60 + 2massw j0740096 + 321203 & j07400965 + 32120320 & l4.5 & & l4.0 & 16.19 & 1.97 & 2010 jan 26 & 120 & 61 ; 61 + & & & & & & & 2012 dec 27 & 120 & 13 ; 13 + sdss j074756.31 + 394732.9 & j07475631 + 3947329 & l0 & & m8.3 & 15.08 & 1.35 & 2008 jan 11 & 120 & 20 ; 20 + sdss j074838.61 + 174332.9 & j07483861 + 1743329 & l7 & & l6.0 & 16.27 & 1.85 & 2013 oct 29 & 120 & 54 ; 54 + denis - p j0751164 - 253043 & j07511645 - 2530432 & l2.5 & & l1.1 & 13.16 & 1.17 & 2009 nov 04 & 120 & 53 ; 53 + 2massi j0753321 + 291711 & j07533216 + 29171190 & l2 & & l2.0 & 15.52 & 1.67 & 2010 jan 28 & 120 & 13 ; 13 + & & & & & & & 2011 apr 03 & 120 & 13 ; 13 + 2mass j07575274 + 0914103 & j07575274 + 0914103 & l4 : : & & l3.7 : & 15.86 & 1.77 & 2013 apr 26 & 120 & 54 ; 54 + sdss j080048.13 + 465825.5 & j08004815 + 46582560 & l2 & & l1.3 & 15.51 & 1.20 & 2011 mar 09 & 75 & 20 ; 20 + 2mass j08041429 + 0330474 & j08041429 + 0330474 & m8.5 & & m9.7 & 13.69 & 1.25 & 2013 oct 29 & 120 & 49 ; 49 + sdss j080959.01 + 443422.2 & j08095903 + 4434216 & & l6 & l6.3 : & 16.44 & 2.02 & 2008 jan 12 & 120 & 34 ; 40 + denis - p j0812316 - 244442 & j08123170 - 24444239 & l2.5+/-1 & & l0.9 & 13.82 & 1.43 & 2010 jan 03 & 120 & 53 ; 53 + & & & & & & & 2012 dec 27 & 120 & 53 ; 53 + sdss j08175749 + 1824050 & j08175749 + 1824050 & l1 & & l2.0 & 15.09 & 1.27 & 2013 apr 26 & 120 & 60 ; 60 + sdss j08181228 + 3310482 & j08181228 + 3310482 & l0 & & l1.3 & 15.99 & 0.93 & 2011 apr 03 & 120 & 60 ; 60 + sdss j081946.02 + 165853.9 & j08194602 + 1658539 & m9 & & m8.6 & 13.79 & 1.17 & 2013 dec 05 & 120 & 74 ; 74 + denis j0823031 - 491201 & j08230313 - 4912012 & l1.5 & l3 & l2.7 & 13.55 & 1.48 & 2013 nov 24 & 120 & 53 ; 53 , 68 + 2mass j08230838 + 6125208 & j08230837 + 61252079 & l2 : & & l2.5 & 14.82 & 1.62 & 2011 apr 03 & 120 & 49 ; 49 + sdss j08264265 + 19392195 & j08264262 + 1939224 & l0 & & m9.2 & 14.78 & 1.20 & 2010 jan 25 & 120 & 48 ; 48 + 2massw j0829066 + 145622 & j08290664 + 1456225 & l2 & & l1.0 & 14.75 & 1.58 & 2008 jan 08 & 120 & 13 ; 13 + gj 1111 & j08294949 + 2646348 & m6.5 & & m7.2 & 8.230 & 0.97 & 2013 dec 12 & 120 & 2 ; 58 + 2massw j0829570 + 265510 & j08295706 + 26550990 & l6.5 & & l6.4 : : & 17.11 & 2.15 & 2013 dec 12 & 120 & 13 ; 13 + 2mass j08303256 + 0947150 & j08303256 + 0947150 & m8 & & m7.8 & 11.89 & 1.13 & 2012 dec 27 & 120 & ; + 2mass j08315564 + 1025466 & j08315564 + 1025466 & m9 & & m8.6 & 13.62 & 1.17 & 2009 mar 21 & 120 & ; 49 + 2massw j0832045 - 012835 & j08320451 - 0128360 & l1.5 & & l1.2 & 14.13 & 1.42 & 2008 jan 10 & 120 & 13 ; 13 + 2mass j08352366 + 1029318 & j08352366 + 1029318 & m7 & & m7.9 & 13.14 & 1.09 & 2013 dec 05 & 120 & 49 ; 49 + sdss j08354533 + 2224308 & j08354537 + 2224310 & l0 & & m8.7 & 15.74 & 1.29 & 2010 jan 25 & 120 & 60 ; 60 + 2mass j08355829 + 0548308 & j08355829 + 05483080 & l3 & & l1.7 & 14.53 & 1.36 & 2010 jan 24 & 120 & 49 ; 55 + & & & & & & & 2011 apr 04 & 120 & 49 ; 55 + sdss j08362199 + 4949315 & j08362199 + 4949315 & l0 & & m9.3 : & 15.42 & 0.88 & 2011 mar 09 & 120 & 60 ; 60 + sdss j08364634 + 0526426 & j08364635 + 0526426 & l0 & & m9.4 & 14.58 & 1.29 & 2008 sep 23 & 120 & 60 ; 60 + 2mass j08391608 + 1253543 & j08391608 + 1253543 & m9 & & m8.8 & 13.75 & 1.16 & 2013 dec 12 & 120 & 74 ; 74 + lhs 2034 & j08402975 + 1824091 & m6 & & m7.0 & 11.05 & 1.01 & 2013 dec 12 & 120 & 1 ; 57 + sdss j08410685 + 6035063 & j08410685 + 6035063 & l4 & & l1.5 & 15.94 & 1.25 & 2011 apr 03 & 120 & 60 ; 60 + sdss j08430794 + 3141292 & j08430794 + 3141292 & l3 & & l2.5 & 15.99 & 1.34 & 2011 apr 03 & 120 & 60 ; 60 + sdss j08433328 + 1024435 & j08433328 + 1024435 & l1 & & l2.7 : & 14.87 & 1.20 & 2013 dec 12 & 120 & 60 ; 60 + sdss j084403.46 + 043436.19 & j08440346 + 04343619 & m8 & & m8.7 & 13.46 & 1.05 & 2013 dec 12 & 120 & 48 ; 48 + sdss j08475148 + 0138110 & j08475148 + 0138110 & l3 & & l3.2 & 16.23 & 1.82 & 2013 dec 05 & 120 & 60 ; 60 + 2mass j08490052 + 0220155 & j08490052 + 0220155 & m8 & & m7.3 & 12.93 & 1.03 & 2013 dec 05 & 120 & 74 ; 74 + 2mass j08533434 - 0329432 & j08533434 - 0329432 & m9e & & m8.6 & 15.04 & 0.52 & 2009 mar 21 & 120 & ; + sdss j08583697 + 2710508 & j08583697 + 2710508 & l0 & & l0.0 & 15.05 & 1.39 & 2011 mar 09 & 120 & 60 ; 60 + 2massi j0859254 - 194926 & j08592522 - 1949279 & l6 : : & l8 & l6.8 & 15.53 & 1.78 & 2008 jan 12 & 120 & 27 ; 27 , 69 + 2mass j08593854 + 6341355 & j08593854 + 6341355 & l0 & & m8.6 & 13.70 & 1.31 & 2008 jan 11 & 120 & 49 ; 49 + 2mass j08594029 + 1145325 & j08594029 + 1145325 & m8 & & m8.4 & 12.74 & 1.25 & 2012 dec 27 & 120 & ; + lhs 2090 & j09002359 + 2150054 & m6.5 & & m7.3 & 9.440 & 1.00 & 2013 dec 12 & 120 & 17 ; 58 + sdss j09002368 + 2539343 & j09002368 + 2539343 & l7 & & l6.7 & 16.43 & 1.77 & 2013 dec 12 & 120 & 60 ; 60 + sdss j090206.90 + 003319.36 & j09020690 + 00331936 & m7 & & m7.3 & 12.11 & 0.95 & 2013 dec 05 & 120 & 48 ; 48 + sdss j09094813 + 1940439 & j09094813 + 1940439 & l1 & & m9.9 & 14.73 & 1.29 & 2013 apr 26 & 120 & 60 ; 60 + denis - p j0909 - 0658 & j09095749 - 0658186 & l0 & & l0.2 & 13.89 & 1.35 & 2008 jan 10 & 120 & 9 ; 52 + 2mass j09161504 + 2139512 & j09161504 + 2139512 & m9 & & m9.8 & 13.22 & 1.15 & 2013 dec 12 & 120 & 74 ; 74 + 2massw j0918382 + 213406 & j09183814 + 21340580 & l2.5 & & l2.7 & 15.66 & 1.76 & 2011 apr 03 & 120 & 11 ; 11 + sdss j09230870 + 2340137 & j09230870 + 2340137 & l1 & & l2.3 & 13.85 & 1.04 & 2011 mar 09 & 120 & 60 ; 60 + 2massw j0928397 - 160312 & j09283972 - 1603128 & l2 & & l1.7 & 15.32 & 1.71 & 2008 jan 12 & 120 & 13 ; 13 + sdss j09311323 + 2802271 & j09311323 + 2802271 & l3 & & l2.5 : & 14.98 & 1.25 & 2011 mar 09 & 120 & 60 ; 60 + sdss j093128.22 + 052821.93 & j09312822 + 05282193 & m7 & & m8.4 & 12.86 & 1.06 & 2013 dec 05 & 120 & 48 ; 48 + sdss j09323747 + 6725145 & j09323747 + 6725145 & l0 & & m9.4 & 15.91 & 0.92 & 2011 apr 03 & 120 & 60 ; 60 + 2mass j09340617 + 0536234 & j09340617 + 0536234 & m8 pec & & m7.5 & 15.57 & 0.81 & 2004 mar 11 & 120 & 45 ; 45 , 75 + 2mass j09352803 - 2934596 & j09352802 - 29345959 & l0 & & l0.3 & 14.04 & 1.21 & 2012 dec 27 & 120 & 49 ; 49 + sdss j09385888 + 0443439 & j09385888 + 0443439 & l0 & & m7.8 & 15.24 & 1.24 & 2013 apr 26 & 120 & 60 ; 60 + sdss j09404788 + 2946530 & j09404788 + 2946530 & l1 & & l0.4 & 15.29 & 1.37 & 2013 apr 26 & 120 & 60 ; 60 + sdss j09413492 + 1009421 & j09413492 + 1009421 & l0 & & m8.7 & 14.58 & 1.21 & 2013 apr 26 & 120 & 60 ; 60 + 2massw j0944027 + 313132 & j09440279 + 3131328 & l2 & & l1.7 & 15.50 & 1.49 & 2008 jan 09 & 120 & 13 ; 13 + 2mass 09474477 + 0224327 & j09474477 + 0224327 & m8 : & & m8.2 & 13.17 & 1.09 & 2013 dec 05 & 120 & 49 ; 49 + lhs 2195 & j09492223 + 0806450 & m8.5 & & m8.5 & 12.31 & 1.10 & 2013 dec 12 & 120 & 49 ; 49 + 2mass j09524622 + 0620410 & j09524622 + 0620410 & m8 & & m7.3 & 12.45 & 0.99 & 2013 dec 12 & 120 & 74 ; 74 + 2mass j09532126 - 1014205 & j09532126 - 1014205 & l0 & & m9.4 & 13.47 & 1.33 & 2008 jan 12 & 120 & 44 ; 44 + 2mass j09532126 - 1014205 & j09532126 - 10142056 & l0 & & l0.0 & 13.47 & 1.33 & 2006 dec 20 & 120 & 44 ; 44 + nltt 22851b & j09532455 + 0526583 & m9.5 & & m9.8 & 15.67 & 1.28 & 2013 dec 05 & 120 & 62 ; 62 + sdss j100319.17 - 010508.15 & j10031917 - 01050815 & m7 & & m8.0 & 12.33 & 1.09 & 2013 dec 12 & 120 & 48 ; 48 + g 196 - 3b & j10042066 + 5022596 & l3beta & & l5.6 : : & 14.83 & 2.05 & 2006 dec 23 & 120 & 7 ; 56 + lhs 5166b & j10043929 - 33351889 & l4 & & l4.5 : & 14.48 & 1.56 & 2013 apr 26 & 120 & 22 ; 22 + lhs 2243 & j10163470 + 2751497 & m8 v & & m7.4 & 11.99 & 1.03 & 2013 dec 12 & 120 & 5 ; 5 + 2massi j1017075 + 130839 & j10170754 + 1308398 & l2 : & l1 & l1.1 & 14.10 & 1.39 & 2008 jan 12 & 120 & 27 ; 27 , 30 + sdss j10174251 + 4310579 & j10174251 + 4310579 & l1 & & l0.9 & 15.55 & 1.03 & 2011 mar 09 & 120 & 60 ; 60 + 2massw j1018588 - 290953 & j10185879 - 29095349 & l1 & & l0.3 & 14.21 & 1.42 & 2012 dec 27 & 120 & 22 ; 22 + denis j1019245 - 270717 & j10192447 - 2707171 & l0.5 & & m9.1 & 13.53 & 1.06 & 2013 dec 12 & 120 & 59 ; 59 + 2mass 10213232 - 2044069 & j10213232 - 2044069 & m9 & & m8.5 & 13.19 & 1.13 & 2013 dec 12 & 120 & 49 ; 49 + 2mass j10220489 + 0200477 & j10220489 + 0200477 & l0 & & m9.0 & 14.10 & 1.20 & 2008 jan 10 & 120 & 49 ; 49 + 2mass j10224821 + 5825453 & j10224821 + 5825453 & l1beta & & l0.6 & 13.50 & 1.34 & 2008 jan 11 & 120 & 49 ; 49 + sdss j102552.43 + 321234.0 & j10255227 + 3212349 & & l7.5+/-2.5 & l5.8 : & 15.91 & 0.84 & 2008 jan 08 & 120 & 40 ; 40 + 2massi j1029216 + 162652 & j10292165 + 1626526 & l2.5 & & l2.8 & 14.29 & 1.67 & 2008 jan 12 & 120 & 13 ; 13 + sdss j10330910 + 1216259 & j10330910 + 1216259 & l0 & & m8.9 & 15.07 & 1.12 & 2013 apr 26 & 120 & 60 ; 60 + sdss j10340567 + 0350163 & j10340567 + 0350163 & l0 & & m8.7 & 14.70 & 1.07 & 2013 apr 26 & 120 & 60 ; 60 + 2massw j1035245 + 250745 & j10352455 + 2507450 & l1 & & l1.1 & 14.76 & 1.47 & 2008 jan 08 & 120 & 13 ; 13 + denis j104617.0 - 421237 & j10461703 - 4212372 & & m8 & m8.2 & 14.29 & 1.01 & 2008 jan 24 & 120 & ; 84 + lp 213 - 68 & j10471381 + 4026493 & m8 & & m8.2 & 12.44 & 1.18 & 2013 dec 12 & 120 & 16 ; 16 + denis - p j1047 - 1815 & j10473108 - 18155739 & l2.5 & & l0.3 & 14.20 & 1.31 & 2012 dec 27 & 120 & 10 ; 10 + sdss j104922.45 + 012559.2 & j10492244 + 01255930 & l5 & & l3.5 : & 15.88 & 1.77 & 2011 apr 04 & 120 & 54 ; 54 + 2mass j10511900 + 5613086 & j10511900 + 56130860 & l2 & & l0.8 & 13.24 & 1.34 & 2012 may 30 & 75 & 49 ; 49 + sdss j10515124 + 13111633 & j10515124 + 13111633 & l0 & & m9.5 : & 14.83 & 1.20 & 2013 apr 26 & 120 & 48 ; 48 + sdss j105547.29 + 080842.64 & j10554729 + 08084264 & m9 & & m8.5 & 12.55 & 1.18 & 2013 dec 12 & 120 & 48 ; 48 + 2massi j1059513 - 211308 & j10595138 - 21130819 & l1 & & l2.8 : & 14.56 & 1.35 & 2012 dec 27 & 120 & 27 ; 27 + lhs 2351 & j11061897 + 0428327 & m7 & & m7.6 & 12.33 & 1.00 & 2013 dec 12 & 120 & 4 ; 39 + 2mass j11073750 - 2759385b & j11073750 - 2759385 & & m7 & m7.0 & & & 2009 dec 29 & 120 & ; 84 + gliese 417bc & j11122567 + 35481310 & l4.5 & & l4.8 & 14.58 & 1.86 & 2007 mar 16 & 120 & 13 ; 13 + lhs 2397a & j11214909 - 1313082 & m8 & & m8.5 & & & 2011 mar 09 & 120 & ; + 2massw j1122362 - 391605 & j11223623 - 39160540 & l3 & & l3.7 : : & 15.71 & 1.83 & 2012 dec 27 & 120 & 22 ; 22 + sdss j11264703 + 5816322 & j11264703 + 5816322 & l3 & & l1.3 & 15.84 & 1.30 & 2011 apr 03 & 120 & 60 ; 60 + lehpm 2 - 333 & j11414421 - 2232204 & & m8 & m8.8 & 12.63 & 1.06 & 2006 apr 09 & 120 & 32 ; 75 + sdss j11491231 - 0153006 & j11491231 - 0153006 & l1 & & m8.5 & 14.67 & 0.96 & 2011 mar 09 & 120 & 60 ; 60 + 2mass j11533966 + 5032092 & j11533966 + 5032092 & l1 : & & l0.3 & 14.19 & 1.34 & 2008 jan 11 & 120 & 49 ; 49 + gj 3693 & j11535267 + 0659561 & m8 & & m7.0 & 11.26 & 0.99 & 2013 dec 12 & 120 & 4 ; 63 + 2mass j11544223 - 3400390 & j11544223 - 3400390 & l0 & & l0.5 & 14.20 & 1.34 & 2008 jan 09 & 120 & 38 ; 52 + 2massw j1155395 - 372735 & j11553951 - 37273499 & l2 & & l2.3 & 12.81 & 1.35 & 2012 dec 27 & 120 & 22 ; 22 + lp 851 - 346 & j11554286 - 2224586 & m7.5 & & m7.7 & 10.93 & 1.05 & 2013 dec 12 & 120 & 24 ; 24 + denis - p j1157480 - 484442 & j11574809 - 48444283 & l0.5+/-1 & & m9.4 & 14.01 & 1.21 & 2008 jan 24 & 120 & 53 ; 53 + denis - p j1159 + 0057 & j11593850 + 00572679 & l0 & & l0.4 & 14.08 & 1.27 & 2013 apr 25 & 120 & 10 ; 20 + sdss j11594072 + 5409386 & j11594072 + 5409386 & l2 & & l0.9 & 15.22 & 1.46 & 2013 apr 24 & 120 & 60 ; 60 + sdssp j120358.19 + 001550.3 & j12035811 + 00155000 & l3 & & l5.0 & 14.01 & 1.53 & 2011 apr 04 & 120 & 12 ; 12 + 2massi j1204303 + 321259 & j12043036 + 32125950 & l0 & m9 & m9.6 & 13.82 & 1.30 & 2013 apr 24 & 120 & 27 ; 27 , 30 + sdss j12061049 + 6242572 & j12061049 + 6242572 & l1 & & l2.1 & 15.65 & 1.70 & 2011 mar 09 & 120 & 60 ; 60 + denis j1206501 - 393725 & j12065011 - 3937261 & l2 & & l1.9 & 14.32 & 1.22 & 2007 mar 17 & 120 & 59 ; 59 + 2mass j12073804 - 3909050 & j12073803 - 39090500 & l2 : & & l0.9 & 14.69 & 1.44 & 2011 mar 11 & 120 & 49 ; 49 + 2massi j1213033 - 043243 & j12130335 - 04324369 & l5 & & l4.2 & 14.68 & 1.67 & 2011 apr 04 & 120 & 27 ; 27 + bri 1222 - 1222 & j12245222 - 1238352 & m9 & & m8.5 & 12.57 & 1.22 & 2006 jun 01 & 120 & 89 ; + 2mass j12312141 + 4959234 & j12312141 + 49592339 & l2 & & l3.4 & 14.62 & 1.48 & 2013 apr 24 & 120 & 44 ; 44 + 2mass j12321827 - 0951502 & j12321827 - 0951502 & l0 & & m9.5 & 13.73 & 1.17 & 2008 jan 14 & 120 & 49 ; 49 + sdss j12455566 + 4902109 & j12455566 + 4902109 & l1 & & m7.8 : & 15.95 & 0.75 & 2011 mar 09 & 120 & 60 ; 60 + 2massw j1246467 + 402715 & j12464677 + 40271500 & l4 & & l4.0 & 15.09 & 1.81 & 2011 apr 03 & 120 & 13 ; 13 + sdss j125128.43 + 624310.7 & j12512843 + 6243107 & l4 & & l4.2 & 15.49 & 1.13 & 2013 apr 24 & 120 & 54 ; 54 + 2mass j12565688 + 0146163 & j12565688 + 0146163 & l2 : & & l1.6 & 14.48 & 1.69 & 2009 jun 30 & 120 & 49 ; 49 + sdssp j125737.26 - 011336.1 & j12573726 - 0113360 & l4 & l5 & l5.8 & 15.94 & 1.82 & 2009 jun 30 & 120 & 21 ; 20 , 34 + 2massw j1300425 + 191235 & j13004255 + 1912354 & l1 & l3 & l1.7 & 12.72 & 1.09 & 2006 apr 10 & 120 & 14 ; 14 , 50 + 2mass j13015465 - 1510223 & j13015464 - 15102229 & l1 & & l0.5 & 14.54 & 1.44 & 2013 apr 26 & 120 & 49 ; 49 + gj 499c & j13054106 + 2046394 & l4 : & & l6.5 & 15.20 & 1.83 & 2009 jun 30 & 120 & 27 ; 27 + 2mass j13061727 + 3820296 & j13061726 + 38202960 & l0 & & l0.6 & 14.63 & 1.41 & 2011 apr 03 & 120 & 49 ; 49 + sdss j131142.1 + 362923.9 & j13114192 + 3629247 & l6 & l5 pec ( blue ) & l6.8 & 15.55 & 1.40 & 2012 jul 09 & 120 & 54 ; 54 , 66 + 2mass j13120700 + 3937440 & j13120700 + 3937440 & l0 : & & m8.8 & 14.14 & 1.25 & 2012 may 30 & 75 & ; + 2massi j1315309 - 264951 & j13153094 - 2649513 & l5.5 & & l6.7 & 15.20 & 1.73 & 2009 jun 30 & 120 & 79 ; 52 + 2mass j13204427 + 0409045 & j13204427 + 04090450 & l3 : : & & l2.5 & 15.25 & 1.63 & 2013 apr 24 & 120 & 49 ; 49 + denis - p j1323 - 1806 & j13233597 - 18063790 & l0 & & l0.0 & 14.90 & 1.24 & 2013 apr 26 & 120 & 10 ; 10 + 2massw j1326201 - 272937 & j13262009 - 2729370 & l5 & & l6.6 : & 15.85 & 1.99 & 2009 jun 30 & 120 & 22 ; 22 + sdss j13271521 + 0759375 & j13271521 + 0759375 & l1 & & l0.5 & 14.60 & 1.36 & 2013 apr 25 & 120 & 60 ; 60 + 2massw j1328550 + 211449 & j13285503 + 2114486 & l5 & & l4.1 & 16.19 & 1.93 & 2009 jun 30 & 120 & 11 ; 11 + sdss j133148.92 - 011651.4 & j13314893 - 01165000 & l6 & l8+/-2.5 & l6.1 & 15.46 & 1.39 & 2011 mar 09 & 120 & 20 ; 20 , 34 + sdss j13331279 + 1509566 & j13331279 + 1509566 & l0 & & m7.8 & 15.84 & 0.75 & 2011 mar 11 & 120 & 60 ; 60 + sdss j13334536 - 0216002 & j13334536 - 0216002 & l3 & & l2.2 & 15.38 & 1.53 & 2013 apr 25 & 120 & 60 ; 60 + 2mass j13364062 + 3743230 & j13364062 + 3743230 & l1 & & l0.4 & 14.41 & 1.31 & 2013 apr 26 & 120 & 44 ; 44 + 2mass j13373116 + 4938367 & j13373115 + 49383670 & l0 & & m8.9 & 13.77 & 1.19 & 2012 may 30 & 75 & 44 ; 44 + 2massw j1338261 + 414034 & j13382614 + 41403420 & l2.5 & & l2.4 & 14.22 & 1.45 & 2012 may 30 & 75 & 13 ; 13 + 2mass j13384944 + 0437315 & j13384944 + 04373150 & l1 & & l0.0 & 14.16 & 1.42 & 2012 may 30 & 75 & 49 ; 49 + 2mass j13411160 - 3052505 & j13411160 - 30525049 & l2 : : & & l2.7 : & 14.61 & 1.53 & 2012 jul 09 & 120 & 49 ; 49 + 2massw j1343167 + 394508 & j13431670 + 39450870 & l5 & & l5.3 : : & 16.16 & 2.01 & 2013 apr 24 & 120 & 13 ; 13 + sdss j13571490 - 1438520 & j13571490 - 1438520 & m7 & & m8.5 & 12.85 & 1.11 & 2012 may 30 & 75 & ; + 2mass j13595510 - 4034582 & j13595510 - 40345819 & l1 & & l3.1 : & 13.65 & 1.08 & 2013 apr 25 & 120 & 49 ; 49 + 2mass j14022235 + 0648479 & j14022235 + 0648479 & m9 & & m8.9 & 13.72 & 1.21 & 2010 jul 07 & 120 & ; + 2mass j14044495 + 4634297 & j14044495 + 4634297 & l0 : & & m9.7 & 14.34 & 1.28 & 2008 jul 13 & 120 & 44 ; 44 + sdss j14060148 + 5249309 & j14060148 + 5249309 & l0 & & m9.8 & 15.56 & 1.00 & 2011 mar 09 & 120 & ; 60 + 2mass j14090310 - 3357565 & j14090310 - 3357565 & l2 & & l1.3 & 14.25 & 1.38 & 2010 jul 07 & 120 & 52 ; 52 + 2massw j1411175 + 393636 & j14111735 + 3936363 & l1.5 & & l1.5 & 14.64 & 1.40 & 2008 jul 13 & 120 & 13 ; 13 + 2mass j14122270 + 2354100 & j14122270 + 2354100 & m9 & & m7.8 & 13.73 & 1.08 & 2009 jun 30 & 120 & ; + 2massw j1412244 + 163312 & j14122449 + 1633115 & l0.5 & & m9.9 & 13.89 & 1.37 & 2008 jul 30 & 120 & 13 ; 13 + sdss j14205830 + 2131566 & j14205830 + 2131566 & l1 & & l0.0 : & 15.12 & 1.06 & 2011 mar 11 & 120 & 60 ; 60 + 2massw 1421314 + 182740 & j14213145 + 1827407 & l0 & & m8.9 & 13.23 & 1.29 & 2008 jul 30 & 120 & 14 ; 49 + sdss j142227.25 + 221557.1 & j14222720 + 2215575 & & l6.5+/-2 & l5.4 : & 17.06 & 1.42 & 2009 jun 30 & 120 & 40 ; 40 + sdss j14225715 + 0827521 & j14225715 + 0827521 & l2 & & l2.1 & 15.10 & 1.45 & 2013 apr 25 & 120 & 60 ; 60 + 2mass j14232186 + 6154005 & j14232186 + 6154005 & & l4 & l4.5 & 16.60 & 1.32 & 2013 apr 26 & 120 & 65 ; 65 + gd 165b & j14243909 + 0917104 & l4 & l3+/-2 & l4.4 & 15.69 & 1.52 & 2009 jun 29 & 120 & 3 ; 11 , 34 + denis - p j142527.97 - 365023.4 & j14252798 - 3650229 & l3 : & l5 & l4.5 : : & 13.75 & 1.94 & 2010 jul 07 & 120 & 31 ; 49 , 31 + 2mass j14261286 + 3130394 & j14261286 + 3130394 & l4 & & l4.0 : : & 16.62 & 1.90 & 2013 apr 25 & 120 & 54 ; 54 + 2mass j14283132 + 5923354 & j14283132 + 59233546 & l4 & & l4.4 & 14.78 & 1.52 & 2012 jul 09 & 120 & 49 ; 49 + lhs 2924 & j14284313 + 3310368 & m9 v & & m9.1 & 11.99 & 1.25 & 2007 jul 04 & 120 & ; + 2massi j1430435 + 291540 & j14304358 + 2915405 & l2 & l0.5 & l1.5 & 14.27 & 1.50 & 2008 jul 13 & 120 & 30 ; 27 , 30 + sdss j14324210 + 3451427 & j14324210 + 3451427 & l1 & & l1.2 & 15.75 & 0.98 & 2011 apr 19 & 120 & 60 ; 60 + sdss j14380829 + 64083631 & j14380829 + 64083631 & l0 & & l0.2 & 12.98 & 1.33 & 2011 mar 11 & 120 & 48 ; 48 + sdss j143832.63 + 572216.9 & j14383259 + 5722168 & l5 & & l4.6 & 15.96 & 1.59 & 2010 jul 07 & 120 & 54 ; 54 + 2massw j1438549 - 130910 & j14385498 - 13091029 & l3 : & & l3.1 & 15.49 & 1.63 & 2012 jul 09 & 120 & 13 ; 13 + sdss j143933.44 + 031759.2 & j14393342 + 03175909 & l1 & & l0.7 & 15.99 & 1.18 & 2011 apr 19 & 120 & 20 ; 20 + sdss j14403025 + 12333391 & j14403025 + 12333391 & m9 & & m8.6 & 14.41 & 1.22 & 2013 mar 27 & 75 & 48 ; 48 + 2massi j1441045 + 271932 & j14410457 + 27193234 & m7 & & m7.0 & 12.99 & 1.02 & 2013 mar 27 & 75 & 44 ; 44 + g 124 - 62bc & j14413716 - 09455900 & l0.5 & & l0.4 & 14.02 & 1.36 & 2008 jul 14 & 120 & 10 ; 13 + & & & & & & & 2011 mar 09 & 120 & 10 ; 13 + g 239 - 25 b & j14422175 + 6603198 & & l0+/-1 & m8.1 & 11.51 & 1.18 & 2010 jul 07 & 120 & 35 ; 78 + ssspm 1444 - 2019 & j14442067 - 2019222 & d / sdm9 & d / sdm7 & m9.0 & 12.55 & 0.61 & 2005 mar 23 & 120 & 33 ; 45 , 75 + 2mass j14442946 + 0048530 & j14442946 + 0048530 & m9 & & m8.1 & 15.95 & 0.76 & 2009 jun 30 & 120 & ; + sdssp j144600.60 + 002452.0 & j14460060 + 00245190 & l6 & l5 & l4.2 & 15.89 & 1.96 & 2011 mar 11 & 120 & 21 ; 20 , 34 + 2massw j1449378 + 235537 & j14493784 + 2355378 & l0 & & m9.5 & 15.82 & 1.51 & 2008 jul 13 & 120 & 13 ; 13 + 2mass j14520183 + 1114590 & j14520183 + 1114590 & l2 & & l2.9 & 15.52 & 1.18 & 2013 apr 26 & 120 & 77 ; 62 + sdss j14525558 + 2723244 & j14525558 + 2723244 & l0 & & l0.2 : & 14.92 & 0.84 & 2011 mar 09 & 120 & 60 ; 60 + sdss j14532589 + 1420418 & j14532589 + 1420418 & l1 & & l4.1 : & 15.07 & 1.18 & 2013 apr 26 & 120 & 60 ; 60 + lehpm 2 - 50 & j14560172 - 2747288 & & l1 & m9.3 & 13.25 & 1.06 & 2006 mar 12 & 120 & 32 ; 75 + lhs 3003 & j1456383 - 280947 & m7 & & m7.7 & 9.970 & 1.04 & 2008 jul 29 & 120 & ; + lehpm 2 - 498 & j14565736 - 2631265 & & m8 & m7.7 & 13.56 & 0.95 & 2006 mar 12 & 120 & 32 ; 75 + pss 1458 + 2839 & j1458245 + 283958 & m8.5 & & m8.5 & 13.08 & 1.23 & 2008 jul 13 & 120 & ; + 2mass j15004572 + 4219448 & j15004572 + 4219448 & m9 & & m8.4 & 13.77 & 1.13 & 2012 may 30 & 75 & ; + tvlm 513 - 46546 & j1501081 + 225002 & m8.5 & & m9.3 & 11.87 & 1.16 & 2008 jul 30 & 120 & ; + 2mass j15101685 - 024107 & j15101685 - 024107 & m9 & & m9.3 & 12.61 & 1.27 & 2009 jun 29 & 120 & ; + sdss j15102955 + 36194699 & j15102955 + 36194699 & m9 & & m8.3 & 14.00 & 1.18 & 2013 aug 14 & 120 & 48 ; 48 + 2mass j15111091 + 4340363 & j15111091 + 4340363 & l5 & & l5.8 & 16.60 & 1.90 & 2013 apr 25 & 120 & 54 ; 54 + sdss j15124067 + 3403501 & j15124067 + 3403501 & l3 & & l0.7 & 15.04 & 1.63 & 2013 apr 25 & 120 & 60 ; 60 + 2mass j15230657 - 2347526 & j15230657 - 2347526 & & l2.5 & l0.3 & 14.20 & 1.30 & 2008 jul 29 & 120 & 47 ; 82 + sdss j153453.33 + 121949.2 & j15345324 + 12194950 & & l4+/-1.5 & l5.5 & 15.33 & 1.51 & 2012 jul 09 & 120 & 40 ; 40 + 2mass j15345700 - 1418480 & j15345700 - 1418480 & m7 & & m8.6 & 11.39 & 1.08 & 2012 may 30 & 75 & ; + denis - p j153941.96 - 052042.4 & j15394189 - 0520428 & l4 : & l2 & l4.2 & 13.92 & 1.35 & 2008 jul 14 & 120 & 31 ; 52 , 31 + lehpm 2 - 287 & j15453990 - 2255167 & & m8 & m8.2 & 13.71 & 1.09 & 2006 apr 09 & 120 & 32 ; 75 + 2mass j15474719 - 2423493 & j15474719 - 2423493 & m9 & & m9.0 & 13.97 & 1.23 & 2008 jul 29 & 120 & 49 ; 49 + sdss j154849.02 + 172235.4 & j15484911 + 17223590 & & l5 & l6.6 & 16.10 & 1.65 & 2013 aug 14 & 120 & 40 ; 40 + 2mass j15485834 - 1636018 & j15485834 - 1636018 & & l2+/-1 & m8.8 & 13.89 & 1.26 & 2010 jul 07 & 120 & 47 ; 47 + sdss j15512086 + 4329303 & j15512086 + 4329303 & l3 & & l3.1 & 15.13 & 1.50 & 2011 apr 19 & 120 & 60 ; 60 + sdss j15525232 - 0035019 & j15525232 - 0035019 & l0 & & m9.6 : & 15.99 & 0.98 & 2011 apr 19 & 120 & 60 ; 60 + 2massw j1552591 + 294849 & j15525906 + 2948485 & l0beta & l1 & l0.5 : & 13.48 & 1.46 & 2008 jul 30 & 120 & 30 ; 56 , 30 + 2massw j1553214 + 210907 & j15532142 + 2109071 & l5.5 & & l6.6 : & 16.70 & 2.02 & 2009 jun 29 & 120 & 11 ; 11 + 2massw j1555157 - 095605 & j15551573 - 0956055 & l1 & & l1.6 & 12.56 & 1.11 & 2008 jul 14 & 120 & 22 ; 22 + sdss j15564435 + 1723089 & j15564435 + 1723089 & l0 & & l0.7 & 14.67 & 1.32 & 2013 aug 14 & 120 & 60 ; 60 + 2mass j15573270 + 1752380 & j15573270 + 1752380 & m7.5 & & m8.3 & 13.54 & 1.09 & 2013 aug 14 & 120 & 44 ; 44 + 2massi j1600054 + 170832 & j16000548 + 1708328 & l1.5 & & l1.8 & 16.05 & 1.37 & 2008 jul 12 & 120 & 13 ; 13 + 2mass j16082460 + 195747 & j16082460 + 195747 & m9 & & m8.6 & 13.52 & 1.17 & 2009 jun 29 & 120 & ; + lsr 1610 - 0040 & j16102900 - 0040530 & sdl : & d / sdm6 & m7.0 : & 12.91 & 0.89 & 2004 mar 12 & 120 & 26 ; 26 , 42 + 2mass j16134550 + 1708270 & j16134550 + 1708270 & m9.5 & & m9.6 & 13.47 & 1.28 & 2009 jun 29 & 120 & ; + 2massw j1615441 + 355900 & j16154416 + 3559005 & l3 & & l3.6 & 14.54 & 1.60 & 2009 jun 30 & 120 & 13 ; 13 + 2mass j16184503 - 1321297 & j16184503 - 1321297 & l0 : & & m9.3 & 14.25 & 1.33 & 2010 jul 07 & 120 & 52 ; 52 + sdss j161928.31 + 005011.9 & j16192830 + 0050118 & l2 & & l1.3 & 14.39 & 1.20 & 2008 jul 30 & 120 & 20 ; 20 + lehpm 2 - 1973 & j16202207 - 2446025 & & m7 & m7.0 & 14.32 & 0.94 & 2006 apr 09 & 120 & 32 ; 75 + gj 618.1b & j16202614 - 0416315 & l2.5 & & l2.4 & 15.28 & 1.69 & 2008 jul 30 & 120 & 18 ; 18 + 2mass j16203450 - 2430200 & j16203456 - 2430205 & m6.5 & & m9.4 : : & 14.20 & 1.66 & 2008 jul 29 & 120 & ; + 2mass j16210822 + 2938480 & j16210822 + 2938480 & m9 & & m9.4 & 15.17 & 0.84 & 2009 jun 30 & 120 & ; + sdss j16260303 + 2113130 & j16260303 + 2113130 & l3 & & l4.7 : & 15.48 & 1.56 & 2013 apr 24 & 120 & 60 ; 60 + 2mass j16301770 - 2120010 & j16301770 - 2120010 & & m9 : & m9.4 & 14.51 & 1.32 & 2008 jul 29 & 120 & 77 ; + & & & & & & & 2008 aug 28 & 120 & 77 ; + 2mass j16304139 + 0938446 & j16304138 + 09384459 & l0 : : & & l0.4 : & 14.87 & 1.57 & 2011 apr 19 & 120 & 49 ; 49 + sdss j16311227 + 32271141 & j16311227 + 32271141 & m7 & & m8.2 & 13.15 & 0.91 & 2013 apr 26 & 120 & 48 ; 48 + sdss j163256.1 + 350507.3 & j1632561 + 3505073 & l1 & & l0.7 & 14.65 & 1.35 & 2012 jul 09 & 120 & 74 ; 74 + sdss j16351918 + 42230531 & j16351918 + 42230531 & m8 & & m8.6 & 12.88 & 1.09 & 2013 apr 26 & 120 & 48 ; 48 + sdssp j163600.79 - 003452.6 & j16360078 - 0034525 & l0 & & m9.0 & 14.59 & 1.18 & 2008 jul 30 & 120 & 12 ; 12 + 2mass j16452207 + 3004071 & j16452207 + 3004071 & l3 & & l2.3 & 15.19 & 1.60 & 2008 jul 13 & 120 & 44 ; 44 + 2mass j16452207 + 3004071 & j16452207 + 30040718 & l3 & & l3.0 & 15.19 & 1.60 & 2012 jul 09 & 120 & 44 ; 44 + 2mass j16490419 + 0444571 & j16490419 + 0444571 & m8 & & m8.1 & 12.96 & 1.08 & 2013 aug 14 & 120 & 49 ; 49 + sdss j16545079 + 3747146 & j16545079 + 3747146 & l2 & & l0.7 & 15.01 & 1.35 & 2013 apr 26 & 120 & 60 ; 60 + vb 8 & j16553529 - 0823401 & m7 v & & m7.0 & 9.780 & 0.96 & 2007 jul 04 & 120 & ; + 2mass j16573454 + 1054233 & j16573454 + 1054233 & l2 & & l1.4 & 14.15 & 1.35 & 2009 jun 29 & 120 & 49 ; 49 + wise j165842.56 + 510335.0 & j16584256 + 5103350 & & l6pec & l5.5 : & 15.06 & 1.40 & 2013 oct 23 & 120 & 69 ; 69 + sdss j16585026 + 1820006 & j16585026 + 1820006 & l0 & & l0.9 & 15.48 & 0.91 & 2010 jul 07 & 120 & ; + sdss j17031670 + 19063603 & j17031670 + 19063603 & l0 & & m8.9 & 14.92 & 1.27 & 2012 jul 09 & 120 & 48 ; 48 + 2mass j17072343 - 0558249b & j17072343 - 05582489 & & l3 & l0.0 & 12.26 & 1.23 & 2006 aug 28 & 120 & 41 ; 41 + denis j1707252 - 013809 & j17072529 - 0138093 & l0.5 & l2 & l1.0 & 14.29 & 1.22 & 2006 aug 28 & 120 & 59 ; 59 , 64 + 2massi j1707333 + 430130 & j17073334 + 4301304 & l0.5 & & m8.7 & 13.97 & 1.35 & 2008 sep 08 & 120 & 27 ; 27 + sdss j17104934 + 33232518 & j17104934 + 33232518 & l0 & & m9.4 & 15.13 & 1.05 & 2011 apr 19 & 120 & 48 ; 48 + 2mass j17111353 + 2326333 & j17111353 + 2326333 & l0 : & & l0.5 : & 14.50 & 1.44 & 2008 jul 13 & 120 & 44 ; 44 + g 203 - 50 b & j17114530 + 4029021 & & l5 + 2 - 1.5 & l5.0 & 15.00 & 1.20 & 2008 sep 09 & 120 & 86 ; 86 + & & & & & & & 2010 jul 07 & 120 & 86 ; 86 + 2mass j10511900 + 5613086 & j17170450 + 1509530 & m7 & & m7.9 & 13.61 & 1.11 & 2012 may 30 & 75 & ; + sdss j171714.10 + 652622.2 & j17171408 + 6526221 & l4 & & l6.2 & 14.95 & 1.77 & 2009 jun 30 & 120 & 20 ; 20 + sdss j17175402 + 64274503 & j17175402 + 64274503 & m8 & & m8.8 & 14.41 & 1.03 & 2013 oct 22 & 120 & 48 ; 48 + 2massi j1721039 + 334415 & j17210390 + 3344160 & l3 & l5+/-1 & l5.3 : & 13.63 & 1.14 & 2008 sep 08 & 120 & 27 ; 27 , 50 + sdss j172244.32 + 632946.8 & j17224432 + 6329470 & l0 & & l0.7 & 15.37 & 1.29 & 2009 jun 30 & 120 & 20 ; 20 + sdss j17254384 + 5325349 & j17254384 + 5325349 & l1 & & m8.0 & 15.16 & 0.95 & 2010 jul 07 & 120 & ; + 2massi j1726000 + 153819 & j17260007 + 1538190 & l3beta & & l3.6 : : & 15.67 & 2.01 & 2008 may 11 & 120 & 13 ; 56 + & & & & & & & 2008 jul 13 & 120 & 13 ; 56 + 2mass j17281134 + 0839590 & j17281134 + 0839590 & m9 pec & & m9.8 & 13.63 & 1.13 & 2009 jun 29 & 120 & ; + denis - p j1733423 - 165449 & j17334227 - 1654500 & l0.5+/-1 & & l0.9 & 13.53 & 1.18 & 2006 apr 11 & 120 & 53 ; 53 + 2massw j1743415 + 212707 & j17434148 + 2127069 & l2.5 & & l2.4 & 15.83 & 1.51 & 2009 jun 30 & 120 & 13 ; 13 + denis - p j1745346 - 164053 & j17453466 - 1640538 & l1.5+/-1 & & l1.3 & 13.65 & 1.24 & 2008 sep 08 & 120 & 53 ; 53 + 2mass j17461199 + 5034036 & j17461199 + 50340362 & l5 & & l5.7 & 15.10 & 1.57 & 2012 jul 09 & 120 & 49 ; 49 + 2mass j18000116 - 1559235 & j18000116 - 1559235 & l5.5 & & l4.3 & 13.30 & & 2012 jul 09 & 120 & 68 ; 68 + 2mass j18064570 + 2923591 & j18064570 + 2923591 & m8 : & & m8.3 & 14.20 & 1.26 & 2010 jul 07 & 120 & ; + lsr 1826 + 3014 & j18261131 + 3014201 & m8.5 & d / sdm8.5 & m9.9 & 11.66 & 0.85 & 2004 sep 09 & 120 & 23 ; 23 , 75 + sdss j183929.17 + 442438 & j18392917 + 442438 & m9 & & m8.9 & 13.43 & 1.08 & 2008 nov 04 & 120 & 48 ; 48 + denis j19013910 - 3700170 & j19013910 - 3700170 & m8 & & l0.1 : : & 14.26 & 1.96 & 2013 apr 24 & 120 & 59 ; 59 + denis - p j1909081 - 193748 & j19090821 - 1937479 & l1+/-1 & & l1.4 & 14.52 & 1.61 & 2008 sep 08 & 120 & 53 ; 53 + vb 10 & j19165762 + 0509021 & m8 v & & m8.0 & 9.910 & 1.14 & 2007 jul 04 & 120 & 90 ; 73 + 2mass j19233810 - 3308410 & j19233810 - 3308410 & m7 & & m8.5 & 13.27 & 1.04 & 2012 sep 27 & 120 & ; + denis jj1934511 - 184134 & j1934511 - 184134 & m8.5 & & m8.5 & 14.28 & 1.15 & 2013 oct 23 & 120 & 59 ; 59 + lehpm 2 - 90 & j19453495 - 2557190 & & m9 & m8.9 & 12.35 & 0.84 & 2006 sep 02 & 120 & 32 ; 75 + 2mass j19561542 - 1754252 & j19561542 - 1754252 & m8 & l0+/-1 & m8.5 & 13.75 & 1.10 & 2007 sep 16 & 120 & 47 ; 49 , 82 + 2mass j20025073 - 0521524 & j200250732 - 05215242 & l6 & & l6.8 : : & 15.32 & 1.90 & 2007 jul 04 & 120 & 44 ; 44 + denis j2013108 - 124244 & j2013108 - 124244 & l1.5 & & l0.7 & 14.52 & 1.21 & 2013 oct 22 & 120 & 59 ; 59 + 2mass j20263647 + 0439400 & j20263647 + 0439400 & m9 : & & m8.5 & 14.16 & 1.23 & 2009 jun 28 & 120 & ; + 2mass j2035203 - 311008 & j2035203 - 311008 & m7 & & m7.9 & 13.19 & 1.03 & 2010 jul 07 & 120 & ; + 2mass j20414283 - 3506442 & j20414283 - 3506442 & l2 : & & l1.9 & 14.89 & 1.49 & 2008 sep 08 & 120 & 44 ; 44 + sdss j204724.7 + 142152 & j2047247 + 142152 & m7.5 & & m8.9 & 13.04 & 1.16 & 2008 nov 03 & 120 & 48 ; 48 + lehpm 2 - 381 & j20522811 - 4758362 & m8 & m8 & m8.1 & 12.94 & 1.06 & 2004 sep 06 & 120 & 32 ; 75 + 2massi j2054358 + 151904 & j20543585 + 1519043 & l1 : & & l0.3 & 16.37 & 1.39 & 2009 jun 28 & 120 & 13 ; 13 + 2massi j2057153 + 171515 & j20571538 + 1715154 & l1.5 & & m9.9 & 15.97 & 1.47 & 2009 jun 28 & 120 & 13 ; 13 + 2mass j20575592 - 0050060 & j20575592 - 0050060 & m9 & & m9.3 & 14.97 & 1.20 & 2009 jun 29 & 120 & ; + 2mass j21075409 - 4544064 & j21075409 - 45440639 & l0 : & & l2.6 & 14.92 & 1.53 & 2013 aug 14 & 120 & 49 ; 49 + hb 2115 - 4518 & j2118317 - 450552 & m8.5 & & m8.6 & 13.43 & 1.06 & 2008 jul 14 & 120 & ; + sdss j211846.77 - 001044.6 & j21184677 - 00104469 & l1 & & l2.2 : & 16.20 & 1.13 & 2012 sep 27 & 120 & 54 ; 54 + sdss j212033.89 + 102159 & j21203389 + 102159 & m8 & & m9.0 & 13.54 & 1.12 & 2008 nov 03 & 120 & 48 ; 48 + 2mass j21233110 - 2345180 & j21233110 - 2345180 & m7.5 & & m8.2 & 13.58 & 1.04 & 2013 jul 17 & 75 & 44 ; 44 + 2mass j21263403 - 3143220 & j21263403 - 3143220 & m9 & & m9.2 & 13.47 & 1.10 & 2009 jun 28 & 120 & ; + hb 2124 - 4228 & j2127261 - 421518 & m7.5 & & m8.9 & 13.32 & 1.14 & 2008 jul 14 & 120 & ; + hb 2126 - 4459 & j2130086 - 444627 & m8.5 & & m8.2 & 14.32 & 1.15 & 2008 jul 14 & 120 & ; + 2massw j2130446 - 084520 & j21304463 - 08452049 & l1.5 & & m8.3 & 14.14 & 1.32 & 2006 sep 01 & 120 & 52 ; 52 + & & & & & & & 2006 sep 02 & 120 & 52 ; 52 + & & & & & & & 2008 jul 13 & 120 & 52 ; 52 + & & & & & & & 2008 sep 09 & 120 & 52 ; 52 + & & & & & & & 2012 sep 27 & 120 & 52 ; 52 + & & & & & & & 2012 oct 27 & 120 & 52 ; 52 + sdss j213240.36 + 102949.4 & j21324035 + 10294940 & & l4.5+/-1 & l4.8 : : & 16.59 & 1.96 & 2012 oct 27 & 120 & 40 ; 40 + sdss j213307.94 + 232159 & j21330794 + 232159 & m9.5 & & m9.6 & 13.74 & 1.18 & 2008 nov 04 & 120 & 48 ; 48 + sdss j213435.61 + 240408 & j21343561 + 240408 & m8 & & m9.2 & 13.57 & 1.17 & 2008 nov 04 & 120 & 48 ; 48 + 2mass j21371044 + 1450475 & j21371044 + 1450475 & l2 & & l0.8 & 14.13 & 1.32 & 2008 sep 08 & 120 & 49 ; 49 + denis j21391360 - 3529500 & j21391360 - 3529500 & l0 & & l2.1 : & 14.47 & 1.11 & 2013 aug 14 & 120 & 59 ; 59 + sdss j214046.55 + 011259.7 & j21404654 + 0112594 & l3 & & l4.4 & 15.89 & 1.47 & 2008 sep 08 & 120 & 20 ; 20 + 2mass j21420580 - 3101162 & j21420580 - 3101162 & l3 & & l2.9 & 15.84 & 1.88 & 2010 jul 07 & 120 & 43 ; 43 + sdss j214527.82 - 073434.2 & j21452782 - 0734342 & m9 & & m8.7 & 15.59 & 1.28 & 2009 nov 04 & 120 & 48 ; 48 + 2mass j21472764 + 0101040 & j21472764 + 0101040 & m9 & & m9.1 & 14.57 & 1.10 & 2009 jun 28 & 120 & ; + 2mass j21481628 + 4003593 & j21481633 + 4003594 & l6 & l6.5pec & l4.0 & 14.15 & 2.38 & 2005 sep 09 & 120 & 51 ; 51 + 2mass j21483083 + 0020540 & j21483083 + 0020540 & m9 & & m8.5 & 15.46 & 1.15 & 2009 jun 28 & 120 & ; + sdss j214956.55 + 060334 & j21495655 + 060334 & m9 & & m8.1 & 13.34 & 1.17 & 2008 nov 03 & 120 & 48 ; 48 + sdss j215339.77 + 295005 & j21533977 + 295005 & m9 & & m8.8 & 13.94 & 1.17 & 2008 nov 03 & 120 & 48 ; 48 + 2mass j21580457 - 1550098 & j21580456 - 15500980 & l4 : & & l4.3 : & 15.04 & 1.85 & 2006 sep 01 & 120 & 52 ; 52 + 2mass j22044198 - 0036510 & j22044198 - 0036510 & m9 & & m8.5 & 15.73 & 1.24 & 2009 jun 29 & 120 & ; + 2massw j2206450 - 421721 & j22064498 - 4217208 & l2 & & l4.2 : : & 15.56 & 1.95 & 2008 jul 14 & 120 & 13 ; 13 + grh 2208 - 2007 & j2210499 - 195224 & m7.5 & & m8.8 & 14.00 & 0.85 & 2008 jul 14 & 120 & ; + 2mass j22114470 + 6856262 & j22114470 + 6856262 & & l2 & l2.2 & 15.67 & 1.65 & 2005 sep 08 & 120 & 61 ; 61 + 2mass j22134491 - 2136079 & j22134491 - 21360789 & l0gamma & & l1.9 : & 15.38 & 1.62 & 2006 sep 02 & 120 & 44 ; 56 + & & & & & & & 2008 aug 29 & 120 & 44 ; 56 + wise j222219.93 + 302601.4 & j22221993 + 3026014 & & l9 & l6.7 : & 16.55 & 1.37 & 2013 oct 23 & 120 & 69 ; 69 + lhs 523 & j22285440 - 1325178 & m6.5 & & m7.7 & 10.77 & 0.92 & 2007 sep 16 & 120 & ; + 2maspj2234330 + 291850 & j2234331 + 291849 & m8 : & & m7.9 & 14.04 & 1.08 & 2008 jul 12 & 120 & ; + 2mass j22355013 + 1227370 & j22355013 + 1227370 & m9 & & m9.2 & 15.21 & 1.17 & 2009 jun 30 & 120 & ; + 2massi j2238074 + 435317 & j22380742 + 4353179 & l1.5 & & l0.6 & 13.84 & 1.32 & 2009 nov 07 & 120 & 27 ; 27 + & & & & & & & 2010 jul 07 & 120 & 27 ; 27 + sdss j22434553 - 08215302 & j22434553 - 08215302 & m8 & & m8.4 & 15.43 & 1.14 & 2013 jul 17 & 75 & 48 ; 48 + sdss j225003.72 + 143046.7 & j22500372 + 1430467 & m9 & & m8.7 & 14.94 & 1.22 & 2009 nov 04 & 120 & 48 ; 48 + & & & & & & & 2009 nov 07 & 120 & 48 ; 48 + 2massi j2254519 - 284025 & j22545194 - 2840253 & l0.5 & l0.5 & l0.2 & 14.13 & 1.18 & 2008 jul 14 & 120 & 27 ; 27 , 31 + sdssp j225529.09 - 003433.4 & j22552907 - 0034336 & l0 : & & m8.6 & 15.65 & 1.21 & 2008 sep 07 & 120 & 19 ; 19 + ulas j22585405 + 0113512 & j22585405 + 0113512 & m9 & & m8.8 & 13.91 & 1.00 & 2013 aug 14 & 120 & 62 ; 62 + sdss j225913.88 - 005158.2 & j22591388 - 0051581 & l2 & & l1.6 & 16.36 & 1.71 & 2009 jun 29 & 120 & 20 ; 20 + sdss j230809.9 - 313122 & j2308099 - 313122 & m7 & & m7.5 & 13.62 & 1.04 & 2008 nov 04 & 120 & 48 ; 48 + denis j2308113 - 272200 & j2308113 - 272200 & l1.5 & & l1.4 & 14.58 & 1.40 & 2013 sep 03 & 120 & 59 ; 59 + ssspm j2310 - 1759 & j23101846 - 1759090 & l0 : & l1 & m9.9 & 14.38 & 1.41 & 2008 jul 14 & 120 & 83 ; 44 , 37 + sdss j23172515 - 00543358 & j23172515 - 00543358 & m9 & & m8.7 & 15.68 & 0.94 & 2012 sep 27 & 120 & 48 ; 48 + 2mass j2320292 + 412341 & j2320292 + 412341 & l1 : & & m9.8 & 14.59 & 1.39 & 2003 aug 12 & 75 & 76 ; + & & & & & & & 2008 sep 08 & 120 & ; + 2mass j23211254 - 1326282 & j23211254 - 1326282 & & l1 & l0.4 & 14.50 & 1.36 & 2003 sep 04 & 75 & 47 ; 47 + & & & & & & & 2008 jul 12 & 120 & 47 ; 47 + & & & & & & & 2008 sep 08 & 120 & 47 ; 47 + sdss j232136.11 - 002819.1 & j23213611 - 0028191 & m9 & & m8.3 & 15.48 & 1.04 & 2009 nov 04 & 120 & 48 ; 48 + sdss j232246.84 - 313323 & j23224684 - 313323 & l0 : & & l1.6 : & 13.58 & 1.25 & 2006 aug 28 & 120 & 48 ; 48 + & & & & & & & 2008 nov 04 & 120 & 48 ; 48 + sdss j232313.4 - 024435 & j2323134 - 024435 & m8.5 & & m8.0 & 13.58 & 1.10 & 2008 nov 03 & 120 & 48 ; 48 + 2mass j23302258 - 0347189 & j23302258 - 0347189 & l1 : & & l0.5 & 14.48 & 1.35 & 2008 jul 14 & 120 & 44 ; 44 + hd 221356bc & j23310217 - 0406248 & m9+l3 & & m8.7 & 12.94 & 0.98 & 2013 dec 05 & 120 & 14 ; 46 + lspm j2331 + 4607n & j23311807 + 4607310 & & d / sdm7 & m7.4 & 15.92 & 0.74 & 2005 sep 07 & 120 & ; 75 + 2mass j23312935 + 1552220 & j23312935 + 1552220 & l0 & & m9.5 & 15.06 & 1.06 & 2009 jun 30 & 120 & ; + sdss j233224.38 - 005025 & j23322438 - 005025 & m8 & & m7.7 & 13.65 & 1.04 & 2008 nov 03 & 120 & 48 ; 48 + sdss j233350.76 - 000011.3 & j23335076 - 0000113 & m9 & & m8.6 & 15.52 & 0.98 & 2009 nov 07 & 120 & 48 ; 48 + 2mass j23335838 + 0050110 & j23335838 + 0050110 & m9 & & m9.3 & 15.01 & 1.17 & 2009 jul 01 & 120 & ; + sdss j23352642 + 0817213 & j23352642 + 0817213 & l0 & & m9.5 & 14.72 & 1.34 & 2012 oct 24 & 120 & 60 ; 60 + sdss j23371664 - 09332480 & j23371664 - 09332480 & m8 & & m7.6 & 13.41 & 1.13 & 2013 aug 14 & 120 & 48 ; 48 + 2mass j2341286 - 113335 & j2341286 - 113335 & m8 & & m7.9 & 13.55 & 1.00 & 2010 jul 07 & 120 & ; + 2mass j23440624 - 0733282 & j23440624 - 0733282 & l4.5 & & l6.0 & 14.80 & 1.57 & 2009 jun 29 & 120 & 52 ; 52 + 2mass j23453903 + 0055137 & j23453903 + 0055137 & m9 & & m9.2 & 13.77 & 1.19 & 2008 jul 14 & 120 & 49 ; 49 + sdss j234654.7 - 315353 & j2346547 - 315353 & m8 & & m8.7 & 13.28 & 1.08 & 2008 nov 04 & 120 & 48 ; 48 + 2mass j2352050 - 110043 & j2352050 - 110043 & m7 & & m8.0 & 12.84 & 1.10 & 2010 jul 07 & 120 & ; + denis - p j2353 - 0833 & j2353594 - 083331 & m8.5 & & m8.6 & 13.03 & 1.10 & 2010 jul 07 & 120 & ; + denis j23545990 - 1852210 & j23545990 - 1852210 & l2 & & l2.6 : & 14.22 & 1.39 & 2013 aug 14 & 120 & 59 ; 59 + ssspm j2356 - 3426 & j23561081 - 3426044 & m9.0 & l0.5 & m9.3 & 12.95 & 0.98 & 2007 sep 16 & 120 & 37 ; 37 + ssspm j2400 - 2008 & j23595762 - 2007394 & m9.5 & l1 & m8.9 & 14.38 & 1.13 & 2008 jul 14 & 120 & 37 ; 37 + ccccc h@xmath25o-@xmath87 & 1.14 - 1.165 & 1.26 - 1.285 & 1.15 @xmath21 m h@xmath25o & 1 + ch@xmath22-@xmath87 & 1.315 - 1.335 & 1.26 - 1.285 & 1.32 @xmath21 m ch@xmath22 & 1 + h@xmath25o-@xmath88 & 1.48 - 1.52 & 1.56 - 1.60 & 1.40 @xmath21 m h@xmath25o & 1 + ch@xmath22-@xmath88 & 1.635 - 1.675 & 1.56 - 1.60 & 1.65 @xmath21 m ch@xmath22 & 1 + h@xmath25o-@xmath89 & 1.975 - 1.995 & 2.08 - 2.10 & 1.90 @xmath21 m h@xmath25o & 1 + ch@xmath22-@xmath89 & 2.215 - 2.255 & 2.08 - 2.12 & 2.20 @xmath21 m ch@xmath22 & 1 + @xmath90 & 2.06 - 2.10 & 1.25 - 1.29 & @xmath91 color & 1 + @xmath88-dip & 1.61 - 1.64 & 1.56 - 1.59 + 1.66 - 1.69 & 1.63 @xmath21 m feh / ch@xmath22 & 2 + @xmath89-slope & 2.06 - 2.10 & 2.10 - 2.14 & @xmath89-band shape / cia h@xmath25 & 3 + @xmath87-slope & 1.27 - 1.30 & 1.30 - 1.33 & 1.28 @xmath21 m flux peak shape & 4 + @xmath87-curve & 1.04 - 1.07 + 1.26 - 1.29 & 1.14 - 1.17 & curvature across @xmath87-band & 4 + @xmath88-bump & 1.54 - 1.57 & 1.66 - 1.69 & slope across @xmath88-band peak & 4 + h@xmath25o-@xmath92 & 1.04 - 1.07 & 1.14 - 1.17 & 1.15 @xmath21 m h@xmath25o & 4 + cl spt vs. ch@xmath22-_h _ & best fit curve : @xmath93 , @xmath28 = 0.0354 . + & select points below the @xmath94 curve . + h@xmath25o-_j _ vs. ch@xmath22-_h _ & intersection of : @xmath95 and @xmath96 . + & select points on lower left corner . + h@xmath25o-_j _ vs. _ h_-bump & intersection of : @xmath97 and @xmath96 . + & select points on upper left corner . + ch@xmath22-_j _ vs. ch@xmath22-_h _ & intersection of : @xmath98 and @xmath99 . + & select points on lower left corner . + ch@xmath22-_j _ vs. _ h_-bump & intersection of : @xmath100 , @xmath101 , and @xmath102 . + & select points on upper left corner . + ch@xmath22-_h _ vs. _ j_-slope & intersection of : @xmath103 , @xmath104 , and @xmath105 . + & select points on upper left corner . + ch@xmath22-_h _ vs. _ j_-curve & best fit curve : @xmath106 , @xmath28 = 0.088 . + & select points above the @xmath107 curve , up to ch@xmath22-_h _ = 1.03 . + ch@xmath22-_h _ vs. _ h_-bump & best fit curve : @xmath108 , @xmath28 = 0.013 . + & select points below the @xmath94 curve , down to _ h_-bump = 0.92 . j_-slope vs. _ h_-dip & intersection of @xmath109 and @xmath104 . + & select points on lower right corner . + _ j_-slope vs. _ h_-bump & intersection of : @xmath110 and @xmath102 . + & select points on upper right corner . + _ k_-slope vs. h@xmath25o-_y _ & best fit curve : @xmath111 , @xmath28 = 0.064 . + & select points above the @xmath107 curve and between _ k_-slope = 0.93 - 0.96 . + _ j_-curve vs. _ h_-bump & best fit curve : @xmath112 , @xmath28 = 0.048 . + & select points above the @xmath107 and greater than _ j_-curve = 2.00 and _ h_-bump = 0.92 . + lccrcccl 2mass j02361794 + 0048548 & l5.0@xmath480.6 & t1.9@xmath481.1 & @xmath113 & 1.05@xmath480.48 & 1.80@xmath480.55 & 1 & + sdss j093113.23 + 280227.1 & l1.4@xmath480.1 & t2.6@xmath480.9 & @xmath114 & 2.22@xmath480.23 & 2.74@xmath480.25 & 1 & v , b + 2mass j13114227 + 3629235 & l4.8@xmath480.6 & t4.1@xmath482.7 & @xmath114 & 2.19@xmath481.02 & 3.14@xmath481.33 & 5 & v + 2mass j13411160 - 30525049 & l1.2@xmath480.3 & t6.3@xmath481.0 & @xmath115 & 3.28@xmath480.53 & 4.82@xmath480.63 & 1 & v + 2mass j14532589 + 1420418 & l1.1@xmath480.0 & t6.0@xmath481.1 & @xmath114 & 3.27@xmath480.46 & 4.42@xmath480.63 & 1 & v , ( b ) + 2mass j20261584 - 2943124 & l0.1@xmath480.5 & t5.8@xmath481.0 & @xmath114 & 3.42@xmath480.40 & 4.82@xmath480.57 & 4 & v + 2mass j02060879 + 22355930 & l5.1@xmath480.5 & t3.2@xmath482.3 & @xmath116 & 1.61@xmath480.89 & 2.36@xmath481.25 & 1 & + denis - p j04272708 - 1127143 & m7.4@xmath480.2 & t5.1@xmath481.5 & @xmath117 & 4.13@xmath480.62 & 4.98@xmath480.84 & 1 & + 2mass j10365305 - 3441380 & l5.2@xmath480.4 & t1.4@xmath480.4 & @xmath114 & 0.51@xmath480.32 & 1.41@xmath480.24 & 1 & + 2mass j10595138 - 2113082 & l0.6@xmath480.4 & t3.4@xmath481.3 & @xmath114 & 2.58@xmath480.32 & 3.30@xmath480.64 & 1 & v + sdss j142227.20 + 221557.5 & l4.2@xmath480.6 & t4.1@xmath482.3 & @xmath118 & 2.36@xmath480.78 & 3.22@xmath481.18 & 1 & b + wise j16235970 - 0508114 & l0.6@xmath480.3 & t6.0@xmath480.8 & @xmath114 & 3.39@xmath480.40 & 4.80@xmath480.53 & 1 & + 2mass j17072529 - 0138093 & l0.7@xmath480.5 & t4.3@xmath482.0 & @xmath113 & 2.87@xmath480.75 & 3.75@xmath481.01 & 1 & + 2massi j1711457 + 223204 & l1.5@xmath480.6 & t2.5@xmath481.0 & @xmath114 & 1.20@xmath480.40 & 3.08@xmath480.64 & 6 & v + 2mass j11181292 - 0856106 & l1.4@xmath480.7 & t2.3@xmath482.3 & @xmath119 & 1.50@xmath480.82 & 2.42@xmath481.11 & & b + sdss j141624.09 + 134826.7 & l4.4@xmath481.1 & t3.9@xmath481.4 & @xmath114 & 2.12@xmath480.42 & 2.99@xmath480.67 & & b + 2mass j15150083 + 4847416 & l5.0@xmath480.6 & t2.7@xmath481.9 & @xmath119 & 1.32@xmath480.61 & 2.17@xmath480.84 & 1 & b + 2mass j17114559 + 4028578 & l4.4@xmath480.3 & t2.7@xmath480.8 & @xmath114 & 1.60@xmath480.18 & 2.32@xmath480.33 & & b + 2mass j03205965 + 1854233 & m7.8@xmath480.1 & t6.0@xmath481.5 & @xmath120 & 4.43@xmath480.73 & 5.47@xmath480.92 & & v + 2mass j03264453 + 1919309 & m8.5@xmath480.0 & t6.7@xmath480.8 & @xmath121 & 4.66@xmath480.50 & 5.91@xmath480.58 & & v + 2mass j03303847 - 2348463 & m7.7@xmath480.3 & t5.6@xmath481.6 & @xmath122 & 3.97@xmath480.85 & 4.58@xmath481.05 & & + 2mass j03301720 + 3505001 & m7.7@xmath480.5 & t5.3@xmath481.7 & @xmath123 & 4.27@xmath480.81 & 5.08@xmath481.03 & & + 2mass j03440891 + 0111249 & l0.6@xmath480.5 & t4.8@xmath482.1 & @xmath124 & 3.34@xmath480.88 & 4.14@xmath481.19 & & + 2mass j04024315 + 1730136 & m7.5@xmath480.2 & t5.4@xmath481.7 & @xmath125 & 4.59@xmath480.78 & 5.29@xmath481.01 & & + 2mass j04430581 - 3202090 & l4.5@xmath480.3 & t1.7@xmath481.0 & @xmath126 & 1.41@xmath480.34 & 2.05@xmath480.39 & 1 & b + 2mass j08433328 + 1024435 & l0.9@xmath480.3 & t4.9@xmath482.0 & @xmath127 & 3.01@xmath480.75 & 3.99@xmath481.03 & & + 2mass j08475148 + 0138110 & l2.0@xmath480.7 & t5.7@xmath482.1 & @xmath128 & 3.19@xmath480.74 & 4.61@xmath481.07 & & + 2mass j14232186 + 6154005 & l1.9@xmath480.8 & t4.6@xmath481.9 & @xmath129 & 2.63@xmath480.67 & 3.78@xmath480.97 & 3 & v + 2mass j14493784 + 2355378 & m9.4@xmath480.3 & t6.2@xmath481.7 & @xmath130 & 4.22@xmath480.69 & 5.39@xmath481.00 & 2 & v + 2mass j15412408 + 5425598 & m7.6@xmath480.3 & t5.4@xmath481.6 & @xmath123 & 4.53@xmath480.75 & 5.22@xmath480.99 & & + 2mass j16403561 + 2922225 & m8.1@xmath480.7 & t5.3@xmath481.7 & @xmath130 & 4.38@xmath480.77 & 5.09@xmath481.00 & & + 2mass j17175402 + 64274503 & m8.5@xmath480.1 & t4.6@xmath481.8 & @xmath131 & 4.18@xmath480.70 & 4.93@xmath481.02 & & + 2mass j19064847 + 4011068 & l0.0@xmath480.4 & t5.9@xmath481.8 & @xmath132 & 3.68@xmath480.75 & 4.85@xmath481.01 & & + 2mass j20472471 + 1421526 & m8.4@xmath480.2 & t5.7@xmath481.6 & @xmath133 & 4.10@xmath480.70 & 5.13@xmath480.95 & & v + 2mass j23311807 + 4607310 & m7.5@xmath480.0 & t5.7@xmath481.5 & @xmath134 & 4.60@xmath480.75 & 5.30@xmath480.95 & & + Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: an empirical tenet in the study of accreting compact objects is that their maximum radiative luminosity is about eddington luminosity , that is the limit where the inward - directed gravitational force is balanced by the outward - directed radiation force . for spherical accretion of fully ionized hydrogen , the eddington limit can be written as @xcite @xmath5 where @xmath6 is the thomson scattering cross section , @xmath7 is the proton mass , and @xmath8 is the black hole ( bh ) mass . the x - ray luminosity is generally used as a good proxy for the bolometric luminosity for accreting neutron stars and stellar mass bhs . the eddington argument proved very valuable in the early stages of x - ray astrophysics , to show that most galactic x - ray binaries are neutron stars @xcite . later on , dynamical mass measurements confirmed the existence of a population of galactic bhs with masses @xmath9@xmath10 ( * ? ? ? * for a review ) : they also satisfy the eddington limit , with peak luminosities @xmath11 erg s@xmath12 . the _ einstein _ satellite revealed for the first time a small number of non - nuclear x - ray sources in nearby galaxies ( particularly , in star - forming galaxies ) with apparent luminosities @xmath13 @xcite . however , it was not immediately clear whether these sources could be a distinct class from galactic stellar mass bh candidates , which were just being discovered around the same years . _ s relatively crude spatial resolution allowed for the possibility of source confusion in compact star - forming regions ; lack of long - term monitoring made it hard to distinguish between persistently luminous sources and transient events , _ e.g. _ , young supernovae ( sne ) missed by optical observers . in the following decade , as nearby galaxies were repeatedly observed with x - ray satellites such as _ rosat _ and _ asca _ , with superior spectral coverage and resolution , it became clear that some of those non - nuclear sources could not be sne , and had apparent luminosities well above the eddington limit of typical galactic stellar mass bhs . for example , the `` most likely interpretation '' of the brightest non - nuclear source in the spiral galaxy ngc1365 was `` an ultra - powerful x - ray binary , with either a highly super - eddington low - mass black hole or a very massive black hole '' with @xmath14@xmath15 , `` which may pose a challenge for stellar evolution models '' @xcite ; a third scenario , beamed emission , was also proposed for other sources @xcite . thirteen years later , distinguishing between those three alternative scenarios remains the unsolved fundamental question for this class of sources , despite the wealth of new discoveries from _ chandra _ , _ xmm - newton _ , _ suzaku _ and _ swift _ , intensive modeling of their x - ray spectral and timing properties , and integrating studies of their multiband counterparts . for a while , different names have been in use to identify this class of accreting bhs : extraluminous x - ray binaries @xcite , superluminous x - ray sources @xcite , intermediate - luminosity x - ray objects @xcite . in recent years , the community consensus has settled on ultraluminous x - ray sources ( ulxs ) , adopting the terminology first used by japanese _ asca _ teams @xcite . a handful ulxs are observed to exhibit strong x - ray variability at time scales as short as minutes @xcite , and many others have shown random variation at long time scales from days , weeks to years @xcite , confirming their compact nature . they are 10100 times more luminous than galactic black hole binaries ( bhbs ) , whose peak luminosity is about @xmath16 . as the eddington limit is proportional to the bh mass , this suggests that ulxs may contain bhs at least 10100 times more massive than galactic ones which are typically of 10 @xmath1 , and thus represents a new class of bhs with masses in the range of @xmath3@xmath4 @xmath1 , intermediate between stellar mass and supermassive bhs . however , some models suggest that the eddington limit could be violated by a factor up to @xmath1710 for stellar mass bhs @xcite , and the x - ray luminosity may be overestimated due to anisotropic emission caused by geometric beaming expected at high accretion rate @xcite . in these circumstances , intermediate mass black holes ( imbhs ) are not required to interpret ulxs . a combination of super - eddington and mildly beamed emission from stellar mass bhs could account for ulxs with an apparent luminosity up to @xmath17@xmath18 @xcite . some objects other than bhbs may appear like ulxs if their timing or multiwavelength properties are unavailable . young sne , especially of type iin , could be as luminous as @xmath19 in the x - ray band @xcite . their x - ray spectrum is usually thermal and soft , and will not show chaotic variability . young x - ray pulsars could theoretically be more luminous than @xmath16 but the number of such sources appears to be very small @xcite . in general , these objects can be distinguished from accreting sources with variability @xcite , but it is possible that a sn leaves behind a compact remnant which becomes x - ray luminous when a fallback disk forms @xcite and mixed properties may be seen . in a survey of ulxs in nearby galaxies with archival _ xmm - newton _ data , it was found that the x - ray spectra of a few sources can be decomposed into a power - law component plus a thermal plasma , and in one of them x - ray variability was detected , suggesting that they may be recent sne with an accreting compact core @xcite . background active galactic nuclei ( agn ) that happen to lie behind a nearby galaxy could be misidentified as ulxs in the galaxy . their cosmological location could be revealed if redshift is measured from optical lines ( _ e.g. _ , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the fraction of background objects is usually high in massive , elliptical galaxies ( _ e.g. _ , * ? ? ? * ) . many ulxs are found to be spatially associated with nebulae , young star clusters , or globular clusters ; these are most likely located in the galaxy rather than foreground or background objects . even if they are identified as bhbs , their nature is still not unique in terms of the companion star ( low mass or high mass ) , the mode of the mass transfer ( roche - lobe overflow or wind - fed ) , or the mass of the accretor ( stellar mass or intermediate mass ) . for example , @xcite found that ulxs in young , star forming galaxies are much more variable than in old , elliptical galaxies ( associated with globular clusters ) . no matter what their nature is , ulxs are of great interest in astrophysics . if they are imbhs , their formation mechanism could be rather different from typical stellar mass bhs ; extreme environment , more massive and/or less abundant progenitors are likely required @xcite . imbhs could have played an important role in the formation of supermassive bhs in the early universe and be important targets to search for gravitational wave emission @xcite . instead , if they are stellar mass bhs , it is interesting to investigate why they are so distinct from their galactic cousins . they may be good candidates for the study of the accretion physics at near or super eddington accretion rates @xcite , and they may shed light on binary stellar evolution channels not seen in our galaxy . ulxs are usually defined as non - nuclear , point - like objects which at least once have been observed at an apparent isotropic x - ray luminosity higher than that of stellar mass galactic bhs : typically , @xmath20 erg s@xmath12 in the 0.310 kev band . some authors ( _ e.g. _ * ? ? ? * ) prefer to use @xmath21 ergs s@xmath12 , that is the eddington limit of the heaviest stellar bh ( @xmath22 ) expected at normal metallicity @xcite , to exclude a large number of `` faint '' ulxs that should be similar to normal galactic bhbs . such an empirical definition may have ambiguities and also include young x - ray pulsars , sne , and sn remnants as mentioned above . the consensus in the ulx community is to restrict the definition of ulxs to accreting bhs . therefore , we define ulxs as _ non - nuclear accreting bhs with peak luminosities inferred assuming isotropic emission above the eddington limit of normal stellar mass bhs_. this definition is confined to compact objects that are most likely accreting from a companion star in a binary system , but more exotic possibilities , for example isolated imbhs accreting via the bondi process from the surrounding medium are not excluded a priori . observationaly , variability information is required to define a ulx . recently , many ulxs are found to have a mechanical power output as high as in radiation @xcite . in one of them , the mechanical output is dominant while the radiative output is way below the ulx threshold @xcite . this source is apparently not a ulx , but speculated to be an offset ulx or quiescent ulx ( see section [ sec : bub ] for details ) . we thus define them as `` ultrapowerful sources ( upss ) '' for completeness , to standard for the same population of sources perhaps in a different state or viewing angle . a _ rosat_/hri catalog of luminous , non - nuclear x - ray sources by @xcite proved to be very important for bringing ulxs to the forefront of research , and for emphasizing the possible observational link with theoretically - predicted intermediate mass bhs , with masses @xmath23 . larger and more systematic x - ray source catalogs based on re - analysis of archival _ rosat_/hri data soon followed : in particular , by @xcite , by @xcite , and by @xcite , which included 106 sources above @xmath16 erg s@xmath12 . the first xmm - newton catalog was done by @xcite . catalog with 154 sources in 82 galaxies was produced by @xcite , and 226 objects were included in the catalog by @xcite , compiled from mixed literature sources . such works allowed the first population studies of these objects as a well recognized group , and in relation to the properties of their host galaxies . they also provided the rationale for longer , targeted observations with _ chandra _ and _ xmm - newton_. today , the largest catalog @xcite is based on the 2xmm serendipitous survey , and contains 470 ulxs , of which 367 not listed in previous compilations . updated catalogs based on _ chandra _ observations have been compiled by @xcite and @xcite . there are three basic ways to explain non - nuclear systems with apparent accretion luminosities @xmath24 erg s@xmath12 , using only standard accretion physics already included in @xcite s model . we can either increase the mass of the bh ( thus , increasing its eddington limit ) , or attribute the enhanced brightness to beamed emission , or require a mass accretion rate high enough to allow the source to exceed the eddington limit by a factor of a few . it is also possible that the most luminous ulxs may be explained by a combination of all three factors . assuming solar abundances , the _ apparent _ luminosity . ] of an accreting bh is : @xmath25 @xcite , where @xmath26 is the beaming factor , and @xmath27 is the dimensionless accretion rate at large radii ( typically , the mass transfer rate from the donor star ) , that is normalized to the eddington accretion rate , typical of efficient disk accretion ; hence , a dimensionless accretion rate @xmath28 produces a luminosity @xmath29 . other authors prefer to define @xmath30 , in which case the eddington limit is reached at @xmath31 . ] : @xmath32 . let us examine the main tenets of these alternative or complementary scenarios . _ strong beaming _ ( @xmath33 ) a decade ago , accreting bhs with strongly beamed emission were already well known in extragalactic astronomy ( bl lac objects and blazars : * ? ? ? * ) , and at least one galactic bh ( ss 433 ) was known to have semi - relativistic jets , massive outflows and likely anisotropic emission . naturally , a possible explanation for the early sample of _ rosat _ ulxs was that they were microblazars , or ss 433-like objects seen face - on @xcite . the relativistic beaming scenario soon ran into several fatal difficulties . it predicts that for every ulxs , there should also be a large number of lower - luminosity beamed sources : @xmath34 sources with apparent @xmath35 erg s@xmath12 for every ulx at @xmath36 erg s@xmath12 ; but there is no evidence of this population @xcite . instead , empirical luminosity functions show only @xmath9@xmath37 sources more luminous than @xmath16 erg s@xmath12 for every ulx at @xmath19 erg s@xmath12 @xcite , consistent with the higher end of the high - mass x - ray binary luminosity function . there is also no multiband evidence of the large number of jet sources beamed in other directions , required by this scenario @xcite . conversely , the presence of photoionized bubbles around several ulxs ( for example holmberg ii x-1 : @xcite ; ngc5408 x-1 : @xcite ) requires quasi - isotropic x - ray emission with luminosities @xmath38 erg s@xmath12 . finally , the majority of ulxs do not have radio counterparts and/or rapid x - ray variability , expected for relativistic beaming . for all those reasons , relativistic beaming is now ruled out as a general scenario . _ mild beaming and/or super - eddington accretion _ ( @xmath39 , @xmath40 ) when the accretion rates exceeds the eddington limit , radiatively - driven outflows are launched from the inner part of the disk , inside the spherization radius where the disk becomes geometrically thick @xcite . it was suggested @xcite that the wind may produce a funnel wall that scatters and collimates the emission along the axis perpendicular to the disk plane . the apparent high luminosity of ulxs is then a combination of this mild collimation ( with @xmath41 , @xcite ) and mild super - eddington emission ( @xmath42 ) . in this scenario , accretion rates @xmath31@xmath43 are enough to explain most or all ulxs up to luminosities @xmath44 erg s@xmath12 with bh masses @xmath45 . recent numerical radiation - magneto - hydrodynamical simulations @xcite supported and advanced earlier analytical results ( slim disk models ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and the radiation - hydrodynamical simulations of , among others , @xcite . an important result of the simulations is that even a moderately super - critical mass supply @xmath46 can produce a total luminosity @xmath47 ( in rough agreement with the standard analytical scaling used earlier ) , and an apparent luminosity @xmath48 for face - on observers , due to mild collimation in the magnetized outflow @xcite . an alternative explanation for ulxs relies on super - critical accretion but does not require collimation of the emerging radiation . it was suggested that radiation pressure - dominated accretion disks have strong density inhomogeneities due to the development of photon - bubble instabilities @xcite . radiation would escape from such disk regions at rates @xmath49 times above the classical eddington limit for stellar mass bhs . _ quasi - isotropic eddington luminosity _ ( @xmath50 , @xmath51 ) although some models and simulations offer us compelling reasons to ignore the eddington limit as a constraint for the ulx interpretation , some authors remain wary of this scenario . they point out that the eddington limit works well for galactic bhs and neutron stars . more significantly , the isotropic bolometric luminosity of the whole population of @xmath52 quasars at redshift @xmath53 in the sloan digitized sky survey is strongly eddington - limited at all bh masses , with only a very small fraction of sources between 1 and 3 @xmath54 @xcite . perhaps like the bumblebee , the eddington limit should not fly , but does . thus , it is argued that the null hypothesis for ulxs should be that their luminosities are @xmath55 , despite the difference on the gas reservoir between agn and binary systems , and their radiative anisotropy is only a factor of @xmath56 ( a standard accretion disk has a beaming factor @xmath57 in the absence of any other collimation ) . these conditions are possible to explain most ulxs up to a few @xmath58 with accreting bhs of masses @xmath59 , which is the theoretical upper limit for a direct collapse of a single stellar core in the current universe ( see section [ sec : bhmass ] ) . sources more luminous than that may require imbhs . for example , hlx-1 in eso243 - 49 , the most luminous ulx found to date , has a peak bolometric luminosity @xmath60 erg s@xmath12 @xcite . if it is eddington limited , a black hole mass of at least @xmath61 is inferred . even if super - eddington emission is allowed , an imbh is required . _ quasi - isotropic sub - eddington luminosity _ ( @xmath62 , @xmath63 ) finally , it is possible that ulxs are accreting at sub - eddington accretion rates with nearly isotropic emission or disk beaming , which requires the compact objects to be imbhs ( @xmath64 ) . then we expect to see similar state transition and x - ray spectral properties from both ulxs and galactic bhbs , scaled only by the mass . the emission scenarios outlined above require three different ranges of bh masses , with different formation mechanisms . _ ordinary stellar mass bhs _ ( @xmath65 ) the first advantage of explaining ulxs with ordinary stellar mass black holes ( smbhs ) is that we already have direct evidence of their existence in the local universe . there are more than 20 bhs identified in x - ray binaries with dynamically measured masses @xmath9@xmath10 ( for reviews see * ? ? ? * ; * ? ? ? * ) . in the milky way , the most massive smbhs are those in grs 1915@xmath66105 ( @xmath67 ; * ? ? ? * ) and in cygnus x-1 ( @xmath68 ; * ? ? ? a similarly massive bh is found in m33 x-7 ( @xmath69 @xmath1 ; * ? ? ? an accreting binary system in the nearby starburst galaxy ic 10 @xcite contains a bh with a mass of @xmath70@xmath71 ( the mass uncertainty is mainly due to the unconstrained companion mass and inclination angle of the binary plane ) . the second advantage is that they are the easiest bhs to form : standard stellar models predict smbhs as the endpoint of individual massive star evolution , for a wide range of stellar masses and metallicities . thus , the strictest constraints to smbh models for ulxs come not from the bh formation process , but from the extreme super - eddington accretion , beaming and outflows required by these models . _ massive stellar bhs _ ( @xmath72 ) the mass of a compact remnant depends both on the initial mass of the stellar progenitor , and on how much mass is lost through stellar winds during the stellar lifetime @xcite . radiatively - driven winds from massive stars are strongly dependent on their metallicity @xcite . therefore , metallicity plays an important role in stellar evolution and determines the maximal mass of a bh that can be formed via core collapse . @xcite argued that the collapse of a single star can form a bh up to @xmath73 at solar metallicity , @xmath74 at 0.3 times solar metallicity , and @xmath75 at 0.01 times solar metallicity ; consistent results ( upper bh mass limit @xmath76 at low metallicities ) that are obtained by @xcite and @xcite . these results can well explain the inferred bh masses in galactic x - ray binaries ( average metallicity close to the solar value ) and in the ic 10 source ( metallicity @xmath77 solar ) . in summary , current stellar evolution models predict a bh remnant distribution peaked around @xmath78 , with a tail extending up to @xmath59 . as the formation of stellar bhs more massive than @xmath79 may require rarer , more extreme conditions , we will refer to them as `` massive stellar black holes ( msbhs ) '' . there is another interesting physical difference between smbh and msbh formation , in addition to different progenitor mass and metallicity : smbhs are probably formed through a sn explosion and subsequent rapid fallback ; msbhs are probably formed from direct core collapse without an explosion @xcite . this is because for the more massive objects , the gravitational energy released during core collapse is not enough to unbind and expel the stellar envelope . at low metallicities , the threshold for direct collapse is an initial stellar mass @xmath80 @xcite ; at solar metallicity , even the most massive stars may lose enough mass in winds and explode as sne . _ intermediate mass bhs _ ( @xmath81@xmath4 @xmath1 ) to produce an imbh of @xmath82@xmath83 , the core collapse of an isolated star in the current epoch is not a viable process . this is not because of the lack of very massive stars ( in fact , several stars with initial masses @xmath84@xmath85 may exist in the lmc ; * ? ? ? * ) , but because helium cores more massive than @xmath86 ( corresponding to initial stellar masses @xmath87 at zero metallicities ) undergo an electron / positron pair - instability explosion that leads to the complete disruption of the star , without bh formation @xcite . however , a bh may form again if the helium core is more massive than @xmath88 , corresponding to initial stellar masses @xmath89 at zero metallicity @xcite . it has been suggested that metal - free population iii stars formed in the very early universe could reach masses of a few hundred @xmath90 @xcite , above the pair - instability limit , and thus may have collapsed into imbhs @xcite . in young and dense star clusters , dynamical friction could lead to massive stars sinking towards the center and undergoing runaway collisions and mergers on timescales @xmath91 yr . numerical simulations suggest that such a process may result in the formation of a huge star , which may subsequently collapse into an imbh @xcite . dynamical evidence for imbhs with masses @xmath92 has been proposed for a few globular clusters , _ e.g. _ , g1 in m31 @xcite , although the issue is still controversial @xcite . @xcite suggested that in massive and/or compact globular clusters , a central seed smbh may grow by up to a factor of 100 via accretion of gas lost by the first generation of cluster stars in their red - giant phase . imbhs may also wander in the halo of major galaxies , after tidal stripping of merging satellite dwarfs that contained nuclear bhs @xcite ; this is currently the most plausible explanations for the most extreme ulx found to date , hlx-1 in eso243@xmath9349 . to summarize , three types of non - nuclear bhs could be powering ulxs ( table [ tab : bh ] ) . cclc type & mass & progenitor & @xmath94 + smbh & @xmath1710 @xmath1 & normal star & @xmath56 + msbh & @xmath95100 @xmath1 & low - z star & @xmath96 + & & pop iii & + imbh & @xmath3@xmath4 @xmath1 & cluster core & @xmath97 + & & stripped nucleus & + assuming that ulxs are binary systems with a bh receiving matter directly from an individual donor star , we can try to fit them into the traditional classification of galactic x - ray sources , as either high - mass or low - mass x - ray binaries ( hmxbs and lmxbs , respectively ) . hmxbs have a massive ob donor star with a characteristic age @xmath98 yrs ; lmxbs have a low - mass donor ( often in the subgiant or giant phase ) with ages @xmath99 yrs ( _ e.g. _ * ? ? ? hmxbs are mostly found in spiral and irr galaxies , with current or recent star formation , while lmxbs populate old spheroidals and globular clusters . ulxs are found both in ellipticals and in spiral / irregular galaxies ; however , the most luminous ones are found predominantly in star - forming galaxies . specifically , more than two thirds of ulxs found in ellipticals are in the lowest luminosity range , with @xmath100 erg s@xmath12 @xcite , and can easily be explained as the upper end of the lmxb distribution , with bh masses @xmath101 . it was even suggested @xcite that all ulx candidates in ellipticals with apparent @xmath102 erg s@xmath12 could be dismissed as background or foreground sources . today , it is generally accepted that there are a few genuine ulxs in ellipticals at luminosities @xmath103@xmath104 erg s@xmath12 , but none has been found above @xmath19 erg s@xmath12 . in spirals , by comparison , one third of ulxs have luminosities @xmath105@xmath106 erg s@xmath12 , and about 10% have luminosities @xmath107 erg s@xmath12 @xcite . both early - type ( hubble types 0 to 4 ) and late - type ( hubble types 5 to 9 ) spirals may contain ulxs , with similar probability , within the current statistical error of the samples ; @xcite find slightly more ulxs in early - type spirals , while @xcite find slightly more in late types , due to the different spatial resolutions of the instruments . furthermore , the fact that a luminous ulx is found in a spiral galaxy does not imply that it has an ob donor star . many ulxs in spirals are not located in regions ( within a few 100s of pc ) with evidence of current or very recent star formation @xcite . using the optical color distribution in a sample of face - on host galaxies , @xcite found that the brightest ulxs turn on @xmath0@xmath108 myr after the end of star formation . characteristic ages @xmath109 myr for the surrounding stellar population were also found for example in the case of ngc1313 x-1 and x-2 @xcite , ic 342 x-1 @xcite , ngc4559 x-1 @xcite , and ngc 1073 ixo 5 @xcite . @xcite normalized the radial distance of all ulx candidates in their sample to the deprojected @xmath110 diameter of their host galaxies . they found that the ulx surface density profile is strongly peaked towards the center . the profile can be fitted with a sersic function , with a best - fit index consistent with a de vaucouleurs profile ( @xmath111 ) , and an exponential scale - radius @xmath112 . this finding suggests that the ulx population is intrinsically related to the stellar population of their host galaxies . less centrally concentrated distributions would be expected if most ulxs were primordial halo relics such as population iii remnants , or nuclear bhs of accreted satellite galaxies . a _ chandra _ sample of ulxs in 58 face - on star - forming galaxies within 15 mpc , studied in details by @xcite , does not show statistically significant associations between ulxs and young , massive ( @xmath113 ) star clusters , within @xmath114 regions . the same conclusion was found by @xcite for ulxs in the antennae galaxies . this is inconsistent with formation scenarios based on runaway core collapse and mergers of o stars inside a compact , young stellar cluster , leading to the possible formation of an imbh of a few hundred solar masses via direct collapse of the resulting superstar @xcite . in fact , most ulx host galaxies do not even have stellar clusters sufficiently massive and compact to satisfy the requirements for runaway core collapse . in theory , imbhs could be formed inside clusters that have since dispersed . however , the evaporation timescale of such clusters would be too long to explain the observed association of many ulxs with young ( @xmath115 myr ) stellar populations . here , we further quantify the association between ulx populations and host galaxy properties . @xcite studied 107 ulxs in a complete sample of 127 nearby galaxies ( @xmath116 mpc ) , selected to be above the completeness limit of both the uppsala galaxy catalog and the _ infrared astronomical satellite _ ( _ iras _ ) survey . the total star formation rate ( sfr ) of the galaxies in the sample is @xmath117 yr@xmath12 based on the _ iras _ data , consistent with other measurements of star formation based on the h@xmath118 and [ o@xmath119ii ] indicators @xcite . this , in turn , corresponds to an sfr density @xmath120 yr@xmath12 mpc@xmath121 in the local universe . finding @xmath122 ulxs in that volume , after correcting for incompleteness and background interlopers , implies that we expect , on average , @xmath123 ulxs for a sfr of @xmath124 yr@xmath12 , and 1 ulx in a volume of @xmath125 mpc@xmath121 . this is consistent with the hmxb luminosity function in local star - forming galaxies : @xcite find @xmath34 sources with @xmath126 erg s@xmath12 ( corresponding to @xmath127 erg s@xmath12 in the ulx definition used in @xcite ) in a sample of galaxies with a total sfr @xmath128 yr@xmath12 . the average specific frequency in star - forming galaxies , in the local universe , is 1 ulx per @xmath129 , for galaxies with dynamical mass of @xmath130 @xcite . this rate decreases for increasing galaxy mass , scaling as @xmath131 @xcite . in elliptical galaxies , the specific rate is 1 ulx per @xmath132 @xcite . thus , it appears that low - mass spirals and irregulars are more efficient at producing ulxs , per unit mass . this is true at least down to galaxy masses @xmath133 . there may be two reasons why low - mass galaxies contain more ulxs per unit mass @xcite . one is that their sfr per unit mass is higher , so they are more efficient at producing ulxs . the other is that they have lower metal abundances , which may favor the formation of heavier stellar bhs from the collapse of massive o stars @xcite . however , it is suggested by @xcite that the metallicity does not affect the ulx production strongly after comparing two pairs of colliding galaxies . finally , it is often noted that many luminous ulxs reside in colliding or strongly interacting galaxies : the antennae @xcite , the cartwheel @xcite , the mice @xcite , ngc4485/90 @xcite , ngc7714/15 @xcite , ngc3256 @xcite , and a few more arp galaxies . this may simply be due to the fact that colliding galaxies have the largest sfr , and the number of ulxs scales with the recent sfr . the cumulative luminosity function ( lf ) of a population of accreting sources ( also known as @xmath134@xmath135 curve ) is expected to have a break or cut - off at the eddington luminosity of the most massive objects in the population . therefore , in the absence of direct mass measurements , the lf can be used as a rough indicator of the maximum mass . in fact , things are not so simple . the lf of x - ray binaries in our galaxy and in other nearby galaxies does not show any features at @xmath136 erg s@xmath12 , eddington limit of the neutron star population . moreover , the location of the upper cutoff may be a function both of mass ( via the eddington limit ) and of the age of the population , since the most luminous sources tend to have a shorter lifetime @xcite . thus , careful modeling is required to interpret a lf . matches the observed distribution better than an unbroken power - law , although the issue remains controversial ( _ cf . at the low - energy end , it also matches the slope and normalization of the hmxb distribution below @xmath16 @xcite . [ fig : lf ] ] with this caveat , we can use the ulx lf to test the scenario that ulxs in star - forming galaxies are mostly the high - luminosity tail of hmxbs ( plus some additional lmxbs ) , and ulxs in ellipticals are the high - luminosity tail of lmxbs . chandra _ and _ xmm - newton _ surveys of discrete x - ray sources in nearby galaxies have produced consistent results in agreement with this interpretation @xcite . in summary , they have all found that the cumulative lf of young x - ray populations in star - forming galaxies is a simple power - law with a slope @xmath137@xmath138 , while the cumulative lf of old x - ray populations in ellipticals and spiral bulges has a slope of @xmath139 . thus , short - lived hmxb - like sources dominate the high - luminosity end of the distribution . the normalization of the hmxb lf is proportional to the sfr ; the normalization of the lmxb lf is proportional to the stellar mass @xcite . what remains controversial is the behavior of the lf for the young ulx population above an x - ray luminosity @xmath38 erg s@xmath12 . from their _ chandra _ survey , @xcite find that a pure power - law @xmath140 provides a very poor fit to the data above that threshold ( figure [ fig : lf ] ) . a cut - off power - law model with @xmath141 provides an acceptable fit , with @xmath142 ( as required by the low - luminosity end of the lf ) and @xmath143 erg s@xmath12 . a very similar result was also found by @xcite , @xcite , and @xcite . the cut - off power - law model does not preclude the possibility of finding ulxs above @xmath144 erg s@xmath12 , but they would be increasingly rarer . it predicts @xcite 1 ulx with @xmath145 erg s@xmath12 within a radius of @xmath146@xmath147 mpc , and 1 ulx with @xmath148 erg s@xmath12 within a radius of @xmath149@xmath150 mpc . if the luminosity of a few ulxs recently discovered by @xcite is confirmed to be @xmath44 erg s@xmath12 , their existence would not be consistent with the cut - off model of @xcite , and may suggest a physically different population of accretors @xcite . and , of course , so does hlx-1 in eso243@xmath9349 with its peak outburst luminosity @xmath60 erg s@xmath12 @xcite . in contrast with those findings , the _ xmm - newton _ survey of @xcite does not see any turnover or cut - off in the cumulative lf above @xmath19 erg s@xmath12 , but their data can not rule out the presence of a similar break . at this stage , the discrepant result may be caused by a different binning technique , or by small - number statistics . both the @xcite and @xcite complete subsamples have only 5 sources with @xmath151 erg s@xmath12 out of a total of @xmath152 ulxs . a few other ulxs above that threshold are known , but can not simply be added to the lf because they do not form part of a complete sample . thanks to the large effective area of _ xmm - newton _ , combined with _ chandra _ observations , the spectral and timing properties of ulxs in the @xmath153@xmath37 kev band have been widely investigated and modeled over the last decade , even though the physical interpretation is still controversial . most of the x - ray studies have focused on bright ulxs ( @xmath154 ) in nearby galaxies ( @xmath155 mpc ) that allows spectral and timing analysis at high signal to noise ratios . in the next two sections , we review x - ray spectral and timing properties of bright ulxs and discuss their physical interpretation in comparisons with well - defined emission properties of galactic bhbs . phenomenologically , the energy spectra of ulxs in the @xmath153@xmath37 kev band can be divided into two groups : those that are consistent with a simple power - law , and those that are more complex @xcite . the latter mainly refers to a mild broad curvature ( convex shape ) over the whole band , a break or steepening above @xmath103 kev , or a soft excess below @xmath103 kev . a typical ulx spectral shape of the latter is sketched in figure [ fig : spec ] . many ulx spectra are well fitted by a single , absorbed power - law model . the photon index has a broad distribution , peaked at @xmath1562.0 @xcite but with a few much harder ( down to @xmath157 ) as well as much softer sources ( up to @xmath158 ) . the distribution does not appear to be bimodal ; there is no evidence of a gap between a `` hard state '' and a `` soft state '' . @xcite found that the power - law photon index tends to be harder at increasing luminosity ; a selection of ulxs with high luminosity ( @xmath159 erg s@xmath12 ) and hard power - law spectrum ( @xmath160 ) is also presented in @xcite . some of the hard sources with multiple observations show strong flux variability ( sometimes by an order of magnitude ) but constant photon index @xcite . similar behavior has been seen from galactic bhbs in the hard state @xcite . the x - ray luminosity in the hard state is usually @xmath161 times eddington @xcite , but during hard - to - soft state transitions the hard state can reach a luminosity up to 30% of eddington @xcite . thus , if hard power - law ulxs belong to the same canonical hard state defined in galactic bhs , their bh masses must be @xmath162 @xcite . a key signature of the hard state is the presence of compact , continuous radio jets . so far , the only possible detection of such jets is from ic 342 x-1 , with _ vla _ observations @xcite . using the fundamental plane of accreting bhs , @xcite estimate a compact object mass @xmath163@xmath164 . further observations are needed to spectrally constrain the nature of the radio core in this system . m82 x-1 is another good candidate for hard state ulxs . this source displays low frequency quasi - periodic oscillations ( qpos ) @xcite above a band - limited broad noise component , and its x - ray spectrum is a featureless power - law with a photon index @xmath165 @xcite . we note that m82 x-1 may be one of the very few ulxs that change their spectral state during outbursts , switching from a hard to a thermal state ( see section [ sec : curve ] ) . the hard spectrum may come from an advection - dominated accretion flow @xcite , a luminous hot accretion flow @xcite , a classical hot corona above the accretion disk @xcite , or from a combination of synchrotron and inverse - compton emission at the base of the jet @xcite . the first mechanism is limited to low accretion rates , but the latters can in principle operate also near the eddington luminosity and be relevant to ulxs @xcite . thus , an alternative interpretation favored by some authors is that hard power - law ulxs are not in the canonical low / hard state ; in other words , that the ulx hard state is not confined to luminosities @xmath166 eddington , and does not switch to a softer , disk - dominated thermal state as the accretion rate increases @xcite . we could call it a persistent high / hard state ( unknown in galactic bhbs ) , which can extend all the way to the eddington luminosity and smoothly morph into the very high state or other super - eddington accretion states . at the opposite end of the broad photon - index distribution , several ulxs with @xmath167 could be classified in the steep power - law state of galactic bhbs , typically found at near - eddington luminosities @xcite . the fact that there are no clear gaps , but instead a continuous distribution in photon index and an overlapping luminosity distribution between hard and soft ulxs may be an indication that the two regimes are smoothly connected . in conclusion , with longer x - ray observations and increasing photon statistics , we have reached a stage where we can identify and model a number of subtle spectral features on top of the simple power - law for many ulxs . in the rest of this section , we discuss a variety of spectral models that account for such features , and their physical implications . -free ) or a warm , thick comptonization model can adequately fit the hard curvature ; in the plot , a @xmath168-free model with @xmath169 kev and @xmath170 has an almost identical spectral shape to a comptt model with @xmath171 kev and @xmath172 , which is shifted vertically for clarity . for comparison , two diskbb models are plotted with @xmath173 and 2.5 kev , respectively . they are more curved and can not fit the energy spectrum over the 0.3 - 10 kev band . [ fig : spec ] ] in many ulxs , excessive residuals at low energies ( @xmath174 kev ) would be seen if one makes a power - law fit at high energies ; an additional disk component could adequately fit the excesses and the inferred disk inner temperature is around 0.10.4 kev . this cool , soft excess was first identified by @xcite in ngc 5408 x-1 ; @xcite showed the emission was point like and applied a disk plus power - law model . similar phenomena were then found in ngc 1313 x-1 and x-2 @xcite , and later on in many other ulxs @xcite . such a fit is analogous to the empirical model for galactic bhbs except that the disk is cooler and more luminous in ulxs than in smbhs at their high state @xcite . for a standard accretion disk @xcite extending to the last stable orbit around a bh , the inner disk temperature @xmath175 is a function of bh mass @xmath8 and total luminosity @xmath176 : @xmath177 where @xmath118 depends on the bh spin ( @xmath178 for a schwarzschild bh , @xmath179 for an extreme kerr bh ) , @xmath180 is a corrective factor that takes into account the no - torque condition at the innermost orbit , and @xmath181 is the spectral hardening factor @xcite . given the same spinning and correction factors , the bh mass is scaled with the disk inner temperature and luminosity as a function of @xmath182 thus , the low - temperature , high - luminosity characters of the assumed disk suggest that the emission arises from accretion onto imbhs . for comparison , the typical disk inner temperature is about 1 kev at a luminosity of @xmath183 for smbhs about 10 @xmath1 . the characteristic temperature for ulxs is usually around 0.10.4 kev and the luminosity is some @xmath16@xmath19 , suggestive of a bh mass of @xmath184 @xmath4 @xmath1 . however , the interpretation of soft excesses as emission from cool accretion disk has confronted challenges . @xcite argued that soft excesses could also be soft deficits depending on the energy band where the power - law is fitted . @xcite found similar soft excesses in both ulxs and less luminous x - ray sources , implying that they are not a unique signature of imbhs in a statistical sense . it is suggested that supercritical accretion may produce massive outflows , which are optically thick with soft , thermal emission at similar temperature , being another possible origin for the detected soft excesses @xcite . other authors @xcite have suggested that the soft excess does come from the disk surface , but much further away than the innermost stable orbit , where the disk may be obscured or replaced by an outflow or a scattering corona ( section [ sec : curve ] ) . in that case , the characteristic radius of the thermally - emitting disk region can not be used to infer the bh mass . repeated observations of an individual ulx may help test the cool disk model @xcite . if the disk touches the innermost stable orbit around the bh , one expects to see the disk luminosity vary to the 4th power of the inner temperature , @xmath185 , as well known in the thermal state of galactic bhbs @xcite . if , instead , the observed inner disk radius is the boundary between an outer standard disk and an inner corona or outflow , we expect that such transition will move further and further out at higher accretion rates when super - eddington . in the simplest outflow scenario , @xmath186 @xcite . this implies @xcite @xmath187 constant for a viscously heated outer disk and @xmath188 for a strongly x - ray - irradiated outer disk ( as is more likely the case in ulxs ) . most other plausible relations between @xmath189 and @xmath27 predict negative correlations between the disk luminosity and temperature , with slopes between @xmath190 and @xmath191 @xcite . a negative correlation between the luminosity and temperature is clearly detected for at least ngc 1313 x-2 and ic 342 x-1 @xcite . from a sample of 8 ulxs , @xcite found that the luminosity of the soft excesses scales with the temperature as @xmath192 , roughly consistent with the model prediction that the soft emission arises from a photosphere due to strong outflows from a supercritical disk and its temperature is determined by the spherization radius @xcite . on the contrary , a positive correlation is seen from ngc5204 x-1 ; however , it has a shallower slope @xmath193 @xcite . @xcite suggested , based on a disk atmosphere model , that the hardening correction factor becomes larger at increasing luminosity , and explains the apparent shallower slope in galactic bhs such as xte j1550@xmath93564 and h1743@xmath93322 . applying such a varying hardening correction , one can recover the @xmath194 relation for ngc5204 x-1 , which would favor the imbh scenario . however , this result is not robust yet : @xcite found that the positive correlation disappears or even becomes negative if one chooses a different lower energy bound for the spectral fitting . also , the disk contributes only @xmath195 of the 0.310 kev luminosity , inconsistent with the thermal dominant state . instead , if ulxs were imbhs , we would expect to see many of them in the canonical thermal state dominated by the soft disk component . in summary , using the disk parameters inferred from x - ray spectra to determine the bh mass is based on assumptions that have proved either incorrect or problematic for most sources . it was noted by @xcite and @xcite that high energy curvature / steepening is a common feature in high - quality ulx spectra . generally , the break / steepening could be fitted by a comptonization model with relatively low energy electrons and high optical depth @xcite . a few ulx spectra are mildly convex , formally consistent with a standard disk model . we thus discuss the physical implications of these models . with _ asca _ observations , a hot multicolor accretion disk model was found to fit the spectra of a number of ulxs ( _ e.g. _ * ? ? ? for those with relatively low temperatures ( @xmath196 kev ) , they could be explained by accretion onto smbhs or msbhs @xcite . the high temperature and luminosity may be explained by large bh spin @xcite , but this does not work for extreme cases where relatively small bh mass is required , which in turn is inconsistent with the observed high luminosity . for _ xmm - newton _ spectra , a single hot disk with absorption is usually unable to fit the data ; an additional component is required and dominates in the low energy band @xcite , and the disk component is not the dominant part @xcite . the presence of a soft component in addition to the hot disk is unusual in bhbs and does not have a simple physical interpretation . in general , the hot disk model is problematic . the standard disk emission is best known in the thermal state in bhbs , in which the x - ray spectrum is dominated by emission from a standard accretion disk ; plus , the x - ray variability is low , with no or weak qpos @xcite . so far , the thermal state may have been identified only in one case , m82 x-1 , constrained using both spectral and timing information . with simultaneous _ chandra _ and _ xmm - newton _ observations of m82 , @xcite found surprisingly that the previously known qpos in the source disappeared . the light curve was no longer highly variable and the power spectrum was consistent with that of white noise . the energy spectrum also changed dramatically from a straight power - law to a disk - like spectrum . fitting with a standard accretion disk model , the inferred luminosity and temperature from three repeated observations are consistent with a 4th power relation . all results are well consistent with that expected for the thermal state . the monitoring data from _ rxte _ indicate that these _ chandra _ and _ xmm - newton _ observations were made during the source outbursts , suggesting that m82 x-1 usually stays in the hard state and could transition to the thermal state during outbursts . fitting with a relativistic disk model , the bh is estimated to be a fast spinning imbh of 200800 @xmath1 . there are a few more sources @xcite which have shown spectral variability possibly consistent with a @xmath194 relation if fitted with a hot standard disk model . another group of five nearby ulxs consistent with the thermal state spectrum , based on x - ray spectral modeling , was discussed by @xcite . these sources all lie at the lower end of the ulx luminosity regime , and could be explained by kerr msbhs in the thermal state . they do not stand for the majority of ulxs , in particular those with a typical luminosity around @xmath19 . also , lack of fast timing information hampers a firm identification . therefore , it is generally accepted that the canonical thermal state is rare in ulxs @xcite . the slim disk can radiate near or moderately above the eddington limit estimated from spherical accretion @xcite . this super - eddington disk is thus a good candidate for ulx emission @xcite . when advective energy transport dominates over radiative cooling in an accretion disk , the radial temperature profile changes from @xmath197 for a standard disk to @xmath198 for a slim disk @xcite . an extended multicolor disk model called the @xmath168-free model was introduced , with the radial dependence @xmath168 on temperature ( @xmath199 ) being an additional free parameter @xcite , and used to explain ulx emission @xcite . @xcite investigated the energy spectra of four ulxs , ngc 5204 x-1 , ngc 4559 x-7/x-10 , and ngc 1313 x-2 , and found that they can be well fitted by the @xmath168-free model . interestingly , the best - fit @xmath168 value was found to be around 0.5 , consistent with that for a slim disk . in the luminosity vs. disk temperature diagram , the four ulxs and galactic bhbs appear to be in line with the same @xmath194 relation , suggesting that ulxs may be bhs of similar or slightly higher masses ( smbhs or msbhs ) but in the slim disk state . another successful test of the @xmath168-free or slim disk model was for m33 x-8 , which is marginally classified as a ulx but a good candidate for deep investigation owing to its relatively small distance ( _ e.g. , _ * ? ? ? * ; * ? ? ? * ) . @xcite first identified that an extra power - law component is required in addition to the @xmath168-free disk to fit the source spectrum . the best - fit @xmath168 value ( @xmath200 ) suggests notable advection in the disk , close to the slim disk nature rather than a standard one . these indicate that the slim disk state may be a natural extension of the canonical thermal state at supercritical accretion . this is consistent with the radiation - magnetohydrodynamic simulations that the hard , thermal and slim states could be produced under a single model with different accretion rates @xcite . however , the @xmath168-free model encounters difficulties in both observation and theory . in some sources , the disk inner temperature was found too high to be physically reasonable when fitted with the @xmath168-free model @xcite . if photon trapping is taken into account , the radial temperature profile is expected to be significantly flatter than @xmath201 within the trapping radius @xcite , in contrast to the spectrum predicted without considering photon trapping @xcite . also , numerical simulations suggest that dense , warm outflows are always associated with supercritical accretion @xcite , which would comptonize and obscure the emission from the underlying disk , see details in the following subsection . in galactic bhbs , high energy spectral curvature is also detected , but usually occurs at tens to hundreds of kev . it is interpreted as comptonization in a thin corona with temperatures typical of @xmath3 kev . however , the spectral curvature in ulxs suggests an electron temperature of a few kev and a scattering optical depth significantly higher than unity , indicative of a warm and thick corona . due to the presence of such a thick corona , the emission from the disk would be distorted and the previous interpretation , a simple summation of a cool disk and a warm corona @xcite , needs be modified consequently . @xcite revisited best quality _ xmm - newton _ data and concluded that two common features simultaneously exist in the energy spectrum of ulxs : soft excesses below 2 kev and high energy curvature above 2 kev . this is not frequently seen in the spectra of galactic bhbs when they are thought to accrete at sub - eddington level . thus , they suggested that ulxs represent a new state distinct from the four well - known states ( quiescent , hard , thermal , and steep power - law ) in galactic bhbs @xcite , and call it ultraluminous state . please refer to @xcite for a more detailed review of this state . the galactic bhb that comes closest to this state is xte j1550@xmath93564 at the peak of its 1998 september outburst . a coupled disk - corona model @xcite was proposed to explain such state , in which the disk and corona are not independent but share the total gravitational energy release . the un - scattered outer disk is responsible for the soft excesses , while the comptonized inner disk produces the hard , curved tail . the bh mass , estimated from the unmasked temperature and radius of the underlying disk , lies in the regime of smbhs or msbhs . numerical simulations of supercritical accretion indicates that radiation pressure would power strong outflows above the disk in the inner region @xcite . the wind is optically thick and warm ( a few kev ) , and would strongly comptonize the disk emission . at high accretion rate , the disk becomes thick and would collimate the radiation along the symmetric axis causing geometric , mild beaming with an apparent isotropic luminosity over the eddington limit by a factor up to @xmath1710 . this picture is consistent with the proposed ultraluminous state . observationally , from the spectral evolution of m33 x-8 , @xcite suggested that the outflow is launched at near - eddington luminosity , and the radius at which it is launched moves further out with increasing luminosity . if the comptonizing region is the fast outflow , rather than a static corona , it is possible that bulk motion @xcite can be as or more important than thermal comptonization , depending on the characteristics of the outflow @xcite . observationally , ulx spectra that can be formally fitted with thermal comptonization models can be equally well fitted with bulk motion models ( _ e.g. _ * ? ? ? the only difference is in the interpretation of the characteristic energy of the upscattering electrons . as noted by @xcite , simple physical considerations , comparisons with other astrophysical classes of outflows , and numerical simulations all suggest that super - eddington ulx outflows have a characteristic speed @xmath202 escape velocity from the launching radius , corresponding to a bulk kinetic energy @xmath203 kev for the outflowing electrons . this is similar to the characteristic energy of the spectral breaks . most ulxs and bhbs have considerable radiation output above 2 kev , while supersoft sources ( ssss ) have most of their energy released below 2 kev . canonical ssss are speculated to be powered by steady nuclear burning on the surface of white dwarfs @xcite . a number of supersoft ulxs have been discovered , whose spectrum is dominated by a cool blackbody component of tens to a hundred ev , being a distinct class from the majority of ulxs . these include m101 ulx-1 @xcite , antennae x-13 @xcite , m81 uls1 @xcite , ngc 4631 x1 @xcite , and ngc 247 ulx @xcite . at long timescales , most of these sources have shown dramatic variability with a luminosity change by a factor up to @xmath184 but with consistently low temperatures . a weak power - law tail is significantly detected in at least one source @xcite , but none of the supersoft ulxs has become a power - law dominated ulx or vice versa . fast timing noise was detected in two of them , m101 ulx-1 @xcite and ngc 247 ulx @xcite , down to a timescale around 100 s , indicating that they are not agn . the nature of these peculiar sources is very uncertain . they are unlikely powered by white dwarfs , because they are too luminous ( a few times @xmath16 ) and some of them are too hot ( @xmath204 ev ) . they are also unlikely to be due to cool disks around imbhs , because such disks can not have a huge variability with a constant temperature . one possible scenario is that their emission is produced by massive outflows from supercritical accreting sources viewed at a high inclination angle @xcite . a unification model ( supercritical accretion with thick outflows , see figure [ fig : model ] ) may be able to explain both normal , broad - band ulxs and less luminous , supersoft ulxs with different viewing angles . so far , we have briefly reviewed a variety of scenarios that can account for the observed ulx spectra ( see also figure [ fig : spec ] ) . in figure [ fig : model ] , we sketch the physical ingredients behind the most common combination of such models , which may be successfully used to explain the multiband spectrum of most ulxs . ] the study of long term spectral variability of ulxs mainly relies on pointed observations with _ chandra _ or _ xmm - newton _ , or quasi - regular snapshots with the x - ray telescope on board _ swift_. in addition , _ rxte _ has been used to monitor the brightest ( in terms of observed flux ) ulx in m82 . the most interesting science with these observations , in addition to the search for orbital periods , is perhaps to find state transitions known in galactic bhbs . canonical transitions between the low / hard and high / soft states were reported in ic 342 x-1 and x-2 with _ asca _ observations @xcite and also in holmberg ix x-1 with multiple missions @xcite . however , almost no such canonical state transitions have been reported in the _ chandra _ and _ xmm - newton _ era . in contrast , transitions between soft and hard states were found but not in a way similar to galactic sources @xcite . a major difference between ulxs and galactic bhbs is that most ulxs are persistent sources and have been bright since their discovery decades ago , while the latter are usually transients especially accreting from a low mass companion with roche lobe overflow . _ swift _ monitoring @xcite of bright ulxs have revealed flaring activity with a luminosity change by a factor of @xmath155 on time scales of days to months . in particular , @xcite found holmberg ix x-1 remains in the hard state as its flux varies , and @xcite found the x - ray spectral state is not correlated with the luminosity in holmberg ii x-1 . also , as mentioned previously in section [ sec : curve ] , @xcite found m82 x-1 may have changed from the hard state to thermal state during outbursts . there is one exceptional case . eso243@xmath9349 hlx-1 shows a low / hard to high / soft jump reminiscent of the canonical state transition @xcite . it also displays outbursts in a `` fast rise exponential decay '' pattern , very similar to those seen in galactic bhbs @xcite . however , the authors argued that the observed variability is unlikely triggered by disk instability due to inconsistent timescale . the power spectrum of galactic bhbs can be decomposed into broad components ( power - law red noise , band - limited noise , flat - top or zero - centered broad lorentzian , etc . ) and narrow components ( qpos ) . although we still know little about the physical origin of these temporal features , their phenomenology has proved useful in defining and identifying accretion states @xcite and possibly in determining bh masses via model independent calibrations ( _ e.g. _ * ? ? ? an important model - independent finding of @xcite , based on the study of sample of 16 bright ulxs , is that those sources can be divided into two groups defined by their short - term x - ray variability , even though they have similar energy spectra and luminosities . one group ( including among others ngc 5204 x-1 ) shows weak or absent variability on timescales at least as large as 100s . the other group ( including among others ngc 5408 x-1 ) has similar variability levels and power spectra as luminous galactic bhbs and agn in the observed frequency bandpass ( @xmath205@xmath206 hz ) . it is still unclear what suppresses short - term variability in one class but not the other . in galactic bhbs , high variability is often associated to non - thermal emission with steady jets ; low variability to thermal emission with no jets @xcite . in the following , we select four ulxs in which interesting structures have been found in their power spectra , shown in figure [ fig : psd ] . _ m82 x-1 , also known as m82 x41.4 + 60 _ @xcite first discovered low frequency qpos around 54 mhz from the m82 galaxy with both _ xmm - newton _ and _ rxte _ observations . by fitting the _ xmm - newton _ images to its point spread function with known source positions from _ chandra _ , @xcite identified that m82 x-1 is the source that produced the qpos , and interestingly , it was not the brightest x - ray source in m82 at the time of qpo discovery . a following _ xmm - newton _ observation revealed that the qpos have changed the central frequency from 54 mhz to 114 mhz , above a flat - top component with a low frequency break at 34 mhz @xcite . @xcite also found that the qpos changed frequency from 107 mhz to 120 mhz during the second observation , and a plausible harmonic ratio of 1:2:3 for qpo frequencies with _ these qpos , however , were undetected with three _ xmm - newton _ observations of similar depth , suggestive of a spectral state transition @xcite . _ ngc 5408 x-1 _ with a short _ xmm - newton _ observation , @xcite found the source is highly variable , and its power spectrum has a break at 2.5 mhz . @xcite reported the identification of narrow qpos around 20 mhz from the source , and confirmed the presence of the frequency break . the qpos are narrow and may show a pair with a 4:3 frequency ratio . a longer _ xmm - newton _ observation later detected the qpos at about 10 mhz and the energy spectral parameters of the source also changed @xcite . the qpo frequency was found to scale with the disk flux and also the spectral index of the power - law component . these correlations are analogous to that seen in galactic bhbs , suggesting that this ulx is in an emission state similar to sub - eddington sources . the qpo properties ( narrow , strong , with a flat - top continuum ) and the temporal - spectral evolution patterns suggest that the qpos are of type c and the compact object in ngc 5408 x-1 is an imbh . however , as the qpo frequency is not proportional to the break frequency of the continuum as expected , the identification as of type c was questioned and so was the imbh scenario @xcite ; they proposed that the source is super - eddington and the turbulence in the comptonization region gives rise to the observed variability . other authors have also rejected the imbh argument @xcite , pointing out that neither lower qpo frequencies nor larger , cooler disks can be used as indicators of higher bh masses in ulxs ; both findings may simply suggest that the visible inner radius of the disk has receded much further than the innermost stable orbit , and has been covered or replaced by a thick comptonizing region or outflow in the inner region . besides , @xcite also found that the source shows a linear correlation between the rms amplitude and flux , similar to that seen in bright bhbs and agn . _ m82 x42.3 + 59 _ this is a transient ulx in m82 @xcite . it could not be detected by _ chandra _ in its low state , but sometimes , was even brighter than m82 x-1 , being the most luminous source in m82 @xcite . with _ chandra _ observations , and also confirmed by a simultaneous _ xmm - newton _ observation once , @xcite discovered low frequency qpos around 3 - 4 mhz in its power spectrum . these qpos are wide and strong , appear above white noise , and can only be detected when the source is brighter than @xmath19 . this is consistent with a type a or b classification , but is inconsistent with the properties of type c qpos . _ ngc 6946 x-1 _ this source has the highest short - term variability found in a ulx , with a fractional rms amplitude of 60% integrated in the frequency range of 1 - 100 mhz @xcite . its power spectrum shows a flat - top continuum that breaks at about 3 mhz with possible quasi - periodic oscillations ( qpos ) near 8.5 mhz . ] the characteristic frequency alone usually is not a direct indicator of the bh mass , as these noise features can vary dramatically in a single source . however , correlations between the frequency and spectral parameters have been found , and the correlation patterns are thought to scale with the bh mass . the most common qpos seen in bhbs are of type c ( _ cf . the frequency of type c qpos are found to vary over a wide range . for example , the lowest detected frequencies in grs 1915 + 105 are around 1 mhz @xcite , even lower than the frequencies of all qpos detected in ulxs . a positive correlation between the qpo frequency and disk flux has been found for type c qpos in xte j1550@xmath93564 , gro j1655@xmath9340 and h 1743@xmath93322 @xcite . the qpo frequency also scales with the photon index of the power - law component in the energy spectrum ( _ e.g. _ * ? ? ? * ; * ? ? ? @xcite suggested that the qpo frequency scales inversely with the bh mass at a fixed power - law photon index , and thus one can infer the mass given spectral index @xcite . on the contrary , type a and b qpos show in a relatively narrow frequency range . the broad continuum seems to be a more reliable determinant of bh mass , as it is detected in both bhbs and agn and the scaling has been tested over a wide frequency and mass range . smbhs and agn appear to populate a `` variability plane '' defined by the bh mass , accretion rate , and @xmath207 , which is the frequency of the lower high - frequency lorentzian component in the power spectrum @xcite . limited by sensitivity , @xmath207 can not be detected in ulxs . based on a correlation between @xmath207 and the frequency of type c qpos , @xcite extended the variability plane to type c qpos that are visible in ulxs . the bh mass can be inferred from the x - ray luminosity ( here we adopt the radiatively efficient accretion case ) and type c qpo frequency as @xmath208 where the inequality is due to the unknown bolometric correction ( @xmath209 ) . although the `` variability plane '' is tested for a wide range of bh masses , the correlation between @xmath207 and @xmath210 has not been tested in ulxs and even the identification as of type c is questioned at least for ngc 5408 x-1 @xcite . attempts to scale bh mass with the variability amplitude ( rms ) have made great success in agn @xcite . however , it is found that ulxs are significantly less variable than agn , and a simple extrapolation could not be obtained @xcite . based on the correlations mentioned above , the bh masses inferred from the characteristic frequencies are all about @xmath184@xmath4 @xmath1 , indicating that ulxs contain imbhs . however , application of these mass scaling methods requires strong caveats . there are at least three problems . first , we are not sure whether the timing features detected in ulxs are the same type found to scale with bh masses in galactic bhbs ; for example , we are not sure whether the qpos in ngc 5408 x-1 are of type c @xcite . for other types of timing features ( _ e.g. _ , type a and b qpos ) , it is still not known whether their frequency is proportional to the bh mass @xcite . second , frequency - luminosity - mass relations have been calibrated and tested for galactic bhbs and agn in sub - eddington accretion , that is when the inner edge of the disk is constant or moves towards the innermost stable orbit as the accretion rate increases ( thus , increasing qpo frequencies ) . there is no empirical evidence that the scaling goes the same way in ulxs , which may be super - eddington systems where the visible inner edge of the disk recedes from the innermost stable orbit for increasing accretion rates ( thus , reducing qpo frequencies ) . the third problem is that we do not have enough signal - to - noise ratio to detect high - frequency qpos , which severely limits possibility of direct comparisons with galactic bhbs . in this section the `` optical counterpart '' refers to the point - like optical source that is spatially associated with the ulx ; sources with resolved structure , such as optical and radio nebulae , will be discussed in the next section . the optical emission from the counterpart could arise from the donor star or outer accretion disk , or both . it provides interesting information regarding the binary evolution history , nature of the donor star , disk geometry , and mode of mass transfer , and would also help constrain the bh mass . several groups have modeled x - ray irradiation in order to distinguish between the contributions of the outer disk and the donor star , and constrain the intrinsic stellar properties . @xcite applied their irradiation model to the photometric data for seven ulxs ; they suggested that donor stars are consistent with spectral type b or later , and the bh masses are in the range of @xmath0@xmath211 . @xcite computed evolution tracks of ulx counterparts on the color - magnitude diagram , assuming a massive companion orbiting an smbh or a msbh ; they applied their models to ngc1313 x-2 , suggesting a bh mass @xmath212@xmath211 . in addition to reproducing the observed colors and luminosity ( after accounting for irradiation and disk contribution ) , a donor star in a ulx must be able to supply a mass accretion rate @xmath213 yr@xmath12 , which requires roche lobe overflow , and to do it over a sufficiently long period of time to explain the observed number of ulxs per galaxy . binary population synthesis models have been developed and used to address this problem @xcite . modeling the binary evolution requires calculations of the mass transfer rate , the corresponding change in orbital separation , and the subsequent fate of the transferred mass . this in turn depends critically on whether we impose a strict eddington limit on the mass transfer rate that can be accepted by the bh ( with the rest being expelled in a wind ) , or instead we assume that the bh can accrete at higher rates . thus , for the same kind of donor star , the outcome in terms of x - ray luminosity and lifetime is very different depending on whether the primary is an eddington - limited stellar bh , a non - eddington - limited stellar bh , or an imbh . @xcite considered three characteristic families of binary evolution tracks , with different bh mass distributions : @xmath214@xmath10 , @xmath214@xmath215 , and a single - mass population of @xmath216 bhs . they built probability diagrams in different slices of parameter space ( optical color - magnitude diagram ; orbital period versus potential x - ray luminosity ; x - ray luminosity versus age ) , representing the total time spent by all the systems in those regions of parameter space . their main conclusion is that the most probable ulx system parameters correspond to high - mass donors ( @xmath217 ) , with orbital period between 1 and 10 days and characteristic age @xmath218 yr . in terms of bh mass , they found that both the @xmath216 bh population , and that with non - eddington - limited bhs up @xmath219 were consistent with the optical and x - ray observations and with the observed number of ulxs per galaxy ; the @xmath220@xmath211 bh mass range has not been tested yet . thanks to the sub - arcsecond resolution of _ chandra _ and _ hst _ , the optical counterparts of more than a dozen ulxs have been identified . these are done mainly by aligning _ chandra _ and _ hst _ images using objects with emission on both , to reduce the _ chandra _ x - ray position error from 0.6 to about 0.20.4 . @xcite collected all available data in the _ hst _ archive to investigate the optical properties of 13 ulxs , which can be summarized as follows . a. the x - ray to optical flux ratios and optical colors are consistent with that of lmxbs , but unlike hmxbs or agn . this indicates that the optical emission from ulxs is mainly due to reprocessing of x - rays on the outer accretion disk . as a consequence , one can not infer the spectral type of the companion star using optical colors directly . b. the broad band optical spectrum can be generally fitted with a power - law model ( @xmath221 ) . the distribution of the spectral index @xmath118 peaks between 1 and 2 , which is also consistent with disk irradiation models . c. the counterpart has unique optical properties within its neighborhood . nearby sources usually show a blackbody spectrum instead of a power - law spectrum . d. the optical emitting region is estimated to be large , of the order of @xmath222 cm , indicative of long binary periods ( @xmath109 days ) for smbh or msbh primaries , or shorter periods for imbhs . e. sources with multiple observations usually show strong , random variability on both magnitudes and colors . there are a few outliers showing peculiar properties . the optical spectra of ngc 2403 x-1 and m83 ixo 82 have a slope consistent with that of the standard disk emission , @xmath223 . m101 ulx-1 and m81 uls1 are more similar to hmxbs than lmxbs in terms of x - ray to optical flux ratios . ic 342 x-1 may show a blackbody - like spectrum instead of a power - law . more details can be found in @xcite . there have been a lot of observational efforts to determine the nature of the companion stars in nearby ulxs . so far , ngc 7793 p13 is the only ulx in which emission from the companion star is directly seen beyond reasonable doubt @xcite ; detection of narrow emission lines identifies the companion star as a late - b type supergiant with a mass @xmath79 . orbital periods may have been detected in a few sources . assuming roche - lobe overflow , the companion density can be simply determined from the binary period , @xmath224 g cm@xmath121 @xcite . m82 x-1 is the strongest case of a positive detection ; a monitoring program with _ rxte _ revealed a steady period of about 62 day @xcite , corresponding to a companion density of @xmath226 g cm@xmath121 . @xcite discovered a 115 day x - ray period from ngc 5408 x-1 with _ swift _ monitoring , and inferred a similar companion density of @xmath227 g cm@xmath121 ; however , the interpretation of that modulation as a binary period has been disputed @xcite . these claimed long periods imply an evolved donor star . possible detections include a 6.1 day period from ngc 1313 x-2 with _ hst _ observations @xcite ; however , low significance of the period was claimed by other groups @xcite . m82 x-1 is located near or within a super star - cluster mcg-11 with an age of 7 - 12 myr @xcite . ngc 1073 ixo 5 appears near a ring of star formation of 816 myr old @xcite . ngc 1313 x-2 is found near a star cluster with an age of 20 myr @xcite . ic 342 x-1 lies in a field with relatively old stellar objects , mostly older than 10 myr @xcite . the stellar field around ngc 1313 x-1 is at least 30 myr old @xcite . holmberg ix x-1 is suggested to be part of a loose cluster with an age @xmath115 myr @xcite . all those findings suggest that ulxs are in relatively old stellar environments with ages of at least 10 myr . an exceptional case is provided by a transient ulx recently discovered in m83 @xcite . it is located in an interarm region with a stellar population older than @xmath228 yr . _ hst _ observations when the system was probably in quiescence constrain the donor star to be @xmath229 . instead , _ observations during the recent outburst revealed a bright , blue counterpart with absolute @xmath230 magnitude @xmath231 , due to the irradiated outer disk and/or irradiated donor star . identification of the optical counterparts to ulxs is the first step for a potential dynamical mass measurement . @xcite found that the broad he ii @xmath232 line from ngc 1313 x-2 changed its velocity by about 380 km s@xmath12 between two observations . if this is due to binary motion of the disk , an upper limit on the bh mass of 50 @xmath1 can be placed . there have been a few attempts to obtain the mass function of a few ulxs using emission - line velocity curves . however , the lines were found to vary randomly instead of showing an ellipsoidal modulation ; thus , they did not provide dynamical constraints on the bh masses @xcite . the orbital period is a key parameter for dynamical mass measurement . with current facilities , it may be more feasible to search for orbital periods in x - rays and then confirm them in optical for extragalactic sources , like the cases of m33 x-7 and ic 10 x-1 . we naturally expect ulxs to ionize the surrounding interstellar medium ( see * ? ? ? * ; * ? ? ? * ) for the basic principles of x - ray ionization . the main difference between uv ionized h@xmath119ii regions and x - ray ionized nebulae is the lack , in the latter , of a sharp boundary ( strmgren sphere ) between the ionized and neutral plasma @xcite . instead , there is an extended warm region where weakly ionized atoms coexist with collisionally excited neutral species . as a result , an x - ray ionized nebula is characterized by high - ionization emission lines such as he@xmath119ii @xmath2334686 and [ o@xmath119iv ] at 25.89 @xmath234 m in the inner regions , and low - ionization forbidden lines such as [ o@xmath119i ] @xmath2336300 in the outer regions . in particular , he@xmath119ii @xmath2334686 acts as a photon counter of the ionizing flux between 54 ev and @xmath235 ev @xcite . by comparing the true source flux inferred from the he@xmath119ii @xmath2334686 line with the apparent flux measured at energies @xmath236 ev with _ chandra _ or _ xmm - newton _ , we can test whether the x - ray emission is isotropic or beamed . the n159f nebula , powered by the smbh lmc x-1 , is the best - known example of a large - scale ( @xmath0 pc ) x - ray ionized nebula in galactic systems @xcite . @xcite opened a new field with their systematic search for similar nebulae around ulxs . they found a spectacular case around the ulx in m81 group dwarf galaxy holmberg ii . the holmberg ii x-1 ionized nebula ( now known as `` foot nebula '' ) has a he@xmath119ii @xmath2334686 luminosity @xmath237@xmath238 erg s@xmath12 @xcite , 30 times higher than the lmc x-1 nebula , which implies an average in the nebula , estimated as @xmath239 yr @xcite . ] ionizing x - ray flux @xmath240@xmath241 erg s@xmath12 . this result has been confirmed by further optical / infrared spectroscopic studies @xcite . on the other hand , the apparent x - ray luminosity of holmberg ii x-1 has been seen to vary over the years between @xmath17 a few @xmath16 erg s@xmath12 and @xmath242 erg s@xmath12 @xcite . the consistency between the two independent measurements gives the strongest evidence to date that beaming is negligible ( @xmath243 ) for this ulx ; this favors models with relatively large bh masses ( section [ sec : bh ] ) . similar arguments have also been applied to the photoionization - dominated nebula around ngc 5408 x-1 . from the flux in high - ionization lines such as he@xmath119ii @xmath2334686 and [ ne@xmath119v ] @xmath2333426 , @xcite inferred a lower limit on the isotropic x - ray luminosity @xmath244 erg s@xmath12 , only a factor of 3 lower than the directly measured x - ray luminosity . as in holmberg ii x-1 , this suggests that beaming factor is @xmath245 . more examples of x - ray ionized nebulae in ulxs are reported by @xcite , @xcite , and @xcite , although in those cases the interpretation is more model dependent , because the observed line spectra are a mix of photo - ionized and shock - ionized plasma emission . + + it is often hard to distinguish between x - ray photoionization and shock excitation , for ulx nebulae just as it is generally difficult to separate the contribution of a starburst nucleus and/or a low - luminosity agn in a liner . diagnostic diagrams with several line ratios are computed with shock and photoionization codes such as mappings iii @xcite and cloudy @xcite . shock - ionized plasma is characterized by high ratios of low - excitation lines such as [ o@xmath119i ] @xmath246 [ s@xmath119ii ] @xmath247,@xmath248 and [ n@xmath119ii ] @xmath249,@xmath250 over h@xmath118 . for shock velocities @xmath251@xmath252 km s@xmath12 , the high - ionization , forbidden transitions [ o@xmath119iii ] @xmath253,@xmath254 are strong ( [ o@xmath119iii ] @xmath233 5007/h@xmath255 for @xmath256 km s@xmath12 ) , while he@xmath119ii @xmath2334686 indicates @xmath257 km s@xmath12 @xcite . the luminosity in the h@xmath258 line is only a function of the total shock power @xmath259 and shock velocity @xcite ; for typical speeds @xmath260@xmath261 km s@xmath12 , @xmath262@xmath263 . supershells were first found around ngc 6946 mf 16 @xcite and holmberg ix x-1 @xcite , and were initially speculated to be powered by unusually energetic sn remnants . the x - ray emission could not be from the nebulae , because of its variability @xcite . following the pioneering work of @xcite , several more huge shock - ionized nebulae have been identified around ulxs ( `` ulx bubbles '' , see figure [ fig : bubble ] for examples ) , with characteristic diameter @xmath264@xmath265 pc , expansion velocity @xmath260@xmath266 km s@xmath12 , characteristic age @xmath267@xmath206 myr , mechanical power @xmath268@xmath19 erg s@xmath12 ( independently determined from the bubble expansion speed and from the h@xmath258 luminosity ; @xcite ) and energy content @xmath269 erg @xcite . the two best - studied examples are the bubbles around ngc 1313 x-2 and holmberg ix x-1 @xcite . ulx bubbles resemble ordinary sn remnants , but are an order of magnitude larger and 10 to 100 times more luminous . they can not come from ( hypothetical ) single hypernova explosions : first , because the surrounding stellar populations are already too old ( typically @xmath270 yr ) and without o stars ; secondly , because it is unlikely that a lower - mass binary companion that survived the hypothetical hypernova explosion could evolve to fill its roche lobe and feed the ulx already within @xmath271 myr @xcite . invoking 10 to 100 ordinary sn explosions over the last 1 myr would require a massive star cluster with @xmath272 a few @xmath273 ( _ e.g. _ , from starburst99 ; @xcite ) . however , none of those ulxs is associated with a massive star cluster : at most , they are inside ordinary ob associations with age @xmath0@xmath108 myr and total stellar mass @xmath92 . the only plausible conclusion is that ulx bubbles are inflated by a continuous jet or outflow from the ulx , with a power comparable to the radiative output @xcite . jet - inflated bubbles around accreting bhs are found in our galaxy , too . the best - known example is the w50 nebula around ss433 @xcite : a spherical shell with two lateral extensions , the ears , that are inflated by interaction of the jets with the interstellar medium . the characteristic diameter along the major axis is @xmath152 pc . both the jet power and the radiative luminosity are @xmath13 erg s@xmath12 . indeed , it is suggested that ss433 would appear as a beamed , super - eddington ulx if it was seen face - on @xcite . another local example is the 5-pc bubble inflated by the smbh cygnus x-1 @xcite : the shock velocity determined from optical emission line ratios is @xmath260@xmath274 km s@xmath12 ; the required jet power @xmath275 erg s@xmath12 is @xmath276@xmath206 times the x - ray luminosity . the ulx bubbles discovered so far have a quasi - spherical appearance , with no direct signature of relativistic , collimated jets ( as in the galactic source ss433 ) . the first extra - galactic super - bubble with collimated jets was recently found in the outskirts of the spiral galaxy ngc7793 @xcite . this source ( figure [ fig : bubble ] ) , known as s26 , has : x - ray and optical hot spots where the jet and counter - jet interact with the interstellar medium ; an optical nebula with a size of @xmath277 pc , expanding at @xmath278 km s@xmath12 ; an associated x - ray and radio nebula with bright radio lobes . the long - term - average jet power @xmath17 a few @xmath19 erg s@xmath12 @xcite , over a characteristic age @xmath279 yrs . the x - ray luminosity of the central bh is only @xmath280 erg s@xmath12 . this might be because the bh emission is highly obscured and we only see a scattered component ; or because it is beamed away from us ; or because the bh is a transient , currently in a radiative - faint state , but may appear as a ulx at other times . whatever the reason for the current radiative dimness , it is important to note that ngc7793 s26 has had collimated jets with a long - term average power at least of order of the eddington power ( and probably much greater ) of a stellar bh ( smbh or msbh ) . in other words , it is a jet - dominated system at super - eddington mass accretion rates . there is no physical principle that prevents mechanical power to exceed the eddington luminosity . however , the fact that it is collimated in a large - scale jet may be surprising . the canonical view @xcite of bh accretion states based mostly on the behavior of galactic transients is that collimated , steady jets are found at @xmath281 , when accretion is radiatively inefficient ( low / hard state ) . whether and how steady , collimated jets can be launched during super - eddington accretion phases remains an unsolved theoretical problem . a possible solution comes from the radiation - magnetohydrodynamic simulations of @xcite , who find that a jet can be accelerated up to mildly relativistic speed by radiation pressure , and collimated by the lorentz force of a magnetic tower made of wound - up toroidal magnetic field lines . the magnetic structure is itself kept collimated by the geometrically thick accretion flow that dominates in the innermost region when @xmath282 . an ultrapowerful ( rather than ultraluminous ) system such as ngc7793 s26 is the local - universe analog of a recently discovered class of quasars , dominated by mechanical power @xcite . in those quasars , the jet power is up to 25 times higher than the bolometric radiative luminosity ; in some of those systems , the jet power is near or above the eddington power . mechanically - dominated , radiatively inefficient quasars in the early universe may grow much faster than eddington - limited sources , and would provide strong mechanical rather than radiative feedback onto the surrounding gas . thus , understanding the inflow / outflow / radiation processes in ulxs has wider astrophysical relevance . moreover , radiative and mechanical feedback from hmxbs and ulxs themselves may have been an additional , important source of heating and reionization of the intergalactic medium at high redshifts @xcite . some ulx bubbles ( from both the photoionized and shock - ionized variety ) are associated to extended radio emission . the best - studied examples are holmberg ii x-1 @xcite and ngc5408 x-1 @xcite . both radio nebulae are dominated by optically thin synchrotron emission , somewhat similar but much larger ( @xmath283 pc and @xmath284 pc , respectively ) and more luminous than typical radio sn remnants . they are also an order of magnitude more luminous than the radio nebula around ss433 @xcite , which well matches the relative values of radiative and mechanical core power in those sources . the total power carried by synchrotron - emitting electrons has been estimated to be between @xmath17 a few @xmath285 erg s@xmath12 and @xmath286 erg s@xmath12 , depending on the very uncertain lower - energy cut - off of the distribution . in any case , both nebulae are much more energetic than w50 around ss433 ( @xmath17 a few @xmath287 erg s@xmath12 : @xcite ) or than the extended lobes of typical galactic microquasars such as grs1758@xmath93258 ( @xmath288 erg s@xmath12 ; @xcite ) . ngc7793 s26 also has a large radio nebula @xcite , interpreted as a combination of flat - spectrum free - free emission ( mostly in the central region ) and steep - spectrum synchrotron emission from the radio hot spots and lobes . the energy stored in relativistic electrons is @xmath17 a few @xmath289 erg s@xmath12 . the energy content in relativistic electrons for typical ulx nebulae is only @xmath290@xmath291 of the core power in particular , of the mechanical power when it can be reliably estimated from bubble size or optical line fluxes integrated over the characteristic age of the bubble . the main reason is that most of the input energy is transferred to non - relativistic electrons , protons and nuclei @xcite . the fraction of energy in non - radiating particles to that in relativistic electrons in a system such as ngc7793 s26 is similar to the values found by @xcite for a sample of agn , seyfert and giant elliptical galaxies . for the same jet power , agn , ulxs and galactic microquasars may be radio louder or radio quieter depending on the amount of plasma entrained by the jets . we speculate that an interesting test may come from comparisons of the radio loudness and lobe energy in low / hard state microquasars , probably dominated by magnetically accelerated jets @xcite , and in ulxs , possibly dominated by radiation - pressure accelerated jets / outflows @xcite , with more entrainment . there is no doubt that our understanding of the phenomenology of ulxs has greatly advanced in the past decade thanks to _ chandra _ , _ xmm - newton _ , and many other space- and ground - based facilities . however , the key physical issue is still unresolved : that is , whether they are powered by imbhs or normal stellar bhs . a few sources have emerged as strong imbh candidates : in particular , eso243@xmath9349 hlx-1 and m82 x-1 . on the other hand , for the majority of ulxs , there is no strong theoretical need nor compelling observational evidence for imbhs . in the absence of direct mass - function measurements from phase - resolved optical spectroscopy , we still have to rely on x - ray spectral and timing modeling and other indirect clues . reviewing the results that we have collected here , we suggest the following conservative statement : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ulxs are a diverse population ; msbhs with moderate super - eddington accretion seem to be the easiest solution to account for most sources up to luminosities @xmath17 a few @xmath19 erg s@xmath12 ; strong beaming ( @xmath292 ) can be ruled out for the majority of ulxs ; imbhs are preferred in a few exceptional cases . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ some of the following kinds of observations may provide further breakthroughs : a. measurements of the fast timing behavior down to a time scale of sub - seconds may probe the inner part of the inflow : ideally , we would search for possible high frequency features ( breaks and qpos ) that are found in galactic bhs at frequencies @xmath82 hz . this would require a new generation of x - ray telescopes with large effective area . b. determining the relative contribution of thermal emission and comptonization component is a key test for competing accretion models . for that , multilayer - coated x - ray telescopes with good sensitivity up to a few tens of kev are needed . c. detection of , or a more solid upper limit to broad iron k lines may constrain the properties of the inner disk . d. high - resolution x - ray spectroscopy with gratings or microcalorimeters may constrain the metal abundance and ionization state of the inflow , the mass and energy of the outflow , and probe the absorption from the surrounding interstellar medium . e. the ulx sample within @xmath293 mpc is almost complete , and it does not contain enough sources to permit statistical studies of the luminosity distribution above @xmath294 a few @xmath58 erg s@xmath12 . a homogeneous all - sky survey , complete for ulxs up to @xmath212 mpc would allow us to determine whether or not the luminosity distribution extends unbroken up to at least @xmath18 erg s@xmath12 . such survey will hopefully be successfully conducted by _ erosita _ ; f. repeated snapshot monitoring of nearby ulxs over the next few years will extend their long - term lightcurves enough to prove that their duty cycles and transient behavior are different from those of galactic bhs . this can be easily done with smaller x - ray missions ; g. phase - resolved spectroscopy would lead to the measurement of radial velocity curves and mass functions . attempts to do so with 8-m class telescopes have so far been inconclusive but have suggested that it will be easily done with future 30-m class telescopes ; h. integral - field spectroscopic studies of star forming galaxies , as well as low - frequency radio studies , may reveal fainter , older ulx bubbles or lobes , still expanding into the interstellar medium but no longer energized by a central source . this would constrain the duration of the ulx active phase ; i. searching for compact radio jets down to a sensitivity of @xmath295jy will constrain the bh mass , in relation to other classes of accreting bhs in the fundamental plane . this may be done in the near future with the _ evla _ and with _ ska _ pathfinders , and in the more distant future with the ska itself . in conclusion , we gather from this list that there is more than one way to skin a cat . many fundamental aspects of ulx physics will be understood with instruments already planned or under construction . thus , we anticipate a fruitful decade of ulx studies beyond the _ chandra _ and _ xmm - newton _ era . we thank andy fabian , sean farrell , luigi foschini , fabien gris , phil kaaret , kip kuntz , manfred pakull , tim roberts , mat servillat , tod strohmayer , doug swartz , dany vanbeveren , and luca zampieri for helpful discussions and comments . hf acknowledges funding support from the national natural science foundation of china under grant no . 10903004 and 10978001 , the 973 program of china under grant 2009cb824800 , the foundation for the author of national excellent doctoral dissertation of china under grant 200935 , the tsinghua university initiative scientific research program , and the program for new century excellent talents in university . rs acknowledges support from a curtin university senior research fellowship , and hospitality at the mullard space science laboratory , tsinghua university and the university of sydney during part of this work . , f. , pakull , m. w. , soria , r. , & motch , c. 2009 , in american institute of physics conference series , vol . 1126 , american institute of physics conference series , ed . j. rodriguez & p. ferrando , 201203 , m. w. , & gris , f. 2008 , in american institute of physics conference series , vol . 1010 , a population explosion : the nature & evolution of x - ray binaries in diverse environments , ed . r. m. bandyopadhyay , s. wachter , d. gelino , & c. r. gelino , 303307 , a. h. , galache , j. , kalogara , v. , linden , t. , kilgard , r. , zezas , a. , wolter , a. , & trinchieri , g. 2010 , in bulletin of the american astronomical society , vol . 36 , american astronomical society meeting abstracts # 215 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: x - ray observations of clusters of galaxies show that in the central regions of most clusters the cooling time of the intracluster gas is significantly less than a hubble time ( edge , stewart & fabian 1992 ; white , jones & forman 1997 ; peres 1998 ) . this cooling is thought to lead to a slow net inflow of material towards the cluster centre ; a process known as a cooling flow ( see fabian 1994 for a review ) . x - ray imaging data show that gas typically ` cools out ' throughout the central few tens to hundreds of kpc in clusters , with @xmath2 , where @xmath3 is the integrated mass deposition rate within radius @xmath4 ( thomas , fabian & nulsen 1987 ) . spatially resolved x - ray spectroscopy has confirmed the presence of distributed cooling gas in cooling flows , with a spatial distribution and luminosity in good agreement with the predictions from the imaging data and cooling - flow models ( allen & fabian 1997 ) . for some years , the primary uncertainty with the standard model for cooling flows was the fate of the cooled matter ( see fabian , nulsen & canizares 1991 ) . however , the discovery of large column densities of intrinsic x - ray absorbing material associated with cooling flows observed with the solid state spectrometer ( sss ) on the einstein observatory ( white 1991 ; johnstone 1992 ) opened one interesting possibility . follow - up spatially - resolved x - ray spectroscopy with the position sensitive proportional counter ( pspc ) on rosat ( allen 1993 ; irwin & sarazin 1995 ; allen & fabian 1997 ) confirmed the presence of intrinsic x - ray absorbing material in cooling - flow clusters and further showed this material to be distributed throughout , but centrally - concentrated within , the cooling flows . the x - ray data thus identify the intrinsic x - ray absorbing material as a plausible sink for the cooled gas deposited by the cooling flows . in this paper we examine the x - ray properties of the cooling flows in a sample of 30 of the most x - ray luminous ( @xmath5 ) clusters of galaxies known . using asca spectra and rosat high resolution imager ( hri ) data , we present independent determinations of the mass deposition rates in the cooling flows and measure the column densities of intrinsic x - ray absorbing material associated with these systems . we show that the cooling flow model provides a consistent description of the x - ray imaging and spectral data and that the observed masses of intrinsic x - ray absorbing material are in reasonable agreement with the masses expected to have been accumulated by the cooling flows over their lifetimes . we also discuss the evidence for mass deposition from the cooling flows in other wavebands . this is the final paper in a series that has examined the impact of cooling flows on mass measurements ( allen 1998 ) , the @xmath6 relation ( allen & fabian 1998a ) and metallicity measurements ( allen & fabian 1998b ) for x - ray luminous clusters . in this paper , we describe the data reduction and analysis procedures used in these works . a large number of other , previous studies have also analysed one or more of data sets included here ( see references in the papers listed above ) , although these studies have not , in general , examined the properties of the cooling flows in the clusters . exceptions are the deprojection analysis of peres ( 1998 ) , and the combined asca / rosat studies of allen ( 1996 ) , schindler ( 1997 ) , bhringer ( 1998 ) and rizza ( 1998 ) . the cosmological parameters @xmath7=50 , @xmath8 and @xmath9 are assumed throughout . our sample was identified from clusters contained in the rosat x - ray brightest abell - type cluster sample ( xbacs ; ebeling 1996 ) and rosat brightest cluster sample ( bcs ; ebeling 1998 ) with x - ray luminosities in the @xmath10 kev band exceeding @xmath11 . from this initial list , we selected for study those targets with asca x - ray spectra and rosat high resolution imager ( hri ) data available on the goddard space flight centre ( gsfc ) public archive as of 1997 july 15 . we supplemented this sample with a number of other , southern x - ray luminous clusters , known to exhibit strong gravitational lensing effects ( pks0745 - 191 , rxj1347.5 - 1145 , ms2137.3 - 2353 , ac114 ) and the distant , luminous cooling - flow clusters abell 1068 , 1704 and iras 09104 + 4109 . the primary goal of this project was to investigate the x - ray properties of cooling flows in x - ray luminous clusters and their effects on the integrated x - ray properties of their host systems . we have not included either the perseus ( abell 426 ) or coma ( abell 1656 ) clusters in our study , although these systems met our selection criteria , since both are too close ( @xmath12 and @xmath13 , respectively ) to allow their integrated cluster properties to be studied using the techniques used in this paper . detailed analyses of these clusters and other nearby cooling flows are discussed by allen ( 1999 ) . the clusters abell 370 , 115 and 1758 also met the selection criteria but were not included in our study since rosat hri images show them not be single , coherent structures but to consist of a number of merging subunits . the integrated x - ray spectra for such clusters will relate to the virial properties of the individual subclusters rather than the systems as a whole , and the assumptions of spherical symmetry and hydrostatic equilibrium required by the x - ray modeling will not apply . the x - ray images for the other clusters included in the sample do not exhibit any dramatic substructure that would suggest these assumptions to be invalid . our final sample consists of 30 clusters , spanning the redshift range @xmath14 , with a mean redshift of 0.21 . the asca ( tanaka , inoue & holt 1994 ) observations were made over a three - and - a - half year period between 1993 april and 1996 december . the asca x - ray telescope array ( xrt ) consists of four nested - foil telescopes , each focussed onto one of four detectors ; two x - ray ccd cameras , the solid - state imaging spectrometers ( sis ; s0 and s1 ) , and two gas scintillation imaging spectrometers ( gis ; g2 and g3 ) . the xrt provides a spatial resolution of @xmath15 arcmin half power diameter ( hpd ) in the energy range @xmath16 kev . the sis detectors provide good spectral resolution [ @xmath17 over a @xmath18 arcmin@xmath19 field of view . the gis detectors provide poorer energy resolution [ @xmath20 but cover a larger circular field of view of @xmath21 arcmin diameter . for our analysis we have used the screened event lists from the rev1 processing of the data sets available on the gsfc asca archive ( for a detailed description of the rev1 processing see the gsfc asca data reduction guide , published by gsfc . ) the asca data were reduced using the ftools software ( version 3.6 ) issued by gsfc , from within the xselect environment ( version 1.3 ) . further data - cleaning procedures as recommended in the asca data reduction guide , including appropriate grade selection , gain corrections and manual screening based on the individual instrument light curves , were followed . a summary of the asca observations , including the individual exposure times after all screening procedures were carried out , is provided in table 1 . spectra were extracted from all four asca detectors ( except in those few cases where the s1 data were lost due to saturation problems caused by flickering pixels in the ccds ) . the spectra were extracted from circular regions , centred on the peaks of the x - ray emission . for the sis data , the radii of the regions used were selected to minimize the number of chip boundaries crossed ( thereby minimizing the systematic uncertainties introduced by such crossings ) whilst covering as large a region of the clusters as possible . data from the regions between the chips were masked out and excluded . the final extraction radii for the sis data are summarized in table 2 . also included in that table are the chip modes used in the observations ( whether 1,2 or 4 chip mode ) and the number of chips from which the extracted data were drawn . for the gis data a constant extraction radius of 6 arcmin was used . [ we note that for abell 2142 , the 2 arcmin ( 3 arcmin ) radius region surrounding the x - ray bright seyfert-1 galaxy 1556 + 274 ( offset by @xmath22 arcmin from the x - ray centre of the cluster ) was masked out and excluded from the analysis of the sis ( gis ) data . ] for the gis observations , and sis observations of clusters in regions of low galactic column density ( @xmath23 ) , background subtraction was carried out using the ` blank sky ' observations of high galactic latitude fields compiled during the performance verification stage of the asca mission . for such data sets , the blank - sky observations provide a reasonable representation of the cosmic and instrumental backgrounds in the detectors . the background data were screened and grade selected in the same manner as the target observations and background spectra were extracted from the same regions of the detectors as the cluster spectra . for the sis observations of clusters in directions of higher galactic column density , background spectra were extracted from regions of the chips that were relatively free from foreground cluster emission . for the sis data , response matrices were generated using the ftools sisrmg software . where the spectra covered more than one chip , response matrices were created for each chip , which were then combined to form a counts - weighted mean matrix . for the gis analysis , the response matrices issued by gsfc on 1995 march 6 were used . the rosat hri observations were carried out between 1991 november and 1995 june . the hri provides a @xmath24 arcsec ( fwhm ) x - ray imaging facility covering a @xmath25 arcmin@xmath19 field of view ( david 1996 ) . reduction of the data was carried out with the starlink asterix package . x - ray images were created on a @xmath26 arcsec@xmath19 pixel scale , from which centres for the cluster x - ray emission were determined . where more than one observation of a source was made , a mosaic was constructed from the individual observations . for the cooling flow and intermediate clusters , the x - ray centres were identified from the peaks of the x - ray surface brightness distributions ( which are easily determined from the hri images ) . for the non - cooling flow clusters , the x - ray emission is not as sharply - peaked and for these systems we have identified the x - ray centres with the results from iterative determinations of the centroids within a circular aperture of fixed radius . for abell 665 , 2163 and ac114 a 1 arcminute radius aperture was used . for abell 2744 , 773 , 2218 and 2219 a 2 arcmin aperture was adopted , and for abell 520 a 3 arcmin aperture was used . for two of the clusters included in the sample , abell 665 and 1413 , rosat hri images were not available at the time of writing and for these clusters pspc imaging data have been used instead . a summary of the rosat observations and the x - ray centers for the clusters is given in table 3 . for the purposes of this paper , we have classified the clusters into subsamples of cooling - flow ( hereafter cf ) and non - cooling flow ( ncf ) systems . cfs are those clusters for which the upper ( 90 percentile ) limit to the central cooling time , as determined from the deprojection analysis of the rosat hri x - ray images ( section 4 ) , is less than @xmath27 yr . ( the ` central ' cooling time is the mean cooling time of the cluster gas in the innermost bin included in the deprojection analysis , which is of variable size . the use of a fixed physical size of 100 kpc for the central bin leads to similar results ; section 4 . ) using this simple classification scheme we identify 21 cfs and 9 ncfs in our sample . the mean redshift for the subsamples of both cf and ncf clusters is @xmath28 . the fraction of cfs in our sample is @xmath29 per cent , the same as that determined by peres ( 1998 ) from a study of a larger , flux - limited sample of clusters ( primarily at lower redshift ; @xmath30 ) using a similar classification scheme . the modeling of the x - ray spectra has been carried out using the xspec spectral fitting package ( version 9.0 ; arnaud 1996 ) . for the sis data , only counts in pulse height analyser ( pha ) channels corresponding to energies between 0.6 and 10.0 were included in the analysis ( the energy range over which the calibration of the sis instruments is best - understood ) . for the gis data only counts in the energy range @xmath31 were used . the spectra were grouped before fitting to ensure a minimum of 20 counts per pha channel , allowing @xmath32 statistics to be used . the spectra have been modeled using the plasma codes of kaastra & mewe ( 1993 ; incorporating the fe l calculations of liedahl , osterheld & goldstein 1995 , in xspec version 9.0 ) and the photoelectric absorption models of balucinska - church & mccammon ( 1992 ) . the data from all four asca detectors were analysed simultaneously ( except where the s1 data were lost due to chip saturation problems ) with the fit parameters linked to take the same values across the data sets . the exceptions to this were the emission measures of the ambient cluster gas in the four detectors which , due to the different extraction radii used and residual uncertainties in the flux calibration of the instruments , were allowed to fit independently . the spectra were examined with a series of spectral models . model a , consisted of an isothermal plasma in collisional equilibrium , at the optically - determined redshift for the cluster , and absorbed by the nominal galactic column density ( dickey & lockman 1990 ) . the free parameters in this model were the temperature ( @xmath33 ) and metallicity ( @xmath34 ) of the plasma and the emission measures in the four detectors . ( the metallicities are measured relative to the solar values of anders & grevesse ( 1989 ) , with the different elements assumed to be present in their solar ratios . ) the second model , model b , was the same as model a but with the absorbing column density @xmath35 also included as a free parameter in the fits . the third model , model c , was the same as model a but with an additional component explicitly accounting for the emission from the cooling flows . the material in the cooling flows is assumed to cool at constant pressure from the ambient cluster temperature , following the prescription of johnstone ( 1992 ) . the normalization of the cooling - flow component was parameterized in terms of a mass deposition rate , @xmath36 , which was a free parameter in the fits . the mass deposition rate is linked to take the same value in all four detectors , scaled by a normalization factor proportional to the total flux measured in that detector . the metallicity of the cooling gas was assumed to be equal to that of the ambient icm . the emission from the cooling flows was also assumed to be absorbed by an intrinsic column density , @xmath37 , of cold gas which was a further free parameter in the fits . the abundances of metals in the absorbing material were fixed to their solar values ( anders & grevesse 1989 ) although the effects of varying these values were examined . finally , a fourth spectral model , model d , was examined which was similar to model c but with the constant - pressure cooling flow replaced with a second isothermal emission component . the temperature and normalization of this second emission component were included as free parameters in the fits . ( as with model c , the normalizations were linked to the same value in all four detectors , scaled by the appropriate normalization factors . ) the second emission component was also assumed to be intrinsically absorbed by a column density of cold gas , @xmath37 , which was a free parameter in the fits . the metallicities of the two emission components were linked to take the same values . 1 shows the asca data and best - fitting spectral models for four of the clusters in the sample ; abell 1704 , 2029 , 2204 and 2219 . table 4 summarizes the fit results for all of the clusters using the four spectral models . for each cluster we list the best - fitting parameter values and 90 per cent ( @xmath38 ) confidence limits . the mass deposition rates ( @xmath39 ) , @xmath40 kev x - ray luminosities ( @xmath41 ) and bolometric luminosities ( @xmath42 ) are the values measured in the g3 detector . note that for the clusters at relatively low redshifts ( @xmath43 ) , we have corrected the @xmath41 and @xmath44 values for the emission arising at radii @xmath45 arcmin by scaling from the david ( 1993 ) results , which were based on ginga and einstein mpc observations . table 5 summarizes the goodness - of - fit measurements obtained with the different spectral models . the tabulated results are the probabilities of exceeding the @xmath32 values obtained , assuming in each case that the model correctly describes the spectral properties of the clusters . goodness - of - fit values @xmath46 may be regarded as indicating a formally unacceptable fit to the data . the results from the spectral analysis show that model a ( the isothermal model with the absorbing column density fixed at the nominal galactic value ) provides a reasonable description of most ( 21/30 ) of the clusters . however , model a fails significantly for the massive cf clusters abell 1068 , 1795 , 2029 and 2204 . the significance of the improvements to the fits obtained with spectral model b over model a have been evaluated using the f - test for the introduction of an additional free parameter ( bevington 1969 ) . these significances are listed in column 6 of table 5 . we see that the improvements obtained when including the column density as a free parameter in the fits with the isothermal models are significant at @xmath47 per cent confidence for 23 of the 30 clusters ( and would also be highly significant for abell 478 if the nominal dickey & lockman 1990 value for the galactic column density , rather than the x - ray value determined by allen & fabian 1997 were used in model a ) . model b provides a statistically acceptable fit for 27 of the 30 clusters . the results on the goodness - of - fit determined with spectral models c and d are summarized in columns 4 and 5 of table 5 . the only clusters that these models do not adequately describe are the two nearest , brightest systems ; abell 1795 and abell 2029 . spectral models c and d include an absorption component acting on the entire cluster emission which is normalized to the appropriate galactic column densities determined from hi studies ( dickey & lockman 1990 ) . we have examined the significance of the improvements to the fits obtained when allowing the galactic column density to also be a free parameter in the fits with these models . the fits were not significantly improved for any of the clusters except pks0745 - 191 , for which the galactic column density is most uncertain . ( the galactic column density to pks0745 - 191 varies between 3.2 and @xmath48 in the surrounding regions studied by dickey & lockman 1990 ) . we find @xmath49 with model c and @xmath50 with model d for pks0745 - 191 ( indicating improvements significant at @xmath51 per cent confidence ) when allowing the galactic column density to be free . the preferred value for the galactic column density for this cluster from the asca data is @xmath52 . for abell 1795 , we also find a marginal improvement to the fits with models c and d when allowing the galactic column density to be a free parameter ( @xmath53 with model c and @xmath54 for model d , which is significant at the 96.5 and 98 per cent level , respectively ) with a preferred value of @xmath55 . the agreement between the galactic column densities inferred from the 21 cm and x - ray data for most of the clusters with spectral models c and d provides further support for the validity of these models . the statistical improvements obtained by introducing an additional emission component into the fits with the single - temperature model ( model b ) are summarized in columns 7 and 8 of table 5 . we list the results obtained for both cases , where the extra emission component is modeled as a cooling flow ( for the cf systems only ) or a second isothermal emission component . ( essentially we use spectral models c and d , but also include the galactic column density as a free parameter it the fits so as to permit a direct comparison with the goodness of fit measured with spectral model b ) . the significance of the improvements to the fits obtained with the multiphase models with respect to model b have been quantified using an f - test for the introduction of 2 or 3 extra parameters ( for models c and d respectively ) . we find that the improvements with model d are significant at @xmath47 per cent confidence for 21 of the 30 clusters . the only clusters for which we find no significant improvement with the introduction of a second emission component are the ncf clusters abell 520 , 773 , 2218 and ac114 , and the more distant and/or less - luminous cf clusters abell 963 , ms1358.4 + 6245 , ms1455.0 + 2232 , ms2137.3 - 2353 and abell 2390 , which have amongst the lowest numbers of counts in their asca spectra . for the cf clusters ( the only clusters to which model c was applied ) the improvements obtained with the cooling - flow model over model b are significant at @xmath47 per cent confidence in 13 of 21 systems , with again those cf clusters with the lowest numbers of total counts in their asca spectra exhibiting the least - significant improvements . we thus find that most cf clusters , and a number of ncf systems ( in general those systems with the best signal - to - noise ratios in their asca spectra ) exhibit significant improvements to their fits with the use of multiphase , over single - phase , models . these improvements generally indicate the presence of emission from material cooler than the mean cluster temperatures ( see table 4 ) . for the cf clusters , this can be naturally understood as being due to emission associated with the cooling flows . for the ncf clusters , the improvements to the fits obtained with model d are likely to reflect the presence of merging subclusters with lower virial temperatures this is supported by optical , radio and x - ray imaging ( edge 1992 ; buote & tsai 1996 ; markevitch 1996 ; feretti , giovannini & bhringer 1997 ; rizza 1998 ) and gravitational lensing studies ( kneib 1995 ; smail 1995 , 1997 ; squires 1997 ; allen 1998 ) of the ncf clusters , which show that these systems typically exhibit complex morphologies and centroid shifts indicative of merger events . the results in table 4 show that spectral model d typically provides at least as good a fit to the asca spectra for the cf clusters as the cooling - flow model ( model c ) . this has sometimes been taken to indicate that the x - ray gas in these systems is distinctly two - phase , with the cooler phase being due to the dominant cluster galaxy ( makishima 1997 ; ikebe 1999 ) . however , when interpreting these results it is important to recall that the two - temperature model provides a more flexible fitting parameterization , with an extra degree of freedom , when applied to asca observations . simulated cooling - flow spectra constructed with spectral model c , including plausible levels of intrinsic absorption ( section 6 ) and observed at the spectral resolution and signal - to - noise levels typical of asca observations , are invariably well - described by two - temperature models . this is illustrated in fig . 2 where we show a simulated asca sis spectrum for a 7 kev cluster with a metallicity of 0.4@xmath56 , containing a cooling flow ( intrinsically absorbed by a column density of @xmath57 ) accounting for 30 per cent of the total @xmath40 kev flux . a count rate of 1ct s@xmath58 , an exposure time of 40ks and a galactic column density of @xmath0 have been assumed . overlaid , we show the best - fitting two - temperature model , which provides a good fit to the simulated data . thus , even in the case where the constant - pressure cooling flow model provides an exact description of the data ( _ i.e. _ in the simulations ) , the two - temperature model provides a similarly good fit . in a real cluster , where spectral model c will undoubtedly over - simplify the true situation , the more flexible two - temperature model ( model d ) is likely to provide a better match to the observations . thus , our finding that spectral model d often provides a better fit to the observed cluster spectra than model c , only shows that the constant pressure cooling flow model over - simplifies the true spectra of the cooling flows in the clusters ( see also section 6.4 ) . we note that such tendencies would be enhanced by the presence of additional cool components in the spectra , for example due to temperature gradients at large radii ( markevitch 1998 ) . finally , we note that the very high luminosities associated with the cool emission components in clusters like rxj1347.5 - 1145 ( with @xmath59 a few @xmath11 ) are difficult to explain with models in which the cool emission is due to the interstellar medium of the central cluster galaxies . the analysis of the hri imaging data has been carried out using an extensively updated version of the deprojection code of fabian ( 1981 ) . azimuthally - averaged x - ray surface brightness profiles were determined for each cluster from the hri images . these profiles were background - subtracted , corrected for telescope vignetting and re - binned to provide sufficient counts in each annulus to allow the analysis to be extended to radii of at least 500 kpc . ( bin sizes of 8 - 24 arcsec were used . ) with the x - ray surface brightness profiles as the primary input , and under assumptions of spherical symmetry and hydrostatic equilibrium , the deprojection technique can be used to study the basic properties of the intracluster gas ( temperature , density , pressure , cooling rate ) as a function of radius . the deprojection code uses a monte - carlo method to determine the statistical uncertainties on the results and incorporates the appropriate hri spectral response matrix issued by gsfc . the cluster metallicities were fixed at the values determined from the spectral analysis ( table 4 ) . the absorbing column densities were fixed at the appropriate galactic values ( table 1 ) . the deprojection code requires the total mass profiles for the clusters ( which define the pressure profiles ) to be specified . we have iteratively determined the mass profiles that result in deprojected temperature profiles ( which approximate the mass - weighted temperature profiles in the clusters ) that are isothermal within the regions probed by the hri data ( the central @xmath60 mpc ) and which are consistent with the spectrally - determined temperatures ( section 3 ) . the assumption of approximately isothermal mass - weighted temperature profiles in the central regions of the clusters is supported by the following evidence : firstly , asca observations of nearby cooling flows show that in the central regions of these systems the gas is multiphase , but that the bulk of the x - ray gas there has a temperature close to the cluster mean ( fabian 1994b ; fukazawa 1994 ; matsumoto 1996 ; ikebe 1999 ) . secondly , combined x - ray and gravitational lensing studies of cf clusters ( allen 1998 ) show that approximately isothermal mass - weighted temperature profiles in the cores of such clusters lead to excellent agreement between their x - ray and gravitational - lensing masses . thirdly , the use of approximately constant mass - weighted temperature profiles implies a more plausible range of initial density inhomogeneities in the clusters than would be the case if the temperature profiles decreased within their cores ( thomas , fabian & nulsen 1987 ) . finally , the use of approximately isothermal mass - weighted temperature profiles in the deprojection analyses leads to independent determinations of the mass deposition profiles in the cooling flows , from the x - ray spectra and imaging data , in excellent agreement with each other ( allen & fabian 1997 ) . we note that the assumption of a constant mass - weighted deprojected temperature profile is consistent with measurements of a decreasing emission - weighted temperatures in the cores of many cf clusters ( waxman & miralda - escud 1995 ) . the mass profiles for the clusters were parameterized as isothermal spheres ( equation 4 - 125 of binney & tremaine 1987 ) with adjustable core radii ( @xmath61 ) and velocity dispersions ( @xmath62 ) . the core radii were adjusted until the temperature profiles determined from the deprojection code became isothermal . the velocity dispersions were then adjusted until the temperatures determined from the deprojection code came into agreement with the spectrally - determined values . errors on the velocity dispersions are the range of values that result in isothermal deprojected temperature profiles that are consistent , at the 90 per cent confidence limit , with the spectrally - determined temperature results . estimates of the thermal gas pressure in the outermost radial bins used in the analysis are also required by the deprojection code and were determined iteratively . ( the uncertainties on the outer pressure estimates do not significantly affect the results presented here . ) although the deprojection method of fabian ( 1981 ) is essentially a single - phase technique , we note that it produces results in good agreement with more detailed multiphase treatments ( thomas , fabian & nulsen 1987 ) and , due to its simple applicability at large radii in clusters , is better - suited to the present project . the mass distributions determined from the deprojection analysis are summarized in columns 3 and 4 of table 6 . these simple parameterizations permit direct comparisons with independent mass constraints from gravitational lensing and dynamical studies ( allen 1998 ) . we note that for a few of the clusters ( primarily the brightest cfs with the best data ) the single - component mass models can not adequately satisfy the requirement for approximately isothermal deprojected temperatures profiles and for these systems a significantly better match was obtained by introducing a second ` linear ' mass component , truncated at a specified outer radius . the clusters requiring the additional mass components were abell 586 ( @xmath63 kpc@xmath58 within the central 20 kpc ) , abell 2029 ( @xmath64 kpc@xmath58 within the central 20 kpc ) , abell 2219 ( @xmath65 kpc@xmath58 within the central 30 kpc ) , abell 478 ( @xmath66 kpc@xmath58 within the central 20 kpc and an adjustment of the cluster parameters to @xmath67 and @xmath68 kpc ) and abell 1795 ( @xmath65 kpc@xmath58 within the central 20 kpc , and an adjustment of the cluster parameters to @xmath69 and @xmath70 kpc ) . the core radii of the mass distributions determined from the deprojection analysis are similar to the values measured from simple ` @xmath71model ' fits to the x - ray surface brightness profiles ( cavaliere & fusco - femiano 1976 , 1978 ) . it is well known that the @xmath72-model does not provide a good match to the x - ray surface brightness profiles of the brightest cf clusters when their central regions are included in the analysis ( jones & forman 1984 ) . however , where @xmath72-models are used to estimate the mass core radii in cf clusters , the central regions of these clusters should not simply be excised from the fits . the central surface brightness peaks in cf clusters trace the presence of sharp central density rises in these systems and are not only the result of excess emission due to cooling gas . ( it is the high central densities that lead to the high cooling rates . ) although fully accounting for the effects of cooling on the central density measurements is difficult with current data , simply excising the inner regions of the clusters can lead to overestimates of the mass core radii and corresponding underestimates of the central gas and total mass densities . the basic results on the cooling flows from the deprojection analysis are summarized in columns @xmath73 of table 6 . for those clusters in common , the results obtained are generally in good agreement with values reported from previous works ( edge 1992 , white 1997 ; peres 1998 ) . the spectral and image deprojection analyses discussed in sections 3 and 4 provide essentially independent estimates of the mass deposition rates in the clusters . a comparison of the results obtained from these analyses therefore provides a test of the validity of the cooling flow model . the deprojection method describes the x - ray emission from a cluster as arising from a series of concentric spherical shells . the luminosity in a particular shell , @xmath74 , may be written as the sum of four components ( arnaud 1988 ) . @xmath75 , \label{swa_eq1}\ ] ] where @xmath76 is the mass deposited in shell @xmath74 , @xmath77 is the enthalpy of the gas in shell @xmath74 , and @xmath78 is the gravitational energy released in crossing that shell . @xmath79 is the mass flow rate through shell @xmath74 , and @xmath80 the change of enthalpy of the gas as it moves through that shell . the first term in equation [ swa_eq1 ] thus accounts for the enthalpy of the gas deposited in shell @xmath74 . the second term is the gravitational work done on the gas deposited in shell @xmath74 . the third and fourth terms respectively account for the gravitational work done on material flowing through shell @xmath74 to interior radii , and the enthalpy released by that material as it passes through the shell . in any particular shell , the densest material in the cooling flow is assumed to cool out and be deposited . since the cooling time of this material will be short compared to the flow time , the cooling can be assumed to take place at a fixed radius . thus , the luminosity contributed by the first term in equation 1 should have a spectrum that can be approximated by gas cooling at constant pressure from the ambient cluster temperature the same spectrum as the cooling component incorporated into the spectral analysis with model c ( section 3 ) . for the bulk of the material continuing to flow inwards towards the cluster centre , the cooling via x - ray emission is assumed to be offset by the gravitational work done on the gas as it moves inwards . the emission accounted for in the second and third terms of equation 1 should therefore have a spectrum that can be approximated by an isothermal plasma at the appropriate ambient temperature for the cluster the spectrum of the isothermal emission component in model c. since the mass - weighted temperature profiles in the clusters are assumed to remain approximately isothermal with radius , the luminosity contributed by the fourth term of equation [ swa_eq1 ] should be negligible . the mass deposition rates , @xmath81 , listed in table 6 are the mass flow rates ( @xmath82 ) determined from the deprojection analysis , at the point where the mean cooling time of the cluster gas first exceeds the hubble time ( @xmath83 yr ) . thus , if the cooling flow model is correct , and the cooling flows have been undisturbed for a significant fraction of a hubble time , the mass deposition rates determined from the deprojection analysis should be similar to those measured independently from the spectral data . 3 shows the mass deposition rates determined from the spectral analysis ( @xmath39 ) versus the image deprojection results ( @xmath81 ) , for those clusters for which both quantities have been measured . we see that the results exhibit an approximately linear correlation ; a fit to the data with a power - law model of the form @xmath84 , using the akritas & bershady ( 1996 ) bisector modification of the ordinary least - squares statistic , gives a best - fitting power - law slope of @xmath85 ( where the error is the standard deviation determined by bootstrap re - sampling ) , although the spectrally - determined mass deposition rates typically exceed the deprojection results by a factor of @xmath86 . however , the spectral analysis presented in section 3 also requires that the cooling gas is intrinsically absorbed by equivalent hydrogen column densities of , typically , a few @xmath1 ( model c ) . the mass deposition rates determined from the deprojection analysis must therefore also be corrected for the effects of this absorbing material . the corrections for the effects of intrinsic absorption on the deprojection results have been carried out by re - running the deprojection analysis with the absorbing column densities set to the total values determined with spectral model c. we assume that the absorption is due to cold gas with solar metallicity . if , however , the intrinsic absorption were instead due to dust , a possibility examined in more detail in section 7 , then the required correction factors would likely be reduced by a few tens of per cent . ( the introduction of a simple oik absorption edge at @xmath87kev , such as might be associated with oxygen - rich , silicate dust grains , generally provides as good a description of the intrinsic absorption detected in the @xmath88kev asca spectra as a cold , gaseous absorber . ) since the intrinsic column densities measured with model c are redshifted quantities , we set the total column densities used in the revised deprojection analysis , which assumes zero redshift for the absorber , to be @xmath89 , where @xmath90 is the galactic column density ) . the absorption - corrected mass deposition rates ( @xmath91 ) so determined are summarized in table 7 . 4 compares the spectrally - determined mass deposition rates with the absorption - corrected deprojected values . we see that once the presence of intrinsic absorption has been accounted for in a consistent manner in the imaging and spectral analyses , the results on the mass deposition rates in the clusters show good agreement . table 8 lists the fractions of the total @xmath40 kev x - ray fluxes from the cf clusters associated with cooling gas , as determined with spectral model c. the results range from @xmath92 per cent for systems like abell 1689 and 2142 to @xmath93 per cent for zwicky 3146 , abell 1068 , rxj1347 - 1145 and abell 2204 . such a range of measurements is consistent with the results of peres ( 1998 ) , from a deprojection analysis of a flux - limited sample of the brightest clusters known . the fraction of the x - ray flux from a cluster due to it s cooling flow can be expected to increase with the age of the flow . it is therefore likely that the cooling flows in abell 1689 and 2142 are younger than those in abell 2204 and zwicky 3146 . we note that the x - ray data are not , in general , of sufficient quality to place firm constraints on the ages of the cooling flows in these clusters ( for a discussion of cooling flow age measurements see allen 1999 ) . it is interesting that the clusters with the largest cooling - flow flux fractions ( with the exception of abell 1413 and 2261 ) are also amongst the most optically line - luminous cooling flows known ( crawford 1999 ) . the production of powerful optical emission lines and associated uv / blue continuum emission is known to be due , at least in part , to the formation of massive stars at the centres of cooling flows ( johnstone , fabian & nulsen 1987 ; allen 1995 ; cardiel 1995 , 1998 ; mcnamara 1996 ; voit & donahue 1997 ) . the results presented here are consistent with the idea that such star formation is typical of undisturbed cooling flows . it is also interesting to note that the clusters with the largest cooling flow flux fractions and most optically line - luminous emission line nebulae also tend to have the shortest cooling times averaged over their central 100 kpc regions , with @xmath94gyr ( see also peres 1998 ) . the equivalent column densities of intrinsic x - ray absorbing material inferred from the spectral analysis are summarized in table 9 . all three of the spectral models incorporating a free - fitting absorption component ( models b , c , d ) reveal the presence of excess x - ray absorption over and above the nominal galactic values along the lines of sight to the clusters ( dickey & lockman 1990 ) , although the measured column densities are sensitive to the spectral model used . using the simple isothermal model with free - fitting absorption ( model b ) , we determine a mean excess column density for the whole sample of 30 clusters of @xmath95 . when we examine the cf and ncf clusters separately , we find that both subsamples exhibit intrinsic absorption , at a similar level on average . for the cf clusters , the mean excess column density measured with spectral model b is @xmath96 . for the ncf systems , the value is @xmath97 . the application of a students t - test ( accounting for the possibility of unequal variances in the two distributions ; press 1992 ) shows the mean excess column densities for the cf and ncf clusters , determined with spectral model b , to differ at only the 45 per cent confidence level . the application of a kolmogorov - smirnov test shows the two subsamples to be drawn from different parent populations at only the @xmath92 per cent confidence level . thus , our analysis with spectral model b suggests that intrinsic absorption is not confined to cf clusters . the results on the column densities are quite different , however , when the more - sophisticated multiphase spectral models are used . as discussed in section 3 , the multiphase models generally provide a better description of the x - ray properties of the cf clusters ( and some of the brighter ncf systems ) . using spectral model c , our preferred model for the cf clusters in that it provides a consistent description of the spectral and imaging x - ray data for these systems ( section 5.1 ) , we determine a mean intrinsic column density acting on the cooling - flow components of @xmath98 . using the two - temperature spectral model ( model d ) , we determine a mean intrinsic column density acting on the cooler emission components in the cf systems of @xmath99 . thus , the intrinsic column densities for the cf clusters measured with the multiphase models ( models c and d ) are similar and approximately an order of magnitude larger than the values inferred using the single - phase model b. these results demonstrate the need for adopting an appropriate spectral when attempting to measure the column densities of absorbing material in cf clusters . we also note that where spectral model d provided a significant improvement to the fits to the ncf clusters , with respect to model b , the measured column densities acting on the cooler emission components were also significantly larger than the values determined using the single - phase models . we have also measured the intrinsic column densities in the cf clusters using one further spectral model , referred to in table 9 as model c. model c is identical to model c except that it assumes that the excess absorption acts on the entire cluster spectrum , rather than just the cooling gas , and that the absorbing material lies at zero redshift . model c has been used in a number of previous studies and is included here for comparison purposes . the mean excess column density for the cf clusters determined with spectral model c is @xmath100 . allen & fabian ( 1997 ) present results from an x - ray colour deprojection study of 18 clusters observed with the rosat pspc . these authors determine intrinsic column densities across the central 30 arcsec ( radius ) regions of abell 478 , 1795 and 2029 , using spectral model c , of @xmath101 , @xmath102 and @xmath103 , respectively . the results for abell 478 and abell 1795 are in excellent agreement with those presented here , although the allen & fabian ( 1997 ) value for abell 2029 is @xmath22 times smaller than our result . abell 2029 is unusual in that it hosts a strong cooling flow without associated optical line emission . this may indicate that the central regions of the cluster have been disrupted ( perhaps by a minor merger event or the strong central radio source in this system ) , complicating the distributions of cooling and absorbing material ( _ cf . _ section 5.2 ) . four of the cf clusters studied here were also examined by white ( 1991 ) , using einstein observatory sss data . these authors measured intrinsic column densities for abell 478 , 1795 , 2029 and 2142 ( in a 3 arcmin radius circular aperture , using spectral model c ) of @xmath104 , @xmath105 , @xmath106 and @xmath107 , respectively . ( the sss result for abell 478 has been corrected to account for the different value of galactic absorption assumed in that study ) . the intrinsic column densities determined by white ( 1991 ) are @xmath108 times larger than the values measured from the asca data . allen & fabian ( 1997 ) compared the results from their x - ray colour deprojection study of pspc data with the white ( 1991 ) sss analysis and concluded that if the intrinsic absorption were due to cold gas , then agreement between the measured column densities could only be obtained if the covering fraction of the x - ray absorbing material were , in general , @xmath109 . we have therefore examined the constraints on the covering fraction , @xmath110 , that can be obtained from the asca data using spectral model c. the results are also listed in table 9 . in all cases we determine a best - fit covering fraction for the intrinsic x - ray absorbing material of unity . for many clusters , covering fractions of @xmath111 per cent can be firmly ruled out . ( the physical significance of these results are further explored in section 6.4 ) . the results on the covering fractions , and the reasonable agreement of the asca and pspc results for abell 478 and 1795 , suggest that the white ( 1991 ) results may have systematically over - estimated the intrinsic column densities in clusters by a factor of a few . before considering in more detail the possible origin and nature of the absorbing material , it is pertinent to consider whether the intrinsic absorption , inferred to be present using a variety of spectral models , could in fact be an artifact due to an unconsidered emission process in the clusters . 5 shows the sis0 spectra for abell 1068 and 1795 , which exhibit two of the clearest absorption signatures . the data between @xmath112 and @xmath113 kev have been fitted with a single - temperature emission model with galactic absorption ( _ i.e. _ spectral model a ) . the residuals to these fits , over the full @xmath88 kev band , are shown in the lower panels . we see that the residuals , which can be naturally explained by the introduction of cool emission components with associated intrinsic absorption ( using spectral models c and d ) , take the form of a large excess of counts between @xmath114 kev . we therefore include the caveat that if some extra emission process which can account for such residuals in the x - ray spectra were identified , then the requirement for intrinsic absorption from the x - ray data would be greatly diminished . finally , we note that the fits with spectral model a over the restricted @xmath115 energy range lead to determinations of the cluster temperatures for abell 1068 and 1795 of @xmath116 and @xmath117 kev , respectively , in excellent agreement with the results obtained for the full data sets with spectral model c. the results presented in section 6.1 show that the cooling flows in our sample are typically intrinsically absorbed by equivalent hydrogen column densities of a few @xmath118 ( taking spectral model c as our preferred model ) . assuming , in the first case , that this absorption is due to cold gas with solar metallicity , the mass of absorbing gas , @xmath119 , within a radius , @xmath120 , may be approximated as @xmath121 where @xmath120 is in units of kpc and @xmath37 in @xmath1 . for metallicities of @xmath122 in the absorbing gas ( the mean emission - weighted value for the x - ray emitting gas in the cf clusters ; allen & fabian 1998b ) the implied masses are a factor @xmath123 larger . for zero metallicity these masses are a factor @xmath15 times larger than the values given by equation 2 . ( the scaling factors for different metallicities in the absorbing material have determined from spectral simulations using the xspec code . ) we have adopted @xmath120 as the radius where the cooling time of the cluster gas first exceeds @xmath124 yr in the clusters ( a plausible age for the cooling flows in most of the cf systems ) . these radii , and the masses of absorbing gas implied by equation 2 ( using the column densities determined with spectral model c ) for both solar metallicity and @xmath125 , are listed in columns @xmath126 of table 10 . the observed masses of absorbing material may be compared to the masses expected to have been accumulated by the cooling flows @xmath127 within the same radii over the past @xmath124 yr . assuming that from time @xmath128 to @xmath129 yr , the integrated mass deposition rate within radius @xmath120 increases approximately linearly with time , the total mass accumulated within radius @xmath120 after @xmath124 yr is approximately @xmath130 at times @xmath131 yr , the mass deposition rate within @xmath120 should remain approximately constant . if instead we assume @xmath132 to be constant during the first 5 gyr , the accumulated mass will be twice the value indicated by equation 3 . ( exactly how the mass deposition rate within @xmath120 grows with time during the first 5 gyr is unclear and will depend upon the evolution of the cluster and the detailed properties of the cluster gas . ) the mass deposition rates within radii @xmath120 ( corrected for the effects of absorption due to cold gas , as discussed in section 5.1 ) and the estimated masses of cooled material accumulated by the cooling flows within these radii over a 5 gyr period , are summarized in table 10 . 6 compares the accumulated masses with the masses of absorbing material determined from the spectral data . the agreement between the results is reasonable and is maximized for a metallicity in the absorbing gas of @xmath133 ( depending on the growth of the cooling flow over the first 5gyr ) . this supports the idea that the observed x - ray absorption may be due to material accumulated by the cooling flows . we note that if the x - ray absorption in the cf clusters were due to dust rather than cold gas , a possibility examined in more detail in section 7 , then the agreement between the observed and predicted masses of absorbing material shown in fig . 6 might still be expected to hold if the dust were contained in the material deposited by the cooling flows . in this case , the mass of the absorber calculated with equation 2 would be the mass of cooled gas associated with the dust when it was deposited from the cooling flow . for a galactic dust to gas ratio , the mass in dust would be @xmath134 times smaller than the associated gas mass . the luminosities absorbed at x - ray wavelengths must eventually be reprocessed in other wavebands . if the absorbing material is dusty ( as is likely to be the case in the central regions of the clusters voit & donahue 1995 ; fabian , johnstone & daines 1994 ; allen 1995 ) then the bulk of this reprocessed emission is likely to emerge in the infrared band ( dwek , rephaeli & mather 1990 ) . table 11 summarizes the reprocessed luminosities ( _ i.e. _ the bolometric luminosities absorbed within the clusters ) measured with spectral model c. the reprocessed luminosities range from @xmath135 for abell 1795 to @xmath136 for rxj1347.5 - 1145 , with a mean value of @xmath137 . allen ( 1999 ) report detections of spatially extended @xmath138 m emission coincident with the x - ray centroids in the nearby centaurus cluster and abell 2199 . the infrared fluxes from those clusters were also shown to be in good agreement with the predicted values due to reprocessed x - ray emission from their cooling flows . we have also used the ipac scanpi software and archival iras data to measure the 60 and @xmath138 m fluxes within a four arcmin ( radius ) aperture centred on the x - ray centroids for the clusters in the present sample . ( the medians of the co - added scanpi results were used . ) the results are listed in table 11 . the error bars associated with the measured fluxes are the root - mean - square deviations in the residuals , external to the source extraction regions , after baseline subtraction . where no detection was made , an upper limit equal to three times the r.m.s . deviation in the residuals is listed . where a detection was made , we also list the in - scan separations ( in arcmin ) between the peak of the @xmath138 m emission ( or the @xmath139 m emission in the case of abell 1835 ) and the x - ray centre . two of the clusters in our sample , iras 09104 + 4109 and abell 1068 have known infrared point sources coincident with their x - ray centroids ( kleinmann 1988 ; moshir 1989 ) . the presence of these sources is confirmed by our analysis . the 60 and @xmath138 m data for abell 1704 and @xmath139 m data for abell 1835 also suggest the presence of unresolved sources , coincident with the x - ray centroids for the clusters . abell 478 and zwicky 3146 have @xmath138 m emission originating from close to their x - ray centroids , which appears spatially extended , although the infrared flux from abell 478 is probably contaminated by galactic cirrus . the data for abell 2142 , 2204 , 2261 and 2390 also provide significant detections within the four arcmin ( radius ) source apertures , although the peaks of the infrared emission from these systems are spatially offset from their x - ray centroids , suggesting that the detected flux is likely to originate , at least in part , from some other source . several of the clusters included in our study ( abell 586 , 963 , 1413 , 1795 , 2029 , 2142 and pks0745 - 191 ) have previously been studied at infrared wavelengths , using iras data , by wise ( 1993 ) and/or cox , bregman & schombert ( 1995 ) . these authors also report no clear detections of infrared emission from these sources . edge ( 1999 ) report measurements of @xmath139 m emission from abell 1835 and 60 and @xmath138 m emission from abell 2390 , using a similar analysis to that presented here . these authors also present detections of @xmath140 m emission from these clusters which , for abell 1835 , they suggest is likely to be due to dust heated by the vigorous star formation observed in the cluster core or an obscured active galactic nucleus . following helou ( 1988 ) and wise ( 1993 ) , we can estimate the total infrared luminosities implied by the observed iras fluxes and flux limits using the relation @xmath141 where @xmath142 and @xmath143 are the 60 and @xmath138 m iras fluxes in units of jy . this relation assumes a dust temperature of @xmath144k ( which is consistent with the observations ; allen 1999 ; edge 1999 ) and an emissivity index , @xmath145 , in the range @xmath146 , where the emissivity is proportional to the frequency , @xmath147 . where measurements , rather than upper limits , were obtained , we associate a systematic uncertainty of @xmath148 per cent with the estimated @xmath149 m luminosities , which is combined in quadrature with the random errors . the @xmath149 m luminosities calculated from this relation are summarized in table 11 . in general , the measurements and upper limits to the observed @xmath149 m luminosities exceed the predicted luminosities due to reprocessed x - ray radiation from the cooling flows . where measurements ( rather than upper limits ) are made , this may suggest the presence of an additional source of infrared flux , such as star formation or agn associated with the central cluster galaxies , or some other contaminating ( possibly galactic ) source . at some level , the massive starbursts at the centres of systems like abell 1068 , 1835 , 2204 2390 and zwicky 3146 ( allen 1995 ; crawford 1999 ; edge 1999 ) must contribute to the detected infrared flux . as discussed in section 6.1 , the measured intrinsic equivalent hydrogen column densities of absorbing material inferred from the asca spectra exhibit significant systematic variation , depending upon whether the single - phase or multiphase emission models are used . we have adopted the results obtained with the cooling - flow model ( model c ) as our preferred values , since this model provides a consistent description of the spectral and imaging x - ray data for the clusters . however , a number of systematic uncertainties affecting the absorption results remain with the cooling flow model . spectral model c undoubtedly over - simplifies the situation in a real cooling flow . in particular , the model assumes that the absorbing material lies in a uniform screen in front of the cooling flows . if the absorber is intrinsically associated with the clusters , then the absorbing material must , to some extent , be distributed throughout the cooling flows . allen & fabian ( 1997 ; see also wise & sarazin 1999 ) discuss two different , limiting geometries for the absorbing material ; partial covering and multilayer absorption . the partial covering model , with a covering factor @xmath150 , can be expected to apply or where there are @xmath151 individual absorbing clouds along each line of sight and/or where the absorbing material is strongly clumped on large scales . however , as discussed in section 6.1 , covering fractions significantly less than unity are found to provide a poor description of the asca spectra , suggesting that the absorbing material is likely to be more smoothly distributed throughout the cooling flows . the multilayer absorption model discussed by allen & fabian ( 1997 ) is applicable where the absorbing material is made up of a large number of similarly sized absorbing clouds , each with a column density much less than the total intrinsic column density ( @xmath37 ) , homogeneously distributed throughout the x - ray emitting medium . for an intrinsic emission spectrum @xmath152 , the observed spectrum , @xmath153 , emerging from the emitting / absorbing region may be written ( where @xmath154 is the absorption cross - section ) as @xmath155 we have simulated cooling flow spectra with multilayer absorption , with total column densities of between @xmath1 and @xmath156 , and fitted these with simple models which assume that the absorbing material acts as a uniform screen in front of the cooling flow ( as with model c ) . for the simulations we adopt an emission spectrum appropriate for a constant pressure cooling flow , with an upper temperature of 7 kev and a metallicity of @xmath157 . we use a response matrix appropriate for the asca sis detectors . the comparison between the fitted ( uniform screen ) column densities and the true total ( multilayer ) values are shown in fig . we see that for multilayer column densities of @xmath158 , the fitted column densities underestimate the true values by factors of @xmath159 . ( these results are not sensitive to any reasonable choice of galactic column density . ) if the multilayer model approximates the true distribution of x - ray emitting and absorbing material in cooling flows , then the results obtained with spectral model c would imply true equivalent column - densities of intrinsic absorbing gas in the cluster cores of , typically , a few @xmath160 . importantly , we also note that for signal - to - noise ratios appropriate for asca observations of bright clusters ( with count rates of @xmath161ct s@xmath58 in the sis detectors and exposure times of @xmath162 ks ) , the simple uniform - screen absorption model provides a good fit to the simulated spectra with multilayer absorption , for total column densities @xmath163 . ( for larger total column densities , complex residuals are detected in the spectra at energies @xmath164kev . ) the large intrinsic column densities implied by the multilayer models would have strong implications for the nature of the absorbing matter . in particular , column densities of @xmath165 of gaseous material with solar metallicity would be very difficult to reconcile with current hi and co limits , even if this material were very cold and highly molecular . ( we also note that in the case of a multilayer distribution , the implied masses of absorbing gas would often exceed the predicted accumulated masses due to the cooling flows and imply , at least in part , some other origin for the absorbing medium . ) in the following section , we explore these issues in more detail and summarize the current observational constraints on the physical nature of the absorbing matter . the results presented in this paper provide a consistent picture for the x - ray properties of cooling flows in x - ray luminous clusters of galaxies . the asca spectra and rosat images provide independent determinations of the mass cooling rates in the clusters in good agreement with each other , when a consistent method of analysis is employed ( section 5.1 ) . the spectral data require the presence of large column densities of intrinsic x - ray absorbing material associated with the cooling gas ( section 6 ) . the implied masses of absorbing material are in reasonable agreement with the masses expected to have been accumulated by the cooling flows over their lifetimes ( section 6.2 ) , identifying this material as a plausible sink for the bulk of the cooled gas from the flows . despite the evidence for intrinsic x - ray absorption associated with cooling flows , however , searches for this material in other wavebands have , to date , proved largely unsuccessful ( at least outside of the central @xmath166 kpc ) . we here briefly review these observations and comment on their implications for the physical state of the x - ray absorbing matter . more detailed discussions of a number of these issues are given by ferland , fabian & johnstone ( 1994 ) , daines , fabian & thomas ( 1994 ) , fabian ( 1994a ) , odea ( 1994a ) , voit & donahue ( 1995 ) , odea & baum ( 1996 ) , arnaud & mushotzky ( 1988 ) and henkel & wiklind ( 1998 ) . as mentioned in section 5.2 , the central ( @xmath166 kpc ) regions of cf clusters are commonly observed to be sites of ongoing star formation with powerful , associated optical line emission ( johnstone 1987 ; heckman 1989 ; mcnamara & oconnell 1989 ; crawford & fabian 1992 ; allen 1995 ; cardiel 1995 , 1998 ; mcnamara 1996 ; voit & donahue 1997 ; crawford 1999 ) . nearly all cf clusters with short central cooling times ( @xmath167gyr ) exhibit such phenomena ( peres 1998 ) , whereas systems with longer central cooling times ( including all ncf clusters ) generally do not . the observed levels of star formation imply the presence of at least moderate masses of cooled gas in the cores of cf clusters , sufficient in a few cases to account for the bulk of the mass deposited by the cooling flows in those regions ( allen 1995 ) and at least some causal connection with the cooling flows . at radio wavelengths , extensive searches have been carried out for 21 cm line - emission associated with warm atomic hydrogen in the cores of cf clusters ( valentijn & giovanelli 1982 ; odea & baum 1996 ) . the negative results obtained ( with the exception of m87 ; jaffe 1992 ) typically constrain the column densities of optically thin material to be less than a few @xmath168 and , for an assumed spin temperature , @xmath169k , the number of clouds along the line of sight to be @xmath170 ( odea & baum 1996 ) . searches have also been made for the 21 cm absorption signature of cold atomic hydrogen ( mcnamara , bregman & oconnell 1990 ; jaffe 1992 ; dwarakanath , van gorkom & owen 1994 ; odea , gallimore & baum 1995 ; odea & baum 1996 ; johnstone 1998 ) . such material has been detected across the central few kpc of a few cf clusters , including abell 426 ( crane , van der hulst & haschick 1982 ; sijbring 1989 ; jaffe 1990 ) 2a0335 + 096 ( mcnamara 1990 ) , hydra a ( dwarakanath , van gorkom & owen 1995 ) , mkw3s ( mcnamara 1990 ) and abell 2597 ( odea , baum & gallimore 1994 ) with typical column densities of a few @xmath171 . however , for most clusters only negative results have been obtained , with upper limits to the column densities of 10k gas ( assuming a covering fraction of unity and a velocity width of @xmath172 kms@xmath58 ; dwarakanath 1994 , odea & baum 1996 ) of @xmath173 . laor ( 1997 ) presents tight limits on the column density of atomic hydrogen towards the core of abell 426 ( @xmath174 ) from ly@xmath175 observations made with the hubble space telescope ( see also johnstone & fabian 1995 ) . together , the results on hi emission and absorption , and ly@xmath175 absorption , imply that the bulk of the material responsible for the observed x - ray absorption can not be in the form of atomic hydrogen . an issue of crucial importance in interpreting the 21 cm hi results and the results from co studies ( see below ) is the expected temperature of x - ray absorber , if in gaseous form , which has proved controversial . johnstone , fabian & taylor ( 1998 ; updating earlier calculations by ferland , fabian & johnstone 1994 and fabian 1994a ) suggest that cooled gas clouds , in equilibrium with the cooling - flow environment , are likely to be very cold . johnstone ( 1998 ) suggest that dust - free clouds form an extended outer envelope , with a column density of @xmath176 and a temperature of 13 - 17k , beyond which the gas drops to the microwave background temperature . for galactic dust / gas ratios , however , the results are significantly modified , such that the cloud temperature drops to the microwave background temperature after a column density of only @xmath177 . at such low temperatures , dusty gas is likely to be highly molecular . however , these predictions differ from those of odea ( 1994a ) and voit & donahue ( 1995 ) who suggest a minimum temperature for such clouds of @xmath178k , even in the presence of dust . braine ( 1995 ) also suggest a minimum temperature of @xmath179k for the absorbing gas . possible reasons for the discrepancies between these results ( with contrasting views ) are discussed by voit & donahue ( 1995 ) and ferland ( in preparation ) . extensive searches have been carried out for co associated with molecular gas in cooling flows . to date co emission has only been detected in abell 426 ( lazareff 1989 ; mirabel , sanders d.b . & kazs i. 1989 ) . braine ( 1995 ) also present limits on co absorption in the central regions of this cluster ) . the upper limits to the column density of molecular hydrogen in the inner regions of other clusters ( for an assumed kinetic temperature of @xmath180k and a covering fraction of unity ) are typically @xmath181 a few @xmath0 ( odea 1994a , antonucci & barvainis 1994 , mcnamara & jaffe 1994 ; braine & dupraz 1995 ; odea & baum 1996 ) which is significantly below the x - ray inferred column densities . the co results also constrain the number of clouds along the line of sight ( for @xmath180k ) to be @xmath182 ( odea 1994a ; odea & baum 1996 ) . the co emission limits appear to exclude the possibility of large column densities ( _ i.e. _ values consistent with the x - ray measurements ) of molecular gas , with a kinetic temperature @xmath183k . however , if the bulk of the x - ray gas cools to the microwave background temperature , as suggested by johnstone ( 1998 ) , the co limits may be consistent with the x - ray data . it should be noted that the determinations of molecular hydrogen masses from co observations are affected by a number of additional uncertainties . firstly , the co / h@xmath184 ratios in cluster cooling flows may differ from galactic values . in particular , where the temperature and metallicity of the absorbing gas is low , the use of a standard galactic co / h@xmath184 ratio may underestimate the mass of molecular gas ( maloney & black 1988 ; madden 1997 ) . it is unclear whether x - ray absorbing gas accumulated by cooling flows could have substantially sub - solar metallicity . allen & fabian ( 1998b ) determine a mean emission - weighted metallicity for the x - ray emitting gas in cf clusters of @xmath185 . these authors also show that cf clusters generally contain metallicity gradients , so that the metallicity of the x - ray gas within the cooling radii of cf systems will , on average , exceed this value . thus , it seems unlikely that the material currently being deposited from the cooling flows will have substantially sub - solar metallicities . however , if the x - ray absorbing material is formed from the matter deposited by the cooling flows over their lifetimes , the metallicity of this material could be somewhat lower . within an inhomogeneous cooling flow ( nulsen 1986 , 1998 ; thomas , fabian & nulsen 1987 ) the denser gas at any particular radius will tend to have a smaller volume filling factor . if the metals are evenly distributed throughout the volume of the cooling flow at any particular radius ( whether this will occur is unclear ) , the least dense gas will contain most of the metals ( reisenegger , miralda - escud & waxman 1996 ) . since the densest material in cooling flows will have the shortest cooling time and be deposited first , the material accumulated by the cooling flows over their histories could have a significantly lower metallicity than the material being deposited today . conversely , if the metals in cooling flows were concentrated in the densest material , the metallicity of the x - ray absorbing material could exceed that of the x - ray emitting gas . fabian ( 1994a ) , have suggested that the masses of molecular gas inferred from co observations could be significantly underestimated if most of the co in the absorbing material has frozen onto dust grains . such freezing should occur rapidly ( on timescales @xmath186 yr ) if even small amounts of dust are initially present in the absorbing gas ( voit & donahue 1995 ) . daines ( 1994 ) and fabian ( 1994a ) discuss how dust is likely to form even in clouds that are initially free of dust , on timescales @xmath187 yr ( although see also voit & donahue 1995 ) . optical , uv and sub - mm studies of the cores of cooling flows show that dust lanes ( sparks , macchetto & golombek 1989 ; mcnamara & oconnell 1992 ) and intrinsic reddening ( hu 1992 ; allen 1995 ; edge 1999 ) are common . elston & maloney ( 1994 ) and jaffe & bremer ( 1997 ) report detections of h@xmath184(1 - 0)s(1 ) emission from warm ( @xmath188k ) molecular hydrogen from k - band spectroscopy of the inner few kpc of a number of cf systems ( including abell 478 and and pks0745 - 191 ) . no detections of such emission have been made in ncf systems . the h@xmath184s emission is likely to arise from the heated skin of cold molecular clouds which could also be responsible for the optical emission - line phenomena discussed above . a final possibility , first discussed in detail by voit & donahue ( 1995 ; see also arnaud & mushotzky 1998 ) , is that the x - ray absorption could be due to dust , with little or no associated gas . the k - absorption edge of oxygen is a primary source of x - ray absorption and its signature is fairly insensitive to the physical state of the oxygen , so long as it is not highly ionized . the introduction of a simple oik absorption edge at @xmath189 kev in the asca analysis ( such as might be associated with oxygen - rich , silicate dust grains ) typically provides at least as good a model for the intrinsic absorption as a cold gaseous absorber . arnaud & mushotzky ( 1998 ) have shown that @xmath190 kev broad band x - ray telescope data for the perseus cluster require excess absorption which is significantly better - explained by a simple oxygen absorption edge ( which they associate with dust ) than by a cold , gaseous absorption model . verification of this important result will be possible in the near future using observations made with the chandra observatory and xmm . optical and uv spectroscopy of the central emission - line nebulosities in cooling flows indicate the presence of significant amounts of dust ( hu 1992 ; allen 1995 ; crawford 1999 ) . allen ( 1995 ) also show that the column densities of absorbing gas inferred from the optical reddening studies are in reasonable agreement with the x - ray values , for galactic dust / gas ratios . the infrared observations reported here and elsewhere , together with the sub - mm results of edge ( 1999 ) , require the presence of significant dust masses in the core regions of at least a few nearby and/or exceptionally massive cooling flows ( for which the best data exist ) and are consistent with dust being a common feature of cf clusters . limits on background counts of galaxies and quasars behind clusters are consistent with extinctions ranging from @xmath191 ( boyle , fong & shanks 1988 ; romani & maoz 1992 ) , although measurements of galaxy colours in nearby clusters constrain the reddening on cluster - wide scales to be @xmath192 mag ( ferguson 1993 ) . voit & donahue ( 1995 ) argue that the levels of dust required to account for the observed x - ray absorption are unlikely to lie entirely within the cores of the clusters , and should be distributed more widely throughout the cluster gas . rosat results on the spatial distribution of the x - ray absorbing material ( allen & fabian 1997 ) show it to be centrally concentrated within the cooling radii of clusters and possibly be confined within cooling flows . however , the x - ray data are also consistent with a scenario in which the absorbing material is distributed in a manner similar to the x - ray emitting gas . this is illustrated in fig . 8(a ) , where we show the equivalent column densities of hydrogen ions in the x - ray emitting gas through two simulated clusters ; one cf and one ncf system . the gas distributions have been modeled as isothermal spheres , with core radii of 50 and 300 kpc , respectively ( and have been normalized to provide total x - ray gas masses within 3mpc of the cluster centres of @xmath193 ) . we see that the distribution of x - ray emitting gas is much more centrally concentrated in the cf system , particularly within the central @xmath134 kpc . 8(b ) shows the corresponding x - ray absorbing column densities as a function of radius ( as would be determined from observations with asca - like ccd detectors in the @xmath88 kev band ) in the case where the intracluster gas has an equivalent metallicity in dust grains of @xmath194 solar . for simplicity , we assume that the dust has an absorption spectrum in the asca band similar to that of metal - rich cold gas ( morrison & mccammon 1983 ; qualitatively similar results are obtained for an oxygen edge at 0.54 kev ) . we also assume that the absorber follows a multilayer distribution within the x - ray emitting medium , but is modeled as a uniform absorbing screen in front of the emitting regions . the results imply mean emission - weighted column densities ( where we assume the x - ray emissivity to be proportional to square of the gas density ) of @xmath195 for the cf cluster , and a few @xmath196 for the ncf system . ( the absolute results on the excess column densities will depend upon the chemical composition of the dust ) . we thus see that if the x - ray absorption is due to dust , with a spatial distribution that follows the x - ray gas , then we can expect absorption signatures to be present in both cf and ncf clusters , but to be stronger in cf systems ( assuming in each case that an appropriate spectral model is used in the analysis ) . we would also expect the absorption to be concentrated towards the cores of cf clusters , in agreement with the observations . we note that if the x - ray absorption were entirely due to a pre - existing distribution of dust grains distributed throughout the clusters , then the agreement between the observed masses of absorbing matter ( calculated assuming that the absorption is due to cold gas confined within the cooling flows ) and the masses predicted to have been accumulated by the cooling flows over their lifetimes ( fig . 6 ) must be regarded as coincidence . however , it remains possible that both material accumulated by the cooling flows and a more extended distribution of large dust grains ( presumably due to supernovae enrichment at early epochs ) could contribute to the x - ray absorption . the lifetime of grains of radius @xmath197 m to sputtering in hot gas of density @xmath145 is @xmath198 ( draine & salpeter 1979 ) . provided that individual grains exceed 10@xmath199 m in radius , they should survive for a hubble time or longer throughout ncf clusters and beyond the cooling radius in clusters with cooling flows . within cooling flows , the gas density rises inward so the grains are increasingly sputtered , releasing the metals into the gas phase which thus becomes increasingly metal rich towards the cluster centre . this provides one possible explanation for the abundance gradients inferred to be present in most cf clusters ( allen & fabian 1998b ; see also irwin & bregman 1999 ) . however , @xmath200 m grains are optically thick to soft x - rays and so some distribution of grain sizes extending to smaller values is required if dust is responsible for the observed x - ray absorption ( laor & draine 1993 ) . we note that the sputtering of small ( @xmath201 m ) grains will significantly modify the reddening law in clusters with respect to the standard galactic relation , reducing the optical / uv reddening ( laor & draine 1993 ) . in conclusion , the available data from other wavebands provide some support for the large column densities of intrinsic absorbing material inferred to be present from the x - ray data , at least in the innermost regions of cooling flows . optical and sub - mm observations of star formation in the cores of cf clusters and k - band observations of emission lines from molecular hydrogen suggest that significant masses of dusty , molecular gas are present in the central few tens of kpc of many cf clusters . the constraints from 21 cm observations suggest that the bulk of the material accumulated by cooling flows can not remain in a long - lived reservoir of atomic hydrogen . substantial masses of molecular gas may be distributed throughout cooling flows and as yet have avoided detection , but this material must be very cold ( @xmath202k ) and dusty , a possibility that remains controversial . the very large x - ray column densities ( @xmath165 ) required if the absorbing material has a multilayer distribution are probably incompatible with a gaseous absorber . it remains possible that dust grains , either present in the cooled material deposited by the cooling flows or distributed throughout the clusters in a manner similar to the x - ray gas , may be responsible for much of the observed x - ray absorption . the main conclusions that may be drawn from this paper may be summarized as follows : 0.2 cm \(i ) we have demonstrated the need for multiphase models to consistently explain the spectral and imaging x - ray data for the cf clusters included in our study . the mass deposition rates from the cooling flows , independently inferred from multiphase analyses of the asca spectra and deprojection analyses of the rosat hri images , exhibit good agreement , especially once the effects of intrinsic x - ray absorption have been accounted for in a consistent manner . the mass deposition rates from the largest cooling flows exceed 1000 , identifying these as some of the most massive cooling flows known . \(ii ) we have confirmed the presence of intrinsic x - ray absorption in the cluster spectra using a variety of spectral models . the measured equivalent hydrogen column densities are sensitive to the spectral models used in the analysis , ranging from a few @xmath0 for a simple isothermal emission model to a few @xmath1 using our preferred cooling - flow models , assuming in each case that the absorber lies in a uniform foreground screen . both the cf and ncf systems exhibit excess absorption at a similar level ( on average ) when analysed with the same , simple isothermal model . \(iii ) the masses of x - ray absorbing material inferred to be present in the cf clusters ( assuming the absorption to be due to cold gas with a covering fraction of unity ) are in reasonable agreement with the masses expected to have been accumulated by the cooling flows over their lifetimes . \(iv ) the asca spectra constrain the covering fraction of the absorbing material acting on the cooling flows to be close to ( or exceed ) unity . if the x - ray absorption is due to many small , similarly - sized clouds along each line of sight , intermixed with the x - ray emitting gas , then the column densities inferred from the spectral analysis may significantly underestimate the true column densities of absorbing material in the cooling flows . \(vii ) we have summarized the constraints on the physical properties of the x - ray absorbing material from observations in other wavebands . substantial 60 and 100@xmath199 m fluxes , sufficient to account for the bolometric luminosities absorbed and reprocessed within the clusters , are detected from several of the largest cooling flows . optical observations of star formation in the cores of cooling flows also suggests that significant masses of molecular gas are present in the central few tens of kpc of many cf systems . constraints from 21 cm observations suggest that the bulk of the material accumulated within the cooling radii can not remain in a long - lived reservoir of atomic hydrogen . substantial masses of molecular gas may be distributed throughout cooling flows and as yet have avoided detection , although this material must be very cold ( @xmath202k ) , a possibility that remains controversial . the very large x - ray column densities ( @xmath165 ) required if the absorbing material has a multilayer distribution are probably incompatible with a gaseous absorber . it remains possible that dust grains , either present in the material accumulated by the cooling flows or distributed more widely in a manner similar to the x - ray gas , may be responsible for much of the x - ray absorption . i thank andy fabian , roderick johnstone and dave white for many helpful discussions and harald ebeling , alastair edge and carolin crawford for their continuing efforts with the bcs project . i acknowledge the support of the royal society . allen s.w . , 1995 , mnras , 276 , 947 allen s.w . , 1998 , mnras , 296 , 392 allen s.w . , fabian a.c . , 1997 , mnras , 286 , 583 allen s.w . , fabian a.c . , 1998a , mnras , 297 , l57 allen s.w . , fabian a.c . , 1998b , mnras , 297 , l63 allen s.w . , fabian a.c . , johnstone r.m . , white d.a . , daines s.j . , edge a.c . , stewart g.c . , 1993 , mnras , 262 , 901 allen s.w . , fabian a.c . , edge a.c . , bhringer h. , white d.a . , 1995 , mnras , 275 , 741 allen s.w . , fabian a.c . , edge a.c . , bautz m.w . , furuzawa a. , tawara y. , 1996 , mnras , 283 , 263 allen s.w . , fabian a.c . , johnstone r.m . , arnaud k.a . , nulsen p.e.j . , 1999 , mnras , submitted akritas m.g . , bershady m.a . , 1996 , apj , 470 , 706 anders e. , grevesse n. , 1989 , geochemica et cosmochimica acta 53 , 197 antonucci r. , barvainis r. , 1994 , aj , 107 , 448 arnaud , k.a . , 1996 , in astronomical data analysis software and systems v , eds . jacoby g. and barnes j. , asp conf . series volume 101 , p17 arnaud k.a . , 1988 , in fabian a.c . , ed , cooling flows in clusters and galaxies . kluwer , dordrecht , p. 31 arnaud k.a . , mushotzky r.f . , 1998 , apj , 501 , 119 balucinska - church m. , mccammon d. , 1992 , apj , 400 , 699 bevington p.r . , 1969 , data reduction and error analysis for the physical sciences . mcgraw - hill , new york binney j. , tremaine s. , 1987 , galactic dynamics , princeton univ . press , princeton bhringer h. , tanaka y. , mushotzky r.f . , ikebe y. , hattori m. , 1998 , a&a , 334 , 789 boyle b.j . , fong r. , shanks t. , 1988 , mnras , 231 , 897 braine j. , dupraz c. , 1994 , a&a , 283 , 407 braine j. , wyrowski f. , radford s.j.e . , henkel c. , lesch h. , 1995 , a&a , 293 , 315 buote d.a . , tsai j.c . , 1996 , apj , 458 , 27 cardiel n. , gorgas j. , aragon - salamanca a. , 1995 , mnras , 277 , 502 cardiel n. , gorgas j. , aragon - salamanca a. , 1998 , mnras , 298 , 977 cavaliere a. , fusco - femiano r. , 1976 , a&a , 49 , 137 cavaliere a. , fusco - femiano r. , 1978 , a&a , 49 , 137 cox c.v . , bregman j.n . , schombert j.m . , 1995 , apjs , 99 , 405 crane p , van der hulst j.m . , haschick a. , 1982 , iaus , 97 , 307 crawford c.s . , fabian a.c . , 1992 , mnras , 259 , 265 crawford c.s . , allen s.w . , ebeling h. , edge a.c . , fabian a.c . , 1999 , mnras , 306 , 857 daines s.j . , fabian a.c . , thomas , p.a . , 1994 , mnras , 268 , 1060 david l.p . , slyz a. , jones c. , forman w. , vrtilek s.d . , arnaud k.a . , 1993 , apj , 412 , 479 david l.p . , harnden f.r . , kearns k.e . , zombeck m.v . , 1996 , the rosat hri calibration report , @xmath203 dickey j.m . , lockman f.j . , 1990 , ara&a , 28 , 215 draine b.t . , salpeter e.e . , 1979 , apj , 231 , 77 dwarakanath k.s . , van gorkom j.h . , owen f.n . , 1994 , apj , 432 , 469 dwarakanath k.s . , owen f.n . , van gorkom j.h . , 1995 , apj , 442 , l1 dwek e. , rephaeli y. , mather j.c . , 1990 , apj , 350 , 104 ebeling h. , voges w. , bhringer h. , edge a.c . , huchra j.p . , briel u.g . , 1996 , mnras , 281 , 799 ebeling h. , edge a.c . , bhringer h. , allen s.w . , crawford c.s . , fabian a.c . , voges w. , huchra j.p . , 1998 , mnras , 301 , 881 edge a.c . , stewart g.c . , fabian a.c . , 1992 , mnras , 258 , 177 edge a.c . , ivison rr.j . , smail i. , blain a.w . , kneib j .- . , 1999 , mnras , 306 , 599 elston r. , maloney p. , 1994 , in infrared astronomy with arrays , assl , 190 , kluwer , dordrecht , p. 169 fabian a.c . , 1994 , a&ar , 32 , 277 fabian a.c . , nulsen p.e.j . , canizares c.r . , 1991 , a&ar , 2 , 191 fabian a.c . , johnstone r.m . , daines s.j . , 1994a , mnras , 271 , 737 fabian a.c . , hu e.m . , cowie l.l . , grindlay j. , 1981 , apj , 248 , 47 fabian a.c . , arnaud k.a . , bautz m.w . , tawara y. , 1994b , apj , 436 , l63 ferguson h.c . , 1993 , mnras , 263 , 343 feretti l. , giovannini g. , bhringer h. , 1997 , newa , 2 , 501 ferland g.j . , fabian a.c . , johnstone r.m . , 1994 , mnras , 266 , 399 fukazawa y. 1994 , pasj , 46 , l55 heckman t.m . , baum s.a . , van breugel w.j.m . , mccarthy p. , 1989 , apj , 338 , 48 helou g. , khan i.r . , malek l. , boehmer l. , 1988 , apjs , 68 , 151 henkel c. , wiklind t. , 1997 , ssrv , 81 , 1 hu e.m . , 1992 , apj , 391 , 608 ikebe y. , makishima k. , fukazawa y. , tamura t. , xu h. , ohashi t. , matsushita k. , 1999 , apj , in press irwin j.a . , bregman j.n . , 1999 , apj , submitted irwin j.a . , sarazin c.l . , 1995 , apj , 355 , 497 jaffe w. , 1990 , a&a , 240 , 254 jaffe w. , 1992 , in fabian a.c . , ed . , clusters and superclusters of galaxies , kluwer , dordrecht , p. 109 jaffe w. , bremer m.n . , 1997 , mnras , 284l , 1 johnstone r.m . , fabian a.c . , 1995 , mnras , 273 , 625 johnstone r.m . , fabian a.c . , nulsen p.e.j . , 1987 , mnras , 224 , 75 johnstone r.m . , fabian a.c . , taylor g.b . , 1998 , mnras , 298 , 854 johnstone r.m . , fabian a.c . , edge a.c . , thomas p.a . , 1992 , mnras , 255 , 431 jones c. , forman w. , 1984 , apj , 276 , 38 kaastra j.s . , mewe r. , 1993 , legacy , 3 , heasarc , nasa kleinmann s.g . , hamilton d. , keel w.c . , wynn - williams c.g . , eales s.a . , becklin e.e . , kuntz k.d . , 1988 , apj , 328 , 161 kneib j .- p . , mellier y. , pell r. , miralda - escud j. , le borgne j .- f . , bhringer h. , picat j .- , 1995 , a&a , 303,27 laor a. , 1997 , apj , 483 , l103 laor a. , draine b.t . , 1993 , apj , 402 , 441 lazareff b. , castets a. , kim d - w . , jura m. , 1989 , apj , 336 , l13 liedahl d.a . , osterheld a.l . , goldstein w.h . , 1995 , apj , 438 , l115 madden s.c . , poglitsch a. , geis n. , stacey g.j . , townes c.h . , 1997 , 483 , 200 maloney p. , black j.h . , 1988 , apj , 325 , 389 makishima k. , 1997 , in makino f. , mitsuda k. , eds . , x - ray imaging and spectroscopy of cosmic hot plasmas , universal academy press , tokyo , p. 137 markevitch m. , 1996 , apj , 465 , l1 markevitch m. , forman w.r . , sarazin c.l . , vikhlinin a. , 1998 , apj , 503 , 77 matsumoto h. , koyama k. , awaki h. , tomida h. , tsuru t. , mushotzky r. , hatsukade i. , 1996 , pasj , 48 , 201 mcnamara b.r . , jaffe w. , 1994 , a&a , 281 , 673 mcnamara b.r . , oconnell r.w . , 1989 , aj , 98 , 2018 mcnamara b.r . , oconnell r.w . , 1992 , apj , 393 , 589 mcnamara b.r . , bregman j.n . , oconnell , 1990 , apj , 360 , 20 mcnamara b.r . , jannuzi b.t . , elston r. , sarazin c.l . , wise m. , 1996 , apj , 469 , 66 mirabel i.f . , sanders d.b . , kazs i. , 1989 , apj , 340 , l9 morrison r. , mccammon d. , 1983 , apj , 270 , 119 moshir m. , 1989 . explanatory supplement to the iras faint source survey . ipac preprint 44 nulsen p.e.j . , 1986 , mnras , 221 , 377 nulsen p.e.j . , 1998 , mnras , 297 , 1109 odea c.p . , baum s.a . , 1996 , in cold gas at high redshift , eds bremer m.n . , van der werf p.p . , rttgering h.j.a . , carilli c.l . , kluwer , dordrecht , p.199 odea c.p . , baum s.a . , gallimore j.f . , 1994b , apj , 436 , 669 odea c.p . , gallimore j.f . , baum s.a . , 1995 , aj , 109 , 26 odea c.p . , baum s.a . , maloney p.r . , tacconi l.j . , sparks w.b . , 1994a , apj , 422 , 467 peres c.b . , fabian a.c . , edge a.c . , allen s.w . , johnstone r.m . , white d.a . , 1998 , mnras , 298 , 416 press w.h . , teukolsky s.a . , vetterling w.t . , flannery b.p . , 1992 , numerical recipes , cambridge university press reisenegger a. , miralda - escud j. , waxman e. , 1996 , apj , 457 , 11l rizza e. , burns j.o . , ledlow m.j . , owen f.n . , voges w. , bliton m. , 1998 , mnras , 301 , 328 romani r.w . , maoz d. , 1992 , apj , 386 , 36 schindler s. , hattori m. , neumann d.m . , bhringer h. , 1997 , a&a , 317 , 646 sijbring d. , de bruyn a.g . , jaffe w.j . , sancisi r. , in meurs e.j.a , fosbury r.a.e . , extranuclear activity in galaxies , european southern observatory , 1989 , garching bei munchen , p107 smail i. , hogg d.w . , blandford r. , cohen j.g . , edge a.c . , djorgovski s.g . , 1995 , mnras , 277 , 1 smail i. , ellis r.e . , dressler a. , couch w.j . , oemler a. , sharples r.m . , butcher h. , 1997 , apj , 479 , 70 sparks w.b . , macchetto f. , golombek d. , 1989 , apj , 345 , 153 squires g. , neumann d.m . , kaiser n. , arnaud m. , babul a. , bhringer h. , fahlman g. , woods d. , 1997 , apj , 482 , 648 tanaka y. , inoue h. , holt s.s . , 1994 , pasj , 46 , l37 thomas p.a . , fabian a.c . , nulsen p.e.j , 1987 , mnras , 228 , 973 valentijn e.a . , giovanelli r. , 1982 , a&a , 114 , 208 voit g.m . , donahue m. , 1995 , apj , 452 , 164 voit g.m . , donahue m. , 1997 , apj , 486 , 242 waxman e. , miralda - escud j. , 1995 , apj , 451 , 451 white d.a . , fabian a.c . , johnstone r.m . , mushotzky r.f . , arnaud k.a . , 1991 , mnras , 252 , 72 white d.a . , jones c. , forman w. , 1997 , mnras , 292 , 419 wise m.w . , sarazin c.l . , 1999 , apj , submitted wise m.w . , oconnell r.w . , bregman j.n . , roberts m.s . , 1993 , apj , 405 , 94 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: phylogenetics is a study of how species are related to each other . evolutionary relationships are most conveniently represented by rooted leaf - labelled trees , where the leaves represent extant species and the root represents their most recent common ancestral species . similarly other internal vertices of evolutionary trees correspond to extinct ancestral species . the arrival of dna and protein sequence data in the last forty years led to explosive growth of phylogenetics . many of the modern phylogenetic methods consider sequence data under probabilistic models of sequence evolution . for such models to be useful for phylogenetic inference , it is important to establish their _ identifiability _ ( i.e. , to show that nonisomorphic trees or different model parameters can not induce the same distribution on the sequences at the leaves under a given model of sequence evolution ) . mathematical theory of phylogenetic trees , especially probabilistic models of sequence evolution , the associated questions of identifiability and statistical consistency ( especially of the maximum likelihood methods ) have been extensively studied @xcite , giving a firm statistical foundation to the study of phylogenetic trees . while phylogenetic trees represent relationships between species , population pedigrees represent how individuals within a population are related to each other . communities all over the world have long been curious about knowing their ancestral histories , and have often kept detailed records of their family trees . in fact this curiosity goes back much further in the past than the interest in constructing evolutionary relationships between species . an example of a fairly detailed record of family histories is the icelandic database slendingab ( the book of icelanders http://www.islendingabok.is ) of genealogical records that covers almost the whole icelandic population and goes back to nearly 1200 years . such ancestral histories are often compiled from a variety of sources such as church records , birth and death records , obituaries etc . that are prone to ambiguities or missing data beyond a few generations in the past . in the last several years large amounts of data on intra - population genetic variation have been recorded . for example , the icelandic biomedical company decode genetics has compiled genomic , genealogical and health data of more than 100000 individuals ( which is a significant proportion of the current icelandic population ) . such data offer promising opportunity to cross check and resolve ambiguities in historic genealogical records besides being useful for other studies such as , for example , genetic factors associated with medical conditions . therefore , there is a renewed interest in accurately inferring pedigrees from genomic data . the statistical and combinatorial foundation for studying reconstruction problems for pedigrees has not been as developed as in phylogenetics . a purpose of this paper is to continue our earlier attempts to develop such a foundation for the problem of reconstructing pedigrees from observations ( sequences ) on extant individuals . to develop such a foundation , we need to establish results along the following lines : developing a biologically realistic model for sequences undergoing mutations and recombinations , and identifiability results for such a model ; statistical consistency results ; and finally results that give estimates for the amount of genomic data necessary to reliably construct pedigrees . this paper is mainly about identifiability questions . in the rest of this section , we discuss the above theoretical motivation in more detail . we begin by informally sketching some well known reconstruction and identifiability results for phylogenetic trees that not only motivate the work in this paper but also are crucially useful in the proof of the main identifiability result of this paper . _ results of zarecki and buneman . _ it was shown in @xcite that a leaf - labelled tree can be uniquely constructed from the pairwise distances between its leaves . the result was strengthened slightly as follows @xcite . suppose that @xmath0 is an _ additive function _ on the family of subsets of cardinality 2 of the vertex set of a leaf - labelled tree . here _ additive _ means that for any two vertices @xmath1 and @xmath2 , we have @xmath3 , where the summation is over all edges @xmath4 on the ( unique ) path from @xmath1 to @xmath2 . buneman showed that knowing @xmath0 on all pairs of leaves of a leaf - labelled tree without vertices of degree 2 is sufficient to uniquely construct the tree and the function . it is not surprising that these results are quite useful in phylogenetics , where observations on extant species ( leaves of an evolutionary tree ) are used to infer a suitably defined distance ( or an additive function ) between pairs of species , and then their phylogenetic tree is constructed uniquely . _ results of steel and chang . _ now suppose the evolutionary process on a ( rooted ) tree is modelled as follows : first the root is assigned a random state from a finite alphabet @xmath5 ( e.g. , @xmath5 may be \{a , t , g , c } in the case of dna sequences ) . each state is assumed to have a nonzero probability of being assigned to the root . each edge of the tree has associated with it a @xmath6 matrix of substitution probabilities . these substitution probabilities determine how the vertex - states evolve away from the root , and induce a distribution of states at the leaves of the tree . the model was formulated in @xcite as described above , while a slightly more general formulation in terms of markov random fields on unrooted trees was given in @xcite . it was independently shown by chang and steel that when the matrices defining the substitution probabilities satisfy certain mild conditions , the ( unrooted ) tree can be uniquely recovered from the joint distribution of states at the leaves of the tree . in particular they showed that the negative logarithm of the determinant of the matrix of substitution probabilities between pairs of vertices is an additive function on the pairs of vertices , and can be computed from the probability distribution on extant sequences . it was further proved in @xcite that the substitution matrices are also identifiable from the marginal distributions on triples of leaves of the tree . special cases of these results for models more commonly used in phylogenetics were known in the phylogenetics literature earlier . how far can we generalise such results if the underlying structure is more general than a tree ? a recent result in this direction is due to @xcite where it is shown that under some mild non - degeneracy conditions the dependency structure of a markov random field can be obtained from sufficiently many independent samples . in this paper we present simple models for recombinations and mutations for population pedigrees , and generalise phylogenetic identifiability results for them . one difference between reconstructing markov random fields and reconstructing pedigrees is that for pedigrees we have observations only on the extant individuals ( e.g. , dna sequences derived from living individuals ) . moreover , in the problem of reconstructing pedigrees , the samples of data ( e.g. , columns in a sequence alignment ) are not _ i.i.d . _ ( independent and identically distributed ) as a result of recombinations . in @xcite , we studied some purely combinatorial reconstruction problems motivated by zarecki s result , for example , the problem of reconstructing a pedigree from the pairwise distances between its extant individuals or from its subpedigrees ( pedigrees of subsets of the extant population ) . in @xcite , we showed that a pedigree can not in general be reconstructed from the collection of its proper subpedigrees . such a result implies that knowing pairwise distances between extant vertices is in general not enough to reconstruct pedigrees . in @xcite , we considered models for sequences evolving on pedigrees , and showed that for certain simple markov models , pedigrees are not identifiable from the distribution of observed states at extant vertices . we did construct examples of processes for which pedigrees could be proved to be identifiable , but the processes lacked the markovian property , which informally states that the state observed at a vertex depends only on the states of its parents . moreover , it seems that pedigrees that are difficult to reconstruct in a purely combinatorial framework ( e.g. , from pairwise distances between extant vertices or from subpedigrees ) are also likely to be difficult to reconstruct in a stochastic framework . for example , if a pedigree can not be reconstructed from its proper subpedigrees , then the marginal distributions of extant sequences on proper subsets of the extant population might be insufficient to uniquely recover the pedigree . on the other hand , negative results in a combinatorial setting may not imply non - identifiability in a stochastic framework . it is therefore important to study combinatorial reconstruction problems ( e.g. , classification of pedigrees that may be difficult to reconstruct combinatorially ) , stochastic identifiability problems for idealised markov models of recombination and mutation , and relationships between these problems . reconstruction problems of purely combinatorial nature are well known to combinatorialists , the foremost among such problems being the vertex reconstruction conjecture @xcite . the conjecture states that all simple undirected unlabelled graphs can be constructed from their collection of unlabelled induced subgraphs . combinatorial reconstruction problems have also been studied in phylogenetics , for example , problems of reconstructing phylogenetic trees from subtrees @xcite . steel and chang proved their phylogenetic identifiability results in two parts : computation of the additive ` log determinant ' function on pairs of leaves from the joint probability distribution on leaf states , and the combinatorial problem of reconstructing a tree from the additive function , which had been solved by buneman and zarecki . similarly the problem of reconstructing pedigrees under a recombination - mutation model may be solved in two parts : in the first part , we would like to reduce the identifiability question to an appropriate combinatorial reconstruction problem , and then in the second part we would like to show that the combinatorial reconstruction problem has a unique solution . to ensure the uniqueness of reconstruction , we will have to compute sufficiently strong combinatorial invariants of the pedigree from the joint distribution of extant sequences . although we noted that distances between extant vertices in a pedigree are not sufficient to reconstruct a pedigree , we sketch a heuristic argument given in @xcite that shows how the distances between extant vertices in a discrete generation pedigree may be obtained from sequence data . suppose @xmath7 and @xmath8 are two extant individuals in a pedigree . suppose further that in the pedigree there are @xmath9 pairs of paths such that one of the paths in each pair ends on @xmath7 and the other ends on @xmath8 , and the two paths in each pair start at a common ancestor of @xmath7 and @xmath8 in the @xmath10-th generation , and the two paths in a pair do not share any other vertex . now if sufficiently many short recombination - free homologous segments of dna of @xmath7 and @xmath8 are compared , then we would expect about @xmath11 of them to have a common ancestor in the @xmath10-th generation . thus it may be possible to estimate @xmath12 . such calculations are theoretically possible for small values of @xmath10 assuming the population is large and the sequences are long . they then tried to use the numbers @xmath12 for all pairs @xmath13 for all @xmath10 to construct the pedigree . they computationally found many pairs of non - isomorphic pedigrees that have the same number of pairs of paths of each length . one such example is the pair of pedigrees shown in figure [ fig - hap ] , which was also mentioned in @xcite . but a more detailed analysis of sequence similarities ( between multiple sequences , if required ) under simple recombination and mutation models may give us more information than just pairwise distances between living individuals . in the main theorem of this paper ( theorem [ thm - paths ] ) , we show that the joint distribution on extant sequences determines a class of combinatorial invariants ( e.g. , certain types of subgraph sequences ) that supersedes pairwise distances between extant sequences and subtrees ( genealogical trees ) in a pedigree . we then show that pedigrees , such as the ones in figure [ fig - hap ] , that earlier seemed difficult to distinguish due to their combinatorial similarities ( including the non - reconstructible pedigrees constructed in @xcite ) are distinguished by the class of invariants . this paper is organised as follows . in section [ sec - prelim ] , we define pedigrees , alignments and subgraph sequences . in section [ sec - models ] , we give a schematic description of the recombination process , and formalise three models for sequences : model r ( a model in which there are recombinations but no mutations ) , model rm ( a model in which there are recombinations and mutations ) , and model m ( a model of mutations for sequences evolving on trees ) . we then formulate identifiability problems for these models . in section [ sec - model - r ] , we analyse pedigrees with two generations under model r. in section [ sec - model - rm ] , we prove the main theorem ( theorem [ thm - paths ] ) and demonstrate its applications . in the last section , we discuss a few open questions . a section on nomenclature follows the references where all symbols appearing in the paper and the number of the page on which they appear first are listed . we use the following notation for number systems and their subsets : @xmath14 for the set of integers , @xmath15 for the set of positive integers , @xmath16 for natural numbers , @xmath17 $ ] for the set @xmath18 . depending on the context , we write @xmath19 $ ] for the set of integers @xmath20 or real numbers @xmath21 ( and similarly @xmath22 , @xmath23 and @xmath24 $ ] for open or half - open intervals in integers and reals ) . the set of all @xmath10-tuples of elements of a set @xmath25 is written as @xmath26\}$ ] . the set of all functions from @xmath27 to @xmath25 is written as @xmath28 . next we introduce some graph theoretic notation . the vertex set and the edge set of a graph @xmath29 are denoted by @xmath30 and @xmath31 , respectively , and their cardinalities by @xmath32 and @xmath33 , respectively . the in - degree and the out - degree of a vertex @xmath34 in a directed graph are denoted by @xmath35 and @xmath36 , respectively . the degree of a vertex @xmath34 in an undirected graph ( or the total degree in a directed graph ) is denoted by @xmath37 . an arc from @xmath34 to @xmath38 in a directed graph and also an edge between @xmath34 and @xmath38 in an undirected graph is written as @xmath39 , and it will be understood from the context whether @xmath39 is meant to be a ( directed ) arc or an ( undirected ) edge . when any two objects @xmath40 and @xmath41 are isomorphic , we write @xmath42 . the isomorphism class of an object @xmath29 is written as @xmath43 . for a collection @xmath44 of labelled objects , we write @xmath45 for the set of isomorphism classes of objects in @xmath44 . let @xmath29 and @xmath46 be two directed or undirected graphs . we write @xmath47 ( or @xmath48 ) if @xmath29 is isomorphic to a subgraph of @xmath46 , and this notation may be used when @xmath29 or @xmath46 is unlabelled ( i.e. , they are just isomorphism classes ) . we write @xmath49 ( or @xmath50 ) when a labelled graph @xmath29 is a subgraph ( or a supergraph ) of a labelled graph @xmath46 . [ defn - general ] a _ general pedigree _ @xmath51 of a set @xmath27 is a directed acyclic graph on a vertex set @xmath52 and a set of arcs @xmath53 such that each vertex has in - degree 0 or 2 . the set @xmath27 is the set of vertices with out - degree 0 . the set @xmath54 is the set of vertices with in - degree 0 . the vertices in @xmath27 are called the _ extant vertices _ ( or the extant individuals in the population ) . the vertices in @xmath54 are called the _ founder vertices _ ( or the founders of the population ) . the _ order _ of the pedigree is @xmath55 . the _ depth _ of a pedigree is the length of ( i.e. , the number of arcs in ) a longest path in the pedigree . two pedigrees @xmath51 and @xmath56 are said to be _ isomorphic _ if there is a one - one map @xmath57 such that @xmath39 is an arc in @xmath58 if and only if @xmath59 is an arc in @xmath60 , and @xmath61 for all @xmath62 . we denote the natural partial order on @xmath63 by @xmath64 , i.e. , @xmath65 if there is a directed path from @xmath34 to @xmath38 ( or @xmath66 ) . we define isomorphism only between pedigrees of the same set of extant individuals and require that it fixes all extant vertices because ( informally speaking ) we would like to treat extant vertices to be labelled and other vertices to be unlabelled . throughout this paper , we will assume that all pedigrees have @xmath27 as their set of extant vertices . [ defn - diploid ] let @xmath51 be a pedigree . suppose that @xmath63 can be partitioned into unordered pairs of vertices such that the following conditions hold : the extant vertices are paired with extant vertices and the founder vertices are paired with founder vertices ; two non - extant vertices @xmath38 and @xmath67 are paired if and only if there are arcs @xmath68 and @xmath69 in @xmath53 ; no two paired vertices have a common parent . then @xmath58 together with one such pairing is called a _ diploid pedigree_. thus any general pedigree is a _ haploid pedigree _ ; a diploid pedigree is a haploid pedigree with a pairing of its vertices ( although not all pedigrees admit such a pairing ) . the pairing in a diploid pedigree is completely determined by the pairing of its extant vertices since all other pairs are determined by the definition . an advantage of a purely combinatorial definition of a diploid pedigree is that we can now assign just one sequence to each vertex . from this point of view , haploid pedigrees are pedigrees of sequences , not of individuals . pedigrees of individuals are diploid pedigrees obtained by pairing sequences in a haploid pedigree . figure [ fig - d2h ] illustrates this point of view . when vertices ( sequences ) @xmath70 and @xmath71 are paired , @xmath72 and @xmath73 must be paired , since they are parents of @xmath70 ; similarly , @xmath74 and @xmath75 must be paired . [ defn - char ] let @xmath76 be a finite set ( called an _ alphabet _ ) and @xmath63 a finite set . a _ character _ on @xmath63 is a map @xmath77 . for @xmath78 , an _ alignment _ of length @xmath79 on @xmath63 is a map @xmath80 . equivalently , an alignment is an @xmath79-tuple @xmath81 of characters on @xmath63 , or a two dimensional array of symbols from @xmath5 , with @xmath82 rows and @xmath79 columns . the rows of the array are called _ sequences _ , and are written as @xmath83 , @xmath84 to @xmath82 . individual entries in the array are written as @xmath85 , where @xmath84 to @xmath82 and @xmath86 to @xmath79 . the columns of an alignment are called _ usually @xmath63 will be the set of vertices of a pedigree , and we will be interested in alignments restricted to the set @xmath27 of extant vertices of the pedigree . the space of characters on @xmath27 is @xmath87 , which is also referred to as the space of _ site patterns_. the set of alignments of length @xmath79 on a set @xmath27 is @xmath88 , which we write as @xmath89 . let @xmath51 be a pedigree and let @xmath90 be an alignment on @xmath58 . if the sequences in @xmath91 have evolved under some process of recombination or mutation , then ( regardless of the details of the model of recombination or mutation ) we may suppose that each site in each sequence is inherited only from one of the two parent sequences . therefore , for each @xmath92 and each @xmath93 $ ] , there is a unique directed path @xmath94 from some founder vertex @xmath95 to @xmath34 that defines the genetic ancestry of @xmath96 . therefore , each site @xmath97 has associated with it a spanning forest @xmath98 , and each alignment of length @xmath79 has an underlying ( usually unknown ) sequence @xmath99 of spanning forests . similarly , each site @xmath97 has associated with it a _ directed subforest _ defined by @xmath100 , and we have a _ directed subforest sequence _ @xmath101 of the alignment . here we define these notions purely graph theoretically ( without reference to alignments or models ) . [ defn - xtree ] let @xmath27 be a finite set . a _ directed @xmath27-forest _ @xmath102 is a directed forest that satisfies the following conditions : the set of vertices with out - degree 0 is @xmath27 , and is called the _ leaf set _ of @xmath102 ; each component has a single vertex of in - degree 0 , called the _ root vertex _ of the component , and all other vertices have in - degree 1 ; all arcs are directed away from the root vertices . undirected @xmath27-forest _ is an unrooted forest with leaf set @xmath27 . suppose @xmath102 is a directed @xmath27-forest . it induces a natural partition of @xmath27 into _ maximal clusters _ @xmath103 such that vertices in cluster @xmath104 have a unique most recent common ancestor ( mrca ) @xmath105 in @xmath102 . we construct a subgraph of @xmath102 induced by @xmath106 $ ] and their descendants in @xmath102 , and then replace its ( directed ) arcs by ( undirected ) edges . the resulting undirected unrooted graph is called the _ undirected @xmath27-forest _ of @xmath102 , written as @xmath107 . it is the maximal undirected @xmath27-forest contained in @xmath102 . in the above definition , the term _ @xmath27-forest _ is meant to be analogous to the term _ @xmath27-tree _ that is commonly used in phylogenetics @xcite . in this paper we will adapt some of the phylogenetic identifiability results for undirected @xmath27-forests that appear as undirected graphs underlying subgraphs of pedigrees . therefore , in the following definition , we specialise the terms for pedigrees . [ defn - subtree ] let @xmath58 be a pedigree . a _ spanning forest _ of @xmath58 is a spanning subgraph @xmath29 of @xmath58 such that the in - degree of each vertex in @xmath29 is 1 unless it is a founder vertex in @xmath58 . a _ directed @xmath27-forest _ of @xmath58 is a subgraph @xmath102 of @xmath58 such that @xmath102 is a directed @xmath27-forest and the root vertex of each component of @xmath102 is a founder vertex in @xmath58 . each spanning forest @xmath29 in @xmath58 contains a unique directed @xmath27-forest of @xmath58 , and we denote it by @xmath108 . an _ undirected @xmath27-forest _ in @xmath58 is the unique undirected @xmath27-forest in any directed @xmath27-forest in @xmath58 . each spanning forest @xmath29 in @xmath58 contains a unique undirected @xmath27-forest of @xmath58 , and we denote it by @xmath109 . note that we use the term _ spanning forest _ in a specific sense : a spanning forest is not any spanning forest in the graph theoretic sense . we illustrate these terms in figure [ fig - subtrees ] , which shows a pedigree and a spanning forest @xmath29 with @xmath110 ( shown by bold arcs ) . in this example , the unique directed @xmath27-forest @xmath108 in @xmath29 has the arc set @xmath111 ; its clusters are @xmath112 and @xmath113 , and the root vertices of its components are @xmath114 and @xmath4 . the unique undirected @xmath27-forest @xmath109 in @xmath29 consists of vertex set @xmath115 and edge set @xmath116 . note that vertex 3 is isolated in @xmath109 since it is the mrca of its cluster , but we need it in our analysis . two ( directed or undirected ) @xmath27-forests @xmath102 and @xmath117 are said to be _ isomorphic _ ( written @xmath118 ) if there is a graph theoretic isomorphism @xmath119 from @xmath102 to @xmath117 such that @xmath61 for all @xmath120 in @xmath27 . the isomorphism class of a directed or an undirected @xmath27-forest @xmath102 , denoted by @xmath121 , is the set of all @xmath27-forests @xmath117 that are isomorphic to @xmath102 . the set of all directed @xmath27-forests of @xmath58 is denoted by @xmath122 . the set of isomorphism classes of directed @xmath27-forests of @xmath58 ( or the set of _ distinct _ directed @xmath27-forests in @xmath58 ) is denoted by @xmath123 . the set of spanning forests of a pedigree @xmath58 is denoted by @xmath124 . the set of undirected @xmath27-forests of @xmath58 is denoted by @xmath125 . the set of isomorphism classes of undirected @xmath27-forests of @xmath58 ( or the set of _ distinct _ undirected @xmath27-forests in @xmath58 ) is denoted by @xmath126 . [ prop - spanning - forests ] each spanning forest of a pedigree @xmath58 has @xmath127 arcs and contains exactly one directed @xmath27-forest . there are @xmath128 spanning forests . each directed @xmath27-forest @xmath102 is contained in @xmath129 spanning forests . each spanning forest of @xmath58 is obtained by selecting one of the two arcs that point to each non - founder vertex . the unique directed @xmath27-forest @xmath102 in a spanning forest @xmath29 is the subgraph of @xmath29 spanned by vertices in @xmath29 that are ancestral in @xmath29 to the extant vertices . finally , @xmath130 is the number of arcs not pointing to vertices in @xmath102 , therefore , @xmath131 is the number of non - founder vertices outside @xmath102 , at each of which we can choose one of the two incoming arcs to construct a spanning forest containing @xmath102 . in any model of recombination , the number of recombination events is determined by the spanning forest sequence underlying an alignment , but we define it for all spanning forest sequences without reference to alignments or models . [ def - rg - sg ] let @xmath58 be a pedigree and let @xmath132 be a spanning forest sequence in @xmath58 . if @xmath68 and @xmath69 are distinct arcs in @xmath58 such that @xmath68 is in @xmath133 and @xmath69 is in @xmath134 , then we say that a recombination has occurred at site @xmath135 at vertex @xmath34 or that @xmath135 is a recombining site . if @xmath68 is an arc in @xmath133 and @xmath134 then we say that there is no recombination at @xmath34 at site @xmath135 . we define the number of recombinations in @xmath136 to be @xmath137 where @xmath138 is the number of recombinations separating @xmath133 and @xmath134 . the number of points of no recombination is @xmath139 the directed @xmath27-forest sequence of @xmath136 is the sequence @xmath140 , and the undirected @xmath27-forest sequence of @xmath136 is the sequence @xmath141 . we assume that in any reasonable model of sequence evolution , sequences are first assigned to the founder vertices , and then subsequent generations of individuals inherit their sequences from their parents sequences subject to recombinations and mutations . we are then interested in the following types of _ identifiability _ questions . [ problem - main ] suppose sequences of equal length over a finite alphabet are assigned to the founder vertices of a pedigree . the sequences then evolve on the pedigree undergoing recombinations and mutations , giving a probability distribution on the space of alignments on the set of extant vertices . can we determine the pedigree uniquely ( i.e. , up to isomorphism ) - with or without the knowledge of the size or the depth of the pedigree or the various probability parameters defining the recombination and mutation processes , with or without restrictions such as discrete generations or constant population , and so on ? in the case of diploid pedigrees , we will be given the distribution on alignments on the set of extant vertices along with a pairing of extant vertices . in this paper , we study the above types of questions under two simple models of recombination and mutation . in _ , we assume that sequences evolve on a pedigrees under a process of recombinations without mutations . in _ model rm _ , we assume that sequences evolving on a pedigree undergo recombinations and mutations . for convenience , we also formalise the mutation part of model rm for the spanning forests of pedigrees and in general for directed @xmath27-forests , and call it _ model m_. in all these models , we assume that first all the founder vertices of a pedigree ( or the root vertices of a spanning forest or a directed @xmath27-forest ) are assigned sequences . these sequences are independently selected from a uniform distribution on @xmath142 , where @xmath76 is a known finite alphabet . then the sequences evolve on the pedigree ( or a spanning forest or a directed @xmath27-forest ) in a top - down manner , i.e. , a vertex is assigned a sequence only after its parents have been assigned sequences . we begin with a schematic description of the recombination process . our description is largely based on chapter 12 of @xcite . figure [ fig - meio ] schematically shows the process of gamete ( sperm or egg ) formation in eukaryotes . initially there is a parent cell with one pair of homologous non - sex chromosomes . each chromosome is then duplicated with the identical sister _ chromatids _ joined together at the _ centromere _ , forming a _ four - strand bundle_. then the two duplicated chromosomes exchange material between _ chiasmata _ ( _ recombination points _ ) . in the diagram , there are three recombination events . the first recombination is between strands 1 and 3 ( counted from top to bottom ) , the second is between strands 2 and 4 , and the third is between strands 2 and 3 . the four chromatids after the exchange of material are shown next . then the cell undergoes two cell divisions to create four haploid gametes , each receiving one of the four chromosomes . as shown in the diagram , at each recombination point a crossover occurs between one strand from the first pair and one strand from the second pair . at each crossover , a strand from the first pair and a strand from the second pair are chosen randomly with equal probability ( independent of other chiasma ) . this independence property is known as the lack of _ chromatid interference_. suppose that the locations of recombinations are modelled as a poisson point process along the sequence ( or on @xmath143 ) with the rate @xmath144 or a bernoulli process with probability @xmath145 ( so that a crossover occurs after a site on a sequence with probability @xmath145 independently of other sites or sequences ) . since exactly two of the four gametes - one from the first pair and one from the second pair - inherit any recombination , any given gamete inherits a recombination with probability @xmath146 . therefore , for the sequence of any fixed gamete , the locations of recombinations are still modelled by a poisson process , but with the rate @xmath147 , ( or bernoulli process with probability @xmath148 ) . therefore , a model may be formalised with just two parent sequences instead of four . a poisson process for the locations of chiasmata was first proposed in @xcite . based on the above description , we formalise models r and rm , in which we assume that crossovers in a finite sequence occur according to a bernoulli process . _ model r : _ consider three sequences @xmath83 of length @xmath79 over alphabet @xmath5 , where @xmath149 and @xmath38 and @xmath67 are parents of @xmath34 . the sequence @xmath150 is obtained by recombining sequences @xmath151 and @xmath152 as follows . let @xmath153 be a markov chain on the state space @xmath154 with transition probabilities @xmath155 if @xmath156 for @xmath157 , and @xmath158 . then , for @xmath159 , @xmath160 if @xmath161 . thus @xmath162 indicates a crossover from one sequence to the other . we refer to this model as _ model r_. _ model rm : _ consider three sequences @xmath83 of length @xmath79 over alphabet @xmath5 , where @xmath149 and @xmath38 and @xmath67 are parents of @xmath34 . the sequence @xmath150 is obtained from the sequences @xmath151 and @xmath152 by a process of recombinations and mutations as follows . let @xmath153 be a markov chain on the state space @xmath154 with transition probabilities @xmath155 if @xmath156 for @xmath157 , and @xmath163 . then if @xmath164 and @xmath165 , then @xmath166 is assigned @xmath1 with probability @xmath167 , and @xmath166 is assigned a state different from @xmath1 with probability @xmath168 . when a state different from @xmath1 is assigned to @xmath166 , each state in @xmath169 has equal probability @xmath170 of being assigned to @xmath166 . we refer to this process as _ model rm_. _ model m : _ this process is defined for spanning forests of a pedigree and directed @xmath27-forests . first each founder or the root vertex in each component of a directed @xmath27-forest is assigned independently and uniformly randomly a state from @xmath5 . suppose @xmath97 is the parent vertex of @xmath135 in a spanning forest of a pedigree or in a directed @xmath27-forest . let @xmath83 and @xmath171 be the sequences of @xmath135 and @xmath97 , respectively , both of equal length @xmath79 over alphabet @xmath5 . then for each @xmath172 $ ] , @xmath173 is assigned the same state as @xmath174 with probability @xmath167 , and @xmath173 is assigned a state different from @xmath174 with probability @xmath168 . when a state different from @xmath174 is assigned to @xmath173 , each state in @xmath175 has equal probability @xmath170 of being assigned to @xmath173 . we refer to this process as _ model m_. model m on a directed @xmath27-forest @xmath102 is equivalent to a similarly formulated model on the undirected @xmath27-forest of @xmath102 . we root each component of the undirected @xmath27-forest arbitrarily , and assign to it a state from @xmath5 uniformly randomly , independent of the roots of other components . the state then evolves away from the root in each component . if a component itself is an isolated vertex , it is simply assigned a state from @xmath5 uniformly randomly . since the mutation model described here is reversible , the same distribution on the site patterns is observed on @xmath27 in the undirected @xmath27-forest as in a directed @xmath27-forest for a given @xmath170 . therefore , when we try to construct a tree from the character distribution on its leaves , we can not construct the directed @xmath27-forest , but we can at best construct the undirected @xmath27-forest in it . therefore , we will consider model m only on undirected @xmath27-forests . thus model rm may be thought of as a synthesis of models r and m so that the recombination - free segments of sequences evolving on a pedigree may be examined under model m with phylogenetic methods . let @xmath58 be a pedigree on @xmath27 . for an alignment @xmath176 , we denote by @xmath177 the probability that sequences of length @xmath79 evolving on the pedigree @xmath58 under model @xmath178 give an alignment @xmath91 on @xmath27 . we use analogous notation when the model @xmath178 is replaced by the model @xmath179 , or when the pedigree @xmath58 is replaced by a directed or an undirected @xmath27-forest @xmath102 and the model under consideration is the mutation model @xmath180 . we denote the various probability spaces by @xmath181 , @xmath182 , @xmath183 , @xmath184 , and so on . for pedigrees @xmath58 and @xmath60 , we will write @xmath185 when @xmath186 for all @xmath187 , and analogously for other models . as in the case of alignments , we treat the spaces of spanning forest sequences , directed @xmath27-forest sequences , and undirected @xmath27-forest sequences as probability spaces , and denote them by @xmath188 , @xmath189 , and @xmath190 , respectively . the spanning forest sequences , and directed and undirected @xmath27-forest sequences , ( and the corresponding sequences of isomorphism classes of spanning forests and directed and undirected @xmath27-forests ) are defined by only the recombination events , therefore , the probability spaces are unchanged if @xmath179 is replaced by @xmath178 . for a spanning forest sequence @xmath191 , we will write @xmath192 for the probability of @xmath193 in the probability space @xmath194 . we will use analogous notation for other sequences and probability spaces . unless stated otherwise , when we will refer to alignments or sequences of spanning forests or other objects , we will mean alignments or sequences of spanning forests or other objects , respectively , from the appropriate probability spaces that are clear in the context . [ def - identifiability ] nonisomorphic pedigrees @xmath58 and @xmath60 in a class @xmath195 are said to be distinguished from each other under model @xmath178 if @xmath196 for some @xmath79 , ( i.e. , for some @xmath79 , there exists @xmath197 such that @xmath198 ) . a pedigree @xmath58 in a class @xmath195 is said to be identifiable under model @xmath178 if it is distinguished from every other pedigree @xmath60 in @xmath195 , ( i.e. , if there is a pedigree @xmath60 in @xmath195 such that @xmath199 for all @xmath200 , then @xmath60 is isomorphic to @xmath58 ) . pedigrees in a class @xmath195 are said to be identifiable under model @xmath178 if all pairs of pedigrees in @xmath195 are distinguished from each other under model @xmath178 . similar terminology will be used for other models and for undirected @xmath27-forests . stronger notions of identifiability may be defined , and correspondingly stronger variants of identifiability questions may be asked . the above definitions assume the model parameters to be fixed . but we may ask if there are nonisomorphic pedigrees @xmath58 and @xmath60 and model parameters @xmath201 such that @xmath202 for all @xmath203 . given the probability distribution a pedigree induces on the space of alignments , we may ask if the pedigree can be recognised to be in a class @xmath195 . we assume in all results in this paper that the model parameters and the class @xmath195 ( typically defined by the size of a pedigree ) to be fixed ( but possibly unknown ) . pedigrees of order 1 are in general not identifiable since for all pedigrees of order 1 , the extant sequence will be uniformly distributed . in fact for all pedigrees , all extant sequences will be uniformly distributed . but when there are more than one extant vertices , the joint distribution of extant sequences will contain some information about the pedigree on which they have evolved , because , for example , some vertices may have common ancestors , so their sequences will be correlated . we will therefore consider identifiability questions for pedigrees of order more than 1 . the models r and rm on a pedigree @xmath58 with @xmath204 arcs may be interpreted as hidden markov models with the set of hidden states @xmath124 and the set of observed states @xmath87 . this is illustrated in figure [ fig - chains ] . the initial probability for each hidden state is @xmath205 . the probability of an observed state conditional on a given hidden state @xmath206 can be easily computed ( and actually depends only on @xmath109 and @xmath170 ) . the probability of transition from state @xmath133 to @xmath207 is given by @xmath208 in most contexts in which hmms are used , one assumes that the set of hidden states in known , but here we do not know the set @xmath124 . we also note that the sequence of @xmath27-forests is not a markov chain . consider directed @xmath27-forests @xmath209 and @xmath210 at sites @xmath97 and @xmath211 , respectively . suppose a vertex @xmath34 in @xmath212 has parents @xmath38 and @xmath67 in the pedigree . the probability that the arc @xmath68 is in @xmath210 depends on the history before the @xmath97-th site . for example , if @xmath68 was in @xmath213 then the probability that it would also be in @xmath210 is @xmath214 . but if @xmath68 was not in @xmath213 the probability that it would be in @xmath210 is @xmath215 . therefore , @xmath216 is not a markov chain , i.e. , @xmath217 may not be the same as @xmath218 if the sequences @xmath219 and @xmath220 are different . therefore , we can not interpret the models as hmms with directed or undirected @xmath27-forests as hidden states . of characters at the extant vertices of a pedigree @xmath58 . among the intermediate chains , only @xmath221 is a markov chain.,width=445 ] in this section , we analyse diploid pedigrees and certain haploid pedigrees that obey many properties of diploid pedigrees under model @xmath222 . we show that , with some exceptions and mild conditions on @xmath145 and @xmath223 , diploid pedigrees of depth 2 are reconstructible from the probability distribution on the alignments on extant vertices . we use very basic techniques such as pairwise comparisons between extant sequences to exploit the correlation between them to reconstruct their pedigree . let @xmath58 be a pedigree and let @xmath102 be an undirected @xmath27-forest . we define @xmath224 . [ prop - ngt ] let @xmath58 be a pedigree with @xmath204 arcs . let @xmath225 be any character . then the probability that the @xmath10-th character in an alignment is @xmath226 is given by @xmath227 in particular , it does not depend on @xmath10 . let @xmath228 be a spanning forest sequence . it is a time - homogeneous markov chain on @xmath124 , with transition probabilities given by @xmath229 it follows that @xmath230 for all @xmath231 . let @xmath232 be a sequence of characters . then , under model rm , @xmath233 & = & \sum_{g \in { \mathcal{g}}_p } { \mathtt{pr}}\{c_k = c\mid g_k = g\ } { \mathtt{pr}}\{g_k = g { \mid}p , rm(p,\mu)\ } \nonumber \\[2 mm ] & = & \frac{1}{2^e}\sum_{g \in { \mathcal{g}}_p } { \mathtt{pr}}\{c_k = c\mid g_k = g\ } \nonumber \\[2 mm ] & = & \sum_{t \in \lvert { \mathcal{u}}_p \rvert } \frac{n(g > t : p)}{2^e } { \mathtt{pr}}\{c\mid t , m(\mu)\}. \nonumber \end{aligned}\ ] ] a similar result holds when model rm is replaced by model r. the above proposition implies that if two pedigrees have the same number of undirected @xmath27-forests of each type and the same number of arcs , then the character frequencies in alignments alone are not sufficient to distinguish the two pedigrees . for example , pedigrees in figure [ fig - hap ] can not be easily distinguished . but it turns out that , under model r , most diploid pedigrees of depth 2 are easily distinguished by making pairwise comparisons between extant sequences and computing the probability that they agree at a site . the above proposition implies that the probability that two sequences in an alignment agree at a site @xmath10 does not depend on @xmath10 in models r and rm . moreover , such a probability can be computed easily under model r. any given undirected @xmath27-forest @xmath102 of a pedigree induces a partition of its leaves so that the leaves within a component are in the same part . under model r , the probability that the sequences at two leaves @xmath135 and @xmath97 are in the same state at a site is 1 if they are in the same component of @xmath102 . otherwise , the probability is @xmath234 . under model rm , the probability depends on @xmath170 if they are in the same component , and is @xmath234 otherwise . [ prop-2seq ] let @xmath58 be a diploid pedigree of depth 2 . let @xmath235 . then the subpedigree of @xmath135 and @xmath97 is determined by the probability distribution induced on @xmath236 under model r. there are four possible ways in which any two vertices @xmath135 and @xmath97 are related in a diploid pedigree , which are shown in figure [ fig - grand ] . for each of them , we give the probability that the sequences @xmath70 and @xmath72 match at any site @xmath10 . in the following , we set @xmath237 . and @xmath97 may be related , width=445 ] _ case 1 : _ if @xmath135 and @xmath97 have the same parents , then @xmath238 . _ case 2 : _ if @xmath135 and @xmath97 have distinct pairs of parents but the same grand parents , then @xmath239 . _ case 3 : _ if @xmath135 and @xmath97 have distinct pairs of parents but exactly one pair of grand parents in common , then @xmath240 . _ case 4 : _ if @xmath135 and @xmath97 have no common parents or grand parents , then @xmath241 . therefore , unless @xmath242 , the above cases are distinguished by the marginal joint distribution on @xmath236 under model r. the assumption that @xmath58 is a diploid pedigree is essential . it implies that @xmath135 and @xmath97 are not related as in the haploid pedigree @xmath243 shown in figure [ fig - hap ] . we can verify that @xmath244 for both @xmath245 and @xmath243 , so they are indistinguishable by the above method , which was also pointed out as a consequence of proposition [ prop - ngt ] . the above probabilities do not depend on @xmath145 , therefore , we have a slightly stronger identifiability statement : if diploid pedigrees @xmath58 and @xmath60 of depth 2 and crossover probabilities @xmath145 and @xmath246 are such that @xmath247 then their subpedigrees of order 2 are correspondingly isomorphic . [ prop - n2 - 1 ] when @xmath248 , pedigrees of depth 2 in which no two vertices have exactly one common parent are identifiable under model r. in particular , when @xmath248 , diploid pedigrees of depth 2 are identifiable under model r. there are only 4 ways in which any two extant vertices @xmath135 and @xmath97 are related . they are illustrated in figure [ fig - grand ] . each of the possible relationships is recognised by proposition [ prop-2seq ] . we denote the 3rd and the 4th types of relationships by @xmath249 and @xmath250 , respectively . suppose that no two extant vertices @xmath251 in a pedigree are related to each other as @xmath249 or @xmath250 . then the pedigree is constructed by adding extant vertices one by one . on each step we add one extant vertex and join to previously added extant vertices as in figure [ fig - grand ] - ( i ) or ( ii ) , whichever is appropriate . therefore , we assume that at least two extant vertices @xmath252 are related as @xmath249 or @xmath250 . let @xmath253 be a ( nonempty ) maximal subset of @xmath27 such that for any two distinct extant vertices @xmath135 and @xmath97 in @xmath253 , either @xmath249 or @xmath254 . every other extant vertex @xmath10 not in @xmath253 is related to some vertex in @xmath253 as in figure [ fig - grand ] - ( i ) or ( ii ) . therefore , once the subpedigree of @xmath253 is constructed , there is only one way to extend it to the whole pedigree . to construct the subpedigree of @xmath253 , we first construct an edge labelled graph with edge set @xmath253 in which edges @xmath135 and @xmath97 are incident if and only if @xmath249 . this is a known problem in graph theory , namely , the problem of constructing an edge labelled graph from its line graph . it was proved @xcite that there are only 4 pairs @xmath255 of connected nonisomorphic edge labelled graphs that have the same line graphs . edge labelled graphs that can not be uniquely constructed from their line graphs must contain components isomorphic to @xmath133 or @xmath256 . we refer to @xcite ( in particular , chapter 15 , problem 1 ) for discussion about reconstructing graphs from their line graphs , in particular , for the complete list of pairs @xmath255 . the first pair is @xmath257 ( with edges of each of them labelled @xmath258 ) . based on the example @xmath257 , we construct pedigrees shown in figure [ fig - k3 ] , in which all pairs of extant vertices are similarly related . we distinguish the two pedigrees in figure [ fig - k3 ] by comparing the probabilities @xmath259 and @xmath260 for any site @xmath2 . in @xmath261 , there are 512 spanning forests . among them there are 192 spanning forests in which two extant vertices have a common grandparent , giving the first term on the rhs below . in the remaining 320 spanning forests , no two extant vertices have a common grandparent , which explains the second term on the rhs below . therefore , @xmath262 in @xmath263 , there are 512 spanning forests . among them there are 16 spanning forests in which @xmath258 have a common grandparent ( giving the first term on the rhs below ) , 144 spanning forests in which two extant vertices have a common grandparent at site @xmath2 ( giving the second term ) , and 352 spanning forests in which @xmath258 have distinct grandparents at site @xmath2 ( giving the third term ) . therefore , @xmath264 whenever @xmath248 , @xmath265 , therefore , @xmath261 and @xmath263 can be distinguished . the two expressions are equal when @xmath266 . similarly , other pedigrees constructed from @xmath267 are distinguished when @xmath248 . since there are only two types of site patterns for three sequences when @xmath266 ( either the three sequences agree at a site or exactly two of them agree ) , the two cases can not be distinguished by considering other site pattern probabilities . in diploid pedigrees , no two vertices have exactly one common parent , therefore , when @xmath248 , they are identifiable under model r. we used only site pattern probabilities in the above proofs . but because of recombinations , consecutive sites in an alignment are not independent . we use the dependence between sites to eliminate the restriction @xmath248 when the crossover probability @xmath145 is sufficiently small . [ prop - n2 - 2 ] when @xmath266 and @xmath145 is sufficiently small , haploid pedigrees @xmath261 and @xmath263 ( shown in figure [ fig - k3 ] ) are distinguished under model @xmath179 . we compute the probability that , in an alignment @xmath91 , there are long runs of sites at which all the three sequences @xmath70 , @xmath72 and @xmath74 are equal . in particular , we compute bounds on @xmath268\}$ ] on the two pedigrees . for @xmath261 , for any fixed @xmath269 , if any two of the three sites @xmath270 are inherited from the same grand parent , then @xmath271 , and if all of them are inherited from distinct grandparents , then @xmath272 is 1/4 . therefore , @xmath273 { \mid}p_2,r(p ) \ } \leq ( 1/2)^{t+1}.\ ] ] for @xmath263 , the probability that the first site of all sequences is inherited from a common grand parent is 1/32 . the probability that at each successive site the three sequences have a common grand parent is @xmath274 . therefore , @xmath275 { \mid}q_2,r(p ) \ } \nonumber \\[2 mm ] & \geq & \frac{\left((1-p)^6+p^3(1-p)^3\right)^t}{32}. \end{aligned}\ ] ] when @xmath145 is sufficiently small and @xmath276 is sufficiently large , the above probability for @xmath263 is more than that for @xmath261 . we do not analyse other examples of pedigrees based on graphs that are not reconstructible from their line graphs ( graphs @xmath277 mentioned in proposition [ prop - n2 - 1 ] ) , but they may be analysed similarly . in this section we develop ideas from section [ sec - model - r ] ( especially proposition [ prop - n2 - 2 ] ) in much more generality . earlier we observed that parts of alignments that are free of recombination may be analysed with phylogenetic methods . since we do not know where the recombinations have occurred in an alignment , we choose long segments of carefully chosen alignments ( or sets of alignments ) and show that they have higher probability of having evolved on one particular @xmath27-forest ( or a sequence of @xmath27-forests ) than any other @xmath27-forest ( or a sequence of @xmath27-forests ) . for example , the method of proposition [ prop - n2 - 2 ] works because @xmath263 contains a tree in which @xmath258 have a common ancestor and @xmath261 does not contain such a tree . therefore , a sufficiently long sequence of characters in which sequences @xmath278 and @xmath279 are in the same state will more likely have evolved on a pedigree such as @xmath263 than on a pedigree that does not contain such a tree . this argument may be generalised to count the number of @xmath27-forests of each type from the distribution of extant sequences . such a generalisation requires identifiability results for phylogenetic trees , which we state in a form suitable for our application . in this section , we state known results on identifiability and statistical consistency of maximum likelihood reconstruction of phylogenetic trees . we need to adapt them slightly since the @xmath27-forests in a pedigree differ from phylogenetic trees in three respects - they may have vertices of degree 2 , they may be unresolved ( i.e. , they may have vertices of degree more than 3 ) , and they may be disconnected ( two extant vertices may not have a common ancestor in a given directed @xmath27-forest in a pedigree ) . we address them in the following identifiability result , which was originally proved for phylogenetic trees in @xcite in the @xmath266 case . identifiability and statistical consistency of maximum likelihood reconstruction of phylogenetic trees were independently proved in full generality for all @xmath280 in @xcite . [ thm - identifiability ] for all @xmath281 and any two undirected @xmath27-forests @xmath282 and @xmath283 with bounded number of edges , if @xmath284 for all characters @xmath285 then @xmath286 . the result follows from the analogous results in @xcite for phylogenetic @xmath27-trees , but we have to clarify three issues : unlike the phylogenetic @xmath27-trees , the @xmath27-forests as defined in this paper may be disconnected , unresolved , and may have vertices of degree 2 . _ connectivity : _ any two extant vertices @xmath287 and @xmath288 are in different components of @xmath282 and @xmath283 if and only if @xmath289 for all @xmath290 , where @xmath291 is the state at the extant vertex @xmath135 in a character @xmath226 . if @xmath135 and @xmath97 are in the same component , then @xmath292 can not be arbitrarily close to @xmath234 if the number of edges ( and hence the distance between @xmath135 and @xmath97 ) is bounded . therefore , we can consider the identifiability question for each component separately . _ unresolved @xmath27-forests : _ given an unresolved phylogenetic tree , there are resolved phylogenetic trees with site pattern probabilities arbitrarily close to the site pattern probabilities for the unresolved tree . therefore , even though unresolved phylogenetic @xmath27-trees are identifiable , statistical consistency of maximum likelihood methods requires that the substitution probabilities on the edges of a phylogenetic tree are bounded below by a positive real number . in our model , corollary [ cor - consistency ] below is possible because the substitution probability @xmath170 is fixed on each edge . _ vertices of degree 2 : _ let @xmath34 and @xmath38 be any two vertices of an @xmath27-forest . suppose that @xmath34 and @xmath38 are of degree 1 or more than 2 . suppose all internal vertices on the path between @xmath34 and @xmath38 have degree 2 . since @xmath170 is fixed for all arcs , the distance between @xmath34 and @xmath38 on a tree is determined by the substitution probability on the @xmath39 path . the substitution probability on the @xmath39 path is determined by the distribution on the space of characters . in the above result , if @xmath282 and @xmath283 were directed @xmath27-forests , then we would be able to conclude that @xmath293 . let @xmath294 . suppose that @xmath295 . we associate with each undirected @xmath27-forest @xmath102 a vector @xmath296 in @xmath297 , where @xmath298 . therefore , the condition @xmath299 for all characters @xmath225 may be equivalently written as @xmath300 . given @xmath301 and a point @xmath302 , let the open ball of radius @xmath1 centred at @xmath303 be denoted by @xmath304 . here the radius may be taken to be in the 1-norm ( i.e. , the distance between points @xmath305 and @xmath306 is defined by @xmath307 ) . let @xmath91 be an alignment on @xmath27 . we define a vector @xmath308 , where @xmath309 are the _ fractional site pattern frequencies _ , i.e. , @xmath309 is the fraction of columns of @xmath91 of type @xmath310 . then the above identifiability result implies a statistical consistency result for maximum likelihood . it informally says that as the length of a random alignment @xmath91 goes to infinity , we expect @xmath311 to be arbitrarily close to @xmath312 with probability approaching 1 if @xmath102 is the true @xmath27-forest , and that the probability that @xmath311 is arbitrarily close to @xmath312 approaches 0 if @xmath102 is not the true @xmath27-forest . [ cor - consistency ] for all @xmath313 , @xmath314 and @xmath315 , there exists @xmath316 such that for any two undirected @xmath27-forests @xmath102 and @xmath117 such that @xmath317 , and an alignment @xmath318 , @xmath319 and @xmath320 we give bounds on the above probabilities in terms of @xmath79 , which we prove using bernstein s inequality . [ lem - bernstein1 ] let @xmath321 be i.i.d . bernoulli random variables with @xmath322 . then for all @xmath323 and @xmath324 , @xmath325 and ( equivalently ) @xmath326 [ lem - eps - l1 ] let @xmath313 and @xmath281 . let @xmath327 . let @xmath102 be an undirected @xmath27-forest . then @xmath328 there are @xmath329 distinct characters , with probabilities @xmath330 , @xmath331 . if @xmath332 , then @xmath333 for some @xmath135 therefore , we apply bernstein s inequality to each distinct character , and write the probability that @xmath334 . we then apply the union bound to get the result . if we set @xmath335 it will ensure that the @xmath27-forests are _ separated _ by open balls of radius @xmath336 in the space of site pattern probability vectors , i.e. , the open balls @xmath337 and @xmath338 are non - intersecting whenever @xmath339 and @xmath209 are non - isomorphic . we will use this value of @xmath336 unless specified otherwise . now for an undirected @xmath27-forest @xmath339 , we define @xmath340 and @xmath341 by selecting a sufficiently large value of @xmath79 we can make @xmath342 arbitrarily small as in lemma [ lem - eps - l1 ] . moreover , if @xmath343 , then @xmath344 hence @xmath345 can be made arbitrarily small as per lemma [ lem - eps - l1 ] . we set @xmath346 , which depends on @xmath79 and @xmath336 . in theorem [ thm - paths ] ( particularly in the proof of inequality [ eq - term1a ] ) we require a concentration inequality similar to the inequality in lemma [ lem - eps - l1 ] for the situation in which @xmath79 sites of an alignment have evolved on an @xmath27-forest @xmath339 and @xmath347 sites ( for a small @xmath348 ) have evolved on another @xmath27-forest @xmath209 . ( we will keep the notation simple by assuming that @xmath347 is an integer . ) therefore , we give the following variants of lemmas [ lem - bernstein1 ] and [ lem - eps - l1 ] . [ lem - bernstein2 ] let @xmath349 be i.i.d . bernoulli random variables with @xmath350 . let @xmath351 be bernoulli random variables . let @xmath352 , where @xmath353 is a positive constant . let @xmath354 and @xmath355 and @xmath356 . if @xmath357 and @xmath358 , then @xmath359 since @xmath360 , we have @xmath361 & \implies & \left(\frac{s_n}{n(1+c ) } - p \leq -r\right ) \;\text{or}\ ; \left(\frac{s_n}{n(1+c ) } - p \geq r\right ) \\[1 mm ] & \implies & \left(\frac{\sum_{i=1}^n x_i}{n(1+c ) } - p \leq -r\right ) \;\text{or}\ ; \left(\frac{cn + \sum_{i=1}^n x_i}{n(1+c ) } - p \geq r\right ) \\[1 mm ] & \implies & \left(\frac{\sum_{i=1}^n x_i}{n(1+c ) } \leq p - r \leq m \right ) \;\text{or}\ ; \left(\frac{cn + \sum_{i=1}^n x_i}{n(1+c ) } \geq p+r \geq m \right ) \\[1 mm ] & \implies & \left(\frac{\sum_{i=1}^n x_i}{n(1+c)}-\frac{p}{1+c } \leq m-\frac{p}{1+c}\right ) \text{or}\ ; \left(\frac{cn + \sum_{i=1}^n x_i}{n(1+c ) } - \frac{p+c}{1+c } \geq m-\frac{p+c}{1+c } \right ) \\[1 mm ] & \implies & \left(\frac{\sum_{i=1}^n x_i}{n}-p \leq -r^\prime \right ) \;\text{or}\ ; \left(\frac{\sum_{i=1}^n x_i}{n}-p \geq r^{\prime\prime}\right ) . \end{aligned}\ ] ] now we apply bernstein s inequality ( lemma [ lem - bernstein2 ] ) to each term and obtain the desired bound . let @xmath91 be an alignment of length @xmath362 . suppose that @xmath79 characters of @xmath91 evolved on an undirected @xmath27-forest @xmath102 and the remaining characters evolved on undirected @xmath27-forests @xmath363 . the following lemma states that if @xmath353 is sufficiently small , then @xmath311 is concentrated near @xmath364 for large @xmath79 . moreover , as in lemma [ lem - eps - l1 ] , if we require @xmath311 to be sufficiently near @xmath364 with probability at least @xmath365 , then the length of the alignment @xmath362 must be @xmath366 . in the following lemma , we do not specify the constants @xmath367 and @xmath368 precisely , but they can be chosen depending on @xmath336 . [ lem - eps - l2 ] let @xmath313 and @xmath281 . let @xmath369 for a suitably chosen positive constant @xmath353 . let @xmath370 be undirected @xmath27-forest . then @xmath371 & \leq & \sum_i^{|\sigma|^{|x|}}\left(\exp \left\{\frac{-l(r_i^\prime)^2}{2p_i(1-p_i ) + 2r_i^\prime/3}\right\ } + \exp \left\{\frac{-l(r_i^{\prime\prime})^2 } { 2p_i(1-p_i ) + 2r_i^{\prime\prime}/3}\right\}\right ) , \end{aligned}\ ] ] where @xmath372 and @xmath368 are positive constants as in lemma [ lem - bernstein2 ] . we apply lemma [ lem - bernstein2 ] for each component of @xmath311 and use the union bound as in the proof of lemma [ lem - eps - l1 ] . for each component , we use @xmath373 as before . constants @xmath372 and @xmath368 ( and @xmath374 , which are implicit ) depend on @xmath336 and the probabilities @xmath330 , @xmath375 . the constant @xmath353 may be taken to be the smallest among @xmath376 . a hidden markov model ( hmm ) is defined by two sequences @xmath377 and @xmath378 of random variables . the sequence @xmath379 takes values in @xmath380 $ ] and is a stationary markov chain with transition matrix @xmath91 and initial distribution @xmath381 to @xmath1 , which is also the stationary distribution of the markov chain . the random variables @xmath378 take values in @xmath382 $ ] , and are independent and identically distributed conditional on @xmath379 . the distribution of @xmath383 depends only on @xmath384 . let @xmath385 be the @xmath386 matrix of conditional probabilities @xmath387 , where @xmath388 $ ] and @xmath389 $ ] . the sequence @xmath378 are the _ observations_. identifiable hidden markov models were characterised in @xcite , where a precise description of conditions on @xmath91 and @xmath385 for which the probability distribution on observed sequences determines @xmath91 and @xmath385 ( up to re - labelling of hidden states ) was given . here _ identifiability up to a re - labelling of hidden states _ means the following : if @xmath25 is an @xmath390 permutation matrix , then the hmm with parameters @xmath391 ( where @xmath119 is treated as a column vector of length @xmath1 ) is equivalent to ( induces the same distribution on the space sequences of observed states as ) the hmm with parameters @xmath392 . therefore , identifiability only means computing the matrices and the initial distribution up to equivalence . we denote the class of models equivalent to @xmath391 by @xmath393 . earlier we noted that models r and rm for sequences evolving on a pedigree @xmath394 define a hidden markov model with the spanning forests in @xmath58 as hidden states and characters from @xmath87 as observed states . we call it @xmath395 and denote its matrices by @xmath396 and @xmath397 . the initial distribution on hidden states is uniform : each spanning forest has the probability @xmath398 if the pedigree has @xmath399 arcs . we informally look at some of the issues about its identifiability . the transition matrix @xmath396 is defined by transition probabilities given in equation ( [ eq - gi - gj ] ) . therefore , @xmath396 will be identical ( up to a permutation of rows and columns ) for all pedigrees with the same number of arcs , for a fixed @xmath145 . but the set of spanning forests ( hidden states ) is unknown . we now describe at a high level how we compute the rows of @xmath397 . suppose the pedigree contains an undirected @xmath27-forest @xmath339 . there are @xmath400 spanning forests @xmath29 of @xmath58 that contain @xmath339 as the unique undirected @xmath27-forest , and corresponding to each of them we have a row of @xmath397 that is equal to @xmath401 . now consider a set @xmath402 of sufficiently long alignments as defined in equation [ eq - mcai ] . we compute the probability of @xmath403 ( i.e. , the probability that a random alignment is in @xmath403 ) . suppose that @xmath404 is the probability that there are no recombination events . thus @xmath404 approaches 1 as @xmath145 approaches 0 . then one of the terms in the expression for the probability of @xmath403 will be @xmath405 , where @xmath406 is the probability that the first site evolved on a spanning forest @xmath29 that contained @xmath339 as the undirected @xmath27-forest . there will be terms for contributions from other undirected @xmath27-forests @xmath407 , but they will be much smaller than the above term because they will contain factors @xmath408 ( as in equation [ eq - p - mcai ] ) . there will also be terms to account for recombinations among the first @xmath79 sites , but they will be small as well since they will contain @xmath145 as a factor ( in contrast to @xmath404 , which is a power of @xmath409 ) . so let us say @xmath410 is @xmath411 . therefore if @xmath145 is sufficiently small and @xmath79 is sufficiently large , then @xmath412 will be roughly equal to the dominating term @xmath413 , which will uniquely determine @xmath414 . in other words , if @xmath60 is another pedigree such that @xmath415 , then @xmath410 and @xmath416 will differ roughly by a multiple of @xmath417 . in the proof of proposition [ prop - n2 - 2 ] , we used a similar idea : the pedigree @xmath263 contains a certain subtree @xmath102 in which @xmath258 have a common ancestor , while the pedigree @xmath261 does not such a subtree . as a result , the alignments that are close to @xmath364 are more likely to have evolved on @xmath263 than on @xmath261 . suppose now that @xmath397 is identified and each of its rows is labelled by the corresponding unlabelled undirected @xmath27-forest . that is , the matrices @xmath397 that appear among the triples in the equivalence class @xmath393 of hmms are constructed . as pointed out above , the matrix @xmath396 and the initial distribution are also known up to relabelling of hidden states . but the equivalence class @xmath418 is not known unless we are able to label the rows and the columns of @xmath396 by unlabelled undirected @xmath27-forests in a manner consistent with the labelling of rows of @xmath397 . in other words , for full identifiability of @xmath395 , we would like to construct an automaton with transition probabilities given by @xmath396 and with its states labelled by unlabelled undirected @xmath27-forests . identifying the pedigree from the _ labelled automaton _ will then be a purely combinatorial problem . in this paper we do not succeed in constructing matrix @xmath396 with rows and columns labelled by undirected @xmath27-forests , but we are able to count certain types of _ walks _ ( to be described next ) on the automaton with vertices labelled by undirected @xmath27-forests . suppose that @xmath419 is a sequence of undirected @xmath27-forests such that no two consecutive ones are isomorphic . analogous to @xmath420 , we define @xmath421 as the number of sequences @xmath422 of spanning forests in @xmath58 such that @xmath423 , where consecutive @xmath133 are separated by just one recombination . ( a single recombination at a site is more likely than multiple recombinations . moreover , if there is a recombination at a site @xmath135 , but the two spanning forests @xmath133 and @xmath134 contain isomorphic undirected @xmath27-forests , then such a recombination has no effect on the emitted characters . these are the reasons why we consider the sequences @xmath339 and @xmath133 as above . ) we then consider a set @xmath424 of alignments of length @xmath425 ( for a suitably large @xmath79 ) obtained by concatenating alignments from @xmath403 for @xmath84 to @xmath269 . we compute the probability of @xmath424 ( as we described for @xmath426 above ) , and show that the dominating term is proportional to @xmath427 , and other terms are of smaller order of magnitude for small @xmath145 . this allows us to compute @xmath428 . a more visual description of the walks may be given as follows . suppose the pedigree has @xmath399 arcs . we define a graph on the vertex set consisting of the spanning forests of the pedigree , with two spanning forests @xmath133 and @xmath207 being adjacent if there is exactly one recombination separating them ( i.e. , @xmath429 ) . the graph is a hypercube . the hidden markov chain on the set of spanning forests jumps on the vertices of the cube . if there is at most one recombination at any site ( which is more likely than more than 1 recombinations at a site ) , then we have a walk on the edges of the cube . we label each vertex @xmath133 of the cube by the undirected @xmath27-forests @xmath430 . our interest is to construct this object for a more complete understanding of the hmm . but problem is made difficult by the fact that the emission probabilities associated with @xmath133 and @xmath207 are identical if @xmath431 . therefore , we construct a weaker object , namely the number of walks of each length on the cube such that consecutive vertices have distinct labels . let @xmath58 be a pedigree . for @xmath432 , we define @xmath433 & : = & n(g_1 > t_1 , g_2 > t_2 , \ldots , g_m > t_m : p ) \\[2 mm ] & : = & |\{\mathbf{g } \in { \mathcal{g}}_p^m : t_u(g_i ) \cong t_i \forall i \in [ m ] \wedge |e(g_{i+1})\bigtriangleup e(g_{i})| = 2 \ ; \forall\ ; i \in [ m-1]\}|,\end{aligned}\ ] ] where the second condition in the last line says that there is exactly one recombination event between consecutive @xmath133 . in the rest of this section , we show how invariants @xmath434 may be computed from the probability distribution on the space of alignments under model rm . in the end , we demonstrate an application to pedigrees @xmath245 and @xmath243 shown in figure [ fig - hap ] . lem - pg ] let @xmath58 be a pedigree with @xmath204 arcs . let @xmath435 be a sequence of spanning forests in @xmath58 . then under model @xmath179 , the probability that @xmath136 is a sequence of site - specific spanning forests is given by @xmath436 where @xmath437 and @xmath438 are as in definition [ def - rg - sg ] . we have a factor @xmath145 for each recombination event and @xmath409 whenever there is no recombination ( i.e. , an arc is contained in two consecutive spanning forests in the sequence ) . the probability that the first spanning forest is @xmath40 is @xmath398 . let @xmath318 be an alignment of length @xmath79 . for an interval @xmath439 \subseteq [ l]$ ] , we write @xmath440 $ ] for the part of the alignment between columns @xmath441 and @xmath442 ( inclusive of columns @xmath441 and @xmath442 ) . for a sequence of alignments @xmath443 $ ] , let @xmath444 denote the alignment obtained by concatenating alignments @xmath445 ( in that order ) . let @xmath446 $ ] be sets of alignments . we define @xmath447\}.\ ] ] [ lem - pa ] let @xmath58 be a pedigree with @xmath204 arcs . the probability of an alignment @xmath318 on @xmath58 is given by @xmath448 & = & \sum_{k=1}^{l}\ ; \sum_{\mathbf{g } \in { \mathcal{g}}_p^k } \frac{(1-p)^{e(l - k)+s(\mathbf{g})}p^{r(\mathbf{g})}}{2^{e } } \sum_{\substack{\mathbf{l } \in { \mathbb{z}}_+^k : \\ \sum\limits_{i=1}^kl_i = l}}\ ; \prod_{i=1}^k{\mathtt{pr}}\{a[l_{i-1}+1,l_i ] { \mid}t_u(g_i),m(\mu ) \ } , \nonumber \\ & & \end{aligned}\ ] ] where @xmath449 , @xmath450 and @xmath451 , and the second summation is over @xmath136 such that consecutive spanning forests @xmath452 and @xmath134 are unequal for @xmath453 $ ] . the probability of an alignment of length @xmath79 is obtained by summing its probability over all spanning subgraph sequences of length @xmath79 . suppose that the recombinations in a spanning subgraph sequence occur only at sites @xmath454 , for @xmath455 , where @xmath456 . we write the spanning forest sequence of length @xmath79 as @xmath457 . then the probability of the alignment is written as a product of probabilities of its segments that have evolved on spanning forests @xmath133 ( i.e. , effectively on @xmath430 ) . this probability is summed over @xmath458 ( with the constraint @xmath459 ) , @xmath460 $ ] and @xmath461 . for a fixed @xmath462 , the probability of @xmath457 is given by lemma [ lem - pg ] . for @xmath463 , we will compute @xmath464 by summing ( [ eq - pa ] ) over @xmath465 . in such a calculation , we will sometimes use the following upper bound in evaluating the last summation in ( [ eq - pa ] ) for fixed values of @xmath10 , and fixed @xmath466 and @xmath136 . [ lem - pa - ub2 ] let @xmath463 . let @xmath467 . let @xmath468 : a \in { \mathcal{a}}\ } , i \in [ k]$ ] . then for any fixed @xmath469 @xmath470 we have equality if @xmath471 . the claim follows from the observation that @xmath472 . the following lemma is used in the proof of equation ( [ eq-3b ] ) , and the reader may skip it until then . [ lem - seq ] let @xmath473 be a finite set . let @xmath474 and @xmath475 be two sequences in @xmath473 , each of length @xmath476 , defined by @xmath477 where @xmath478 and @xmath479 $ ] , and @xmath480 where @xmath481 and @xmath482 $ ] . suppose that @xmath474 and @xmath475 satisfy the constraints : @xmath483 $ ] ; @xmath484 $ ] ; @xmath485 ; if @xmath486 , then @xmath487 for some @xmath135 . then there is at least one block @xmath488 of @xmath474 over which @xmath474 and @xmath475 mismatch everywhere ; hence there are at least @xmath489\}$ ] mismatches between the two sequences . suppose that the claim is false ; so in each block @xmath488 of @xmath474 , there is a matching symbol in @xmath475 . therefore , @xmath419 is a subsequence of @xmath475 and @xmath490 . this , together with @xmath483 $ ] and @xmath491 , implies that @xmath486 and @xmath492 $ ] , which contradicts the assumption that when @xmath486 , there is some @xmath135 for which @xmath487 . [ thm - paths ] let @xmath58 and @xmath60 be any two pedigrees with @xmath493 arcs . let @xmath494 be any sequence of undirected @xmath27-forests in which consecutive @xmath27-forests are non - isomorphic . then for all @xmath281 , there exists @xmath495 such that for all @xmath496 , the following statement is true : if @xmath497 , then @xmath498 . let @xmath499 , where @xmath500 and @xmath336 are as defined in equations ( [ eq - eps0 ] ) and ( [ eq - mcai ] ) , respectively . we will choose @xmath501 ( defined at the end of section [ sec - consistency ] ) and @xmath79 ( that depends on @xmath501 ) suitably later . the probability of @xmath424 on @xmath58 and @xmath60 is written by summing ( [ eq - pa ] ) over all @xmath465 . but based on theorem [ thm - identifiability ] , corollary [ cor - consistency ] and lemma [ lem - eps - l2 ] , we can make the following qualitative and somewhat informal statement : if @xmath79 is large enough , then @xmath502 will get significantly higher contribution from spanning forest sequences @xmath503 of length @xmath504 such that @xmath505 for all @xmath506 $ ] and @xmath507 are all close to @xmath79 , than from spanning forest sequences @xmath508 for which there are many mismatches ( in terms of isomorphism ) between sequences @xmath509 and @xmath510 or if they require more than @xmath511 recombinations . let @xmath512 , where @xmath513 is the joint probability of @xmath424 and the event that there are exactly @xmath10 recombinations . furthermore , we write @xmath514 , where @xmath515 is the contribution from spanning forest sequences @xmath516 of length @xmath517 such that @xmath505 for all @xmath506 $ ] , and @xmath518 is the remaining contribution to @xmath519 , i.e. , from spanning forest sequences @xmath520 of length @xmath517 such that either @xmath521 ( i.e. , the @xmath511 recombinations occur at fewer than @xmath511 sites ) , or @xmath522 and @xmath523 for some @xmath506 $ ] . we will show that only @xmath515 makes a significant contribution to @xmath524 . in particular , we will show that the various contributions to @xmath525 take the following form : @xmath526 \sum_{k < m-1}p_k(p ) + p_{(m-1)b}(p ) & < & \delta_1 \\[1 mm ] \sum_{k \geq m } p_k(p ) & < & \delta_2 , \end{aligned}\ ] ] where @xmath527 does not depend on the pedigree ( but depends on the undirected @xmath27-forest sequence @xmath474 ) . moreover , @xmath527 , @xmath528 and @xmath529 depend on @xmath530 and @xmath501 . we will show that when @xmath145 and @xmath501 are sufficiently small and @xmath79 is sufficiently large , @xmath528 and @xmath529 are very small compared to @xmath527 . it will imply that for @xmath502 and @xmath531 to be equal , @xmath532 and @xmath533 must be equal . ( otherwise , there would be a difference between @xmath502 and @xmath531 that is of the order of a multiple of @xmath527 . ) _ a lower bound on @xmath515 . _ since the consecutive @xmath27-forests in @xmath474 are nonisomorphic and @xmath505 for all @xmath506 $ ] , there is at least one recombination between consecutive spanning forests @xmath133 . and since there are exactly @xmath511 recombinations , there must be exactly one recombination between consecutive spanning forests . therefore , @xmath534 & & \times \left ( \sum_{a \in { \mathcal{a } } } \sum_{\substack{\mathbf{l } \in { \mathbb{z}}_+^m : \\ \sum\limits_{i=1}^ml_i = m^2l}}\ ; \prod_{i=1}^m{\mathtt{pr}}\{a[l_{i-1}+1,l_i ] { \mid}t_u(g_i),m(\mu ) \}\right ) \nonumber \\ & : = & n(\mathbf{g } \geq \mathbf{t}:p ) \delta(\mathbf{t } ) , \end{aligned}\ ] ] where @xmath449 , @xmath450 . the inner summation is evaluated for any fixed choice of @xmath535 such that @xmath505 for all @xmath506 $ ] , because for any fixed @xmath536 $ ] , and spanning forest sequences @xmath537 and @xmath538 , if @xmath539 for all @xmath506 $ ] , then for each alignment @xmath91 , we have @xmath540 { \mid}g_i^{l_i},m(\mu ) \ } = { \mathtt{pr}}\{a[l_{i-1}+1,l_i ] { \mid}(g_i^{\prime})^{l_i},m(\mu ) \ } = { \mathtt{pr}}\{a[l_{i-1}+1,l_i ] { \mid}t_u(g_i),m(\mu ) \}$ ] . therefore , the rhs is a product of @xmath532 and a factor that does not explicitly depend on the pedigree , but only on @xmath474 . we have @xmath541 to prove the lower bound , we sum ( [ eq - pa ] ) over spanning forest sequences of the type @xmath503 where @xmath505 for all @xmath506 $ ] and @xmath507 are such that the @xmath511 recombinations occur at sites @xmath542 , for a small positive constant @xmath353 , ( i.e. , we ignore contributions from the spanning forest sequences in which some of the recombinations are not near the boundaries of the blocks @xmath543 of @xmath424 ) . we can choose the recombination sites in @xmath544 ways . since @xmath545 and @xmath546 , the probability of @xmath136 is @xmath547 . we write @xmath548 , where @xmath549 are blocks of @xmath136 of length @xmath425 each . then we have @xmath550 for each @xmath136 and for all @xmath135 , provided we choose @xmath551 and @xmath501 appropriately according to lemma [ lem - eps - l2 ] . _ an upper bound on @xmath552 . _ we have @xmath553 & \leq & \sum_{k=0}^{m-1 } \frac{1}{2^e } \left(\sum_{l=0}^{k}n_{kl}\binom{m^2 l -1}{l } ( \epsilon_{max})^{m - l}\right ) p^k(1-p)^{e(m^2l-1)-k } \nonumber \\[2 mm ] & \leq & c_2 l^{m-1 } \epsilon_{max } : = \delta_1 , \end{aligned}\ ] ] where @xmath554 is the number of spanning forest sequences @xmath555 in which @xmath10 recombinations occur at @xmath476 sites , and @xmath556 is a constant that depends on @xmath4 and @xmath269 . we explain below how the bound is obtained . we fix @xmath557 such that @xmath558 . for @xmath559 , since we are only interested in the contribution @xmath560 , we fix @xmath136 with the additional restrictions in the definition of @xmath560 . suppose that the @xmath10 recombinations occur at @xmath476 distinct sites @xmath561 , where @xmath562 . so we write @xmath563 . there are @xmath554 choices for @xmath564 . for each choice of @xmath565 , there are @xmath566 choices for the recombining sites @xmath567 . each @xmath136 has a probability @xmath568 , where @xmath569 and @xmath570 . we show that @xmath571 is bounded above by @xmath572 for each @xmath136 with @xmath476 recombining sites that satisfies the above constraints . let @xmath573 be a spanning forest sequence of length @xmath504 . for @xmath506 , j \in [ m]$ ] , we refer to the subsequences @xmath574)$ ] as _ blocks _ of @xmath136 , and subsequences @xmath575)$ ] as _ subblocks _ of @xmath136 . we say that subblock @xmath576 is _ recombination - free _ if there are no recombinations between any two sites of the subblock . ( the subblock may have recombinations at its boundaries . ) by lemma [ lem - seq ] , there is at least one block , say the @xmath135-th block , over which the sequences @xmath474 and @xmath577 mismatch everywhere . since there are @xmath269 subblocks in each block and @xmath578 , there are at least @xmath579 recombination - free subblocks in the @xmath135-th block of @xmath136 . let these subblocks be denoted by @xmath580 . we have @xmath581 for each of them . therefore , @xmath582 _ an upper bound on @xmath583 . _ we use the following fact about binomially distributed random variables : if @xmath584 , then @xmath585 . ( this result is a consequence of the union bound . ) since there are @xmath586 points at which a recombination can possibly occur , we have @xmath587 where @xmath588 is a constant that depends on @xmath4 and @xmath269 . we write similar bounds for @xmath60 . now suppose that @xmath589 but @xmath590 . therefore , equations [ eq - term1a ] and [ eq - term1b ] imply that @xmath591 & = & |n(\mathbf{g } \geq \mathbf{t}:p ) - n(\mathbf{g } \geq \mathbf{t}:q)|\delta(\mathbf{t } ) \nonumber \\[2 mm ] & \geq & \frac{(2cl)^{m-1}(1-p)^{(em^2 l - m - e+1)}p^{(m-1 ) } ( 1-\epsilon_{max})^{(m^2)}}{2^{e}}. \end{aligned}\ ] ] but that is impossible if we can choose @xmath501 , @xmath79 , and @xmath145 so that @xmath592 , or @xmath593 in other words , the discrepancy @xmath594 can not be compensated for by the remaining terms in @xmath595 and @xmath596 . such a choice is possible . for example , we first choose @xmath597 , let us say , @xmath598 , where @xmath599 . we then set @xmath600 and @xmath601 so that @xmath602 the choice of @xmath79 must guarantee phylogenetic consistency as in corollary [ cor - consistency ] . moreover , we must set @xmath79 as small as possible so as to get a better bound on @xmath145 . by lemma [ lem - eps - l2 ] , @xmath79 is required to grow only logarithmically in @xmath603 , so let @xmath604 for a suitable choice of @xmath605 . now conditions ( [ eq - cond ] ) are satisfied for @xmath606 for @xmath607 . then ( [ eq - final ] ) is satisfied for sufficiently large @xmath608 . in the following , we illustrate theorem [ thm - paths ] with a general result on counting @xmath27-forests and a couple of specific examples . if the conditions of theorem [ thm - paths ] are satisfied , then @xmath58 and @xmath60 have the same number of undirected @xmath27-forests of each type . we apply theorem [ thm - paths ] for @xmath609 . suppose that @xmath58 is a pedigree with @xmath204 arcs . let @xmath102 be an undirected @xmath27-forest . suppose that we want to count the number of copies @xmath610 of @xmath102 in p. let @xmath611 be the distinct copies of @xmath102 in @xmath58 . for each @xmath339 , let @xmath612 be the undirected @xmath27-forests in @xmath58 that contain @xmath339 . ( to be precise , we have subgraphs @xmath613 and @xmath614 in @xmath58 such that when the ( directed ) arcs of @xmath613 and @xmath614 are replaced by ( undirected ) edges , we get the undirected @xmath27-forests @xmath339 and @xmath615 . but we do not make this distinction in the following . ) note that @xmath615 are all distinct @xmath27-forests . there are @xmath616 non - founder vertices in @xmath58 at which we can choose one of the two incoming arcs to construct a spanning forest @xmath29 that contains @xmath339 . therefore , for each @xmath339 there are @xmath617 spanning forests that contain @xmath339 . therefore , @xmath618 now by grouping terms on the rhs by isomorphism classes of @xmath615 , we obtain @xmath619 since @xmath620 for all undirected @xmath27-forests @xmath621 , we also have @xmath622 for all undirected @xmath27-forests @xmath102 . the examples of non - reconstructible pedigrees given in @xcite can be distinguished from the probability distribution on the space of alignments under model rm . pedigrees in these examples do not have the same number of @xmath27-forests of each type . for example , one pedigree in each pair contains a common ancestor of all extant vertices while the other does not . this may also be observed in the examples in figure [ fig - k3 ] . but knowing the number of @xmath27-forests of each type is not always enough to distinguish pedigrees , as verified by pedigrees @xmath245 and @xmath243 shown in figure [ fig - hap ] . they both have the same number of directed ( therefore , also undirected ) @xmath27-forests of each type : they have 4 directed @xmath27-forests in which 1 and 2 have a common ancestor . we denote their undirected @xmath27-forest by @xmath282 , which is a path of length 4 with end vertices 1 and 2 . there are 12 directed @xmath27-forests in which 1 and 2 do not have a common ancestor . we denote their undirected @xmath27-forest by @xmath283 , which consists of two isolated vertices 1 and 2 . moreover , @xmath623 and @xmath624 . also , for @xmath625 ( and for @xmath626 ) , we check by direct counting that @xmath627 . but @xmath245 and @xmath243 can nevertheless be distinguished as shown below . [ cor - pq ] pedigrees @xmath245 and @xmath243 in figure [ fig - hap ] are distinguished from the probability distribution on the space of alignments under model rm provided the crossover probability is sufficiently small . let @xmath282 and @xmath283 be the @xmath27-forests as described above . we apply theorem [ thm - paths ] for @xmath628 with @xmath629 . we can verify that @xmath630 and @xmath631 . therefore , @xmath245 and @xmath243 give different distributions on the space of alignments . in the following , we describe how @xmath632 and @xmath633 are counted . let @xmath634 denote the directed @xmath27-forests in @xmath245 and @xmath243 consisting of paths @xmath635 and @xmath636 . similarly , we write @xmath637 for other directed @xmath27-forests in @xmath245 and @xmath243 , rooted at @xmath8 , @xmath353 and @xmath114 , respectively . these are the 4 directed @xmath27-forests that have @xmath282 as their undirected @xmath27-forest . all other directed @xmath27-forests have @xmath283 as their undirected @xmath27-forest . _ counting _ @xmath638 : since @xmath639 , we count 16 different contributions to @xmath638 depending on the choices for @xmath640 and @xmath641 in @xmath642 . please refer to figure [ fig - p ] . when @xmath643 and @xmath644 ( left ) , and when @xmath643 and @xmath645 ( right),width=445 ] when @xmath643 and @xmath646 : this is possible only if @xmath647 . there are 4 choices of @xmath40 , and for each of them there are 4 choices for @xmath41 depending on which arc of @xmath634 is replaced to obtain @xmath41 . thus we have 16 sequences @xmath136 for which @xmath648 and @xmath646 . when @xmath643 and @xmath644 : the dashed arcs @xmath649 and @xmath650 in figure [ fig - p ] on the left must be obtained by replacing @xmath651 and @xmath652 , and there are two sequences @xmath136 that achieve this : either @xmath653 or @xmath654 . also , there are 4 possible ways to include arcs pointing to @xmath0 and @xmath655 in @xmath40 . therefore , there are 8 sequences @xmath136 such that @xmath643 and @xmath644 . when @xmath643 and @xmath645 : there are only 2 sequences @xmath136 for which this is possible , since the arcs @xmath656 and @xmath657 ( shown in bold in figure [ fig - p ] on the right ) must already be in @xmath40 . when @xmath643 and @xmath658 : this case is similar to the case @xmath645 . thus there are 28 choices of @xmath136 such that @xmath643 , and similarly 28 choices each for @xmath659 , @xmath660 , and @xmath661 . therefore , there are 112 sequences @xmath136 such that @xmath662 . _ counting _ @xmath633 : again we count 16 different contributions to @xmath633 depending on the choices for @xmath640 and @xmath641 in @xmath663 . please refer to figure [ fig - q ] . when @xmath664 and @xmath644 ( left ) , and when @xmath643 and @xmath658 ( right),width=445 ] when @xmath643 and @xmath646 : as in case of @xmath245 , we have 16 sequences @xmath136 for which @xmath643 and @xmath646 . when @xmath643 and @xmath644 : the dashed arcs @xmath650 and @xmath665 in figure [ fig - q ] on the left must be obtained by replacing @xmath652 and @xmath666 , and arc @xmath667 must already be in @xmath40 . there are two sequences @xmath136 that achieve this : either @xmath653 or @xmath668 . also , there are 2 choices for arcs pointing to @xmath0 , therefore , there are 2 choices for @xmath40 . therefore , there are 4 sequences @xmath136 such that @xmath643 and @xmath669 . when @xmath643 and @xmath645 : this case is similar to the case in which @xmath644 , therefore , there are 4 sequences such that @xmath643 and @xmath645 . when @xmath643 and @xmath658 : in this case , the arcs @xmath670 and @xmath671 must already be in @xmath40 . thus there are only two sequences @xmath136 that are counted depending on whether @xmath672 or @xmath668 . thus there are 26 choices of @xmath136 such that @xmath643 , and similarly 26 choices each for @xmath659 , @xmath660 , and @xmath661 . therefore , there are 104 sequences @xmath136 such that @xmath662 . thus we have verified that @xmath638 and @xmath633 are unequal for @xmath673 , implying that @xmath245 and @xmath243 are distinguished by the probability distribution they induce on the extant sequences under model rm . in this paper we have presented a rigorous mathematical framework for studying pedigree reconstruction problems under probabilistic models . we extended phylogenetic identifiability results to reconstruct pedigrees under an idealised model of recombination and mutation . the main result of this paper is the computation of a class of combinatorial invariants from the joint distribution of extant sequences . as a corollary , we were able to show that certain known examples of pedigrees that could not be distinguished from path lengths alone , could be distinguished by a more detailed analysis of subgraph sequences . here we identify a few open problems and directions for future investigation . _ identifiability of computationally tractable invariants : _ the invariants of a pedigree @xmath58 defined by @xmath434 may be quite difficult to apply in general . theorem [ thm - paths ] may be difficult to use computationally for bigger pedigrees . even for other reconstruction problems in graph theory , computational verifications are difficult . for example , ulam s reconstruction conjecture has been computationally verified to be true only for graphs on at most 11 vertices @xcite . even on restricted classes of graphs , computational reconstruction experiments are difficult to perform . it will be useful to derive from @xmath674 ( or independently ) other identifiable invariants that may be easier to use in computational experiments . theorem [ thm - paths ] only states that if two pedigrees induce the same joint distribution on extant sequences under model rm , then they agree on the invariants @xmath675 . but it will be important to prove a converse or a result of the type : two pedigrees induce the same distribution on extant sequences under model rm if and only if they take the same value for a class of combinatorial invariants . such a result would reduce the identifiability problem to a purely combinatorial problem of proving or disproving that the class of invariants is complete . such a class of invariants may be @xmath676 or it may be somewhat stronger than @xmath677 . it may be possible to compute , by the methods of theorem [ thm - paths ] , other invariants , for example , spanning forest sequences in which consecutive spanning forests are not necessarily separated by just one recombination . _ improving the bound on @xmath145 : _ in the lower bound on @xmath527 in equation ( [ eq - term1b ] ) , we have @xmath398 in the denominator , while @xmath678 is roughly @xmath679 , where @xmath4 is the number of arcs in a pedigree . therefore , it is possible to obtain better bounds on @xmath145 for the applicability of the main theorem for pedigrees with fewer arcs . therefore , we would significantly improve the upper bound on @xmath145 if we showed that a pedigree can be reconstructed from the collection of its subpedigrees ( pedigrees of subsets of the extant population ) of order @xmath10 for some small @xmath10 . in @xcite we made a conjecture about how small @xmath10 may be for pedigrees of order @xmath680 in which the population remains constant over generations . thus solving the purely combinatorial problem of reconstructing pedigrees from their subpedigrees , while important in its own right , will be useful for improving bounds on @xmath145 in theorem [ thm - paths ] . on the other hand it is also likely ( although we do not conjecture ) that theorem [ thm - paths ] is valid without restrictions on @xmath145 , or for all but finitely many values of @xmath145 , or for all @xmath145 except when @xmath223 takes small values . but we do not have good intuition as to why proposition [ prop - n2 - 2 ] ( with @xmath266 ) requires a more complicated argument and an upper bound on @xmath145 . _ maximum likelihood computation of the invariants @xmath681 _ : it will be of interest to derive statistical consistency results and bounds on sequence lengths , analogous to corollary [ cor - consistency ] and lemma [ lem - eps - l1 ] , for computing the invariants @xmath675 . for example , we would like to make the following qualitative statement precise . suppose @xmath58 is a pedigree with @xmath4 arcs . let @xmath607 be given . suppose @xmath474 is an undirected @xmath27-forest sequence of length @xmath269 with no two consecutive @xmath27-forests isomorphic . then there is a sufficiently large @xmath682 such that if a collection of sequences of length @xmath683 evolved on @xmath58 giving an alignment @xmath91 , then the likelihood ratio @xmath684 is large for all pedigrees @xmath60 such that @xmath685 . we expect that @xmath682 would be of the order of @xmath686 . i would like to thank jotun hein and mike steel for many valuable comments and for pointing me to many useful references relevant to this work . in particular , mike s comments on and references to the literature on phylogenetic identifiability results and results on the consistency of ml were very useful . this project was funded by the grant `` from complete genomes to global pedigrees '' from epsrc ( engineering and physical sciences research council ) , u.k . i would like to thank jotun hein and epsrc for supporting my postdoctoral stay at the university of oxford under this grant . i am currently working on the project `` tree - graphs and incidence matrices : theory and applications '' funded by cnpq , brasil ( processo : 151782/2010 - 5 ) at the instituto de matemtica e estatstica , universidade de so paulo , where i completed a significant revision of this paper and improvements on the bounds in the main theorem . i would also like to thank yoshiharu kohayakawa with whom i had very useful conversations during the revision of this paper . guy bresler , elchanan mossel , and allan sly . reconstruction of markov random fields from samples : some observations and algorithms . in a. goel _ et al . _ , editor , _ 11th international workshop , approx 2008 , and 12th international workshop , random 2008 , lncs 5171 _ , pages 343356 , berlin heidelberg , 2008 . springer - verlag . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the problem of time synchronization in large distributed systems consists of giving all the physically disjoint elements of the system a common time scale on which to operate . this common time scale is usually achieved by periodically synchronizing the clock at each element to a reference time source , so that the local time seen by each element of the system is approximately the same . time synchronization plays an important role in many systems in that it allows the entire system to cooperate and function as a cohesive group . time synchronization is an old problem @xcite , but the question of scalability is not . recent advances in sensor networks show a clear trend towards the development of large scale networks with high node density . for example , a hardware simulation - and - deployment platform for wireless sensor networks capable of simulating networks with on the order of 100,000 nodes was recently developed @xcite . as well , for many years the smart dust project sought to build cubic - millimeter motes for a wide range of applications @xcite . also , there is work in progress on the drastic miniaturization of power sources @xcite . these developments ( and many others ) indicate that large scale , high density networks are on the horizon . large scale , high density networks have applications in a variety of situations . consider , for example , the military application of sniper localization . large numbers of wireless nodes can be deployed to find the shooter location as well as the trajectory of the projectile @xcite . since the effective range of a long - range sniper rifle can be nearly @xmath0 km , in order to fully track the trajectory of the projectile it may be essential that our deployed network be tightly synchronized over distances of a few kilometers . another example might be the implementation of a distributed radio for communication . in extracting information from a deployed sensor network , it may be beneficial for the nodes to cooperatively transmit information to a far away receiver @xcite . such an application would require that nodes across the network be well synchronized . as a result , a need for the synchronization of large distributed systems is very real and one that requires careful study to understand the fundamental performance limits on synchronization . the synchronization of large networks has been studied in fields ranging from biology to electrical engineering . the study of synchronous behavior has generally taken one of two approaches . the first approach is to consider synchronization as an emergent behavior in complex networks of oscillators . in that work , models are developed to describe natural phenomena and synchronization emerges from these models . the second approach is to develop and analyze algorithms that synchronize engineering networks . nodes are programmed with algorithms that estimate clock skew and clock offset to achieve network synchronization . however , both of these approaches have significant limitations . emergent synchronization properties in large populations has been the object of intense study in the applied mathematics ( @xcite ) , physics ( @xcite ) , and neural networks ( @xcite ) literature . these studies were motivated by a number of examples observed in nature : * in certain parts of south - east asia , thousands of male fireflies congregate in trees and flash in synchrony at night @xcite . * pacemaker cells of the heart , which on average cause 80 contractions a minute during a person s lifetime @xcite . * the insulin - secreting cells of the pancreas @xcite . for further information and examples , see @xcite , and the references therein . a number of models have been proposed to explain the emergence of synchrony , but perhaps one of the most successful and well known is the model of _ pulse - coupled oscillators _ by mirollo and strogatz @xcite , based on dynamical systems theory . consider a function @xmath1\to[0,1]$ ] that is smooth , monotone increasing , concave down ( i.e. , @xmath2 and @xmath3 ) , and is such that @xmath4 and @xmath5 . consider also a phase variable @xmath6 such that @xmath7 , where @xmath8 is the period of a cycle . then , each element in a group of @xmath9 oscillators is described by a state variable @xmath10 $ ] and a phase variable @xmath11 $ ] as follows : * in isolation , @xmath12 . * if @xmath13 then @xmath14 , and if @xmath15 then @xmath16 . * when @xmath17 for any of the @xmath18 s and some time @xmath19 , then for all other @xmath20 , @xmath21 @xmath22 where @xmath23 denotes an infinitesimal amount of time after @xmath19 . that is , oscillator @xmath18 reaching the end of a cycle causes the state of all other oscillators to increase by the amount @xmath24 , and the phase variable to change accordingly . the state variable @xmath25 can be thought of as a voltage . charge is accumulated over time according to the nonlinearity @xmath26 and it discharges once it reaches full charge , resetting the charging process . upon discharging , it causes all other charges to increase by a fixed amount of @xmath24 , up to the discharge point . for this model , it is proved in @xcite that for all @xmath9 and for almost all initial conditions , the system eventually becomes synchronized . for the network to converge into a synchronous state , one key assumption is that the behavior of every single oscillator is governed by the same function @xmath27 . this means that all oscillators must have the same frequency . from the literature , it appears that this requirement is nearly always needed . as far as we are aware , for a fully synchronous behavior to emerge , the oscillators need to have the same , or nearly the same , oscillation frequencies . the need for nearly identical oscillators presents a significant limitation for emergent synchronization . this emergence of synchrony is clearly desirable and it has been considered for communication and sensor networks in @xcite . however , whether or not these techniques can be adapted to synchronize networks with nodes that have arbitrary oscillator frequencies ( clock skew ) is still unclear . thus , in order to overcome this limitation and find techniques capable of synchronizing a more general class of networks , we turn to algorithms designed to estimate certain unknown parameters such as clock skew . there have been many synchronization techniques proposed for use in sensor networks . these algorithms generally allow each node to estimate its clock skew and clock offset relative to the reference clock . reference broadcast synchronization ( rbs ) @xcite eliminates transmitter side uncertainties by having a transmitter broadcast reference packets to the surrounding nodes . the receiving nodes then synchronize to each other using the arrival of the reference packets as synchronization events . tiny - sync / mini - sync @xcite and the timing - sync protocol for sensor networks ( tpsn ) @xcite organize the network into a hierarchial structure and the nodes are synchronized using pair - wise synchronization . in lightweight tree - based synchronization ( lts ) @xcite , pair - wise synchronization is also employed but the goal of lts is to reduce communication and computation requirements by taking advantage of relaxed accuracy constraints . the flooding time synchronization protocol ( ftsp ) @xcite achieves one - hop synchronization by having a root node broadcast timing information to surrounding nodes . these surrounding nodes then proceed to broadcast their synchronized timing information to nodes beyond the broadcast domain of the root node . this process can continue for multi - hop networks . the problem with each of these traditional synchronization techniques is that synchronization error will increase with each hop . since each node is estimating certain synchronization parameters , i.e. clock skew , there will be inherent errors in the estimate . as a result , a node multiple hops away from the node with the reference clock will be estimating its parameters from intermediate nodes that already have estimation errors . therefore , this introduces a fundamental _ scalability problem _ : as the number of hops across the network grows , the synchronization error across the network will also grow . current trends in network technology are clearly moving us in the direction of large , multi - hop networks . first , sensors are decreasing in size and this size decrease will most likely be accompanied by a decrease in communication range . thus , more hops will be required to traverse a network deployed over a given area . second , as we deploy networks over larger and larger areas , then for a given communication range , the number hops across the network will also increase . in either case , the increased number of hops required to communicate across the network will increase synchronization error . therefore , it is essential that we develop techniques than can mitigate the accumulation of synchronization error over multiple hops . to decrease the error increase in each hop , we need to decrease the estimation error . there are two primary ways of achieving this . first , each node can increase the amount of timing information it obtains from neighboring nodes . for example , from a received timing packet , the node may be able to construct a data point telling it the approximate time at the reference clock and the corresponding time at its local clock . using a collection of these data points , the node can estimate clock skew and clock offset . so instead of using , say , five packets with timing information , a node can wait for ten packets . more data points will generally give better estimates . the drawback to such an approach is the increase in the number of packet exchanges . the second way in which to reduce estimation error is to increase the quality of each data point obtained by the nodes . this can be achieved through improving packet exchange algorithms and time stamping techniques . however , we believe that there is one fundamentally new approach to improving data point quality that has not be carefully studied . this is to use spatial averaging to improve the quality of each data point . the motivation for this approach is very simple . assume that each node has many neighbors . if all nodes in the network are to be synchronized , then the neighbors of any given node will also have synchronization information . is it possible to simultaneously use information from all the neighbors to improve the quality of a timing observation made by a node ? furthermore , it would seem to make sense that with more neighbors , hence more available timing information , the quality of the constructed data point should improve . if this is indeed the case , then achieving synchronization through the use of spatial averaging will provide a fundamentally new trade - off in improving synchronization performance . network designers would simply be able to increase the number and density of nodes to obtain better network synchronization . the study of cooperative time synchronization using spatial averaging is the focus of this work . to obtain a model for developing cooperative synchronization in large wireless networks , we begin by looking at the signals observed by a node in a network with @xmath9 nodes uniformly deployed over a fixed finite area . to start , we assume propagation delay to negligible ( the general case is considered in section [ sec : timedelay ] ) . all nodes transmit a pulse @xmath28 and a node @xmath29 will see a signal @xmath30 which is the superposition of all these pulses , @xmath31 in this expression , @xmath28 is the basic pulse transmitted by each node ( assumed to be the same for all nodes ) . @xmath32 is the ideal pulse transmit time , but since we assume imperfect time synchronization among the nodes we have @xmath33 modelling random errors in the pulse transmission time . @xmath34 models the amplitude loss in the signal transmitted by the @xmath18th node . @xmath35 is the maximum magnitude transmitted by a node . we scale each node s transmission by @xmath9 so that as the network density grows , the total power radiated does not grow unbounded . this model thus describes the received signal seen at a node @xmath29 for a network with @xmath9 nodes and this holds for any @xmath9 . increasing @xmath9 will have two effects : ( a ) node density will increase since the network area is fixed and ( b ) node signal transmission magnitude will decrease due to the @xmath36 scaling . therefore , by increasing @xmath9 this model allows us to study the scalability of networks as node density grows and node size decreases . given that these are the signals observed at each node , we ask : is it possible for @xmath30 to encode a time synchronization signal that will enable all nodes in the network to synchronize their clocks with bounded error , as @xmath37 ? the answer is yes , and the key to proving all our results is the law of large numbers . our key idea is the following . if all nodes were able to determine when time @xmath32 ( in the reference time ) arrives , then by transmitting @xmath28 at time @xmath32 , the signal observed at any node @xmath29 would be @xmath38 , which is a suitably scaled version of @xmath28 centered at @xmath32 . in reality however , there will be some error in the determination of @xmath32 , which we account for by allowing for a node - dependent random error @xmath39 . but , if the distribution of @xmath39 satisfies certain conditions , then the effects of that timing error can be averaged out . a pictorial representation of why this should be the case is shown in fig . [ fig : why - sync - holds ] . therefore , intuitively we can see how the technique of _ cooperative time synchronization _ using spatial averaging can average out the inherent timing errors in each node . even though there is randomness and uncertainty in each node s estimates , by using cooperation among a large number of nodes it is possible to recover _ deterministic _ parameters from the resulting aggregate waveform ( such as the location of certain zero - crossings ) in the limit as node density grows unbounded . thus more nodes will give us better estimates . this is because the random waveform converges to a deterministic one as more and more nodes cooperatively generate an aggregate waveform . at the same time , the average power required by each node will decrease since smaller nodes send smaller signals . therefore , by programming suitable dynamics into the nodes , in this paper we show how it is possible to generate an aggregate output signal with _ zero - crossings in the limit of asymptotically dense networks . thus , the detection of these zero - crossings plays the same role as that of an externally generated time reference signal based on which all nodes can synchronize . we develop this synchronization technique in three main steps . one , we set up the model for @xmath30 . two , we specify characteristics of the model ( i.e. the distribution of @xmath33 ) that allow us to prove desirable properties of the aggregate waveform ( such as a center zero - crossing at @xmath32 ) . three , we develop the estimators needed for our synchronization technique and show that the estimators give us the desired characteristics . the main contributions presented in this paper are the following ; * the definition of a probabilistic model for the study of the time synchronization problem in wireless networks . this model does contain the classical mirollo - strogatz model as a special case , but its formulation and the tools used to prove convergence results are of a completely different nature ( purely probabilistic , instead of based on the theory of dynamical systems ) . * using this model , we provide a rigorous analysis of a new cooperative time synchronization technique that employs spatial averaging and has favorable scaling properties . as the density of nodes increases , synchronization performance improves . in particular , in the limit of infinite density , deterministic parameters for synchronization can be recovered . * we show that cooperative time synchronization works perfectly for negligible propagation delay . when propagation delay is considered , we find that asymmetries at the boundaries reveal some limitations that need to be carefully considered in designing algorithms that take advantage of spatial averaging . in analyzing the proposed cooperative time synchronization technique , our goal is to show that the proposed technique can average out all random error and provide deterministic parameters for synchronization as node density grows unbounded . this asymptotic result can be viewed as a _ convergence in scale _ to synchrony . the result serves as a theoretical foundation for allowing a new trade - off between node density and synchronization performance . in particular , higher node density can yield better synchronization . the rest of this paper is organized as follows . the general model is presented in section [ sec : systemmodel ] . of particular interest here is section [ sec : specialcase ] , where we show how our model contains the model of mirollo and strogatz for pulse - coupled oscillators as a special case @xcite . in section [ sec : timesync - setup ] we specialize the general model for our synchronization setup and develop waveform properties that will be used in time synchronization . in section [ sec : timesynchronization ] we develop the cooperative time synchronization technique for no propagation delay . we extend the cooperative synchronization ideas to the case of propagation delay in section [ sec : timedelay ] . the paper concludes in section [ sec : conclusion ] with a detailed discussion on the scalability issue and how the technique proposed in this work lays the theoretical foundation for a general class of cooperative time synchronization techniques that use spatial averaging . we consider a network with @xmath9 nodes uniformly distributed over a fixed finite area . the behavior of each node @xmath18 is governed by a clock @xmath40 that counts up from @xmath41 . the introduction of @xmath40 is important since it provides a consistent timescale for node @xmath18 . by maintaining a table of pulse - arrival times , node @xmath18 can utilize the arrival times of many pulses over an extended period of time . the clock of one particular node in the network will serve as the reference time and to this clock we wish to synchronize all other nodes . we will call the node with the reference clock node @xmath42 and the clocks of other nodes are defined relative to the clock of node @xmath42 . we never adjust the frequency or offset of the local clock @xmath40 because we wish to maintain a consistent time scale for node @xmath18 . the clock of node @xmath42 , @xmath43 , will be defined as @xmath44 where @xmath45 . taking @xmath43 to be the reference clock , we now define the clock of any other arbitrary node @xmath18 , @xmath40 . we define @xmath40 as @xmath46 where * @xmath47 is an unknown offset between the start times of @xmath40 and @xmath43 . * @xmath48 is a constant and for each @xmath18 , @xmath49 $ ] where @xmath50 are finite . this bound on @xmath51 means that the frequency offsets between any two nodes can not be arbitrarily large . * @xmath52 is a stochastic process modeling random timing jitter . thus , this model assumes that there is a bounded constant frequency offset between the oscillators of any two nodes as well as some random clock jitter . it is important to note that node @xmath42 does not have to be special in any way ; its clock is simply a reference time on which to define the clocks of the other nodes . this means that our clock model actually describes the relative relationship of all the clocks in the network by using an arbitrary node s clock as a reference . from section [ sec : intromodel ] , we see that we are interested in studying the aggregate waveform observed at a node @xmath29 . as a result , we are only concerned with the aggregate signal magnitude and do not care about the particular signal contribution from each surrounding node . with this in mind , we can develop a random model for pathloss that , for dense networks , gives the appropriate aggregate signal magnitude at node @xmath29 . such a model is ideal for our situation since we are studying asymptotically dense networks . we start with a general pathloss model @xmath53 , where @xmath54 for all distances @xmath55 , is non - increasing and continuous . @xmath53 is a fraction of the transmitted magnitude seen at distance @xmath56 from the transmitter . for example , if the receiver node @xmath29 is at distance @xmath56 from node @xmath18 , and node @xmath18 transmits a signal of magnitude @xmath57 , then node @xmath29 will hear a signal of magnitude @xmath58 . we derive @xmath53 from a power pathloss model since any pathloss model captures the _ average _ received power at a given distance from the transmitter . this average received power is perfect for modelling received signal magnitudes in our problem setup since we are considering asymptotically dense networks . due to the large number of nodes at any given distance @xmath56 from the receiver , using the average received magnitude at distance @xmath56 as the contribution from each node at that distance will give a good modelling of the amplitude of the aggregate waveform . the random pathloss variable @xmath59 will be derived from @xmath53 . to understand how @xmath59 and @xmath53 are related , we give an intuitive explanation of the meaning of @xmath59 as follows : the @xmath60 $ ] is the fraction of nodes at distances @xmath56 from node @xmath29 such that @xmath61 , where @xmath62 is a small constant . this means that , roughly speaking , for any given scaling factor @xmath63 , @xmath64 is the fraction of received signals with magnitude scaled by approximately @xmath65 , where @xmath66 is the probability density function of @xmath59 . thus , if we scale the transmit magnitude @xmath57 from every node @xmath18 by an independent @xmath59 , then as the number of nodes , @xmath9 , gets large , node @xmath29 will see @xmath67 signals of approximate magnitude @xmath68 , and this holds for all @xmath65 in the range of @xmath59 . this is because taking a large number of independent samples from a distribution results in a good approximation of the distribution . thus , this intuition tells us that scaling the magnitude of the signal transmitted from every node @xmath18 by an independent sample of the random variable @xmath59 gives an aggregate signal at node @xmath29 that is the same magnitude as if we generated the signal using @xmath53 directly . even though the signals from two nodes at the same distance from a receiver have correlated magnitudes , we do not care about the signal magnitude from any particular node but only that an appropriate number of all possible received signal magnitudes contribute to the aggregate waveform . for a receiving node @xmath29 , we choose therefore to work with the random variable @xmath59 instead of directly with @xmath53 because , for the goals of this paper , doing so has two major advantages : ( a ) we can obtain desirable limit results by placing very minimal restrictions on the distribution of the @xmath59 s ( and hence on @xmath53 ) and ( b ) we can apply tools from probability theory ( basically , the strong law of large numbers ) to carry out our analysis . from the above intuition we can define the cumulative distribution function of @xmath59 as @xmath69 \\ 1 & k\in(1,\infty ) \end{array } \right.\ ] ] where * @xmath70 is the total area of the network , * @xmath71 is the area of the network contained in a circle of radius @xmath72 centered at node @xmath29 , * @xmath73 . from the above discussion we see that the distribution of @xmath59 is only a function of node @xmath29 , the receiving node . we illustrate the relationship among node @xmath29 , @xmath53 , @xmath74 , and @xmath75 in fig [ fig : propmodel - illus ] . we sometimes write @xmath34 with @xmath18 used to index each node surrounding node @xmath29 . @xmath18 is thus indexing a sequence of independent random variables @xmath34 for fixed @xmath29 . therefore , for a given @xmath29 , @xmath34 s are independent and identically distributed ( i.i.d . ) with a cumulative distribution function given by ( [ eq : prop_cdf ] ) for all @xmath18 . we assume that @xmath59 has the following properties : * @xmath59 is independent from @xmath76 for all @xmath29 , @xmath77 , and @xmath78 . * @xmath79 , @xmath80 , and @xmath81 . the requirements on the random variable @xmath59 places restrictions on the model @xmath53 . any function @xmath53 that yields a @xmath59 with the above requirements can be used to model pathloss . in this section we develop a more complex model to simultaneously model propagation delay and pathloss . this leads to the joint development of the delay random variable @xmath82 and a corresponding pathloss random variable @xmath59 . since we want to develop a model for both pathloss and time delay , we start by keeping the pathloss function @xmath53 defined in section [ sec : prop - model ] . the general delay model assumes a function @xmath83 that models the time delay as a function of distance . @xmath83 describes the time in terms of @xmath43 that it takes for a signal to propagate a distance @xmath56 . for example , if node @xmath18 and node @xmath29 are distance @xmath84 apart , then a pulse sent by node @xmath18 at time @xmath85 will be seen at node @xmath29 at time @xmath86 . we make the reasonable assumption that @xmath83 is continuous and strictly monotonically increasing for @xmath55 . as with the pathloss only model , we want to define a delay random variable @xmath82 for each receiving node @xmath29 . recall that this means that for every node @xmath29 there is a random variable @xmath82 associated with it since , in general , each node @xmath29 will see different delays . there is a correlation between the delay random variable @xmath82 and the pathloss random variable @xmath59 . this correlation arises for two main reasons . first , since in section [ sec : prop - model ] we define @xmath53 to be monotonically decreasing and continuous , it is possible for @xmath87 for @xmath88 , @xmath89 . this might be the case for a multi - hop network . in this situation , there will be a set of nodes whose transmissions will never reach node @xmath29 ( i.e. infinite delay ) even though according to @xmath83 these nodes should contribute a signal with finite delay . second , a small @xmath59 value would represent a signal from a far away node . as a result , the corresponding @xmath82 value should be large to reflect large delay . therefore , keeping these two points in mind , we proceed to develop a model for both pathloss and propagation delay . we define the cumulative distribution function of @xmath82 as @xmath90 \\ a(x-\delta(r))+\frac{a(j , r)}{a_{t } } & x\in(\delta(r),\delta(r+\delta r ) ] \\ 1 & x\in(\delta(r+\delta r),\infty ) \end{array } \right.\ ] ] where @xmath91 , @xmath92 is a constant , @xmath93 , and @xmath94 recall that @xmath71 , defined in section [ sec : prop - model ] , is the area of the network contained in a circle of radius @xmath72 centered at node @xmath29 and @xmath70 is the total area of the network . note that @xmath95 can be infinite . using the delay random variable @xmath82 with the cumulative distribution function in ( [ eq : delay_cdf ] ) , we define @xmath59 as @xmath96 where @xmath97 is the deterministic pathloss function from section [ sec : prop - model ] and @xmath98 is the inverse function of the deterministic delay function @xmath99 . note that @xmath100 exists since @xmath99 is continuous and strictly monotonically increasing on @xmath101 . to understand the distribution of @xmath82 , we need to consider the definition of @xmath59 as well . recall that a signal arriving with delay @xmath82 is scaled by the pathloss random variable @xmath59 . let us consider the cumulative distribution in two pieces , @xmath102 $ ] and @xmath103 . the case for @xmath104 is trivial . first , for @xmath102 $ ] , the probability that @xmath82 takes a value less than or equal to @xmath105 is simply the fraction of the network area around node @xmath29 such that the nodes are at distances @xmath56 with @xmath106 . the intuition is the same as that for the development of @xmath59 in section [ sec : prop - model ] . second , for @xmath103 , the situation is more complex . note that a transmitted signal from a node at distance @xmath107 from @xmath29 will arrive at node @xmath29 with infinite delay since @xmath87 for @xmath107 . since any delay values in @xmath103 correspond to distances @xmath108 , the corresponding scaling value will be zero because @xmath59 and @xmath82 are related by ( [ eq : djkjconnection ] ) . as a result , it does not matter what delay values we assign to the fraction of the network area outside a circle of radius @xmath95 centered at node @xmath29 as long as their delay value @xmath105 is such that @xmath109 . thus , we can arbitrarily choose a constant @xmath110 value and construct a piecewise linear portion of the cumulative distribution function of @xmath82 on @xmath103 . the probability that @xmath111 will be the fraction of the network area outside a circle of radius @xmath95 around node @xmath29 . and since @xmath111 will have a corresponding @xmath59 value that is zero , this fraction of nodes will not contribute to the aggregate waveform at node @xmath29 . it is clear that the correlated @xmath82 and @xmath59 random variables work together to accurately model a signal arriving with both pathloss and propagation delay . an illustration of how @xmath53 , @xmath83 , node @xmath29 , and @xmath112 are related can be found in fig . [ fig : delaymodel - illus ] . we require that @xmath82 is bounded , has finite expectation , and has finite variance for all @xmath29 . note that @xmath113 by the requirement that @xmath114 . as well , since the cumulative distribution in ( [ eq : delay_cdf ] ) is continuous , and often absolutely continuous , we assume that @xmath82 has a probability density function @xmath115 . when we write @xmath116 , the @xmath18 indexes each node surrounding node @xmath29 . thus , the @xmath116 s are independent and identically distributed in @xmath18 for a given @xmath29 and have a cumulative distribution given by ( [ eq : delay_cdf ] ) . using the @xmath59 and @xmath82 developed in this section to simultaneously model pathloss and propagation delay , respectively , we will be able to closely approximate the received aggregate waveform at any node @xmath29 as @xmath37 . to summarize , we see that our choice of the pathloss and delay random variables will depend on what we want to model . if we only consider pathloss and not propagation delay , then we will use the random variable @xmath59 defined in section [ sec : prop - model ] . if we account for both pathloss and delay , then we will use the delay random variable @xmath82 in this section ( section [ sec : delay - model ] ) and the pathloss random variable @xmath59 defined by ( [ eq : djkjconnection ] ) . the exchange of pulses is the method through which the network will maintain time synchronization . each node @xmath18 will periodically transmit a scaled pulse @xmath117 , where @xmath118 is a constant and @xmath28 , in general , can be any pulse . we call the interval of time during which a synchronization pulse is transmitted a _ synchronization phase_. what each node does with a set of pulse arrival observations is determined by the pulse - connection function @xmath119 for node @xmath18 . the pulse - connection function is a function that determines the time , in the time scale of @xmath40 , when node @xmath18 will send its @xmath120th pulse . it can be a function of the current value of @xmath121 and past pulse arrival times . this function basically determines how any node @xmath18 reacts to the arrival of a pulse . the system model that we presented thus far is powerful because it is very general . in this section we show that it is a generalization of the pulse - coupled oscillator model proposed by mirollo and strogatz @xcite . as a result , the results presented in that paper will hold under the simplified version of our model . in setting up the system model , mirollo and strogatz make four key assumptions : * pathloss model : the first assumption that is made is that there is all - to - all coupling among all @xmath9 oscillators . this means that each oscillator s transmission can be heard by all other oscillators . thus , for our model we ignore pathloss , i.e. @xmath122 , to allow any node s transmission to be heard by each of the other @xmath123 nodes . * delay model : the second assumption is that there is instantaneous coupling . this assumption is the same as setting @xmath124 . in such a situation we would use our pathloss only model . * synchronization pulses : the third key assumption made in @xcite is that there is non - uniform coupling , meaning that each of the @xmath9 oscillators fire with strengths @xmath125 . we modify the parameters in our model by making node @xmath18 transmit with magnitude @xmath126 . they also assume that any two pulses transmitted at different times will be seen by an oscillator as two separate pulses . in our model , we may choose any pulse @xmath28 that has an arbitrarily short duration and each node will detect the pulse arrival time and pulse magnitude . * clock model : the fourth important assumption made by mirollo and strogatz is that the oscillators are identical but they start in arbitrary initial conditions . we simplify our clock model in ( [ eq : clock ] ) by eliminating any timing jitter , i.e. @xmath127 , and making the clocks identical by setting @xmath128 for @xmath129 . we leave @xmath47 in the model to account for the arbitrary initial conditions . we also assume that the phase variable in the pulse - coupled oscillator model increases at the same rate as our clock . that is , the time it takes the phase variable to go from zero to one and the time it takes our clock to count from one integer value to the next are the same . now that we have identical system models , what remains is to modify our model to mimic the coupling action detailed in @xcite . this is accomplished by defining a proper pulse - connection function @xmath119 . to match the coupling action in @xcite , we choose a pulse transmit time function @xmath130 that is a function of pulse receive times and also the time of node @xmath18 s @xmath131th pulse transmission time . @xmath132 is the time in terms of @xmath40 that node @xmath18 receives its @xmath65th pulse since its last pulse transmission at @xmath133 . in this case , @xmath119 will be a function that updates node @xmath18 s @xmath120th pulse transmission time each time node @xmath18 receives a pulse . let @xmath134 where it is node @xmath18 s @xmath120th pulse transmission time after observing @xmath65 pulses since its last pulse transmission . node @xmath18 will transmit its pulse as soon as @xmath135 where @xmath121 is node @xmath18 s current time . as soon as the node transmits a pulse at @xmath119 the function will reset and become @xmath136 . the node is now ready to receive pulses and at its first received pulse , the next transmission time will become @xmath137 . @xmath119 will thus be defined as @xmath138 , \qquad k>0\\ x_{n , i}^{c_{i}}(0 ) & = & x_{n-1,i}^{c_{i}}+1 \label{eq : pco_estimator2}\end{aligned}\ ] ] where the pulse received at @xmath132 is a pulse of magnitude @xmath139 and the function @xmath1\to [ 0,1]$ ] is the smooth , monotonic increasing , and concave down function defined in @xcite . equations ( [ eq : pco_estimator ] ) and ( [ eq : pco_estimator2 ] ) fundamentally say that each time node @xmath18 receives a pulse , node @xmath18 s next transmission time will be adjusted . this is in line with the behavior of the coupling model described by mirollo and strogatz since each time an oscillator receives a pulse , its state variable is pulled up by @xmath140 thus adjusting the time at which the oscillator will next fire . to see how equations ( [ eq : pco_estimator ] ) and ( [ eq : pco_estimator2 ] ) relate to the coupling model in @xcite , let us consider an example with two pulse coupled oscillators . consider two oscillators @xmath57 and @xmath141 illustrated in fig . [ fig : strogatzmodel ] . in fig . [ fig : strogatzmodel](a ) , we have that oscillator @xmath57 is at phase @xmath142 and oscillator @xmath141 is just about to fire . below the pulse - coupled oscillator model we have a time axis for node @xmath18 corresponding to our clock model going from time @xmath133 to @xmath143 . our time axis for node @xmath18 models the behavior of oscillator @xmath57 , that is , we want node @xmath18 to behave in the same way as oscillator @xmath57 under the influence of oscillator @xmath141 . if oscillator @xmath141 did not exist , then the phase variable @xmath142 will match our clock in that @xmath142 reaches @xmath42 at the same time our clock reaches @xmath144 and oscillator @xmath57 will fire at the same time our model fires . in fig . [ fig : strogatzmodel](b ) , oscillator @xmath141 has fired and has pulled the state variable of oscillator @xmath57 up by @xmath140 . this coupling has effectively pushed the phase of oscillator @xmath57 to @xmath145 and decreased the time before @xmath57 fires . in fact , the time until oscillator @xmath57 fires again is decreased by @xmath56 . we can capture this coupling in our model since we can calculate the lost time @xmath56 . the time at which oscillator @xmath141 fires is @xmath146 and it is clear that @xmath147 . thus , if the time that oscillator @xmath57 will fire again is decreased by time @xmath56 due to the pulse of @xmath141 , then we adjust our node firing time by decreasing the firing time to @xmath148 . this is exactly the expression in ( [ eq : pco_estimator ] ) for @xmath149 . this relationship between our model for calculating the node firing time and the pulse - coupled oscillator coupling model can be easily extended to @xmath9 oscillators . we can see then that the pulse - coupled oscillator model proposed by mirollo and strogatz in @xcite is a special case of our model . our model generalizes this pulse - coupled oscillator model by considering timing jitter , pulses of finite width , propagation delay , non - identical clocks , and an ability to accommodate arbitrary coupling functions . just as we could specialize our model to the pulse - coupled oscillator model of mirollo and strogatz , we now specialize the model for our proposed synchronization technique . we start under the assumption of no propagation delay and develop the synchronization technique for this case . propagation delay is considered in section [ sec : timedelay ] . we proceed in three steps . in section [ sec : signal - reception ] , we specify the model for @xmath150 , the received waveform at any node @xmath29 . second , in section [ sec : pulseproperties ] , we prove that given certain characteristics of the model , @xmath150 has very useful limiting properties . third , we show in section [ sec : timesynchronization ] that estimators ( i.e. , the pulse connection function ) developed for our synchronization technique give @xmath150 the desired properties . for our synchronization technique , we specialize the general model by making the following assumptions on @xmath51 and @xmath52 for @xmath151 : * a characterization of the @xmath152 is given by a known function @xmath153 with @xmath154 $ ] that gives the percentage of nodes with any given @xmath155 value . thus , the fraction of nodes with @xmath155 values in the range @xmath156 to @xmath157 can be found by integrating @xmath153 from @xmath156 to @xmath157 . we assume that @xmath158 , for some constant @xmath159 . we keep this function constant as we increase the number of nodes in the network ( @xmath37 ) . given any circle of radius @xmath95 that intersects the network , the nodes within that circle will have @xmath51 s that are characterized by @xmath153 . @xmath95 is the maximum @xmath56 such that @xmath160 . this means that the set of nodes that any node @xmath29 will hear from will have its @xmath51 s characterized by a known function . note that @xmath95 can be infinite , and in that case , any node @xmath29 hears from all nodes in the network . fundamentally , @xmath153 means that as we increase node density , the new nodes have @xmath155 parameters that are well distributed in a predictable manner . * @xmath52 is a zero mean gaussian process with samples @xmath161 , for any @xmath162 , and independent and identically distributed samples for any set of times @xmath163 $ ] , @xmath65 a positive integer . we assume @xmath164 and note that @xmath165 is defined in terms of the clock of node @xmath18 . we assume that @xmath52 is gaussian since the rms ( root mean square ) jitter is characterized by the gaussian distribution @xcite . we maintain the full generality of the pathloss model from section [ sec : prop - model ] . note that throughout this work we assume no transmission delay or time - stamping error . this means that a pulse is transmitted at exactly the time the node intends to transmit it . we make this assumption since there will be no delay in message construction or access time @xcite because our nodes broadcast the same simple pulse without worrying about collisions . also , when a node receives a pulse it can determine its clock reading without delay since any time stamping error is small and can be absorbed into the random jitter . for our proposed synchronization technique , the aggregate waveform seen by node @xmath29 at any time @xmath78 is @xmath166 where @xmath150 is the waveform seen at node @xmath29 written in the time scale of @xmath43 and @xmath167 for all @xmath18 . @xmath35 is the maximum transmit magnitude of a node . @xmath33 is the random timing offset suffered by the @xmath18th node , which encompasses the random clock jitter and estimation error . this model says that each node @xmath18 s pulse transmission occurs at the ideal transmit time @xmath32 plus some random error @xmath33 . in the next section , section [ sec : pulseproperties ] , we find properties for @xmath33 that will give us desirable properties in @xmath150 . then , in section [ sec : timesynchronization ] , we show that our proposed steady - state synchronization technique and its associated pulse - connection function will give us the desired properties . there are two comments about ( [ eq : timesync - aggwaveform ] ) that we want to make . first , note that even though we sum the transmissions from all @xmath9 nodes in ( [ eq : timesync - aggwaveform ] ) , we do not assume that node @xmath29 can hear all nodes in the network . recall from the pathloss model that if we have a multi - hop network , then there will be a nonzero probability that @xmath168 . thus , node @xmath29 will not hear from the nodes whose transmissions have zero magnitude . second , it may be possible that the nodes are told that there are @xmath169 nodes in the network while the actual number of functioning nodes is @xmath9 . in which case , each node will transmit with signal magnitude @xmath170 and ( [ eq : timesync - aggwaveform ] ) will have a factor of @xmath171 . other than for this factor , however , the theoretical results that follow are not affected . to model the quality of the reception of @xmath150 by node @xmath29 , we model the reception of a signal by defining a threshold @xmath172 . @xmath172 is the received signal threshold required for nodes to perfectly resolve the pulse arrival time . if the maximum received signal magnitude is less than @xmath172 then the node does not make any observations and ignores the received signal waveform . we assume that @xmath173 . in our work we will assume that @xmath28 takes on the shape @xmath174 where @xmath175 is expressed in terms of @xmath43 . we assume @xmath176 for @xmath177 , @xmath178 only on @xmath179 , @xmath180 , and @xmath181 is uniformly continuous on @xmath182 . thus , we see that @xmath28 has at most three jump discontinuities ( at @xmath183 ) . @xmath184 should be chosen large compared to @xmath185 , i.e. @xmath186 , where @xmath187 is the value of @xmath165 translated from the time scale of @xmath40 to @xmath43 . this way , over each synchronization phase , with high probability a zero - crossing will occur . for each node , the duration in terms of @xmath43 of a synchronization phase will be @xmath188 . note that we assume @xmath184 is a value that is constant in any consistent time scale . this means that even though nodes have different clocks , identical pulses are transmitted by all nodes . we define a pulse to be transmitted at time @xmath78 if the pulse makes a zero - crossing at time @xmath78 . similarly , we define the _ pulse receive ( arrival ) time _ for a node as the time when the observed waveform first makes a zero - crossing . a _ zero - crossing _ is defined for signals that have a positive amplitude and then transition to a negative amplitude . it is the time that the signal first reaches zero . for the exchange of synchronization pulses , we assume that nodes can transmit pulses and receive signals at the same time . this simplifying assumption is not required for the ideas presented here to hold , but simplifies the presentation . we mention a way to relax this assumption in section [ sec : nosimulttxrx ] . in ( [ eq : timesync - aggwaveform ] ) and in the discussions above , we have focused on characterizing the aggregate waveform for any one synchronization phase . that is , ( [ eq : timesync - aggwaveform ] ) is the waveform seen by any node @xmath29 for the synchronization phase centered around node @xmath42 s transmission at @xmath189 , @xmath32 a positive integer . we can , however , describe a synchronization pulse train in the following form , @xmath190 where @xmath191 is the integer value of @xmath78 at the @xmath192th synchronization phase , and @xmath193 is the error suffered by the @xmath18th node in the @xmath192th synchronization phase . we seek to create this pulse train with equispaced zero - crossings and use each zero - crossing as a synchronization event . an illustration of such a pulse train is shown in fig . [ fig : pulsetrain ] . for simplicity , however , most of the theoretical work is carried out on one synchronization phase . in this section , we characterize the properties of @xmath33 that give us desirable properties in the aggregate waveform . from ( [ eq : timesync - aggwaveform ] ) , the aggregate waveform seen at each node @xmath29 in the network has the form @xmath194 we have dropped the @xmath29 and @xmath43 for notational simplicity since in this section we deal solely with the received waveform at a node @xmath29 in the time scale of @xmath43 . as we let the number of nodes grow unbounded ( @xmath37 ) , the properties of this limit waveform can be characterized by theorem [ theorem : main ] . these properties will be essential for asymptotic cooperative time synchronization . as a note , in theorem [ theorem : main ] we present the case for gaussian distributed @xmath33 but similar results hold for arbitrary zero - mean , symmetrically distributed @xmath33 with finite variance . [ theorem : main ] let @xmath28 be as defined in equation ( [ eq : poft ] ) and @xmath195 with @xmath196 a constant and @xmath197 for all @xmath18 , @xmath141 a constant . also , let @xmath198 be defined as in section [ sec : prop - model ] and be independent from @xmath33 for all @xmath18 . then , @xmath199 has the properties * @xmath200 , * @xmath201 for @xmath202 , and @xmath203 for @xmath204 for some @xmath205 . * @xmath206 is odd around @xmath189 , i.e. @xmath207 for @xmath208 * @xmath209 is continuous . @xmath210 the properties outlined in theorem [ theorem : main ] will be key to the synchronization mechanism we describe . the specific value of @xmath211 will be determined by our choice of the pulse - connection function . before we prove theorem [ theorem : main ] in section [ sec : theoremproof ] we develop and motivate a few important related lemmas . at time @xmath212 , we have that @xmath213 where @xmath214 . we have the mean of @xmath215 being @xmath216 where @xmath217 is the gaussian pdf @xmath218 it is clear that the @xmath215 s , for different @xmath18 s , do not have the same mean and do not have the same variance since the two quantities depend on the @xmath51 value . since the @xmath51 s are characterized by @xmath153 ( defined in section [ sec : systemparameters ] ) , we write the gaussian distribution for @xmath8 as @xmath219 and @xmath215 is in fact a function of @xmath220 as well , denoted @xmath221 . using @xmath222 and @xmath221 , the notation makes it clear that we can average over the @xmath51 s that are characterized by @xmath153 . we use the results of lemmas [ lemma : polarity_positive ] and [ lemma : polarity_negative ] to prove the polarity result for @xmath206 in section [ sec : theoremproof ] . [ lemma : polarity_positive ] given the sequence of independent random variables @xmath215 with @xmath223 , @xmath224 , and @xmath225 . then , for all @xmath18 , @xmath226 @xmath227 for some constants @xmath228 , @xmath229 , and @xmath230 and @xmath231 almost surely , where @xmath232 [ lemma : polarity_negative ] given the sequence of independent random variables @xmath215 with @xmath233 , @xmath224 , and @xmath225 . then , for all @xmath18 , @xmath234 @xmath235 for some constants @xmath228 , @xmath229 , and @xmath230 and @xmath236 almost surely , where @xmath237 the results of lemma [ lemma : polarity_positive ] and lemma [ lemma : polarity_negative ] are intuitive since given that @xmath28 is odd and the gaussian noise distribution is symmetric , it makes sense for @xmath206 to have properties similar to an odd waveform . since the proofs of the two lemmas are very similar , we only prove lemma [ lemma : polarity_positive ] . the proof can be found in the appendix . knowing only the polarity of @xmath206 is not entirely satisfying since we would also expect that the limiting waveform be continuous . the proof of lemma [ lemma : continuity1 ] is once again left for the appendix . [ lemma : continuity1 ] using @xmath28 in ( [ eq : poft ] ) , @xmath238 is a continuous function of @xmath78 , where @xmath239 we can proceed in a straightforward manner to show that @xmath240 . for @xmath241 , @xmath242 where @xmath243 . since our goal is to apply some form of the strong law of large numbers , we first examine the mean of @xmath244 . we have that @xmath245 @xmath246 @xmath247 . furthermore , @xmath248 since @xmath249 is odd and @xmath217 is even because it is zero - mean gaussian . thus , @xmath250 . we next consider the variance of @xmath244 : @xmath251 where we have used the fact that @xmath252 and @xmath253 . from the preceding discussion we see that the @xmath244 s are a sequence of zero mean , finite ( but possibly different ) variance random variables . from stark and woods @xcite , we know that if @xmath254 , then we have strong convergence of the @xmath244 s : @xmath255 with probability-1 as @xmath37 . but it is easy to see that @xmath256 so the condition is satisfied . as a result , @xmath257 as @xmath258 . we have that @xmath206 is continuous from lemma [ lemma : continuity1 ] . thus , next we need to show that @xmath201 for @xmath259 , and @xmath203 for @xmath260 for some @xmath261 . we show the case for @xmath262 by simply applying lemma [ lemma : polarity_positive ] . since lemma [ lemma : polarity_positive ] holds for all @xmath223 , there clearly exists a @xmath263 such @xmath264 for @xmath259 . the case for @xmath265 comes similarly from lemma [ lemma : polarity_negative ] . lastly , it remains to be shown that @xmath206 is odd around @xmath189 . this , however , is evident from the form of @xmath266 . since @xmath222 is even in @xmath267 about @xmath41 and @xmath249 is odd about @xmath41 , it is clear that @xmath268 as a function of @xmath78 is odd about @xmath32 . thus , @xmath266 is odd around @xmath32 . this then completes the proof for theorem [ theorem : main ] . in this work we want to show that as we let @xmath37 then we can recover deterministic parameters that allow for time synchronization . such a result would provide rigorous theoretical support for a new trade - off between network density and synchronization performance . to simplify the study , we focus on the steady - state time synchronization properties of asymptotically dense networks . in particular , we develop a cooperative technique that constructs a sequence of equispaced zero - crossings seen by all nodes which allows the network to maintain time synchronization indefinitely given that the nodes start with a collection of equispaced zero - crossings . starting with a few equispaced zero - crossings allows us to avoid the complexities of starting up the synchronization process but still allows us to show that spatial averaging can be used to average out timing errors . if we are able to maintain indefinitely a sequence of equispaced zero - crossing using cooperative time synchronization , then it means that spatial averaging can average out all uncertainties in the system as we let node density grow unbounded . this recovery of deterministic parameters is our desired result . here , we overview the estimators needed for cooperative time synchronization . let @xmath269 be the time , with respect to clock @xmath270 , that the @xmath18th node sees its @xmath120th pulse . in dealing with the steady - state properties , we start by assuming that each node @xmath18 in the network has observed a sequence of @xmath271 pulse arrival times , @xmath272 , that occur at integer values of @xmath78 , @xmath271 is an integer . recall that @xmath272 is defined as a set of @xmath271 pulse arrival times in the time scale of @xmath40 . therefore , even though @xmath272 occur at integer values of @xmath78 ( the time scale of @xmath43 ) , these values are not necessarily integers since they are in the time scale of @xmath40 . note also that in our model the pulse arrival time is a zero - crossing location . using these @xmath271 pulse arrival times , each node @xmath18 has two distinct , yet closely related tasks . the first task is time synchronization . to achieve time synchronization , node @xmath18 wants to use these @xmath271 pulse arrival times to make an estimate of when the next zero - crossing will occur . if it can estimate this next zero - crossing time , then it can effectively estimate the next integer value of @xmath78 . this estimator can then be extended to estimate arbitrary times in the future which gives node @xmath18 the ability to synchronize to node @xmath42 . the second task is that node @xmath18 needs to transmit a pulse so that the sum of all pulses from the @xmath9 nodes in the network will create an aggregate waveform that , in the limit as @xmath37 , will give a zero - crossing at the next integer value of @xmath78 . this second task is very significant because if the aggregate waveform gives the exact location of the next integer value of @xmath78 , then each node @xmath18 in the network can use this new zero - crossing along with @xmath273 to form a set of @xmath271 zero - crossing locations . this new set can then be used to predict the next zero - crossing location as well as node @xmath18 s next pulse transmission time . recall that determining the pulse transmission time is the job of the pulse - connection function @xmath119 . with such a setup , synchronization would be maintained indefinitely . the zero - crossings that always occur at integer values of @xmath78 would provide node @xmath18 a sequence of synchronization events and also illustrate how cooperation is averaging out all random errors . the waveform properties detailed in theorem [ theorem : main ] play a central role in accomplishing the nodes task of cooperatively generating an aggregate waveform with a zero - crossing at the next integer value of @xmath78 . from ( [ eq : waveformgeneralform ] ) , if the arrival time of any pulse at a node @xmath29 is a random variable of the form @xmath274 , where @xmath32 is the next integer value of @xmath78 and @xmath33 is zero - mean gaussian ( or in general any symmetric random variable with zero - mean and finite variance ) , then theorem [ theorem : main ] tells us that the aggregate waveform will make a zero - crossing at the next integer value of @xmath78 . this idea is illustrated in fig . [ fig : example ] . thus , for achieving time synchronization in an asymptotically dense network we need to address two issues . first , we need to develop an estimator for the next integer value of @xmath78 given a sequence of @xmath271 pulse arrival times that occur at integer values of @xmath78 . we will call this the _ time synchronization estimator _ and let us write @xmath275 as the time synchronization estimator that determines the time , in the time scale of @xmath40 , when node @xmath18 predicts it will see its @xmath120th zero - crossing . two , we need to develop the pulse - connection function @xmath119 such that node @xmath18 s transmitted pulse will arrive at a node @xmath29 with the random properties described in theorem [ theorem : main ] . here we establish the conditions for estimating the next pulse arrival time , or equivalently the next integer value of @xmath78 , given @xmath271 pulse arrival times . these conditions apply most directly to the time synchronization estimator @xmath275 since we want to synchronize in some desired manner . the problem of synchronization is the challenge of having the @xmath18th node accurately and precisely predict when the next integer value of @xmath78 will occur . in our setup , the reception of a pulse by node @xmath18 tells it of such an event . let us explicitly model the time at an integer value of @xmath78 in terms of the clock of node @xmath18 . assume @xmath32 is an integer value of @xmath78 and at this time , node @xmath18 will observe its @xmath120th pulse . thus , from ( [ eq : clock ] ) we have that @xmath276 the equation makes use of the clock model of node @xmath18 ( [ eq : clock ] ) to tell us the time at clock @xmath40 when node @xmath42 is at @xmath32 , where @xmath32 is an integer in the time scale of @xmath43 . we are also starting with the assumption that the zero - crossing that occurs at an integer value of @xmath78 is observed by node @xmath18 at this time . from ( [ eq : state - eqns ] ) we see that the pulse receive time at node @xmath18 , @xmath277 , is a gaussian random variable whose mean is parameterized by the unknown vector @xmath278 $ ] . thus , to achieve synchronization node @xmath18 will try to estimate the random variable @xmath277 using a series of @xmath271 pulse receive times as observations ( recall that @xmath271 is known ) . note that the observations are also random variables with distributions parameterized by @xmath279 . we want the time synchronization estimator of node @xmath18 to make an estimate of @xmath277 , denoted @xmath280 which is a function of past observations @xmath281 , that meets the following criteria : @xmath282 = e_{\vartheta}(t_{n , i}^{c_{i } } ) \label{eq : opt1}\ ] ] @xmath283 \label{eq : opt2}\ ] ] for all @xmath279 . the subscript @xmath279 means that the expectation is taken over the distributions involved given any possible @xmath279 . the first condition comes from the fact that given a finite @xmath271 , it is reasonable to want the expected value of the estimate to be the expected value of the random variable being estimated for all @xmath279 . as in the justification for unbiased estimators , this condition eliminates unreasonable estimators so that the chosen estimator will perform well , on average , for all values of @xmath279 @xcite . the second condition is the result of seeking to minimize the mean squared error between the estimate and the random variable being estimated for all @xmath279 . for the time synchronization estimator , node @xmath18 will seek to estimate @xmath277 given @xmath272 . from ( [ eq : state - eqns ] ) , we see that @xmath284^{t}$ ] is a jointly gaussian random vector parameterized by @xmath279 . recall that we assume @xmath52 is a zero mean gaussian process with independent and identically distributed samples @xmath285 , for any @xmath78 . also , since we re assuming that the zero - crossings at node @xmath18 occur at consecutive integer values of @xmath78 , the random variable @xmath286 is gaussian with @xmath287 for some @xmath288 $ ] . we also notice that @xmath289 since each noise sample is independent , we see that the distribution of @xmath290 parameterized by @xmath279 can be written as @xmath291 where @xmath292\ ] ] and @xmath293 . as a result , for any @xmath271 consecutive observations , we can simplify notation by using the model @xmath294 where @xmath295^{t } = [ t_{n - m , i}^{c_{i } } \quad t_{n - m+1,i}^{c_{i } } \dots t_{n-1,i}^{c_{i}}]^{t}$ ] and @xmath296 = \left [ \begin{array}{c } \alpha_{i}(\tau_{0}-m-\bar{\delta}_{i})\\ \alpha_{i } \end{array } \right]\ ] ] with @xmath297^t\ ] ] and @xmath298^{t}$ ] . since @xmath52 is a gaussian noise process , @xmath299 with @xmath293 . using the simplified notation in ( [ eq : simple - obs ] ) , we want to estimate @xmath300 , where @xmath300 is jointly distributed with @xmath301 as @xmath302 \sim { \mathcal n } ( \left [ \begin{array}{c } { \mathbf m}\\ \theta_{1}+m\theta_{2 } \end{array } \right ] , \left [ \begin{array}{cc } \sigma & 0\\ 0 & \sigma^{2 } \end{array } \right]).\ ] ] using this notation , we can rewrite the synchronization criteria as : @xmath303=e_{\theta}(y_{m+1})\ ] ] @xmath304,\ ] ] where @xmath305 is the estimator for @xmath300 . condition ( [ eq : opt1-simple ] ) implies that our estimate must be unbiased . condition ( [ eq : opt2-simple ] ) is equivalent to @xmath306.\ ] ] to see this equivalence , note that @xmath307 } \nonumber \\ & = & e_{\theta}\big[(\hat{y}_{m+1}(y_{1 } , y_{2 } , \dots , y_{m})-(\theta_{1}+m\theta_{2})-w_{m+1 } ) ^{2}\big ] \nonumber \\ & = & e_{\theta}\big[(\hat{y}_{m+1}(y_{1 } , y_{2 } , \dots , y_{m})- ( \theta_{1}+m\theta_{2 } ) ) ^{2}\big]+e\big [ w_{m+1}^{2}\big],\end{aligned}\ ] ] where the last inequality follows from the independence of @xmath308 from all other noise samples . since the distribution of of @xmath308 is independent of @xmath309 , @xmath310 } \nonumber \\ & = & \textrm{argmin}_{\hat{y}_{m+1 } } e_{\theta}\big[(\hat{y}_{m+1}(y_{1 } , y_{2 } , \dots , y_{m})- ( \theta_{1}+m\theta_{2 } ) ) ^{2}\big ] \nonumber.\end{aligned}\ ] ] with these two conditions , from @xcite we see that the desired estimate for @xmath300 will be the uniformly minimum variance unbiased ( umvu ) estimator for @xmath311 . using the above linear model , from @xcite we know the maximum likelihood ( ml ) estimate of @xmath309 , @xmath312 , is given by @xmath313 this estimate achieves the cramer rao lower bound , hence is efficient . the fisher information matrix is @xmath314 and @xmath315 . this means that @xmath312 is umvu . again from @xcite , the invariance of the ml estimate tells us that the ml estimate for @xmath316 is @xmath317 . first , it is clear that @xmath318 , where @xmath319 $ ] . as a result , we first see that @xmath320 so @xmath321 is unbiased . next , to see that @xmath321 is also minimum variance we compare its variance to the lower bound . @xmath322 the extension of the cramer rao lower bound in @xcite to a function of parameters tells us that @xmath323 with @xmath324 . in this case , @xmath325 $ ] so the lower bound to the mean squared error is @xmath326 as a result , we see that @xmath321 is umvu . since @xmath321 is the desired estimate of where the next pulse arrival time will be , it is the time synchronization estimator . thus , @xmath327 note that @xmath328 has a variance that goes to zero as @xmath329 . we now need to develop the pulse - connection function so that the conditions for @xmath33 in theorem [ theorem : main ] are satisfied . recall we are developing the synchronization technique under the assumption of no propagation delay , i.e. @xmath330 . given a sequence of @xmath271 pulse arrival times , the time synchronization estimator @xmath275 given in ( [ eq : timesync - estimator ] ) gives each node the ability to predict the next integer value of @xmath78 . what remains to be considered is the second part of the synchronization process : developing a pulse - connection function @xmath119 such that the aggregate waveform seen by a node @xmath29 will have the properties described in theorem [ theorem : main ] . let us first consider the distribution of @xmath275 . from ( [ eq : timesync - estimator - distribution ] ) , we have that @xmath331 using ( [ eq : clock ] ) , we can translate @xmath332 into the time scale of @xmath43 as @xmath333 which gives @xmath334 this means that @xmath335 under our assumption of @xmath330 , any transmission by node @xmath18 will be instantaneously seen by any node @xmath29 . as a result , the random variable @xmath336 will be seen as the pulse arrival time at node @xmath29 , in the time scale of @xmath43 . due to the assumption of no propagation delay , defining @xmath337 will give us the desired properties in the aggregate waveform . to see this , let us compare the distribution of @xmath338 to the assumptions of theorem [ theorem : main ] . since @xmath32 is the ideal crossing time in the time scale of @xmath43 , we have @xmath339 therefore , we see that @xmath340 where @xmath211 from theorem [ theorem : main ] is @xmath341 we have shown that using the pulse connection function @xmath337 satisfies the conditions of theorem [ theorem : main ] . thus , all the results of the theorem apply . as a result , we have established a time synchronization estimator @xmath336 and a pulse - connection function @xmath338 . in the case of @xmath330 , we have that @xmath337 , or in the time scale of @xmath40 , @xmath342 . when each node in the network uses the pulse - connection function @xmath343 we have a resulting aggregate waveform that has a zero - crossing at the next integer value of @xmath78 as @xmath37 . this fact follows from applying theorem [ theorem : main ] . thus , we have an asymptotic steady - state time synchronization method that can maintain a sequence of equispaced zero - crossings occurring at integer values of @xmath78 . an interesting feature of this synchronization technique is that no node needs to know any information about its location or its surrounding neighbors . before ending this section , let us comment on the assumption of simultaneous transmission and reception . one way to relax this assumption is to divide the network into two disjoint sets of nodes , say the odd numbered nodes and the even numbered nodes , where each set is still uniformly distributed over the area . then , the odd nodes and the even nodes will take turns transmitting and receiving . for example , the odd numbered nodes can transmit pulses at odd values of @xmath78 and the even numbered nodes will listen . the even numbered nodes will then transmit pulses at the even values of @xmath78 and the odd numbered nodes will listen . with such a scheme , nodes do not transmit and receive pulses simultaneously , but can still take advantage of spatial averaging . the odd numbered nodes will see an aggregate waveform generated by a subset of the even numbered nodes and the even numbered nodes will receive a waveform cooperatively generated by the odd numbered nodes . let us take a more detailed look at this scheme . in fig . [ fig : nosimulttxrx ] we assume that @xmath32 is an even integer value of @xmath78 and use @xmath344 . each even numbered node will use the aggregate signals occurring at @xmath345 , @xmath346 , and @xmath347 to estimate @xmath32 and cooperatively the even nodes will generate the aggregate signal at @xmath32 . the odd numbered nodes will then use the aggregate signals occurring at @xmath348 , @xmath349 , and @xmath32 to generate the aggregate signal at @xmath350 . therefore , the odd and even numbered nodes can take turns transmitting and receiving signals and nodes never need to simultaneously transmit and receive . of course , such a setup would require a modification of the estimators used by the nodes . nodes will receive a vector of @xmath271 observations @xmath351 with @xmath352 = \alpha_{i}(\tau_{0}+1 - 2(m - l)-\bar{\delta}_{i})+\psi_{i}$ ] for @xmath353 . with such a mechanism , the @xmath354 matrix in equation ( [ eq : simple - obs ] ) would change to @xmath355^t\ ] ] and @xmath309 becomes @xmath296 = \left [ \begin{array}{c } \alpha_{i}(\tau_{0}+1 - 2m-\bar{\delta}_{i})\\ \alpha_{i } \end{array } \right].\ ] ] to estimate the location @xmath32 in the time scale of @xmath40 , we can proceed as in section [ sec : timesyncestimator ] : @xmath356 will be distributed @xmath357 and @xmath312 is umvu . this leads to the umvu estimate @xmath358 , where @xmath359 $ ] , and @xmath360 . in this case , the variance of @xmath321 will be @xmath361 , and thus we have that @xmath362 converted to the time scale of @xmath43 we have @xmath363 comparing equations ( [ eq : comptonosimult ] ) and ( [ eq : nosimult ] ) , we see that they have the same form . as a result , we can again set @xmath342 and achieve cooperative time synchronization . we now extend the ideas of cooperative time synchronization to the situation where signals suffer not only from pathloss but also propagation delay . it turns out that the effect of propagation delay can also be addressed using the concept we have been using throughout this paper averaging out errors using the large number of nodes in the network . in this section , we use the pathloss and propagation delay model detailed in section [ sec : delay - model ] . we introduce a time delay function @xmath83 . for generality , we explicitly model a multi - hop network where we have a @xmath53 function that is zero for @xmath56 greater than some distance @xmath95 , i.e. @xmath364 for @xmath365 . such a model implies that the aggregate signal seen at any node @xmath29 is influenced only by the set of nodes inside a circle of radius @xmath95 centered at node @xmath29 . with this we can effectively divide the network into two disjoint sets , a set of _ interior nodes _ and a set of _ boundary nodes_. an interior node @xmath29 is defined to be a node whose distance from the nearest network boundary is greater than or equal to @xmath95 . a boundary node is thus defined to be a node that is a distance less than @xmath95 away from the nearest network boundary . we make this distinction since the synchronization technique for each set of nodes is different . please note that if a pathloss function where @xmath364 for @xmath365 is unreasonable , then we simply choose @xmath95 to be infinite and consider all nodes in the network to be boundary nodes . using the propagation delay model , @xmath116 will obviously modify the general received aggregate waveform seen at any node @xmath29 . in fact , equation ( [ eq : timesync - aggwaveform ] ) will now be written as @xmath366 for @xmath9 large , this model will give an accurate characterization of the aggregate waveform seen at node @xmath29 . from equation ( [ eq : timesync - aggwaveform - withdelay ] ) , it is clear that the aggregate waveform will not have a zero - crossing at @xmath32 for every node @xmath29 because of the presence of the @xmath116 random variables . therefore , to average out propagation delay , the idea we employ is to have each node introduce a _ random _ artificial time shift that counteracts the effect of the time delay random variable . more precisely , we want to introduce another random variable @xmath367 such that @xmath368 will have zero mean and a symmetric distribution . at the same time , we assume each node knows @xmath97 and @xmath99 and will also introduce an artificial scaling factor @xmath369 to simplify the analysis of the aggregate waveform . this means that instead of using the scaling factor @xmath370 , each node @xmath18 will scale its transmitted pulse by @xmath371 . for the motivation in this section , let us assume that node @xmath29 is an interior node . to find the distribution of @xmath367 , we consider the following . @xmath82 has density @xmath115 and let @xmath372 be the density of @xmath367 . since @xmath82 and @xmath367 are independent , we know that the density of @xmath373 , @xmath374 , will be the convolution of @xmath115 and @xmath372 . therefore , by the properties of the convolution function , if we set @xmath375 , then we have that @xmath374 is symmetric , i.e. @xmath376 . as well , since @xmath82 has finite expectation , it is easy to see that @xmath377 . given a sequence of @xmath271 zero - crossings that we know to be occurring at integers of @xmath78 , we can still use @xmath336 ( from ( [ eq : timesync - estimator ] ) in the time scale of node @xmath42 ) as the time synchronization estimator . however , with propagation delay , the pulse - connection function will now be @xmath378 . with @xmath367 and @xmath379 included , we can rewrite equation ( [ eq : timesync - aggwaveform - withdelay ] ) as @xmath380 it is important to see that since @xmath82 has the same distribution for _ all _ interior nodes @xmath29 , equation ( [ eq : timesync - aggwaveform - withdelayandfix ] ) holds for every node @xmath29 that is an interior node . this means that for the network to cooperatively generate the waveform in ( [ eq : timesync - aggwaveform - withdelayandfix ] ) each transmit node @xmath18 needs to have the following additional knowledge : ( 1 ) the distribution of @xmath367 whose density is @xmath375 , where @xmath29 is an interior node , and ( 2 ) the functions @xmath97 and @xmath99 to generate @xmath379 . with this knowledge , we can use equation ( [ eq : timesync - aggwaveform - withdelayandfix ] ) to study the aggregate waveform seen at any interior node @xmath29 . in fact , we find that the aggregate waveform has limiting properties that are similar to those outlined in theorem [ theorem : main ] . these properties are described in theorem [ theorem : main - delay ] . [ theorem : main - delay ] let @xmath28 be as defined in equation ( [ eq : poft ] ) and @xmath195 with @xmath196 a constant and @xmath197 for all @xmath18 , @xmath141 a constant . @xmath34 and @xmath116 are defined as in section [ sec : delay - model ] and @xmath367 with density @xmath375 is independent from @xmath116 . @xmath369 and let @xmath116 , @xmath367 , and @xmath33 be mutually independent for all @xmath18 . then , for any interior node @xmath29 with @xmath381 as defined in ( [ eq : timesync - aggwaveform - withdelayandfix ] ) , @xmath382 has the properties * @xmath383 , * @xmath384 is odd around @xmath189 , i.e. @xmath385 for @xmath208 . @xmath210 the proof of theorem [ theorem : main - delay ] is left for the appendix . from the arguments so far , it seems that time synchronization with delay , at least for interior nodes , can be solved simply by modifying the pulse - connection function @xmath386 and changing the scaling factor to @xmath371 . theorem [ theorem : main - delay ] tells us that the limiting aggregate waveform makes a zero - crossing at the next integer value of @xmath78 and the waveform is odd . thus , we can use this zero - crossing as a synchronization event and maintain synchronization in a manner identical to the technique used in the situation without propagation delay . this , however , unfortunately is not the case . in order to implement the above concept , we need to find the random variable , @xmath387 , in the time scale of @xmath40 , that corresponds to @xmath367 such that @xmath388 this means that we need @xmath389 . however , each node @xmath18 can not find @xmath387 that satisfies this since it does not know its @xmath51 . since the @xmath18th node does not know its own value of @xmath51 , to do time synchronization with propagation delay we can have each node estimate its @xmath51 value . however , this estimate will not be perfect and we may no longer have the symmetric limiting aggregate waveform described by theorem [ theorem : main - delay ] . this means that the center zero - crossing might occur some @xmath140 away from @xmath32 , @xmath32 an integer value of @xmath78 . however , steady - state time synchronization can be maintained if the network can use a sequence of @xmath271 equispaced zero - crossings that occur at @xmath390 , where @xmath32 is an integer value of @xmath78 , to cooperatively generate a limiting aggregate waveform that has a zero - crossing at @xmath391 . in such a situation , the network will be able to construct a sequence of equispaced zero - crossings and maintain the occurrence of these zero - crossings indefinitely . the idea is the same as in the case without propagation delay , but the only difference here would be that the zero - crossings do not occur at integer values of @xmath78 . let us give a more formal description of this idea . using notation from section [ sec : timesyncestimator ] , we start with the assumption that each interior node @xmath18 has a sequence of @xmath271 observations that has the form @xmath392 where @xmath393 and @xmath140 is known . to develop the time synchronization estimator @xmath394 and the pulse - connection function @xmath343 , we consider the observations made by each node . if we assume that each node knows the value of @xmath140 , the vector of observations can be written as in ( [ eq : simple - obs ] ) @xmath395 where the matrix @xmath396 in this case is @xmath397^t.\ ] ] using this model , we can follow the development in section [ sec : timesyncestimator ] to find the the time synchronization estimator @xmath398 where @xmath399 $ ] . this estimator will give each node the ability to optimally estimate the next integer value of @xmath78 . note that the variance of the time synchronization estimator is @xmath400 using the time synchronization estimator , we can choose the pulse - connection function as @xmath401 where each time node @xmath18 makes the estimate @xmath402 it also estimates @xmath403 as @xmath404 @xmath405 $ ] . we find that @xmath406 . since , from section [ sec : propdelay_motivation ] , we know we want @xmath389 , we have set @xmath407 . notice that since @xmath387 is simply a realization of @xmath367 multiplied by node @xmath18 s estimate of @xmath51 , node @xmath18 can use the realization of @xmath367 and find @xmath369 . with our choice of @xmath343 in ( [ eq : pulse - connection - propdelay ] ) , we see that @xmath408 where @xmath409 , and @xmath410 . because of the random factor @xmath411 , we see that @xmath412 is no longer a symmetric distribution . as a result , the limiting aggregate waveform @xmath413 may not have a zero - crossing at @xmath189 . thus , if we can find an @xmath140 such that each node @xmath18 using a set of observations of the form ( [ eq : observation_form ] ) allows the network to cooperatively generate the waveform in ( [ eq : aggregate - waveform - timedelay ] ) that has its zero - crossing occurring at @xmath414 ( in the time scale of @xmath43 ) , then we have steady - state time synchronization . this is because the network would be able to use a sequence of @xmath271 observations to generate the next observation that gives the same information as any of the previous observations . thus , by always taking the @xmath271 most recent observations , the process can continue forever and maintain synchronization . each node @xmath18 would need to know distribution of @xmath367 , the value of @xmath140 , and the functions @xmath97 and @xmath99 . therefore , we find that steady - state time synchronization of the interior nodes is possible under certain conditions . as a note , no interior node needs to know any location information . before we consider the synchronization of boundary nodes , we note that the key requirement for each boundary node @xmath18 is to have a pulse - connection function given in equation ( [ eq : pulse - connection - propdelay ] ) . the reason that this must be the pulse - connection for every boundary node @xmath18 is because the analysis for the interior nodes assumes that the aggregate waveform seen by any interior node @xmath29 is created by pulse transmissions occurring at a time determined by ( [ eq : pulse - connection - propdelay ] ) . since the aggregate waveform seen by some interior nodes are created by pulse transmissions from boundary nodes , each boundary node must have the appropriate pulse - connection function . this requirement , however , proves to be extremely problematic and reveals a limitation of the elegant technique of averaging out timing delay when we come to boundaries of the network . the problem comes because @xmath415 already does not have a symmetric distribution if @xmath29 is a boundary node . recall that @xmath416 when @xmath29 is an interior node and @xmath417 when @xmath29 and @xmath77 are both interior nodes . however , @xmath418 when @xmath29 is an interior node and @xmath77 is a boundary node . as a result , @xmath415 is no longer symmetric if @xmath29 is a boundary node . in fact , it is clear that the distribution of @xmath415 is a function of node @xmath29 s location near the boundary . because of this additional asymmetry , let us assume for a moment that the sequence of zero - crossings observed by boundary node @xmath18 occur @xmath419 away from an integer value of @xmath78 . that is , if every node in the network , including the boundary nodes , transmitted a sequence of pulses where each pulse was sent according to ( [ eq : pulse - connection - propdelay ] ) , then boundary node @xmath18 would observe the sequence of observations @xmath420 where @xmath393 and @xmath419 is known . this boundary node @xmath18 could then use the time synchronization estimator given by ( [ eq : timesync - estimator - propdelay ] ) but where the matrix @xmath396 is now replaced with @xmath421 @xmath422^t.\ ] ] thus , for this boundary node @xmath18 we have @xmath423 in this case , however , the variance of the time synchronization estimator depends on @xmath419 @xmath424 the fact that the variance depends on @xmath419 is the root of the problem . the pulse - connection function @xmath425 is _ not _ the same as that given by ( [ eq : pulse - connection - propdelay ] ) . to correct for this , we can make the strong assumption that each boundary node @xmath18 knows is own @xmath51 . we address the reasoning behind this assumption in section [ sec : propdelay_assumption ] . if we use this assumption , then each boundary node @xmath18 can get an observation sequence of the form ( [ eq : observation_form ] ) simply by adding @xmath426 to each of the @xmath271 observations of the form given in ( [ eq : observation_form_boundary ] ) , where we assume that node @xmath18 knows both @xmath140 and @xmath419 . with such an observation sequence , boundary node @xmath18 will have the time synchronization estimator ( [ eq : timesync - estimator - propdelay ] ) and , more importantly , the pulse - connection function ( [ eq : pulse - connection - propdelay ] ) . thus , maintaining time synchronization for the case of propagation delay would be possible . what we have then is that boundary node synchronization would require only the boundary nodes to know their @xmath51 parameters . with this strong assumption only for the boundary nodes , the network is effectively synchronized . even though the boundary nodes do not see the same zero - crossing as the interior nodes , they can calculate this time and thus have all the required synchronization information . the assumption that each boundary node @xmath18 knows @xmath51 is a strong assumption . even though the fraction of nodes that are boundary nodes is small for multi - hop networks requiring many hops to send information across the network , we believe that the assumption is still very artificial . there are two reasons that we make the assumption for the presentation of results on time synchronization with propagation delay . first , the assumption allows us to give an elegant presentation of the main concept of this paper which is to use high node density to average out errors in the network . throughout this work we have used high node density to average out inherent errors present in the nodes . we were able to average out random timing jitter that is present in each node and provide the network with a sequence of zero - crossings that can serve as synchronization events . we then applied this technique to averaging out the errors introduced by time delay . to this end we were partially successful in that the interior nodes can average out these errors assuming the boundary nodes have additional information . but this is of interest since the goal of this paper is to understand the theory of spatial averaging for synchronization and discover its fundamental advantages and limitations . second , the problem encountered at the boundaries is one that opens up an entirely new area of study which is the target of our future work . the issue that we encounter is that the waveform seen by some nodes in the network will have a zero - crossing that is shifted from the ideal location . this implies that different nodes will observe different zero - crossings . furthermore , these zero - crossings will now evolve in time since we do not have the same observations over the entire network . this problem is similar to what we encounter if we consider finite sized networks . for finite @xmath9 , the zero - crossing location will be random and thus introduce another source of error . as well , different nodes will see different zero - crossing locations . therefore , we will turn our attention to the case of finite @xmath9 and develop a different set of tools that will be needed to understand what types of synchronization are achievable under the situation where zero - crossing locations evolve in time . using this understanding , we hope to return to the issue of propagation delay in asymptotically dense networks and characterize the behavior of the network . to conclude , we revisit the scalability issue under the light of work developed in this paper . in the introduction ( section [ sec : estimate - params ] ) , we mentioned that most existing proposals for time synchronization suffer from an inherent scalability problem . the problem with those existing proposals lies in the fact that synchronization errors accumulate : if node 2 can synchronize to node 1 with some small error , and node 3 can synchronize to node 2 with the same small error , these errors accumulate , and the synchronization of node 3 to node 1 is worse . therefore , synchronization error increases with the number of hops in the network , and this problem is especially apparent in the regime of high densities . to make these ideas precise , we first determine the maximum number of hops over which synchronization information must travel and then study how the error in a generic pairwise synchronization mechanism depends on this number of hops . to obtain an estimate for the maximum number of hops @xmath427 in a network in the regime of high densities ( fixed area , @xmath37 ) , we approximate the transmission range of a node by the minimum required transmission distance , @xmath428 , to maintain a fully connected network with high probability . from @xcite , we have that for @xmath9 nodes uniformly distributed over a @xmath429\times [ 0,1]$ ] square , the graph is connected with probability-1 as @xmath37 if and only if each node s transmission distance @xmath428 is such that @xmath430 for some @xmath431 . let us , therefore , approximate @xmath428 as @xmath432 thus , @xmath433 , and thus @xmath434 as @xmath37 . now , we assume there are @xmath435 nodes arranged in a linear ordering , numbered @xmath42 to @xmath435 . to synchronize , each node @xmath18 forms an estimate of its own @xmath51 , based on @xmath271 pulses transmitted from node @xmath436 . as before , node @xmath42 will have the reference clock @xmath44 . node @xmath42 starts by sending @xmath271 pulses at times @xmath437 for @xmath393 . as a result , node @xmath0 will get a vector of observations @xmath438 , where @xmath439 = \alpha_{2}(\tau_{1}-\bar{\delta}_{2})+\psi_{2}$ ] and the @xmath440th element of @xmath438 is @xmath441=\alpha_{2}(\tau_{1}-\bar{\delta}_{2})+l\alpha_{2}+\psi_{2}$ ] . this is similar to the situation we had in ( [ eq : simple - obs ] ) and we can therefore estimate @xmath442 using @xmath443 where @xmath444 $ ] . we find that @xmath445 . node @xmath0 will now transmit @xmath271 pulses at times , in terms of @xmath446 , @xmath447 , for @xmath393 . note that @xmath448 is now a fixed value since node @xmath0 has estimated @xmath442 . in terms of @xmath43 , these pulses occur at @xmath449 for @xmath393 , where @xmath450 . thus , if we translate these times into the time scale of @xmath451 , we will have the vector of observations , @xmath452 , made by node @xmath453 . we find that the @xmath440th element of @xmath452 is @xmath454 = \alpha_{3}((\tau_{2 } + l\frac{\hat{\alpha}_{2}}{\alpha_{2}}- \frac{\psi_{2}}{\alpha_{2}})-\bar{\delta}_{3})+\psi_{3 } \sim { \mathcal n}\bigg(\alpha_{3}(\tau_{2}-\bar{\delta}_{3})+l\alpha_{3}\frac{\hat{\alpha}_{2}}{\alpha_{2 } } , \sigma^{2}\big(\frac{\alpha_{3}^{2}}{\alpha_{2}^{2}}+1\big)\bigg).\ ] ] this vector of observations is of the form @xmath455 where @xmath456 = \left [ \begin{array}{c } \alpha_{3}(\tau_{2}-\bar{\delta}_{3})\\ \alpha_{3}\frac{\hat{\alpha}_{2}}{\alpha_{2 } } \end{array } \right]\ ] ] with @xmath297^t\ ] ] and @xmath298^{t}$ ] . @xmath457 with @xmath458 . with this vector of observations , we can use the estimator @xmath459 where @xmath444 $ ] . we find that @xmath460 if we continue this reasoning , we find that @xmath461 will be the estimate of node @xmath435 . from the above analysis , we see that each node @xmath18 s estimate suffers from jitter variance of the same form . however , there is an accumulation of error because node @xmath18 s estimate has a mean that is dependent on node @xmath436 s estimate . as a result , if node @xmath436 has some small error , then that error will propagate to the estimate of node @xmath18 . a good way to see this is if we consider the special case where @xmath462 . this is the case where the clock frequencies are the same , but nodes do not know this . in this case , we find that node @xmath435 s estimate can be written as @xmath463 where @xmath464 . this is intuitively obvious because node @xmath18 s estimate @xmath403 will be the mean of the gaussian random variable @xmath465 . therefore , it is obvious that the error variance grows linearly with the number of hops . in fact , this behavior is observed in experimental work . with reference broadcast synchronization ( rbs ) , from @xcite the authors find that the synchronization error variance of an @xmath435 hop path is approximately @xmath466 , where @xmath165 is the one hop error variance . therefore , we have that the synchronization error between our two nodes will grow linearly as @xmath467 , which is strictly monotonically increasing . as a result , as @xmath37 , we have that synchronization error will grow unbounded . this scalability problem , however , can potentially be avoided using cooperative time synchronization as @xmath37 . this is because in the limit of infinite density , the cooperative time synchronization technique allows every node in the network to see a set of identical equispaced zero - crossings . as a result , in steady - state the synchronization error does not grow across the network . this comes about by using the high node density to average out random timing errors . thus , we find that cooperative time synchronization has very favorable scalability properties in the limit as @xmath37 . the cooperative synchronization technique described in this paper provides us deterministic parameters that we can use for time synchronization in the limit as node density grows unbounded . in fact , as the node density grows , the observations that can be used for synchronization improve . this means that our cooperative synchronization technique provides an effective trade - off between network density and synchronization performance . such a trade - off has not existed before and will provide network designers an additional dimension over which to improve network synchronization performance . the fundamental idea behind cooperative time synchronization is that by using spatial averaging , the errors inherent in each node can be averaged out . by using observations that are an average " of the information from a large number of surrounding nodes , synchronization performance can be improved due to the higher quality observations . from this point of view , it is clear that the particular technique described in this paper is but one example of using spatial averaging to improve synchronization . other techniques can also be developed using spatial averaging . for example , nodes may not necessarily have to send odd - shaped pulses and use zero - crossing observations . even though this setup takes advantage of the superposition of pulses , it has its drawbacks . to keep the signals in phase , the jitter variance will limit the maximum frequency at which signals can be sent . instead , nodes may transmit ultra wideband pulses . if the nodes surrounding a particular node @xmath29 each transmit an impulse at their estimate of an integer value of @xmath78 , then due to timing errors in the surrounding nodes , node @xmath29 will see a cluster of pulse arrivals around this integer value of @xmath78 . node @xmath29 can then take the sample mean of this cluster of pulses and use that as an observation , just like we used the zero - crossing as an observation in this paper . this idea is illustrated in fig . [ fig : pulsetraincluster ] . such a technique based on ultra wideband pulses will also provide similar scalability properties . as a result , cooperative time synchronization really describes a class of techniques that can take advantage of spatial averaging to improve synchronization performance . with the goal of developing practical cooperative synchronization mechanisms , two keys areas of interest are cooperative synchronization in finite - sized networks and algorithm development . first , the analysis of performance for finite - sized networks is very important . determining when the asymptotic properties presented in this work are good predictors of performance in networks that may be large but still finite in size is important in terms of bridging the gap between our proposed ideas and practical systems . preliminary , simulation - based work along these lines can be found in @xcite . second , developing practical techniques for cooperative time synchronization is essential for implementing spatial averaging in real networks . along these lines , one area of interest is determining what types of pulses should be used , i.e. odd - shaped pulses or ultra wideband pulses . furthermore , the ideas in this paper suggest a few other areas of interest for future work . one is the issue of distributed modulation methods . if we have the ability to generate an aggregate waveform with equispaced zero - crossings , by controlling the location of these crossings we can modulate information onto this waveform and use it to communicate with a far receiver . preliminary work along these lines can be found in @xcite . another issue is to study how the idea of spatial averaging that is so prevalent in this work contributes to synchronization that is observed in nature . * proof of lemma [ lemma : polarity_positive ] . * to show ( [ eqn : cond3a ] ) , we consider @xmath468 since @xmath223 , we have that @xmath469 implying that @xmath249 is shifted to the left and the zero - crossing of @xmath249 occurs at a negative value . @xmath249 is odd about its zero - crossing and @xmath217 is symmetric about zero and strictly monotonically increasing on @xmath470 $ ] for all positive finite variance values . thus , it is clear that @xmath471 which makes @xmath472 . now , the expectation will vary with the variance of @xmath33 and the variance will range from a positive upper bound of @xmath473 to a positive lower bound of @xmath474 , where recall that @xmath211 is a value determined by our choice of the pulse connection function . if we consider @xmath475 to be a function of the variance of @xmath33 , then we see that it is bounded and continuous on the compact domain @xmath476 $ ] . since we showed in the previous paragraph that @xmath472 whenever @xmath33 has a nonzero finite variance , clearly @xmath472 when @xmath477 $ ] . thus , it is clear that @xmath228 and @xmath229 exist and ( [ eqn : cond3a ] ) is shown . to show ( [ eqn : cond3b ] ) , we consider @xmath478 where the second to last inequality follows from the fact that @xmath479 is upper bounded by @xmath42 . the last inequality follows since @xmath480 by the fact that @xmath481 . thus , we have shown ( [ eqn : cond3b ] ) . [ theorem : feller ] the convergence of the series @xmath484 implies that the strong law of large numbers will apply to the sequence of independent random variables @xmath215 . that is , again from @xcite , for every pair @xmath485 , @xmath486 , there corresponds an @xmath9 such that @xmath487 for all @xmath488 . @xmath210 we have shown ( [ eqn : cond3b ] ) so we have @xmath489 . thus @xmath490 and we have convergence by the direct comparison test . therefore , we can apply theorem [ theorem : feller ] and get that for any pair @xmath485 , @xmath486 , we can find an @xmath9 such that @xmath491 for all @xmath488 . by ( [ eqn : cond3a ] ) we have that @xmath492 . thus , we can clearly see that @xmath493 furthermore , since we keep the function @xmath153 constant as we increase the number of nodes in the network we get that @xmath494 converges to a constant @xmath495 given by @xmath496 the above expression comes from the fact that since each @xmath497 is a function of @xmath51 , @xmath494 will converge to the average of the @xmath498 over @xmath153 , the function that characterizes the set of @xmath51 s . therefore , given any @xmath140 , we can find an @xmath499 such that @xmath500 for all @xmath501 . note that since @xmath502 , we have that @xmath503 . since @xmath504 using ( [ eqn : slln ] ) and ( [ eq : means - converge ] ) we have @xmath505 for all @xmath488 , where @xmath506 . thus , we have @xmath507 almost surely . this completes the proof of lemma [ lemma : polarity_positive ] . @xmath210 * proof of lemma [ lemma : continuity1 ] . * first , we start by finding an analytical expression for @xmath508 . from the proof of lemma [ lemma : polarity_positive ] we have that @xmath509 therefore , @xmath508 can be written as @xmath510 f_{t}(\psi , s ) f_{\alpha}(s)d\psi ds| \\ & \leq & a_{max } \int_{\alpha_{low}}^{\alpha_{up}}\int_{-\infty}^{\infty } |p(t-\tau_{o}-\psi)-p(t_{o}-\tau_{o}-\psi)|f_{t}(\psi , s ) f_{\alpha}(s)d\psi ds \\ & = & a_{max } \int_{\alpha_{low}}^{\alpha_{up}}\int_{-\tau_{nz}+t_{o}-\tau_{0}-|t - t_{o}|}^{\tau_{nz}+t_{o}-\tau_{0}+|t - t_{o}| } f_{t}(\psi , s)f_{\alpha}(s ) d\psi ds,\end{aligned}\ ] ] where @xmath511 . the change in the limits of integration in the last equality comes from the fact that @xmath512 outside of @xmath513 $ ] . this is the maximum interval over which @xmath514 can be non - zero . there is no need to take the absolute value of @xmath222 and @xmath153 since they are always non - negative . our second step is to bound the inner integral . before doing so , we first show that the inside integral is in fact riemann integrable . for any given @xmath78 and @xmath515 , the inside integral is taken over a closed interval . over a closed interval , we know from strichartz @xcite that any bounded function that is continuous except at a finite number of points is riemann integrable . furthermore , also from @xcite we know that the sums and products of continuous functions are continuous . as well , if a function is continuous then the absolute value of that function is also continuous . @xmath28 has at most @xmath516 locations at which it is discontinuous and over any open interval not containing a discontinuity , @xmath28 is uniformly continuous since @xmath181 is uniformly continuous . @xmath222 has @xmath517 discontinuities in @xmath267 for an given @xmath220 since it is gaussian for any @xmath220 . and since @xmath518 $ ] , @xmath519 for all @xmath267 and @xmath220 ( @xmath520 occurring when @xmath521 and @xmath522 ) . thus , since @xmath28 and @xmath222 are continuous except at a finite number of points , we see that for given @xmath220 , @xmath78 , and @xmath162 @xmath523 is also continuous in @xmath267 except at a finite number of points ( at most @xmath524 points ) . this function is also bounded since the product of two bounded functions is bounded . as a result , we see that the integral is riemann integrable over any closed interval . we now proceed to bound from above the value of this integral by first bounding the maximum value of the integral assuming no discontinuities and then introducing another term that bounds the maximum area contributed by the discontinuities . if we ignore the discontinuities and assume @xmath28 is uniformly continuous , for any @xmath525 there exists a @xmath526 such that @xmath527 for all @xmath78 and @xmath515 . as a result , @xmath514 can be made as small as desired by choosing the proper @xmath120 thus giving us @xmath528 for all @xmath267 for an appropriate choice of @xmath120 . furthermore , we note that @xmath529 because @xmath253 and @xmath519 . the maximum possible jump at a discontinuity in the function @xmath530 is thus @xmath531 and for any @xmath532 , the maximum area contributed by each discontinuity is @xmath533 . as a result , for all @xmath524 discontinuities , the maximum area contribution will be no more than @xmath534 . for the third term we want @xmath546 this gives us @xmath547 since the only requirement is @xmath536 for @xmath120 chosen by any given @xmath525 , we can always choose @xmath532 as small as desired . thus , this condition can be satisfied . with the second term we want the condition @xmath548 which means that @xmath549 again , this condition can be satisfied since we can choose @xmath550 as large as we want and @xmath532 as small as we want as long as @xmath536 for a given @xmath550 . thus , for any @xmath542 , we first choose @xmath551 . then , we find an @xmath552 such that @xmath553 implies that @xmath554 for all @xmath78 and @xmath515 if we remove the discontinuities in @xmath28 . then , if necessary , @xmath555 is increased to @xmath120 so that @xmath536 implies that @xmath556 and @xmath557 . if no increase is necessary , then @xmath558 . with this choice of @xmath526 , @xmath559 . as a result , for any @xmath271 , we can find an @xmath120 such that @xmath536 implies that @xmath560 . thus , @xmath206 is continuous . * proof of theorem [ theorem : main - delay ] . * let us start by writing ( [ eq : timesync - aggwaveform - withdelayandfix ] ) as @xmath561 where @xmath562 . recall that the dependence on @xmath220 comes from the fact that the density of @xmath33 is a function of @xmath51 which is characterized by @xmath153 . this notation is analogous to the notation used in section [ sec : pulseproperties ] . following the steps in the proof of lemma [ lemma : polarity_positive ] , we can quickly show that the limiting aggregate waveform at node @xmath29 will take on the form @xmath563 where @xmath564 with @xmath565 . therefore , we can prove theorem [ theorem : main - delay ] in two steps : * to show that @xmath266 is odd about @xmath32 , we need to show that @xmath566 is odd in @xmath78 about @xmath32 , i.e. @xmath567 for @xmath568 . * to show a zero - crossing at @xmath32 , show that @xmath569 . we first show that @xmath567 for @xmath208 . using the fact that @xmath570 and @xmath571 , we have the following : @xmath572)\big ) \\ & \stackrel{\small{(a)}}{= } & -e\big(a_{max}g(-d_{fix } ) g(d_{j , i } ) p(-\xi+[t_{i}+d_{fix}+d_{j , i}])\big ) \\ & = & -a_{max}\int_{-\infty}^{\infty } \int_{-\infty}^{0 } \int_{0}^{\infty}g(-y)g(x)p(-\xi+[\psi+y+x])f_{d_{j}}(x ) f_{d_{fix}}(y ) f_{t}(\psi , s)dx dy d\psi \\ & \stackrel{\small{(b)}}{= } & a_{max}\int_{\infty}^{-\infty } \int_{\infty}^{0 } \int_{0}^{-\infty}g(z)g(-u)p(-\xi-[w+z+u])f_{d_{j}}(-u ) f_{d_{fix}}(-z ) f_{t}(-w , s)du dz dw \\ & \stackrel{\small{(c)}}{= } & -a_{max}\int_{-\infty}^{\infty } \int_{-\infty}^{0 } \int_{0}^{\infty}g(-u)g(z)p(-\xi-[w+u+z])f_{d_{j}}(z ) f_{d_{fix}}(u ) f_{t}(w , s)dz du dw \\ & = & -e\big(a_{max}g(-d_{fix } ) g(d_{j , i } ) p(-\xi-[t_{i}+d_{fix}+d_{j , i}])\big ) \\ & = & -e(\tilde{m}_{i}(\tau_{0}-\xi , s)),\end{aligned}\ ] ] where @xmath573 follows because @xmath574 and at @xmath575 we did a change of variables with @xmath576 , @xmath577 , and @xmath578 . @xmath579 follows from @xmath580 and @xmath581 . we thus have @xmath567 for @xmath568 . an - swol hu was born in new york state and grew up in california . he received his b.s . in electrical engineering from stanford university in 2002 . currently he is a ph.d . candidate in the school of electrical and computer engineering at cornell university . his research interests include applied statistics and statistical signal processing , with applications to sensor networks . sergio d. servetto was born in argentina , on january 18 , 1968 . he received a licenciatura en informatica from universidad nacional de la plata ( unlp , argentina ) in 1992 , and the m.sc . degree in electrical engineering and the ph.d . degree in computer science from the university of illinois at urbana - champaign ( uiuc ) , in 1996 and 1999 . between 1999 and 2001 , he worked at the ecole polytechnique federale de lausanne ( epfl ) , lausanne , switzerland . since fall 2001 , he has been an assistant professor in the school of electrical and computer engineering at cornell university , and a member of the field of applied mathematics . he was the recipient of the 1998 ray ozzie fellowship , given to `` outstanding graduate students in computer science , '' and of the 1999 david j. kuck outstanding thesis award , for the best doctoral dissertation of the year , both from the dept . of computer science at uiuc . he is also the recipient of a 2003 nsf career award . his research interests are centered around information theoretic aspects of networked systems , with a current emphasis on problems that arise in the context of large - scale sensor networks . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: magnetic fields are ubiquitous in diverse astrophysical settings on various scales ranging from proto - stellar discs , young stellar objects ( ysos ) , microquasars , quasars , galaxies and grandiose astrophysical jets to clusters of galaxies . in some cases , magnetic fields are relatively weak so that dynamical processes are almost unaffected by their presence . however , there do exist numerous cases where magnetic fields are necessary and important for both dynamics and diagnostics , especially in spiral galaxies and accretion disc systems ( e.g. , sofue et al . 1986 ; beck et al . 1996 ; balbus & hawley 1998 ; tagger & pellat 1999 ; widrow 2002 ) . in general , it is challenging to model magnetic fields realistically because of their complexities ( sofue et al . 1986 ; kronberg 1994 ; caunt & tagger 2001 ; widrow 2002 ) . in the problems involving protostar formation and disc galaxies , shu & li ( 1997 ) presumed the so - called ` isopedic ' magnetic field configuration where the mass - to - magnetic flux ratio in a razor thin disc remains constant . using a singular isothermal disc ( sid ) model ( mestel 1963 ) that is isopedically magnetized , shu et al . ( 2000 ) studied stationary coplanar perturbations and proposed that these global modes are bifurcations to either secularly or dynamically unstable configurations . along a separate yet complementary line , lou ( 2002 ) studied an azimuthally magnetized singular isothermal disc ( msid ) and derived two different global stationary magnetohydrodynamic ( mhd ) perturbation modes . based on the msid model , lou ( 2002 ) explored the manifestation of interlaced optical and magnetic field spiral arms in the outer portion of a disc with a nearly flat rotation curve such as the case of the nearby spiral galaxy ngc 6946 ( beck & hoernes 1996 ; fan & lou 1996 ; lou & fan 1998a , 2002 ; frick et al . 2000 , 2001 ; lou et al . 2002 ) . in contexts of spiral galaxies , it is natural and important to consider a composite disc system consisting of gravitationally coupled gaseous and stellar discs . this is because various physical processes on different scales occur in the gaseous disc yet large - scale gas dynamics and environment are significantly affected by large - scale structures in the stellar disc ( lou & fan 1998b ) . to construct global coplanar perturbation structures in a systematic manner , we started with a composite sid system of two fluid discs without magnetic field ( lou & shen 2003 ) to generalize the work of shu et al . ( 2000 ) for a single sid . in terms of the global axisymmetric stability for such a composite sid system , we provided a more straightforward @xmath12criterion ( shen & lou 2003 ) in contrast to the axisymmetric stability criteria of elmegreen ( 1995 ) and jog ( 1996 ) . to be more general than a background sid profile , we recently constructed global stationary perturbation patterns in a composite system of two fluid scale - free discs without magnetic field ( shen & lou 2004a , b ) to further generalize the work of syer & tremaine ( 1996 ) . with proper adaptations , these global stationary perturbation solutions may be utilized to model large - scale structures spiral galaxies . while a scale - free disc has no characteristic scales , one may impose boundary conditions to describe a modified scale - free disc , e.g. , by cutting a central hole in a disc ( e.g. , zang 1976 ; evans & read 1998a , b ) . with such boundary conditions , discrete eigenfunctions for perturbations may be constructed as growing normal modes . on the other hand , by specifying a phase relation for a postulated reflection of spiral waves from the origin @xmath13 , goodman & evans ( 1999 ) could also define discrete normal modes even for an unmodified gaseous sid . shu et al . ( 2000 ) speculated that the swing amplification process ( e.g. , goldreich & lynden - bell 1965 ; julian & toomre 1966 ; toomre 1977 ; fan & lou 1997 ) across corotation allows a continuum of normal modes ( lynden - bell & lemos 1993 ) while proper ` boundary conditions ' may select from this continuum a discrete spectrum of unstable normal modes ; they also suggested that zero - frequency ( stationary ) disturbances do signal the onset of instability . on the basis of the work of shu et al . ( 2000 ) for a single isopedically magnetized sid and of lou & shen ( 2003 ) for a composite sid system without magnetic field , lou & wu ( 2004 ) analyzed coplanar mhd perturbation structures in a composite sid system with one of the disc being isopedically magnetized ( shu & li 1997 ) . as x - ray emitting hot gases can freely flow along open magnetic field lines anchored at the gaseous disc , global stationary solutions constructed by lou & wu ( 2004 ) set an important preliminary stage for modelling spiral galactic mhd winds . in particular , if activities of star formation , starbursts and supernovae etc . are mainly responsible for producing x - ray emitting hot gases streaming out of the galactic plane from both sides , we would expect stronger winds from the circumnuclear starburst region ( e.g. lou et al . 2001 ) and along spiral arms than those winds from the remainder of the galactic disc plane . both sid and msid models offer valuable physical insight for disc dynamics and belong to a wider family of scale - free discs ( syer & tremaine 1996 ) where physical variables in the axisymmetric background equilibrium scale as powers of radius @xmath14 ( e.g. , the rotation curve @xmath15 ) . it turns out that coplanar perturbations in such scale - free discs in a proper range of @xmath8 can be treated globally and analytically ( lemos et al . 1991 ; syer & tremaine 1996 ) without invoking the usual wkbj or tight - winding approximation to solve the poisson equation for density wave perturbations ( lin & shu 1964 ; binney & tremaine 1987 ; bertin & lin 1996 ) . in other words , it becomes feasible to include azimuthal magnetic fields in a scale - free disc and to construct global coplanar mhd perturbation configurations that are stationary in an inertial frame of reference . such analytical solutions , still idealized and limited , are making important further steps and are extremely valuable for bench - marking numerical codes and for initializing numerical mhd simulations . for an azimuthally magnetized scale - free disc with an infinitesimal thickness , the axisymmetric background equilibrium without radial flow is characterized by following features : a surface mass density @xmath0 , a rotation curve @xmath2 , an azimuthal magnetic field @xmath1 and a barotropic equation of state @xmath16 where @xmath5 is the vertically integrated gas pressure with constant @xmath17 for a warm disc and @xmath18 as the barotropic index . to maintain a radial force balance at all radii , there exists an explicit relationship among @xmath7 , @xmath8 , @xmath9 and @xmath6 such that only one parameter is independent . as will be seen in section 2 , it is convenient to use @xmath8 as this independent parameter ( syer & tremaine 1996 ; shen & lou 2004b ) . in order to satisfy relevant constraints , the prescribed range of @xmath8 turns out to be @xmath10 that includes the previous special case of @xmath19 ( i.e. , the msid case ) studied by lou ( 2002 ) , lou & zou ( 2004a ) and lou & wu ( 2004 ) . in section 2 , we introduce coplanar mhd perturbations and present the linearized perturbation equations . we construct and analyze global stationary aligned and logarithmic spiral perturbation configurations in section 3 and as an example of illustration , we discuss specifically the case of @xmath11 with the background azimuthal magnetic field being force - free . in section 4 , we derive phase relationships between perturbation enhancements of surface mass density and magnetic field for both aligned and logarithmic spiral cases , and discuss the problem of angular momentum conservation . finally , we summarize our results in section 5 . for the convenience of reference , relevant technical details are collected in the appendices . in the magnetofluid approximation , the disc is taken to be razor - thin ( i.e. , we use vertically integrated magnetofluid equations and neglect vertical derivatives of physical variables ) and large - scale stationary aligned and spiral coplanar disturbances develop in a background mhd rotational equilibrium of axisymmetry ( syer & tremaine 1996 ; lou 2002 ; lou & zou 2004a ; lou & wu 2004 ) . the background magnetic field is taken to be azimuthal to avoid the magnetic field winding dilemma ( lou & fan 1998a ) . for the sake of simplicity at this stage , non - ideal effects such as viscosity , resistivity and thermal diffusion etc . are ignored for large - scale perturbations . using cylindrical coordinates @xmath20 , we have the following two - dimensional nonlinear mhd equations for a razor - thin disc geometry : the mass conservation equation @xmath21 where @xmath22 is the surface mass density , @xmath23 is the radial bulk flow velocity , @xmath24 is the specific angular momentum in the vertical @xmath25-direction and @xmath26 is the azimuthal linear velocity ; the radial component of the momentum equation @xmath27\ , \end{split}\ ] ] where @xmath5 is the vertically integrated gas pressure ( sometimes referred to as two - dimensional pressure ) , @xmath28 is the total gravitational potential including that of an axisymmetric dark matter halo in contexts of disc galaxies , and @xmath29 and @xmath30 are the azimuthal and radial components of the coplanar magnetic field @xmath31 , respectively ; the azimuthal component of the momentum equation @xmath32\ ; \end{split}\ ] ] the poisson integral equation @xmath33^{1/2}}\ , \ ] ] where @xmath34 is defined as the ratio of the potential arising from the disc to that arising from the entire system including a massive dark matter halo that is presumed not to respond to coplanar perturbations in the disc plane ( syer & tremaine 1996 ; shu & li 1997 ; shu et al . 2000 ; lou & shen 2003 ; shen & lou 2004a , b ; lou & zou 2004a ) with @xmath35 for a full disc and @xmath36 for partial discs ; the divergence - free condition of magnetic field @xmath37 @xmath38 the radial component of the magnetic induction equation @xmath39 the azimuthal component of the magnetic field induction equation @xmath40 and the barotropic equation of state @xmath41 where constants @xmath17 and @xmath18 . among the three equations ( [ divergfree ] ) , ( [ radiaminduct ] ) and ( [ azimuminduct ] ) , we can freely choose two independent ones . by barotropic equation of state ( [ barotropic ] ) , the sound speed @xmath42 is defined by @xmath43 where subscript @xmath44 indicates the background equilibrium . an isothermal sound speed corresponds to a barotropic index @xmath45 . we now proceed to derive properties of an axisymmetric rotational mhd equilibrium characterized by physical variables associated with a subscript @xmath44 . in our notations , such a background equilibrium has the following power - law scalings : surface mass density of @xmath0 , a radial velocity profile @xmath46 and @xmath47 , a purely azimuthal magnetic field with @xmath48 and @xmath49 . substitutions of these power - law scalings into equations @xmath50 lead to the following condition for the radial force balance , namely @xmath51 where the alfvn speed @xmath52 in a thin disc is defined by @xmath53 and the poisson integral gives @xmath54 with the numerical factor @xmath55 explicitly defined by @xmath56 and @xmath57 being the standard gamma function . this expression ( [ p0 ] ) can also be included in a more general form of @xmath58 as in expression ( [ pm ] ) derived later with the limiting result of @xmath59 as @xmath60 . to satisfy the radial force balance equation ( [ sfeqn ] ) at all radii ( i.e. , scale - free ) would necessarily require ) in general and for the special case of @xmath11 , we have @xmath61 for the equilibrium magnetic field to be force - free ( e.g. , low & lou 1990 ) . the non - force - free msid case studied by lou ( 2002 ) has @xmath62 , @xmath19 , @xmath63 and @xmath45 . ] @xmath64 which immediately leads to the explicit expressions of indices @xmath7 , @xmath9 and @xmath6 all in terms of @xmath8 @xmath65 in this formulation , @xmath66 remains finite only if @xmath67 which implies @xmath68 . for the total gravity force arising from the equilibrium surface mass densities to be finite , a larger @xmath69range @xmath70 is allowed . the @xmath69range is also constrained by @xmath10 where the first inequality is required for @xmath18 and the second inequality is imposed such that the central point mass will not diverge . by these considerations , the plausible range of @xmath8 falls in @xmath71 ( see syer & tremaine 1996 and shen & lou 2004a , b ) . according to equations ( [ sfeqn ] ) and ( [ pararelation ] ) , we can explicitly write all physical variables of the background equilibrium state in power - law scalings of @xmath14 in terms of parameter @xmath8 , namely @xmath72 } { 2\pi g(2\beta{\cal p}_0)r^{1 + 2\beta}}\ , \ \ \qquad a^2=\frac{a^2}{(1 + 2\beta)r^{2\beta}}\ , \ \ \qquad c_a^2\equiv q^2a^2=\frac{q^2a^2}{(1 + 2\beta)r^{2\beta}}\ , \\ & v_0=\frac{ad}{r^{\beta}}\ , \ \ \qquad j_0\equiv rv_0=\frac{ad}{r^{\beta-1}}\ , \ \ \qquad \omega\equiv\frac{j_0}{r^2}=\frac{ad}{r^{1+\beta}}\ , \ \ \qquad\kappa\equiv [ ( 2\omega / r)d(r^2\omega)/dr]^{1/2}=[2(1-\beta)]^{1/2}\omega\ . \end{split}\ ] ] by introducing a reference radius @xmath73 , the constant @xmath74 is actually related to a scaled sound speed @xmath75 , the constant @xmath76 $ ] is a scaled rotational mach number , the constant @xmath77 is the ratio of the alfvn speed @xmath52 to the sound speed @xmath42 , @xmath78 and @xmath79 are the equilibrium angular speed and the epicyclic frequency , respectively ( lou 2002 ; shen & lou 2004a , b ) . the first expression of equation ( [ propequilib ] ) puts an upper limit on the magnetic field strength , namely , @xmath80 . alternatively , once a magnetic field parameter @xmath81 is chosen , the scaled rotational mach number @xmath82 must meet the following physical requirement @xmath83 for a positive equilibrium surface mass density @xmath84 . note that for @xmath85 , inequality ( [ restrt ] ) is automatically satisfied without actually restricting @xmath81 and the relevant physical requirement is simply @xmath86 . not all stationary solutions of @xmath87 can satisfy condition ( [ restrt ] ) . this then implies limits on allowed magnetic field strength for a known background rotational mhd equilibrium . it is also important to note that for the axisymmetric background rotational mhd equilibrium under consideration , the magnetic force arising from the azimuthal magnetic field is radially inward when @xmath88 , radially outward when @xmath89 and is zero when @xmath11 [ i.e. , the azimuthal magnetic field is force - free ( lou & fan 1998a ) and the equilibrium density distribution reduces to that of the single unmagnetized disc case ( lemos et al . 1991 ; syer & tremaine 1996 ) ] . these different possibilities can be physically understood in terms of the competition between the magnetic pressure and tension forces . in order to see this more specifically for the background , we can write @xmath90 $ ] referring to equation ( [ radialmomeneqn ] ) . the first term in the square brackets stands for the magnetic pressure force and the second term stands for the magnetic tension force for a purely azimuthal magnetic field @xmath91 in the cylindrical geometry . as @xmath92 scales @xmath93 , we have @xmath94 . therefore for @xmath95 or equivalently @xmath88 , the magnetic tension force dominates and the total magnetic force is radially inward ; for @xmath96 or equivalently @xmath89 , the magnetic pressure force dominates and the total magnetic force is radially outward . the two magnetic forces balance each other for @xmath61 ( or @xmath11 ) . on the basis of the full mhd equations @xmath97 , we readily derive the linearized coplanar mhd perturbation equations as @xmath98 @xmath99 -\frac{1}{\sigma_0}\int\frac{dzb_{\theta}}{4\pi r } \frac{\partial(rb_0)}{\partial r}\ , \end{split}\ ] ] @xmath100 @xmath101^{1/2}},\ ] ] @xmath102 @xmath103 @xmath104 where we use subscript @xmath105 to denote associations with small disturbances in physical variables and @xmath106 stands for the coplanar magnetic field perturbation . as the background rotational mhd equilibrium is stationary and axisymmetric , these perturbed physical variables can be decomposed in terms of fourier harmonics with the periodic dependence @xmath107 where @xmath108 is the angular frequency and @xmath109 is an integer for azimuthal variations . more specifically , we write @xmath110 where @xmath111 , @xmath112 , @xmath113 , @xmath114 , @xmath115 and @xmath116 are all radial variations of the corresponding perturbed physical variables and can be complex in general for possible radial oscillations . without loss of generality , we take @xmath117 in our analysis . with fourier decomposition ( [ fourier ] ) , it is then straightforward to cast coplanar mhd perturbation equations @xmath118 into the following forms of @xmath119 @xmath120\ , \end{split}\ ] ] where @xmath121 is a short - hand notation , and @xmath122 @xmath123 @xmath124 @xmath125 @xmath126 where equation ( [ perturb07 ] ) can be derived by combining equations ( [ perturb05 ] ) and ( [ perturb06 ] ) . we now rearrange the time - dependent coplanar mhd perturbation equations by taking @xmath127 from equations ( [ perturb01])@xmath128([perturb07 ] ) ; and the special case of @xmath129 will be analyzed in details at the end of this subsection . from equations ( [ perturb05 ] ) and ( [ perturb06 ] ) , we readily obtain @xmath130 a substitution of equation ( [ zr ] ) into the radial and azimuthal components of the momentum equation ( [ perturb02 ] ) and ( [ perturb03 ] ) yields @xmath131 \bigg(\frac{u}{\omega - m\omega}\bigg)\ , \end{split}\ ] ] and @xmath132 respectively . in order to construct global solutions without the wkbj approximation , we are mainly interested in stationary configurations with zero pattern speed @xmath133 , also referred to as neutral @xmath134 modes ( e.g. , syer & tremaine 1996 ; shu et al . 2000 ; lou 2002 ; shen & lou 2004a , b ; lou & zou 2004a ) . by setting @xmath134 and taking @xmath127 , equations ( [ perturb01 ] ) , ( [ zr ] ) , ( [ radial ] ) and ( [ azimuthal ] ) can be readily reduced to @xmath135 @xmath136 @xmath137\frac{\hbox{i}u}{\omega}\ , \end{split}\ ] ] @xmath138 equations ( [ stationary1 ] ) through ( [ stationary4 ] ) together with equation ( [ perturb04 ] ) are the basic mhd perturbation equations for constructing stationary non - axisymmetric @xmath127 configurations of both aligned and unaligned logarithmic spiral cases . note that aligned and spiral stationary global solutions both involve propagations of fast and slow mhd density waves ( lou 2002 ) . we next present the basic coplanar mhd perturbation equations for stationary axisymmetric @xmath129 configurations with or without radial propagations . starting from equations ( [ perturb01 ] ) , ( [ perturb06 ] ) , ( [ perturb07 ] ) , ( [ radial ] ) and ( [ azimuthal ] ) by setting @xmath129 and @xmath139 , we obtain @xmath140\hbox{i}u\ , \\ & \omega j=\frac{r\kappa^2}{2\omega}\hbox{i}u\ , \\ & r=0\ , \qquad\qquad \omega z=\frac{\partial}{\partial r}(b_0\hbox{i}u)\ . \end{split}\ ] ] this set of equations is applicable as we proceed to investigate stationary axisymmetric coplanar mhd perturbations ( lou & zou 2004a , b ) . for coplanar hydrodynamic perturbations or coplanar mhd perturbations in an isopedically magnetized sid ( shu et al . 2000 ) , there is no essential difference for changing the order of limiting procedure for @xmath129 and @xmath141 . however , for msid with a coplanar azimuthal background magnetic field , the results would be different by changing the order of limiting procedure for @xmath129 and @xmath141 ( lou 2002 ; lou & fan 2002 ; lou & zou 2004a ) . to solve the poisson equation connecting the perturbed surface mass density and the perturbed gravitational potential , we may consistently assume disturbances to carry proper scale - free forms ( lynden - bell & lemos 1993 ) for aligned and logarithmic spiral perturbations . for aligned perturbations , @xmath111 contains only an amplitude variation in @xmath14 of perturbed surface mass density and does not involve phase variation in @xmath14 so that the maximum density perturbations at different radii line up in the azimuth . for spiral perturbations in comparison , in addition to an amplitude variation in @xmath14 , @xmath111 also involves a phase variation in @xmath14 such that a spiral pattern emerges . for aligned global coplanar mhd perturbations , we select those perturbations that carry the same power - law variation in @xmath14 of the background equilibrium , namely @xmath142 where @xmath143 is a small constant coefficient and the numerical factor @xmath58 is given explicitly by @xmath144 where @xmath145 ( qian 1992 ; syer & tremaine 1996 ; shen & lou 2004a , b ) . in the limit of @xmath60 , we simply have @xmath146 consistent with earlier results on sids and msids ( shu et al . 2000 ; lou 2002 ; lou & shen 2003 ; lou & zou 2004a ; lou & wu 2004 ) . in fact , a more general class of self - consistent potential - density pairs satisfying the poisson integral ( [ perturb04 ] ) is also available ; see footnote 3 in shen & lou ( 2004b ) . for aligned neutral modes of axisymmetry ( @xmath129 ) , we can start from equation ( [ m0 ] ) by setting @xmath134 ; this implies @xmath147 and gives no constraints on @xmath148 . as @xmath149 , the scale - free condition requires @xmath150 with @xmath151 and @xmath152 . it turns out that such stationary coplanar mhd perturbations are simply alternative states for the axisymmetric background equilibrium with a proper rescaling factor ( shu et al . 2000 ; lou 2002 ) . as this is somewhat trivial , we now turn our attention mainly to cases of non - axisymmetric @xmath127 stationary aligned coplanar mhd perturbations . for the potential - density pair @xmath149 and @xmath153 , we have @xmath154 and @xmath152 according to equations @xmath155 . it then follows that @xmath156}{2m\omega r^2}\hbox{i}u\ , \\ & m\omega j+r(1-\beta)\omega \hbox{i}u=\frac{ma^2}{r\sigma_0}\bigg(\frac{r\sigma_0}{a^2}2\pi g { \cal p}_m-1\bigg)rs+\frac{c_a^2(1 - 4\beta)}{2\omega r}\hbox{i}u\ , \\ & z=-\bigg(\frac{1}{2}-2\beta\bigg)\frac{b_0\hbox{i}u}{m\omega r}\ , \qquad\qquad\qquad r=\frac{b_0u}{\omega r}\ . \end{split}\ ] ] rearranging the first three equations of ( [ alignedmge1 ] ) , we obtain a set of three algebraic equations , namely @xmath157 where coefficients @xmath158 , @xmath159 , @xmath160 ( @xmath161 ) are explicitly defined by @xmath162\ , \quad\qquad b_2\equiv m\omega+\frac{[(1 - 4\beta)^2 - 2m^2]c_a^2}{2m\omega r^2}\ , \quad\qquad c_2\equiv\frac{2\omega}{r}\ , \\ & a_3\equiv -mr\bigg[2\pi g{\cal p}_m(\beta ) -\frac{a^2}{\sigma_0 r}\bigg]\ , \qquad\qquad b_3\equiv r(1-\beta)\omega-\frac{(1 - 4\beta)c_a^2}{2\omega r}\ , \qquad\qquad c_3\equiv m\omega\ . \end{split}\ ] ] for nontrivial solutions of @xmath163 , the determinant of algebraic equations ( [ suj ] ) must vanish , namely @xmath164 using coefficient definitions ( [ abc ] ) in condition ( [ aligned0 ] ) , we obtain a more informative form of stationary dispersion relation @xmath165}{[m^2\omega^2r^2 -c_a^2(m^2 - 1/2 - 8\beta^2 + 4\beta ) ] [ ( 1 + 2\beta)\omega^2r^2-c_a^2(1/2 - 2\beta)]}\ \\ & \qquad\qquad\qquad -\frac{4\beta a^2 - 8\beta\pi g{\cal p}_m\sigma_0r + ( 1 - 4\beta)c_a^2}{2[m^2\omega^2r^2 -c_a^2(m^2 - 1/2 - 8\beta^2 + 4\beta)]}\ . \end{split}\ ] ] note that equation ( [ aligned1 ] ) takes a simpler form in the limit of @xmath60 ( i.e. , a flat rotation curve of a sid ) and reduces to equation ( 3.1.9 ) of lou ( 2002 ) as expected . by using background mhd equilibrium variables ( [ propequilib ] ) in condition ( [ aligned1 ] ) , we obtain the stationary dispersion relation for global aligned @xmath127 coplanar mhd configurations in the form of a quadratic equation in terms of @xmath166 , namely @xmath167 where the three coefficients @xmath168 ( @xmath161 ) are defined by @xmath169q^2 -{\cal a}_m(1-f { \cal c}{\cal p}_m)\ , \\ & c_0\equiv -\frac{1 - 4\beta}{4(1 + 2\beta ) } \bigg[\frac{-2m^2+(1 - 4\beta)(1 - 2\beta)}{1 + 2\beta } f{\cal c}{\cal p}_m+1 - 4\beta\bigg]q^4 -\frac{-2m^2+(1 - 4\beta)(1 - 2\beta)}{2(1 + 2\beta ) } ( 1-f{\cal c}{\cal p}_m)q^2\ , \end{split}\ ] ] with auxiliary parameters explicitly defined by @xmath170\ . \end{split}\ ] ] which are exactly the same as those defined in shen & lou ( 2004b ) for aligned cases of global perturbations in a purely hydrodynamic composite disc system . here , the effect of coplanar magnetic field is associated with the @xmath81 parameter in definitions ( [ alignc2c1c0 ] ) . the determinant of quadratic equation ( [ aligned ] ) , @xmath171 , depends on various parameters and can become negative under some circumstances ( i.e. , no physical solutions for @xmath166 ) . for @xmath172 , the two real solutions of @xmath166 to equation ( [ aligned ] ) are @xmath173 as a first check of necessary consistency , we consider the limiting case of vanishing magnetic field , that is , @xmath174 . then , equation ( [ aligned ] ) simply reduces to @xmath175 which gives a non - trivial solution @xmath176 this @xmath87 solution ( [ alignednomsol ] ) is exactly the same result for the case of a single scale - free gas disc without magnetic field ( see syer & tremaine 1996 and subsection 3.1 of shen & lou 2004b ) . as a second check of necessary consistency , we consider another limiting case of @xmath60 which corresponds to the msid case with a flat rotation curve studied by lou ( 2002 ) . one can readily show that equation ( [ aligned ] ) reduces to equation ( 3.2.11 ) of lou ( 2002 ) as expected . in contrast to the case of a single fluid disc studied by syer & tremaine ( 1996 ) without azimuthal magnetic fields , there are now two branches of @xmath87 solutions of algebraic equation ( [ aligned ] ) in general . a @xmath87 solution is considered to be physical when it satisfies both the condition @xmath86 and inequality ( [ restrt ] ) . when the magnetic field becomes sufficiently weak , among these two branches of @xmath87 solutions , @xmath177 is the counterpart of the single disc case and @xmath178 is additional due to the coupling between the surface mass density and magnetic field . we recall that in a composite system of one stellar disc and one gaseous disc coupled through the mutual gravitational interaction , there are also two branches of @xmath87 solutions ( lou & fan 1998b , 2000 ; lou & shen 2003 ; shen & lou 2004b ) . in a composite stellar - gaseous disc system , the two different classes of perturbation modes correspond to either in - phase or out - of - phase of surface mass density perturbation enhancements ( see also lou & wu 2004 ) . in a single azimuthally magnetized disc , the two coplanar mhd modes @xmath177 and @xmath178 will be distinguished by either in - phase or out - of - phase perturbation enhancements of the surface mass density and the azimuthal magnetic field in the wkbj or tight - winding approximation . in the wkbj regime , the @xmath177 and @xmath178 branches correspond to stationary fast and slow mhd density waves , respectively ( fan & lou 1996 ; lou & fan 1998a ) . although the phase relationships for perturbation enhancements of the surface mass density and the azimuthal magnetic field become @xmath179 for open spiral structures ( lou 2002 ; lou & fan 2002 ) , we shall still refer to @xmath177 and @xmath178 solutions as global stationary fast and slow mhd density waves , respectively . we shall come back to this discussion of ` phase relationship ' in more details in section 4 . as an example of illustration , we now focus on the specific case of @xmath11 with inequality ( [ restrt ] ) automatically satisfied . in such a magnetized disc system , the rotation curve scales as @xmath180 , the surface mass density scales as @xmath181 , the barotropic index is @xmath182 and the azimuthal magnetic field scales as @xmath183 ( i.e. , the background equilibrium magnetic field is force - free ) . for this specific case , the determinant @xmath184 of algebraic equation ( [ aligned ] ) remains always positive for both full and partial discs ( see proofs in appendix a ) . as seen from @xmath87 solutions ( [ solalign ] ) , the sign of @xmath185 determines whether @xmath177/@xmath178 is the upper / lower branch or reverse . for @xmath186 , we always have @xmath187 and therefore @xmath177 and @xmath178 remain to be the higher and lower branches , respectively . the situation becomes somewhat involved for the @xmath188 case when the sign of @xmath185 is dependent on the value of @xmath189 parameter . we show in fig . 1 the @xmath166 solution branches for @xmath188 and @xmath190 cases , respectively , versus the gravitational potential ratio @xmath189 . for the @xmath188 case , there exists a diverging point at @xmath191 $ ] for @xmath192 solution branch where @xmath193 . the value of this critical @xmath194 is numerically determined to be @xmath195 . more specifically , we have @xmath196 for @xmath197 , while we have @xmath187 for @xmath198 . this sign variation of @xmath185 is indicated by the relative locations of @xmath177 and @xmath178 solution branches as shown in panel @xmath199 of fig . in other words , for @xmath200 , @xmath177 remains negative and the only physical solution is @xmath178 . on the other hand for @xmath198 , both @xmath177 and @xmath178 solution branches become physically plausible . for @xmath11 , the explicit form of algebraic equation ( [ aligned ] ) becomes @xmath201d^4 -\bigg[\bigg(\frac{2{f}{\cal c}{\cal p}_m}{3 } + 1\bigg)m^2q^2+\bigg(m^2+\frac{3}{4}\bigg)(1-{f } { \cal c}{\cal p}_m)\bigg]d^2+\frac{2}{3}(1-{f } { \cal c}{\cal p}_m)m^2q^2=0\ , \ ] ] where auxiliary parameters @xmath202 , @xmath203 , @xmath204 , @xmath205 and @xmath206 are all evaluated at @xmath11 . it is fairly straightforward to show that @xmath207 for all @xmath127 when @xmath11 ( shen & lou 2004b ) . as @xmath208 , we then have @xmath209 . in our formulation , stationary surface mass density perturbations are characterized by @xmath210 . for aligned coplanar mhd perturbations , we took @xmath211 where @xmath212 is a real exponent . in section 3.1 , we have chosen @xmath213 that carries the same power - law dependence as the equilibrium disc does . for a complex @xmath212 , a perturbation pattern would appear spiral , that is , a logarithmic spiral in the form of @xmath214 $ ] where @xmath215 and @xmath216 are real and imaginary parts of @xmath212 . to ensure the gravitational potential arising from this perturbed surface mass density being finite , we should require @xmath217 ( e.g. , qian 1992 ) . in fact , there exists a more general class of self - consistent potential - density pairs as indicated in footnote 7 of shen & lou ( 2004b ) . however , for the analysis of axisymmetric stability problem in section 3.2.1 , equation ( [ spirperturb ] ) is used such that dispersion relation ( [ m0u ] ) derived later on is real on both sides ( e.g. , lemos et al . 1991 ; syer & tremaine 1996 ; shu et al . 2000 ) . to be specific here , we consistently take the logarithmic spiral surface mass density perturbations and the resulting gravitational potentials as ( kalnajs 1971 ; lemos et al . 1991 ; shu et al . 2000 ; lou 2002 ; lou & shen 2003 ; shen & lou 2004b ; lou & zou 2004a ) @xmath218 where @xmath143 is a small constant , @xmath219 is a parameter related to the radial wavenumber and the kalnajs function @xmath220 is defined by @xmath221 ( kalnajs 1971 ) . as @xmath220 is an even function of @xmath219 , a consideration of @xmath222 would suffice . in our convention of notations , @xmath223 and @xmath224 correspond to leading and trailing logarithmic spiral waves , respectively , and @xmath225 relates the parameter @xmath219 to the radial wavenumber @xmath226 . we note that @xmath220 decreases monotonically with increasing @xmath223 , and @xmath227 for @xmath127 ; for @xmath129 , @xmath228 is positive and can be greater than 1 when @xmath219 becomes sufficiently small . the choice of unaligned perturbations is not unique , and the perturbation potential - density pairs may involve @xmath8 parameter such that the background plus coplanar mhd perturbations are altogether scale - free ( e.g. , syer & tremaine 1996 ) . as an example of illustration , we here presume coplanar logarithmic spiral perturbations defined by the density - potential pair ( [ spirperturb ] ) in the following analysis . for the specific @xmath11 case , the background surface mass density and the perturbed surface mass density bear the same radial scaling @xmath229 . we start from the axisymmetric coplanar mhd perturbation equation ( [ m0 ] ) with @xmath139 . for radial oscillations with a logarithmic potential - density pair , we presume a surface mass density @xmath230 , a gravitational potential @xmath231 and thus a radial speed @xmath232 according to the first of equation ( [ m0 ] ) . after proper rearrangements , the set of equations ( [ m0 ] ) can be combined to form a single equation in terms of @xmath112 , namely @xmath233u\\ & = \frac{a^2}{r^2}\bigg\{\bigg[1-{f}{\cal c } { \cal n}_0\bigg(1+d^2-\frac{1 - 4\beta}{2 + 4\beta}q^2\bigg)\bigg ] \bigg(\frac{1}{4}+\xi^2\bigg ) + q^2\bigg(\frac{1}{4}+\xi^2-\beta\bigg)\bigg\}u\ , \end{split}\ ] ] where @xmath234 , @xmath235 $ ] and @xmath236 is extremely small . by inspection , equation ( [ m0u ] ) bears similar features of the dispersion relation in the classic wkbj regime involving magnetic field . as the right - hand side of equation ( [ m0u ] ) is real , axisymmetric instabilities first set via neutral @xmath134 modes . it should be emphasized that the scale - free condition is met only for neutral modes . with @xmath237 , the marginal stability curves are then given by @xmath238\}(\xi^2 + 1/4)+q^2(\xi^2-\beta+1/4)}{{f } { \cal c}{\cal n}_0(\xi^2 + 1/4)+(1 + 2\beta)(-2 + 2\beta)}\ .\ ] ] as a check of necessary consistency , the axisymmetric marginal stability curve for an msid when @xmath19 ( @xmath239 ) then gives @xmath240 ( \xi^2 + 1/4)}{{f}{\cal n}_0(\xi^2 + 1/4)-2}\ .\ ] ] one can readily show ) is automatically satisfied for @xmath85 and we need only to examine the range of @xmath88 . when the denominator of the right - hand side of equation ( [ msc ] ) is negative , the numerator must also be negative for @xmath86 . it then follows that @xmath241 and inequality ( [ restrt ] ) is met . when the denominator of the right - hand side of equation ( [ msc ] ) is positive , it is easy to show that inequality ( [ restrt ] ) holds . as solution ( [ mscbeta0 ] ) is a special case of solution ( [ msc ] ) , inequality ( [ restrt ] ) is thus met . ] that stationary dispersion relations ( [ msc ] ) and ( [ mscbeta0 ] ) automatically satisfy inequality ( [ restrt ] ) for physical solutions @xmath86 . in the msid case with different values of @xmath77 parameter for magnetic field strengths , one readily obtains the marginal stability curves that separate two distinct unstable regimes , namely , the collapse regime for large radial spatial scales and the ring fragmentation regime for relatively short radial wavelengths ( lemos et al . 1991 ; shen & lou 2003 , 2004a ) . there exists a critical @xmath242 where @xmath87 diverges and this @xmath243 is determined by @xmath244 . for @xmath245 , we have @xmath246 and thus an enhancement of @xmath77 tends to suppress the collapse boundary . for @xmath247 , we have @xmath248 and thus an enhancement of @xmath77 tends to raise the ring fragmentation boundary . in short , an enhancement of ring magnetic field tends to reduce dangers of both collapse and ring fragmentation instabilities in the msid case . while the enhancement of ring magnetic field tends to suppress ring fragmentation instabilities in general , it only suppresses collapse instabilities for @xmath88 and tends to aggravate collapse instabilities for @xmath89 ( see fig . 2 for details ) . it has been shown that the magnetic force is radially inward and outward for @xmath88 and @xmath89 , respectively . why does the trend of variations for the collapse - stability appear seemingly paradoxical ? this situation can be understood because the unperturbed background is in an mhd radial force balance and the surface mass density , the rotation speed and the magnetic field are coupled explicitly through condition ( [ propequilib ] ) . furthermore , for fast mhd density waves with @xmath129 ( lou & fan 1998a ) , the gas pressure and magnetic pressure together is associated with the radial wavenumber squared , while the surface mass density and self - gravity are associated linearly with the radial wavenumber ( see the first line of dispersion relation [ m0u ] with small @xmath236 ) . for @xmath88 , the surface mass density becomes smaller for stronger magnetic field strength , while for @xmath89 , the situation reverses . for the ring fragmentation instability at relatively large radial wavenumbers , the dominant magnetic pressure effect tends to stabilize the disc . for the jeans collapse instability at relatively small radial wavenumbers , the dominant self - gravity effect is proportional to the background surface mass density ; the roles of magnetic field for two different situations can then be readily understood . it is the coupling effect between the surface mass density and the magnetic field of the background that gives rise to this collapse feature . the marginal stability curves for scale - free discs with @xmath249 and without magnetic fields are @xmath250 that are consistent with those for the single - disc case of shen & lou ( 2004b ; e.g. , see their subsection 3.2.1 ) . the marginal axisymmetric stability curves generally consist two branches , namely , the collapse branch and the ring - fragmentation branch ( lemos et al . 1991 ; shu et al . 2000 ; lou 2002 ; lou & shen 2003 ; shen & lou 2003 ) . the lowest value of @xmath87 for stability is the maximum of the collapse branch and the highest value of @xmath87 for stability is the minimum of the ring - fragmentation branch . we note that the maximum of the collapse branch is always located at @xmath251 for @xmath252 , while for @xmath253 the maximum of the collapse branch may locate at @xmath223 with @xmath251 corresponding to a local minimum of @xmath87 . it was proven in appendix c of shen & lou ( 2004b ) that @xmath251 is always a local extremum for @xmath87 . variations of the stable range for @xmath87 with parameters @xmath254 , @xmath8 and @xmath77 are shown in panels ( a ) and ( b ) of fig . finally we note that the axisymmetric destabilization involves only the fast mhd wave branch because the slow mhd wave branch is negative and thus unphysical . for non - axisymmetric ( @xmath255 ) perturbations , both fast and slow mhd density waves are possible and stationary global configurations may represent transitions from stable to unstable situations ( e.g. , shu et al . 2000 ) . here , the stability of our slow mhd density waves differs from those models involving finite disc thickness where either vertical or horizontal weak magnetic fields may induce velikhov - chandrasekhar - balbus - hawley instability and magneto - rotational instabilities ( mri ) through slow mhd modes ( chandrasekhar 1961 ; balbus & hawley 1998 ; kim & ostriker 2000 ; lou , yuan & fan 2001 ) . it would be interesting to recall the axisymmetric stability analysis in the wkbj approximation . the local dispersion relation for fast mhd waves in an azimuthal magnetic field is ( fan & lou 1996 ) @xmath256 where @xmath226 is the radial wavenumber and other notations have the conventional meanings . equation ( [ drwkbj ] ) gives the marginal stability curve for axisymmetric @xmath129 perturbations as @xmath257 by using scale - free disc equilibrium condition ( [ propequilib ] ) and inserting @xmath258 for large @xmath226 and @xmath219 into dispersion relation ( [ stability ] ) , we readily obtain the corresponding marginal stability curve in the wkbj regime as @xmath259\xi } { f{\cal c}\xi+(1 + 2\beta)(-2 + 2\beta)}\ , \ ] ] which bears a strong resemblance to expression ( [ msc ] ) of our global analysis , especially in view of the asymptotic expression for @xmath220 , namely @xmath260 when @xmath261 . another equivalent way of this comparison is to note the parallel terms in the wkbj condition ( [ stability ] ) and in the first line of the corresponding global condition ( [ m0u ] ) , also in reference to the above asymptotic expression for @xmath220 . a specific illustrating example of such a comparison is shown in fig . 3 with parameters @xmath19 ( @xmath239 ; flat rotation curve ) , @xmath35 ( full disc ) and @xmath262 . in the short - wavelength limit ( i.e. , @xmath263 ) , the wkbj approximation is well justified and the results are qualitatively consistent , while in the long - wavelength of jeans collapse regime , the wkbj analysis differs from the global exact treatment significantly . , @xmath35 and @xmath262 . the marginal stability curves for the long - wavelength jeans collapse and the short - wavelength ring fragmentation regimes are denoted . for @xmath263 ( i.e. , the short - wavelength limit ) , the wkbj approximation is well applicable , while for small @xmath219 the results in the wkbj approximation deviates from those of the exact global treatment considerably ( see shen & lou 2003 , 2004b ) . ] in parallel with the aligned @xmath127 cases , we now construct unaligned logarithmic spiral solutions from coplanar mhd perturbation equations ( [ stationary1])@xmath128([stationary4 ] ) . for the density - potential pair @xmath230 and @xmath264 , we then have radial flow speed @xmath265 and @xmath266component specific angular momentum @xmath267 . coplanar mhd perturbation equations @xmath155 can therefore be written in the forms of @xmath268\hbox{i}u\ , \\ & m\omega j+r(1-\beta)\omega \hbox{i}u = m\bigg(2\pi g{\cal n}_m -\frac{a^2}{r\sigma_0}\bigg)rs+\frac{c_a^2(1 - 4\beta ) } { 2\omega r}\hbox{i}u\ , \\ & z=-\hbox{i}\xi\frac{b_0iu}{m\omega r}\ , \qquad\qquad\qquad\qquad r=\frac{b_0u}{\omega r}\ . \end{split}\ ] ] the first three equations of set ( [ spiralmge1 ] ) immediately lead to @xmath269 where coefficients @xmath158 , @xmath159 and @xmath160 ( @xmath161 ) are explicitly defined by @xmath270c_a^2}{m\omega r^2}\ , \qquad c_2\equiv \frac{2\omega}{r}\ , \\ & a_3\equiv -mr\bigg(2\pi g{\cal n}_m-\frac{a^2}{\sigma_0 r}\bigg)\ , \qquad\qquad b_3\equiv r(1-\beta)\omega-\frac{(1 - 4\beta)c_a^2}{2\omega r}\ , \qquad\qquad c_3\equiv m\omega\ . \end{split}\ ] ] for nontrivial solutions of the set of homogeneous algebraic equations ( [ sujspiral ] ) , the determinant must vanish @xmath271 by using the set of expressions ( [ abcspiral ] ) and applying equilibrium conditions ( [ propequilib ] ) , we obtain the final dispersion relation for stationary logarithmic spiral @xmath127 mhd configurations in the form of a quadratic equation in terms of @xmath166 , namely @xmath272 where the coefficients @xmath185 , @xmath273 and @xmath274 are explicitly defined by @xmath275q^2 -{\cal a}_m(1-{f}{\cal c}{\cal n}_m)\ , \\ & c_0\equiv -\frac{(1 - 4\beta ) } { 4(1 + 2\beta ) } \bigg[\frac{(-2m^2 - 2\xi^2 - 2\beta+1/2)}{(1 + 2\beta ) } { f}{\cal c}{\cal n}_m+1 - 4\beta\bigg]q^4 -\frac{(-2m^2 - 2\xi^2 - 2\beta+1/2)}{2(1 + 2\beta ) } ( 1-{f}{\cal c}{\cal n}_m)q^2\ , \end{split}\ ] ] with auxiliary parameters defined by @xmath276\ . \end{split}\ ] ] that are the same as those in shen & lou ( 2004b ) for unaligned logarithmic spiral cases . for a positive determinant @xmath277 , the two real solutions to quadratic equation ( [ spiral ] ) are given by @xmath278 similar to @xmath127 cases for aligned perturbations in the limit of vanishing magnetic field ( i.e. , @xmath279 ) , equation ( [ spiral ] ) reduces to @xmath280 that leads to a nontrivial solution @xmath281 solution ( [ spiralnomsol ] ) is exactly the same result for the single - disc case discussed in section 3.2 of shen & lou ( 2004b ) . another limiting regime of @xmath60 has been analyzed by lou ( 2002 ) for global stationary perturbation structures in a single msid . for stationary spiral mhd configurations , we here analyze again the @xmath11 case as a specific example of illustration . the background equilibrium ring magnetic field is force - free and inequality ( [ restrt ] ) is satisfied . the determinant @xmath184 of quadratic equation ( [ spiral ] ) remains always positive as shown in appendix a and hence two real solutions of @xmath282 to equation ( [ spiral ] ) exist . in parallel with the aligned @xmath11 case , we solve quadratic equation ( [ spiral ] ) for given values of parameter @xmath283 and display in diagrams the two branches of solutions @xmath177 and @xmath178 ( @xmath166 ) versus @xmath219 variation . as @xmath284 for @xmath188 and when @xmath283 becomes sufficiently small , there is a diverging point for @xmath192 solution where @xmath193 . since @xmath285 increases monotonically with increasing @xmath219 for @xmath222 and attains its minimum value at @xmath251 ( @xmath285 is an even function of @xmath219 ; see appendix c of shen & lou 2004b ) , the existence of such a critical point @xmath243 for a diverging @xmath166 requires @xmath286 for the specific @xmath11 case , this inequality ( [ criticalf ] ) implies a critical @xmath283 value of @xmath287 $ ] below which there exists one divergent point of @xmath243 where @xmath177 diverges . numerically , we have determined @xmath288 [ note that this @xmath289 is of the same value as that of the aligned @xmath188 case because @xmath290 . for selected @xmath254 values in the range of @xmath291 , the corresponding critical @xmath243 at which @xmath292 can be numerically computed ; for example , we have @xmath293 for @xmath294 . it becomes much simpler for cases of @xmath186 because @xmath185 remains always positive and there is no diverging point for @xmath87 solutions . therefore , @xmath177 and @xmath178 remain always as the upper and lower branches , respectively . for @xmath188 and @xmath190 cases with different parameters of @xmath283 and @xmath77 , we show in four panels @xmath295 of fig . 4 relevant curves of @xmath166 versus @xmath219 variation to illustrate the basic features . to be more explicit for the case of @xmath11 in quadratic equation ( [ spiral ] ) , we write @xmath296d^4 -\bigg[(m^2+\xi^2)\bigg(\frac{2}{3 } { f}{\cal c}{\cal n}_m+1\bigg)q^2\\ & \hbox { } \qquad\qquad\qquad\qquad\qquad\qquad\qquad + \bigg(m^2+\xi^2+\frac{3}{4}\bigg ) ( 1-{f}{\cal c}{\cal n}_m)\bigg]d^2 + \frac{2}{3}(m^2+\xi^2)(1-{f}{\cal c}{\cal n}_m)q^2=0\ , \end{split}\ ] ] where auxiliary parameters @xmath297 , @xmath203 , @xmath204 , @xmath205 and @xmath206 are all evaluated for the special value of @xmath11 . for multi - wavelength observations of nearby spiral galaxies ( sofue et al . 1986 ; kronberg 1994 ; beck et al . 1996 ) , it is possible to identify large - scale spatial phase relationships of spiral patterns using various observational diagnostics ( e.g. , mathewson et al . 1972 ; visser 1980a , b ; neininger 1992 ; beck & hoernes 1996 ; frick et al . 2000 , 2001 ; ferrire 2001 ; lou et al . 2002 ) . for this reason , we derive below spatial phase relationships for various coplanar mhd perturbation variables , keeping in mind of potential applications to magnetized spiral galaxies ( e.g. , fan & lou 1996 ; lou & fan 1998a ) . here we systematically examine the spatial phase relationships among enhancements of gas surface mass density , magnetic field and velocity perturbations by analysing the relationships among @xmath298 , @xmath148 , @xmath73 , @xmath299 and @xmath300 for aligned and logarithmic spiral cases , separately . starting from the set of homogeneous algebraic equations ( [ suj ] ) for the aligned cases , we obtain the relationships among @xmath298 , @xmath301 and @xmath299 in the following , namely @xmath302 where @xmath158 , @xmath159 and @xmath160 ( @xmath161 ) are defined by expressions ( [ abc ] ) . by using definitions ( [ abc ] ) and equations ( [ alignedmge1 ] ) , we obtain @xmath303 for the aligned cases , where the common dimensionless real factor @xmath304 is defined below by @xmath305 similarly , by using the set of homogeneous algebraic equations ( [ sujspiral ] ) , definitions ( [ abcspiral ] ) and equations ( [ spiralmge1 ] ) , we readily derive for the cases of global stationary logarithmic spiral configurations @xmath306\ , \end{split}\ ] ] where the common dimensionless complex factor @xmath307 is defined below by @xmath308 the two sets of algebraic equations ( [ relsurjz ] ) and ( [ relsurjzspiral ] ) are the most general phase relationships for aligned and logarithmic spiral cases , respectively . once a @xmath8 value is specified and a particular @xmath87 for a stationary solution inserted , we immediately obtain the phase relationship between any two perturbation variables . as an example of illustration , we examine again the special @xmath11 case and determine the relevant phase relationships in the next subsection . when @xmath11 , the results of phase relations derived in the last subsection 4.1 can be simplified considerably as all terms involving the factor @xmath309 vanish . meanwhile , inequality ( [ restrt ] ) is automatically satisfied . the only requirement for a physical @xmath87 solution is simply @xmath86 . as already noted in subsections 3.1.3 and 3.2.3 , there are two branches of @xmath87 solutions . we insert relevant physical @xmath87 solutions in equations ( [ relsurjz ] ) and ( [ relsurjzspiral ] ) to determine phase relationships among perturbation variables . in particular , we are interested in the phase relationship between the perturbed surface mass density and the perturbed azimuthal magnetic field . we now analyze below this mass - magnetic field phase relationships for aligned and spiral cases , separately . some illustrative examples are shown in fig . 5 and fig . 6 to provide more direct visual impressions . for aligned @xmath11 cases , we readily find @xmath310 which means the azimuthal magnetic field remains unchanged . from equation ( [ relsurjz ] ) , we derive @xmath311 meanwhile , equation ( [ align0d25 ] ) can be rewritten in the form of @xmath312 } = \frac{d^2\{d^2-(1-f{\cal c}{\cal p}_m)/ [ f{\cal c}{\cal p}_m+3(m^2 - 3/2)/(2m^2 + 3/2)]\ } } { d^2-(1-f{\cal c}{\cal p}_m)/(f{\cal c}{\cal p}_m+3/2 ) } \ , \ ] ] that contains two distinct @xmath87 solutions according to subsection 3.1.3 , namely , the plus - sign solution @xmath177 and the minus - sign solution @xmath178 in the solution form ( [ solalign ] ) , respectively . as shown in appendix b , we emphasize that the plus - sign solution @xmath177 ( if positive and thus physical ) makes @xmath313 and the minus - sign solution @xmath178 makes @xmath314 on the right - hand side of equation ( [ q2aligned ] ) , respectively . therefore , the two branches of @xmath87 solutions ( if positive and thus physical ) will give either @xmath315 for the plus - sign @xmath177 solution or @xmath316 for the minus - sign @xmath178 solution . for spiral cases with @xmath11 , we readily obtain from equations ( [ relsurjzspiral ] ) and ( [ thetas ] ) that @xmath317(3 + 2\hbox{i}\xi)\ .\ ] ] following the same procedure , we can show that the two branches of @xmath87 solutions , namely , the plus - sign solution @xmath177 and the minus - sign solution @xmath178 given by equation ( [ solspiral ] ) , if positive and thus physical , correspond to the factor @xmath318 being positive and negative , respectively ( see appendix b for a detailed proof ) . we now examine the phase relationship between coplanar perturbation enhancements of surface mass density and magnetic field . in the short - wavelength limit ( i.e. , @xmath263 with a logarithmic spiral pattern tightly wound ) , we have @xmath319 and @xmath320 where the plus- and minus - signs correspond to the @xmath177 and @xmath178 solutions , respectively . the phase relations between perturbation enhancements of surface mass density and magnetic field are in - phase for the @xmath177 solution ( if positive and physical ) and out - of - phase for the @xmath178 solution . for @xmath186 , @xmath177 remains always greater than @xmath178 . in contexts of magnetized spiral galaxies , these solution properties in the wkbj regime are consistent with the physical identifications of the @xmath177 mode with a stationary fast mhd density wave and the @xmath178 mode with a stationary slow mhd density wave , respectively ( fan & lou 1996 ; lou & fan 1998a ; lou & fan 2002 ) . it is now possible to model magnetized spiral galaxies in terms of coplanar stationary fast and slow mhd density waves with more general rotation curves including the case of a flat rotation curve for a magnetized singular isothermal disc ( shu et al . 2000 ; lou 2002 ; lou & fan 2002 ; lou & zou 2004a , b ; lou & wu 2004 ) . in the long - wavelength limit ( i.e. , @xmath321 with a logarithmic spiral pattern widely open ) , we have @xmath322 and @xmath323 where the minus- and plus - signs correspond to the @xmath177 and @xmath178 solutions , respectively . therefore , the coplanar perturbation enhancement of magnetic field is either ahead of or lagging behind the perturbation enhancement of surface mass density by a phase difference of @xmath324 . for intermediate radial wavelengths , the phase difference between perturbation enhancements of surface mass density and magnetic field in a logarithmic spiral pattern can be readily determined for given specific conditions . on the basis of our model analysis here , it appears that except for global stationary fast mhd density waves in the extreme wkbj regime , magnetic and gas spiral arms tend to be phase shifted relative to each other for a global stationary logarithmic spiral pattern for a much broader set of rotation curves including the special case of a flat rotation curve . one can readily show that for aligned global stationary mhd perturbation configurations , there is no net radial angular momentum flux as aligned stationary mhd density waves do not propagate radially ( shu et al . 2000 ; lou 2002 ) . we here briefly discuss the process of angular momentum transfer in association with the stationary logarithmic spiral mhd modes obtained in this paper . according to the analyses of lynden - bell & kalnajs ( 1972 ) and fan & lou ( 1999 ) , the total angular momentum flux carried by coplanar mhd density waves should contain three separate contributions , namely , the advective transport flux @xmath325 defined by @xmath326 the gravity torque flux @xmath327 defined by @xmath328\bigg[\frac{\partial\re(\phi_1 ) } { \partial r}\bigg]\\ ] ] and the magnetic torque flux @xmath329 defined by @xmath330 the perturbed three - dimensional gravitational potential @xmath331 associated with a fourier component of a coplanar logarithmic spiral perturbation in surface mass density @xmath332 is then @xmath333 where @xmath334 is the cylindrical bessel function of order @xmath109 with an argument @xmath23 ( e.g. , binney & tremaine 1987 ) . in fact , integral ( [ 3dphi1 ] ) gives a simple form of kalnajs function in the disc plane at @xmath335 with the real part of @xmath331 explicitly given by @xmath336 it is fairly straightforward to show that @xmath325 and @xmath329 are independent of @xmath14 where the phase relationships for stationary logarithmic spiral cases derived in section 4.1 have been use . while mathematically tedious , one can further show that @xmath327 is also independent of @xmath14 ( see appendix c ) . therefore , the total angular momentum flux @xmath337 remains constant at all radii , a natural result of angular momentum conservation given that perturbation surface mass density falls off as @xmath181 . therefore , in order to maintain a stationary logarithmic spiral pattern , there must be a source or a sink at the disc center . the perturbation energy flux , however , is zero at all radii for the stationary coplanar mhd configurations [ see equation ( 4.17 ) of fan & lou ( 1999 ) in the wkbj regime ] . the mass flux is also zero at all radii since the first - order mass flux @xmath338 , and the second - order mass flux @xmath339 [ where @xmath340 is the second - order perturbed radial velocity , this equation holds due to the second - order mass conservation ( fan & lou 1999 ) ] . therefore our model result differs from that of spruit ( 1987 ) for stationary logarithmic spiral shocks , where only the gravity of the central object is considered . in spruit s model , there is a net mass accretion through the disc and an effective viscosity parameter `` @xmath7 '' ( shakura & sunyaev 1973 ) may be estimated . physically , the central source of angular momentum required to sustain global stationary logarithmic spiral patterns may be plausibly identified with a fast spinning supermassive black hole ( smbh ) in the nucleus of a spiral galaxy . the kerr spacetime associated with such a spinning smbh excites and sustains mhd density waves in the surrounding disc through a long - range gravitational interaction . for trailing spiral patterns , a net flux of angular momentum is transport radially outward in the disc . conceptually , this process should cause the spinning smbh to gradually slow down , entirely analogous to the spin - down of a rotating neutron star by emitting electromagnetic waves that carry a net angular momentum away . in this paper , we have explored and analyzed stationary perturbation structures in coplanarly magnetized razor - thin scale - free discs in a more general manner . to be specific , we start from a rotational and magnetized background equilibrium of axisymmetry that is dynamically self - consistent with a surface mass density @xmath341 , a rotation curve @xmath2 , a purely azimuthal ( ring ) magnetic field @xmath1 with a vertically integrated barotropic equation of state in the form of @xmath16 . the radial force balance at all radii in an mhd disc implies several simple relationships among these power indices @xmath7 , @xmath8 , @xmath9 and @xmath6 , as stated explicitly in equation ( [ pararelation ] ) . without loss of generality , we can simply use @xmath8 as an independent parameter to specify properties of an important class of radial variations for rotational mhd background equilibria . the allowed range of @xmath8 falls within the interval @xmath71 ( syer & tremaine 1996 ; shen & lou 2004a , b ) with the special case of @xmath19 corresponding to a magnetized singular isothermal disc ( msid ) system ( shu et al . 2000 ; lou 2002 ; lou & fan 2002 ; lou & zou 2004a , b ; lou & wu 2004 ) . for clarity and convenience of our analysis , we introduced several dimensionless parameters . the first parameter is the partial disc parameter @xmath34 for the ratio of the gravitational potential ( @xmath342 ) arising from the disc to that ( @xmath28 ) arising from the entire system ( syer & tremaine 1996 ; shu et al . 2000 ; lou 2002 ) including an axisymmetric halo mass distribution that is involved in the background equilibrium but is presumed to be unresponsive to coplanar mhd disc disturbances ( a massive dark matter halo is an example in mind ) ; for a partial disc system , we have @xmath343 , while for a full disc system , we have @xmath344 by conventional definitions . the second parameter is an effective rotational mach number @xmath82 defined as the ratio of the disc rotation speed to the sound speed yet with an additional scaling factor @xmath345 ; @xmath82 is constant at all radii . the third parameter @xmath77 is a measure for the background ring magnetic field strength defined as the ratio of the azimuthal alfvn speed to the sound speed ; and @xmath77 remains constant at all radii . for @xmath89 , there is no constraint on @xmath77 and the net lorentz force of the background azimuthal magnetic field is radially outward ( the magnetic pressure force is stronger than the magnetic tension force ) to resist the self - gravitation ; for @xmath11 , the background azimuthal magnetic field is force - free ( the magnetic pressure and tension forces cancel each other exactly ) ; for @xmath88 , the net lorentz force of the background azimuthal magnetic field points radially inward ( the magnetic pressure force is weaker than the magnetic tension force ) to aggravate the self - gravity and thus a restriction [ i.e. , inequality ( [ restrt ] ) ] is imposed on the magnetic field strength in order to guarantee that the surface mass density be positive in equation ( [ propequilib ] ) . inequality ( [ restrt ] ) provides a necessary physical criterion for plausible @xmath87 solutions . with such a rotational mhd background equilibrium chosen , we introduce coplanar mhd disturbances to construct global stationary perturbation structures as viewed in an inertial frame of reference . perturbation variables are expressed in terms of fourier components for either aligned or logarithmic spiral pattern forms ( syer & tremaine 1996 ; shu et al . 2000 ; lou 2002 ; lou & shen 2003 ; shen & lou 2003 ; lou & zou 2004a ) . we derive analytical solutions for stationary coplanar mhd perturbation configurations in both aligned and logarithmic spiral cases . we now summarize key results below . 1 . cases of aligned coplanar mhd perturbations 0.2 cm + for aligned cases , we choose perturbation variables that bear the same power - law dependence in radius @xmath14 as the background mhd equilibrium does . for the special aligned case of @xmath129 , the stationary coplanar mhd perturbations actually describe alternative equilibria to the axisymmetric background equilibrium with proper rescaling factors ( shu et al . 2000 ; lou 2002 ; lou & zou 2004a ) and are therefore somewhat trivial . + for aligned cases of @xmath127 , we derive a quadratic equation ( [ aligned ] ) in terms of @xmath166 . the resulting positive @xmath87 is the rotational parameter required for sustaining a stationary nonaxisymmetric mhd perturbation configuration . other parameters involved in quadratic equation ( [ aligned ] ) include @xmath8 , @xmath109 , @xmath283 and @xmath77 . for a general combination of @xmath8 , @xmath109 , @xmath283 and @xmath77 , the determinant @xmath184 of quadratic equation ( [ aligned ] ) may not be always positive definite . on the other hand , for specific values of @xmath8 , equation ( [ aligned ] ) always gives two different real solutions ( see appendix a ) . once we prescribe parameters @xmath8 , @xmath109 , @xmath283 and @xmath77 , the two solutions of @xmath87 can be readily determined . for physical solutions , @xmath87 must be positive and satisfy inequality ( [ restrt ] ) simultaneously . + as a specific case study , we examined the case of @xmath11 . here the background azimuthal magnetic field remains force - free and the @xmath87 solution is free from constraint ( [ restrt ] ) . it has been further shown in appendix a that for this @xmath11 case , equation ( [ aligned ] ) always has two distinct real solutions , the plus - signed solution @xmath177 and the minus - signed solution @xmath178 , as contained in equation ( [ solalign ] ) . all these aspects contribute to a significant simplification of the relevant computational procedures . we derive two branches of @xmath87 solutions and need to require only @xmath346 . there is yet a qualitative difference between the @xmath188 case and the @xmath186 cases . for @xmath188 , there exists one critical value @xmath347 ; for @xmath283 below @xmath289 , @xmath177 and @xmath178 are negative and positive , respectively , while @xmath283 above @xmath289 , both branches of @xmath87 solutions are positive and thus physical . for @xmath186 in comparison , both branches of @xmath87 solutions remain always positive and thus physical . moreover , @xmath177 remains always to be the upper branch for azimuthal stationary fast mhd density waves and represents the hydrodynamic counterpart of the unmagnetized single disc case ; complementarily , @xmath178 remains always to be the lower branch for azimuthal stationary slow mhd density waves ( fan & lou 1996 ; lou & fan 1998a ) and is caused by the very presence of the azimuthal magnetic field ( detailed computations reveal that @xmath178 must be smaller than @xmath348 which seems to be too small for galactic applications ) . when a disc rotates sufficiently fast , it could only support the upper @xmath177 stationary mhd configurations . an increase of magnetic field parameter @xmath77 will raise both physical branches of @xmath87 solutions as displayed in fig . the variation trend of the upper @xmath177 branch appears to be more sensitive . therefore for a much stronger magnetic field , a sufficiently rapidly rotating disc may sustain stationary coplanar mhd configurations in terms of the upper @xmath349 branch . + 0.2 cm 2 . cases of logarithmic spiral configurations for coplanar mhd perturbations 0.2 cm + for cases of stationary spiral perturbation structures , we choose coplanar mhd perturbations in terms of kalnajs logarithmic spirals ( kalnajs 1971 ; lemos et al . 1991 ; shu et al . 2000 ; lou 2002 ; lou & fan 2002 ; lou & shen 2003 ; lou & zou 2004a ; lou & wu 2004 ) . for the special @xmath129 case with radial propagations , it happens that inequality ( [ restrt ] ) is automatically satisfied and the stationary coplanar mhd configurations with @xmath350 obtained here represent marginal stability curves ( syer & tremaine 1996 ; shu et al . 2000 ; shen & lou 2003 , 2004b ; lou & zou 2004b ) . only when the rotational parameter @xmath87 falls within a specific finite range can the magnetized disc be stable against axisymmetric disturbances at all wavelengths . a disc with too slow a rotation speed will succumb to jeans instability in the collapse regime corresponding to long wavelength perturbations , while a disc with too fast a rotation speed will suffer the ring - fragmentation instability corresponding to relatively short wavelength perturbations ( safronov 1960 ; toomre 1964 ; lemos et al . 1991 ; lou & fan 1998a , 2000 ; shu et al . 2000 ; shen & lou 2003 , 2004a ; lou & zou 2004b ) . in comparison to a hydrodynamics disc system , the enhancement of azimuthal magnetic field tends to suppress ring - fragmentation instabilities as expected from the perspective of the mhd @xmath351 parameter ( lou & fan 1998a ; lou 2002 ; lou & zou 2004a , b ) , while an enhancement of azimuthal magnetic field tends to suppress jeans collapse instabilities for @xmath88 , to aggravate jeans collapse instabilities for @xmath89 and to bear no effect on jeans collapse instabilities when @xmath11 . we have provided intuitive interpretations for the roles of magnetic field earlier . + in parallel with the aligned cases , we derive quadratic equation ( [ spiral ] ) in terms of @xmath166 for logarithmic spiral cases of @xmath127 . in addition to the four parameters @xmath8 , @xmath109 , @xmath283 and @xmath77 , we now have one more parameter @xmath219 for the radial wavenumber . while the determinant @xmath184 of quadratic equation ( [ spiral ] ) may not be always positive in general , we prove in appendix a that @xmath352 for the specific @xmath11 case . once the five parameters @xmath8 , @xmath109 , @xmath283 , @xmath77 and @xmath219 are specified , the @xmath87 solutions can be readily obtained . for physical solutions , @xmath87 must be non - negative and also satisfy inequality ( [ restrt ] ) . + as an example of illustration , we study the specific @xmath11 case inequality ( [ restrt ] ) being satisfied automatically . not surprisingly , the aligned cases and the logarithmic spiral cases parallel with each other very well , because of a mere change of wave propagation direction . for the special spiral case of @xmath188 , there again exists the same critical @xmath347 for @xmath283 value as in the aligned @xmath188 case ; for @xmath353 , there is one critical point of @xmath243 where @xmath354 diverges . more specifically , we have unphysical @xmath355 for @xmath356 , while we have physical @xmath357 for @xmath247 . there is no divergent point for @xmath87 solution branch and we always have @xmath358 ( @xmath359 when @xmath174 ) . the solution structures become simpler for @xmath186 cases when both @xmath87 branches of solutions are non - negative and there is no divergent point of @xmath354 . moreover , @xmath177 remains always to be the upper branch for stationary spiral fast mhd density waves and is the hydrodynamic counterpart for those in the unmagnetized single disc case , while @xmath178 remains always to be the lower branch for stationary spiral slow mhd density waves ( fan & lou 1996 ) and is caused by the very presence of the magnetic field ( detailed computations reveal @xmath178 must be smaller than @xmath348 ) . + 0.2 cm 3 . phase relationships among coplanar mhd perturbation variables 0.2 cm + so far , we have constructed global stationary coplanar mhd configurations for both aligned and logarithmic spiral cases in magnetized discs with a range of rotation curves . as expected , the inclusion of azimuthal magnetic field gives rise to one additional distinct physical solution for stationary slow mhd density waves to the problem ( fan & lou 1996 ; lou & fan 1998a ; lou 2002 ; lou & zou 2004a ) ; the alfvnic modes are excluded by our restricted consideration for coplanar mhd perturbations . the two @xmath87 branches of solutions are at least mathematically reasonable and may both have applications to astrophysical disc systems such as magnetized disc galaxies and so forth . these devised mhd disc problems are highly idealized with yet precious global analytical perturbation solutions , and conceptually still belong to a subclass of more general time - dependent mhd perturbation solutions . for applications to magnetized spiral galaxies , it would be of considerable interest to discuss pertinent physical aspects and identify relevant observational diagnostics . + for physical solutions of @xmath86 , the @xmath177 branch would be applicable to mhd discs with relatively fast rotation while the @xmath178 branch would be applicable to mhd discs with relatively slow rotation in general . for spatial phase relationships , the two branches of @xmath86 solutions give different results . we take the specific @xmath11 case as an example of illustration . for corresponding aligned cases , the plus - solution @xmath177 will lead to @xmath315 and the minus - solution @xmath178 will lead to @xmath316 ; there are no azimuthal magnetic field perturbations for aligned @xmath11 cases . for tightly wound logarithmic spiral cases , perturbations of both radial and azimuthal magnetic field are approximately in phase with the perturbation enhancement of the surface mass density for the @xmath177 branch , while they are approximately out of phase for the @xmath178 branch . for open logarithmic spirals , the radial and azimuthal magnetic field perturbations are either ahead of or lagging behind the enhancement of the surface mass by a significant phase difference . these phase relationships together with those associated with flow perturbations ( visser 1980a , b ) are valuable to interpret the spatial phase shifts between optical arms and magnetic arms in lopsided , barred or other spiral galaxies ( mathewson et al . 1972 ; neininger 1992 ; beck & hoernes 1996 ; beck et al . 1996 ; fan & lou 1996 ; lou & fan 1998 , 2000 , 2002 ; frick et al . 2000 , 2001 ; lou et al . + 0.2 cm 4 . angular momentum transfer in steady logarithmic spirals for coplanar mhd perturbation configurations 0.2 cm + stationary logarithmic spiral configurations of mhd density waves carry constant angular momentum flux either outward or inward associated with the advective transport , the gravity torque and the magnetic torque ( lynden - bell & kalnajs 1972 ; goldreich & tremaine 1978 ; fan & lou 1999 ) . therefore , the net angular momentum flux in the entire disc system is conserved and there must be either a source or a sink at the disc center . for the @xmath177 branch with @xmath86 , we find that the total angular momentum flux is inward for leading ( i.e. , @xmath223 ) spiral mhd density waves and is outward for trailing ( i.e. , @xmath224 ) spiral mhd density waves . for the @xmath178 branch in comparison , the total angular momentum flux is outward for leading ( i.e. , @xmath223 ) spiral mhd density waves and is inward for trailing ( i.e. , @xmath224 ) spiral mhd density waves . as a real mhd disc system typically rotates at a relatively large @xmath87 and inequality ( [ restrt ] ) can also rule out small @xmath87 for @xmath360 , the lower @xmath178 branch might be rarely useful in galactic applications . in either case , however , the net mass accretion rate is zero since if we focus on materials between two concentric circles , at any time the influent angular momentum from the inner ring is equal to the effluent angular momentum from the outer ring such that materials inside this belt neither gain nor lose any angular momentum and therefore no net mass accretion occurs . this research has been supported in part by the asci center for astrophysical thermonuclear flashes at the university of chicago under department of energy contract b341495 , by the special funds for major state basic science research projects of china , by the tsinghua center for astrophysics , by the collaborative research fund from the national natural science foundation of china ( nsfc ) for young outstanding overseas chinese scholars ( nsfc 10028306 ) at the national astronomical observatory , chinese academy of sciences , by nsfc grant 10373009 at the tsinghua university , and by the yangtze endowment from the ministry of education through the tsinghua university . affiliated institutions of y.q.l . share this contribution . 9 balbus s. a. , hawley j. f. , 1998 , rev . , 70 , 1 beck r. , brandenburg a. , moss d. , shukurov a. , sokoloff d. , 1996 , ara&a , 34 , 155 beck r. , hoernes p. , 1996 , nat , 379 , 47 binney j. , tremaine s. , 1987 , galactic dynamics . princeton university press , princeton , new jersey caunt s. e. , tagger m. , 2001 , a&a , 367 , 1095 chandrasekhar s. , 1961 , hydrodynamic and hydromagnetic stability . dover , new york elmegreen b. g. , 1995 , mnras , 275 , 944 evans n. w. , read j. c. a. , 1998a , mnras , 300 , 83 evans n. w. , read j. c. a. , 1998b , mnras , 300 , 106 fan z. h. , lou y .- q . , 1996 , nat , 383 , 800 fan z. h. , lou y .- q . , 1999 , mnras , 307 , 645 ferrire k. m. , 2001 , rev . mod . phys . , 73 , 1031 frick p. , beck r. , shukurov a. , sokoloff d. , ehle m. , kamphuis j. , 2000 , mnras , 318 , 925 frick p. , beck r. , berkhuijsen e. m. , patrickeyev i. , 2001 , mnras , 327 , 1145 goodman j. , evans n. w. , 1999 , mnras , 309 , 599 goldreich p. , lynden - bell d. , 1965 , mnras , 130 , 125 goldreich p. , tremaine s. , 1978 , icarus , 34 , 240 jog c. j. , 1996 , mnras , 278 , 209 julian w. h. , toomre a. , 1966 , apj , 146 , 810 kalnajs a. j. , 1971 , apj , 166 , 275 kim w .- , ostriker e. c. , 2000 , apj , 540 , 372 kronberg p. p. , 1994 , reports on progress in physics , 57 , 325 lemos j. p. s. , kalnajs a. j. , lynden - bell d. , 1991 , apj , 375 , 484 lin c. c. , shu f. h. , 1964 , apj , 140 , 646 lou y .- q . , 2002 , mnras , 337 , 225 lou y .- q . , fan z. h. , 1998a , apj , 493 , 102 lou y .- q . , fan z. h. , 1998b , mnras , 297 , 84 lou y .- q . , fan z. h. , 2000 , mnras , 315 , 646 lou y .- q . , fan z. h. , 2002 , mnras , 329 , l62 lou y.q . , walsh w. m. , han j. l. , fan z. h. , 2002 , apj , 567 , 289 lou y. q. , yuan c. , fan z. h. , 2001 , apj , 552 , 189 lou y. q. , yuan c. , fan z. h. , leon s. , 2001 , apj , 553 , l35 lou y .- q . , shen y. , 2003 , mnras , 343 , 750 ( astro - ph/0304270 ) lou y .- q . , wu y. , 2004 , mnras , submitted lou y .- q . , zou y. , 2004a , mnras , 350 , 1220 ( astro - ph/0312082 ) lou y .- q . , zou y. , 2004b , mnras , submitted low b. c. , lou y .- q . , 1990 , apj , 352 , 343 lynden - bell d. , kalnajs a. j. , 1972 , mnras , 157 , 1 lynden - bell d. , lemos j. p. s. , 1993 , preprint ( astro - ph/9907093 ) mathewson d. s. , van der kruit p. c. , brouw w. n. , 1972 , a&a , 17,468 mestel l. , 1963 , mnras , 126 , 553 neininger n. , 1992 , a&a , 263 , 30 qian e. , 1992 , mnras , 257 , 581 rdiger g. , kitchatinov l. l. , 2000 , astro . nachr . , 3 , 181 safronov v. s. , 1960 , ann . , 23 , 979 shakura n. i. , sunyaev r. a. , 1973 , a&a , 24 , 337 shen y. , lou y .- q . , 2003 , mnras , 345 , 1340 ( astro - ph/0308063 ) shen y. , lou y .- q . , 2004a , chjaa , in press ( astro - ph/0404190 ) shen y. , lou y .- q . , 2004b , mnras , 353 , 249 ( astro - ph/0405444 ) shu f. h. , laughlin g. , lizano s. , galli d. , 2000 , apj , 535 , 190 shu f. h. , li z .- y . , 1997 , apj , 475 , 251 sofue y. , fujimoto m. , wielebinski r. , 1986 , ara&a , 24 , 459 spruit h. c. , 1987 , a&a , 184 , 173 syer d. , tremaine s. , 1996 , mnras , 281 , 925 tagger m. , pellat r. , 1999 , a&a , 349 , 1003 toomre a. , 1964 , apj , 139 , 1217 toomre a. , 1977 , ara&a , 15 , 437 visser h. c. d. , 1980a , a&a , 88 , 149 visser h. c. d. , 1980b , a&a , 88 , 159 widrow l. m. , 2002 , rev . phys . , 74 , 775 zang t. a. , 1976 , phd thesis , mit , cambridge ma for @xmath127 , we first examine the solution criterion of quadratic equation ( [ aligned ] ) for the aligned case . the determinant @xmath184 of quadratic equation ( [ aligned ] ) is @xmath361 where the three coefficients @xmath362 ( @xmath363 ) are defined explicitly by @xmath364 ^ 2\\ & \qquad\qquad\qquad\qquad\qquad + \frac{(1 - 4\beta)}{(1 + 2\beta ) } \bigg[\frac{-2m^2+(1 - 4\beta)(1 - 2\beta)}{(1 + 2\beta)}{f } { \cal c}{\cal p}_m+1 - 4\beta\bigg]{\cal h}_m\ , \\ c_1\equiv\ & 2(1-{f}{\cal c}{\cal p}_m)\bigg\{{\cal a}_m \bigg[\frac{(3 - 4\beta)m^2-(1 - 4\beta)(1 - 6\beta+4\beta^2)}{2(1 + 2\beta ) } { f}{\cal c}{\cal p}_m+m^2-(1 - 4\beta)(2-\beta)\bigg]\\ & \qquad\qquad\qquad\qquad\qquad + \frac{-2m^2+(1 - 4\beta)(1 - 2\beta)}{(1 + 2\beta)}{\cal h}_m\bigg\}\ , \\ c_0\equiv\ & { \cal a}_m^2(1-{f}{\cal c}{\cal p}_m)^2\ , \end{split}\ ] ] respectively . for the quadratic expression ( [ a01 ] ) in terms of @xmath81 , we further introduce another determinant @xmath365 such that @xmath366 [ ( 1 + 8\beta)m^2-(1 - 4\beta)(1 + 4\beta-8\beta^2)]\ .\ ] ] the determinant @xmath184 of quadratic equation ( [ aligned ] ) in terms of @xmath87 can be either positive or negative in general for different parameter combinations of @xmath8 , @xmath109 , @xmath77 and @xmath283 . it is fairly straightforward to obtain @xmath87 solutions by assigning values of all relevant parameters . in particular , for one important case of @xmath19 ( i.e. , the msid case with a flat rotation curve and an isothermal equation of state ) , the condition that @xmath172 for @xmath127 and @xmath367 when @xmath188 holds for all values of @xmath81 in a full msid with @xmath344 and the condition that @xmath352 for @xmath127 holds in a partial msid with @xmath343 ( lou 2002 ) . for the magnetic force - free case of @xmath11 with a declining rotation curve with increasing @xmath14 and a polytropic index @xmath182 , we have @xmath352 for @xmath127 and an arbitrary @xmath368 in both full and partial magnetized discs . we now proceed to prove this statement rigorously . for @xmath11 of the aligned case , we explicitly have @xmath369 ^ 2\ , \qquad c_1\equiv 2(1-{f}{\cal c}{\cal p}_m)m^2 [ { \cal a}_m(2{f}{\cal c}{\cal p}_m/3 + 1)-4{\cal h}_m/3]\ , \qquad c_0\equiv { \cal a}_m^2(1-{f}{\cal c}{\cal p}_m)^2\ , \ ] ] where @xmath209 and @xmath370 for @xmath11 and @xmath365 introduced above simply reduces to @xmath371 it is apparent that for @xmath372 , we have @xmath373 . as @xmath374 , it follows immediately that @xmath352 in definition ( [ a01 ] ) . on the other hand for @xmath375 , we have @xmath376 and therefore @xmath352 by definition ( [ a01 ] ) . for @xmath11 in the logarithmic spiral case , we again have the determinant @xmath352 for quadratic equation ( [ spiral ] ) , where in parallel with definitions ( [ a01 ] ) and ( [ a02 ] ) of the aligned case , the relevant parameters in the spiral case are @xmath377 ^ 2\ , \ c_1=2(1-{f}{\cal c } { \cal n}_m)(m^2+\xi^2)\bigg[{\cal a}_m\bigg(\frac{2}{3}{f } { \cal c}{\cal n}_m+1\bigg)-\frac{4}{3}{\cal h}_m\bigg]\ , \ c_0={\cal a}_m^2(1-{f}{\cal c}{\cal n}_m)^2\ , \ ] ] with @xmath378 and @xmath370 for @xmath11 . and @xmath365 simply reduces to @xmath379 hence , as the aligned @xmath11 case , it is apparent that for @xmath372 , we have @xmath373 . as @xmath374 , it follows immediately that @xmath352 . on the other hand for @xmath380 , we have @xmath376 and therefore @xmath352 . in this appendix , we first provide a proof that for the aligned case of @xmath11 , the physical @xmath87 solutions of the plus @xmath357 and the minus @xmath381 determined from quadratic equation ( [ q2aligned ] ) correspond to the factor @xmath382 being positive and negative , respectively . according to appendix a , the determinant @xmath184 for both aligned and spiral cases remains always positive for the case of @xmath11 . we start from equation ( [ q2aligned ] ) by casting it in the following compact form of @xmath383 where @xmath166 and in reference to equation ( [ q2aligned ] ) , @xmath384 , @xmath385 and @xmath386 are constants defined explicitly by @xmath387 } \ .\ ] ] it is apparent that @xmath385 is always positive . while @xmath384 and @xmath386 remain always positive for @xmath186 , their signs depend on the value of @xmath283 for the special case of @xmath188 . in fact , both @xmath384 and @xmath386 diverge at a critical value of @xmath283 for the @xmath188 case as already discussed in subsection 3.1.3 . let us first analyze the simpler cases of @xmath186 . as @xmath388 and @xmath389 , we have two @xmath87 solutions @xmath390^{1/2}}{2 } & > \frac{\wp_2+\wp_3+|\wp_2-\wp_3|}{2}\ge \wp_2\ ; \\ y_2=\frac{\wp_1+\wp_3-[(\wp_1+\wp_3)^2 - 4\wp_2\wp_3]^{1/2}}{2 } & = \frac{\wp_1+\wp_3-[(\wp_1+\wp_3 - 2\wp_2)^2 + 4\wp_2(\wp_1-\wp_2)]^{1/2}}{2}\\ & < \frac{\wp_1+\wp_3-|\wp_1+\wp_3 - 2\wp_2|}{2}\le \wp_2\ . \end{split}\ ] ] we next examine the @xmath188 case . the situation becomes somewhat involved since @xmath391 , @xmath392 when @xmath291 where the critical @xmath283 value @xmath393\simeq 0.6842 $ ] is determined from the condition @xmath292 ( see subsection 3.1.3 ) . on the other hand , we have @xmath394 when @xmath395 in general . for the aligned case and @xmath178 are consistent with the general solution form ( [ solalign ] ) for the aligned case . ] and for @xmath291 , we therefore have two @xmath87 solutions satisfying inequalities @xmath396^{1/2}}{2}<0\ ; \\ 0<y_2&=\frac{\wp_1+\wp_3+[(\wp_1+\wp_3)^2 - 4\wp_2\wp_3]^{1/2}}{2 } \\ & = \frac{\wp_1+\wp_3+[(\wp_1+\wp_3 - 2\wp_2)^2 + 4\wp_2(\wp_1-\wp_2)]^{1/2}}{2}\\ & < \frac{\wp_1+\wp_3+|\wp_1+\wp_3 - 2\wp_2|}{2}=\wp_2\ , \end{split}\ ] ] and for @xmath395 the situation remains the same as ( [ b2 ] ) . for the spiral case of @xmath11 in parallel , we can repeat this same procedure of analysis . as before , we cast quadratic equation ( [ beta0d25s ] ) in the form of @xmath397 } \\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad = \frac{d^2\ { d^2-(1-{f}{\cal c}{\cal n}_m)/ [ { f}{\cal c}{\cal n}_m+3(m^2 - 3/2)/(2m^2 + 2\xi^2 + 3/2)]\ } } { d^2-(1-{f}{\cal c}{\cal n}_m)/ ( { f}{\cal c}{\cal n}_m+3/2)}\ , \end{split}\ ] ] which yields two distinct @xmath87 solutions according to the analysis in subsection 3.2.3 . the following treatment is identical to that for the aligned @xmath11 case described in the first part of this appendix . we here provide a proof to show that the angular momentum flux associated with stationary coplanar mhd perturbations due to the gravity torque @xmath327 , the advective transport @xmath325 and the magnetic torque @xmath329 remains constant at all radii for the entire range @xmath10 . we start from expression ( [ rmphi ] ) for @xmath398 to derive the following two expressions in a straightforward manner : @xmath399j_m(k^{\prime}r ) \int_0^{\infty}r_a^{\prime\ -1/2}\cos(m\theta-\xi\ln r_a^{\prime})j_m(k^{\prime}r_a^{\prime})dr_a^{\prime}\ . \end{split}\ ] ] then in integral ( [ gravangm ] ) , we first integrate the kernel over @xmath400 to yield @xmath401dkdk^{\prime}\ , \ ] ] where @xmath402dr_adr_a^{\prime}\ .\ ] ] we next integrate over the vertical coordinate @xmath403 to yield the final result @xmath404dkdk^{\prime}\ .\ ] ] by introducing the following integral transformation @xmath405 we reduce expression ( [ final ] ) to @xmath406dxdx^{\prime}\ , \ ] ] where @xmath407dr_bdr_b^{\prime}={\cal n}_mx^{-1/2}x^{\prime\ -1/2}\sin\bigg[\xi\ln\bigg(\frac{x}{x^{\prime}}\bigg)\bigg]\ , \ ] ] where @xmath297 is the kalnajs function ( kalnajs 1971 ) . with these technical preparations , we finally determine the angular momentum flux associated with coplanar stationary mhd perturbations due to the gravity torque as @xmath408 where @xmath409dxdx^{\prime}\ , \ ] ] which is clearly independent of radius @xmath14 . in the tight - winding regime or the wkbj limit ( i.e. , @xmath410 ) , we may directly apply the result of fan & lou ( 1999 , see their eqn . 4.13 ) by noting @xmath411 such that @xmath412 where @xmath413 has been defined by equation ( [ thetas ] ) ; here , @xmath414 is inward for leading ( @xmath223 ) spiral waves and outward for trailing ( @xmath224 ) spiral waves . in general , @xmath415 can be calculated via numerical integrations and the specific result is displayed in fig . c1 . for fairly open mhd density waves ( @xmath416 ) with @xmath417 , the angular momentum flux associated with the gravity torque is therefore inward for leading ( @xmath223 ) spiral waves and outward for trailing ( @xmath224 ) spiral waves ; with increasing @xmath418 , @xmath415 falls off rapidly and oscillates around zero with the envelope profile closely fitting the wkbj result better and better . versus @xmath219 , shown in heavy solid line . we plot @xmath419 in light dashed line for comparison . for increasing @xmath418 , @xmath415 oscillates around zero and its envelope profile fits the @xmath419 curve with an increasing accuracy . ] meanwhile , we compute the other two perturbation contributions to the angular momentum flux , namely , those from the advective transfer and the magnetic torque . for the part of advective transport , we start from equation ( [ adveangm ] ) and make use of phase relationship ( [ relsurjzspiral ] ) to explicitly derive @xmath420r^{-1/2}\cos(\xi\ln r - m\theta)\\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad\ -(\omega\sigma/\sigma_0)[(1/2+\beta)\im(\theta_s ) -\xi\re(\theta_s)]r^{-1/2}\sin(\xi\ln r - m\theta)\ . \end{split}\ ] ] it immediately follows that the advective torque is given by @xmath421=m\pi^2g\sigma^2\xi^a\ , \ ] ] where @xmath422 } { { f}{\cal c}[1+d^2-(1 - 4\beta)q^2/(2 + 4\beta)]}\ , \ ] ] that is independent of radius @xmath14 . by specifying parameters @xmath8 , @xmath109 , @xmath283 , @xmath77 and @xmath219 , one finds that for the @xmath177 branch @xmath423 for @xmath223 and @xmath424 for @xmath224 . therefore , the advective angular momentum flux is radially inward for leading stationary spiral density waves and radially outward for trailing stationary spiral density waves for the @xmath177 branch of @xmath87 solutions . for the @xmath178 branch in comparison , the situation switches and the advective angular momentum flux is outward for leading stationary spiral density waves and inward for trailing stationary spiral density waves . practically , to achieve the @xmath178 branch of stationary logarithmic spiral configurations requires a much lower @xmath87 which would be unusual for real magnetized disc systems . an example of illustration for @xmath190 , @xmath344 and @xmath11 is shown in panel ( a ) of fig . c2 . for the momentum flux associated with the magnetic torque , we start from equation ( [ magnangm ] ) and make use of phase relationship ( [ relsurjzspiral ] ) to first derive @xmath425 \ r^{-3/2}\sin(\xi\ln r - m\theta ) + [ mb_0\sigma\im(\theta_s)/\sigma_0]\ r^{-3/2 } \cos(\xi\ln r - m\theta)\ , \\ & \re(b_\theta)=[\xi b_0\sigma\re(\theta_s ) /\sigma_0]\ r^{-3/2}\sin(\xi\ln r - m\theta ) + [ \xi b_0\sigma\im(\theta_s)/\sigma_0 ] \ r^{-3/2}\cos(\xi\ln r - m\theta)\ . ] it immediately follows that the momentum flux @xmath329 associated with the magnetic torque is given by @xmath426 where @xmath427 } \ , \ ] ] which is again independent of radius @xmath14 ; here , @xmath329 is inward for leading spiral density waves ( i.e. , @xmath223 ) and outward for trailing spiral density waves ( i.e. , @xmath224 ) . an example of illustration for @xmath190 , @xmath344 and @xmath11 is shown in panel ( b ) of fig . c2 . by comparing the three contributions to the total angular momentum flux transport , we reach the following conclusions . first , the contribution from the advective transport part tends to be more dominant over those from the magnetic torque and the gravity torque . secondly , for the @xmath177 branch for stationary fast mhd density waves that has a counterpart in the unmagnetized case , leading spiral density waves ( @xmath223 ) transport angular momentum inward and trailing spiral density waves ( @xmath224 ) transport angular momentum outward . thirdly , for the @xmath178 branch for stationary slow mhd density waves caused by the very presence of the azimuthal magnetic field , leading spiral density waves ( @xmath223 ) carry angular momentum outward and trailing spiral density waves ( @xmath224 ) carry angular momentum inward . note that all these results are only valid for stationary logarithmic spiral mhd density waves . there might be some reasons to rule out the @xmath178 branch in galactic contexts . for example , the @xmath178 branch is of lower @xmath87 which can not be sustained in real galactic discs . furthermore , for rotation curve parameter @xmath360 , the @xmath178 branch may be ruled out by inequality ( [ restrt ] ) . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: at the end of their life , stars of low- and intermediate - mass evolve along the asymptotic giant branch , where they experience recurrent thermal instabilities and substantial mass loss . during this phase , the stars undergo a very rich and unique nucleosynthesis . moreover , recurrent dredge - up events enrich the stellar surface with the freshly synthesized nuclides which are then ejected into the interstellar medium through the strong winds . the thermally pulsing agb stars thus play a crucial role in the chemical evolution of galaxies . on the other hand , agb stars are important contributors to the integrated luminosity of stellar systems in a wide range of ages . a coverage of this phase is needed in particular for the treatment of the luminosity functions and color magnitude diagrams of star clusters . last but not least , the corresponding stellar models are basic tools to study carbon stars , oh / ir stars , planetary nebulae and their central objects ( white dwarfs ) . in the last 15 years , extensive observational data has become available that changed and highly constrained the theoretical view of the agb evolution . most of the progresses came from magellanic cloud studies , which revealed for example the absence of very luminous carbon stars ( which were previously predicted to originate from stars with initial masses between 5 and @xmath5 ) , and the unexpected existence of relatively faint and lower mass carbon stars ( blanco et al . 1980 , mould & aaronson 1982 , 1986 ; wood et al . 1983 ; reid & mould 1984 ; aaronson & mould 1985 ; smith & lambert 1989 ) . this so - called `` carbon star mystery '' ( iben 1981 ) has led to the concept of hot - bottom convective envelope burning ( iben 1975 ; sackmann et al . 1974 ; scalo et al . 1975 ; sackmann & boothroyd 1992 ) . this mechanism , which reconverts c - rich to o - rich envelopes and produces @xmath2 , was confirmed by the existence of lithium enriched agb stars ( smith & lambert 1989 , 1990b ; plez et al . 1993 ; smith et al . 1995 ) . on the other hand , the discovery , inside meteorites , of interstellar grains which have been formed in the wind of cool carbon stars ( see e.g. zinner et al . 1991a ) , complemented the stellar abundance analysis , and brought crucial informations on agb star nucleosynthesis . in spite of important improvements , many important issues of agb evolution still remain to be addressed , observationally as well as theoretically . this paper is the first of a series aimed at examining all the physical processes that may induce abundance anomalies in the surface of evolved agb stars . in the present work , we specifically investigate the case of intermediate - mass stars . stars with initial masses between 3 and @xmath0 and with metallicities @xmath6 of 0.02 and 0.005 have been evolved , starting from the pre - main sequence until they had gone through a number of helium - shell flashes along the tp - agb . the @xmath0 ( 6 and @xmath0 ) stars with @xmath7 ( 0.005 ) ignite off - center carbon burning and will not be discussed in detail . for each other stellar mass , full evolutionary models were computed up to the fourth `` full amplitude '' thermal pulse in the so - called `` asymptotic regime '' . synthetic evolution was then used to follow global surface properties and nucleosynthesis up to the planetary nebula ejection . this self - consistent and extended set of models is particularly relevant for population synthesis and chemical evolution models . very complete figures and tables containing informations about the center and surface evolution of the structure and chemical composition for all our models are available . in the present computations , which include the latest input physics , the evolution of the chemical composition is followed by only taking into account nuclear reactions and mixing inside convective zones ( we assume instantaneous mixing , except when hot - bottom burning occurs , which requires a time - dependent convective diffusion algorithm ) . we neglect all other mixing mechanism of non - convective origin which may contribute to transport chemical elements in radiative regions , like rotation - induced mixing or diffusion . in the stellar mass range we consider , this approximation is supported by the observed surface abundances at different evolutionary stages . this is however certainly not the case for lower mass stars , for which some `` non - standard '' mixing processes have to be invoked to explain various chemical anomalies . the physical ingredients of our models and numerical aspects of the computations are presented in sects . 2 and 3 , respectively . the general properties of the stars and their chemical evolution during the phases prior to the tp - agb one are summarized in sects . 4 and 5 , respectively . in sect . 6 , we briefly present the structural features of the thermal pulses and discuss the occurrence of the third dredge - up in our models . we derive a core - mass luminosity relation for our less massive stars that do not undergo hot - bottom burning . we then present the extrapolation procedure that we use to follow both the evolution and the nucleosynthesis in the asymptotic regime " up to the tp - agb tip . we also give the initial mass final mass relation resulting from our models , that appears to be in good agreement with the observed one . in sect . 7 , we review the various nucleosynthesis sites inside tp - agb stars , namely the burning shells ( he , @xmath8 and h ) , the convective tongue that develops during a thermal runaway and the base of hot convective envelopes . we investigate in detail the nucleosynthetic processes that involve the li , cno , f , nena and mgal elements and neutrons . we then discuss which specific nuclear region , inside a tp - agb star , mainly contributes to the surface abundance change of each nuclide when a third dredge - up occurs . in sect . 8 , we finally present our predictions concerning _ ( i ) _ the evolution of all the surface isotopic ratios by including our evolutionary models and synthetic extrapolations up to the convective envelope removal and _ ( ii ) _ the chemical yields ejected at different ages for all the species up to @xmath9 . as in the previous sections , comparisons between our predictions and observations are attentively performed . in this paper version , sects . 2 , 3 , 5 and 8 have been somewhat reduced . our equation of state is analytic . it includes * the ionization of h , he , c , n and o. complete ionization is assumed when oix / oviii @xmath10 ; * the electrostatic corrections through the statistical model of debye - hckel for partially ionized regions and through the interpolation formalism of koester ( 1976 ) , between the debye - hckel and the thomas - fermi models , for the completely ionized matter ; * the electronic degeneracy as soon as its contribution to the total pressure exceeds @xmath11 ; * the production of neutrino pairs through four different processes , namely the photo - neutrinos , plasma neutrinos , @xmath12 pair annihilations and bremsstrahlung neutrinos . energy loss contributions comes from munakata et al . ( 1985 ) , itoh et al . ( 1992 ) and kohyama et al . ( 1993 ) . at low temperatures , i.e. below 8000 k , we use the alexander & ferguson ( 1994 ) opacity tables , very well suited for cool red giant envelopes and atmospheres . at temperatures above 8000 k , we use the opal opacity tables computed by rogers & iglesias ( 1992 ) . the abundance distribution of the heavy elements being scaled from the solar composition given by anders & grevesse ( 1989 ) and updated by grevesse ( 1991 ) , as for low - temperature opacities . for both sets of opacity tables , we first interpolate bi - linearly in @xmath13 and @xmath14 for each table the composition of which is close to the model composition and then linearly in @xmath15 , @xmath6 ( and @xmath16o@xmath17 ) to obtain the right opacity coefficient corresponding to each shell . finally , we used the hubbard & lampe ( 1969 ) program to generate conductive opacity tables corresponding to the same chemical compositions and @xmath18 grid points as for the opal tables . the stellar structure is integrated from the center to a very low optical depth ( @xmath19 ) in the atmosphere . we constraint the atmospheric temperature profile to correspond to those coming from realistic atmosphere models obtained by integrating the radiative transfer equation . more specifically , we use atmosphere models _ ( 1 ) _ from plez ( 1992 ) for @xmath20@xmath21 k , _ ( 2 ) _ from eriksson ( 1994 , private communication , using a physics similar to bell et al . 1976 ) up to 5500 k and _ ( 3 ) _ computed with the kurucz atmosphere program above 5500 k. at each time step , we linearly interpolate a @xmath22 profile corresponding to our model @xmath20 , @xmath23 and @xmath6 from these grids of atmosphere models and then correspondingly modify the radiative pressure and gradient . the structure of the convective regions is computed using the classical mixing - length theory ( mlt ) , as prescribed by kippenhahn et al . our models are standard in the sense that _ ( 1 ) _ the schwarzschild criterion is considered to delimit the convective zones and _ ( 2 ) _ neither overshooting nor semi - convection have been considered in the present computations . the ratio @xmath24 of the mixing - length free parameter over the pressure scale height has been put to a value of 1.5 , i.e. rather close to @xmath25 with which we fit the solar structure . > from the main sequence up to the thermally pulsing agb phase , we used the empirical reimers ( 1975 ) formula @xmath26 with @xmath27 up to the central he exhaustion . > from the beginning of the agb phase , we adopted variable @xmath28 values , depending on the total mass of the star . more specifically , we took @xmath29 , 3 , 3.5 or 4 for our 3 , 4 , 5 or @xmath30 models respectively , whatever @xmath6 . this @xmath28 increase with the total mass was found by bryan et al . ( 1990 ) to give rather good results along the agb . however , later during the thermally pulsing phase , blcker ( 1995 ) recently showed that the bryan et al . ( 1990 ) formula underestimates the observed mass loss rate . this leads us to further increase @xmath28 ( see sect . however , as mentioned by vassiliadis & wood ( 1993 ) , whatever empirical formula , the mass loss rate increase used in the models along the thermally pulsing agb phase still remain uncertain by more than a factor of two . as demonstrated in sect . 8 , this important uncertainty on the mass loss rate formula conditions the agb phase duration and consequently , the surface composition evolution through the successive third dredge - up events . we note that along the thermally pulsing agb phase , the thermal pulse features slightly depend on the way one estimates the rate @xmath31 of gravitational energy release or gain ( see e.g. kippenhahn & weigert 1991 ) . so , let us specify that we wrote @xmath32 where @xmath33 is the specific ( i.e. by unit mass ) internal energy . we follow the abundance evolution , through the whole stars , of 45 nuclides , namely the neutrons , all the 31 stable nuclides up to @xmath34 as well as @xmath35 , @xmath36 , @xmath37 , @xmath38 , @xmath39 , @xmath40 , @xmath41 , @xmath42 , @xmath43 , @xmath44 , @xmath45 , @xmath46 and @xmath47 . these nuclides interact through a network containing 172 nuclear ( neutron , proton and @xmath48 captures ) and decay reactions . many nuclear reaction rates are taken from caughlan & fowler ( 1988 ) . let us just mention the nuclear reactions for which we selected another rate prescription . * @xmath49 from nagai et al . ( 1991 ) , * @xmath50 from caughlan et al . ( 1985 ) , * @xmath51 from raman et al . ( 1990 ) , * @xmath52 from beer et al . ( 1991 ) , * @xmath53 from wiesher et al . ( 1990 ) , * @xmath54 from funck & langanke ( 1989 ) , * @xmath55 , @xmath56 , @xmath57 and @xmath58 from fowler et al . ( 1967 ) , * @xmath59 from brehm et al . ( 1988 ) , * @xmath60 from de oliveira ( 1995 ) , * @xmath61 from allen et al . ( 1971 ) , * @xmath62 from wagoner ( 1969 ) , * @xmath63 and @xmath64 from landr et al . ( 1990 ) , * @xmath65 from thielemann ( 1991 ) , * @xmath66 and @xmath67 from bao & kppeler ( 1987 ) , * @xmath68 from kious ( 1990 ) , * @xmath69 , @xmath70 , @xmath71 , @xmath72 , @xmath73 and @xmath74 from beer et al . ( 1991 ) , * @xmath75 and @xmath76 from hammer ( 1991 ) , * @xmath77 , @xmath78 , @xmath79 and @xmath80 from woosley et al . ( 1978 ) , * @xmath81 and @xmath82 from champagne et al . ( 1988 ) , * @xmath83 and @xmath84 from illiadis ( 1990 ) , * and @xmath85 and @xmath86 from vogelaar ( 1989 ) . on the other hand , the nuclear reaction rates for @xmath87 , @xmath88 , @xmath89 , @xmath90 , @xmath91 , @xmath92 , @xmath93 , @xmath94 as well as the 45 nuclear reactions involving si , p and s are fitted with a procedure developed by rayet ( 1993 , private communication ) let us finally mention recent improvements not yet incorporated in our network : various @xmath95 reactions rates on light nuclei ( nagai et al . 1995 ) and @xmath24 capture rates on @xmath96 by kppeler et al . ( 1994 ) . finally , the nuclear screening factors are parameterized by using the graboske et al . ( 1973 ) formalism , including weak , intermediate and strong screening cases . neutrons can be produced during the thermally pulsing agb phase . they can , in particular , be captured by nuclides heavier than @xmath34 , mainly through @xmath95 reactions . as we do not follow the abundance of such elements , we created an additional nuclide , called @xmath97 , the mass fraction of which is set to @xmath98 this sum counts in fact 253 stable nuclides up to @xmath99 . in order to have a good estimation of the free neutron abundance , we follow a method described by jorissen & arnould ( 1989 ) . if nuclear reactions occur inside a convective zone , most of them proceed in general with a longer time scale than the turn - over time scale associated to the motion of convective cells . in such a case , one usually consider the convective mixing as instantaneous . this allows to treat nucleosynthesis in one shot , by taking mass - weighted averages of the number abundances and nuclear reaction rates over the whole convective region . however , if some key nuclides nuclearly evolve more rapidly than they are mixed , we have to consider the convection transport together with the nuclear reactions , so that nucleosynthesis equations become diffusion equations of the form @xmath100 \ , + \ , \makebox{nuclear terms } \,,\ ] ] where @xmath101 , the ratio of the mass fraction over the atomic mass of each nuclide @xmath102 , and @xmath103 , the convective diffusion coefficient , with @xmath104 and @xmath105 being the mean velocity and turn - over time of the convective cells , respectively . it is very cpu time consuming to treat the nucleosynthesis and convective mixing together through such diffusion equations for each nuclide in each convective shell for all the time steps . consequently , as usual , instantaneous mixing has been assumed by default for our stellar evolution computations , except in two cases that potentially require a time - dependent treatment of the convective mixing : _ ( 1 ) _ the convective tongue associated with thermal pulses and _ ( 2 ) _ the base of the convective envelope when it is hot enough , both along the agb phase . we verified that inside thermal pulses , the elements that are the most affected by a diffusion treatment of the convective mixing are the very unstable nuclides ( like @xmath42 ) , neutrons and protons . however , this do not change very significantly the nucleosynthesis global results ( see however sect . 7.5 ) . on the contrary , the mean turn - over time scale of the convective motions is much longer inside the convective envelope than inside a thermal pulse . if nuclear burning occurs at the base of the convective envelope , some important nuclear reactions can , depending on the bottom temperature , occur faster than the convective mixing so that a diffusion treatment is required . this is in particular the case for some of the reactions involved in the @xmath2 synthesis ( see sect . 7.2 ) . we have built initial models of 3 , 4 , 5 , 6 and @xmath0 allowing us to start our computations at the beginning of the pre - main sequence phase ( along the hayashi track ) , when central temperatures are below @xmath106 k. for our @xmath1 ( 0.02 ) models , the hydrogen mass fraction has been set to @xmath107 ( 0.687 ) and that of helium to @xmath108 ( 0.293 ) . the abundances of the heavier elements have been scaled on the anders & grevesse ( 1989 ) abundance distribution . we solve the five one - dimensional hydrodynamical equations for the evolution of the stellar structure with the mass coordinate as independent variable . we first convert the five non - linear stellar structure differential equations into difference equations following the henyey s scheme by defining discrete shells in mass . we then solve the resulting linearized system of equations by the newton - raphson relaxation procedure ( see e.g. press et al . typically , 600 shells are required for a model on the main sequence phase , while along the agb , 1000 to 1300 shells are needed ( 200 to 500 more shells being added during a thermal pulse ) . on the other hand , typically 200 ( 700 ) models are necessary to model the central hydrogen ( helium ) burning phase . during the agb phase , 2500 to 4500 models are required to model the time separating two successive thermal pulses , while the thermal pulse itself requires 100 to 350 models . let us stress that these rather small numbers of mass shells and time steps have been chosen in order to save computer time and present a very large set of nucleosynthesis predictions . straniero et al . ( 1996 ) very recently showed that this is probably in the disadvantage of the occurrence of the third dredge - up along the tp - agb phase . finally , the mass @xmath109 lost during each time step @xmath110 is suppressed proportionally to the shell masses of the whole envelope above the burning regions . the nucleosynthesis equations are solved using the wagoner ( 1969 ) numerical technique , well suited for our purposes . however , if a diffusion treatment is required to simultaneously compute the time - dependent convective mixing and the nuclear burning , the diffusion eqs . ( 4 ) are then solved with the same newton - raphson algorithm . theoretical hertzsprung - russell diagrams ( hrd ) for the ensemble of our models are shown in fig . [ hrd ] , from the early pms phase up to the beginning of the first thermal pulse ( or he - shell flash ) along the agb phase . the lifetimes are given in table [ times ] for the pms , the h - burning phase ( from the zero - age - main sequence to central h - exhaustion ) , the rgb phase ( from central h - exhaustion to central he - ignition ) , the he - burning phase ( up to central he - exhaustion ) , the early ( e ) agb phase ( up to the first thermal pulse ) and the thermally pulsing ( tp ) agb phase . this last duration includes our extrapolation procedure up to the convective envelope exhaustion ( see sect . 6.3 ) . lccccccc & & + & & + & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath110 [ pms ] @xmath114&1.20@xmath115&5.88@xmath116&3.70@xmath116&2.19@xmath116&1.10@xmath115&5.82@xmath116&3.50@xmath116 + @xmath110 [ ms ] @xmath114&2.80@xmath117&1.34@xmath117&7.69@xmath115&5.09@xmath115&2.59@xmath117&1.30@xmath117&7.88@xmath115 + @xmath110 [ rg ] @xmath114&1.78@xmath115&6.39@xmath116&2.86@xmath116&1.22@xmath116&1.45@xmath115&5.17@xmath116&2.38@xmath116 + @xmath110 [ hb ] @xmath114&1.55@xmath117&5.26@xmath115&2.64@xmath115&1.08@xmath115&9.37@xmath115&4.30@xmath115&2.59@xmath115 + @xmath110 [ e - agb ] @xmath114&6.16@xmath116&2.88@xmath116&1.25@xmath116&5.98@xmath118&4.29@xmath116&1.69@xmath116&6.41@xmath118 + @xmath110 [ tp - agb ] @xmath114&1.22@xmath116&2.26@xmath118&1.33@xmath118&9.12@xmath119&9.03@xmath118&1.59@xmath118&7.97@xmath119 + the occurrence of the first thermal pulse marks the end of the e - agb phase and the beginning of the tp - agb phase . this precise time , marked by the maximum surface luminosity just preceding the first major thermal instability , will be called @xmath120 in the following . table [ t0 ] presents some characteristics of our models at @xmath120 . lccccccc & & + & & + & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath121&4.739@xmath117&2.034@xmath117&1.123@xmath117&6.624@xmath115&3.851@xmath117&1.880@xmath117&1.117@xmath117 + @xmath122 & 3921&13274&19367&25743 & 5420&14761&21479 + @xmath123&182&398&486&554&182&339&387 + @xmath124 \ , ( \rm{r_{\odot}})$ ] & 3&15&17&18 & 3&10&10 + @xmath125&2.92&3.77&4.70&5.72&2.81&3.51&4.31 + @xmath126&2.0@xmath127&1.5@xmath128&2.5@xmath128&3.6@xmath128&3.6@xmath127&1.8@xmath128&2.8@xmath128 + @xmath129&3381&3102&3088&3104&3667&3450&3550 + @xmath130 & 0.38 & -0.19 & -0.26 & -0.29 & 0.36 & -0.08 & -0.10 + @xmath131 & 0.96 & 0.87 & 0.80 & 0.71 & 0.94 & 0.79 & 0.66 + @xmath132 & 0.09 & 0.08 & 0.07 & 0.06 & 0.06 & 0.13 & 0.10 + @xmath133&20&30&37&39&20&32&39 + @xmath134&2.11@xmath117&3.48@xmath117&3.97@xmath117&4.60@xmath117&2.44@xmath117&3.91@xmath117&4.54@xmath117 + @xmath135 & 0.320 & 0.521 & 0.643 & 0.745 & 0.373 & 0.608 & 0.736 + @xmath136 \ , ( \rm{r_{\odot}})$ ] & 0.00008 & 0.00003 & 0.00003 & 0.00002 & 0.00008 & 0.00004 & 0.00003 + @xmath137 \ , ( \rm{m_{\odot}})$ ] & 0.520 & 0.732 & 0.815 & 0.894 & 0.594 & 0.814 & 0.903 + @xmath138 \ , ( \rm{m_{\odot}})$ ] & 0.08 & 0.03 & 0.02 & 0.24 & 0.04 & 0.01 & 0.23 + @xmath139 \ , \rm{(k)}$]&1.43@xmath117&1.61@xmath117&1.64@xmath117&1.68@xmath117&1.37@xmath117&1.64@xmath117&1.68@xmath117 + @xmath140 \ , \rm{(g \ , cm^{-3})}$]&13.8 & 7.7 & 5.9 & 5.9&11.2 & 5.3 & 4.6 + @xmath141 \ , ( \rm{m_{\odot}})$]&4.0@xmath142&4.7@xmath143&1.5@xmath143&6.0@xmath144&3.3@xmath142&3.8@xmath143&1.2@xmath143 + @xmath145 \ , ( \rm{r_{\odot}})$ ] & 0.00016 & 0.00006 & 0.00004 & 0.00003 & 0.00015 & 0.00006 & 0.00005 + @xmath146 \ , ( \rm{m_{\odot}})$ ] & 0.542 & 0.738 & 0.818 & 0.895 & 0.609 & 0.818 & 0.905 + @xmath147 \ , ( \rm{m_{\odot}})$]&1.2@xmath148&3.2@xmath149&1.8@xmath149&5.5@xmath148&9.7@xmath149&3.4@xmath149&2.2@xmath149 + @xmath150 \ , \rm{(k)}$]&5.07@xmath115&6.43@xmath115&7.10@xmath115&7.77@xmath115&5.75@xmath115&7.00@xmath115&7.74@xmath115 + @xmath151 \ , \rm{(g \ , cm^{-3})}$]&2.0@xmath152&1.4@xmath152&1.3@xmath152&1.1@xmath152&2.4@xmath152&1.7@xmath152&1.6@xmath152 + @xmath153 \ , ( \rm{r_{\odot}})$ ] & 0.0042 & 0.0008 & 0.0002 & 0.0001 & 0.0044 & 0.0008 & 0.0002 + @xmath154 \ , ( \rm{m_{\odot}})$ ] & 0.544 & 0.738 & 0.818 & 0.896 & 0.612 & 0.819 & 0.905 + @xmath155 \ , \rm{(k)}$]&2.68@xmath116&5.72@xmath116&1.58@xmath115&4.27@xmath115&2.71@xmath116&6.53@xmath116&2.25@xmath115 + @xmath156 \ , \rm{(g \ , cm^{-3})}$]&1.1@xmath148&4.1@xmath148&5.7@xmath152 & 0.9&1.0@xmath148&6.0@xmath148 & 0.2 + for the main sequence duration , our results for @xmath7 are in better agreement with bressan et al . ( 1993 ) predictions than with the slightly higher values given by schaller et al . the central helium burning lifetime crucially depends on the amount of overshooting . as our stellar models are computed without overshooting , we can only compare our predictions at this phase with those of the padova group calculated in the same conditions ; we obtain very similar results . the other important quantity for the agb phase is the core mass ( i.e. the mass contained up to the hbs ) at the end of the central he - burning phase . indeed , its value determines the surface luminosity ( through the core mass luminosity relation ; see e.g. boothroyd & sackmann 1988a ) , the mass loss rate , and consequently the agb phase duration . last but not least , along the tp - agb phase , the thermal pulses are stronger and the third dredge - up deeper when the core mass is higher . our @xmath111 star with @xmath7 enter the agb phase with a core mass of @xmath157 ( see table [ t0 ] ) , a value very close to that of the comparable model of boothroyd & sackmann ( 1988a ) or lattanzio ( 1986 ) , also computed without overshooting . always for a reference @xmath111 star , the tp - agb phase of the straniero et al . ( 1995 ) models begins with a core mass of @xmath158 ( 2 % lower ) . at a comparable evolution stage however , the @xmath113 tp - agb star of vassiliadis & wood ( 1993 ) , also computed without any kind of extra - mixing , has a core mass @xmath159 % higher than ours . this could be due to the somewhat lower metallicity of their models ( @xmath160 ) . in conclusion , the core mass at the beginning of the tp - agb phase , although very important for this phase , still depends on various uncertainties related to the treatment of convection during the previous evolutionary phases . during the first and second dredges - up , convective mixing and induced dilution modify the surface abundances which were unaltered until this phase ( iben 1964 ) . table [ dup12 ] presents the resulting changes for the principal elements altered by h - burning . let us emphasize the main points . a second dredge - up occurs in our @xmath161 @xmath112 models , also leading to surface abundance changes . for our @xmath161 @xmath113 models , the convective envelope penetrates so deep that it pushes down the h he discontinuity . llccccccc & & & + & & & + & & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath162&_initial_&3.22@xmath163&3.19@xmath163&3.13@xmath163&3.12@xmath163&2.71@xmath163&2.71@xmath163&2.66@xmath163 + & _ top rg_&1.55@xmath149&1.05@xmath149&8.52@xmath163&7.32@xmath163&1.71@xmath149&1.28@xmath149&1.08@xmath149 + & _ at @xmath164_&1.54@xmath149&1.03@xmath149&7.79@xmath163&6.65@xmath163&1.66@xmath149&1.13@xmath149&8.56@xmath163 + @xmath48&_initial _ & 0.2932 & 0.2932 & 0.2932 & 0.2932 & 0.2495 & 0.2495 & 0.2495 + & _ top rg _ & 0.3047 & 0.3039 & 0.2995 & 0.3017 & 0.2585 & 0.2505 & 0.2498 + & _ at @xmath164 _ & 0.3054 & 0.3052 & 0.3367 & 0.3421 & 0.2624 & 0.2629 & 0.3014 + @xmath2&_initial_&9.90@xmath165&9.90@xmath165&9.90@xmath165&9.90@xmath165&2.47@xmath165&2.47@xmath165&2.47@xmath165 + & _ top rg_&1.44@xmath166&1.44@xmath166&1.43@xmath166&1.44@xmath166&3.14@xmath167&3.51@xmath167&3.86@xmath167 + & _ at @xmath164_&1.20@xmath166&9.85@xmath167&7.39@xmath167&7.53@xmath168&1.49@xmath167&1.68@xmath167&3.59@xmath169 + @xmath170/@xmath8&_initial_&83.1&83.1&83.1&83.1&83.1&83.1&83.1 + & _ top rg_&19.3&19.0&18.8&18.5&18.7&18.4&18.3 + & _ at @xmath164_&19.1&18.6&17.8&17.5&18.1&17.5&17.4 + @xmath170/@xmath171&_initial_&2.745&2.745&2.745&2.745&2.745&2.745&2.745 + & _ top rg _ & 0.725 & 0.714 & 0.769 & 0.735 & 0.636 & 0.888&1.100 + & _ at @xmath164 _ & 0.712 & 0.684 & 0.565 & 0.538 & 0.581 & 0.610 & 0.496 + @xmath171/@xmath172&_initial _ & 253 & 253 & 253 & 253 & 253 & 253 & 253 + & _ top rg _ & 1243 & 1281 & 1210 & 1279 & 1397 & 1033 & 856 + & _ at @xmath164 _ & 1270 & 1346 & 1654 & 1774 & 1546 & 1510 & 1841 + @xmath170/@xmath173&_initial _ & 0.316 & 0.316 & 0.316 & 0.316 & 0.316 & 0.316 & 0.316 + & _ top rg _ & 0.205 & 0.207 & 0.210 & 0.211 & 0.194 & 0.208 & 0.229 + & _ at @xmath164 _ & 0.204 & 0.205 & 0.206 & 0.207 & 0.191 & 0.192 & 0.197 + @xmath173/@xmath174&_initial_&2467&2467&2467&2467&2467&2467&2467 + & _ top rg _ & 314 & 409 & 562 & 617 & 258 & 869&1814 + & _ at @xmath164 _ & 311 & 391 & 508 & 579 & 245 & 323 & 358 + @xmath173/@xmath96&_initial _ & 442 & 442 & 442 & 442 & 442 & 442 & 442 + & _ top rg _ & 589 & 582 & 577 & 577 & 616 & 573 & 522 + & _ at @xmath164 _ & 591 & 588 & 588 & 588 & 623 & 619 & 611 + @xmath162 is produced at the beginning of the t tauri phase from the initial @xmath175 burning . on the main sequence , a peak of @xmath162 appears , due to the competition of the reactions @xmath176@xmath177 and @xmath178 . during the first dredge - up , the surface abundance of @xmath162 thus increases . however , the higher the stellar mass , the lower the @xmath162 peak , and the lower the surface enhancement of @xmath162 after the first dredge - up . the first ( and second ) dredge - up phase(s ) lead to a decrease of the @xmath170/@xmath8 , @xmath170/@xmath171 , @xmath172/@xmath171 and @xmath173/@xmath174 isotopic ratios , the extent of which depends on both stellar mass and metallicity . * from an initial value equal to 83 , the @xmath170/@xmath8 ratio decreases to @xmath179 . the higher the stellar mass , the lower the final carbon isotopic ratio . + this ratio slightly further decreases after the second dredge - up . * for @xmath7 , the maximum depth of the convective envelope reaches the deepest step of the @xmath171 profile , for @xmath161 @xmath111 stars . this implies that @xmath170/@xmath171 , as well as @xmath172/@xmath171 have almost the same value in these models after the first dredge - up . on the contrary , for the @xmath1 objects , the lower the stellar mass , the less efficient the dredge - up , and consequently , the less @xmath171 is up - heaved to the surface . this explains the trend for the final @xmath180c/@xmath181n ratio versus stellar mass ( see charbonnel 1994 for more details ) . + after the second dredge - up , whatever @xmath6 , the convective envelope of @xmath161 @xmath112 stars penetrates deeper for increasing stellar mass as attested by both @xmath170/@xmath171 and @xmath172/@xmath171 ratios . * the theoretical post dredge - up value of @xmath173/@xmath174 increases with the stellar mass . moreover , since for @xmath1 the deepening convective envelope of our 4 and @xmath113 stars does not reach the regions where the on cycle operated , the post dredge - up values of @xmath173/@xmath174 are higher than for @xmath7 . + after the second dredge - up however , @xmath174 enriched matter is up - heaved in the convective envelope leading to very low @xmath173/@xmath174 . * the predicted post dredge - up value of @xmath173/@xmath96 slightly decreases with increasing stellar mass . our @xmath162 surface abundance increases are in rather good agreement with the predictions by bressan et al . ( 1993 ) . concerning the cno elements , our results are in general good agreement with the predictions by schaller et al . ( 1992 ) , bressan et al . ( 1993 ) and el eid ( 1994 ) . the only important discrepancy between theoretical predictions of the different groups for intermediate - mass stars concerns the post dredge - up @xmath173/@xmath174 ratio , which value highly depends on the adopted @xmath182 and @xmath183 reaction rates . we use the rates given by landr et al . ( 1990 ) , as do schaller et al . ( 1992 ) , and we obtain @xmath173/@xmath174 ratios very similar to those of the geneva group . on the other hand , bressan et al . ( 1993 ) use the lower @xmath174 proton capture rates given by caughlan & fowler ( 1988 ) , and obtain much lower @xmath173/@xmath174 ratios [ see also el eid ( 1994 ) for a discussion of the influence of the adopted @xmath174 destruction rate on the resulting @xmath173/@xmath174 ratio ] . in our intermediate - mass stars , the theoretical post dredge - up values of the carbon isotopic ratio are slightly lower than the observations by gilroy ( 1989 ) in galactic cluster giants . red giants in this evolutionary phase present @xmath173/@xmath174 ratios between 300 to 1000 and @xmath173/@xmath96 ratios in the range 400 to 600 ( harris et al . 1988 ; smith & lambert 1990a ) . this is in very good agreement with our predictions . in particular , our prediction is in perfect agreement with the observed value in @xmath24 uma ( harris et al . 1988 ) , which estimated mass is roughly @xmath111 . this point is in favor of high @xmath174 destruction rates . this agreement between standard models and observations confirms that no extra - mixing ( diffusion , rotation - induced mixing , ) is expected to occur in intermediate - mass stars , as already discussed in charbonnel ( 1994 , 1995 ) . very complete tables containing informations about the center and surface evolution of the structure and chemical composition for our 3 , 4 , 5 , 6 and @xmath0 models with @xmath1 and 0.02 , from the beginning of the pms phase up to the end of the e - agb phase , can be found through the internet network at the following address and path : http://gag.observ-gr.fr/liens/starevol/evol.html more specifically , we present , for each star , the evolution of * the age , mass , mass loss rate , surface luminosity , bolometric magnitude , effective temperature , total radius , surface density and gravity ; * the central temperature , density , pressure , degeneracy parameter and the mass fraction of @xmath184 , @xmath48 , @xmath170 and @xmath173 ; * the mass and radius positions of the central convective core boundary and of the bottom of the convective envelope as well as its temperature and density ; * the nuclear luminosities of the h- and he - burning shells ; * the mass and radius positions of the h- and he - burning shells ; * the surface mass fractions of @xmath184 , @xmath162 , @xmath48 , @xmath2 , @xmath185 , @xmath170 , @xmath8 , @xmath171 , @xmath172 , @xmath173 , @xmath174 , @xmath96 , @xmath3 , @xmath186 , @xmath187 , @xmath188 , @xmath4 , @xmath189 , @xmath190 , @xmath191 , @xmath43 , @xmath192 as the stars are ascending the agb , the features of the thermal pulses evolve , and some of the main characterizing quantities of these events reach asymptotic values " . the most important feature is the thermal pulse intensity that can be defined in terms of the hebs luminosity @xmath194 reached at the maximum of the he - shell flash . @xmath194 is quite low during the first thermal pulses , but it rapidly increases with core mass ( or flash number ) . the increasing rate is greater for lower metallicities or higher stellar masses . after typically 10 ( 5 ) thermal pulses in our @xmath195 ( 0.005 ) models , the thermal pulse intensity is very high and only slightly increases from pulse to pulse , monotonously with the core mass . this second part of the tp - agb phase is usually called the `` asymptotic regime '' or `` full amplitude '' of the thermal pulses . for each of the modeled stars , we have computed at least four full amplitude thermal pulses . let us briefly discuss the main evolutionary properties of these thermal runaways . the thermal runaways are stronger inside stars of lower @xmath6 , which facilitates dredge - up events ( see sect . the intensities of our he - shell flashes are comparable to other published values for stars of about the same core masses ( see e.g. boothroyd & sackmann 1988b and table 4 ) . while the stars evolve along the tp - agb , the hebs and hbs advance in mass , and the mean inter - shell mass slightly decreases with time . from one pulse to another , the top of the convective tongue at its maximum extension gets closer from the hbs , even if contact between both regions has never been found due to the fact that the hbs is still active at that time . moreover , for all the thermal runaways we have computed , the overlap in mass between two successive thermal pulses is ranged between 0.6 and 0.4 for agb stars of increasing total mass . this overlap is slightly larger for our lower @xmath6 models . the maximum ( minimum ) temperature @xmath196 ( @xmath197 ) and density @xmath198 ( @xmath199 ) reached at the base ( top ) of a thermal pulse both increase ( decrease ) from pulse to pulse . for a given he - shell flash , @xmath196 increases with stellar mass and metallicity , while @xmath198 decreases with stellar mass but increases with metallicity . in the asymptotic regime , the growth rate of @xmath196 also tends to an asymptotic value . by comparison with other works for the @xmath111 ( z = 0.02 ) star , we found that our @xmath196 are typically 4 % higher than those of boothroyd & sackmann ( 1988b ) and straniero et al . . this could be mainly due to _ ( i ) _ different prescriptions to compute @xmath31 , _ ( ii ) _ different envelope masses due to different mass loss rates and _ ( iii ) _ different mlt treatments ( see sect . the temperature and density differences between the base and top of each convective tongue are rather large . this point is important for nucleosynthesis purposes ( see sect . the duration of the convective tongue associated with each thermal pulse ( @xmath200 ) and , to a lower extent , that of the inter - pulse phase ( @xmath201 ) , are crucial quantities for the nucleosynthesis too . the stronger a he - flash , the shorter it is and the quickest it deactivates . both quantities are thus smaller for higher core masses and lower metallicities . moreover , @xmath200 tends to decrease as the stars evolve along the tp - agb . @xmath201 also decreases after the first he - shell flashes , but it then reaches an almost constant value during the asymptotic regime , only very slightly decreasing with the increasing core mass . along the tp - agb phase , the bottom of the convective envelope can penetrate deep inside regions that have been nuclearly processed just after a thermal pulse , if it is strong enough ( in order to give rise to a large enough entropy increase in the inter - shell region ) . contrarily to the first and second dredges - up , this so - called third dredge - up , when it occurs , successively mixes up to the surface _ ( i ) _ material that has experienced h - burning in the thin hbs but also _ ( ii ) _ part of the region where the thermal pulse nucleosynthesis operated just before . both regions give rise to very different and sometimes opposite chemical pollutions of the convective envelope ( see sects . 7.1 and 7.6 ) . in our @xmath7 models , the hbs region is already dredged - up from the 14th , 5th and 3rd thermal pulse for the 3 , 4 and @xmath113 stars , respectively , while for the @xmath1 models , such a mixing already occurs from the 2nd and 1st thermal pulse for the 3 and @xmath112 stars , respectively . the @xmath30 ( @xmath113 ) star with @xmath7 ( 0.005 ) continuously mixes h - burning products up to the surface as it experiences h - burning at the bottom of its convective envelope ( the so - called `` hot - bottom burning '' or hbb ; see sect this kind of dredge - up already modifies the surface composition for some species ( like @xmath43 ; see sects . 7.4 and 7.6 ) . the complete third dredge - up ( one usually simply calls the third dredge - up , 3dup hereafter ) , i.e. reaching the upper part of the previous thermal pulse region , is more difficult to obtain numerically . following models ( wood 1981 ; iben 1983 ; lattanzio 1987 ; boothroyd & sackmann 1988c ) , the dredge - up of thermal pulse material is easier for stars of lower metallicity @xmath6 . this is supported by observations ( see e.g. blanco et al . 1978 ) . in our models , we just found a 3dup from the 9th thermal pulse of the @xmath112 with @xmath1 star . however , based on the evolution of the post - thermal pulse entropy profile from pulse to pulse , we can predict that such a dredge - up will also occur , within the next two thermal pulses , for our 5 and @xmath30 stars with @xmath7 and for the @xmath113 star with @xmath1 ( see table [ endagb ] for the number of computed thermal pulses for each object ) . some additional thermal pulses are still needed for lower mass objects . several years ago , the 3dup was found to occur too late along the tp - agb phase for agb stars of decreasing mass and/or increasing metallicity ( see e.g. iben 1976 , fujimoto & sugimoto 1979 ) . this was a common problem ( see lattanzio 1989 for a review ) . recent works showed that with the new opal opacities , the 3dup can now be obtained even for low - mass pop i objects ( see e.g. frost & lattanzio 1996 or straniero et al . 1995 ) . one of the clearest signature of the 3dup is the significant @xmath170 pollution of the convective envelope , brought into the inter - shell region by the convective tongue of a thermal pulse ( see sect . as the thermal pulse intensity ( and consequently the probability of 3dup occurrence ) increases with time , as do the core mass and the total luminosity , the models tend predict the formation of carbon ( c ) stars ( for which the surface @xmath170/@xmath173 becomes greater than unity thanks to the repetitive 3dup events ) at too high luminosities , i.e. to late during the tp - agb phase , especially for solar metallicity stars . solutions to increase the depth of the convective envelope during a dredge - up , e.g. by treating in some way semi - convection ( see e.g. lattanzio 1986 ) or by invoking undershooting ( see alongi et al . 1991 ) have been suggested . they indeed help the 3dup to occur but do not yet solve completely the problem . fundamentally , this could be due to our bad knowledge of convection inside stars . very recently , straniero et al . ( 1996 ) suggested that by considerably increasing the number of mass shells and especially the number of time steps to model thermal pulses , the 3dup occurs earlier during the tp - agb phase . this consists in a real numerical progress in that field . the fact we find 3dup latter than them along the agb phase can be due to our lower number of time steps ( typically by a factor of 3 ) . it was first discovered by paczyski ( 1975 ) that during the asymptotic tp - agb phase , there is a linear relation between the surface luminosity @xmath202 and the core mass @xmath203}$ ] ( see table [ t0 ] ) which is only slightly dependent on the total mass of the agb star . blcker & schnberner ( 1991 ) however found that this kind of relation does not exist for the most massive agb stars for which the bottom of the convective envelope goes down to the hbs so that part of the h - burning occurs inside the convective envelope ( hbb ) . > from our models for the last four thermal pulses of the 3 and @xmath112 with @xmath7 and the @xmath111 with @xmath1 , we have derived a core mass luminosity relation in the form @xmath204 \,,\end{aligned}\ ] ] where @xmath205 , @xmath206 , @xmath207 , @xmath208 , @xmath209 and @xmath210 . it is in good agreement with other @xmath211 relations , for @xmath212 and @xmath213 ( see e.g. lattanzio 1986 or boothroyd & sackmann 1988a ) . the more massive stars experience hbb and do not fit , consequently , to the same relation . in table [ endagb ] , we however give a linear relation of the form @xmath214 for the @xmath113 ( @xmath112 ) agb star with @xmath7 ( 0.005 ) that do not experience a very strong hbb . let us stress that during the tp - agb phase , @xmath202 suffers very important variations during and after each thermal pulse . frequently , the core mass luminosity relation is derived by taking solely account of the maximum surface luminosity preceding each thermal pulse . the probability to observe an agb star just at that time is very weak ( @xmath215 ; see table [ endagb ] ) . we thus preferred to derive our relations with a more realistic surface luminosity that is averaged over each pulse - inter - pulse cycle . on the other hand , it is well known that the rate of mass advance of a burning shell ( here @xmath216 for the hbs ) is related to the rate of nuclear energy production by @xmath217 with @xmath218 being the luminosity produced by the hbs , @xmath219 the mean hydrogen mass fraction over this burning shell and @xmath220 the amount of energy produced by the conversion of 1 g of @xmath184 to @xmath48 through the cno bi - cycle . along the tp - agb , most of the surface luminosity @xmath202 is produced by the hbs . as indicated in table [ t0 ] , the hebs contributes much less than the hbs to the nuclear luminosity ( except during a thermal runaway ) . gravitation also constitutes a relatively small energy source , especially for the less massive agb stars . in conclusion , we can write @xmath221 with @xmath222 depending on @xmath193 . by eliminating @xmath218 in eq . ( 6 ) with the core mass luminosity relation , one finds that the core mass @xmath216 increases exponentially with time during the asymptotic tp - agb phase . the time scale for the core mass growth is @xmath223 ( @xmath224 ) yr for a @xmath111 ( @xmath30 ) agb star , that is , interestingly , comparable to the total duration of the tp - agb phase . table [ mcl ] demonstrates that eq . ( 6 ) predicts rates of hbs advance in mass that are very similar to those coming from our numerical models , except for the @xmath30 ( @xmath113 ) object with @xmath7 ( 0.005 ) , probably due to the strong hbb . lccccccc & & + & & + & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath225&9.94@xmath118&1.13@xmath118&4.56@xmath119&2.32@xmath119&4.39@xmath118&5.53@xmath119&2.13@xmath119 + @xmath226&19&13&13&14&8&9&9 + @xmath122 & 9750&18200&24150&33650 & 9250&19700&27600 + @xmath227&1.662&2.718&3.642&4.575&1.869&2.547&3.124 + @xmath228&0.6052&0.7563&0.8290&0.9022&0.6411&0.8277&0.9101 + @xmath24&see rel.&see rel.&264550&&see rel.&383550& + @xmath229&see rel.&see rel.&0.73729&&see rel.&0.77564& + @xmath230&2.1@xmath115&1.2@xmath115&5.1@xmath116&1.9@xmath116&2.4@xmath115&6.2@xmath231&1.1@xmath116 + @xmath232 & 38000 & 9200 & 3700 & 1600 & 45000 & 7500 & 2500 + @xmath233 & 95 & 24 & 14 & 9 & 104 & 20 & 13 + @xmath234 & 19 & 10 & 7 & 6 & 20 & 9 & 8 + @xmath235 & 0.0115 & 0.0038 & 0.0021 & 0.0011 & 0.0119 & 0.0031 & 0.0015 + @xmath236 & 0.0066 & 0.0019 & 0.0010 & 0.0005 & 0.0069 & 0.0017 & 0.0008 + 3dup ? & no & no & no & no & no & yes & no + lccccccc & & + & & + & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath219 & 0.49&0.54&0.52&0.64&0.54&0.58&0.69 + @xmath237 & 3450&10690&14480&17250&4600&10320&12890 + eq . ( 6 ) prediction & 6.7@xmath238&1.9@xmath127&2.7@xmath127&2.6@xmath127&8.1@xmath238&1.7@xmath127&1.8@xmath127 + @xmath239 & 6.4@xmath238&1.7@xmath127&2.5@xmath127&3.1@xmath127&7.8@xmath238&1.8@xmath127&2.4@xmath127 + one of our ambitions is to extrapolate our predictions coming from the complete evolution models not only for the global surface properties but also for the chemical composition ( 5 ) and ( 6 ) already allow to extrapolate rather accurately the core mass and surface luminosity from our last computed models towards the end of the tp - agb phase . for the @xmath30 ( @xmath113 ) star with @xmath7 ( 0.005 ) , independent fits of @xmath202 and @xmath216 as a function of time have been calculated . we also need fits for the surface radius [ in order to follow the mass loss rate increase through eq . ( 1 ) ] as well as the temperature and density at the base of the convective envelope ( to follow the hbb nucleosynthesis ) . we found , on grounds of the last four computed thermal pulses for each star , very good fitting relations for these quantities , which are presented in table [ fits ] . note that they are linear with time during the last part of the tp - agb phase , contrarily to @xmath216 and @xmath202 that increase exponentially during this phase . the main interest of such an extrapolation procedure is that very detailed stellar evolution computations ( including a very complete description of the nucleosynthesis ) would take prohibitive computation time . however , we had to make two important approximations ( compared to full evolution models ) that must be mentioned . * the successive 3dup events progressively pollute the convective envelope in @xmath48 , @xmath170 and heavier elements to the detriment of @xmath184 . this increases the mean molecular weight ( or @xmath6 ) and the mean opacity in the sub - photospheric region , probably leading to an overestimation of the surface luminosity . * very few computations are existing for the very end of the tp - agb phase ( vassiliadis & wood 1993 ; blcker 1995 ) , that can not maintain to give accurate predictions concerning the evolution of the surface isotopic composition . one of their interests is to indicate that the last few thermal pulses seem to have different properties than the full amplitude ones ( e.g. longer inter - pulse durations ) . on the other hand , it has been noted by boothroyd & sackmann ( 1988a ) that the increasing mass loss rate and @xmath6 possibly prevent the 3dup to occur after these last thermal pulses . as a consequence , assuming the full amplitude regime up to the agb tip probably makes our following predictions quite indefinite concerning the very end of the tp - agb phase . llccccccc & & & + & & & + & & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath240&@xmath24 & 3.17@xmath149&1.16@xmath148&1.76@xmath148&1.95@xmath148&9.03@xmath163&1.33@xmath148&3.48@xmath148 + & @xmath229 & 8.53@xmath241&4.11@xmath242&5.11@xmath242&6.35@xmath242&2.37@xmath242&3.57@xmath242&4.06@xmath242 + @xmath243&@xmath24 & 1.81@xmath244&2.88@xmath241&5.13@xmath242&4.21@xmath242&3.53&2.81@xmath242&1.07@xmath245 + & @xmath229 & 4.05@xmath246&1.14@xmath247&2.05@xmath247&6.34@xmath247&3.16@xmath246&8.82@xmath246&4.50@xmath247 + @xmath248&@xmath24 & 1.16@xmath166&2.51@xmath127&1.56@xmath163&3.32@xmath163&4.71@xmath249&3.86@xmath128&1.24@xmath149 + & @xmath229 & 1.72@xmath148&2.28@xmath152&5.40@xmath152&2.47&9.82@xmath149&1.04@xmath148&3.53@xmath244 + such extrapolation procedures have already been realized by groenewegen & de jong ( 1993 and following papers ) . these synthetic evolution models are very powerful to compare observations with computed models concerning the evolution of the global surface properties of agb stars . however , they contain shortcomings that do not allow to predict in detail the evolution of the surface isotopic composition . mainly , _ ( i ) _ only the most abundant elements are followed ( @xmath48 , @xmath170 , @xmath8 , @xmath171 and @xmath173 ) and _ ( ii ) _ the hbb nucleosynthesis is not followed in detail whereas it becomes a very influent process concerning the convective envelope composition towards the tp - agb tip . recent progresses have been made by busso et al . ( 1995 ) and lambert et al . ( 1995 ) to follow in detail the nucleosynthesis as late as possible along the agb phase . in sect . 8 , we both present the nucleosynthesis encountered in our full evolution models and the extension of the predictions concerning the convective envelope composition towards the end of the agb phase corresponding to the seven computed stars . for this extrapolation , we proceed as follows . 1 . from the last computed inter - pulse phase for each star , we integrate , during the successive inter - pulse phases , the evolution of the nucleosynthesis occuring inside the convective envelope by solving eqs . ( 4 ) , suited for the hbb , for each nuclide while imposing a temperature and density increases at the bottom of the convective envelope given by the fits presented in table [ fits ] . we took the inter - pulse durations of table [ endagb ] ; indeed , they are only very slightly decreasing with the core mass increase along the asymptotic tp - agb phase . 2 . since the 3dup is almost ready to occur in our models ( see sect . 6.2 ) , we mix the envelope with the hbs and inter - shell regions recurrently , every @xmath201 . to quantify the adopted depth of each 3dup , let us remember the definition of the usual parameter + @xmath250 + where @xmath251 represents the amount of mass that is dredged - up ( difference between the normal depth of the convective envelope and its maximum downwards penetration during a dredge - up ) and @xmath252 the increase in mass of the hbs between two successive thermal pulses . we adopted a 3dup depth corresponding to @xmath253 for all the stars . this value has been suggested by groenewegen & de jong ( 1993 ) because it reproduces rather well the carbon star luminosity function of the lmc ( that can not be fitted on grounds of self - consistent evolution models only ; see sect . + during such a dredge - up , we assumed that the material coming from the convective tongue of a thermal pulse has abundance profiles corresponding to end of the last computed thermal pulse . of course , we can not do better without further computing thermal pulses with detailed nucleosynthesis up to the end of the tp - agb phase that is , again , not yet realistic . however , this appears to be rather justified for many important chemical elements as they reach almost constant abundances after the few last computed thermal pulses ( see discussion in sect . 7 ) . taking a constant value for @xmath254 is probably not strictly right . as far as the surface composition is concerned , increasing @xmath254 more efficiently pollutes the convective envelope from pulse to pulse . there is another quantity that greatly influences the evolution of the envelope composition : the mass loss rate . indeed , decreasing the mass loss rate allows to increase the total number of thermal pulses ( and consequently the total number of 3dup episodes ) before the end of the tp - agb phase . we had to increase our @xmath28 values in eq . ( 1 ) for each star , in order to reach the highest mass loss rates ( called super - winds ) observed among very evolved agb stars ( see the discussion in sect . 2.1.5 ; see also zuckerman et al . 1986 ; wannier & sahai 1986 ; knapp et al . 1989 ) : we adopted @xmath255 , 10 , 15 and 20 for the 3 , 4 , 5 and @xmath30 agb models , respectively , whatever @xmath6 . as the exact way following which the mass loss rate increases with time along the asymptotic tp - agb phase is still questionable by more than a factor of two ( see sect . 2.1.5 ) , the resulting uncertainty on the evolution of the surface composition up to the agb tip dominates the effect of @xmath254 ( see our discussion in sect . 8.1 , where our mass loss rates are varied by a factor of four ) . in table [ mlos ] , we present the global mass loss increases we obtained , for the seven tp -agb stars , from the beginning of the tp - agb phase up to its end ( i.e. including our extrapolation computations ) . llccccccc & & & + & & & + & & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath126&_at @xmath164 _ & 2.0@xmath127&1.5@xmath128&2.5@xmath128&3.6@xmath128&3.6@xmath127&1.8@xmath128&2.8@xmath128 + & _ agb tip _ & 7.6@xmath128&4.3@xmath163&9.4@xmath163&1.5@xmath149&5.6@xmath128&4.0@xmath163&1.0@xmath149 + one of the most important constraint for evolutionary models towards the agb tip is the reproduction of the observed correlation between the initial mass of stars ( @xmath256 ) and their final one ( @xmath257 ) , i.e. their core mass ( become a hot white dwarf ) when the convective envelope has been completely removed after the planetary nebula ( pn ) ejection . we present in fig . [ mimf ] the comparison between observations analyzed by weidemann & koester ( 1983 ) and weidemann ( 1987 ) and the final core masses we obtained after our evolutionary plus extrapolations computations . we note that agreement with observations is slightly better than for vassiliadis & wood ( 1993 ) , especially around @xmath258 @xmath111 where they only reproduce the upper limits . on the contrary , the final core mass obtained by boothroyd & sackmann ( 1988b ) is rather low compared to observations . note we have not taken into account of the core mass reduction following each 3dup event . however , as explained below , such a reduction , even if it increases the agreement with the observed relation , is rather negligible compared to other uncertainties ( core mass at the first thermal pulse , thermal pulse number depending on the adopted mass loss rate , ) . we observe that changing the mass loss rate by a factor of four during the asymptotic tp - agb phase ( see also sect . 8.1 concerning surface abundance predictions ) does not significantly change the agreement with observations . we claim that this is due to the fact that while almost 50% of the c o core mass is built during the central he - burning phase , roughly 35 to 45% ( in mass ) are added during the e - agb phase . more precisely , during the whole tp - agb phase , we found that our core masses increased by 14 ( 12 ) , 7 ( 4 ) , 5 ( 3 ) and 4% for the 3 , 4 , 5 , and @xmath30 stars with @xmath7 ( 0.005 ) , respectively . the final core mass is consequently almost decided before the super - wind occurs and this observed relation is not very compulsive for the empirical mass loss rate to adopt during the tp - agb phase . > from the pioneer work of iben ( 1975 ) , one realized that inside tp - agb stars , a very rich nucleosynthesis occurs . during the recurrent third dredge - up events , the synthesized nuclides are mixed inside the convective envelope and then ejected into the interstellar medium ( ism ) through the strong mass loss . in conclusion , tp - agb stars play a crucial role for the chemical evolution of galaxies , especially for some elements like @xmath162 , @xmath2 , @xmath3 , @xmath259 , nuclides heavier than @xmath260 , it is the reason why we will give , in sect . 8.3 , the yields resulting from our intermediate - mass star computations . let us analyze the various nucleosynthesis processes occuring inside intermediate - mass tp - agb stars . in fact , three very different sites have to be distinguished . as indicated in table [ t0 ] , most of the nuclear energy release on the agb is due to the hbs , mainly through the cno bi - cycle . more precisely , the six most important nuclear reactions that energetically supply the star are @xmath261 ( @xmath262 ) , @xmath263 ( @xmath264 ) , @xmath265 ( @xmath266 ) , @xmath40 @xmath267 ( @xmath268 ) , @xmath39 @xmath267 ( @xmath269 ) and @xmath270 ( @xmath271 ) , all occuring inside the hbs . these numbers are quite similar ( within a few % ) for all our modeled stars . whatever the agb star total mass , the hbs slightly depletes @xmath170 and @xmath172 and significantly depletes @xmath174 , @xmath96 and @xmath3 , while @xmath171 is produced . the case of @xmath8 is more difficult : while its abundance is very slightly increased in the hbs of @xmath111 stars ( whatever @xmath6 ) , it is somewhat decreased in the hbs of @xmath30 stars . the rather high temperature inside the hbs also allows the activation of the nena and mgal chains , even if they are much less influent for the global energetics . the most interesting consequences of the nena chain are the production of @xmath188 and @xmath4 in @xmath272 @xmath113 stars and their slight destruction in more massive agb stars . the mgal chain is very sensitive to the agb total mass . in @xmath273 @xmath113 objects , @xmath190 is destroyed to the benefit of @xmath259 ( roughly half in @xmath43 and half in @xmath44 that quite instantaneously decays in @xmath191 ) ; a small quantity of @xmath191 is also destroyed in @xmath192 . in more massive agb stars , the hbs temperature is high enough to substantially produce @xmath190 through the @xmath274 reaction . this in turn leads to a much important @xmath43 production . however , as the hbs mean temperature always increases with time , @xmath43 begins to be partially destroyed by @xmath275 reactions late along the tp - agb . we refer to forestini et al . ( 1991 ) for the first detailed study of @xmath43 production inside agb stars , by radiative proton burning ( i.e. in the hbs ) . in guelin et al . ( 1995 ) , more details can be found , including the possibility to synthesize @xmath43 by proton burning at the bottom of the convective envelope ( i.e. by hbb ) . it was however a parametric study . first predictions concerning the enrichment of the agb star surfaces , on grounds of full evolutionary models are presented in sects . 7.6 and 8.1 . the comparatively small amount of nuclear energy supplied by the hebs comes from the @xmath276 ( @xmath277 ) and @xmath50 ( @xmath278 ) reactions . consequently , the hebs mainly produces @xmath170 and , to a lower extent , @xmath173 , while during the previous central he - burning stage , the @xmath173 production was preponderant . at the beginning of a thermal pulse , it is the @xmath276 reaction that becomes the dominant energy source ( see also sect . 7.1.3 for the thermal pulse energetics ) . during the long inter - pulse phase that follows a 3dup episode , the @xmath8 coming out of the hbs can be burnt in the inter - shell region , close to the hebs top , mainly through the @xmath279 reaction . this reaction is highly energetic . the first published work reporting the existence of this inter - shell @xmath8 burning is straniero et al . the produced neutrons are then captured by light species up to @xmath260 mainly , but part on them are also used to synthesize @xmath280 elements . by decreasing order of amplitude , the nuclear energy liberated in the inter - shell region during the inter - pulse phase comes from these @xmath95 reactions up to @xmath260 ( @xmath281 ) , @xmath282 ( @xmath283 ) , the so - called @xmath284 process reactions ( @xmath285 ) , @xmath286 ( @xmath269 ) and @xmath287 ( @xmath288 ) . note the important nuclear energy released by the neutron captures . this energy production creates a third burning shell between the hebs and hbs , even if its nuclear activity is considerably lower than that of both the hebs and hbs . it continues to exist until @xmath8 has been destroyed there . this happens , in general , almost when the following thermal pulse begins . as a consequence , when the convective tongue of a thermal pulse engulfs the inter - shell matter , the amount of @xmath8 that is mixed is reduced compared to what was previously thought , i.e. @xmath289 x$](@xmath8 ) @xmath290 $ ] , where @xmath290 $ ] designates , as before , the inter - shell region . we show , in fig . [ 13csh ] , the nuclear energy production rate ( @xmath291 ) profile in the case of our @xmath111 agb star . this clearly demonstrates the three - burning shell structure . such a structure is also found in the more massive agb stars . let us emphasize that this third burning shell could at least partially explain the production of the @xmath280 nuclides ( see sect . 7.5 ) in a radiative zone . this has been suggested by straniero et al . ( 1995 ) , even if their amount of @xmath8 in the inter - shell was artificially increased in order to study the @xmath284process nucleosynthesis . if the temperature @xmath292 at the base of the deep convective envelope is high enough ( i.e. @xmath293 k ) , i.e. if the base of the convective envelope reaches the hbs , proton burning occurs inside it . this hbb occurs in models with initial masses @xmath294 @xmath113 for @xmath7 and @xmath294 @xmath112 for @xmath1 . these lower mass limits and this @xmath6 dependence are common features of all the agb models ( see e.g. boothroyd et al . 1993 ) . as mixing up to the cool surface is very efficient ( the mean turn - over time @xmath105 of the convective envelope is roughly 0.5 yr ) , the effective nuclear reaction rates are lower than in radiative h - burning , leading to specific nucleosynthesis signatures . for @xmath295 k already , @xmath2 can be produced from @xmath162 following a scenario that can not operate in radiative h - burning ( see sect . 7.2 for more details ) . this li synthesis has been qualitatively suggested by cameron & fowler ( 1971 ) and quantitatively first demonstrated on grounds of evolutionary models by sackmann & boothroyd ( 1992 ) . for @xmath296 k , the cn cycle begins to operate partially , i.e. @xmath170 is converted to @xmath8 that accumulates . as first investigated by boothroyd et al . ( 1993 ) , this conversion can efficiently reduce the @xmath170/@xmath173 ratio and eventually , if hbb is strong enough , this can prevent the formation of c stars ( see also sects . 8.1 and 8.2 ) . if @xmath292 @xmath297 @xmath298 k typically , the cn cycle almost operates at equilibrium through the whole envelope , leading to a @xmath170/@xmath8 isotopic ratio lower than 10 ( depending on @xmath292 ) . in our models , this happens for the @xmath30 ( @xmath113 ) agb star with @xmath7 ( 0.005 ) and also , later along the tp - agb , for our @xmath113 star with @xmath7 ( see sect . 8.1 ) . at very high @xmath292 , other reactions begin to occur ; in particular , in extreme cases ( i.e. the end of the tp - agb phase of our @xmath30 models with @xmath7 ) , @xmath43 can eventually be directly produced inside the convective envelope through the @xmath83 reaction on @xmath190 supplied by @xmath274 ( see sect . 8.1 ) . the thermal runaway , inside the hebs , leading to form a convective tongue is principally due to the @xmath276 reaction excitement . when this convective tongue finally penetrates the inter - shell region , the hbs ashes are ingested , i.e. not only @xmath48 but also @xmath8 , @xmath171 and @xmath43 mainly . as these elements are partially burnt in the bottom of the inter - shell region during the inter - pulse phase ( globally , we find that half of the @xmath8 is so destroyed ) , they mainly survive in the upper part of the inter - shell . they are consequently engulfed by the convective tongue when it reaches its maximum extension . at that time , the @xmath276 reaction is already dominant for the energetics . we consequently find that the energetics associated with the @xmath8 burning through the @xmath282 reaction as well as the neutron captures that follow is rather negligible ( less than 1 % ) . this is also found by straniero et al . ( 1995 ) . this conclusion could however be modified if more @xmath8 was synthesized in the inter - shell region , as required for the @xmath284process ( see sect . the important energetic effects reported by bazan & lattanzio ( 1993 ) concerning the @xmath8 burning inside thermal pulses was probably due to _ ( i ) _ the fact they did not give account of partial inter - shell @xmath8 burning and _ ( ii ) _ the very high @xmath8 amount artificially ingested into their thermal pulses in order to study the consequent @xmath284process . part of produced neutrons participate to @xmath299 reactions , mainly on @xmath43 and , to a lower extent , on @xmath171 . the resulting protons , at these high temperatures characterizing he - burning during a thermal pulse ( @xmath300 k at the base of the convective tongue ) , are rapidly captured by various elements ( see sects . 7.3 to 7.5 for more details ) . let us stress that the simultaneous presence , inside a mixed he - burning region ( the convective tongue ) , of neutrons and protons leads to a very specific ( and unique ) nucleosynthesis . we will analyze in more details the nucleosynthesis processes associated with the thermal pulses in sects . 7.3 to 7.5 . sackmann & boothroyd ( 1992 ) were the first ones to quantitatively demonstrate how the cameron & fowler ( 1971 ) scenario leads to @xmath2 synthesis in intermediate - mass agb stars . this major result was obtained by using a time - dependent convective diffusion algorithm suited for hot - bottom convective envelopes . indeed , at the base of such envelopes , some nuclear reactions involved in @xmath2 production occur faster than mean turn - over time associated with the convective motions ( see fig . [ reacthp ] ) . we also use a time - dependent treatment to study the @xmath2 synthesis [ see eqs . ( 4 ) in sect . we show in fig . [ libe ] the typical resulting @xmath35 and @xmath2 profiles inside the convective envelope of our @xmath30 agb models with @xmath7 . when hbb occurs ( see sect . 7.1.2 ) , @xmath162 first begins to burn through the @xmath301 and @xmath302 reactions , the second one being significantly slower than the first one . as in deep stellar interiors , the @xmath35 destruction by electron capture and proton capture are much longer than the time scale associated with the convective mixing , the produced @xmath35 is efficiently up - heaved towards cooler regions where it progressively decays in @xmath2 . let us emphasize that without fast convective mixing , @xmath2 could not accumulate at the surface . indeed , at the base of the convective envelope , _ ( i ) _ @xmath35 would be partially destroyed by electron and proton captures and _ ( ii ) _ the @xmath2 resulting from @xmath35 decay would be quite instantaneously destroyed by proton capture too . let us now compare our predictions with sackmann & boothroyd ( 1992 ) results and with observations . a direct comparison with their figs . 2a and 2b reveals nearly identical features . in fig . [ lli ] , we present the evolution of the surface @xmath2 abundance during the tp - agb phase of all our modeled stars . in our @xmath111 models , the temperature at the base of the convective envelope is never high enough for @xmath35 to be produced efficiently , and the surface @xmath2 abundance remains almost constant at a very low value resulting from the first dredge - up . we also see that our @xmath112 star with @xmath1 begins to produce some @xmath2 when it enters the asymptotic regime , but the surface abundance of this light element always stays at least one order of magnitude below the interstellar medium ( or cosmic ) value @xmath303(@xmath2 ) @xmath304 . note the @xmath2 destruction inside the @xmath112 star with @xmath7 attesting that its convective bottom temperature is high enough to burn @xmath2 through @xmath305 but not enough to produce @xmath35 by @xmath162 burning . finally , our 5 and @xmath30 models present real high @xmath2 surface enhancements , compared to the cosmic value . the maximum abundances we obtain range between @xmath303(@xmath2 ) @xmath306 to 5.1 , depending on the stellar mass and metallicity . our super - lithium - rich stars are predicted to appear in a luminosity range between @xmath307 and @xmath308 , i.e. for @xmath309 between -6.25 and -6.65 . these values are in good agreement with those found by sackmann & boothroyd ( 1992 ) . as the evolution proceeds , the surface @xmath2 abundances finally decrease again very rapidly . this indicates that @xmath162 has been almost completely burned in the envelope [ remember that the @xmath301 is dominant ] , so that the @xmath2 production can no more be supported and it is then destroyed by proton capture . this fact clearly puts an upper limit in luminosity to observe super - lithium - rich agb stars . this last feature is also present in sackmann & boothroyd ( 1992 ) , even if it appears somewhat later during the tp - agb phase of their most massive objects . this could be due to different mass loss rate prescriptions . how do these predictions compare with observational data ( see fig . [ lli ] ) ? in the magellanic clouds , the large majority of lithium - rich giants are observed within a narrow luminosity range , @xmath310 to @xmath311 ( smith & lambert 1989 , 1990b ; plez et al . 1993 ; smith et al . 1995 ) , in very good agreement with the predictions . the observed abundance range is also well reproduced , even if no star has been observed with @xmath303(@xmath2 ) higher than 4.5 in the clouds . moreover , the mass estimate from pulsation theory for some lithium - rich giants of both clouds is consistent with the predicted mass range for surface @xmath2 enhancement ( smith et al . 1995 ) . comparisons with @xmath2 in galactic giants are more delicate because of the larger observational uncertainties on luminosity and mass . galactic super - lithium - rich agb stars can exhibit @xmath303(@xmath2 ) values as high as 5.4 ( abia et al . 1991 ) , but these carbon stars have @xmath309 laying between -5 and -6.2 , i.e. rather lower than the predicted luminosities . this could be due to a bad luminosity determination , as it seems difficult to synthesize @xmath2 inside the convective envelope of agb stars with such relatively low luminosities . however , most of the galactic super - lithium - rich agb stars have lower masses than the ones we model here , and they will be addressed in a forthcoming paper . we now review all the species ( from c ) by indicating in which nuclear reactions they are implicated inside the convective tongue of the successive thermal pulses , as a function of the initial mass @xmath193 and metallicity @xmath6 of the computed tp - agb models . for the discussion , we refer to figs . 7 to 9 of the electronic version of the paper . the mean mass fractions @xmath312 are calculated by taking the average of the mass fraction profile of each nuclide @xmath313 over the maximum region covered by the convective tongue of each thermal pulse , just after its disappearance , i.e. @xmath314 we also refer to fig . [ xenv302 ] that presents the mean mass fractions of key nuclides ( calculated as for the convective tongue ) in each burning region ( i.e. the hbs and inter - shell regions ) compared to their convective envelope content , for the @xmath111 star with @xmath7 . similar figures for the other six agb stars are only available in the electronic version . similarly , fig . [ xthp302 ] presents the abundance variations of the same nuclides inside typical convective thermal pulses . again , similar figures for the other six agb stars can solely be found in the electronic form . all these figures are supporting the following nucleosynthesis descriptions . during the growth of the convective tongue , the mean @xmath170 mass fraction first decreases due to the @xmath48 mixing from the inter - shell region . then , at the thermal pulse maximum , @xmath170 is produced by the @xmath276 reaction and also enhanced due to the downwards penetration of the convective tongue into the he - burnt shell . although this penetration is commonly found , its amplitude seems somewhat more important in our case , bringing more @xmath170 into the thermal pulses . note this increased amount of @xmath170 makes the 3dup episodes more efficient to convert an agb star into a c star . during the first thermal pulses , the amount of @xmath170 that is produced increases due to the higher temperatures reached at the base of the convective tongue . this bottom temperature increase from pulse to pulse is a common feature of all the stars and it traduces that the thermal pulse intensity increases . the last computed thermal pulses of the @xmath111 object with @xmath7 are even hot enough to partially destroy @xmath170 by the @xmath50 reaction . most of the engulfed @xmath8 is destroyed very efficiently through @xmath282 that produces neutrons . the trend of the remaining mean @xmath8 mass fraction to increase from pulse to pulse is due to the fact that stronger thermal pulses develop convective tongues reaching regions always closer to the hbs , i.e. engulfing more material from the inter - shell zone . as mentioned in sect . 6.1 , the bottom of the hbs can not be reached however . @xmath38 is only produced by the capture of part of the neutrons on the engulfed @xmath171 , through @xmath59 . note that this reaction contributes to the production of protons . for initial masses above @xmath112 ( @xmath111 ) with @xmath7 ( 0.005 ) however , the convective tongue finally becomes hot enough ( base temperature @xmath315 k ) to significantly destroy the produced @xmath38 by the @xmath54 reaction . the relatively low @xmath38 production inside the @xmath111 with @xmath7 object is due to the relatively lower amount of @xmath8 that is ingested ( partially due to the third @xmath8 burning shell ; see sect . 7.1.1 ) , leading to a smaller quantity of neutrons . however , the last thermal pulses of this star ingest increasing amounts of @xmath8 and @xmath171 from the inter - shell , leading to an increasing production of @xmath38 . due to its short lifetime ( @xmath316 yr ) , @xmath38 partially decays during the inter - pulse phases of our less massive agb models . this depletion is however reduced by the @xmath38 production inside the radiative @xmath8 burning shell . for the most massive agb stars that have inter - pulse durations shorter that @xmath317 , @xmath38 mostly behaves like a stable nuclide , even if strongly destroyed by @xmath24 captures . consequently , as explained with more details in forestini et al . ( 1996 ) , intermediate - mass agb stars are globally not important producers of @xmath38 . apart from the first two thermal pulses ( whatever the initial mass or metallicity ) , @xmath171 is almost completely destroyed inside thermal pulses through @xmath318 . this reaction , as first noted by iben ( 1976 ) , also significantly contributes to the energetics of a thermal pulse . only a small part of @xmath171 is destroyed through the @xmath287 reaction . indeed , the inter - shell mean mass fraction of @xmath171 is ranged between @xmath319 for the @xmath111 star with @xmath7 and @xmath320 for the @xmath113 star with @xmath321 , while the production level of @xmath38 is @xmath322 ( @xmath323 ) , in mass fraction , for the @xmath7 ( 0.005 ) models ( except for the @xmath111 with @xmath7 star ; see above ) . as are @xmath8 and @xmath171 , @xmath172 is engulfed in a thermal pulse from the inter - shell region . the only significant reaction that produces this element inside the convective tongue itself is @xmath324 ( see below for the @xmath96 production ) . on the other hand , @xmath172 is mainly destroyed by the @xmath60 reaction . its rate is lower than that of @xmath318 so that @xmath172 is efficiently destroyed in all the stars only in the asymptotic regime , i.e. when thermal pulses become very intense . remember that it arrives earlier for lower @xmath6 objects . note that the @xmath111 stars ( whatever @xmath6 ) maintain a relatively higher @xmath172 abundance , especially during the full amplitude thermal pulses . indeed , while these thermal pulses efficiently destroy @xmath172 whatever the initial mass of the star , its initial abundance inside the convective tongue is higher for @xmath111 objects . this is due to the fact that the engulfed inter - shell region of @xmath111 stars , that contains @xmath172 , is significantly thicker ( in mass ) compared to that of more massive ones . @xmath173 being partially destroyed in the hbs , the mixing of the inter - shell region reduces its mean mass fraction . on the other hand , at the maximum extent of the thermal pulses , it is enhanced , like @xmath170 , due to the downwards penetration of the convective tongue . from pulse to pulse , it also becomes significantly produced by @xmath50 . this is especially visible for @xmath1 stars , the initial @xmath173 content of which is lower . this primary @xmath173 production by the thermal pulses finally stabilizes around @xmath319 ( in mass fraction ) . the decrease observed during the last thermal pulses of the @xmath111 agb star with @xmath7 traduces the more important role of mixing of inter - shell material ( already emphasized for @xmath171 and @xmath172 ) . the @xmath174 previously contained in the inter - shell region is also mixed in the convective tongue . it is not produced inside thermal pulses . on the contrary , it can be destroyed mainly through @xmath325 and @xmath326 , the first reaction being in general at least ten times faster than the second one . this weakly contributes to produce neutrons inside the thermal pulses . this @xmath174 destruction already operates from the first thermal pulses , whatever the initial @xmath193 and @xmath6 . in later thermal pulses , the @xmath174 mass fraction stabilizes around 4 to @xmath327 , that represents a balance between the amount of @xmath174 brought from the inter - shell region and destroyed by the thermal pulse . again , our @xmath111 models with @xmath7 indicate a still dominant role of the inter - shell mixing . it is less true for the @xmath1 models because at lower @xmath6 , thermal pulses are somewhat hotter , that favors @xmath174 depletion . contrarily to the case of @xmath174 , the inter - shell mass fraction of @xmath96 is very low ( @xmath328 typically ) . its presence inside thermal pulse consequently results from its production by the @xmath329 reaction . note that the rapid @xmath41 @xmath330decay to @xmath96 also significantly contributes to the thermal pulse energetics . this element is destroyed by three reactions that are , by decreasing order of nuclear time scales , @xmath331 , @xmath332 and @xmath333 . this last reaction can even become faster than @xmath332 when significant amounts of protons are produced . this especially arrives for the late thermal pulses of the @xmath111 star with @xmath7 . the convective tongues being hotter for more massive agb stars and/or later along the tp - agb phase of a given star , the @xmath96 destruction , which is very sensitive to temperature , becomes more and more efficient from pulse to pulse . this point was already noted by boothroyd & sackmann ( 1988b ) . the thermal pulse equilibrium mean mass fraction between production and destruction is ranged between @xmath334 and @xmath335 depending on @xmath193 and @xmath6 . as for @xmath96 , @xmath3 is destroyed in the hbs . it is however produced inside thermal pulses . as first demonstrated by forestini et al . ( 1992 ) , the @xmath3 synthesis involves many nuclear reactions , so that it constitutes a good tracer of the nucleosynthesis conditions prevailing inside thermal pulses . @xmath3 is produced by the @xmath60 reaction . the rather rapid increase of the mean @xmath3 mass fraction during the first thermal pulses of each agb star is due to the increasing importance of the inter - shell @xmath172 destruction . when convective tongues become hot enough to completely destroy the engulfed @xmath172 , the following thermal pulses only ingest @xmath172 from matter that has not experienced the preceding thermal pulse ( i.e. @xmath336 of the convective tongue extension in mass typically ; see table [ endagb ] ) . however , as we have explained above , @xmath172 can be produced by the chain @xmath329@xmath337@xmath333 . neutrons are mainly coming from the @xmath282 reaction , and to a lower extent , from @xmath332 , @xmath76 , @xmath338 ( see below ) and @xmath325 . protons are coming from @xmath299 reactions , mostly on @xmath43 and @xmath171 ( see below ) . however , _ ( i ) _ only part of the neutrons are available to produce protons as many @xmath95 reactions also occur ( see sect . 7.5 ) and _ ( ii ) _ a small ( @xmath339 ) part of the protons are also captured by other nuclides than @xmath96 [ mainly by @xmath340 . the success of this reaction chain mainly depends on the rate of @xmath96 destruction , mainly by @xmath24 captures ( see above ) . this destruction becomes more efficient from pulse to pulse so that only a small fraction of @xmath96 is available for the @xmath341 reaction . on the other hand , when convective tongues become very hot ( base temperature @xmath342 k ) , @xmath3 begins to be destroyed by @xmath343 and @xmath66 ( if significant amounts of neutrons are available ) . in conclusion , three facts concur to considerably decrease the @xmath3 production after the first thermal pulses ( whatever @xmath193 and @xmath6 ) . indeed , from pulse to pulse and especially for the most massive agb stars , * the amount of @xmath8 engulfed in the convective tongue of a thermal pulse decreases ( due to the partial destruction of @xmath8 in the inter - shell region during the inter - pulse phase ; see sect . 7.1.1 ) ; * the amount of @xmath172 ingested from the inter - shell region also decreases ; * the @xmath24 captures on @xmath96 are more frequent ; * the @xmath3 destruction becomes significant . this can in particular lead to a partial destruction of the @xmath3 synthesized in the previous thermal pulses for stars with @xmath344 @xmath112 . the mean @xmath186 abundance is roughly unchanged in the hbs as well as by the thermal pulse nucleosynthesis . in number , the contribution of @xmath345 is rather negligible . @xmath187 is significantly destroyed in the hbs . inside the convective tongues of quite hot thermal pulses ( base temperatures above @xmath346 k ) , it is produced by the small proportion of @xmath96 that is destroyed through the @xmath332 reaction . there is also a small contribution by @xmath326 . only a very small @xmath187 fraction can be destroyed by the @xmath338 at the end of very hot thermal pulses ( base temperatures above @xmath347 k ) . @xmath188 is somewhat produced in the hbs . its abundance is considerably enhanced inside thermal pulses ( by typically a factor of 100 ) due to its production by @xmath331 . however , its abundance from pulse to pulse becomes constant ( or even slightly decreases inside full amplitudes thermal pulses of the most massive stars and/or those with lower @xmath6 ) . at their maximum extent , such convective tongues are effectively hot enough ( base temperatures @xmath348 k ) to activate the @xmath349 and @xmath350 reactions . however , as noted below , only a small part of the @xmath188 nuclides are converted into @xmath191 or @xmath190 , respectively ( typically one hundredth ) . nevertheless , this consists in a non - negligible neutron source for these thermal pulses , as first noted by kppeler et al . ( 1990 ) and emphasized by straniero et al . ( 1995 ) . concerning @xmath4 , one has to distinguish between the 3 , 4 , @xmath113 with @xmath7 and @xmath111 with @xmath1 stars on the one hand , and the @xmath30 with @xmath7 , 4 and @xmath113 with @xmath1 stars on the other . inside the former ones , @xmath4 is significantly produced at the bottom of the hbs and very slightly produced by the hebs [ during the inter - pulse phase , through the @xmath351 reaction ] . furthermore , whatever @xmath193 and @xmath6 , @xmath4 is almost unchanged by the thermal pulse nucleosynthesis . as a consequence , its mean mass fraction inside the convective tongue does not evolve significantly from pulse to pulse . at the opposite , inside the latter ones , the bottom of the hbs is hot enough to somewhat destroy @xmath4 while it is produced in its upper part . however , the hebs substantially produces this nuclide during the inter - pulse phase . consequently , when the convective tongues of such stars penetrate the inter - shell region , they engulf material that is impoverished in @xmath4 ; this explains its mean mass fraction decrease from pulse to pulse . the @xmath189 abundance is almost unchanged by the thermal pulse nucleosynthesis too . furthermore , the hbs of @xmath111 stars is not hot enough to destroy it significantly . its mean mass fraction in the convective tongue region of such stars consequently remains almost constant from pulse to pulse . the situation is quite different for more massive stars ( whatever @xmath6 ) as they destroy @xmath189 at the bottom of their hbs . dilution of inter - shell material by the successive convective tongues thus traduces in a mean mass fraction decrease with time . intermediate nuclide of the mgal chain , @xmath190 is modified by the proton burning inside the hbs . we refer to sect . 7.1.1 for more details ( dependence on @xmath193 and @xmath6 ) . however , the evolution of its mean mass fraction inside the convective tongues of thermal pulses is dominated by its production through the @xmath76 reaction [ operating at a quite comparable rate than @xmath75 , following our present knowledge of the corresponding nuclear reaction rates ] . indeed , even if this reaction is very slow inside thermal pulses of relatively low mass agb stars ( especially for their first thermal pulses ) , @xmath188 is much more abundant than @xmath190 ( by typically a factor of one hundred ) in that region , that explains the sensible increase of the @xmath190 mean mass fraction . @xmath191 is slightly ( significantly ) depleted in the hbs of the @xmath195 ( 0.005 ) models . on the other hand , it is substantially produced by the thermal pulse nucleosynthesis _ ( i ) _ by the @xmath75 reaction ( same remark as for @xmath190 above ) and , to a lower extent , _ ( ii ) _ by the @xmath352 reaction ( see below ) . the slope with which its mean mass fraction increases form pulse to pulse is consequently higher for hotter convective tongues , i.e. steeper for full amplitude thermal pulses of massive agb stars . @xmath43 is one of the most important by - product of the hbs ( see above ) . with a lifetime @xmath353 yr , i.e. much longer than the inter - pulse duration of intermediate - mass agb stars , it essentially behaves like a stable nuclide in the inter - shell region . when engulfed inside the convective tongue of a thermal pulse , it is mainly destroyed through the @xmath352 reaction and , at least ten times slower , by @xmath354 . as a result , @xmath43 is the most important proton source inside thermal pulses , whatever @xmath193 and @xmath6 . its destruction is however partial , due to the relatively low neutron abundance ( see sect . 7.5 below ) . note that @xmath352 is also the principal neutron source in the radiative @xmath8 burning shell , as reported by wasserburg et al . ( 1994 ) too . @xmath192 is significantly produced in the hbs , especially in the most massive agb stars . it does not significantly participate to the thermal pulse nucleosynthesis . as a consequence , it is engulfed by the convective tongue from the inter - shell region and its corresponding mean mass fraction slightly increases from pulse to pulse . note however that with time , most of the h - burning of the @xmath113 agb star with @xmath1 occurs inside its convective envelope . this considerably reduces the @xmath192 production . si , p and s are almost not concerned by charged particle nuclear reactions inside the convective tongue of thermal pulses . on the other hand , they are also rather unchanged in the hbs . the only role played by these elements concerns neutron captures that we now briefly discuss . many evolved agb stars show considerable overabundances in elements heavier than iron ( see e.g. smith & lambert 1986 ) . the original discovery by merrill ( 1952 ) of tc in the spectra of some s stars definitively proved that these heavy elements are synthesized inside agb stars . iben ( 1975 ) first demonstrated how such species can indeed be produced inside the thermal pulses by slow neutron captures ( the so - called @xmath284process ) , those neutrons coming from the @xmath76 reaction . however , iben s computations concerned a very massive agb star ( @xmath0 ) , while ever since , all the observed stars showing @xmath280 elements enrichment have been identified as being of lower mass ( i.e. @xmath273 @xmath30 and for many of them @xmath273 @xmath111 ) . more recent observations ( aaronson & mould 1985 ) and evolutionary models ( malaney & boothroyd 1987 ) then strongly suggested that the major neutron source had to be the @xmath282 reaction . @xmath8 is engulfed by the convective tongues of thermal pulses from the inter - shell region where the hbs ashes accumulate . however , the amount of @xmath8 spread out by the hbs is by far ( i.e. at least a factor of 10 or more ) insufficient to explain the @xmath284process that requires more than one neutron by seed @xmath260 . this so - called `` @xmath284process mystery '' ( sackmann & boothroyd 1991 ) is a common failure of all the recent stellar evolution models of tp - agb stars . it is moreover reinforced by the fact that other lighter nuclides also capture part of the available neutrons ( mainly , by increasing order of atomic mass , @xmath170 , @xmath171 , @xmath174 , @xmath3 , @xmath189 , @xmath190 , @xmath43 , @xmath9 , @xmath355 , @xmath356 , @xmath357 , @xmath358 , @xmath34 , ) . one has to note that while the neutron abundance has to be increased to solve the @xmath284process problem , the present computations probably slightly overestimate the abundances of the above mentioned nuclides . more specifically , in our thermal pulse models , the ratio of the engulfed mass of @xmath8 to the mass of @xmath260 inside the convective tongue at its maximum extent ranges between 0.002 to 0.03 . this upper limit is reached during the asymptotic regime of all our modeled stars and thus appears to be rather independent of the initial mass or metallicity . inside their first thermal pulses however , @xmath111 stars already produce neutrons ( ratio roughly equal to 0.02 ) while at the opposite , the corresponding thermal pulses of @xmath30 stars do not engulf substantial amounts of @xmath8 . furthermore , we noted ( see sect . 7.4 ) that the @xmath188 neutron source operates after the maximum of very intense thermal pulses , at a rate that is growing with the maximum temperature of the convective tongue , i.e. with initial mass and/or pulse number . let us stress that such low quantities of @xmath8 ingested by thermal pulses are partly explained by the partial @xmath8 radiative burning , operating at the bottom of the inter - shell region during the inter - pulse phase ( see sect . consequently , it seems clear that the @xmath284process occuring inside tp - agb stars has two distinct origins ( in time and space ) : * during the inter - pulse phase , _ radiative _ @xmath284process occurs in a radiative zone that is the lower part of the inter - shell region ; the only neutron source is @xmath8 ; * during and inside the convective tongue associated with thermal pulses , the _ classical _ @xmath284process takes place with two possible neutron sources : _ ( i ) _ the remainder amount of @xmath8 that is engulfed by the convective tongue and , to a lower extent ( maximum a few percents ) , _ ( ii ) _ the @xmath188 , close to the end of hot thermal pulses . let us emphasize that the @xmath359 and accompanying @xmath95 reactions have non - negligible energetic effects on the structural evolution of the convective tongues . we compute again some thermal pulses by treating together nucleosynthesis and time - dependent convective mixing through eqs . we observed an abundance gradient inside the convective tongue for a few nuclides that are involved in very fast nuclear reactions ( neutrons , protons and some unstable nuclides like @xmath41 or @xmath42 ) . globally however , the final abundances resulting from the thermal pulse nucleosynthesis were similar to within @xmath360 compared to standard computations . more specifically , neutrons were systematically found to be much more abundant ( by typically five orders of magnitude ) at the base of the convective tongues , meaning that the ingested @xmath8 is first transported down before it is destroyed by @xmath282 . the liberated neutrons are very rapidly captured , as attested e.g. by a greater proton abundance at the base of the convective tongue too . this leads us to the conclusion that matter that is actually irradiated by the neutron flux is only the bottom part of each convective tongue . the number of free neutrons by seed @xmath260 is so enhanced . such an approach has already been suggested by malaney et al . the conclusion , however , remains that in order to reproduce the observed distribution of @xmath280 elements in the primitive solar system , higher neutron fluxes are still needed . the most natural way to conciliate observations and theoretical models concerning the production of the @xmath280 elements should be to increase the amount of @xmath8 in the inter - shell region . this can be done if protons are transported down , e.g. during the 3dup . @xmath8 could then be produced by the @xmath270 reaction . such a scenario would enhance the @xmath284process in both sites where it can occur . in our opinion , the present failure to build this extra - amount of @xmath8 could be related to our bad treatment of convective boundaries , especially when they rapidly penetrates very inhomogeneous regions , like it is the case during the 3dup events . the possible occurrence of such a slow - particle transport process able to transport protons down into the inter - shell region and its exact efficiency will be quantitatively investigated in the context of low - mass agb stars . these stars have indeed longer inter - pulse and 3dup durations that could allow to transport significant amounts of hydrogen downwards ( see sect . 9 ) . straniero et al . ( 1995 ) have recently shown that such enhanced amounts of @xmath8 in the inter - shell region could indeed allow the radiative @xmath284process to occur . most interestingly , they found that the resulting signatures are rather similar to those coming from classical @xmath284process computations and are even reached faster . this work is very important as it clearly demonstrates that this so - called `` @xmath8 pocket '' allows to explain the observed @xmath280 element synthesis in agb stars . in order to summarize our detailed discussion of nucleosynthesis in the previous sections , we now identify which specific nuclear region , inside a tp - agb star , mainly contributes to the surface abundance change of each nuclide , when the third dredge - up occurs . concerning the light elements , let us just recall ( see sect . 7.2 ) that @xmath2 can be produced by hbb in the convective envelope of @xmath297 @xmath361 ( @xmath112 ) agb stars with @xmath7 ( 0.005 ) . when a third dredge - up occurs , the convective envelope first penetrates the inter - shell region that is depleted in @xmath170 and then reaches a deeper region previously enriched in @xmath170 by the convective tongue . as a result , @xmath170 globally increases in the convective envelope . the situation is reversed for @xmath8 . after a 3dup , the envelope is somewhat impoverished in @xmath8 as this nuclide is almost completely destroyed inside the convective tongue . consequently , the isotopic ratio @xmath170/@xmath8 significantly increases at each 3dup , all the more as the convective envelope mass is reduced . however , in our @xmath294 @xmath113 ( @xmath112 ) tp - agb stars with @xmath7 ( 0.005 ) , hbb is strong enough to partially convert @xmath170 in @xmath8 through the cn cycle so that the surface @xmath170/@xmath8 drastically decreases . for these objects , the 3dup and hbb have opposite effects . last but not least , @xmath38 , somewhat produced by thermal pulses and the third burning shell , is up - heaved to the surface during 3dup events , where it however substantially decays in the less massive agb stars . @xmath171 is produced in the inter - shell region but is strongly destroyed inside thermal pulses . @xmath172 is destroyed in the inter - shell region while it is somewhat produced by the first few thermal pulses and then destroyed inside full amplitude thermal pulses , i.e. those that can be followed by a 3dup . as a consequence , the surface isotopic ratio @xmath171/@xmath172 is expected to slightly increase from 3dup to 3dup . again , as the cn cycle operates inside the convective envelope of the most massive agb stars , @xmath171 is produced from @xmath170 and @xmath172 is destroyed , thus leading to very high @xmath171/@xmath172 ratios . if hbb occurs at very high temperatures ( i.e. @xmath362 k ) a rather long time enough ( i.e. for the lowest mass loss rate ) , the approach towards the cno bi - cycle equilibrium leads to an increase of the @xmath172 abundance and consequently , a decreasing @xmath171/@xmath172 ratio . this mainly concerns our @xmath30 ( @xmath113 ) agb star with @xmath7 ( 0.005 ) . @xmath173 , like @xmath170 , is depleted in the inter - shell region , but is produced inside the thermal pulses . during 3dup events , its surface abundance very slightly increases . @xmath174 is always considerably depleted in both the inter - shell and thermal pulse regions , so that a 3dup decreases its surface abundance . finally , at the end of full amplitude thermal pulses , @xmath96 has a lower abundance as inside the convective envelope . as it is also completely destroyed in the inter - shell region , its abundance drastically decreases from 3dup to 3dup . consequently , both @xmath173/@xmath174 and @xmath173/@xmath96 surface isotopic ratios significantly increase during each dredge - up . if hbb is strong enough however , the operation of the on cycle increases the amount of @xmath174 to the expense of @xmath173 and substantially destroys @xmath96 . so , in massive agb stars , @xmath173/@xmath174 can decrease while @xmath173/@xmath96 further increases much more than in stars that do not experience hbb . @xmath3 is destroyed in the hbs but it is significantly produced by thermal pulses . as a consequence , 3dup events must lead to correlated @xmath3 and @xmath170 surface enhancements , since both elements are produced in the same region . as already explained , @xmath186 is almost unchanged in the hbs and thermal pulse nucleosynthesis . @xmath187 is destroyed inside the inter - shell region but produced inside thermal pulses , this production slightly increasing from pulse to pulse , whatever @xmath193 or @xmath6 . third dredges - up consequently slightly decreases the @xmath186/@xmath187 isotopic ratio . in our @xmath30 ( @xmath113 ) agb models with @xmath7 ( 0.005 ) however , @xmath187 is partially destroyed by hbb . on the other hand , the @xmath186/@xmath188 ratio substantially decreases from 3dup to 3dup due to large @xmath188 production in the hbs and especially inside the thermal pulses . during 3dup events , the surface abundance of @xmath4 slightly increases for agb stars with initial masses up to @xmath112 ( @xmath111 ) and @xmath7 ( 0.005 ) , due to its production in the hbs . the hotter hbs of more massive agb stars do not produce @xmath4 , globally . our @xmath30 ( @xmath113 ) models for the same respective @xmath6 even slightly destroy @xmath4 by hbb . @xmath189 is somewhat destroyed in the hbs of our most massive agb stars , especially those with @xmath1 , while its abundance does not change significantly after the thermal pulse nucleosynthesis . @xmath190 is depleted in the inter - shell region of our @xmath111 models with @xmath195 as well as in the @xmath1 models of all masses . it is however produced in the hbs of @xmath297 @xmath112 agb stars with @xmath7 . nevertheless , its surface abundance evolution following 3dup events is mainly conditioned by its large abundance ( compared to the envelope one ) in the region where thermal pulses occur , especially inside the most massive agb stars . therefore , the surface @xmath189/@xmath190 isotopic ratio necessarily decreases in all agb stars . the same is true for the @xmath189/@xmath191 ratio that roughly behaves in a similar way . reasons are also rather similar , even if @xmath191 is less modified than @xmath190 in the hbs . at the surface of our @xmath30 ( @xmath113 ) with @xmath7 ( 0.005 ) agb stars , the abundances of both @xmath190 and @xmath191 can be enhanced to the expense of @xmath189 , if hbb is strong enough , leading to very low @xmath189/@xmath190 and @xmath189/@xmath191 ratios . as we will see in sect . 8.1 however , this only concerns the end of the tp - agb phase in extreme situations allowing hbb to occur a long time enough ( e.g. in case of relatively low mass loss rates ) . only a very small amount of @xmath43 is present in the convective envelope of e - agb stars as a result of the first ( and eventually the second ) dredge(s)-up . along the tp - agb phase , @xmath43 is largely produced in the hbs and most of it accumulates in the inter - shell region . inside full amplitude thermal pulses however , it is somewhat destroyed . for its part , @xmath192 is slightly produced in the hbs , especially inside the most massive agb stars . consequently , @xmath43/@xmath192 significantly increases with time during the tp - agb phase , due to the repetitive 3dup episodes . of course , dilution by convective mixing being more important for increasing @xmath193 , the highest ratios are expected at the surface of the less massive agb stars . however , the same extreme cases mentioned in the case of the mg isotopic ratios can lead to very high @xmath43/@xmath192 ratios for the most massive agb stars . > from si , the 3dup does not change significantly the surface isotopic ratios . however , as stressed in sect . 7.5 , when models will be able to reproduce the @xmath280 element enhancements at the surface of evolved agb stars , si and s isotopic ratios could be slightly modified by the high neutron flux . in this last section , we present our predictions concerning _ ( i ) _ the evolution of all the surface isotopic ratios along the tp - agb phase ( by including our evolutionary models and extrapolations up to the convective envelope removal ) and _ ( ii ) _ the total amount of material ejected at different ages for all the species up to @xmath9 . we refer to the previous sections for detailed explanations . thus , in the following , we will just comment these predictions . figs . [ xsurf302a ] and [ xsurf302b ] display , for the @xmath111 agb star with @xmath7 , the evolution of the isotopic ratios as a function of the total remaining mass that decreases due to mass loss . similar figures for the other six agb stars are only displayed in the electronic version . let us stress that we stop our calculations at the last thermal pulse that occurs before the convective envelope mass is reduced below @xmath363 . this stage indeed roughly corresponds to the pn ejection that we do not maintain to model here . this explains the quite large remaining total stellar masses in the figures , especially for those computed with the highest mass loss rates . the first comment concerns the formation of carbon stars . @xmath111 agb stars with @xmath7 and 3 and @xmath112 agb stars with @xmath1 become carbon stars during their tp - agb phase . the surface @xmath170/@xmath173 becomes greater than unity more rapidly for _ ( i ) _ decreasing total mass and/or _ ( ii ) _ decreasing @xmath6 . this metallicity dependence is confirmed by the increase of c stars compared to m stars from the galactic bulge to lmc and smc ( blanco et al . 1978 ) . our @xmath30 ( @xmath113 ) agb stars with @xmath7 ( 0.005 ) can also become c stars with our lowest mass loss rates . in such cases indeed , in spite of very efficient hbb transforming @xmath170 in @xmath8 and @xmath171 , many 3dup events can still occur with very reduced envelope masses , so that the relatively short inter - pulse phases are not long enough to substantially destroy @xmath170 . we will see however , together with other isotopic ratios , that these very low mass loss rates seem actually extreme . let us emphasize that the only super - lithium - rich star ( see sect . 7.2 ) we found to be also a carbon star is the @xmath113 with @xmath321 one ( with our medium mass loss rate ) . however , as already mentioned in sect . 6.2 , observations indicate that @xmath7 carbon stars are formed easier ( i.e. at lower luminosities ) than predicted by all the models . let us also stress the very high @xmath170/@xmath8 isotopic ratios that can be expected at the surface of @xmath272 @xmath112 agb stars , whatever @xmath6 . on the other hand , more massive agb stars experience hbb that progressively leads the @xmath170/@xmath8 isotopic ratio close to its cn cycle equilibrium value ( formation of j stars ) . the case of @xmath112 stars with @xmath1 very clearly shows the competition between the 3dup episodes ( increasing @xmath170/@xmath8 ) and hbb ( decreasing it ) . for lower mass loss rates , the 3dup events are more numerous for the same total mass decrease , leading to a more rapid @xmath170/@xmath8 increase . however , the corresponding longer tp - agb phase allows the base of the convective envelope to reach higher temperatures that finally destroy @xmath170 through the cn cycle . this efficient transformation of @xmath170 to @xmath8 also reduces the @xmath170/@xmath173 ratio , i.e. the possibility to form carbon stars . it happens if the temperatures at the base of the convective envelopes become higher than @xmath364 k. such stars have corresponding surface luminosities @xmath297 @xmath365 ( @xmath366 ) for @xmath7 ( 0.005 ) , i.e. @xmath367 ( -6.45 ) . we consequently predict that c stars can not be formed at higher luminosities than this threshold , due to strong hbb . this is in agreement with boothroyd et al . ( 1993 ) predictions . this also corresponds to the observed threshold for agb stars of the lmc ( cohen et al . 1981 ) . finally , note the surface pollution in @xmath38 that is more important , compared to that of @xmath8 , for decreasing total masses . for @xmath111 agb stars , the surface @xmath8/@xmath38 ratio is quite independent of the mass loss rate . this is due to the inter - pulse duration that is significantly greater than the @xmath38 lifetime , so that the @xmath38 decay is the dominant factor for the @xmath8/@xmath38 ratio evolution at the surface of such stars . the quite complicated behavior of the @xmath8/@xmath38 isotopic ratio at the surface of our most massive models is again due to the competition between the 3dup that adds @xmath38 into the convective envelope and hbb that can partially destroy it . by extrapolation , it is reasonable to expect even lower @xmath8/@xmath38 ratios in low - mass agb stars . as mentioned in forestini et al . ( 1996 ) , @xmath38 could become detectable for @xmath273 @xmath111 evolved agb stars . due to hbb , those stars showing very low surface @xmath170/@xmath8 ratios are also expected to have very high @xmath171/@xmath172 ones . it has to be stressed indeed that in our most massive agb stars , the cn cycle operating inside the convective envelope produces large amounts of @xmath171 . the @xmath173/@xmath174 and @xmath173/@xmath96 surface isotopic ratios also evolve as expected from our discussions in sect . note that quite low @xmath173/@xmath174 and very high @xmath173/@xmath96 ratios , which are clear signatures of the operation of the on cycle at the base of the convective envelope ( hbb ) , appear rather simultaneously with the cn cycle signatures in our most massive objects . also remark the progressive @xmath173/@xmath96 decrease in our @xmath30 ( @xmath113 ) agb star with @xmath7 ( 0.005 ) . for these objects , the temperature at the base of the convective envelope finally becomes high enough for @xmath173 to partially burn . the @xmath3 surface enhancements we predict are displayed in a special form , corresponding to the observation presentations adopted by jorissen et al . compared to their fig . 8 , two conclusions can be stated . * there is a clear observed correlation between the @xmath3 excess and the degree of @xmath170 surface pollution , that is perfectly understandable from the identified nucleosynthetic processes and the mixing episodes ( see sect . furthermore , our models clearly indicate a trend : the less massive the agb star , the more efficient the @xmath3 surface enhancement . in fact , our most massive models even show a surface @xmath3 depletion due to its partial destruction by hbb . * our @xmath7 models well reproduce only the less @xmath3 enriched stars in the [ @xmath3/@xmath173 ] versus @xmath170/@xmath173 diagram . based on the trend just mentioned , we predict that the agb stars showing very large @xmath3 surface enhancements have initial total masses @xmath273 @xmath111 . we plan to verify this statement , i.e. that low - mass agb stars are the most efficient @xmath3 galactic producers ( see sect . 9 ) . as expected , the @xmath189/@xmath190 and @xmath189/@xmath191 surface isotopic ratios are much less sensitive to the 3dup episodes and hbb . more precisely , only our most massive models with the lowest mass loss rates show large @xmath189 depletions due to a very strong hbb . again , these are however quite extreme situations . last but not least , we predict substantial @xmath43 surface enhancements due to the repetitive dredges - up of the hbs and inter - shell regions . the @xmath43/@xmath192 isotopic ratio tends to decrease with increasing total mass due to _ ( i ) _ the greater dilution as mixing occurs in a more massive convective envelope and _ ( ii ) _ the thinner hbs and inter - shell regions ( in mass ) . nevertheless , @xmath43/@xmath192 could well be increasing again if @xmath43 was produced inside the convective envelope itself by strong hbb . note however that such a very high @xmath43 production level is actually expected to be accompanied by very low mg isotopic ratios , as @xmath43 is produced by the mgal chain in the extreme situations mentioned above . let us finally mention that wasserburg et al . ( 1994 ) presented computations in which the evolution of the @xmath43 surface abundance has been followed self - consistently , by including the effect of neutron irradiation due to an enhanced amount of @xmath8 in the inter - shell region ( to engender a @xmath284process ) . our results compare very well with those obtained by boothroyd et al . ( 1994 , 1995 ) concerning the o isotopic ratios in intermediate - mass stars . they also mention the problem to reproduce the @xmath173/@xmath174 ratio , especially in low - mass carbon agb stars . we can not go further in this comparison without having yet computed the tp - agb phase of low - mass stars . boothroyd et al . ( 1995 ) already suggested that to conciliate all the data with observations of low - mass stars , one probably has to take into account of slow - particle transports inside radiative zones , what they called `` cool - bottom precessing '' . that is what we plan to do ( see sect . evolutionary models of tp - agb stars can be strongly constrained by the confrontation between predicted and observed surface isotopic ratios , especially if different ratios are observed for the same stars . for most of the observed agb stars , only the c and o isotopic ratios have been determined ( see e.g. harris & lambert 1984 ; harris et al . 1985 ; lambert et al . 1986 ; harris et al . 1987 ; harris & lambert 1987 ; kahane et al . globally , we found agreements . general features come out . * as the temperatures required at the base of the convective envelope to allow @xmath2 production ( see sect . 7.2 ) are significantly lower than those allowing efficient @xmath170 destruction by the cn cycle , we predict that luminous c stars enriched in @xmath2 can exist , especially for lower @xmath6 objects ( that have lower @xmath173 content ) . there are indeed a few % of the observed c stars that are also super - lithium - rich and there are more numerous in the lmc compared to our galaxy ( see e.g. smith & lambert 1990b for the lmc and feast 1974 for the galaxy ) . * there is no observed c star with @xmath170/@xmath8 @xmath369 . most of the j stars show @xmath170/@xmath8 ratios between 3 et 4 , i.e. very close to the cn cycle equilibrium value . * j stars also show rather low @xmath173/@xmath174 isotopic ratios while only upper limits have been found for the @xmath173/@xmath96 ratio . most of the other agb stars present @xmath173/@xmath174 ratios ranged between 300 and 4000 and @xmath173/@xmath96 ratios ranged between 500 and 5000 . only stars with rather high @xmath173/@xmath174 ratios are difficult to explain with our intermediate - mass agb models ( see also boothroyd et al . also note that the corresponding o isotopic ratios in the ism are @xmath173/@xmath174 @xmath370 and @xmath173/@xmath96 @xmath371 . again , there is a problem with @xmath174 . let us however stress that important uncertainties remain concerning the nuclear cross sections of proton captures on @xmath174 . on the other hand , low - mass agb stars are present among those observations . presently , there is only one star for which all the isotopic ratios have been determined . it is cw leo . this c star is one of the nearest evolved agb star ever observed . its distance is estimated to @xmath372 pc ( see e.g. claussen et al . 1987 ; keady et al . 1988 ; griffin 1990 ) . this corresponds to a surface luminosity @xmath373 @xmath374 and a present mass loss rate of roughly @xmath375 @xmath376@xmath377 . such a high wind makes it quite optically thick . a very extended circumstellar envelope has been detected to surround cw leo , irc+10216 . irc+10216 shows evidences for a bi - polar structure ( see e.g. gulin et al . 1993 ) , attesting that cw leo is very evolved and close to the end of its tp - agb phase . consequently , its present convective envelope mass is probably very reduced ( @xmath378 @xmath379 typically ) . the circumstellar envelope of cw leo allowed very accurate determinations of the isotopic ratios ( see kahane et al . 1988 and 1992 ; gulin et al . 1995 ; forestini et al . 1996 ) : @xmath170/@xmath8 @xmath380 , @xmath8/@xmath38 @xmath381 , @xmath171/@xmath172 @xmath382 , @xmath173/@xmath174 @xmath383 , @xmath173/@xmath96 @xmath384 , @xmath189/@xmath190 @xmath385 , @xmath189/@xmath191 @xmath386 and @xmath43/@xmath192 @xmath387 ( to be confirmed ) . furthermore , the si isotopic ratios appear to have interstellar medium values , i.e. unaltered by the star . as derived in gulin et al . ( 1995 ) and forestini et al . ( 1996 ) , almost all these isotopic ratios seem to be consistent ( within the error bars ) , following our predictions , with an agb star that had a main sequence total mass between 4 and @xmath113 . however , @xmath173/@xmath174 is again not well explained . as various surface isotopic ratios ( especially @xmath171/@xmath172 and @xmath173/@xmath96 ) are significantly changing between 4 and @xmath113 agb models , we are presently computing the tp - agb phase of a @xmath361 star with @xmath7 in order to improve this last prediction . finally , another very interesting way to constraint stellar evolution models of agb stars is to compare the predicted surface isotopic ratios with those determined in primitive meteorites ( mainly through grains included inside carbon rich chondrites ) . most of the @xmath170/@xmath8 , @xmath171/@xmath172 and @xmath43/@xmath192 isotopic anomalies measured in grains ( see e.g. ott 1991 ; zinner et al . 1991a ) are consistent with those coming from the surface of evolved tp - agb stars of intermediate - mass . this is of course in favor of the presence , when the solar system has been formed , of one or a few agb stars in its surroundings . such grains are indeed well known to be formed in the cool atmosphere of evolved agb stars ( i.e. when mass loss rates become rather high ) . concerning the measured o isotopic ratios , the situation is not so clear as many grains have ratios significantly out of the corresponding predicted ranges . some of these grains at least can have been formed in low - mass agb stars . again , as mentioned in sect . 8.1 , boothroyd et al . ( 1995 ) recently suggested that in such stars , cool - bottom processing can help to reduce these discrepancies . on the contrary , some other grains perfectly agree with our predictions . this is for example the case of a @xmath388 @xmath389 grain of the bishunpur ll3.1 chondrite by huss et al . the derived ratios , @xmath173/@xmath174 @xmath390 , @xmath173/@xmath96 @xmath391 and @xmath43/@xmath192 @xmath392 , are in perfect agreement with ratios corresponding to a @xmath393 @xmath112 agb star of nearly solar metallicity . finally , other sic @xmath43-rich grains raise problems , like that discovered by zinner et al . ( 1991b ) , exhibiting a very high @xmath43/@xmath192 ratio of @xmath394 . it seems hard to explain such a value from our models . however , this grain also exhibits an anomalously large @xmath395/@xmath396 isotopic ratio that seems to exclude its agb star origin . tables [ yield1 ] to [ yield3 ] give the total mass ejected by the wind ( in @xmath376 ) for various elements _ ( i ) _ at the top of the first rgb , _ ( ii ) _ at the beginning of the tp - agb phase and _ ( iii ) _ at the end of the tp - agb phase ( thus including our extrapolated nucleosynthesis calculations ) . as such data mainly concerns the chemical evolution of galaxies , we also present the net yield ( positive or negative ) for each element . the data of tables [ yield1 ] to [ yield3 ] correspond to the medium mass loss rates ( see sect . 8.1 ) , i.e. the solid lines of figs . [ xsurf302a ] and [ xsurf302b ] . brief comments are made about these predictions in the electronic version . llccccccc & & & + & & & + & & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath125&_at rg top_&2.9978&3.9985&4.9970&5.9980&2.9996&3.9999&4.9997 + & _ at @xmath164_&2.9151&3.7732&4.7009&5.7167&2.8053&3.5067&4.3149 + & _ at agb tip_&1.0989&1.0143&1.0750&1.3735 & 0.9471&1.2519&1.3818 + @xmath162&_at rg top_&2.59@xmath397&1.51@xmath397&2.93@xmath397&1.59@xmath397&7.73@xmath398&9.70@xmath165&3.54@xmath398 + & _ at @xmath164_&1.30@xmath144&2.35@xmath144&2.41@xmath144&1.96@xmath144&3.26@xmath144&5.70@xmath144&6.52@xmath144 + & _ at agb tip_&2.87@xmath143&3.03@xmath143&1.57@xmath143&5.00@xmath144&3.30@xmath143&3.04@xmath143&9.19@xmath144 + & _ net yield _ & 2.25@xmath143 & 2.07@xmath143 & 3.46@xmath144&-9.44@xmath144 & 2.74@xmath143 & 2.29@xmath143&-4.27@xmath399 + @xmath48&_at rg top_&6.53@xmath143&4.28@xmath143&8.69@xmath143&5.84@xmath143&1.15@xmath143&1.89@xmath144&8.10@xmath144 + & _ at @xmath164_&2.59@xmath400&6.91@xmath400&9.59@xmath400&9.13@xmath400&5.11@xmath400&0.13&0.18 + & _ at agb tip_&0.59&0.91&1.32&1.59&0.55&0.72&1.28 + & _ net yield _ & 3.76@xmath400 & 3.20@xmath400 & 0.17 & 0.23 & 3.56@xmath400 & 3.70@xmath400 & 0.37 + @xmath2&_at rg top_&1.59@xmath167&1.02@xmath167&2.49@xmath167&1.45@xmath167&1.42@xmath401&2.67@xmath402&1.26@xmath401 + & _ at @xmath164_&2.63@xmath167&3.35@xmath167&5.18@xmath167&3.87@xmath167&4.47@xmath403&1.15@xmath167&1.96@xmath167 + & _ at agb tip_&2.92@xmath166&6.35@xmath167&7.33@xmath398&1.08@xmath397&8.23@xmath167&4.60@xmath166&3.57@xmath167 + & _ net yield_&-1.85@xmath398&-2.95@xmath398 & 3.44@xmath398 & 6.23@xmath398&-5.00@xmath165&-6.34@xmath165&-8.92@xmath165 + @xmath170&_at rg top_&6.44@xmath399&4.25@xmath399&9.11@xmath399&5.84@xmath399&2.14@xmath397&4.01@xmath398&1.90@xmath397 + & _ at @xmath164_&1.72@xmath143&4.53@xmath143&5.90@xmath143&5.56@xmath143&9.01@xmath144&2.34@xmath143&3.43@xmath143 + & _ at agb tip_&1.26@xmath400&9.71@xmath142&5.97@xmath142&2.74@xmath142&1.06@xmath400&4.40@xmath142&9.68@xmath143 + & _ net yield _ & 6.53@xmath142 & 1.33@xmath143&-6.62@xmath142&-1.21@xmath400 & 9.00@xmath142 & 2.19@xmath142&-1.93@xmath142 + @xmath8&_at rg top_&1.26@xmath397&8.52@xmath398&1.48@xmath397&1.14@xmath397&1.14@xmath398&2.17@xmath165&1.04@xmath398 + & _ at @xmath164_&8.75@xmath399&2.41@xmath144&3.23@xmath144&3.09@xmath144&4.90@xmath399&1.31@xmath144&1.91@xmath144 + & _ at agb tip_&1.95@xmath143&3.12@xmath143&1.00@xmath142&6.67@xmath143&5.05@xmath144&9.09@xmath144&1.93@xmath143 + & _ net yield _ & 1.22@xmath143 & 1.97@xmath143 & 8.50@xmath143 & 4.89@xmath143 & 3.06@xmath144 & 6.43@xmath144 & 1.58@xmath143 + @xmath38&_at rg top_&4.89@xmath404&4.77@xmath405&3.22@xmath406&5.69@xmath407&2.99@xmath408&9.59@xmath409&3.23@xmath406 + & _ at @xmath164_&6.92@xmath410&4.77@xmath405&3.62@xmath406&5.56@xmath169&1.68@xmath411&4.92@xmath412&3.38@xmath413 + & _ at agb tip_&6.20@xmath165&8.93@xmath165&2.90@xmath165&1.16@xmath165&8.54@xmath166&1.43@xmath165&4.35@xmath403 + & _ net yield _ & 6.20@xmath165 & 8.93@xmath165 & 2.90@xmath165 & 1.16@xmath165 & 8.54@xmath166 & 1.43@xmath165 & 4.35@xmath403 + @xmath171&_at rg top_&3.41@xmath399&2.22@xmath399&3.92@xmath399&2.99@xmath399&3.25@xmath397&4.47@xmath398&1.69@xmath397 + & _ at @xmath164_&2.35@xmath143&6.53@xmath143&9.26@xmath143&9.01@xmath143&1.53@xmath143&3.60@xmath143&4.66@xmath143 + & _ at agb tip_&5.24@xmath142&8.52@xmath142&1.84@xmath400&2.89@xmath400&1.58@xmath142&2.07@xmath142&1.04@xmath400 + & _ net yield _ & 3.01@xmath142 & 5.03@xmath142 & 1.38@xmath400 & 2.35@xmath400 & 9.75@xmath143 & 1.26@xmath142 & 9.33@xmath142 + @xmath172&_at rg top_&8.89@xmath165&5.83@xmath165&1.28@xmath398&8.02@xmath165&2.40@xmath166&4.36@xmath167&2.02@xmath166 + & _ at @xmath164_&1.92@xmath397&4.93@xmath397&6.33@xmath397&5.89@xmath397&1.01@xmath397&2.56@xmath397&3.70@xmath397 + & _ at agb tip_&4.13@xmath399&6.32@xmath399&1.97@xmath399&2.66@xmath399&1.02@xmath399&9.13@xmath397&5.13@xmath399 + & _ net yield_&-4.65@xmath399&-7.47@xmath399&-1.62@xmath144&-1.87@xmath144&-1.35@xmath399&-2.26@xmath399 & 9.53@xmath397 + llccccccc & & & + & & & + & & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath173&_at rg top_&2.23@xmath144&1.47@xmath144&3.00@xmath144&2.00@xmath144&1.10@xmath399&1.92@xmath397&8.22@xmath397 + & _ at @xmath164_&8.33@xmath143&2.20@xmath142&2.84@xmath142&2.67@xmath142&4.68@xmath143&1.21@xmath142&1.69@xmath142 + & _ at agb tip_&1.92@xmath400&2.86@xmath400&3.61@xmath400&3.96@xmath400&5.34@xmath142&6.74@xmath142&4.17@xmath142 + & _ net yield_&-1.36@xmath143&-1.66@xmath142&-3.72@xmath142&-7.33@xmath142 & 1.27@xmath143&-2.29@xmath143&-5.01@xmath142 + @xmath174&_at rg top_&2.18@xmath398&1.15@xmath398&1.51@xmath398&1.26@xmath398&4.19@xmath165&2.09@xmath166&4.38@xmath166 + & _ at @xmath164_&2.62@xmath399&5.56@xmath399&5.45@xmath399&4.51@xmath399&1.87@xmath399&3.63@xmath399&3.17@xmath399 + & _ at agb tip_&5.87@xmath144&7.26@xmath144&7.14@xmath144&1.17@xmath143&1.94@xmath144&2.03@xmath144&1.05@xmath144 + & _ net yield _ & 5.09@xmath144 & 6.03@xmath144 & 5.52@xmath144 & 9.80@xmath144 & 1.73@xmath144 & 1.75@xmath144 & 6.78@xmath399 + @xmath96&_at rg top_&4.74@xmath398&3.13@xmath398&6.61@xmath398&4.29@xmath398&1.79@xmath165&3.36@xmath166&1.59@xmath165 + & _ at @xmath164_&1.42@xmath399&3.76@xmath399&4.87@xmath399&4.57@xmath399&7.55@xmath397&1.97@xmath399&2.86@xmath399 + & _ at agb tip_&3.10@xmath144&4.84@xmath144&2.09@xmath144&5.32@xmath399&7.66@xmath399&1.06@xmath144&3.81@xmath399 + & _ net yield_&-1.26@xmath144&-2.01@xmath144&-6.91@xmath144&-1.01@xmath143&-4.11@xmath399&-5.16@xmath399&-1.69@xmath144 + @xmath3&_at rg top_&9.54@xmath166&6.23@xmath166&1.27@xmath165&8.46@xmath166&4.88@xmath167&8.38@xmath403&3.55@xmath167 + & _ at @xmath164_&3.70@xmath398&9.54@xmath398&1.22@xmath397&1.13@xmath397&2.07@xmath398&5.27@xmath398&7.34@xmath398 + & _ at agb tip_&1.87@xmath399&1.68@xmath399&2.07@xmath399&1.75@xmath399&4.31@xmath397&3.41@xmath397&1.62@xmath397 + & _ net yield _ & 1.06@xmath399 & 3.96@xmath397 & 3.84@xmath397&-2.33@xmath397 & 2.11@xmath397 & 4.65@xmath398&-2.26@xmath397 + @xmath186&_at rg top_&3.79@xmath399&2.49@xmath399&5.07@xmath399&3.39@xmath399&1.92@xmath397&3.24@xmath398&1.39@xmath397 + & _ at @xmath164_&1.46@xmath143&3.89@xmath143&5.12@xmath143&4.85@xmath143&8.34@xmath144&2.11@xmath143&2.93@xmath143 + & _ at agb tip_&3.25@xmath142&5.05@xmath142&6.69@xmath142&7.90@xmath142&8.64@xmath143&1.17@xmath142&1.58@xmath142 + & _ net yield_&-1.87@xmath399&-6.51@xmath144&-3.03@xmath144&-2.98@xmath144&-1.51@xmath144&-4.53@xmath399 & 2.59@xmath144 + @xmath187&_at rg top_&9.81@xmath165&6.45@xmath165&1.31@xmath398&8.76@xmath165&5.31@xmath166&8.75@xmath167&3.71@xmath166 + & _ at @xmath164_&3.83@xmath397&1.03@xmath399&1.35@xmath399&1.27@xmath399&2.37@xmath397&5.86@xmath397&7.85@xmath397 + & _ at agb tip_&1.01@xmath144&1.42@xmath144&9.29@xmath399&2.71@xmath399&2.83@xmath399&3.36@xmath399&1.15@xmath399 + & _ net yield _ & 1.79@xmath399 & 1.21@xmath399&-7.85@xmath399&-1.75@xmath144 & 5.92@xmath397 & 3.60@xmath397&-2.80@xmath399 + @xmath188&_at rg top_&2.99@xmath397&1.97@xmath397&4.06@xmath397&2.69@xmath397&1.38@xmath398&2.54@xmath165&1.11@xmath398 + & _ at @xmath164_&1.06@xmath144&2.81@xmath144&3.66@xmath144&3.45@xmath144&5.85@xmath399&1.54@xmath144&2.21@xmath144 + & _ at agb tip_&9.45@xmath143&6.81@xmath143&7.75@xmath143&8.94@xmath143&2.40@xmath143&1.39@xmath143&1.44@xmath143 + & _ net yield _ & 6.83@xmath143 & 2.69@xmath143 & 2.35@xmath143 & 2.56@xmath143 & 1.69@xmath143 & 4.44@xmath144 & 1.98@xmath144 + @xmath4&_at rg top_&8.40@xmath398&5.48@xmath398&1.07@xmath397&7.41@xmath398&5.66@xmath165&7.33@xmath166&2.92@xmath165 + & _ at @xmath164_&4.11@xmath399&1.13@xmath144&1.55@xmath144&1.49@xmath144&2.60@xmath399&6.08@xmath399&7.76@xmath399 + & _ at agb tip_&9.47@xmath144&1.49@xmath143&2.16@xmath143&2.51@xmath143&2.87@xmath144&3.50@xmath144&2.80@xmath144 + & _ net yield _ & 2.76@xmath144 & 4.30@xmath144 & 7.72@xmath144 & 8.80@xmath144 & 1.06@xmath144 & 1.07@xmath144&-3.98@xmath399 + llccccccc & & & + & & & + & & @xmath111&@xmath112&@xmath113&@xmath30&@xmath111&@xmath112&@xmath113 + @xmath189&_at rg top_&1.20@xmath399&7.90@xmath397&1.61@xmath399&1.08@xmath399&6.09@xmath398&1.03@xmath398&4.40@xmath398 + & _ at @xmath164_&4.62@xmath144&1.23@xmath143&1.63@xmath143&1.54@xmath143&2.64@xmath144&6.68@xmath144&9.29@xmath144 + & _ at agb tip_&1.03@xmath142&1.60@xmath142&2.12@xmath142&2.31@xmath142&2.74@xmath143&3.70@xmath143&2.28@xmath143 + & _ net yield_&-1.35@xmath399&-2.55@xmath144&-2.09@xmath144&-2.09@xmath143&-6.05@xmath399&-4.73@xmath399&-2.65@xmath143 + @xmath190&_at rg top_&1.60@xmath397&1.05@xmath397&2.14@xmath397&1.42@xmath397&8.26@xmath165&1.40@xmath165&5.96@xmath165 + & _ at @xmath164_&6.12@xmath399&1.63@xmath144&2.11@xmath144&1.99@xmath144&3.57@xmath399&9.04@xmath399&1.24@xmath144 + & _ at agb tip_&1.42@xmath143&2.22@xmath143&2.84@xmath143&5.17@xmath143&3.88@xmath144&5.13@xmath144&1.73@xmath143 + & _ net yield _ & 6.17@xmath399 & 7.72@xmath399 & 3.19@xmath399 & 1.85@xmath143 & 2.00@xmath399 & 2.06@xmath399 & 1.08@xmath143 + @xmath191&_at rg top_&1.82@xmath397&1.20@xmath397&2.44@xmath397&1.63@xmath397&9.31@xmath165&1.56@xmath165&6.69@xmath165 + & _ at @xmath164_&7.00@xmath399&1.87@xmath144&2.51@xmath144&2.39@xmath144&4.05@xmath399&1.03@xmath144&1.44@xmath144 + & _ at agb tip_&1.68@xmath143&2.57@xmath143&3.48@xmath143&4.06@xmath143&4.50@xmath144&5.98@xmath144&5.43@xmath144 + & _ net yield _ & 1.23@xmath144 & 1.22@xmath144 & 2.52@xmath144 & 2.57@xmath144 & 2.84@xmath399 & 3.39@xmath399&-1.99@xmath144 + @xmath414&_at rg top_&1.20@xmath169&2.83@xmath402&1.01@xmath402&1.93@xmath401&1.29@xmath168&4.29@xmath406&1.11@xmath406 + & _ at @xmath164_&6.15@xmath169&1.39@xmath166&5.63@xmath165&2.44@xmath398&3.10@xmath401&6.12@xmath166&2.73@xmath165 + & _ at agb tip_&2.03@xmath397&8.87@xmath398&4.03@xmath397&3.67@xmath399&4.34@xmath398&9.02@xmath165&7.69@xmath399 + & _ net yield _ & 2.03@xmath397 & 8.87@xmath398 & 4.03@xmath397 & 3.67@xmath399 & 4.34@xmath398 & 9.02@xmath165 & 7.69@xmath399 + @xmath192&_at rg top_&1.36@xmath397&8.91@xmath398&1.82@xmath397&1.22@xmath397&6.89@xmath165&1.16@xmath165&4.97@xmath165 + & _ at @xmath164_&5.21@xmath399&1.39@xmath144&1.84@xmath144&1.74@xmath144&2.99@xmath399&7.57@xmath399&1.05@xmath144 + & _ at agb tip_&1.18@xmath143&1.83@xmath143&2.45@xmath143&2.94@xmath143&3.27@xmath144&4.38@xmath144&7.42@xmath144 + & _ net yield _ & 1.49@xmath399&-5.26@xmath398 & 3.78@xmath399 & 1.00@xmath144 & 1.24@xmath399 & 1.60@xmath399 & 1.87@xmath144 + @xmath9&_at rg top_&1.53@xmath399&1.00@xmath399&2.05@xmath399&1.37@xmath399&7.76@xmath398&1.31@xmath398&5.60@xmath398 + & _ at @xmath164_&5.87@xmath144&1.57@xmath143&2.07@xmath143&1.96@xmath143&3.36@xmath144&8.52@xmath144&1.18@xmath143 + & _ at agb tip_&1.31@xmath142&2.04@xmath142&2.70@xmath142&3.18@xmath142&3.49@xmath143&4.73@xmath143&6.07@xmath143 + & _ net yield_&-1.47@xmath399&-2.76@xmath144&-1.35@xmath144&-1.84@xmath144&-5.95@xmath399&-2.20@xmath399&-1.85@xmath144 + we tried to confirm and improve our present knowledge of the structural and nucleosynthetic evolution of intermediate - mass agb stars . to do that , we have first presented in detail _ ( i ) _ the physics and numerical aspects of our stellar evolution code , _ ( ii ) _ results concerning evolution phases prior to the agb one and _ ( iii ) _ comparisons with other works . these informations are essential to better understand and appraise the large set of predictions we discuss about the thermally pulsing agb stage . we also compare our predictions with various observations ( at different evolutionary stages ) in order to clearly identify the processes that should be included in future models . * concerning the structure ( intensities , temperatures ) of the thermal pulses as well as the thermal properties at the base of the convective envelope , the most decisive quantities are the total mass and core mass . both change with time in a way that is governed by the increasing mass loss rate . this translates into a great sensitivity of the various nucleosynthesis processes to the adopted mass loss rate . consequently , the evolution of the thermal pulse characteristics crucially depend on this last quantity , unfortunately still rather badly known . in addition , the lower the mass loss rates , the longer the tp - agb phase , the more third dredge - up events have time to occur and the greater are the surface abundance changes due to hbb . this is clearly demonstrated by the high sensitiveness of the predicted surface evolution of almost all the isotopic ratios to the mass loss rate dependence on time . * on the other hand , at a given total stellar mass , core masses are greater for lower metallicities , as a result of the central burning phases . as a consequence , for the reason mentioned at the former point , tp - agb stars with @xmath1 mostly behave like @xmath7 ones with a total mass greater by @xmath415 @xmath416 . however , due to important differences in initial compositions that also influence the nuclear reactions , this is not completely true as far as isotopic ratios are concerned . at the end of the tp - agb phase indeed , some yields are quite different for both @xmath6 we have studied . * the neutron captures on intermediate - mass and heavy nuclides occur in two distinct sites inside tp - agb stars , namely _ ( i ) _ at the base of the inter - shell zone during the inter - pulse phase and , as already known , _ ( ii ) _ inside the convective tongues associated with thermal pulses . along the asymptotic tp - agb , the amount of produced neutrons is rather independent of the initial total mass and metallicity ( at least for intermediate - mass objects ) . * concerning chemical evolution of galaxies , intermediate - mass stars , at the end of their existence , appear to be rather significant producers of @xmath162 , @xmath48 , @xmath2 ( most massive and/or higher @xmath6 stars ) , @xmath8 , @xmath171 , @xmath174 , @xmath3 , @xmath188 , @xmath4 , @xmath191 and @xmath192 . a found trend , compared to observations , however clearly supports the idea that most of the @xmath3 present in the interstellar medium has to come from low - mass agb stars ( that point has to be confirmed by forthcoming evolutionary models ) . + intermediate - mass agb stars also substantially produce radio - nuclides , namely @xmath38 and @xmath259 . the former one mostly comes from lower mass stars while it is the contrary for the last one . + these stars partially deplete the interstellar medium content in @xmath2 ( less massive stars ) , @xmath172 , @xmath173 , @xmath96 , @xmath186 , @xmath189 and @xmath9 . * we have not yet identified the @xmath8 additional source needed to quantitatively explain the @xmath284process . this process is however required in order to account for many observations unquestionably indicating that agb stars are responsible for the main component of the solar system heavy element distribution . some suggestions have been made . radiative diffusion operating during the tp - agb phase of low - mass stars could substantially help . although already invoked , such a process has never been yet included in agb star evolution models . * the rather high level of @xmath2 production detected at the surface of some evolved but relatively faint agb stars of our galaxy still remains unaccountable . thus , the cameron - fowler mechanism does not explain all the super - lithium - rich agb stars . as these galactic objects have relatively low initial masses , the explanation could also come from the radiative slow - transport processes above mentioned . * last but not least , some c stars are observed , especially in the galactic bulge , with low luminosities at which the models fail at predicting the occurrence of the third dredge - up . this is probably due to our bad knowledge ( and treatment ) of finely tuned convection motions in boundaries of stellar regions having strong chemical composition gradients . clearly , some kind of extra - mixing ( deeper inside nuclearly processed regions ) has to be invoked . however , no self - consistent mechanism has been settled and tested through complete evolution models up to now . other confrontations with observations require the modeling of low - mass agb stars . a clear distinction between low- and intermediate - mass stars is justified . indeed , numerous observations indicate that `` non - standard '' particle transport processes are acting inside low - mass stars at different phases of their evolution , that substantially modify the chemical structure compared to what is obtained in classical models . during the various dredge - up episodes , matter up - heaved to the surface has consequently a different composition . this is not the case in intermediate - mass stars . in conclusion , low - mass agb stellar models , maintaining to make detailed predictions concerning the evolution of surface isotopic ratios , have to include such slow - particle transport processes . they are currently being calculated and will be presented in a next - coming paper . we are first indebted to maurizio busso , the referee , for its very careful reading of our manuscript that helped us to significantly improve and clarify it . we also thank him and roberto gallino for the very instructive discussions we already had together about these very complex agb stars . let us finally thank lionel siess for his always valuable contributions to improve the stellar evolution code and related utilities . part of the computations presented in this paper ( roughly representing 7 months of cpu time ) were performed at the `` center de calcul intensif de lobservatoire de grenoble '' . most of them have been realized at `` imag '' on a ibm sp1 computer financed by the mesr , cnrs and rgion rhne - alpes . this work was supported by grants from the gdr `` structure interne des etoiles et des plantes gantes '' ( cnrs ) . aaronson , m. , mould , j. 1985 , apj , 288 , 551 abia , c. , boffin , h.m.j . , isern , j. , rebolo , r. 1991 , a&a , 245 , l1 abia , c. , boffin , h.m.j . , isern , j. , rebolo , r. 1993 , a&a , 272 , 455 alexander , d. r. , ferguson , j. w. 1994 , apj , 437 , 879 allen , b.j . , machlin , r.l . , gibbons , j.h . 1971 , adv . , 4 , 205 alongi , m. , bertelli , g. , bressan , a. , chiosi , c. 1991 , in formation and evolution of star clusters , ed . j. kanneth ( asp conf . 13 ) , 223 anders , e. , grevesse , n. 1989 , geochim . acta , 53 , 197 bao , z.y . , kppeler , f. 1987 , atomic data nucl . data tables , 36 , 411 bazan , g. , lattanzio , j.c . 1993 , apj , 409 , 762 beer , h. , wiescher , m. , kppeler , f. , gorres , j. , koehler , p.e . 1991 , preprint bell , r.a . , eriksson , k. , gustafsson , b. , nordlund , a. 1976 , a&a , 23 , 37 blanco , b.m . , blanco , v.m . , mccarthy , m.f . 1978 , nat , 271 , 638 blanco , v.m . , mccarthy , m.f . , blanco , b.m . 1980 , apj , 242 , 938 blcker , t. 1995 , a&a , 297 , 727 blcker , t. , schnberner , d. 1991 , a&a , 244 , l43 boothroyd , a.i . , sackmann , i .- j . 1988a , apj , 328 , 641 boothroyd , a.i . , sackmann , i .- j . 1988b , apj , 328 , 653 boothroyd , a.i . , sackmann , i .- j . 1988c , apj , 328 , 671 boothroyd , a.i . , sackmann , i .- j . , ahern , s.c . 1993 , apj , 416 , 762 boothroyd , a.i . , sackmann , i .- , wasserburg , g.j . 1994 , apj , 430 , l77 boothroyd , a.i . , sackmann , i .- j . , wasserburg , g.j . 1995 , apj , 442 , l21 brehm , k. , becker , h.w . , rolfs , c. , kppeler , f. , ratynski , w. 1988 , phys . a. atomic nuclei , 330 , 167 bressan , a. , fagotto , f. , bertelli , g. , chiosi , c. 1993 , a&as , 100 , 647 bryan , g.l . , volk , k. , kwok , s. 1990 , apj , 365 , 301 busso , m. , lambert , d.l . , beglio , l. , gallino , r. , raiteri , c.m . , smith , v.v . 1995 , apj , 446 , 775 cameron , a.g.w . , fowler , w.a . 1971 , apj , 164 , 111 caughlan , g.r . , fowler , w.a . 1988 , atomic data nucl . data tables , 40 , 283 caughlan , g.r . , fowler , w.a . , harris , m.j . , zimmerman , b.a . 1985 , atomic data nucl . data tables , 32 , 197 champagne , a.e . , cella , c.h . , konzes , r.t . , loury , r.m . , magnus , p.v . , smith , m.s . , mao , z.q . 1988 , nucl . a , 487 , 433 charbonnel , c. 1994 , a&a , 282 , 811 charbonnel , c. 1995 , apj , 453 , l41 claussen , m.j . , kleinmann , s.g . , joyce , r.r . , jura , m. 1987 , apjs , 65 , 385 cohen , j.g . , frogel , j.a . , persson , s.e . , elias , j.h . 1981 , apj , 249 , 481 de oliveira , f. 1995 , ph . d. thesis ( universit de paris xi , france ) el eid , m.f . 1994 , a&a , 285 , 915 feast , m.w . 1974 , in stellar instability and evolution , eds . p. ledoux , a. noels and a.w . rodgers ( dordrecht , reidel ) , 93 forestini , m. , gulin , m. , cernicharo , j. 1996 , a&a , _ in press _ forestini , m. , paulus , g. , arnould , m. 1991 , a&a , 252 , 597 forestini , m. , goriely , s. , jorissen , a. , arnould , m. 1992 , a&a , 261 , 157 fowler , w.a . , caughlan , g.r . , zimmerman , b.a . 1967 , ara&a , 5 , 525 frost , j.c . , lattanzio , j. 1996 , apj , in press fujimoto , m.y . , sugimoto , d. 1979 , pasj , 31 , 1 funck , c. , langanke , k. 1989 , apj , 344 , 46 gilroy , k.k . 1989 , apj , 347 , 835 graboske , h.c . , dewitt , h.e . , grossman , a.s . , cooper , m.s . 1973 , apj , 181 , 457 grevesse , n. 1991 , in evolution of stars : the photospheric abundance connection , eds . g. michaud and a. tutukov ( dordrecht , kluwer ) , 63 griffin , i.p . 1990 , mnras , 247 , 591 groenewegen , m.a.t . , de jong , t. 1993 , a&a , 267 , 410 gulin , m. , lucas , r. , cernicharo , j. 1993 , a&a , 280 , l19 gulin , m. , forestini , m. , valiron , p. , ziurys , l.m . , anderson , m.a . , cernicharo , j. , kahane , c. 1995 , a&a , 297 , 183 hammer , j.w . 1991 , preprint harris , m.j . , lambert , d.l . 1984 , apj , 281 , 739 harris , m.j . , lambert , d.l . 1987 , apj , 318 , 868 harris , m.j . , lambert , d.l . , smith , v.v . 1985 , apj , 299 , 375 harris , m.j . , lambert , d.l . , smith , v.v . 1988 , apj , 321 , 761 harris , m.j . , lambert , d.l . , hinkle , k.h . , gustafsson , b. , eriksson , k. 1987 , apj , 316 , 294 hubbard , w.b . , lampe , m. 1969 , apjs , 18 , 297 huss , g.r . , fahey , a.j . , gallino , r. , wasserburg , g.j . 1994 , apj , 430 , l81 iben , i. 1964 , apj , 140 , 1631 iben , i. 1975 , apj , 196 , 525 iben , i. 1976 , apj , 208 , 165 iben , i. 1981 , apj , 246 , 278 iben , i. 1983 , apj , 275 , l65 illiadis , c. 1990 , nucl . phys . a , 512 , 509 itoh , n. , mutoh , h. , hikita , a. , kohyama , y. 1992 , apj , 395 , 622 jorissen , a. , arnould , m. 1989 , a&a , 221 , 161 jorissen , a. , smith , v.v . , lambert , d.l . 1992 , a&a , 261 , 164 kahane , c. , gmez - gonzles , j. , cernicharo , j. , gulin , m. 1988 , a&a , 190 , 167 kahane , c. , cernicharo , j. , gmez - gonzles , j. , gulin , m. 1992 , a&a , 256 , 235 kppeler , f. , gallino , r. , busso , m. , picchio , g. , raiteri , c.m . 1990 , apj , 354 , 630 kppeler , f. et al . 1994 , apj , 437 , 396 keady , j.j . , hall , d.n.b . , ridgway , s.t . 1988 , apj , 326 , 832 kious , m. 1990 , ph.d . thesis ( universit de paris xi , france ) kippenhahn , r. , weigert , a. 1991 , stellar structure and evolution ( springer - verlag ) , 23 kippenhahn , r. , weigert , a. , hofmeister , e. 1968 , methods of computational physics , 7 , 129 knapp , g.r . 1989 , apj , 336 , 822 koester , d. 1976 , a&a , 52 , 415 kohyama , y. , itoh , n. , obama , a. , mutoh , h. 1993 , apj , 415 , 267 lambert , d.l . , gustafsson , b. , eriksson , k. , hinkle , k.h . 1986 , apjs , 62 , 373 lambert , d.l . , smith , v.v . , busso , m. , gallino , r. , straniero , o. 1995 , apj , 450 , 302 landr , v. , prantzos , n. , aguer , p. , bogaert , g. , lefbvre , a. , thibaud , j .- 1990 , a&a , 240 , 85 lattanzio , j. 1986 , apj , 311 , 708 lattanzio , j. 1987 , apj , 313 , l15 lattanzio , j. 1989 , in evolution of peculiar red giants , eds . johnson and b. zuckerman ( cambridge university press , cambridge ) , 161 malaney , r.a . , boothroyd , a.i . 1987 , apj , 320 , 866 malaney , r.a . , savage , m.j . , boothroyd , a.i . 1988 , apj , 324 , 948 merrill , p.w . 1952 , apj , 116 , 21 mould , j.r . , aaronson , m. 1982 , apj , 263 , 629 mould , j.r . , aaronson , m. 1986 , apj , 303 , 10 munakata , h. , kohyama , y. , itoh , n. 1985 , apj , 296 , 197 nagai , y. , igashira , m. , takeda , k. , mukai , n. , motoyama , s. , uesawa , f. , kitazawa , h. , fukuda , t. 1991 , apj , 372 , 683 nagai , y. et al . 1995 , in nuclei in the cosmos iii , eds . m. busso , r. gallino and c.m . raiteri ( aip , new york ) , 201 ott , u. 1991 , protostars and planets iii ( university of arizona press , tuscon ) , 144 paczyski , b. 1975 , apj , 202 , 558 plez , b. 1992 , a&as , 94 , 527 plez , b. , smith , v.v . , lambert , d.l . 1993 , apj , 418 , 812 press , , w.h . , flannery , b.p . , teukolsky , s.a . , vetterling , w.t . 1986 , numerical recipes ( cambridge university press ) , 588 raman , s. , igashira , m. , dozono , y. , kitazawa , h. , mizimoto , m. , lynn , j.e . 1990 , phys . c , 41 , 458 reid , n. , mould , j. 1984 , apj , 284 , 98 reimers , d. 1975 , mem . soc . lige , 6th ser . , 8 , 369 rogers , f.j . , iglesias , c.a . 1992 , apjs , 79 , 507 sackmann , i .- j . 1980 , apj , 235 , 554 sackmann , i .- j . , boothroyd , a.i . 1991 , in iau symp . 145 , eds . g. michaud and a. tutukov ( dordrecht , reidel ) , 275 sackmann , i .- j . , boothroyd , a.i . 1992 , apj , 392 , l71 sackmann , i .- smith , r.l . , despain , k.h . 1974 , apj , 187 , 555 scalo , j.m . , despain , k.h . , ulrich , r.k . 1975 , apj , 196 , 805 schaller , g. , schaerer , d. , meynet , g. , maeder , a. 1992 , a&as , 96 , 269 schwarzschild , m. , hrm , r. 1965 , apj , 142 , 855 smith , v.v . , lambert , d.l . 1986 , apj , 311 , 843 smith , v.v . , lambert , d.l . 1989 , apj , 345 , l75 smith , v.v . , lambert , d.l . 1990a , apjs , 72 , 387 smith , v.v . , lambert , d.l . 1990b , apj , 361 , l69 smith , v.v . , plez , b. , lambert , d.l . 1995 , apj , 441 , 735 straniero , o. , gallino , r. , busso , m. , chieffi , a. , raiteri , c.m . , limongi , m. , salaris , m. 1995 , apj , 440 , l85 straniero , o. et al . 1996 , in torino workshop on nucleosynthesis of s- and r - nuclei i ( in press ) thielemann , f .- k . 1991 , in nuclei in the cosmos i , eds . h. oberhummer and c. rolfs ( springer - verlag , berlin ) , 171 vassiliadis , e. , wood , p.r . 1993 , apj , 413 , 641 vogelaar , r.b . 1989 , ph.d . thesis ( california institute of technology , california ) wagoner , r.v . 1969 , apjs , 162 , 247 wannier , p.g . , sahai , r. 1986 , apj , 311 , 335 wasserburg , g.j . , busso , m. , gallino , r. , raiteri , c.m . 1994 , apj , 424 , 412 weidemann , v. 1987 , a&a , 188 , 74 weidemann , v. , koester , d. 1983 , a&a , 121 , 77 wiesher , m. , gorres , j. , thielemann , f .- d . 1990 , apj , 363 , 340 wood , p.r . 1981 , in physical processes in red giants , eds . i. iben and a. renzini ( dordrecht , reidel ) , 135 wood , p.r . , bessell , m.s . , fox , m.w . 1983 , apj , 272 , 99 woosley , s.e . , fowler , w.a . , holmes , j.a . , zimmerman , b.a . 1978 , atomic data nucl . data tables , 22 , 371 zinner , e. , amari , s. , lewis , r. 1991b , in meteoritic society meeting ( monterey , california ) , 17 zinner , e. , amari , s. , anders , e. , lewis , r. 1991a , nat , 349 , 51 zuckerman , b. , dyck , h.m . , claussen , m.j . 1986 , apj , 304 , 401 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: though well - studied for over fifty years , the spectral theory of hyperbolic surfaces still presents basic unresolved questions @xcite . for example , does there exist a noncompact , finite area hyperbolic surface whose laplacian has no nonconstant square - integrable eigenfunctions ? this question has been the subject of many investigations including @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite . the triangle @xmath0 defined by @xmath1 and @xmath2 . ] as a model problem , phillips and sarnak @xcite suggested studying the neumann eigenvalue problem on the domain @xmath3 pictured in figure [ trisect ] . in this paper , we prove the following : [ thm : main ] for all but at most countably many @xmath40,1[$ ] , the neumann laplacian on the geodesic triangle @xmath0 in the hyperbolic plane has no nonconstant ( square - integrable ) eigenfunction . the group @xmath5 of hyperbolic isometries generated by reflections in the geodesic arcs that bound @xmath0 is discrete if and only if @xmath6 for @xmath7 an integer . for example , if @xmath8 , then @xmath5 contains an index two subgroup that is naturally isomorphic to @xmath9 . it follows from the seminal work of a. selberg @xcite that if @xmath10 or @xmath11 , then the neumann laplacian has infinitely many nonconstant eigenfunctions . in particular , for some special @xmath12 , there do exist square - integrable solutions to the neumann problem . is an integer not equal to @xmath13 , or @xmath11 , phillips and sarnak asked whether the domain @xmath14 has no nonconstant neumann eigenfunctions @xcite . we should point out that since theorem [ thm : main ] allows for countably many exceptional @xmath12 , it does not directly answer their question . ] in @xcite , theorem [ thm : main ] was verified under an additional and as of yet unjustified assumption concerning the spectral multiplicities of the neumann laplacian acting on @xmath15 . the proof consisted of studying the singular perturbation problem associated with letting @xmath12 tend to @xmath16 . similar singular perturbations were studied in the context of degenerating hyperbolic surfaces @xcite and unitary characters over a fixed hyperbolic surface @xcite . in @xcite , @xcite , @xcite , and all prior work on this problem , it was necessary to make assumptions about the multiplicities of the spectrum of the limiting surface . the angles of a geodesic triangle in the hyperbolic plane determine the isometry class of the triangle . the angles of @xmath0 are @xmath17 . it is not difficult to extend theorem [ thm : main ] to triangles with angles @xmath18 ( see [ section : alltriangles ] ) . [ thm : gennocf ] the set of @xmath19 for which the hyperbolic triangle with angles @xmath20 admits a nonconstant neumann laplace eigenfunction has lebesgue measure zero and is contained in a countable union of nowhere dense sets . in other words , the generic hyperbolic triangle with one cusp has no non - constant neumann eigenvalue where ` generic ' can be taken in both a topological and a measurable sense . theorem [ thm : main ] gives the existence of a triple of angles @xmath18 for which there are no nonconstant neumann eigenfunctions . theorem [ thm : gennocf ] then results from applying a general and well - understood principle concerning analytic perturbations ( see , for example , @xcite ) . on the other hand , the proof of theorem [ thm : main ] is much more involved . in particular , the proof will rely upon a refined analysis of ` crossings ' of eigenvalue branches . to prove theorem 1.1 , we further develop the method of ` asymptotic separa- tion of variables ' that we introduced in [ hlrjdg11 ] to study generic simplicity of eigenvalues . this method facilitates the study of real - analytic eigenvalue branches in situations where a geometric domain degenerates onto a lower dimensional domain . there is a vast literature for instance , @xcite , @xcite , @xcite concerning perturbations involving degeneration onto lower dimensional domains , but most of these studies do not address analytic eigenvalue branches . in contrast , our results depend crucially on a study of real - analytic eigenbranches and their crossings . we now describe the content of each section . in [ section : prelim ] , we establish notation and recall basic features of the spectral theory of the laplacian acting on functions on a domain in the hyperbolic plane having one cusp . we describe the fourier decomposition associated to the cusp . the zeroth fourier mode is responsible for an essential spectrum of @xmath21 . following @xcite and @xcite , we will replace the dirichlet quadratic form @xmath22 with a modification @xmath23 obtained by ` truncating ' the zeroth fourier coefficient above @xmath24 . an eigenfunction @xmath25 of @xmath23 corresponds to an eigenfunction of @xmath26 if and only if the zeroth fourier coefficient of @xmath25 vanishes identically . we will call such an eigenfunction a _ cusp form_. , these are ` cusp forms ' in the sense of the theory of automorphic forms , but otherwise there is no discrete group , and hence they are not cusp forms in the traditional sense . in this paper we will always be considering ` even ' cusp forms , that is , eigenfunctions satisfying neumann conditions . ] the operator associated to @xmath23 has compact resolvent and hence discrete spectrum . this makes @xmath23 a much better candidate for the application of methods from spectral perturbation theory . in [ sec : generic ] , we recall and make precise some ideas familiar in the perturbational study of cusp form existence . we consider a real - analytic family , @xmath27 , of quadratic forms that have the same domain as @xmath23 . we say that a real - analytic family @xmath28 of eigenfunctions of @xmath29 is a _ cusp form eigenbranch _ if and only @xmath30 is a cusp form for each @xmath12 . we demonstrate a dichotomy : either the family @xmath27 has a real - analytic cusp form eigenbranch or the set of @xmath12 such that @xmath29 has a cusp form is countable . in [ section : alltriangles ] , we consider arbitrary real - analytic paths in the space of hyperbolic triangles with one cusp . we apply the results of [ sec : generic ] to deduce theorem [ thm : gennocf ] under the assumtption that there exists a triangle with no nonconstant neumann eigenfunction . the remainder of the paper is devoted to proving theorem [ thm : main ] which will give the existence of such a triangle . in [ section : expansion ] we specialize to the family @xmath0 . after renormalizing by a factor of @xmath31 , we find that for each @xmath25 , the function @xmath32 has a taylor expansion at @xmath33 . we compute the leading order terms of this expansion . in [ section : approx ] we show that the method of the asymptotic separation of variables introduced in @xcite may be used to analyse the family of quadratic forms @xmath34 . in particular , we define a reference quadratic form @xmath35 to which separation of variables applies and that is asymptotic to @xmath34 at ` first order ' . by separation of variables we mean that each eigenfunction of @xmath35 is of the form @xmath36 with @xmath37 and @xmath38 a solution to a renormalized form of the equation for a modified bessel function with imaginary parameter . we show that one may deduce a non - concentration estimate ( proposition [ prop : noncon ] ) . in [ section : branches ] we use the non - concentration estimate to derive information concerning the real - analytic eigenbranches @xmath39 of @xmath34 . first , we show that there exists an integer @xmath40 so that @xmath41 limits to @xmath42 as @xmath12 tends to zero . second , we find that if the spectral projection of @xmath30 onto the space @xmath43 spanned by functions of the form the @xmath44 is relatively small , then the derivative @xmath45 is of order @xmath46 . finally , we show that if @xmath47 is a cusp form eigenbranch , then @xmath41 can not limit to zero . in [ sec : proof ] we prove theorem [ thm : main ] . by the dichotomy of [ sec : generic ] , it suffices to show that real - analytic cusp form eigenbranches do not exist . we suppose to the contrary that the real - analytic family @xmath34 has a cusp form eigenbranch @xmath47 . by the results of [ section : branches ] , we have that @xmath41 limits to @xmath42 where @xmath48 . by improving the analysis of @xcite , we show that there exists an eigenbranch , @xmath49 , of @xmath35 that ` tracks ' @xmath41 at order @xmath12 in the sense that @xmath50 we will obtain a contradiction to this estimate by estimating @xmath51 from below . indeed , we show that when the norm of the projection @xmath52 of @xmath30 onto @xmath43 is relatively large with respect to @xmath53 , the function @xmath54 is of order @xmath55 , whereas when @xmath56 is relatively small , the function @xmath54 is of order @xmath57 . by controling the sizes of the sets where @xmath56 is respectively small and large relative to @xmath53 and by integrating , we will contradict ( [ trackingestimate ] ) . the key observation is the following : since @xmath41 limits to @xmath58 , it has to ` cross ' each of the eigenbranches of @xmath35 that limit to zero . we show that near such a crossing , the branch @xmath30 must ` interact ' with the functions in @xmath59 to such an extent that the projection onto @xmath43 can not be too large . the effect of each interaction is made precise by careful estimates of the off - diagonal terms in the quadratic form @xmath60 ( appendix a ) . by summing the effects of these interactions , we eventually prove that there exists @xmath61 so that @xmath62 thus contradicting ( [ trackingestimate ] ) . in this section , we describe some basic spectral theory of the neumann laplace operator acting on the square - integrable functions on domains in the hyperbolic plane with a cusp . we define the laplacian and associated dirichlet quadratic form , describe the fourier decomposition of eigenfunctions along horocycles , construct a modification of the dirichlet form whose eigenfunctions include the eigenfunctions ( cusp forms ) of the standard laplacian but whose spectrum is discrete . the half - plane @xmath63 equipped with the riemannian metric @xmath64 is the poincar - lobachevsky model for the @xmath65-dimensional hyperbolic space @xmath66 . the measure associated to the riemannian metric @xmath67 is given by integrating @xmath68 in the present context , a _ cusp of width @xmath69 and height @xmath70 _ is the subset @xmath71 \times [ y_0 , \infty[$ ] of the upper half plane . a domain @xmath72 is said to _ have one cusp _ if @xmath73 is the union of a cusp and a compact set . we will assume that the boundary of @xmath73 is the the union of finitely many geodesic arcs and that the interior of @xmath73 is connected . let @xmath74 denote the set of functions @xmath75 such that @xmath25 is the restriction to @xmath73 of a compactly supported smooth function defined on some neighbourhood of @xmath73 . the @xmath76-inner product of two functions @xmath25 and @xmath77 in @xmath74 is defined by @xmath78 abusing notation slightly , we will often write @xmath79 in place of @xmath80 . let @xmath81 denote the completion of @xmath74 with respect to the norm @xmath82 . to define the neumann laplacian we consider the bilinear form defined on @xmath74 by @xmath83 where @xmath84 satisfies @xmath85 for all vector fields @xmath86 and smooth functions @xmath87 . let @xmath88 denote the value of the quadratic form @xmath89 . one computes that @xmath90 let @xmath91 denote the completion of @xmath74 with respect to the norm @xmath92 . we will consider @xmath26 as a nonnegative symmetric bilinear form on @xmath93 with domain @xmath91 . as such it is densely defined and closed , and hence there exists a unique , densely defined , self - adjoint operator @xmath94 on @xmath93 such that for each @xmath95 and @xmath25 in the domain of @xmath94 we have @xmath96 the operator @xmath94 is called the _ neumann laplacian_. it can be shown that @xmath97 if and only if @xmath98 and @xmath99 where in the definition of @xmath26 the partial derivatives are to be taken in the distributional sense . it is well - known that @xmath94 has essential spectrum equal to @xmath100 . ( for example , see @xcite or @xcite ) . apart from this essential spectrum , @xmath94 may also have eigenvalues either smaller than @xmath101 or embedded in the continuous spectrum . from ( [ laplacian ] ) we see that @xmath25 is an eigenfunction of @xmath94 with eigenvalue @xmath102 if and only if @xmath97 and for each @xmath95 @xmath103 for each integer @xmath104 , define @xmath105 and define @xmath106 . the collection @xmath107 is an orthonormal basis for @xmath108)$ ] . hence , the functions @xmath109 provide an orthonormal basis of @xmath110)$ ] . for positive @xmath111 , let @xmath112 \times [ y_0 , \infty[$ ] be a cusp of width @xmath69 and height @xmath70 . for each @xmath25 in @xmath113 and almost every @xmath114 , the function @xmath115 belongs to @xmath110)$ ] . thus we can write @xmath116 where @xmath117 belongs to @xmath118 . we refer to @xmath119 as the @xmath120 fourier coefficient of @xmath25 . more generally , if @xmath73 is a domain with a cusp @xmath121 , then we define the @xmath120 fourier coefficient of a function @xmath122 to be the @xmath120 fourier coefficient restriction of @xmath77 to @xmath121 . parseval s theorem gives @xmath123 where @xmath124 denotes the characteristic function of a set @xmath86 . [ lem : ode ] if @xmath125 is a neumann eigenfunction of @xmath26 with eigenvalue @xmath102 , then for each @xmath126 and each @xmath127 , the fourier coefficient @xmath119 satisfies @xmath128 if @xmath77 is a smooth function on @xmath73 , then since @xmath25 is a neumann eigenfunction of @xmath26 , integration by parts gives @xmath129 by letting @xmath130 where @xmath131 is a smooth function with compact support in @xmath132y_0,\infty[$ ] , we find that @xmath133 it follows that @xmath119 satisfies ( [ eq : oden ] ) in the distributional sense in @xmath134 by elliptic regularity , @xmath119 is actually smooth and satisfies ( [ eq : oden ] ) in the strong sense . in particular , @xmath135 for some constants @xmath136 and @xmath137 with @xmath138 . if @xmath139 , then the real part of @xmath140 equals @xmath141 , and hence @xmath142 does not belong to @xmath118 unless both @xmath136 and @xmath137 equal zero . therefore , we have the following . [ coro : cuspform ] if @xmath25 is a neumann eigenfunction with eigenvalue @xmath143 , then the zeroth fourier coefficient @xmath142 vanishes identically on @xmath144 . in the classical spectral theory of a quotient of @xmath145 by a lattice in @xmath146 , a laplace eigenfunction @xmath25 with vanishing zeroth fourier coefficient in each cusp is called a _ ( weight zero ) maass cusp form_. even though most of the domains that we will consider are not fundamental domains for discrete groups of isometries , we will adapt this terminology . if @xmath25 is an eigenfunction for the neumann laplacian on a domain with a cusp , and @xmath147 , then we will call @xmath25 a _ cusp form_. traditionally , the neumann eigenfunctions for @xmath14 are called _ even cusp forms _ whereas the solutions to the dirichlet eigenvalue problem are called _ odd cusp forms_. we will not consider odd cusp forms in this paper . we wish to apply analytic perturbation theory to study the behavior of eigenfunctions of @xmath26 on @xmath0 as we vary @xmath12 . because the eigenvalues of @xmath26 might lie inside the essential spectrum , standard perturbation theory does not apply directly . following @xcite and @xcite , we will use a modification of @xmath26 first constructed by p. lax and r. phillips @xcite @xcite . in this section , we recall the construction , show that the eigenvalues of the modification are isolated , and relate the eigenfunctions of the modification to those of @xmath26 . for @xmath148 , let @xmath149 denote the set of @xmath150 such that for each @xmath151 we have @xmath152 . let @xmath153 denote the hilbert space completion of @xmath149 with respect to @xmath154 . let @xmath155 denote the restriction of @xmath156 to @xmath157 . let @xmath158 denote the hilbert space completion of @xmath149 with respect to the norm @xmath159 . the restriction , @xmath23 , of @xmath26 to @xmath158 is a closed , densely defined quadratic form on @xmath157 . a simple argument shows that @xmath160 and @xmath161 in the sequel it will be more convenient to replace @xmath162 by the following other set . [ def : wbet ] define @xmath163 to be the set of functions @xmath25 in @xmath164 such that * @xmath25 extends to a continuous function on the closure @xmath165 of @xmath166 * @xmath25 is smooth on @xmath167 observe that since @xmath168 , the closure of @xmath163 with respect to the norm @xmath159 is @xmath169 the latter assertion says that @xmath163 is a core of the quadratic form @xmath170 let @xmath171 denote the unique operator such that @xmath172 and that satisfies @xmath173 for each @xmath174 . [ lem : relation ] for each @xmath148 , the resolvent of @xmath171 is compact . hence , the spectrum of @xmath23 with respect to @xmath155 is discrete and each eigenspace is finite dimensional . using the fourier decomposition , one shows that for each @xmath175 , the set of @xmath176 such that @xmath177 and @xmath178 is compact in @xmath157 ( see lemma 8.7 @xcite ) . it follows that @xmath171 has compact resolvent . hence , by standard spectral theory , the spectrum is discrete and the eigenspaces are finite dimensional . [ def : cf ] we will say that an eigenfunction @xmath25 of @xmath23 with respect to @xmath155 is a _ cusp form _ if and only for each @xmath127 we have @xmath152 . [ lem : betacf ] the following assertions are equivalent : 1 . @xmath25 is a cusp form of @xmath26 with respect to @xmath156 . there exists @xmath148 such that @xmath25 is a cusp form of @xmath23 with respect to @xmath155 . 3 . for each @xmath148 , the function @xmath25 is a cusp form for @xmath23 with respect to @xmath155 . @xmath179 : if @xmath25 is an eigenfunction of @xmath26 with eigenvalue @xmath102 , then by lemma [ lem : ode ] the zeroth fourier coefficient @xmath142 satisfies the differential equation @xmath180 . since @xmath181 vanishes for @xmath182 , it must vanish for @xmath183 . @xmath184 : fix a smooth function @xmath185 such that @xmath186 for @xmath187 and @xmath188 for @xmath189 if @xmath152 for each @xmath182 , then @xmath190 and @xmath191 for each @xmath95 . thus , if @xmath192 , then @xmath193 . @xmath194 : follows from the equivalence of ( 1 ) and ( 2 ) . not every eigenfunction @xmath25 of @xmath23 is an eigenfunction of @xmath26 . for example , if @xmath195 and @xmath196 is an eisenstein series whose zeroth fourier coefficient vanishes at @xmath24 , then @xmath197 is an eigenfunction of @xmath23 . this function is not smooth across @xmath24 , but elliptic regularity implies that each eigenfunction of @xmath26 is smooth . let @xmath198\times [ y_0,\infty)$ ] and let @xmath199 in this section , we consider a fixed domain @xmath73 that contains the cusp @xmath200 and a real - analytic family @xmath27 of quadratic forms defined on @xmath201 that represents the cusp of width @xmath202 for @xmath182 and some real - analytic function @xmath203 ( see definition [ def : represent ] below ) . we prove the following dichotomy : either there exists a real - analytic eigenfunction branch consisting of ` cusp forms ' or the set of @xmath12 such that @xmath29 has a ` cusp form ' eigenfunction is countable . this fact is fundamental to the proofs of both theorem [ thm : main ] and theorem [ thm : gennocf ] . let @xmath198\times [ y_0 , \infty[$ ] and let @xmath204 . for any @xmath205 , we can define a transformation @xmath206 between @xmath207 and @xmath208 by asking that , for any @xmath209 the function @xmath210 is defined by @xmath211 since , in the sequel @xmath70 will be fixed , we will drop the index @xmath70 and denote by @xmath212 the cusp of width @xmath213 it is straighforward that @xmath206 is an isometry between @xmath207 and @xmath214 moreover @xmath206 also preserves @xmath164 in the sense that @xmath215 if and only if @xmath216 we may thus define @xmath217 the quadratic form obtained by pulling - back @xmath170 on @xmath212 using @xmath218 this quadratic form is then closed on the domain @xmath219 , and for each @xmath220 @xmath221 [ def : represent ] let @xmath73 be a domain that has @xmath200 as a cusp and @xmath222 let @xmath223 be a quadratic form closed over the domain @xmath224 we will say that @xmath223 represents the cusp of width @xmath69 for @xmath225 if , for any @xmath226 that is supported in @xmath227 we have @xmath228 for such a quadratic form , we will say that an eigenfunction @xmath25 is a cuspform if @xmath181 vanishes on @xmath229 the aim of this section is to prove that being a cuspform is a real - analytic condition . to make this statement precise we have to consider a family @xmath29 of quadratic forms that satisfies the following assumptions . [ ass : qt ] let @xmath230 and @xmath231 . for each @xmath232t_- , t_+[$ ] , let @xmath202 be a positive real - analytic function on @xmath233 let @xmath29 denote a nonnegative , closed quadratic form with domain @xmath234 that represents the cusp of width @xmath202 for @xmath235 lastly , we assume that the family @xmath27 is real - analytic of type ( a ) in the sense of @xcite . that is , for each @xmath236 , the map @xmath32 is real - analytic . a straightforward application of analytic perturbation theory@xcite vii gives the following . [ thm : eigenbasis_exist ] let @xmath27 satisfy the assumptions above [ ass : qt ] . then there exist a collection of real - analytic paths @xmath237 and a collection of real - analytic functions @xmath238 so that for each @xmath12 , the set @xmath239 is an orthonormal basis for @xmath153 , and for each @xmath240 , the function @xmath241 is an eigenfunction of @xmath29 with eigenvalue @xmath242 since the embedding from @xmath243 into @xmath244 is compact , for any @xmath12 the spectrum of @xmath29 consists only in eigenvalues . the proof is then similar to the proof of theorem 3.9 in @xcite viii.3.5 . for @xmath245 , define @xmath246 [ lem : nocf_deriv_criterion ] an eigenfunction @xmath25 of @xmath29 is a cusp form if and only if @xmath247 . @xmath248 is the left - sided derivative of @xmath142 at @xmath249 . since @xmath25 is an eigenfunction , @xmath245 and @xmath142 is a solution to a second order ordinary differential equation on @xmath250 $ ] with @xmath251 . thus , @xmath142 vanishes identically on @xmath250 $ ] if and only if @xmath247 . for real - analytic eigenbranches we have the following . [ lem : lanalytic ] let @xmath252 be an analytic eigenbranch of @xmath29 then the mapping @xmath253 is analytic on @xmath132t_-,t_+[.$ ] the zeroth mode @xmath254 of @xmath30 is a solution to the ode @xmath255 on @xmath256 $ ] with dirichlet boundary condition at @xmath257 denote by @xmath258 the unique solution to this ordinary differential equation that satisfies @xmath259 since the coefficients of the ordinary differential equation depend analytically on the parameter @xmath260 , the mapping @xmath261 is analytic ( with values in @xmath262)$ ] for instance ) . moreover , for each compact set @xmath263t_-,t_+[$ ] we can find @xmath264 such that , for each @xmath265 , @xmath266 . since @xmath30 is a multiple of @xmath267 , we then have @xmath268 analyticity on @xmath269 then follows from the analyticity of @xmath270 and @xmath271 and the choice of @xmath272 . we will say that real - analytic eigenfunction branch @xmath30 of @xmath29 is _ a cusp form eigenbranch _ if and only if for each @xmath273 , the eigenfunction @xmath30 is a cusp form ( see definition [ def : represent ] ) . using the real - analyticity proved in lemma [ lem : nocf_deriv_criterion ] we obtain the following [ coro : cf_discrete ] if @xmath30 is a real - analytic eigenfunction branch that is not a cusp form eigenbranch , then the set of @xmath273 such that @xmath30 is a cusp form is discrete . we now proceed to prove that if @xmath29 has no real - analytic cusp form eigenbranch then , for a generic @xmath12 , the form @xmath29 has no cusp form . as it turns out , we will actually first prove that the spectrum of @xmath29 is generically simple . let @xmath274 denote the set of @xmath273 such that @xmath29 has an eigenspace of dimension at least two . [ prop : mult ] if @xmath29 does not have a real - analytic cusp form eigenbranch , then @xmath274 is countable . let @xmath275 and @xmath276 be as in theorem [ thm : eigenbasis_exist ] . for each @xmath277 , let @xmath278 . since each eigenspace of @xmath29 is spanned by a finite collection of @xmath279 , the union @xmath280 equals @xmath274 . the function @xmath281 is analytic , and hence @xmath278 is either countable or equals @xmath282 . thus to prove the claim , it suffices to show that it is not possible for @xmath283 to equal @xmath282 . suppose that there exists @xmath284 and @xmath40 so that @xmath241 and @xmath285 are real - analytic eigenbranches so that @xmath286 for each @xmath287 . to prove the proposition , it suffices to produce a linear combination @xmath30 of @xmath241 and @xmath285 so that for each @xmath273 , the function @xmath30 is a real - analytic cusp form eikgenbranch . by hypothesis , neither @xmath241 nor @xmath285 are cusp form eigenbranches . by corollary [ coro : cf_discrete ] , the set @xmath288 of @xmath12 such that either @xmath241 or @xmath285 is a cusp form is discrete . for each @xmath289 , define @xmath290 its suffices to show that @xmath291 extends to a real - analytic function on @xmath282 . indeed , since @xmath292 is linear , we have @xmath293 for each @xmath289 . by corollary [ lem : lanalytic ] , the real - analytic extension would satisfy @xmath294 . the order of vanishing of @xmath295 ( resp . @xmath296 ) is finite at each @xmath297 . if the order of vanishing of @xmath298 at @xmath299 is at least the order of vanishing of @xmath300 at @xmath301 , then the ratio @xmath302 has a real - analytic extension near @xmath301 . hence , factorizing @xmath300 we obtain that @xmath303 have real - analytic extensions near @xmath301 . if the order of vanishing of @xmath298 at @xmath301 is at most the order of vanishing of @xmath300 at @xmath301 , then a similar argument applies by factorizing @xmath298 everywhere . thus , @xmath30 extends to a real - analytic cusp form eigenbranch . let @xmath304 denote the set of @xmath273 such that @xmath29 has at least one cusp form eigenfunction . [ alt : alt ] if @xmath29 has no real - analytic cusp form eigenbranch , then @xmath304 is countable . if there exists a cusp form eigenbranch , then @xmath305 . therefore , we have the following dichotomy : either the set @xmath304 countable or the family @xmath27 has a real - analytic cusp form eigenbranch . let @xmath306 be as in theorem [ thm : eigenbasis_exist ] . by corollary [ coro : cf_discrete ] , the zero set @xmath307 is countable . if each eigenspace @xmath102 of @xmath29 is one - dimensional , then there exists a unique @xmath284 such that @xmath102 equals the span of @xmath241 . thus , if @xmath12 does not belong to @xmath274 or to any @xmath308 , then @xmath12 does not belong to @xmath304 . in other words , @xmath309 . by proposition [ prop : mult ] , the set @xmath274 is countable , and hence so is @xmath304 . in this section we use the results of the previous section to explain how theorem [ thm : gennocf ] can be deduced from the existence of a triangle with a cusp that has no nonconstant neumann eigenfunctions . this fact may probably be considered as folklore and follows the general philosophy of using analyticity to prove generic spectral results ( see @xcite ) . the main task here is thus construct a real - analytic family of quadratic forms that is associated with each real - analytic path in the moduli space of triangles . first , we discuss the parametrization of the set of triangles with one cusp . the statement of theorem [ thm : gennocf ] makes use of the fact that hyperbolic triangles with one cusp are parametrized by the two nonzero vertex angles . but in order to prove theorem [ thm : gennocf ] , it will be more convenient to use an alternate set of parameters . for each geodesic triangle @xmath310 in the hyperbolic upper half plane @xmath66 having ( exactly ) one cusp , there exists a unique @xmath3110,1[$ ] and @xmath312 so that @xmath310 is isometric to the domain @xmath313 see figure [ trigeneric ] . the triangle @xmath314 in the upper half plane . ] in this way , the set of hyperbolic triangles may be identified with the euclidean triangle @xmath315 we say that a subset of the set of triangles has measure zero if and only if the corresponding subset of @xmath316 has measure zero . similarly , a subset of the set of triangles is said to be a real - analytic curve if and only if the corresponding subset of @xmath316 is a real - analytic curve . these notions are equivalent to those used in the statement of theorem [ thm : gennocf ] because the relationship between the angles @xmath19 and the parameters @xmath317 is real - analytic . indeed , we have @xmath318 and @xmath319 . see figure [ trigeneric ] . to prove theorem [ thm : gennocf ] , we will apply perturbation theory . the following fact makes this approach feasible . [ prop : allcuspform ] each nonconstant neumann eigenfunction on @xmath314 is a cusp form and hence an eigenfunction of the modified quadratic form @xmath23 . the eigenvalue of a nonconstant neumann eigenfunction on @xmath310 is at least 1/4 @xcite . thus , the claim follows from corollary [ coro : cuspform ] and lemma [ lem : relation ] . let @xmath320 denote the set of @xmath321 such that there exists @xmath322 so that the modified quadratic from @xmath23 has a cusp form . to prove theorem [ thm : gennocf ] it will suffice to show that @xmath320 has measure zero and is a countable collection of nowhere dense sets . to show that @xmath320 is non - generic , we will use analytic perturbation theory and proposition [ alt : alt ] . in order to use analytic perturbation theory we will have to normalize the hilbert space and the domains of the quadratic forms . to accomplish this , we let @xmath323 \times [ 1 , \infty[$ ] and for each @xmath317 we define a @xmath324 diffeomorphism @xmath325 such that onto @xmath200 . the mapping that we define here is more complicated because it must preserve the notion of zeroth fourier coefficient for all @xmath326 above some point . in particular , the vertical displacement of vertical lines should not depend on @xmath327 for large @xmath326 . in @xcite , we considered dirichlet boundary conditions , and in that context there is no need to truncate the zeroth fourier coefficient . ] 1 . the restriction of @xmath328 is the identity for @xmath329 . 2 . for each path @xmath330 , the family @xmath331 is a real - analytic path . to construct @xmath328 , we use the fact that the map @xmath332 defines a fibration of @xmath314 over @xmath333 $ ] and a fibration of @xmath200 over @xmath334 $ ] . we define @xmath328 by sending the fiber over @xmath335 onto the fiber over @xmath336 . [ lem : defnb ] for each @xmath3370,{\underline{\alpha}}[$ ] , there exists a unique cubic polynomial @xmath338 so that * @xmath339 , * @xmath340 . * @xmath341 , * @xmath342 the coefficients of @xmath338 are real analytic for @xmath3430,{\underline{\alpha}}[$ ] . moreover if @xmath344 then , for all @xmath345,$ ] and @xmath346 $ ] , we have @xmath347 . since @xmath348 satisfying the two conditions on @xmath349 is equivalent to the existence of some @xmath136 such that @xmath350 denote by @xmath351 the cubic polynomial defined by @xmath352 by integration , there exists some @xmath353 such that @xmath354 evaluating at @xmath355 and using the condition on @xmath356 we find @xmath357 observe that @xmath358 if @xmath359 and , under this condition , we can solve the last equation on @xmath338 to find @xmath360 we obtain @xmath361\\ & = \frac{1-{\alpha}}{2{\underline{\alpha } } } \left [ { \alpha}^2+{\underline{\alpha}}^2 - 2{\underline{\alpha}}({\alpha}+1)\right],\\ & = \frac{(1-{\alpha})\left [ { \alpha}^2 -2 { \underline{\alpha}}{\alpha}+ { \underline{\alpha}}^2 - 2{\underline{\alpha}}\right ] } { 2{\underline{\alpha } } } \end{split}\ ] ] it follows that we have a unique solution provided @xmath362 and @xmath363 and that the coefficients are real - analytic in @xmath364 we now check the last statement . for @xmath365 , we have @xmath366 so that the claim follows . for @xmath367 we observe that the numerator of @xmath136 is a cubic polynomial that has three roots at @xmath368 thus , if @xmath369 then @xmath16 is the smallest root . since this cubic polynomial is positive for large negative @xmath355 and the denominator also is positive , it follows that @xmath136 is positive for @xmath370 so @xmath371 is a concave function , and by construction @xmath372 and @xmath373 . the claim follows . we will use the notation @xmath374 as well as the notation @xmath375 define @xmath376 by @xmath377 where @xmath378 define @xmath379 by @xmath380 observe that the conditions on @xmath137 imply that @xmath381 and @xmath382 are continuous on @xmath314 so that @xmath328 is @xmath383 we will use this function @xmath328 to normalize the triangle @xmath384 this is made possible by the following lemma . [ lem : bprop ] suppose that @xmath385 for each @xmath386 then the map @xmath328 is a @xmath324 diffeomorphism from @xmath314 onto @xmath200 . + in particular , for each @xmath387 and each @xmath388 , the mapping @xmath328 is a @xmath324 diffeomorphism from @xmath314 onto @xmath389 it suffices to show that the map @xmath390 is a @xmath324 diffeomorphism from @xmath391 onto @xmath392 . by assumption @xmath393 for each @xmath327 . we have @xmath394 , @xmath395 and @xmath396 . since @xmath390 is the identity for @xmath397 , we find that @xmath390 is a @xmath324 diffeomorphism from @xmath391 onto @xmath398 $ ] . for each @xmath399 and each @xmath400 , we define @xmath401 and @xmath402 by @xmath403 we then have [ lem : phi_analytic ] for each @xmath404 , each of the following maps is analytic on @xmath401 : 1 . @xmath405 2 . @xmath406 3 . @xmath407 if , for each @xmath408 the assumption of lemma [ lem : bprop ] holds then the map @xmath409 is also analytic on @xmath402 . + moreover , each restriction extends analytically to an open neighbourhood . the coefficients of the cubic polynomial @xmath338 depend analytically on @xmath410 and hence @xmath411 is analytic . the map @xmath412 is analytic and hence it follows that map ( 1 ) is analytic . maps ( 2 ) and ( 3 ) are therefore analytic . since @xmath411 is analytic and @xmath413 for @xmath414 , the implicit function theorem ( theorem 2.1.2 in @xcite ) implies that there exists a function @xmath415 which is analytic and a solution to @xmath416 we then have @xmath417 and , since @xmath418 is analytic , the claim follows . in the rest of the section , @xmath387 will be fixed so that we can use lemmas [ lem : bprop ] and [ lem : phi_analytic ] . we use the family of diffeomorphisms @xmath328 to define a quadratic form @xmath29 with domain @xmath419 that is unitarily equivalent to @xmath23 on @xmath420 . define @xmath421 by @xmath422 where @xmath423 is the operator that returns the jacobian matrix of a map . [ lem : domain ] the isometry @xmath424 is a unitary isomorphism from @xmath425 onto @xmath426 and it maps @xmath234 onto @xmath427 . + on functions that are supported in @xmath428 , @xmath424 coincides with @xmath218 we have @xmath429 it follows that @xmath424 is a unitary isomorphism from @xmath425 onto @xmath426 . let @xmath430 since @xmath328 is a @xmath431 diffeomorphism and @xmath432 is continuous on @xmath314 and smooth away from @xmath24 , then @xmath433 is continuous and in @xmath434 . the jump formula implies that @xmath435 + since , for @xmath436 , @xmath437 , the last statement is a direct verification . define the quadratic form @xmath438 on @xmath439 by @xmath440 [ lem : cuspequivalence ] @xmath25 is a cusp form for @xmath438 if and only if @xmath441 is a cusp form for @xmath26 on @xmath314 . if @xmath442 , then @xmath443 . it follows that if @xmath442 , then @xmath152 if and only if @xmath444 . for @xmath445 , the function @xmath446 is a solution to a second order ordinary differential equation , and hence @xmath444 for @xmath447 if and only if @xmath444 for @xmath448 . it will be convenient to have the following alternate form for @xmath29 . [ prop : alternate ] we have @xmath449 where @xmath450 is defined by @xmath451 and @xmath452 is defined by @xmath453 moreover , @xmath438 represents the cusp of width @xmath69 for @xmath235 this is a straightforward calculation using the chain rule and the change of variables formula . let @xmath454t_-,t_+[$ ] and let @xmath330 be a real - analytic path . [ thm : analyticity ] the family quadratic forms @xmath455 is analytic of type ( a ) in the sense of @xcite . for each @xmath12 , the quadratic form @xmath456 is a closed form with domain @xmath457 . it suffices to show that for each @xmath236 , the function @xmath458 is real - analytic . by proposition [ prop : alternate ] , we have @xmath459 where @xmath460 if @xmath461 \times [ \underline{\alpha } , \infty[$ ] , then the matrix @xmath462 is given by @xmath463 and @xmath464 . thus , the second integral on the right of ( [ eq : splitintegral ] ) depends analytically on @xmath12 . it remains to consider the integral over @xmath334\times [ 1,\underline{\alpha}]$ ] . the integrand @xmath465 can be expanded into a finite sum of terms of the form @xmath466 where @xmath467 is a function that is obtained by multiplying @xmath468 or its derivatives and the entries of @xmath469 and @xmath69 is one of the @xmath470 functions obtained by making the product @xmath471 where both @xmath472 are either @xmath25 or one of its partial derivatives . by lemma [ lem : phi_analytic ] , the coordinates of @xmath328 and @xmath473 are analytic functions of @xmath317 . it follows that @xmath474 and @xmath475 are analytic ( in a neighborhood of @xmath476 \times [ 1 , \underline{\alpha}]$ ] ) . in all possible choices , the function @xmath467 then is analytic . the analyticity of @xmath458 follows from lemma [ lem : intanalytic ] below . [ lem : intanalytic ] if @xmath477 \times [ 1 , \underline{\alpha}]$ ] is analytic , then for each @xmath478 \times [ 1 , \underline{\alpha}])$ ] , the function @xmath479 is analytic on @xmath282 . there exists an open neighborhood @xmath480 of @xmath476 \times [ 1 , \underline{\alpha}]$ ] such that the map @xmath481 extends to a holomorphic function on @xmath482 . since @xmath334 \times [ 1,\underline{\alpha}]$ ] is compact , @xmath483 converges uniformly to @xmath484 as @xmath12 approaches @xmath140 . it follows that the ( complex ) @xmath12-derivative of the map in ( [ eq : intanalytic ] ) exists at each @xmath485 . given theorem [ thm : analyticity ] , we now explain why the generic triangle @xmath314 has no cusp forms provided that one triangle has none . if there exists a point @xmath486 such that @xmath26 on @xmath487 has no nonconstant eigenfunction , then @xmath320 has measure zero and is a countable union of nowhere dense sets . by proposition [ prop : allcuspform ] , the quadratic form @xmath23 on @xmath487 has no cusp form , and hence by lemma [ lem : cuspequivalence ] , the quadratic form @xmath488 has no cusp form . to show that @xmath320 has measure zero , we apply fubini s theorem in a fashion similar to @xcite : let @xmath489 and apply lemma [ alt : alt ] to find that the set @xmath137 of @xmath69 such that @xmath490 is countable . for each @xmath491 , let @xmath492 and apply lemma [ alt : alt ] to find that the intersection @xmath493 of the line @xmath494 with @xmath320 is countable . hence for each @xmath491 , the set @xmath493 has measure zero with respect to the linear measure @xmath495 . hence , the measure of @xmath320 equals the measure of @xmath496 . since @xmath137 is countable , the measure equals zero . for @xmath497 , let @xmath498 be the set of @xmath321 such that @xmath26 on @xmath499 has a cusp form with eigenvalue at most @xmath500 . using the continuity of @xmath501 and the continuity of linear functional @xmath292 , one can show that @xmath498 is a closed . thus , it suffices to show that @xmath498 is nowhere dense . given a point @xmath502 , let @xmath503 \rightarrow { { \mathcal m}}$ ] be a real - analytic path joining @xmath504 to @xmath317 . since @xmath23 on @xmath487 has no cusp forms , the family @xmath455 has no cusp form eigenfunction branch . it follows from lemma [ alt : alt ] , that for each open neighborhood @xmath482 of @xmath317 , there exists @xmath505 $ ] such that @xmath506 and @xmath507 has no cusp forms . hence @xmath498 is nowhere dense . in the remainder of this paper we consider the specific family of triangles @xmath508 defined in the introduction . in particular , we will study the spectral properties of @xmath509 for small @xmath12 . the family @xmath509 of quadratic forms does not converge as @xmath12 tends to zero nor do its real - analytic eigenbranches . but a simple renormalization will give convergence . fix @xmath322 and @xmath399 such that @xmath510 . let @xmath137 be the function defined in lemma [ lem : defnb ] . when @xmath355 tends to @xmath16 , the function @xmath511 converges to @xmath16 uniformly for @xmath512.$ ] thus , there exists @xmath513 such that if @xmath514 and @xmath515 then @xmath516 choose @xmath301 such that @xmath517 then , for each @xmath518 and each @xmath519 @xmath520 so that @xmath521 we may thus use lemmas [ lem : bprop ] and [ lem : phi_analytic ] . the methods and results of section [ subsec : analyticity ] then apply and we define the quadratic form @xmath509 as previously . for each @xmath522 , define the renormalized quadratic form by @xmath523 with domain @xmath234 . by theorem [ thm : analyticity ] , the family @xmath27 is real - analytic of type ( a ) for @xmath40,1[$ ] . in particular , the results of [ subsec : generic ] apply . to study the limiting properties of the family @xmath29 , we re - express @xmath29 in a more convenient form : for each @xmath324 function @xmath524 define @xmath525 recall that @xmath526 is the inverse of @xmath338 and set @xmath527 . define @xmath528 and @xmath529 where the subscript ( or first argument ) in each @xmath530 and @xmath137 is @xmath531 . when comparing @xmath532 and @xmath533 ( @xmath469 and @xmath534 ) we see that we only miss some powers of @xmath12 than eventually cancel in the computation leading to proposition [ prop : alternate ] . this shows that for each @xmath236 @xmath535 where @xmath536\times [ 1 , \underline{\alpha}]$ ] and @xmath537\times [ \underline{\alpha } , \infty]$ ] . by arguing as in the proof of theorem [ thm : analyticity ] , one can show that @xmath538 is analytic at @xmath33 . is not closed on the domain @xmath234 , the family @xmath29 is _ not _ analytic at @xmath33 in the sense of @xcite . ] we will now compute the first few terms in the taylor series in @xmath12 for @xmath539 and @xmath540 these functions are analytic on a neighbourhood of @xmath541 in particular , in the following , the expressions like @xmath542 are uniform with respect to @xmath543 and may be differentiated with respect to @xmath544 and @xmath545 . we first compute @xmath546 and @xmath547 since @xmath548 is analytic and @xmath549 , it follows from ( [ eq : fexp ] ) that @xmath550 moreover , using analyticity , this asymptotic expansion may be differentiated with respect to @xmath551 we thus obtain , @xmath552 substituting these into ( [ defn : rhotilde ] ) , and differentiating , we find that @xmath553 using ( [ eq : fexp ] ) , ( [ eq : fderivexp ] ) , ( [ exp : yalpha ] ) , and ( [ exp : yderivalpha ] ) we find that @xmath554 where @xmath282 is the identity matrix , @xmath542 is a matrix whose operator norm is bounded by a constant times @xmath31 as @xmath12 tends to zero , and @xmath555 is the polynomial @xmath556 to prove theorem [ thm : main ] we will need to know that @xmath557 . [ lem : p_positive ] @xmath558 . by construction we have @xmath559 by differentiating with respect to @xmath355 and setting @xmath365 we get @xmath560 since @xmath561 the claim follows . in this section we apply the method of asymptotic separation variables developed in @xcite to the family @xmath29 . using the small @xmath12 asymptotics derived in [ section : expansion ] , we approximate @xmath29 to first order with a family of quadratic forms @xmath35 for which separation of variables apply . we also derive a non - concentration estimate for eigenfunctions of @xmath29 . in this section and the following sections , we will use @xmath562 in place of @xmath563 as coordinates for @xmath564 \times [ 1 , \infty[$ ] and unless it is specified otherwise @xmath565 is the norm in @xmath566 we begin by using the expansions obtained in [ section : expansion ] to determine the forms used to approximate @xmath29 . in particular , by substituting the expansions ( [ exp : rho ] ) and ( [ exp : q ] ) into ( [ eq : qdefn ] ) we are led to define @xmath567 and @xmath568 where the operator @xmath569 is defined by ( [ eq : gradient ] ) , @xmath536 \times [ 1 , \underline{\alpha}]$ ] , and @xmath570 we wish to show that @xmath29 is asymptotic to @xmath35 at first order in the sense of @xcite . it will also be used to help derive a key estimate for crossing eigenbranches . however , although @xmath35 is a positive quadratic form , the bottom of its spectrum tends to @xmath571 so that it is more convenient to use the quadratic form @xmath572 that we now define to control quantities . [ def : tildea ] the quadratic form @xmath572 is defined on @xmath573 by @xmath574 the following proposition can be seen as the beginning of an asymptotic expansion for @xmath575 [ prop : q - a - tb ] there exists @xmath353 such that for each @xmath576 @xmath577 we have @xmath578 if @xmath579 , then @xmath580 is identically equal to @xmath16 and @xmath581 is identically equal to @xmath282 . hence , by subsituting ( [ eq : nablarhow ] ) into ( [ eq : qdefn ] ) , we find that @xmath582 is the sum of the following four terms @xmath583 in order to estimate these four terms , we use the asymptotic expansions of [ section : expansion ] . for example , by ( [ exp : q ] ) we have that ( [ leading ] ) is equal to @xmath584 since the operator norm of the matrix @xmath542 is bounded by a constant @xmath353 times @xmath31 , we can apply the cauchy - schwarz inequality to find that the norm of ( [ exp : q ] ) is bounded by @xmath585 . similar arguments show that there is a constant @xmath353 so that * ( [ mixed ] ) is bounded above by @xmath586 * ( [ mixed2 ] ) is bounded above by @xmath587 * ( [ uv ] ) is bounded above by @xmath588 the claim follows . we recall the fourier decomposition of section [ sec : fourier ] . since now @xmath589 we thus have , for each @xmath590 @xmath591 where the latter makes sense for almost every @xmath326 and defines an element of @xmath592 as above , let @xmath593 denote the set of functions @xmath594 such that @xmath77 is the restriction of a compactly supported , smooth function defined in a neighborhood of @xmath200 . if @xmath595 , then each @xmath119 is smooth , and a straightforward computation shows that is the operation defined by @xmath596 . ] @xmath597 we define @xmath598 to be the set of compactly supported , smooth functions defined on @xmath599 . for @xmath600 , each integer @xmath40 , and each @xmath601 , we define @xmath602 for @xmath603 in @xmath604 , the inner product is defined by @xmath605 let @xmath606 denote the subspace consisting of those functions whose support lies in @xmath607 $ ] . for each @xmath126 , the quadratic form @xmath608 extends to a closed , densely defined form on the completion of @xmath609 with respect to @xmath610 . for @xmath611 , we will restrict the domain of @xmath608 to be the completion of those smooth functions whose support lies in @xmath607 $ ] . if @xmath25 is an eigenfunction of @xmath35 with eigenvalue @xmath260 , then for each @xmath77 in the domain of @xmath608 , we have @xmath612 and hence @xmath119 is an eigenfunction of @xmath608 with eigenvalue @xmath260 with respect to @xmath613 . thus , each eigefunction @xmath25 of @xmath35 may be written uniquely as @xmath614 where the @xmath120 fourier coefficient @xmath119 is an eigenfunction of @xmath608 . moreover , the spectrum of @xmath35 with respect to @xmath615 is the union of the spectra of @xmath608 with respect to @xmath613 . in what follows we will often suppress the subscript @xmath326 from the notation . the next two lemmas identify the eigenfunctions of @xmath608 for each @xmath40 . we begin with the nonzero modes . [ lem : nonzero_fourier ] for each @xmath12 and @xmath40 and eigenvalue @xmath260 of @xmath608 with respect to @xmath613 , the associated eigenspace consists of functions of the form @xmath616 such that @xmath617 moreover , when @xmath12 varies , the spectrum is organized into eigenvalues branches @xmath618 each of these is a growing function of @xmath12 and @xmath619 integrate by parts as in the proof of lemma [ zerolemma ] , make the change of variables @xmath620 , and use the boundary conditions . the second part follows from theorem [ thm : eigenbasis_exist ] . the zero modes are given by the following lemma . [ zerolemma ] the spectrum of @xmath621 with respect to @xmath613 is the set @xmath622 the eigenspace associated to @xmath623 is spanned by the eigenvector @xmath624 suppose that @xmath77 is an eigenfunction , that is @xmath625 for all @xmath69 . this implies first that @xmath626 holds in the distributional sense . ellipticity then yield that @xmath77 is smooth . moreover , by integrating by parts against a smooth function that is identically equal to @xmath16 near @xmath627 , we also find that @xmath628 let @xmath140 be such that @xmath629 . then two linearly independent solutions are given by @xmath630 and @xmath631 if @xmath632 and by @xmath633 and @xmath634 if @xmath635 . the condition that @xmath636 is real and nonnegative implies either that @xmath637 with @xmath638 , that @xmath639 or that @xmath635 . if @xmath640 , then the boundary conditions @xmath641 and @xmath642 imply that the solutions take the form given in in ( [ zeroeigenvector ] ) with @xmath643 . if @xmath644 $ ] , then there are no solutions that satisfy the boundary conditions . as a consequence of the identification of the eigenfunctions we have the following poincar type inequality . [ lem : poincare ] for each @xmath645 and each @xmath220 we have @xmath646 we have @xmath647 lemmas [ lem : nonzero_fourier ] and [ zerolemma ] imply @xmath648 and @xmath649 for @xmath650 . in the sequel we will use different kind of projections associated either with the fourier decomposition @xmath651 or with the spectral decomposition of @xmath652 more precisely , for each @xmath653 define the orthogonal projection @xmath654 by @xmath655 and let @xmath656 denote the image of @xmath657 . for each @xmath658 , we define @xmath659 the projection onto @xmath660 . we also define @xmath661 to be the orthogonal projection onto the eigenspace of @xmath35 associated to the eigenvalue @xmath260 of @xmath35 . for a fixed interval @xmath282 , define the @xmath35-spectral projection in the energy interval @xmath282 to be @xmath662 for each eigenvalue @xmath260 of @xmath35 , the associated @xmath35-eigenspace @xmath663 is the orthogonal direct sum @xmath664 where @xmath665 is the @xmath260-eigenspace of @xmath666 and @xmath667 is the span of @xmath668 . it follows that @xmath669 more generally , for a quadratic form @xmath545 the notation @xmath670 will always denote the spectral projection onto the interval @xmath233 we prove that @xmath35 and @xmath29 are asymptotic at first order in the sense of @xcite . in the following , we let @xmath671 ( resp . @xmath672 ) denote the derivative of @xmath29 ( resp . @xmath35 ) in @xmath12 . [ prop : q - a_asymp ] there exists a constant @xmath353 and @xmath301 such that , for all @xmath673 and all @xmath674 @xmath675 one argues as in the proof of proposition [ prop : q - a - tb ] paying a little more attention to the terms ( [ mixed ] ) , ( [ mixed2 ] ) , and ( [ uv ] ) . for example , to estimate ( [ mixed ] ) , use the cauchy - schwarz inequality and ( [ exp : rho ] ) to obtain @xmath676 the cauchy - schwarz inequality and lemma [ zerolemma ] give @xmath677 by combining ( [ est : nabla_rho ] ) and ( [ est : v_nabla ] ) and using ( [ exp : rho ] ) we find that @xmath678 switching the roles of @xmath25 and @xmath77 , we obtain the same bound for the expression in ( [ mixed2 ] ) . similar methods apply to bound the other terms . the estimate for @xmath679 is obtained in a similar way . since @xmath29 is asymptotic to @xmath35 at first order and @xmath35 and @xmath672 are nonnegative quadratic forms , each real - analytic eigenvalue branch @xmath41 of @xmath29 converges to a finite limit @xmath680 as @xmath12 tends to zero ( theorem 3.4 of @xcite ) . for the dirichlet eigenvalue problem on @xmath0 , we showed in @xcite that each limit @xmath680 has the form @xmath42 where @xmath40 is an integer . the methods of @xcite can be applied to show that the same fact is true in the present context . in this section we highlight the necessary modifications . we also show that if the eigenvalue branch is associated to a cusp form , then @xmath40 must be positive . this latter fact will be used crucially in the proof of theorem [ thm : main ] . the proof of convergence depends crucially on the following ` non - concentration ' result proved for the dirichlet problem in @xcite . we will denote by @xmath681 the domain of the quadratic form @xmath682 and by @xmath683 the quadratic form defined for @xmath684 by @xmath685 [ prop : real_noncon](compare proposition 9.1 of @xcite ) let @xmath653 , let @xmath269 be a compact subset of @xmath132(\pi \ell)^2 , \infty[$ ] , and let @xmath686 . there exist positive constants @xmath301 and @xmath687 ( that only depend on @xmath688 and @xmath353 ) so that if @xmath689 , if @xmath518 , and if for each @xmath690 , the function @xmath691 satisfies @xmath692 then @xmath693 if @xmath694 this is proposition 9.1 of @xcite with @xmath695 and @xmath696 . if @xmath697 and we let @xmath698 , then ( [ nonconcentrate ] ) holds . in the language of semi - classical analysis , propostion [ prop : real_noncon ] asserts that a quasimode @xmath77 of order @xmath12 at energy @xmath102 does not concentrate at @xmath699 if @xmath700 . in 12 of @xcite , we used non - concentration to derive indirect estimates for @xmath701 . the following proposition and corollary make these estimates more transparent and simpler to apply . [ prop : noncon ] let @xmath653 and let @xmath702 ( \pi \ell)^2,\infty[$ ] be compact . for each @xmath703 , there exists @xmath704 and @xmath705 such that for each @xmath691 and @xmath518 @xmath706 where @xmath707 from ( [ defn : ak ] ) we find that @xmath708 and hence @xmath709 if the claim is not true , then for each @xmath704 , there exists a sequence @xmath710 tending to zero and sequences @xmath711 @xmath712 such that @xmath713 in particular , since @xmath714 , we have @xmath715 . it follows that for each @xmath716 @xmath717 we fix @xmath718 such that @xmath719+k \subset ( ( \ell \pi)^2 , \infty),$ ] and we set @xmath720}(\tilde{v}_n ) = p_{{\widetilde{a}_t^\ell}}^{[e_n+1-\delta , e_n+1+\delta]}(\tilde{v}_n).$ ] reasoning as in the proof of lemma 2.3 in @xcite we find that @xmath721 observe that the sequences @xmath722 and @xmath723 depend on the initial choice of @xmath724 but the preceding estimate gives a constant @xmath353 that is independent of @xmath724 such that @xmath725 this implies in particular @xmath726 so that , for @xmath727 large enough we have @xmath728 in equation ( [ eq : tvnisqm ] ) we replace the test function @xmath69 by @xmath729}(w)$ ] and use that the spectral projector is self - adjoint and commutes with @xmath730 we obtain that for each @xmath731 , @xmath732}(w ) \right)^{\frac{1}{2}}\\ & \leq c \cdot t_n \cdot \| v_n\| \|w\| , \end{split}\ ] ] where we have used that @xmath733 is controlling @xmath734 and that @xmath735}(w)\right ) \,\leq\ , ( \sup(k)+\delta)\|w\|^2\ ] ] by definition of a spectral projector . since @xmath736 and @xmath737 is a non - negative quadratic form , we also have @xmath738 equation ( [ eq : vnisqm ] ) implies that we may use proposition [ prop : real_noncon ] to find @xmath739 since ( [ eq : vnisqm ] ) also implies @xmath740 using ( [ eq : adot_express ] ) we find that @xmath741 on the other hand , the contradiction assumption implies that @xmath742 the implied constant @xmath353 does not depend on @xmath724 so if we take @xmath743 then choosing @xmath744 small enough yields the contradiction . this proposition yields an estimate for @xmath745 from below in terms of the projection @xmath746 [ coro : noncon ] let @xmath48 and let @xmath747 be a compact subset of @xmath132(\pi k)^2 , \infty[$ ] . for each @xmath748 , there exists @xmath749 @xmath705 such that if @xmath689 , @xmath750 , and @xmath751 , then @xmath752 where @xmath753 the functional @xmath754 is equivalent to the @xmath755-norm of @xmath756 where here @xmath757 is the operator such that @xmath758 for each @xmath759 . since @xmath672 is block diagonal with respect to the sum @xmath760 and @xmath761 , we have @xmath762 we may apply proposition [ prop : real_noncon ] with @xmath763 to each term on the right hand side to find that @xmath764 where @xmath687 is the minimum of the @xmath724 coming from proposition [ nonconcentrate ] . for each @xmath765 , and @xmath766 , we have @xmath767 and hence @xmath768 . we also have , @xmath769 , and the claim follows . the bounds proved in [ sec : noncon ] depend on a bound on @xmath770 . in this subsection , we show that if @xmath69 is an @xmath35-spectral projection of a @xmath29-eigenfunction in an interval containing the eigenvalue @xmath102 , then @xmath770 is of order @xmath12 . we start with a real - analytic eigenfunction branch @xmath30 for @xmath29 with associated real - analytic eigenvalue branch @xmath41 . we let @xmath771 denote the associated spectral projection . let @xmath680 denote the limit of @xmath41 as @xmath12 tends to zero . the following two lemmas express the fact that the projection @xmath772 is an order @xmath12 quasimode for @xmath35 at energy @xmath41 . first we have a lemma that is comparable to lemma 2.3 in @xcite . [ lem : u_w_close ] if @xmath282 is a compact interval whose interior contains @xmath680 , then there exist @xmath705 and @xmath353 such that if @xmath518 , then @xmath773 using the fact that @xmath30 is an eigenfunction of @xmath29 and that @xmath35 and @xmath29 are asymptotic at first order , for each @xmath774 @xmath775 observe that letting @xmath776 yields that @xmath777 moreover the former equation can be rewritten as @xmath778 where $ \tilde{e}_t:= e_t+1.$ we may now follow the proof of lemma 2.3 in \cite{hj10 } observing that $ p_{a_t}^i = p_{{\widetilde{a}_t}}^{i+\{1\}}.$ this yields a constant $ c$ such that \begin{equation * } \begin{split } a_t(u_t - w_t^i)\,+\,\|u_t - w_t^i\|^2\ , & \leq \ , c\cdot t^2 \cdot { \widetilde{a}_t}(u_t)\\ & \leq \frac{c\cdot t^2}{1-c\cdot t^2 } \|u_t\|^2,\\ & \leq c'\cdot t^2 \|u_t\|^2 . \end{split}\ ] ] the claim follows . observe that by orthogonality the preceding lemma also imples the following estimate that roughly expresses the fact that all the mass of @xmath30 is supported by @xmath779 @xmath780 [ lem : quasimode ] if @xmath282 is a compact interval whose interior contains @xmath680 , then there exist @xmath705 and @xmath353 such that if @xmath518 , then @xmath781 for each @xmath782 we have @xmath783 so that the cauchy - schwarz inequality and the preceding lemma imply @xmath784 we also have using cauchy - schwarz and the preceding lemma . @xmath785 we now start again from ( [ eq : afo ] ) . first in the bounding term , we have already seen that we could replace @xmath786 by @xmath787 thus from the triangle inequality , ( [ eq : afo ] ) and the two preceding estimates we obtain @xmath788 the claim follows using ( [ eq : normwequivnormu ] ) . lemma [ lem : quasimode ] has the following corollary that expresses , in the language of semiclassical analysis , that @xmath779 is an order @xmath12 quasimode . [ coro : wisqm ] if @xmath282 is a compact interval that contains @xmath680 there exists @xmath353 and @xmath705 such that , for @xmath518 and each @xmath789 , we have @xmath790 since @xmath791 is a spectral projector , we have @xmath792 and hence lemma [ lem : quasimode ] implies @xmath793 since @xmath794 , the claim follows . by combining lemma [ lem : quasimode ] with corollary [ coro : noncon ] we prove the following . [ thm : limitsq ] let @xmath47 be an eigenbranch of @xmath29 then there exists @xmath795 such that @xmath796 suppose to the contrary that @xmath680 is not of the form @xmath797 where @xmath40 is an integer . let @xmath798 . choose a compact interval @xmath799(n-1)^2\pi^2,n^2\pi^2[$ ] whose interior contains @xmath680 . let @xmath30 be a real - analytic eigenfunction branch of @xmath29 associated to @xmath41 . let @xmath800 be the projection of @xmath30 onto the modes of @xmath35 that have energy lying in @xmath282 . if @xmath801 , then each eigenvalue of @xmath666 is at least @xmath802 . ( this follows from lemma [ lem : lambda_asymp ] or simply the nonnegativity of @xmath672 . ) thus , since @xmath803 , equation ( [ eq : projection_relation ] ) implies that @xmath804 let @xmath353 be as in lemma [ lem : quasimode ] , and apply lemma [ coro : noncon ] with @xmath805 to obtain @xmath687 so that @xmath806 it follows that @xmath807 is not integrable . this contradicts theorem 4.2 of @xcite . the next proposition will be the starting point of the contradiction argument in the following sections . it says that a cusp form eigenbranch can not limit to @xmath571 . heuristically , the zeroth fourier coefficient of a cusp form vanishes identically whereas an eigenvalue branche that limit to @xmath571 must eventually live have nontrivial zeroth fourier mode . however , because we have made a nontrivial change of variable , this fact requires an argument . [ prop : cflimnot0 ] if @xmath41 is a real - analytic cusp form eigenvalue branch of @xmath29 , then the integer @xmath40 appearing in ( [ eq : elimit ] ) is positive . suppose to the contrary that @xmath808 . set @xmath809 $ ] and consider @xmath810 defined in ( [ defn : wti ] ) . if @xmath694 , the restriction of @xmath811 to @xmath656 is bounded below by @xmath812 , thus we have @xmath813 on the other hand , the projection of @xmath30 onto @xmath814 equals @xmath815 . let @xmath816 denote the projection of @xmath817 onto @xmath59 . since each @xmath656 is a direct sum of eigenspaces of @xmath811 , we have @xmath818 the quadratic form @xmath811 is nonnegative and the restriction of @xmath35 to @xmath814 is bounded below by @xmath819 . hence @xmath820 . by lemma [ lem : u_w_close ] we have @xmath821 therefore , @xmath822 , and hence @xmath823 for small @xmath12 . lemma [ lem : cuspequivalence ] implies that @xmath824 is a cusp form for @xmath26 . in particular , for each @xmath825 $ ] @xmath826 where @xmath827 and @xmath828 . inspection shows that there exists @xmath353 such that @xmath829\times[1,\beta]\right \ } \ , \leq \ , c \cdot t \ ] ] for small @xmath12 . therefore , using the cauchy - schwarz inequality we find that @xmath830 using the fundamental theorem of calculus and the cauchy - schwarz inequality we find that @xmath831 thus , since there exists @xmath832 so that @xmath833\times[1,\beta]\right \ } \leq c ' \cdot t\ ] ] for small @xmath12 , we have @xmath834 since @xmath35 and @xmath29 are asymptotic at first order @xmath835 for sufficiently small @xmath12 . by combining these estimates , we find that @xmath836 since @xmath837 we can combine ( [ eq : zero_zero ] ) , ( [ est : etaminus1 ] ) and ( [ est : gradient ] ) to find that @xmath838 squaring and integrating over @xmath825 $ ] , we find that @xmath839 without loss of generality , we have @xmath840 . therefore , if @xmath41 limits to zero , then ( [ est : zerosmall ] ) will contradict ( [ est : zerolarge ] ) for small @xmath12 . we can also use the nonconcentration of the spectral projection to give an @xmath841 lower bound on the first variation @xmath842 in the case that the projection onto the small modes is significant . [ prop : adotbound ] let @xmath282 be a compact interval whose interior contains @xmath680 and @xmath843 . for each @xmath844 , there exists @xmath704 and @xmath705 so that if @xmath845 and @xmath846 then @xmath847 using the cauchy - schwarz inequality and the nonnegativity of @xmath672 we have @xmath848 it follows from ( [ eq : defa_t ] ) that for all @xmath849 @xmath850 and hence @xmath851 . moreover , by combining this with lemma [ lem : u_w_close ] we find @xmath852 thus , from ( [ est : adotu_adotw ] ) we obtain @xmath853 hence by applying lemma [ lem : u_w_close ] we have @xmath854 for @xmath12 sufficiently small . as in the proof of theorem [ thm : limitsq ] , we have @xmath855 . since @xmath701 is non - negative and ` block - diagonal ' we have @xmath856 let @xmath353 be as in lemma [ lem : quasimode ] , and apply lemma [ coro : noncon ] with @xmath857 to obtain @xmath687 so that @xmath858 for @xmath12 sufficiently small . estimate ( [ eq : normwequivnormu ] ) implies that @xmath859 for @xmath12 small , and therefore by combining the above inequalities , we prove the claim . in contrast to proposition [ prop : adotbound ] , we have the following . [ edotupperestimate ] there exists @xmath705 and @xmath353 such that if @xmath845 , then @xmath860 it follows from ( [ eq : defa_t ] ) that for all @xmath849 @xmath850 from ( [ eq : q - a ] ) , there exists @xmath353 so that for sufficiently small @xmath12 @xmath861 and @xmath862 thus , if @xmath30 is the real - analytic eigenfunction branch of @xmath29 associated to @xmath41 , then @xmath863 by choosing @xmath301 sufficiently small , we obtain the claim . proposition [ alt : alt ] reduces the proof of theorem [ thm : main ] to the following . [ thm : nobranch ] the family @xmath27 does not have a real - analytic cusp form eigenbranch . the proof of theorem [ thm : nobranch ] will be by contradiction . we will assume that there exists a real - analytic cusp form eigenvalue branch , @xmath41 . by the results of [ section : branches ] , we have @xmath864 where @xmath40 is positive . we aim to contradict the positivity of @xmath40 . the proof of theorem [ thm : nobranch ] will rely on the estimates of solutions to ordinary differential equations made in appendix [ appendix : airyb ] . to make the estimates less tedious , we will choose @xmath249 to be sufficiently close to 1 , where ` sufficiently close ' will be determined by the integer @xmath40 that appears in ( [ eq : limitk ] ) . however , the construction of the quadratic form @xmath29 depends on @xmath249 . therefore , the integer @xmath40 that appears in ( [ eq : limitk ] ) depends a priori on @xmath249 . in order to avoid circularity of reasoning , we will prove the following . [ prop : choosebeta ] let @xmath41 be a real - analytic eigenvalue branch associated to a real - analytic cusp form eigenfunction branch of @xmath865 . for each @xmath866 , the family @xmath867 has a real - analytic cusp form eigenfunction branch with associated eigenvalue branch @xmath41 .. ] for each fixed @xmath12 , since @xmath41 corresponds to a cusp form , it belongs to the spectrum of @xmath868 for all @xmath869 ( see lemma [ lem : betacf]).by theorem [ thm : eigenbasis_exist ] , there exists a real - analytic eigenvalue branch @xmath870 of @xmath871 such that @xmath872 . since @xmath873 has only countably many real - analytic eigenvalue branches , there exists some branch @xmath870 such that the set of @xmath874 with @xmath875 has an accumulation point . thus , by real - analyticity , we have @xmath875 for all @xmath874 . if for each @xmath12 , the dimension of the eigenspace @xmath876 of @xmath877 associated to @xmath878 is greater than one , then one can argue as in the proof of proposition [ prop : mult ] to obtain a real - analytic cusp form eigenfunction branch of @xmath877 associated to @xmath41 . otherwise , by real - analyticity , for each @xmath12 in the complement of discrete set @xmath136 of @xmath12 we have @xmath879 . let @xmath880 be a real - analytic eigenfunction branch of @xmath877 associated to @xmath41 . let @xmath881 denote a real - analytic eigenfunction branch of @xmath882 associated to @xmath41 . for each @xmath12 , the pull - back of @xmath30 by the diffeomorphism @xmath883 is a cusp form of @xmath26 on @xmath0 . in turn , for each @xmath12 , the pull - back of @xmath884 by @xmath885 is a cusp form eigenfunction of @xmath877 . hence , if @xmath886 , then the eigenfunction @xmath887 is a cusp form for @xmath886 . thus , by corollary [ coro : cf_discrete ] , the branch @xmath880 is a real - analytic cusp form eigenbranch of @xmath877 . as a consequence of proposition [ prop : choosebeta ] , we may fix @xmath249 to satisfy is most probably not necessary but it will simplify the arguments in the appendix . in particular it implies that , on @xmath888 $ ] and for @xmath889 , the sturm - liouville equations associated with @xmath890 have no turning point . ] @xmath891 it follows that for each @xmath889 and @xmath892 $ ] we have @xmath893 as soon as @xmath12 is small enough . in what follows , we will drop @xmath249 from the notation for @xmath882 . in this section we show that there exists a real - analytic eigenvalue branch @xmath894 of @xmath608 such that @xmath895 is at most of order @xmath12 . in [ sec : relative ] , we will show to the contrary that @xmath895 is at least of order @xmath896 . this will provide the desired contradiction . [ tracking ] if @xmath41 is a cusp form eigenvalue branch with positive limit @xmath58 , then there exists @xmath705 , @xmath686 , and a real - analytic eigenvalue branch @xmath49 of @xmath608 so that for each @xmath518 , @xmath897~ = ~ \{\lambda^*_t\}. \ ] ] it suffices to show that there exist positive constants @xmath301 and @xmath353 so that the distance from @xmath41 to the spectrum of @xmath608 is at most @xmath898 for @xmath518 . indeed , if this were so , then for each @xmath12 we would have an eigenvalue @xmath894 of @xmath608 so that @xmath899 . suppose that there existed a sequence @xmath900 and eigenvalues @xmath901 of @xmath902 so that @xmath903 . then we would have @xmath904 this would contradict the ` super - separation ' of eigenvalues given in theorem 10.4 in @xcite . hence there exists @xmath905 such that if @xmath906 , then at most one eigenvalue @xmath894 of @xmath608 satisfies @xmath899 . since the eigenvalues of the family @xmath608 belong to real - analytic branches , the function @xmath907 would be real - analytic . we first proceed to prove that there exists a sequence @xmath744 going to @xmath571 such that @xmath41 is at a distance of order @xmath12 of the spectrum of @xmath908 let @xmath30 denote a real - analytic eigenfunction branch associated to @xmath41 . by corollary [ coro : wisqm ] and the fact that @xmath909 is an orthogonal projection that commutes with @xmath35 there exists constants @xmath910 and @xmath301 such that for each @xmath77 and @xmath911 then @xmath912 let @xmath137 denote the set of @xmath601 so that the distance from @xmath41 to spectrum of @xmath608 is at least @xmath913 , where @xmath910 is the constant that appears in ( [ quasimodetrack ] ) . by the remark above , we want to show that there exists @xmath914 so that @xmath9150,t[$ ] is empty . for each @xmath9160,s[$ ] , we apply a resolvent estimate to ( [ quasimodetrack ] ) and obtain @xmath917 . orthogonality then implies that for each @xmath9160,s[$ ] , there exists @xmath918 such that @xmath919 by applying corollary [ coro : noncon ] and using the fact that @xmath701 is both nonnegative and ` block - diagonal ' with respect to @xmath920 , we find that @xmath921 for all @xmath9220 , t_3[\ , \cap b$ ] . by theorem 4.2 in @xcite , the function @xmath923 is integrable over @xmath1320 , t_3[$ ] . thus , by ( [ adotbound ] ) , the function @xmath46 is integrable over @xmath9150,t_3[$ ] . it follows that there exists a sequence @xmath924 with @xmath900 . let @xmath925 be a real - analytic eigenbranch of @xmath902 with @xmath926 . suppose that there exists a sequence @xmath927 with @xmath928 . we claim that there exists another constant @xmath353 and another sequence @xmath929 with @xmath930 so that @xmath931 . to see this , first note that by taking subsequences if necessary , we may assume that @xmath932 for each @xmath727 . let @xmath933 $ ] be the connected component of @xmath934)$ ] that contains @xmath744 . the graph of @xmath925 must cross the lines @xmath935 , and hence @xmath936 this establishes the existence of the desired sequence @xmath937 . by lemma 4.3 in @xcite , the function @xmath938 is bounded . thus , since @xmath939 , we can apply ( [ adotbound ] ) to find a constant @xmath749 so that @xmath940 for all sufficiently large @xmath727 . hence , @xmath941 , and thus @xmath942 but this contradicts lemma [ lambdadot ] below . [ lambdadot ] let @xmath650 and let @xmath744 be a sequence converging to zero . for each @xmath943 , let @xmath925 be a real - analytic eigenbranch of the family @xmath608 . if @xmath944 , then @xmath945 lemma [ lem : lambda_asymp ] as stated does not directly apply since the eigenvalue branch @xmath925 depends on @xmath727 . by keeping track of the constants in the proof of lemma [ lem : lambda_asymp ] , one can produce a version that directly implies lemma [ lambdadot ] . we prefer to give a direct proof here . let @xmath946 be a unit norm eigenfunction of @xmath902 with eigenvalue @xmath947 . by the standard variational formula @xmath948 since @xmath946 is an eigenfunction of @xmath902 with eigenvalue @xmath947 , we have @xmath949 it suffices to show that the right hand side of ( [ variation ] ) tends to zero as @xmath727 tends to infinity . let @xmath703 . since @xmath947 tends to @xmath42 , there exists @xmath844 so that if @xmath950 , then @xmath951 and thus @xmath952 to estimate the remaining integral over @xmath953 , we will apply a standard convexity estimate from the theory of ordinary differential equations . if @xmath727 is sufficiently large , @xmath954 , and hence there exists @xmath955 so that if @xmath956 , then @xmath957 it follows that there exists @xmath301 so that if @xmath751 , then @xmath958 since @xmath959 for sufficiently large @xmath727 , we may combine ( [ deltaplus ] ) with ( [ delta ] ) to show that ( [ variation ] ) is less than the given @xmath960 for sufficiently large @xmath727 . in this subsection , we show that @xmath961 is smaller than @xmath962 for @xmath140 near a _ crossing time _ , a value of the parameter @xmath12 such that @xmath41 belongs to the spectrum of @xmath621 . then we show that there exists a sequence of crossing times @xmath744 and intervals of width @xmath963 about the crossing times on which @xmath56 is smaller than @xmath53 . the proof of the first result depends on the analysis contained in appendix [ appendix : airyb ] . [ prop : xings - kpart ] given @xmath964 , there exists @xmath955 and @xmath705 such that if @xmath518 and @xmath965 then @xmath966 let @xmath967 be an eigenfunction of @xmath621 with eigenvalue @xmath968 satisfying @xmath969 . we have @xmath970 and hence @xmath971 in appendix [ appendix : airyb ] , we prove that there exists @xmath972 so that @xmath973 hence by applying the cauchy - schwarz inequality to the left hand side of ( [ eq : brhs ] ) , we find that @xmath974 let @xmath975 , and use ( [ dist53 ] ) to find that @xmath976 the claim follows by choosing @xmath301 sufficiently small . [ prop : xings ] for all @xmath977 there exists @xmath978 such that if @xmath979 , @xmath980 , and @xmath981 $ ] , then @xmath982 let @xmath983 be the eigenvalue branch of @xmath621 such that @xmath984 . by lemma [ zerolemma ] , we have @xmath985 for some @xmath61 , and hence @xmath986 . using the fact that @xmath35 and @xmath29 are asymptotic and the fact that @xmath701 is non negative , there exists a constant @xmath353 such that @xmath987 for all sufficiently small @xmath12 . thus , for even smaller @xmath12 we obtain @xmath988 since @xmath989 , integration over @xmath990 $ ] and exponentiation gives @xmath991 if @xmath992 , then @xmath993 where the last inequality holds for @xmath994 . by combining this with ( [ est : ratio_eigen ] ) , one finds that for @xmath995 $ ] , we have @xmath996 using lemma [ edotupperestimate ] , and @xmath997 , we have that @xmath998 an argument similar to the one above gives that @xmath999 for @xmath1000 $ ] . since by assumption @xmath1001 , there exists @xmath1002 so that if @xmath1003 , then @xmath1004 . therefore , by ( [ est : lambda_first ] ) , we have that @xmath1005 is bounded above by @xmath1006 for @xmath1007 . in sum , if @xmath1008 and @xmath995 $ ] , then @xmath1009 the claim follows by chosing @xmath718 carefully . we wish to estimate from below the size of the set of @xmath12 for which ( [ dist53 ] ) holds true . this is accomplished by the following proposition . [ prop : seqxings ] let @xmath844 . there is a sequence @xmath744 of crossing times such that @xmath1010 and if @xmath1011 are large enough , then the intervals @xmath1012 $ ] and @xmath1013 $ ] are disjoint . by lemma [ lem : lambda_asymp ] , there exists , @xmath1014 so that @xmath1015 . it also follows from proposition [ tracking ] that there exists @xmath1016 so that @xmath1017 for sufficiently small @xmath12 . by lemma [ zerolemma ] , the eigenvalues of @xmath621 have the form @xmath1018 where @xmath1019 and @xmath1020 is the increasing sequence of positive solutions to the equation @xmath1021 . standard asymptotic analysis shows that @xmath1022 fix @xmath1023 for each @xmath1024 $ ] and each @xmath1025 , there exists a unique @xmath1026 so that @xmath1027 we drop the dependence in @xmath1028 from the notation for a moment . if we set @xmath1029 and @xmath1030 then ( [ eqn : imp_+ ] ) becomes @xmath1031 by the analytic implicit function theorem , for @xmath327 near @xmath571 , there exist a unique analytic function @xmath1032 so that @xmath1033 by inspecting of the first few coefficients in the taylor expansion of @xmath1034 , we find that @xmath1035 thus , since @xmath1036 , and @xmath1037 we find that @xmath1038 where @xmath1039 . choose @xmath703 so that if @xmath1040 , then @xmath1041 . define @xmath1042 . by applying the intermediate value theorem to @xmath1043 , there exists @xmath1044t_n^- , t_n^+[$ ] so that @xmath1045 . see figure [ intermediate ] . since @xmath1046 is increasing to infinity , the sequence @xmath744 is decreasing to zero . . [ intermediate ] ] moreover , since @xmath1047 @xmath1048 where @xmath1049 . from ( [ eqn : r_n_asymp ] ) we have @xmath1050 where @xmath1051 . by substituting this into ( [ eq : tn+_asymp ] ) and ( [ eq : tn_asymp ] ) we find that @xmath1052 and @xmath1053 the first claim follows . moreover , since @xmath1054 , we have @xmath1055 it follows that , for all sufficiently large @xmath727 , we have @xmath1056\subset [ t_n^-,t_n^+]$ ] . since the intervals @xmath1057\}$ ] are disjoint , the claim is proven . in this section we derive the desired contradiction . in particular , we prove the following . [ prop : dt ] if @xmath41 is a cusp form eigenvalue branch with a positive limit , then there exists @xmath705 and @xmath1058 so that if @xmath845 , then @xmath1059 the proof will consist of two types of lower estimates . the first depends on the fact that near each crossing the ` relative variation ' @xmath1060 is at least of order @xmath841 . the second shows that away from the crossings the relative variation is not too negative . define @xmath10610,t]~ \left|~ \left\| w_s^k \right\|~ \leq~ \rho \cdot \|u_s\| \right . \right\}.\ ] ] if @xmath964 , then it follows from proposition [ prop : adotbound ] that there exists @xmath749 so that for @xmath1062 we have @xmath1063 hence , since @xmath1064 , there exists @xmath1065 so that if @xmath1066 and @xmath964 , then @xmath1067 for each @xmath1062 . we will integrate this estimate near the crossings to obtain the following . [ lem : onk ] for each @xmath964 , there exists @xmath705 and @xmath1068 so that for each @xmath518 , we have @xmath1069 by ( [ eq : edotkappa ] ) the integrand is positive on @xmath1070 , it suffices to show the same estimate holds for a subset @xmath1071 of @xmath1070 . to define this subset , we first combine proposition [ prop : xings - kpart ] , proposition [ prop : xings ] , and proposition [ prop : seqxings ] to find @xmath844 , @xmath1072 , and a monotone sequence @xmath1073 so that for each @xmath727 @xmath1074 \ ] ] belongs to @xmath1075 , the intervals @xmath1076 and @xmath1077 are disjoint , and for each @xmath1078 @xmath1079 where @xmath1080 . the subset @xmath1071 will be defined as a union of @xmath1081 over sufficiently large @xmath727 . we have @xmath1082 and hence there exists @xmath1083 so that if @xmath1084 we have @xmath1085 thus , from ( [ t_n_bound ] ) we find that if @xmath1086 , then @xmath1087 since the intervals @xmath1088 are disjoint , by combining ( [ eq : edotkappa ] ) , ( [ eq : intoveri ] ) , and ( [ eq : sumn53 ] ) , we find that @xmath1089 where @xmath1090 and @xmath1091 let @xmath1092 . if @xmath845 , then @xmath1093 $ ] for some @xmath1086 . we have @xmath1094 and hence @xmath1095 and @xmath1096 . therefore , ( [ est : int_g ] ) implies the claim . to bound the relative variation on the complement of @xmath1070 , we will use the following . [ prop : fdotbetweenxings ] there exists @xmath353 and @xmath705 such that , if @xmath518 , then @xmath1097 by arguing as in ( [ est : e_dot_lower ] ) we have @xmath1098 let @xmath1099 denote the orthogonal projection of @xmath52 onto the eigenfunction branch of @xmath35 that corresponds to @xmath1100 from theorem [ tracking ] . let @xmath1101 . we have @xmath1102 , @xmath1103 and @xmath1104 where @xmath1105 is an eigenfunction of @xmath608 with eigenvalue @xmath1100 . using the cauchy - schwarz inequality and the nonnegativity of @xmath672 we have @xmath1106 since @xmath1107 is an eigenfunction , we have @xmath1108 . using ( [ defn : ak ] ) and the fact that @xmath1107 is an eigenfunction that is orthogonal to @xmath1109 , we find that @xmath1110 since @xmath1111 , we have @xmath1112 the large @xmath326 asymptotics of @xmath1113 and @xmath1114 can be analysed using the same methods used in appendix [ appendix : airyb ] for @xmath1115 . in particular , the estimate proven in lemma [ prop : expdecay ] holds with @xmath1116 estimated in lemma [ lem : restimate ] . in particular , for each @xmath1117 , there exists a constant @xmath353 so that for sufficiently small @xmath12 @xmath1118 and @xmath1119 if @xmath1120 , then @xmath1121 for sufficiently small @xmath12 . therefore , by splitting the domain of integration into @xmath1122 $ ] and @xmath1123 and using the cauchy - schwarz inequality , we find that @xmath1124 for sufficiently small @xmath12 . we claim that @xmath1125 . indeed , by applying lemma [ lem : quasimode ] with @xmath1126 , we have @xmath1127 for some constant @xmath353 . thus , since the eigenvalue @xmath1128 satisfies @xmath1129 we find that @xmath1130 by definition , @xmath1109 is a projection onto eigenspaces of @xmath608 whose associated eigenvalues are distinct from @xmath1128 . by lemma [ lem : lambda_asymp ] , there exists @xmath844 so that such eigenvalues differ from @xmath1128 by at least @xmath1131 . because of ( [ est : preresolv ] ) , we can thus apply a resolvent estimate ( e.g. lemma 2.1 @xcite ) to find that @xmath1132 by substituting ( [ est : resolvent_est ] ) into ( [ est : pre_resolvent ] ) and setting @xmath1133 , we find a constant @xmath832 so that @xmath1134 by combining this estimate with ( [ eq : at_ak_inner ] ) , ( [ est : adot ] ) , and ( [ eq : edot ] ) , we obtain a constant @xmath1135 so that @xmath1136 by orthogonality @xmath1137 , and hence by ( [ est : resolvent_est ] ) and lemma [ lem : lambda_asymp ] , we have a constant @xmath1138 so that @xmath1139 the desired result follows . [ coro : nottoonegative ] there exists @xmath832 such that for each @xmath11400,1[$ ] , there exists @xmath705 such that if @xmath1141 , then @xmath1142 \setminus k(t,\rho ) } \left(\dot{e}_s~ -~ \dot{\lambda}^*_s\right)~ ds~ \geq~ c ' \cdot \left(\rho^2- 1 \right ) \cdot t^{{\frac{2}{3}}}. \ ] ] by definition , if @xmath1143\setminus k(t,\rho)$ ] , then @xmath1144 and hence from lemma [ prop : fdotbetweenxings ] , we find that @xmath1145 by using lemma [ lem : lambda_asymp ] , we find @xmath832 and @xmath301 so that for @xmath518 @xmath1146 the claim follows from integration . finally , we use lemma [ lem : onk ] and corollary [ coro : nottoonegative ] to prove theorem [ prop : dt ] . this will complete the proof of the main theorem . apply lemma [ lem : onk ] with @xmath1147 . then apply corollary [ coro : nottoonegative ] with @xmath1148 such that @xmath1149 since @xmath1150 is positive on @xmath1151 , we find that @xmath1152 since @xmath1153 , we have the desired conclusion . in this appendix , we compute the asymptotics of each real - analytic eigenvalue branch of @xmath890 for each @xmath1154 . [ lem : lambda_asymp ] let @xmath1155 and let @xmath907 be a real - analytic eigenvalue branch of @xmath666 for @xmath601 . then @xmath1156 where @xmath1157 and @xmath1158 is a zero of the derivative of the airy function @xmath1159 defined in ( [ eq : airy_asymp ] ) . moreover , @xmath1160 to prove proposition ( [ lem : lambda_asymp ] ) , we will first transform the eigenvalue problem into an eigenvalue problem that is easier to analyse . if @xmath77 is an eigenfunction of @xmath666 with respect to @xmath1161 with eigenvalue @xmath260 , then for each @xmath1162 and @xmath601 , we have @xmath1163 where we have set @xmath1164 hence @xmath1165 by making the change of variable @xmath1166 , letting @xmath1167 and @xmath1168 , and dividing by @xmath1169 , we find that @xmath1170 where @xmath1171 and @xmath1172 this leads us to set @xmath1173 and define for each @xmath1174 the quadratic forms @xmath1175 and @xmath1176 define @xmath1177 and @xmath1178)$ ] ( i.e. the set of functions @xmath1179 such that the distributional derivative also is in @xmath76 ) . then for each @xmath1180 , the form @xmath1181 is a bounded quadratic form on @xmath1182 and @xmath1183 is a closed quadratic form on @xmath1182 with domain @xmath1184 . since @xmath1185 maps bijectively @xmath1186)$ ] onto @xmath1187)$ ] , the function @xmath1188 is an eigenfunction of @xmath1183 with respect to @xmath1181 with eigenvalue @xmath1189 . it follows from the perturbation theory of generalized eigenvalue problems ( see vii.6 in @xcite ) that the eigenvalues of @xmath1183 with respect to @xmath1181 can be organized into real - analytic eigenvalue branches for @xmath1180 . the domains of @xmath1183 and @xmath1181 change , and hence analytic perturbation theory can not be applied . ] since the generalized eigenvalue problem @xmath1190 corresponds to a sturm - liouville problem with neumann condition at @xmath1191 , the eigenspaces are 1-dimensional . hence , we may enumerate the real - analytic eigenvalue branches @xmath1192 so that for each @xmath1193 and @xmath1180 we have @xmath1194 [ lem : bounded_branches ] for each @xmath1195 , there exists @xmath1196 and @xmath353 so that if @xmath979 then @xmath1197 in particular , there exists @xmath811 so that for small @xmath1180 @xmath1198 moreover , @xmath1199 is a zero of the derivative of the airy function @xmath1159 . first , we show that each @xmath1200 is bounded . to this end , define @xmath1201 since @xmath1202 is bounded above by 2 , we have @xmath1203 for each @xmath1180 and @xmath1174 . note that for each @xmath1204 , we have @xmath1205 and hence @xmath1206 integration by parts shows that the eigenfunctions of @xmath1207 with respect to @xmath1208 are solutions to the sturm - liouville problem @xmath1209 standard convexity estimates on solutions to ordinary differential equations imply that each eigenfunction belongs to the domain @xmath1184 of @xmath1183 for each @xmath1180 . in particular , the sum of the first @xmath1195 eigenspaces of @xmath1207 with respect to @xmath1208 belongs to @xmath1184 . therefore , using ( [ est : btoa ] ) , the minimax principle , and ( [ order ] ) we find that @xmath1192 is bounded by the @xmath1210 eigenvalue of @xmath1207 with respect to @xmath1208 . in the remainder of the argument we drop the superscript @xmath1195 and focus on an individual real - analytic eigenfunction branch @xmath1211 with eigenvalue @xmath1212 . for @xmath1180 , we have @xmath1213 where @xmath1214 indicates differentiation with respect to @xmath140 . a computation gives that for each @xmath69 @xmath1215 and @xmath1216 let @xmath1211 be a real - analytic eigenfunction branch of @xmath1183 with respect to @xmath1181 associated to the real - analytic eigenvalue branch @xmath1212 . integration by parts gives @xmath1217 let @xmath1016 be the upper bound on @xmath1212 proven above . if @xmath1204 and @xmath1218 , then @xmath1219 and hence @xmath1220 for @xmath1204 . it follows that @xmath1221 for @xmath1222 . thus , since @xmath1223 is finite , we find that for @xmath1224 @xmath1225 integrating from @xmath1226 to @xmath1227 , we find a constant @xmath353 ( that depends on @xmath1226 ) such that , for @xmath1228 we have @xmath1229 from this we find constants @xmath353 such that that @xmath1230 a similar argument shows that @xmath1231 therefore , ( [ est : nu_dot ] ) holds , and via integration we find @xmath811 so that ( [ eq : nu_asymp ] ) holds true . continuity of solutions to ordinary differential equations with respect to coefficients applies to ( [ eq : a - n - ode ] ) with fixed initial conditions @xmath1232 and @xmath1233 . in particular , we have a solution @xmath1234 to @xmath1235 it follows that @xmath1236 is a solution to @xmath1237 . estimate ( [ a - n - convexity ] ) applies to @xmath1234 , and hence it follows from ( [ eq : airy_asymp ] ) that @xmath77 is a multiple of @xmath1159 . the function @xmath25 satisfies the neumann condition @xmath1238 and hence @xmath1239 as desired . if @xmath1240 is a real - analytic eigenfunction branch of of @xmath666 associated to the eigenvalue branch @xmath894 , then @xmath1241 is a real - analytic eigenfunction branch of @xmath1183 with eigenvalue branch @xmath1242 . lemma [ lem : bounded_branches ] implies that @xmath1243 by differentiating @xmath1244 , we find that @xmath1245 by lemma [ lem : bounded_branches ] , both @xmath1246 and @xmath1212 are bounded . therefore , @xmath1247 and hence @xmath1248 . let @xmath39 be a real - analytic eigenbranch of @xmath29 such that @xmath1249 for some positive integer @xmath40 . for a fixed constant @xmath686 , let @xmath1250.\ ] ] as in [ sec : nota ] , let @xmath202 denote the orthogonal projection of @xmath30 onto the sum of the eigenspaces of @xmath35 whose eigenvalues lie in @xmath282 . [ prop : blower ] let @xmath955 . there exists @xmath749 , @xmath1251 and @xmath705 such that , if @xmath518 and if @xmath1252 is an eigenfunction of @xmath621 with eigenvalue @xmath1253 satisfying @xmath1254 then @xmath1255 the quadratic form @xmath1257 is controlled by @xmath35 meaning that there exists a constant @xmath353 such that for @xmath1258 @xmath1259 thus , proposition [ lem : u_w_close ] and the fact that @xmath1260 imply that observe also that proposition [ lem : u_w_close ] also implies that @xmath1263 in the limit @xmath1264 so that we can freely replace @xmath53 by @xmath1265 and vice - versa in each ( multiplicative ) estimate . by the discussion [ subsec : separation ] and [ sec : nota ] , for each @xmath12 we can uniquely write @xmath1266 where each @xmath1267 is an eigenfunction of @xmath666 with eigenvalue @xmath1268 . set @xmath1269 and @xmath1270 by linearity @xmath1271 from ( [ eq : defb_t ] ) we have @xmath1272 where @xmath1273 $ ] and @xmath1274 is defined in ( [ defn : p ] ) . since @xmath1275 , and @xmath1276 , we find that @xmath1277 if @xmath1278 , then @xmath1279 , and so @xmath1280 . for @xmath1281 , apply lemma [ lem : bwellpsi ] below to find that @xmath1282 since @xmath1283 and @xmath1284 are orthogonal if @xmath1285 , we have @xmath1286 thus , by summing ( [ eq : vlower ] ) over @xmath1287 , we obtain @xmath1288 for @xmath1289 , we have @xmath1290 thus , from lemma [ lem : bwkpsi ] and lemma [ lem : p_positive ] , there exists @xmath704 so that @xmath1291 for some @xmath704 . the latter estimate , combined with ( [ eq : linearity ] ) , ( [ eq : sumvl ] ) , and the triangle inquality , yield the claim . multiply both sides of ( [ eq : rdefn ] ) by a smooth function with compact support @xmath1302 and integrate over @xmath1303 , then integrate by parts to obtain @xmath1304 observe that @xmath1305 so that by applying lemma [ coro : wisqm ] to the test function @xmath1306 , there exists @xmath705 and @xmath832 such that for @xmath518 , we have @xmath1307 recalling that the @xmath76-norm on the right hand side has the weight @xmath1308 this implies that @xmath1309 the claim follows since @xmath1310 on the interval over which we integrate . the strategy of the proof of lemma [ lem : bwellpsi ] is as follows . by ( [ eq : rdefn ] ) , the function @xmath1311 is a solution to the inhomogeneous equation @xmath1312 where @xmath1313 and @xmath1314 the function @xmath1315 is a solution to the homogeneous equation @xmath1316 where @xmath1317 . our choice of @xmath249 in ( [ eq : beta ] ) implies that @xmath1318 is bounded below by a constant @xmath1319 for all small @xmath12 , @xmath889 , and @xmath1320 . hence we can use wkb type estimates to find a basis @xmath1321 of solutions to the homogeneous equation ( [ eq : homoode ] ) . we will then use ` variation of parameters ' to express each solution to to ( [ eq : inhomoode ] ) in terms of this basis , and we use lemma [ lem : restimate ] to provide control of the inhomogeneous term @xmath1116 . finally , we will estimate the integral in ( [ eq : ellsmall ] ) using a riemann - lebesgue type estimate . for @xmath889 and @xmath1320 , we have @xmath1322 , and hence we can apply theorem 6.2.1 in @xcite , to obtain a basis @xmath1323 of solutions to the homogeneous equation ( [ eq : homoode ] ) that satisfy @xmath1324 and @xmath1325 where , for @xmath1320 and @xmath889 , the smooth functions @xmath1326 and @xmath1327 have @xmath1328 norm that is uniformly @xmath1329 . observe that @xmath1330 have @xmath1331)$ ] norms that are uniformly bounded above and away from @xmath1332 moreover , since @xmath1333 and @xmath1334 are highly oscillatory for small @xmath12 , an integration by parts argument shows that the @xmath1331)$ ] inner product @xmath1335 is @xmath1329 . it follows that there exists @xmath1336 such that if @xmath1337 , then @xmath1338 for all sufficiently small @xmath12 . here @xmath1339 denotes the @xmath1331)$ ] norm . in particular , for each @xmath765 and each @xmath12 , there exists @xmath1345 so that the function @xmath38 of ( [ eq : vt ] ) satisfies @xmath1346 the eigenfunction @xmath1252 of @xmath1347 satisfies ( [ eq : generic_inhomo ] ) with @xmath1348 , and hence there exists @xmath1349 so that @xmath1350 the integral in ( [ eq : ellsmall ] ) is equal to @xmath1351 by expanding the product of sums , one obtains a sum of @xmath1352 integrals . by substituting the expressions ( [ eq : wkbderivate ] ) and ( [ eq : wkb ] ) , integration by parts , and applying standard estimates , we find that each integral is @xmath1329 . for example , consider the terms of the form @xmath1353 since @xmath694 , an elementary computation shows that there exists @xmath844 so that if @xmath1354 $ ] and @xmath12 is sufficiently small , then @xmath1355 thus , we may integrate by parts to find a constant @xmath353 so that the integral in ( [ eq : aa ] ) is at most @xmath1356 . it follows that all the terms of this form are bounded above by from ( [ eq : wkb ] ) and ( [ eq : wkbderivate ] ) we find that there exists @xmath1360 so that if @xmath1361 , then @xmath1362 . from ( [ eq : wkb ] ) and ( [ eq : ftl - defn ] ) we find that for each @xmath765 @xmath1363}~ \left|v_{e_t,\mp}^{\ell}(y ) \right|~ \leq~ \frac{2}{\sqrt{\delta_1 } } \ ] ] for all sufficiently small @xmath12 . hence , using the cauchy - schwarz inequality and lemma [ lem : restimate ] we have , for all @xmath1364 $ ] , @xmath1365 for all sufficiently small @xmath12 . for the derivative of @xmath1366 , we have @xmath1367 applying cauchy - schwarz and lemma [ lem : restimate ] gives @xmath1368 finally , we apply integration by parts to ( [ eq : aa ] ) . the resulting terms that do not contain @xmath1369 have uniformly bounded @xmath1328 norm . the term that contains @xmath1369 can be bounded using ( [ eq : boundadotpm ] ) . it follows that all the term of this form are bounded by @xmath1370 the final step consists in bounding @xmath1371 by @xmath1372 to control the terms of eq . ( [ eq : type1 ] ) . using ( [ eq : basis_equivalence ] ) and ( [ eq : defv_var ] ) we have @xmath1373}\{|h_+(y)|+|h_-(y)|\}\ ] ] by orthogonality we have @xmath1374 and , using the bound on @xmath1375 we finally obtain @xmath1376 this finishes the proof . as in the previous subsection , the function @xmath1377 is a solution to the inhomogeneous equation ( [ eq : inhomoode ] ) with @xmath1313 and @xmath1116 defined by ( [ eq : rdefn ] ) . however , for @xmath1289 , the function @xmath1378 is negative for large @xmath326 . in fact , since @xmath41 decreases to @xmath42 , the function @xmath1379 changes sign nearer and nearer to @xmath627 . since the solution @xmath1377 belongs to @xmath1380 , we expect it to decay exponentially as soon as @xmath326 moves away from @xmath16 . for @xmath326 near 1 , we will approximate @xmath38 using airy functions . in this subsection we will make these approximations precise and use them to give a proof of lemma [ lem : bwkpsi ] by lemma [ zerolemma ] , @xmath1252 is a constant multiple of @xmath1381 defined in ( [ zeroeigenvector ] ) . because both sides of the estimate in lemma [ lem : bwkpsi ] are homogeneous functions of degree 1 in @xmath1252 , it suffices to assume that @xmath1382 . we have @xmath1388 where @xmath1389 @xmath1390 and @xmath1391 . in particular , for sufficiently small @xmath12 @xmath1392 thus , since @xmath1393 and @xmath1394 the triangle inequality applied to ( [ eq : psi_0_decomp ] ) implies that ( [ est : psi_0_sup ] ) holds for sufficiently small @xmath12 . by proposition [ tracking ] and lemma [ lem : lambda_asymp ] , there exists @xmath353 so that if @xmath12 is sufficiently small , then @xmath1402 hence , for @xmath1403 , one finds that @xmath1404 thus , since @xmath1405 , there exists @xmath905 such that if @xmath1406 then for all @xmath1407 we have @xmath1408 for each smooth function @xmath1409 with support in @xmath1321+t^{\alpha } , \infty[$ ] , define @xmath1410 extend @xmath1411 to a self - adjoint operator on @xmath1412 . it follows from ( [ eq : lowerbound ] ) that the spectrum of @xmath1411 lies in @xmath1413 . hence @xmath1411 is invertible and the operator norm of @xmath1414 is bounded above by @xmath1415 therefore @xmath1416 the function @xmath1417 is a solution to ( [ eq : inhomoode ] ) on @xmath1418 , and hence @xmath1419 is a solution to the homogeneous equation ( [ eq : homoode ] ) . it follows from ( [ eq : lowerbound ] ) that @xmath1420 if @xmath1421 . in particular , @xmath1422 is convex , moreover @xmath1422 is non - negative and in @xmath1423 since @xmath1424 and @xmath1425 this implies that @xmath1426 . indeed , since @xmath1422 is convex , @xmath1427 has a limit @xmath1428 in @xmath1429 if this limit is positive then it implies that @xmath1430 for large @xmath326 and this contradicts the fact that @xmath1431 is integrable . in particular , @xmath1427 is bounded so that by integrating ( [ est : w2convex ] ) we find that @xmath1432 the argument also shows that @xmath1427 is non positive for large @xmath326 so that @xmath1422 has a limit when @xmath1433 since @xmath1422 is integrable , this limit is @xmath1332 for each @xmath1434 the function @xmath1435 that is defined by @xmath1436 satisfies @xmath1437 with @xmath1438 and @xmath1439 . therefore , by comparison with ( [ est : w2convex ] ) , and using the maximum principle , we find that if @xmath1440 , then @xmath1441 . applying this to @xmath1442 we find that for each @xmath1443 @xmath1444 for small @xmath12 , the function @xmath1455 has a simple zero near @xmath627 . thus , to approximate solutions of @xmath1456 near @xmath627 we will use solutions to airy s differential equation @xmath1457 where @xmath1458 . we first describe the link to airy s equation . if we define @xmath1459 , then we have @xmath1460 where @xmath1461 . let @xmath532 be the smooth function that satisfies @xmath1462 by substituting the latter expression into ( [ eq : ode2 ] ) and by dividing by @xmath1463 , we find that @xmath1464 setting @xmath1465 we have @xmath1466 where @xmath1467 we have @xmath1472 , and hence by lemma [ lem : restimate ] , @xmath1473 let @xmath1474 be the tracking eigenvalue branch associated to @xmath1475 . the eigenvalue @xmath49 corresponding to @xmath1476 is in a @xmath1329 neighbourhood of @xmath1477 moreover , when @xmath12 tends to @xmath571 , proposition ( [ lem : lambda_asymp ] ) implies that @xmath894 is at a distance of order @xmath896 of the rest of the spectrum . using ( [ eq : lambda_asymp ] ) , ( [ eq : rdefn ] ) , lemma [ lem : restimate ] and a resolvent estimate , we have @xmath1478 , and hence @xmath1479 the tracking eigenfunction @xmath1476 satisfies ( [ eq : inhomoode ] ) with @xmath1348 . hence by proposition [ prop : expdecay ] , if @xmath1480 , then @xmath1481 observe that , by orthogonality , @xmath1482 , hence @xmath1483 since @xmath1484 ( because @xmath1470 ) , we may thus take @xmath1485 and the biggest term is then of order @xmath1486 . the claim follows . we first construct solutions to the associated homogeneous equation @xmath1489 using airy functions . in particular , it is well - known that there exists a basis @xmath1490 of solutions to @xmath1491 such that @xmath1492 and @xmath1493 one checks that @xmath1494 defines a basis of solutions to ( [ eq : homoairy ] ) . it follows from well - known identities that the wronskian of @xmath1495 is @xmath65 . hence the wronskian of @xmath1496 is @xmath1497 . therefore , by the method of variation of constants , for each @xmath1498 , the function @xmath1499 is a solution to ( [ eq : inhomoairy ] ) . [ lem : estwp ] for each compact set @xmath1500 , there exists @xmath353 such that , for each @xmath1498 , if @xmath1501 and @xmath1502 $ ] , then @xmath1503 and @xmath1504 where @xmath1505 denotes the @xmath76 norm of @xmath1506 over @xmath1507 $ ] . moreover , there exists @xmath1016 so that if @xmath1508$],then @xmath1509 and @xmath1510 using the cauchy - schwarz inequality , we have for @xmath1511 $ ] @xmath1512 and @xmath1513 thus , from ( [ eq : wparticular ] ) and the triangle inequality , we have @xmath1514 estimate ( [ est : wparticular ] ) then follows from lemma [ lem : wdt ] below . the proof is a straightforward consequence of the continuity and known asymptotics of @xmath1525 and @xmath1526 . from ( [ eq : airy_asymp ] ) and integration by parts we find that @xmath1527 as @xmath25 tends to @xmath1528 . thus there exists @xmath1529 so that if @xmath1530 , then @xmath1531 therefore , for @xmath1530 , @xmath1532 and , using ( [ eq : airy_deriv_asymp ] ) , @xmath1533 the expressions on the left hand sides of ( [ est : aplusaminus ] ) and ( [ est : aplusaminusderiv ] ) are continuous in @xmath25 , and hence are bounded by a constant @xmath353 for @xmath1534 where @xmath1535 . moreover , from ( [ est : aplusaminus ] ) and ( [ est : aplusaminusderiv ] ) we have @xmath1541 and @xmath1542 provided @xmath1543 . let @xmath1544 . if @xmath1545 , then @xmath1546 and @xmath1547 . the desired estimates in the @xmath1548/@xmath1549 case follow . the estimates in the @xmath1549/@xmath1548 case are proved in a similar fashion . the function @xmath1555 satisfies the inhomogeneous equation ( [ eq : inhomoairy ] ) with @xmath1556 , and the inhomogeneity @xmath1469 satisfies ( [ eq : boundaw ] ) . to estimate @xmath1555 and hence ( [ eq : small_x_integral ] ) we write @xmath1557 where @xmath1558 is the particular solution to ( [ eq : inhomoairy ] ) defined by ( [ eq : wparticular ] ) for @xmath1559 . the function @xmath1560 is a solution to the associated homogeneous equation . integration by parts gives @xmath1562 using lemmas [ lem : airyapprox ] and [ lem : estwp ] , one finds that there exists @xmath353 such that , for @xmath1563 $ ] @xmath1564 thus , using ( [ lem_psi_0_bound ] ) , we conclude that the first term on the right side of ( [ eq : wpt_integral ] ) is @xmath1565 . to bound the second term on the right side of ( [ eq : wpt_integral ] ) , we take @xmath1566 and estimate the integrals over @xmath1567 $ ] and @xmath1568 $ ] separately . observe that since @xmath1569 then @xmath1570 using lemmas [ lem : airyapprox ] and [ lem : estwp ] , we find @xmath353 so that for all sufficiently small @xmath12 @xmath1571 and @xmath1572 by combining these estimates and using ( [ eq : psi_zero_deriv ] ) , we find that @xmath1573 the claim will follow provided we can choose @xmath1574 and @xmath1575 so that each power of @xmath12 appearing on the righthand side is greater than @xmath1576 . this gives us a solution set which is the open triangle in @xmath1577 bounded by the lines @xmath1578 , @xmath1579 , and @xmath1580 . the two latter lines intersect for @xmath1581 the claim follows . as above we consider @xmath1586 and take some @xmath1587 using lemmas [ lem : airyapprox ] and [ lem : estwp ] , one finds that @xmath1588 } & \leq c \cdot t^{\frac{5{\alpha}}{2}-\frac{3{\widetilde{\alpha}}}{4}-\frac{1}{6 } } \|u_t\| \\ \| w_{p , t}\|_{[t^{{\widetilde{\alpha}}},3t^{\alpha } ] } & \leq c \cdot t^{2{\alpha}-\frac{{\widetilde{\alpha}}}{2}-\frac{2}{3 } } \|u_t\| . \end{split}\ ] ] the claim will follow provided we can find @xmath1589 and @xmath1590 such that @xmath1591 and @xmath1592 the solution set is here a quadrilateral whose projection on the @xmath355-axis is the interval @xmath132\frac{7}{33},{\frac{1}{3}}[.$ ] finally , we consider the integral corresponding to the homogeneous part @xmath1560 of @xmath1555 : @xmath1593 there exist constants @xmath1594 , @xmath1595 , depending on @xmath12 , such that @xmath1596 where @xmath1597 and @xmath1598 are as defined in ( [ eq : defn_w ] ) with the parameter @xmath140 and @xmath1599 defined in ( [ eq : s_z_s ] ) . [ lem : norms ] for all @xmath1400\frac{7}{33 } , \frac{1}{3}[,$ ] there exists @xmath1584 such that @xmath1602}~ = ~ o(t^\infty ) \cdot \| u_t\| \ ] ] where @xmath1603 is a function that is of order @xmath1604 for each @xmath727 , and @xmath1605}~ \geq~ { \frac{1}{2}}\cdot \left ( \|w_t^k\| -c\cdot t^\delta \|u_t\|\right ) . \ ] ] using the behavior of the norm of @xmath1606 we find that @xmath1607}~ = ~ o(t^\infty ) \cdot \|a_+w_+\|_{[2t^{\alpha},3t^{\alpha } ] } \ ] ] and @xmath1608}~ \leq~ c \cdot \|a_- w_-\|_{[0,2t^{\alpha}]}. \ ] ] we thus have @xmath1609 } & = & o(t^\infty ) \cdot \|a_+w_+\|_{[2t^{\alpha},3t^{\alpha}]}\\ & \leq & o(t^\infty ) \cdot \left ( \|w\|_{[2t^{\alpha},3t^{\alpha}]}\,+\ , \|a_- w_-\|_{[2t^{\alpha},3t^{\alpha}]}\,+\,\|w_{p , t}\|_{[2t^{\alpha},3t^{\alpha}]}\right ) \\ & \leq & o(t^\infty ) \cdot \left ( \|u_t\| \,+\,\|w_-\|_{[0,2t^{\alpha } ] } \,+\ , t^\delta \|u_t\|\right ) \\ & \leq & o(t^\infty ) \cdot \left ( \|u_t\| \,+\ , \|w\|_{[0,2t^{\alpha}]}\,+\ , \|a_+w_+\|_{[0,2t^{\alpha}]}\,+\,\|w_{p , t}\|_{[0,2t^{\alpha}]}\right ) \\ & \leq & o(t^\infty)\left ( \|a_+w_+\|_{[0,2t^{\alpha}]}\,+\,\|u_t\|\right ) . \ ] ] estimate ( [ est : a+w+ ] ) then follows by absorbing the norm of @xmath1610 into the left hand side . to prove estimate ( [ est : a - w- ] ) , we first observe that by using the triangle inequality we find that @xmath1611 } \leq \|w_{p , t}\|_{[0,2t^{\alpha } ] } \,+\,\|a_- w_-\|_{[0,2t^{\alpha}]}\,+\,\|a_+w_+\|_{[0,2t^{\alpha}]}.\ ] ] the first term on the righta hand side is @xmath1612 and the last one is @xmath1613 so that we obtain @xmath1614 } \geq \left ( \|w\|_{[0,2t^{\alpha } ] } - o(t^\delta ) \|u_t\|\right ) .\ ] ] the claim then follows by observing that corollary [ coro : normlocal ] implies that @xmath1615 } \geq { \frac{1}{2}}\left ( \|w_t^k\| - o(t^{1-{\alpha } } ) \|u_t\|\right).\ ] ] from ( [ eq : psi_prime ] ) , we have @xmath1617 thus , the integral we want to estimate can be written as @xmath1618 where we have set @xmath1619 denote by @xmath1620 the integral @xmath1621 integration by parts shows that @xmath1622 where we have set @xmath1623 since @xmath1624 and @xmath140 is of order @xmath12 , and since @xmath1159 is rapidly decreasing , the boundary term at @xmath1625 is @xmath1626 observe that we have a global @xmath1627 prefactor in front of the integral term . thus , when the @xmath1628 is applied to @xmath1629 , we gain @xmath1630 , that is , something of order @xmath1631 when @xmath1628 hits the airy function , we lose a @xmath1632 so that the global prefactor is of order @xmath1633 which is @xmath1634 . summarizing , integrating by parts gains at least a prefactor @xmath1635 by repeated integration by parts we thus observe that we can write , for each @xmath500 @xmath1636 where the @xmath1637 are some constants and the remainder term @xmath1638 can be written @xmath1639 for some smooth functions @xmath1640 if we fix some order @xmath1641 then , using that @xmath1159 and all its derivatives are rapidly decreasing we can find @xmath500 such that the remainder @xmath1638 is @xmath1642 this tells us that @xmath1620 admits a complete asymptotic expansion of the form @xmath1643 from the first integration by parts we see that @xmath1644 and the second term is then of order @xmath1635 the claim follows by taking the imaginary part . let @xmath1158 be the zero of @xmath1159 that comes from lemma [ lem : nonzero ] . since @xmath1159 is a nontrivial solution to a second order differential equation , @xmath1159 can not vanish at a zero of the derivative @xmath1649 . hence , for sufficiently for small @xmath12 , we have @xmath1652 . by arguing as in the proof of lemmas [ lem : nonzero ] and [ lem : wdt ] and using @xmath1653 , we find @xmath1654 so that @xmath1655 where @xmath1656 and @xmath12 is sufficiently small . in particular , @xmath1657 hence the claim follows from lemma [ w- ] . c. m. judge . _ small eigenvalues and maximal laminations on complete surfaces of negative curvature . _ in the tradition of ahlfors - bers . iv , 93 - 105 , contemp . , 432 , amer . , providence , ri , 2007 . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: photoionization models for chemically homogeneous gaseous nebulae of constant density predict an almost constant temperature ; consequently , observers often assume a constant temperature to determine chemical abundances . the spatial temperature inhomogeneities in regions are usually characterized by @xmath1 the mean square temperature variation in three dimensions . recent reviews on the temperature structure of gaseous nebulae are those by peimbert ( 1995 , 2002 ) , esteban ( 2002 ) , stasiska ( 2002 ) , liu ( 2002 ) , torres - peimbert & peimbert ( 2003 ) . from uniform - density chemically - homogeneous photoionization models it has been found that @xmath1 is in the 0.002 to 0.025 range , for objects with solar or subsolar chemical abundances , with typical values around 0.005 ( e.g. , gruenwald & viegas 1992 , garnett 1992 ) . baldwin et al . ( 1991 , henceforth b91 ) produced a photoionization model for the orion nebula , in which grains and gas are well mixed , that included photoelectric heating and cooling of the gas by grain ionization and grain collisions respectively ; for this model peimbert , storey , & torres - peimbert ( 1993 ) found that @xmath1 = 0.004 . alternatively from observations of ngc 346 in the smc ( peimbert , peimbert , & ruiz 2000 ) , 30 doradus ( peimbert 2003 ) , and the orion nebula ( peimbert et al . 1993 , esteban et al . 1998 , henceforth epte98 ) it has been found that @xmath1 is in the 0.02 to 0.04 range . the @xmath1values were derived by combining the temperature derived from the ratio of the @xmath3 4363 , 5007 lines [ @xmath4 with the temperature derived from the ratio of the balmer continuum to @xmath5 [ @xmath6 and by combining @xmath7 with the temperatures derived from the ratio of @xmath8 5007 to the recombination lines of multiplet 1 of o ii and the ratio of the of @xmath3 1906 , 1909 [ c iii ] lines to the @xmath8 4267 c ii recombination line . the differences between the observed @xmath1 values and those predicted by models have two profound implications for the study of ionized gaseous nebulae : they significantly affect the derived abundances and they imply the presence of one or several physical mechanisms not considered by the models . to study this problem further we decided to obtain an independent measurement of @xmath1 based on over a million temperatures determined across the face of the orion nebula . the subarcsecond resolution of the hst makes it ideal for this purpose . each of these temperatures represents an average along a given line of sight . since the columnar temperatures already represent an average temperature it is not possible to obtain directly from them the full @xmath1 value . instead it is possible to obtain the mean square temperature variation across the plane of the sky and from this value it is possible to estimate @xmath1 . in 2 we discuss the observations . in 3 and 4 we present the method for deriving the columnar temperature ( @xmath9 ) image together with its analysis . in 5 we formally define the columnar temperature , and the mean square temperature variation measured in the plane of the sky ( @xmath10 ) , then determine @xmath10 from the data . we then estimate @xmath1 based on @xmath10 and additional data from the @xmath9 image and temperatures . in 6 we compare our results with those of previous work . a new model for the structure of the nebula near the ionization front is presented in 7 . finally , the conclusions are presented in the intent of this observational program was to establish accurate 4363 and 5007 line ratios across the orion nebula using the wfpc2 camera ( holtzman et al . 1995 ) of the hubble space telescope ( hst ) and its narrow - band interference filters . this required imaging the nebula in the filters intended to isolate these lines ( f437n and f502n , respectively ) . since the extinction varies at small scales across the face of the nebula ( odell & yusef - zadeh 2000 ) , the magnitude of the extinction was determined on the same size scale as the line observations . this required determination of the h@xmath11 and h@xmath12 line ratios , which demanded observations with the f656n and f487n filters . since the former filter allows some emission from the nearby doublet to contaminate the h@xmath11 signal , it was also necessary to observe with the f658n filter , which is dominated by the 6583 line . there was the possibility of the h@xmath13 line contaminating the f437n signal . the anticipation was that this could be handled by scaling from the h@xmath12 observations in f487n . each emission line filter also passes the nebular continuum radiation . in the case of the emission line signal being very strong as compared with the underlying continuum it is sufficient to measure the visual continuum with the f547 m filter and extrapolate to the wavelength of the emission line filter ( odell & yusef - zadeh 2000 , odell 1998a ) . however , the continuum of the orion nebula is quite strong ( odell & hubbard 1965 , b91 ) because the large optical depth of dust particles lying just beyond the main ionization front ( mif ) back - scatter a significant fraction of the visual starlight reaching it . this back - scatter produces a continuum much stronger than that arising from atomic processes . the observations most sensitive to continuum contamination are those of the intrinsically weak 4363 line . fortunately , the f469n filter provided a good measure of the strength of the local continuum because it is free of strong emission lines when used to observe a low ionization object like the orion nebula . it is known that the he ii 4686 line ( which the f469n filter was designed to isolate in high ionization objects ) is undetected in the orion nebula even in spectroscopic studies reaching nearly 10@xmath14 the intensity of h@xmath12 ( b91 , epte98 ) . combining the f547 m and f469n observations allowed a linear extrapolation or interpolation of the wavelength dependence of the continuum flux to the wavelength of each filter . the filters employed and their pre - launch characteristics ( wfpc2 instrument handbook , version 6.1 available at http://www.stsci.edu ) are summarized in table 1 . this table also gives the wfpc2+hst system throughput as recently determined in orbit . because of the host of filters that had to be used and the length of the observing times necessary for producing an adequate signal in the fainter lines , only one field of view was covered . this was selected to include regions that had been spectroscopically studied recently using high dynamic range detectors ( b91 and epte98 ) , in order to allow a precise calibration of the filters , and to avoid the presence of the bright trapezium stars . as will be noted in 3.3 , the brightest trapezium star ( @xmath0 ori c ) was so close ( 7 ) to the ccd1 detector that low - level scattered light problems affected the long exposure images with that detector , rendering them useless for this program . the image is shown in figure 1 . the four ccd detectors of the wfpc2 are numbered counter - clockwise , with the high - angular resolution , but smaller field of view , ccd being designated as ccd1 . all exposures were made with exactly the same pointing on 2002 january 21 , 22 , and 26 and at the same gain ( electrons / adu ) values . multiple exposures of the same duration were made with each filter in order to allow correction for cosmic ray induced events on the detectors . the individual exposure times and number of exposures were : f437n(1100 s , 6x ) , f469n(1100 s , 6x ) , f487n(350 s , 3x ) , f502n(160 s , 3x ) , f547m(230 s , 3x ) , f656n(140 s , 2x ) , f658n(200 , 2x ) . all images were `` on - the - fly '' calibrated by the space telescope science institute ( stsci ) and then processed using their stsdas and the national optical astronomy observatories iraf tasks . the final product of the routine processing was a combined image such as shown in figure 1 . a considerable number of steps were necessary to convert the photometric observations with wfpc2 into 4363 and 5007 line ratios and from these line ratios into values . in this section we first describe the method of calibration of the filter properties and other components of the photometric system and their application to obtaining line ratios . we then describe the method of correcting for interstellar reddening . finally , we derive the pixel by pixel values of and estimate the statistical noise of the values . the method of calibration adopted was identical to that employed by odell & doi ( 1999 , henceforth od99 ) . however , in the present case we are only interested in the relative line fluxes , so that their equations for an absolute calibration can be simplified . the reader interested in absolute calibrations should refer to od99 , but note that an editing error occurs in their equation ( 5 ) . the term that reads @xmath15 should read @xmath16 , this error does not affect any of the numerical results in that paper . the basic approach is that one uses the wfpc2 observations of spectroscopically well studied regions to determine the calibration constants , which can then be applied to all the pixels of the image since the filter profile and system sensitivity does not vary across the image ( after the appropriate flat - field corrections have been applied ) . the most important calibration constants are determined directly from the observations of the reference regions and only the secondary constants draw on filter characteristics that were last determined before launch of the wfpcs . since the filter and detector characteristics can change with time , this method of calibration permits accurate determination of the most important characteristics . our method of deriving relative calibrations relates n@xmath17 ( the recorded number of adu counts per unit time ) to the characteristics of the source and the filter and camera system through the equation @xmath18 where we have used the following nomenclature : @xmath19=system throughput of the telescope - camera - detector system for a particular line ( excluding the filter ) @xmath20=transmission at a particular line of a filter @xmath21=surface brightness of a particular line in photons @xmath22 @xmath23=system throughput of the telescope - camera - detector system for the continuum ( excluding the filter ) @xmath24=effective width of the filter in @xmath25=surface brightness of the continuum in photons @xmath26 . it is important to note that this calibration is done in units of the number of photons . when necessary , the resulting numbers can be converted to the relative flux . in this program multiple lines contributed to the signal from a single filter only in the case of f656n ( h@xmath11 - 6563 , -6548 +6583 ) and f658n ( -6583 , h@xmath11 - 6563 ) . we considered the contribution of h@xmath13 4340 to the f437n filter , but found it to be negligible . the reference regions were all taken from b91 . they observed with a long slit of east - west orientation located to the east of @xmath0 ori c. the exact location was incorrectly expressed in b91 , but is correctly described in od99 . b91 gives line ratios and continuum to h@xmath12 ratios for 21 samples along their slit . we used the six samples that overlapped with the ccd4 portion of our images ( shown in figure 1 ) that included 4363 relative intensities . each sample was 2x 15.76 in size . the results of this calibration gave the following equations for the two important line ratios , h@xmath11/h@xmath12 and 4363/5007 . @xmath27}{(1 - 1.28\frac{n_{469}}{n_{487}})},\ ] ] and @xmath28}{[1 - 1.62\frac{n_{469}}{n_{502}}(33.4\frac{n_{469}}{n_{547}}+0.423)]}.\ ] ] equation ( 2 ) uses a full - image average value for the color of the continuum and scales the continuum intensity from the closer continuum filter . equation ( 3 ) derives the color of the continuum at each pixel and then extrapolates or interpolates from the two continuum filters , this step being necessary as the continuum correction is quite important in the f437n measurements . the continuum typically contributes about 73% of the total signal in the f437n filter . in a similar investigation of planetary nebulae rubin et al . ( 2002 , henceforth r2002 ) argued that there should be a small correction in the f437n filter for contamination in the blueward wing by h@xmath13 4340 emission . if one takes the pre - launch transmission profiles in the instrument handbook at face value , this would seem to be the case . however , it must be expected that the profiles of the filters change as they age in space . these changes are why we wanted to calibrate our observations within the data set . one would expect that the transmission near the principal lines to not change much with time , since they are located near the peak of the filters transmissions , whereas the h@xmath13 line falls far out on the wing of the filter curve . the instrument handbook indicates that the ccd sensitivities and throughput of the system have been remarkably constant . this combination of factors allowed us to empirically search for evidence of h@xmath13 contamination . we could determine the ratio @xmath29 with and without a correction for h@xmath13 contamination . with the correction the ratio was 0.42 and without the correction the ratio was 0.68 . the instrument handbook values indicate an expected value for the ratio of 0.62 . this indicates that the correction to the 4363 line surface brightness for h@xmath13 is about 10% . since the method of calibration essentially forces the calibration constants to be such as to give the right intensity ratios for the lines assumed to be present , this means that the effects of h@xmath13 correction will only enter in regions where the line and the h@xmath13 line vary considerably in relative intensity . this is not the case for the area covered by the present images ( b91 , epte98 ) . the scatter of results from the six different samples in the calibration region indicates that the uncertainty of the balmer line ratios will be 7% , while that of the line ratios will be about 12% . since the reddening corrections between 4363 and 5007 are small , the effects of the absolute uncertainty of the balmer line ratios are negligible . a 12% systematic uncertainty of the line ratios would make the derived values be off systematically by @xmath30 k. this will not significantly affect our search for fine scale variations in . as we will see in 3.3 , there is excellent agreement between the electron temperatures derived in b91 and those derived in this study . the extinction corrections are based on determining the balmer line ratio for each pixel and then calculating the total extinction from comparison with the intrinsic ratio . this total extinction was then used to correct the line ratios . we assumed with epte98 that the intrinsic balmer line flux ratio was 2.91 . using the same extinction law as epte98 means that the extinction corrected ( @xmath31 ) flux ratios could be calculated from the observed flux ratios ( @xmath32 ) by the relation @xmath33 the extinction values derived from the balmer line ratios are in good agreement with those of odell & yusef - zadeh ( 2000 ) , this being the case for both their results obtained from radio continuum to h@xmath11 ratios and those obtained from their balmer line ratios . the reddening corrected line ratios were then used to derive local values of . a convenient formulation of the relation of line ratio and is given in r2002 . their equation ( 8) can be restated as @xmath34 where we have restated the relation using the common - usage designation of the wavelengths in air ( rather than the more physically meaningful vacuum wavelengths employed by r2002 ) . since we are dealing with large numbers that are not usually determined to high precision , we sometimes will express by the frequent nomenclature @xmath35@xmath36/10,000 . the resulting image is shown in figure 2 . figure 2 shows multiple features . the most common are the `` scars '' that result from stars . these continuum sources are often saturated on some of the images used to derive . even when they are not saturated , the formal value of at the star positions is subject to big uncertainties owing to the multiple subtractions that occur in the derivation . within ccd1 there are multiple features that are related to scattered light from @xmath0 ori c and the more westward members of the trapezium . any apparent fluctuation in within ccd1 is suspect as being an artifact . there are a few features characterized by very low values of , which are discussed in 4.3 . this derivation gives for the six sample regions from the b91 study @xmath37@xmath35@xmath38=0.92@xmath390.01 , while values given in b91 for the same samples gives @xmath37@xmath35@xmath38=0.91@xmath390.02 . a comparison with the epte98 study is not possible because only their position 2 fell within a part of our image that was not useful as it fell within ccd1 and their position 1 was north of our field . it is important to assess the effect of the expected poisson noise on the determination of because of the low signal level in some of the filters . in the region that was used for calibration the total number of recorded electrons ( 7 times the number of adu s recorded per pixel ) for the key filters were f437n(1260 ) , f469n(1260 ) , f547m(8400 ) , and f502n(10500 ) . in the following discussion , we ll assume that the root mean square deviation of a signal n is n@xmath40 . this is what we will mean as the probable error ( p.e . ) . we will assume that for multiplicative factors the p.e.s add quadratically . unfortunately , one can not determine the p.e . directly from the observations since the nebula is intrinsically variable in surface brightness . an assessment of the p.e . per pixel was done by creating noisy images ( using iraf task _ mknoise _ ) of the appropriate noise level for each of the observationally determined components of equation ( 3 ) , then determining the p.e . of the resulting image . the p.e . was 20% . in most of the following discussions we will assume this uncertainty . equation ( 5 ) was then used to determined the effect of uncertainties of the line ratio on the derived value of . a 20% photometric p.e . produces a p.e . in of 5.7% . throughout this paper we will present uncertainties as probable errors ( p.e . ) calculated as the reciprocal of the root - mean square deviation of the recorded signal . the assumptions used in 3.1 for the derivation of eq . 3 are not fully justified since there are two weak lines that affect the signal in the more important continuum filter , f469n . the line appears at 4702 and the line at 4713 , where the transmissions are 0.51 and 0.15 , respectively . as we will demonstrate in this section , the effect of these lines is small in determination of in the nebula . the presence of these weak lines will make the continuum at f469n appear brighter than it is . if we call the enhanced signal in the f469n n@xmath41 and the signal that would be observed if there were only continuum radiation n@xmath42 , then their ratio will be given by @xmath43 the strength of the continuum is usually measured with respect to the h@xmath12 line at 4861 . this was done in b91 , where their figure 2 presents a quantity which is related to the equivalent width ( @xmath44 ) by @xmath8/@xmath44 . the average value of their plotted quantity is 9.5 , which gives an average @xmath44= 510 , which is in good agreement with earlier filter photometry over a wide range of the nebula ( odell & hubbard 1965 ) which showed that @xmath44 is nearly constant in the inner region of the nebula , then decreases outward . having values of @xmath44 allows us to express the observed - quantity ratios of equation ( 6 ) as s@xmath45/s@xmath46=@xmath44s@xmath47/s@xmath48 and s@xmath49/s@xmath46=@xmath44s@xmath50/s@xmath48 . making this substitution and inserting the known values allows the restating of equation ( 6 ) as @xmath51 if the ratio of line intensities with respect to h@xmath12 are constant across the region observed , then there is no net effect , since they will be constant in the calibration region and their affects will cancel out . however , if the line ratios vary , there will be an effect on the derived line ratios . anticipating that this effect is small , we address only the effect on equation ( 3 ) . since the correction term in the denominator of that equation is only about 0.015 , we can demonstrate adequately the effect of varying contaminating line strengths by considering only the terms in the numerator . if ( n@xmath41 / n@xmath42)@xmath52 is the value of the ratio of these quantities in the region used for calibration , we can express the local effect by multiplying the n@xmath53 rates by ( n@xmath41 / n@xmath42)/ ( n@xmath41 / n@xmath42)@xmath52 . the global average values ( taken from a section of ccd2 free of the peculiar effects due to shocks that are discussed in 4.2 are n@xmath54/n@xmath55= 0.0094 , n@xmath53/n@xmath56=1.04 , and n@xmath53/n@xmath57=0.016 . inserting these values into the numerator of equation ( 3 ) and applying a constant correction of 0.015 for the subtractive term in the denominator , gives @xmath58\}.\ ] ] the weak contaminating lines are seen only at high spectral resolution . in the study of osterbrock , tran , & veilleux ( 1992 ) , fluxes relative to h@xmath12 were 0.00245 and 0.00686 for 4702 and 4713 respectively ( for an east - west region 47 long centered 58 north of @xmath0 ori c ) . epte98 found ratios of 0.00179 and 0.00645 for a region 45 north of @xmath0 ori c and 0.00186 and 0.00621 for a region25 south and 10 west of @xmath0 ori c. these three samples would indicate values of n@xmath41 / n@xmath42 of 1.066 , 1.056 , and 1.055 respectively . the contaminating lines were not measured in b91 due to their lower spectra resolution . there is another set of data that provide even more information . co - authors antonio and manuel peimbert together with mara teresa ruiz obtained a long slit spectrum with the vlt3 telescope ( melipal ) and the fors1 spectrograph at the paranal observatory in chile on 2002 september 10 . the slit was oriented at a position angle of 340 and passed through the feature called hh 202-s in odell et al.1997 ( henceforth o97)and designated in figure 2 . to estimate the intensity of the 4702 line , they measured the intrinsically stronger line at 4658 and we scaled it to the 4702 intensity by using the line ratio of 0.32 from epte98 . likewise , they also measured the strength of the 4713 feature , a blend of [ ar iv ] line at 4711 and the at 4713 the [ ar iv ] line contributing about 15% of the blend intensity . the average of a 2section , which is close to the east - west calibration samples from b91 , gave ratios relative to h@xmath12 of 0.0061 and 0.0060 for 4702 and 4713 respectively , giving a value of n@xmath41 / n@xmath59=1.0551 . the scatter of this value , taken from six 20subsamples was @xmath390.0064 . a similar scatter was found from an analysis of both 20 and 2 samples . recalling that the effect of this 5% level contamination cancels out in our calibration procedure leaves us only to consider the effect of the variation in the value of due to the varying level of contamination . the variations in contamination indicated by the paranal data indicate that the corresponding variations in the derived is + 80 and -120 k. the `` temperature image '' can be used to look for variations in at large and fine scales . in this section we describe the results of a search for systematic changes , and explain the regions of apparently very low and high values of . we also present the evidence for small scale changes , and give the results of a fourier analysis of the temperature fluctuations . we looked for variations of with distance from the dominant ionizing star @xmath0 ori c. nine samples of 10x10 were taken from each of ccd2 - 4 . in each case there were no scars from bright stars and none of the low regions discussed in the next section . the average gave @xmath37@xmath608890@xmath39496 k. no statistically significant correlation was found when trying to fit a linear relation to the observed values of and distance from @xmath0 ori c. we conclude that there are real variations in across the region sampled in this study , but there is no obvious relation to the position in the nebula . the scatter of 496 k is significant since the scatter due to poisson statistics within each sample would be only 5 k and we estimated in 3.5 that the scatter due to variations in the contaminating weak lines would be about @xmath39100 k. there are several regions within ccd2-ccd3-ccd4 that have anomalously low values of the derived , as seen in figure 2 . these all fall within the province of known herbig - haro ( hh ) objects hh 202 ( cant et al . 1980 , o97 ) and hh 269 ( walter et al . 1995 , bally , odell , & mccaughrean 2000 ) . the former is probably the result of a pre - main sequence star s blueshifted jet striking the foreground veil of neutral material(o97 ) . the latter is probably the result of blueshifted jet outflow from a pre - main sequence star that lies behind the mif and has yet to be identified ( odell & doi 2003 ) . it is almost certain that hh 269 has the same source as that producing the series of shocks extending to the east and designated as hh 529 ( bally , odell , & mccaughrean 2000 ) and it is likely ( odell & doi 2003 ) that this is also the source for the jets producing hh 202 and the double object hh 203+hh 204 that lies to the southeast , near @xmath61 ori a. these anomalously low features are probably not real , rather , they are the result of contamination of the f469n filter . our method of calculation of the line ratios assumed that the f469n filter was free of any emission lines , i.e. it was used as a measure of the continuum . in all of the nebula for which spectra exist this assumption is good since the star producing most of the ionization of the nebula ( @xmath0 ori c ) is too cool to doubly ionize helium and there are only the two weak contaminating lines ( c.f . hartigan ( 1999 ) has shown that if a fast jet forms a bow shock in material that is already photoionized then one can expect he ii to be present . he used ground based images to look for evidence for emission in the he ii 4686 line . the only region where he thought that this line was present was in hh 202 , but with his low resolution images he could not conclude that 4686 emission was present . the present high resolution images show that there are anomalies at several of the shocks known to exist in this field of view ( but certainly not in all the known shocks ) . the anomalous feature to the southwest from hh 202-s corresponds to a faint shock visible only in h@xmath11 and , but not in ( o97 ) . the paranal spectra described in 3.5 were centered on hh 202-s . the spectra show no 4686 emission . however , there is a considerable enhancement of emission . in a sample centered on hh 202-s , the flux ratio of 4658 to h@xmath12 was 0.031 , which predicts a 4702 to h@xmath12 ratio of 0.010 ( using the flux ratio from epte98 ) . together with a normal 4713 contribution , this would raise the ratio n@xmath41 / n@xmath42 to 1.18 , which would artificially reduce the 4363 ratio by a factor of 4.6 , and decrease the derived value of to 6300 k. since the paranal sample on hh 202-s was 20 long and the object is much smaller , the contamination would be much larger on the object itself and the artificially derived correspondingly lower . direct confirmation of contamination by 4702 was obtained by high velocity resolution spectra obtained by co - author odell and will henney with the manchester echelle spectrometer ( meaburn , et al . 1984 ) on the 2.1-m telescope at san pedro martir observatory on 2002 october 23 , where one sees significant enhancement of that line , peaking at the optically brightest portion of hh 202-s , with no indication of 4686 emission . this means that enhanced emission is a common feature of the anomalously low spots seen in figure 2 , rendering useless the values there . why is enhanced in these shocks is in itself an interesting problem that should be addressed . several proplyds occur within our sample , the brightest within ccd2 - 4 are 141 - 301 and 182 - 413 . these proplyds and the fainter , less well resolved ones all show enhanced values of . in this case the anomaly is probably due to the mis - application of equation ( 5 ) , which is applicable only in the limit of electron densities ( @xmath62 ) low as compared with the critical density ( @xmath63 ) for . @xmath63 is the density at which collisional de - excitation of the upper state giving rise to an emission line is equal in probability to a spontaneous emission . this is only of concern for the 5007 line as its @xmath63value is much lower than that for the 4363 line . rubin et al . ( 2003 , henceforth r2003 ) give @xmath63(5007 ) @xmath64 6.4x10@xmath65 @xmath66 , which is much higher than densities encountered in the nebula , but comparable to characteristic values for the proplyds ( odell 1998a ) . this means that in the local ionization fronts of the proplyds the 5007 emission is being suppressed by collisional de - excitation and the value derived from equation ( 5 ) will be an upper limit . this same interpretation is applied by r2003 to their high for the proplyd 159 - 350 , which lies in the center of our ccd1 image and fell in their long slit spectrum . in their models of the proplyds close to @xmath0 ori c henney et al . ( 2002 ) argue that the formal temperatures derived without consideration of collisional de - excitation would be about 20,000 k. if the densities of the more distant proplyds are much lower , as indicated by their surface brightnesses , then the effects of collisional de - excitation are less and some of the temperature increase we find is real . it is not clear if the similar local increase in at the jet feature just south of the dark rim ( odell & yusef - zadeh 2000 ) in the southeast corner of ccd4 is due to this same effect . the jet has a known high radial velocity ( odell et al . 1997 ) and proper motion ( odell & doi 2003 ) , so it may be that is enhanced where this flow shocks against ambient nebular gas . figure 2 indicates that there are variations in even though we ve shown in 4.1 that these variations are not a function of distance from @xmath0 ori c. the smaller scale changes have been illuminated by looking at the three sample regions depicted in figure 2 , these samples being selected to be free of the [ fe iii ] contamination . the results are shown in figure 3 . in each case the range of temperature is @xmath39 1000 k. for reference , random noise at the level of 5.7 % probable error is shown , this being the level expected for the bright portions . the southwest sample is about one fourth the surface brightness of the other samples . the noise level was brought down to that of the other samples by taking samples of 2x2 pixels in this region . examination of figure 3 shows that the face of the nebula is mottled with small scale variations of with angular dimensions of about 10 and amplitudes of about 400 k. in order to gain a more quantitative evaluation of the data , we performed a fourier analysis of two regions of the @xmath35 image . the first was a 400x400 pixel section in the northwest portion of ccd2 and the second was 250x250 pixels in the northeast section of ccd4 , both being free of camera scars and the local apparent drops that are due to shocks , as discussed in the analysis was done using stsdas tasks with iraf . the procedure was the same for each sample . the first step was to subtract the average value of the sample from each pixel and then to make a power spectrum of the resulting image , using the task _ powerspec_. a radial profile was taken of the power spectrum image using the center as the origin . the result was a profile that dropped down to a constant value at large spatial frequency values and represents the noise level due to the poisson noise in the data . this constant value was then subtracted from the radial profile of the power spectrum . the resulting noise - subtracted power spectrum was then converted to the physically more meaningful length space by multiplying the power spectrum in spatial frequency by the square of the frequency and binning the result in the spatial dimension . the results are shown in figure 4 and figure 5 . these plots show that there is a large amount of power at the finest scales ( sub - arcsec ) but that power is found at all scales . this analysis is similar to that of vannier et al . ( 2001 ) of h@xmath67 emission arising near the imbedded bn - kl sources to the north of ccd4 . the aims of this section are to estimate representative values of the temperature variations [ @xmath1(o@xmath68 ) , @xmath1(o@xmath69 ) , and @xmath1(h@xmath69 ) ] in the orion nebula in order to compare them with predictions of photoionization models and to be able to study the physical conditions of the nebula including its chemical composition . we proceed in three steps : 1 . ) use of the @xmath9 map obtained in this paper to determine @xmath10(o@xmath68 ) across the plane of the sky ( this is the first detailed 2-dimensional set of temperatures across of the orion nebula ) , 2 . ) estimate the 3-dimensional value of @xmath1(o@xmath68 ) ; 3 . ) estimate @xmath1(h@xmath69 ) including both the o@xmath70 and the o@xmath68 regions . in order to obtain the observed mean square temperature variation in the plane of the sky , @xmath10(o@xmath68 ) , we will define a set of equations . the average temperature , @xmath71 , and the mean square temperature fluctuation , @xmath72 , are given by @xmath73 and@xmath74 respectively , where @xmath75 and @xmath76 are the local electron temperature and density , @xmath77 is the local ion density corresponding to the observed emission line , and @xmath78 is the observed volume ( peimbert 1967 ) . by recognizing that the observations correspond to averages over a given line of sight , it becomes necessary to define the columnar temperature , , as @xmath79 where @xmath11 is the right ascension and @xmath80 is the declination of a given line of sight ( corresponding to a given pixel ) . then , the average temperature can be written as @xmath81 and the mean square temperature variation over the plane of the sky , @xmath10 , can be defined as : @xmath82 in table 2 we list a set of @xmath10(o@xmath68 ) values obtained from equation ( 13 ) and from the @xmath83 map presented in figure 2 . to determine @xmath10(o@xmath68 ) we have excluded several features in order to diminish the noise of the samples : the pixels with very low flux , where we expect the temperature deviations to be mostly due to errors and not to physical inhomogeneities ; the shocks , where the destruction of dust could be enhancing the 4702 line and thus producing spurious temperatures ; the proplyds , whose density is not well represented by equation ( 5 ) ; the seams between the ccds , where there are border effects . keeping in mind these exclusions we defined the six regions from table 2 as follows : regions 1 and 2 include most of the pixels of the se and sw regions as defined in figure 3 ; region 3 includes most of the pixels in a 54 `` @xmath84 73 '' region in the eastern part of ccd4 ; region 4 is the largest sample , it includes most of the pixels from ccd2 , ccd3 , and ccd4 ; finally region 5 includes approximately 72% of the pixels of region 4 where , in order to enhance the s / n ratio , the southern 10% of ccd2 , the southwestern 50% of ccd3 , and the western 25% of ccd4 have been removed , the regions removed are the ones of relatively low emission measure . although the signal in ccd1 was high , the presence of scattered light artifacts ( c.f . 2 ) prevents its use in our analysis . column 2 , of table 2 , presents the number of independent temperatures used in the region ; column 3 presents the average temperature , weighted by @xmath85 ( defined in equation ( 12 ) ) ; while column 4 presents the measured @xmath10(o@xmath68 ) for each region . the @xmath10 of regions 1 , 2 , and 3 is in the 0.0050.016 range . the differences in @xmath10 are real since they are larger than the expected observational errors . this could have 2 explanations : a ) region 3 is more homogeneous than regions 1 and 2 ; b ) another possibility is that @xmath1 is similar in the 3 regions , and the variations along each line of sight in region 3 are larger , while there are not so many variations along the lines of sight in regions 1 and 2 , thus giving a set of more homogeneous values of @xmath9 for region 3 than for regions 1 and 2 . we added most of the pixels from ccd2 , ccd3 , and ccd4 to generate region 4 . figure 6 presents a histogram of the 1.5 million temperatures included in region 4 . the mean square temperature deviation obtained from the histogram is @xmath86 ; which is not @xmath10 . it is not possible to derive directly @xmath10 from the histogram since each temperature has to be weighted by the corresponding emission measure ( see equation ( 13 ) ) . after weighting each value for the emission measure we obtain a @xmath87 for region 4 . we still have to subtract a value associated with the noise . the average p.e . in @xmath88 for the whole sample is 5.7 % , however this value is for an average pixel , while the measured @xmath89 gives more weight to the brightest pixels , the ones with the smaller errors . by considering that the average pixel is 1.89 times fainter than the average pixel weighted by emission measure , we expect the representative p.e . to be 4.2 % , which corresponds to an error on @xmath10 of 0.0017 , that is a @xmath90 . if we take out the noisiest parts of region 4 ( the outermost parts ) , we have region 5 . here the measured @xmath91 . for region 5 we expect the noise to be about 2/3 the noise of the entire region , that is a p.e . of 3.8 % per pixel . in region 5 the average pixel is 1.31 times fainter than the average pixel weighted by emission measure ; thus we expect the representative p.e . to be 3.3% , which corresponds to an error on @xmath10 of 0.0011 , that is a @xmath92 . we expect the @xmath10 result for region 5 to be more accurate ; however , by excluding the outermost parts we seem to be ignoring a fraction of the object which is intrinsically cooler . therefore we will adopt the value derived from region 4 , @xmath93 , as representative for the whole object . this value is similar to the one obtained by r2003 where they find @xmath94 values of 0.0095 , for their slit 1 , and 0.0104 , for their slit 2 . unfortunately , their slits 1 and 2 are located in our unusable ccd1 , which precludes a direct comparison . up to this point we have presented the results in terms of @xmath94(o@xmath68 ) , but what is of physical interest are the variations in three dimensions @xmath1(@xmath96 ) . in order to obtain the total @xmath1(@xmath96 ) value we need to consider the variations along the line of sight . it can be shown that the relevant equation is : @xmath97 where @xmath98 , the variation along a given line of sight , is given by : @xmath99 and the average over all lines of sight is given by : @xmath100 there are two estimates that we can make of @xmath98 : the first based on geometrical considerations and the information contained in @xmath101 ; and the second based on the ratio of @xmath96 collisionally excited lines to @xmath102 recombination lines . from geometrical considerations one would expect the inhomogeneities of one additional dimension , @xmath98 , to be at least half as large as the inhomogeneities of two dimensions , @xmath94 ; this works for large scale inhomogeneities . the power included in @xmath98 increases in the presence of small scale inhomogeneities . this is because the many independent thermal elements along each line of sight will be masked in the averaging process , greatly reducing the total @xmath94 compared to @xmath95 . therefore , @xmath98 would be expected to be much larger than @xmath94 . from the previous considerations , we estimate that @xmath103 ; this result is very model dependent which is reflected by the large range in the estimated @xmath104 value . this estimate together with the @xmath94(o@xmath68 ) value derived in the previous subsection yield @xmath105 . from the ratio of o ii recombination lines to lines in two regions of the orion nebula epte98 find that @xmath95(o@xmath106 . this @xmath95 value is representative of a very small fraction of the nebula , so we expect it to be a lower bound to the total @xmath95(o@xmath68 ) . at the same time it must be bigger than @xmath98(o@xmath68 ) because it represents an average over more than one line of sight . this argument , together with the @xmath94(o@xmath68 ) value estimated in the previous subsection , yields @xmath107 . by combining both estimates we find @xmath108 , which implies that there is power in small scale inhomogeneities . our prefered value of @xmath98 together with the @xmath94 measured in the previous subsection and equation ( 14 ) gives us a @xmath109 value . we are interested in the global @xmath95 value that we will define as : @xmath110 in order to derive the physical condtions in the region . in an region , like the orion nebula , where oxygen is only once or twice ionized , it is useful to define a value @xmath13 , that characterizes the degree of oxygen ionization , as : @xmath111 dl da};\ ] ] for such an region @xmath1 ( ) is given by ( peimbert , peimbert , & luridiana 2002 ) : @xmath112 ^ 2 \over t_0({\rm h}^+)^2},\ ] ] the first term corresponding to the temperature variations in the o@xmath68 zone , the second term corresponds to the variations in the o@xmath70 zone , and the third term corresponds to the thermal variations introduced by having two zones with different average temperatures . from the observations of peimbert & torres - peimbert ( 1977 ) and epte98 it follows that @xmath13 decreases from about 0.85 near the center of the nebula to about 0.15 in the outer regions . we will adopt a value of @xmath113 as representative of the whole object . it is necessary to estimate @xmath1(o@xmath70 ) . since @xmath114 is rarely measured we will use @xmath115 as a surrogate since we expect both temperatures to be similar as they arise from the same narrow + zone . r2003 present values of @xmath116 ; their measured values are in very good agreement with their measured values of @xmath89 ; therefore we will assume @xmath117 . it is impossible , however , to estimate @xmath118 from any set of observations available . from the geometrical arguments presented in the previous section and the similarities between the @xmath89 and @xmath119 presented by r2003 we will assume also that @xmath120 , and consequently that @xmath121 . to compute the final term in equation ( 19 ) we need to estimate the differences in temperatures between the o@xmath70 and o@xmath68 zones ( again we will use @xmath115 as a surrogate ) . our observations do not yield @xmath115 values , therefore we are forced to look for them in the literature . we are more interested in finding @xmath122 than @xmath115 to avoid possible biases , since the calibration is uncertain by a few hundred degrees ( see final paragraphs of sections 3.1 and 3.5 ) ; an additional advantage of such comparison is that papers in the literature cover regions considerably smaller than the one presented in this paper . epte98 have measured @xmath123 k and @xmath124 k ; while r2003 , for their slits 1 and 2 and assuming @xmath125 , obtain @xmath126 k and @xmath127 k. by averaging the innermost 9 regions of b91 , we find @xmath123 k and @xmath128 k. these three results give us an average @xmath129 k. as mentioned before , we have adopted a @xmath113 , giving a @xmath130 approximately 700 k higher than @xmath131 . the inputs presented above yield values of 0.011 , 0.010 , and 0.007 for the three terms in equation ( 19 ) . therefore the total magnitude of the temperature fluctuations becomes @xmath132 . notice that for values of @xmath13 in the 0.3 to 0.7 range the change in total @xmath1 is very small since @xmath121 , thus changes in @xmath13 will make very small changes in the sum of the first 2 terms of equation ( 19 ) , while the third term will change by less than 15% for that range of @xmath13 . the total @xmath133 value derived in this paper is considerably larger than the 0.004 value derived by peimbert et al . ( 1993 ) from the one dimensional model of the orion nebula presented by b91 . the type of temperature variations reported here have not been seen before because previous slit spectra determinations usually employed integrations over long areas of the nebula . the best slit spectrum sub - sampling until r2003 is in b91 , where the sample sizes were as short as 16 and one could see sample to sample changes in of about 300 k ( b91 , table 5 ) . their changes were comparable to the uncertainties in their derivations of from the line ratios , so this variation did not attract attention . however , the b91 variations are consistent with the small scale variations we see in our figure 3 . a much more valuable study is that of r2003 , where they employed the hst s stis ( space telescope imaging spectrometer ) in a long slit mode to determine in samples of 0.5x 0.5 in both and . they presented results for about 400 samples from each of four slit settings , two within our ccd1 field and the others near or off the southeast corner of our ccd2 field . it is possible to use this data to estimate a lower limit to @xmath95(h ii ) using the same procedure as in the previous section . from their slits 1 and 2 , those included in our map , it is found that @xmath134 k and @xmath135 k ; notice that these @xmath136 s corresponds to two rectangular columns of the nebula and not to a large volume . from these slits we also find @xmath137 and @xmath138 . from r2003 observations it is not possible to obtain direct estimates of the o@xmath68 and n@xmath69 temperature fluctuations along any given line of sight ; so we will use @xmath89 and @xmath116 as lower bounds to @xmath139 and @xmath140 . to estimate the two zone temperature term , the term due to the difference between @xmath141 and @xmath142 ( see equation ( 19 ) , we will adopt @xmath143 . this , together with @xmath141 and @xmath142 , yields a contribution of 0.009 . finally , adding these 3 terms , we obtain @xmath95(h ii)@xmath144 . this lower limit to @xmath95(h ii ) is consistent with our determination and with the @xmath95 determination by epte98 . moreover it is already a factor of 5 higher than the values predicted by one - dimensional photoionization models of the orion nebula , like the one by b91 . in this section we will review the large scale structure of the orion nebula ( 7.1 ) , present a conundrum arising from the presence of fine scale structure in images ( 7.2 ) , demonstrate that high resolution spectra and line ratio changes argue for a three dimensional fine scale structure ( 7.3 ) , argue that the ionization front is advancing into a highly clumpy pdr ( 7.4 ) , show how these clumps can lead to low shadows within the ionized layer that can help to explain the @xmath1 phenomenon in the region of low ionization regions ( 7.5 ) , and present a preliminary discussion of the effect of these clumps on the anomalous line - broadening in orion ( 7.6 ) . the accepted model for the orion nebula is that of a blister . the visible nebula is the result of a thin layer of photoionized gas on the front of a giant molecular cloud , with a dense photon - dominated region ( pdr ) ( van der werf et al . 1996 ) , also known ( young owl et al . 2000 ) as a photodissociation region , on the neutral side of the ionization front . gas is accelerated away from this ionization . the dominant ionizing star is @xmath0 ori c which lies at about 0.25 pc in front of the ionization front . in the foreground is an irregular veil of neutral material that produces most of the extinction ( odell 2002 ) . this model is reviewed in detail in a recent review article ( odell 2001b , henceforth od2001 . the basic physics is discussed in a review paper by ferland ( 2001 ) . the shape of the ionization front has been derived ( wen & odell 1995 , henceforth wo95 ) using a basic process originally proposed by gary ferland in b91 . this draws on the fact that at a substellar point , the surface brightness in a recombination line ( for example h@xmath11 ) is directly proportional to the local flux of ionizing photons . wo95 solved the more general problem , where the line of sight to a point in the nebula does not pass through the ionizing star . they showed that the ionization front is an irregular concave form , with an escarpment that explains the bright bar feature running near @xmath61 ori a and has a dominant hill to the southwest of @xmath0 ori c , where the orion - s radio and infrared sources are located . odell & doi ( 2003 ) have shown that the latter region is the source of multiple optical and radio outflows and probably contains a secondary source of star formation . this model was derived using a resolution of 2 . because of the progressive nature of the calculation of the shape of the surface , it is most accurate near the direction of @xmath0 ori c. the only diagnostics that we have along the line of sight into the nebula are the increasingly blue - shifted emission from the higher ionization states that are found further from the mif and forbidden line doublet intensities that indicate a decreasing density with increasing distance from the mif ( od2001 ) . these few tests that we have are consistent with the model . a detailed hydrodynamic model of the orion nebula has not been calculated , however , it is expected that the high density in the pdr region is the result of a shock moving into the molecular cloud and that the ionization front is slowly progressing into the neutral region . the hst images of the orion nebula are filled with detail , especially in the inner regions . at first blush , this appears to present a fundamental quandary if the standard model for the nebula is correct . one can derive a formal thickness ( l ) for the ionized layer , using the assumption that the density is constant and knowing the emissivity in the recombination line being observed . wo95 show that this value is characteristically about 0.13 pc for the central part of the nebula , which corresponds to an angle of 58 if the distance to the nebula is 460 pc ( bally , odell , & mccaughrean 2000 ) . they also show that if the density has an exponential decrease , its scale - height will be l/2 ( 29 ) and the emissivity would decrease with a scale - height of l/4 ( 15 ) . this means that in the standard model one should not be able to see variations in the structure of the nebula with size scales smaller than about 15 . one certainly see lots of fine - scale ( less than 15 ) structure in ground - based and hst images . some of this can be attributed to variations in the amount of extinction ( odell & yusef - zadeh 2000 ) , most of which is caused by the foreground neutral veil of material ( od2001 ) . however , the extinction corrected images in h@xmath11 show significant structure right down to the limit of the extinction corrections , which was about 1.7 . this is shown in figure 7 , which is a profile of the extinction corrected h@xmath11 image of the nebula ( odell & yusef - zadeh 2000 ) derived with a resolution of 1.7 and taken across the velocity sample region designated in figure 1 . there are even smaller features , which appear to be a continuation of the structure , as shown in the hst images , while the smallest features are due to the outflows from young stellar objects and the shocks they create in the low density ambient gas ( odell & doi 2003 , od2001 ) and are seen in the hst images to be sub - arcsecond in extent . the presence of fine scale structure associated with the main emitting layer of ionized gas presents a fundamental challenge to the accepted model of the nebula . the structure is there , for we see it , but if the ionized gas was simply a smooth flow of gas away from a homogeneous ionization front , we should not see any small scale structure . we demonstrate in 7.5 and 7.6 , that the resolution of the quandary lies in the emitting layer having a much more complex structure than assumed in published models . there are two additional lines of evidence indicating that small scale structure exists within the ionized layer , these being the structure seen in the velocities and fine scale variations in the ionization . both argue for structure down to at least a few arcseconds . the internal velocities within the orion nebula have been the subject of investigation for nearly a century , beginning with a fabry - perot study by fabry ( buisson , fabry , & bourget 1914 ) , an extensive mapping using photographic detectors ( wilson et al . 1959 ) , and a series of studies in various emission lines using ccd detectors ( [ o iii ] , castaeda 1988 ; [ o ii ] , jones 1992 ; [ o i ] , odell & wen 1992 ; [ s iii ] , wen & odell , 1993 ) at the kitt peak national observatory coud feed spectrograph at selected position angles with respect to the trapezium stars . this work was used to compare the radial velocities with the predictions of turbulence within a thin ionized layer ( von hrner 1951 ) and the results are summarized in od2001 . a more ambitious and complete mapping of the radial velocities is now being created as part of the rice university phd thesis of takao doi , who is creating velocity images of the huygens region at resolutions of 10 km s@xmath145 and 2 in the h@xmath11 , 5007 , and 6583 lines , and a similar study at the instituto de astronoma , unam , morelia , mxico in the red doublet lines by w. j. henney & m. t. garca - daz that uses in part data obtained by co - author odell . with the permission of our colleagues we have used their north - south orientation slits to create figure 8 and figure 9 . these are spectra taken from the velocity sample region indicated in figure 1 and used for the creation of the surface brightness profiles of figure 7 . the instrumental full width at half maximum intensity ( fwhm ) was very close to 10 km s@xmath145 in each case . although the original samples along the slit were at intervals of 0.64 , the seeing was characteristically about 2 during the observations . the former figure shows a half tone depiction of the spectra in lines characterizing various parts of the ionized layer and the latter presents contour diagrams of the same data . the power of this approach lies in the fact that various emission lines are characteristic of different samples of the ionized layer . although this fact is implicit in the discussion of several texts , perhaps the best illustration is in a discussion of structure of the helix nebula ( odell 1998b ) . emission comes from throughout the ionized layer , its emissivity being weighted towards the lower regions along the line of sight . the forbidden line emissivity in , , and are all weighted towards the higher regions along the same line of sight . delineation of information about structure along the line of sight is also provided by the ionization differences . emission will arise exactly in the ionization front , where ionization of hydrogen rapidly transitions from nearly unity to nearly zero . a good fraction of the emission will occur in the adjacent thin + zone , while the emission will arise in the outermost ( closest to the observer ) + zone . we see in figure 8 and figure 9 that there is considerable velocity variation along the slit and that the velocity in one ion is not well correlated with that in the other ions . thermal broadening of the h@xmath11 line precludes a detailed analysis of its velocity structure , but is low enough in the other , more massive , ions to see the differences clearly . one sees that the narrower the expected emitting region , the more structure there is along the slit . emission gives evidence for velocity structure with a scale of the seeing values of 2 and emission has a characteristic size scale down to about 5 . these data argue that structure exists within the main ionized layer at sizes down to at least 2 and that gas of different states of ionization is moving at different velocities . it is the spatially unresolved average velocity along the slits which give the characteristic progression to more blueshifted radial velocities at higher states of ionization that is the earmark of a thin blister of gas expanding towards the observer . the second line of evidence for fine scale structure within the nebula is from comparison of emission from very different states of ionization or conditions of emissivity . for this comparison we used two pairs of images , whose closeness in wavelength essentially eliminate the effects of interstellar extinction . the first pair used the f487n images , which are primarily due to h@xmath12 emission whose emissivity is like that of h@xmath11 in that it selectively comes from the lower regions , and the f502n images , which are primarily emission from 5007 high collisionally excited regions . the second pair used the f656n h@xmath11 filter and the f658n filter . the considerations for the emissivity are similar in the second pair of filters , but the emitting layer of the emission should be thinner . the hydrogen emission arises from throughout the ionized zone , but the and emission arise from selected zones , as described earlier in this section . the results of these comparisons are shown pictorially in figure 10 . one sees that there are wide variations in these ratios . some of these changes are due to shocks caused by outflows from pre - main sequence stars ( odell & doi 2003 ) while others such as the bright bar region in the southwest corner of the images is due to large scale topographic features of the ionization front ( odell & yusef - zadeh 2000 ) . in addition to these , there are many additional features . a more quantitative depiction is shown in figure 11 and figure 12 , where profiles along the same region as the velocity samples are presented . we see fine scale structure in the ratios down to nearly the size of the 1x2 bins over which the data was averaged , although a characteristic value would be 3 to 4 . the probable errors of the ratios were derived from the total counts used in their calculation and are 1.8 % for the f658n / f656n ratio and 1.2 % for the f502n / f487n ratio . as expected , the peaks are sharper in the f658n / f656n ratio , indicating that the sources are small concentrated regions . the fact that there is a general anti - correlation of the ratios indicates that the variations are largely due to ionization structure , rather than temperature differences . in summary , we can say that both the spatially resolved high resolution velocity data and the line ratio data indicate structure in the main ionized layer of 3 to 5 . a comparison with a profile of the results of the determination of is also useful and is shown in figure 13 , where a profile along the same region as the velocity data is shown . we see that the proplyd 141 - 301 shows the common characteristic of the other proplyds , in having a higher than normal formal value of @xmath35 , but , as shown in 4.3 , this is probably due to collisional de - excitation of the electron state giving rise to the 5007 emission . the `` jet '' feature shows a similar enhancement , which may also be due to having densities comparable to the critical density for this nebular transition , but it could be real , being the result of shock heating at the interface with the ambient nebular gas . aside from these features , there are variations in all along the slit . in the well exposed region from 50 to 150 we calculate that the probable error of is 0.5% i.e. 45 k. , which is small as compared with the observed variations . there are no obvious general patterns of changes with the extinction corrected surface brightness in h@xmath11 , the velocity changes , nor the line ratios . there is considerable evidence that the pdr behind the ionization front of the orion nebula is very clumpy ( tauber , goldsmith , & dickman 1991 , hogerheijde , jansen , & van dishoeck 1995 ) . hcn and hco@xmath69 observations by young owl et al . ( 2000 ) demand that the portion of the pdr seen obliquely behind the bright bar feature possesses dense clumps ( 3 x 10@xmath146 @xmath66 ) embedded within an interclump medium of lower density ( 5 x 10@xmath147 @xmath66 ) . more direct observations come from imaging in h@xmath148 . van der werf et al . ( 1996 ) show that clumps exist from a scale of 40 down to their resolution limit of 1.5 . the highest angular resolution study in h@xmath148 is that of vannier et al . ( 2001 ) who studied a region northwest of @xmath0 ori c at a resolution of 0.15 . again , they saw clumpy structure , in this case going down to 1.5 - 2.0 . a clumpy structure in the pdr is not unexpected and there would be several instability mechanisms that operate there ( capriotti 1973 , brand 1981 , vishniac 1994 , garca - segura & franco 1996 , williams 1999 ) . the implications of these clumps for the nature of the ionized layer of gas in the nebula are very important since it is into these clumps and the interclump medium that the ionization front advances . where the density is lower , the advance will be more rapid . the expected rate of advance of the ionization front is not known as the hydrodynamic problem has not been solved . the orion nebula is old enough that it is in a state of quasi - equilibrium and the velocity will depend on many factors , including whether or not the photoionizing star @xmath0 ori c has a velocity component towards or away from the molecular cloud . this means that one would expect the ionization front advancing into a clumpy pdr to move beyond the dense clumps and be retarded at the clumps , so that a microscopic view of the ionization front would be like that of a stubble - field . the region behind a dense knot would then be shadowed from ionizing lyman continuum ( ) radiation coming directly from @xmath0 ori c and would be illuminated only by diffuse ( sometimes called scattered ) photons . where the density is insufficient to form a quasi - static ionization front , the clumps will simply retard the advance of the ionization front , while the dense clumps will form a small local ionization front that will exist until it has lost sufficient material through photoevaporation that the ionization front can penetrate , then totally disperse it . the shadowed regions behind the knots within the ionized layer can be sources of low temperature emission and of temperature variations , therefore they will increase the @xmath1 value . the photoionization equilibrium in such a shadowed region was considered 30 years ago ( van blerkom & arny 1972 , kirkpatrick , 1972 , capriotti 1973 ) , but it is only recently that the more general case has been considered ( cant et al . 1998 , henceforth c98 ) . the region illuminated only by diffuse lyc will see a lower temperature radiation field because most of the recombinations of that produce the diffuse lyc will result in a photon just above the ionization threshold , while the direct stellar field will have more high energy photons and produce higher regions . c98 have solved the general problem , rather than trying to apply their calculations to only a specific object , including the role of ambient gas density and the time variability of the shadowed zone structure . when the combination of the stellar radiation density ( the diffuse radiation field density is about 15 % of that ) and gas density is right an ionization front can form immediately at the edge of the shadowed cylinder of gas . as the radiation density increases and/or the density decreases , the ionization front can form within the shadowed cylinder and even produce a condition where the shadowed cylinder is completely photoionized . c98 argue that within the ionized shadow zone will be about 6000 k , considerably cooler than that of the directly illuminated gas . the radiation from the shadowed zone can be seen by the distant observer for two reasons . in some cases the angle between the line of sight from the observer to @xmath0 ori c can be quite different from the angle between @xmath0 ori c and a local spot on the ionization front , allowing direct view of the shadowed region . even when that is not the case , a significant fraction of the radiation from the shadowed region will be scattered by dust in the nearby pdr , in the same way that radiation from the ionized layer is scattered and produces the observed redward asymmetry of the emission lines ( odell , walter , & dufour 1992 , henney 1998 ) . shadowed regions have already been detected in the orion nebula . in a study of both the cometary knots in the helix nebula and the proplyds in the orion nebula , odell ( 2000 ) showed that the proplyds cast ionization shadows as long as 0.2 pc . the proplyds are optically thick to lyc radiation ( odell 1998a ) and in that sense resemble the optically thick clumps we argue exist at the main ionization front of the nebula . the proplyd ionization shadows are most visible when the geometry is such that the shadow falls into the ionized layer that provides most of the radiation identified as the nebula . odell ( 1998a ) showed that the shadows of the proplyds covered about 0.5% of the view from @xmath0 ori c. by assuming : pressure equilibrium , @xmath149 , @xmath150(n@xmath70 ) = 10,100 k ( b91 , epte98 , r2003 ) for the directly illuminated regions , and @xmath150(n@xmath70 ) = 6,000 k ( c98 ) for the shadowed regions we obtain a contribution of 0.0093 to @xmath1(o@xmath70 ) ( see equation ( 10 ) ) , that is a contribution of 0.0043 to the total @xmath1 ( due to second term of equation ( 19 ) ) . to reach the estimated 0.011 contribution to @xmath1 due to the o@xmath70 zone , we need an additional 0.7% of the volume of the nebula to be in a different type of ionized shadowed region . in what follows we will propose the presence of high density clumps inside the ionized layer left behind by the advancing ionization front . in the model we propose ( illustrated in cartoon form in figure 14 ) the clumps would not be permanent , rather , they would exist in the ionized layer only until they are destroyed by photoevaporation . the cool columns of gas behind them would then come to the higher equilibrium appropriate for illumination by @xmath0 ori c rather than diffuse lyc . one can approximate the results of an ionization front advancing into a clumpy pdr , as described in 7.4 . we dont know the velocity of this advance . it must be small , because the tracers of the ionization front ( [ o ii ] and [ s ii ] ) are blueshifted 2.5 km s@xmath145 with respect to the pdr tracers ( co , c ii ) ( odell 2001a ) . this blueshift is probably due to the hot gas tracers coming from a layer that has already been accelerated by the pressure gradient , rather than being the result of @xmath0 ori c moving away from the pdr . the radial velocity of @xmath0 ori c is highly uncertain because of variability probably arising in the extended atmosphere of the star . it is unlikely that this star is moving at a high velocity with respect to the other cluster stars and their velocity is the same as the pdr ( odell 2001a ) within the uncertainty of 3 km s@xmath145 . in the discussion below we will assume that the ionization front is advancing into the pdr at a rate of 2 km s@xmath145 , but it should be recognized that this could be off by a factor of two . in 7.5 we showed that the fluctuations in the surface brightness of the nebula indicate structure with a height of about 0.025 pc . this would mean that the characteristic clump will have been illuminated by lyc radiation for about 1.3x10@xmath147 years , which we will call the kinematic age . if we can determine the column density and rate of loss of atoms per unit time and area , we can calculate the time that the clumps can survive photoevaporation and have a check on our basic model . if the clump diameters are about 0.0044 pc ( corresponding to 2 ) and the densities are 3x10@xmath146 @xmath66 ( young owl et al . 2000 ) the average column density of a clump would be 2.8x10@xmath151 atoms @xmath152 . the corresponding mass would then be about 7x10@xmath153 g ) . the photoevaporation rate for these clumps will be very similar to those of the proplyds , which henney et al . ( 2002 ) have shown to be 8x10@xmath154 m@xmath155 yr@xmath145 . since these objects are comparable in size to the clumps at the ionization front , their average rate of loss of atoms is relevant , this being 2x10@xmath156 atoms @xmath152 s@xmath145 . in this case , the lifetime of the clumps would be about 5,000 yrs . however , the proplyd measured by henney et al . ( 2002 ) must be much closer to @xmath0 ori c than the ionization front , so that the rate of loss has been overestimated and the survival age of the clumps underestimated . this means that the survival age of the clumps and their kinematic ages are compatible . they are , however , close enough that the process involved must be a dynamic one , with any one clump surviving for only about 10@xmath147 years , but the number of clumps present at any time would be about constant as new clumps are revealed by the advancing ionization front . we can estimate the effect of these shadowed regions if we know their emissivities and volumes . if the shadowed regions exist long enough that dynamical equilibrium is established with their surroundings ( c98 ) , then their volume emissivity will be four times that of the ambient lower density and higher gas . if pressure equilibrium is not established , then the emissivity increase will be about two times that of the ambient gas . if one interprets the non - jet fluctuations in the extinction corrected h@xmath11 surface brightness ( figure 7 ) as being due to knots at various distances from @xmath0 ori c , then they lie in a range of about 0.025 pc or 11 , which is comparable to the emissivity scale - height of 15 derived in 7.2 . this means that the shadowed regions occur within the zone where most of the emission occurs . to calculate the contribution to @xmath1(o@xmath70 ) and @xmath1 due to the shadows behind these clumps we need an estimate of the fraction of the emission produced in the shadowed regions . assuming : a ) the typical clump density to be @xmath157 @xmath66 , b ) the average density of the ionized media to be @xmath158 @xmath66 , and c ) that half of the ionized matter originated from the photoevaporation of these clumps , it follows that the volume originally occupied by the clumps must be about 1/1200 of the total volume . however , the shadowed volume is larger than the volume occupied by the clumps ; we know that the shadows are approximately 0.025 pc long , while each clump is approximately 0.0044 pc across , so the shadowed volume can amount to approximately 1/200 of the volume of the region closest to the main ionization front , before these clumps are photoevaporated . since these clumps survive long enough to cast 0.025 pc shadows , and the emissivity scale - height of the orion nebula is of 0.034 pc the total volume of the shadowed regions is about 1/250 of the effective emitting volume . there are three possible physical conditions in the shadowed regions : a ) that they are completely neutral , in this case they will not contribute to @xmath1(o@xmath70 ) nor to @xmath1 ; b ) that they are completely ionized , but not in pressure equilibrium with their surroundings , then their contribution to @xmath1(o@xmath70 ) would be about 0.0038 and to @xmath1 of about 0.0016 ; and c ) that they are completely ionized and in pressure equilibrium with their surroundings , then their contribution to @xmath1(o@xmath70 ) would be about 0.0075 and to @xmath1of about 0.0032 . the contribution of the shadows , including the shadows of the clumps and of the proplyds , amounts to @xmath159 , that corresponds to @xmath160 . note that in the shadowed regions oxygen is either neutral or once ionized , and that therefore the shadowed regions do not contribute to the @xmath1(o@xmath68 ) value it can be shown that the ionized part of the proposed clumps has a very small effect on @xmath1 . the clumps are expected to be close to the pdr , and their ionized part will have a larger density than the interclump medium , so we expect the oxygen in the ionized fraction of the clumps to be singly ionized . additionally the clumps cover a total solid angle of about 0.03 steroradians , and therefore absorb less than 1/400 of the ionizing photons , about 1/800 . this implies that their emissivity amounts to about 1/800 of the total emissivity of the nebula and that the physical conditions in the high density photoevaporated region will produce a negligible effect in the increase of @xmath1(o@xmath70 ) or @xmath1 . there is an unexplained component of line broadening in the orion nebula that may be explained by the clumpy model for the ionized layer that we propose . this anomaly first surfaced with castaeda s ( 1988 ) study of , and remained with the inclusion of other massive ions ( , jones 1992 ; , odell & wen 1992 ; , wen & odell 1993 ) . after quadratic subtraction of the instrumental , thermal , and turbulent components of broadening , there remains a large and unexplained line width having fwhm@xmath6410 km s@xmath145 . the large thermal width of the lines makes it difficult to accurately determine the extra width there , however , wilson , et al.s ( 1997 ) study of the h64@xmath11 line indicates an unexplained broadening component ( called the turbulent velocity in their paper ) of 19.6@xmath390.9 km s@xmath145 . this extra broadening component has been re - addressed using spectra made with the keck telescope for the study of proplyds and their jets ( henney & odell 1999 , bally , odell , & mccaughrean 2000 . the keck data were very useful as they have the highest velocity resolution ( 6 km s@xmath145 ) of any of the optical wavelength studies and allow examination of all of the primary lines of , , , , he i , h@xmath11 , and h@xmath12 at exactly the same positions . samples were taken from 14 long spectra centered on 137 - 349 , 150 - 353 , 163 - 357 , 170 - 337 , 177 - 341 , 182 - 413 , and 244 - 440 ( the designation system is described in odell & wen 1994 ) and 28 long spectra taken at the positions 137 - 349 , 150 - 353 , and 163 - 357 . after correction for instrumental , thermal , and turbulent ( assumed to be 1 km s@xmath145 ) broadening , the remaining line widths ( fwhm in km s@xmath145 ) were h@xmath11 20.9@xmath391.3 , h@xmath12 18.8@xmath392.2 , he i 5876 18.4@xmath392.9 , 6583 10.6@xmath391.4 , 6300 9.0@xmath392.1 , 5007 13.0@xmath393.7 , 4959 11.3@xmath392.1 , 6731 11.3@xmath392.4 , and 6312 11.8@xmath391.9 . the only source of [ o ii ] data are those found in the thesis of michael jones ( 1992 ) who found for 10.5@xmath392.5 for the central 80x 80 . the longer keck slit spectra were subjected to a separate analysis using various fractions of the total slit length and no variation of the results were seen , i.e. there was not a relation between sample size and the residual line width . the radio ( h64@xmath11 , 19.6@xmath390.9 ) and optical ( h@xmath11 , 20.9@xmath391.3 and h@xmath12 , 18.8@xmath392.2 ) are in agreement within their individual uncertainties . the unexplained broadening in the emission lines that arise from the lower temperature component of the nebula is twice that of lines arising from the higher temperature component . we see that the recombination lines of h i and he i that preferentially come from cool regions have a clearly larger unexplained width ( average 19.4 km s@xmath145 ) than the collisionally excited forbidden lines that preferentially come from hot regions ( average 10.9 km s@xmath145 ) . this strongly argues against the broadening arising from alvn waves ( od2001 ) since they would have little dependence upon . the explanation certainly does not lie with there being a systematic difference in , because the sense of the correlation is opposite to that expected . the clearcut differences in the extra broadening in collisional and recombination lines is probably related to the the mechanism giving rise to the temperature inhomogeneities . can the broadening be due to the clumpy structure that we argue can explain the temperature inhomogeneities ? odell & wen ( 1992 ) argue that clumps having sizes of less than 2.3 could produce the extra broadening , but this argument was made before knowledge of the larger broadening of the hydrogen lines . the steady process of destruction of the clumps by photoevaporation would certainly give rise to flows away from the clumps . however , the line broadening due to this should selectively be a blueward distortion of the integrated line - of - sight line profile , except for the final stage of destruction , when material would be flowing in all directions . however , the initial phase of a shadowed region would be accompanied by material flowing inward towards the low pressure zone that would exist when the shadowed drops and the shadowed region is moving towards a state of hydrostatic equilibrium . it is obvious that these situations need to be theoretically modeled . although it is not now possible to quantitatively link them , it is likely that the explanation of the electron temperature fluctuations and the line widths are related . from our @xmath9(o@xmath68 ) map of the orion nebula , that includes 1.5 x 10@xmath146 independent temperature determinations , we have found that @xmath10(o@xmath68)@xmath161 . from our @xmath10(o@xmath68 ) value , together with geometrical considerations and other observations in the literature , we estimate that @xmath1(o@xmath68)@xmath162 . note that the total @xmath1(o@xmath68 ) is larger than @xmath10(o@xmath68 ) because in addition to the variations across the plane of the sky it includes the temperature variations along the line of sight . from our @xmath1(o@xmath68 ) value and comparisons between the temperatures in the low- and high - ionization zones , the o@xmath69 and o@xmath68 zones , we find that @xmath1(h ii)@xmath163 . @xmath1(h ii ) is different to @xmath1(o@xmath68 ) because in addition to the variations in the o@xmath68 zone includes the variations in the o@xmath69 zone and the difference in the average temperature between both zones . our derived @xmath1(h ii ) value is 7 times higher than those obtained from homogeneous one - dimensional photoionization models of the orion nebula . a combination of sources needs to be found to explain the large @xmath95 values observed in the orion nebula . many possible causes of temperature variations have been discussed in the literature like : density variations , deposition of mechanical energy , deposition of magnetic energy , presence of shadowed regions , chemical inhomogeneities , dust heating , and transient effects due to changes in the ionizing flux . from observations of variations in surface brightness , columnar electron temperature , radial velocity , and ionization , we have established the presence of significant small scale structure at the 1 - 5 arc sec level . the small scale velocity and ionization structures are probably strongly linked with the temperature structure . the presence of this small scale structure is compatible with a new model for the structure of the nebula near the ionization front . we posit that this front is advancing into the highly clumpy photodissociation region revealed by radio and infrared observations . these clumps possess ionized shadows that are of low . a quasi - static equilibrium exists in which any one clump survives the process of photoevaporation only for a short time . the process may be common to most regions which have a portion that is ionization bounded . there is an unexplained component of line broadening in the orion nebula that may be explained by the clumpy model for the ionized layer that we propose . we have discussed the presence of shadowed regions due to proplyds and to high density clumps near the pdr . we estimate that the shadowed regions can contribute @xmath160 to the total @xmath95 value , which together with the homogeneous one dimensional ionization structure provide only from about one third to about half of the observed @xmath95 value . additional sources of temperature variations are needed to explain the total @xmath95 observed value . moreover , since in the shadowed regions o is only once ionized , these regions could be responsible for temperature variations in the o@xmath70 zone , but not in the o@xmath68 zone , and since the @xmath1(o@xmath68 ) value is very large we need one or several additional sources of temperature variations to explain it . there is an anomalous broadening mechanism operating in the orion nebula which is unexplained . it broadens h and he recombination line emission by about 19 km s@xmath145and the forbidden line emission from heavy ions by about 11 km s@xmath145 . since these two types of lines selectively rise from the low and high temperature components along a line of sight , it appears likely that this process is related to the @xmath1 problem . michael richer and will henney provided valuable support in obtaining the high velocity resolution spectra at the mexican national observatory at san pedro martir . will henney also provided several useful comments on an earlier version of this paper . takao doi kindly shared before publication the results of his orion nebula radial velocity program , which allowed us to generate figure 8 and figure 9 . henney also provided observations on the expected rate of motion of the ionization front in the orion nebula . gary j. ferland provided estimates for the heating rates . christopher a. coco measured the keck spectra line widths as part of his senior thesis at rice university . the authors are grateful to patrick hartigan of rice university and robert rubin of the ames research center for discussions on some aspects of this work . cro s work was supported in part by grant go-9141 from the stsci to vanderbilt university . mp s work was supported in part by grant in 114601 from dgapa unam . baldwin , j. a. , ferland , g. j. , martin , p. g. , corbin , m. r. , cota , s. a. , peterson , b. m. , & slettebak , a. 1991 , apj , 374 , 580 ( b91 ) bally , j. , odell , c. r. , & mccaughrean , m. j. 2000 , aj , 119 , 2919 brand , p. w. j. l. 1981 , mnras , 197 , 217 buisson , h. , fabry , c. , & bourget , h. 1914 , apj , 40 , 241 cant , j. , goudis , c. , johnson , p. g. , & meaburn , j. 1980 , a&a , 85 , 128 cant , j. , raga , a. , steffen , w. , & shapiro , p. r. 1998 , apj , 502 , 695 ( c98 ) capriotti , e. r. 1973 , apj , 179 , 495 castaeda , h. o. 1988 , apjs , 67 , 93 esteban , c. 2002 , rev . mexicana astron . , 12 , 56 esteban , c. , peimbert , m. , torres - peimbert , s. , & escalante , v. 1998 , mnras , 295 , 401 ( epte98 ) ferland , g. j. 2001 , pasp , 113 , 41 garca - segura , g. , & franco , j. 1996 , apj , 469 , 171 garnett , d.r . 1992 , aj , 103 , 1330 . gruenwald , r. , & viegas , s.m . 1992 , apjs , 78 , 153 hartigan , p. 1999 , apj , 526 , 274 henney , w. j. 1998 , apj , 503 , 760 henney , w. j. , & odell , c. r. 1999 , aj , 118 , 2350 henney , w. j. , odell , c. r. , meaburn , j. , garrington , s. t. , & lpez , j. a. 2002 , apj , 566 , 315 hogerheijde , m. r. , jansen , d. j. , & van dishoeck , e. f. 1995 , a&a , 294 , 792 holtzman , j. a. , burrows , c. j. , castertano , s. , hester , j. j. , trauger , j. t. , watson , a. m. , & worthey , g. 1995 , pasp , 107 , 1065 jones , m. r. 1992 , phd thesis , rice univ . , houston , tx kirkpatrick , r. c. 1972 , apj , 176 , 381 liu , x .- w . 2002a , rev . mexicana astron . , 12 , 70 meaburn . j. , blundell , b. , carling , r. , gregory , d. f. , keir , d. , & wynne , c. g. 1984 , mnras , 210 , 463 odell , c. r. 1998a , aj , 115 , 263 odell , c. r. 1998b , aj , 116 , 1346 odell , c. r. 2000 , aj , 119 , 2311 odell , c. r. 2001a , pasp , 113 , 29 odell , c. r. 2001b , araa , 39 , 99 ( od2001 ) odell , c. r. 2002 , rev . mexicana astron . , 12 , 12 odell , c. r. , & doi , t. 1999 , pasp , 111 , 1316 ( od99 ) odell , c. r. , & doi , t. 2003 , aj , 125 , 277 odell , c. r. , hartigan , p. , lane , w. m. , wong , s. k. , burton , m. g. , raymond , j. , & axon , d. j. 1997 , aj , 114 , 730 ( o97 ) odell , c. r. , hartigan , p. , bally , j. , & morse , j. a. 1997 , aj , 114 , 2016 odell , c. r. , & hubbard , w. b. 1965 apj , 142 , 591 odell , c. r. , walter , d. k. , & dufour , r. j. 1992 , apj , 399 , l67 odell , c. r. , & wen , z. 1992 , apj , 387 , 229 odell , c. r. , & wen , z. 1994 , apj , 436 , 194 odell , c. r. , & yusef - zadeh , f. 2000 , aj , 120 , 382 osterbrock , d. e. , tran , h. d. , veilleux , s. 1992 , apj , 389 , 305 peimbert , a. 2003 , apj , 584 , 735 peimbert , a. , peimbert , m. , & luridiana , v. 2002 , apj , 565 , 668 peimbert , m. 1967 , apj , 150 , 825 peimbert , m. 1995 , in the analysis of emission lines , eds . r. e. williams , & m. livio ( cambridge ) , 165 peimbert , m. 2002 , rev . mexicana astron . , 12 , 275 peimbert , m. , peimbert , a. , & ruiz , m. t. 2000 , apj , 541 , 688 peimbert , m. , storey , p. j. , & torres - peimbert , s. 1993 , apj , 414 , 626 . peimbert , m. , & torres - peimbert , s. 1977 , mnras , 179 , 217 rubin , r. h. , bhatt , n. j. , dufour , r. j. , buckalew , b. a. , barlow , m. j. , liu , x .- w . , storey , p. j. , balick , b. , ferland , g. j. , harrington , j. p. , & martin , p. g. 2002 , mnras , 334 , 777 ( r2002 ) rubin , r. h. , martin , p. g. , dufour , r. j. , ferland , g. j. , blagrave , k. p. m. , liu , x .- w . , nguyen , j. f. , & baldwin , j. a. 2003 , mnras , in press , ( astro - ph/0212244 ) , ( r2003 ) stasiska , g. 2002 , rev . mexicana astron . , 12 , 62 tauber , j. a. , goldsmith , p. f. , & dickman , r. l. 1991 , apj , 375 , 635 torres - peimbert , s. , & peimbert , m. 2003 , in planetary nebulae and their role in the universe , iau symposium 209 , in press , ( astro - ph/0204087 ) van blerkom , d. , & arny , t. t. 1972 , mnras , 156 , 91 van der werf , p. p. , stutzki , j. , sternberg , a. , & krabbe , a. 1996 , a & a , 313 , 633 vannier , l. , lemaire , j. l. , field , d. , pineau des forts , g. pijpers , f. p. , & rouan , d. 2001 , a&a , 366 , 651 vishniac , e. t. 1994 , apj , 428 , 186 von hrner , s. 1951 , zap , 30 , 17 walter , d. k. , odell , c. r. , hu , x. , & dufour , r. j. 1995 , pasp , 107 , 686 wen , z. , & odell , c. r. 1993 , apj , 409 , 262 wen , z. , & odell , c. r. 1995 , apj , 438 , 784 ( wo95 ) williams , r. j. r. 1999 , mnras , 310 , 789 wilson , o. c. , mnch , g. , flather , e. m. , & coffeen , m. f. 1959 , apjs , 4 , 199 wilson , t. l. , filges , l , codella , c. , reich , w. , & reich , p. 1997 , a&a , 327 , 1177 young owl , r. c. , meixner , m. m. , wolfire , m. , tielens , a. g. g. m. , & tauber , j. 2000 , 540 , 886 lccccccl e@xmath17 & 16.56 & 17.40 & 19.87 & 22.79 & 601.9 & 22.0 4 & 31.19 + @xmath164 & 0.50 & 0.52 & 0.59 & 0.64 & & 0.78 & 0.79 + @xmath165 & 0.005 & & & & & 0.24 & 0.045 + s & 0.061 & 0.074 & 0.084 & 0.091 & 0.125 & 0.145 & 0.145 + Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: it has been recognized for many years that observations of clusters of galaxies can be applied effectively to test theories of the evolution of the universe . as the largest gravitationally bound objects emerging from the growth and subsequent collapse of primordial density fluctuations , their spatial distribution at different epochs is a relic containing evidence which can test cosmological models of structure formation . x - ray astronomy has enhanced the results of such studies greatly , for a number of reasons . first , clusters are x - ray sources of relatively high luminosity , with a luminosity function sufficiently well defined to allow , in principle at least , cosmic structure to be studied out to epochs of @xmath6 and beyond . further , ambiguities in the identification of clusters , caused by chance line - of - sight superposition of objects , are fewer in x - ray than in optical sky surveys . this is a consequence of the sharply peaked x - ray brightness distribution in most clusters , and of the relatively low surface density of bright x - ray sources . second , the x - ray properties of clusters are defined by events reaching back to the earliest epoch of cluster formation , and therefore may be used to test cosmological models . the x - ray luminosity and temperature of the intracluster gas depend on the cluster mass , so that the shape and amplitude of the luminosity and temperature functions of the cluster population are dependent on the spectrum of the early density perturbations . further , the morphology of clusters in the whole population may be dependent on the characteristics of the expansion , and is a diagnostic of cluster merging processes . finally , the abundance of heavy elements in the intracluster gas is a measure of galaxy evolution and ram - pressure stripping , and of the proportion of primordial gas in the cluster . therefore much effort has been given to the production of statistically complete samples of x - ray emitting clusters . the first were produced using @xmath7 ( schwartz 1978 ) and @xmath8 ( mchardy 1978 ) observations , and the heao-1 all - sky surveys ( piccinotti et al . 1982 ; kowalski et al . 1984 ) , and contained up to 76 clusters . the flux limits of these surveys ( @xmath9 , expressed here in the rosat band @xmath1 ) , and their angular resolution were limited by the use of proportional counters , and improvements were possible only after the first missions using imaging x - ray instruments . edge et al . ( 1990 ) reported the use of einstein and exosat observations to refine the results of the earlier surveys , producing a sample of 55 clusters free of , for example , the effects of source confusion . in the einstein extended medium sensitivity survey ( emss ) it was possible , by examining the fields of einstein pointings covering a total of 740 @xmath10 , to create a sample with a flux limit of @xmath11 and containing clusters with redshifts out to @xmath12 ( henry et al . 1992 ) . the rosat x - ray telescope , equipped with an imaging proportional counter in the focal plane , allowed for the first time complete detailed surveys of large areas of sky ( trmper 1983 ) . the analysis of the rosat all - sky survey ( rass ) proceeded in two phases rass-1 and rass-2 , described below in section 3.1 , in which the second phase incorporated a number of improvements suggested by rass-1 ( voges et al . . the first x - ray flux - limited samples of clusters based on analysis of a large area of sky in the rosat survey were obtained by romer et al.(1994 ) , romer ( 1995 ) , ebeling et al.(1996 ) and ebeling et al . ( 1998 ) . the sgp sample described in romer et al.(1994 ) , based on analysis of the rass-1 data base , contained 161 cluster candidates with a limiting flux of @xmath13 . redshifts were obtained for 128 clusters in this sample , and this subset was used in a study of the cluster spatial correlation function ( romer et al . 1994 ) . the bcs survey by ebeling et al . ( 1998 ) addressed the whole extragalactic sky ( @xmath14 ) in the northern hemisphere , and reanalysed the fields in which clusters had been detected in the rass-1 analysis . the survey yielded a sample of 201 clusters at @xmath15 , having a limiting flux of @xmath16 , which is statistically complete at the 90 percent level . in a recent publication the bcs sample has been extended to an x - ray flux limit of @xmath17 , at which the completeness was 75 percent ( ebeling et al . 2000 ) . a recent survey of clusters in the northern sky ( noras ) , based on x - ray sources classified as extended in the rass analysis , is described by bhringer et al . deeper surveys have been possible in areas around the ecliptic poles ( henry et al . 2001 , gioia et al . 2001 , mullis et al . 2001 and voges et al . 2001 ) , where the rosat survey achieved unusually long exposure times , and in areas where deep pointings have been made ( romer et al . 2000 , jones et al . 1998 , rosati et al . 1998 , vikhlinen et al . 1998 and collins et al . the deepest survey was made by rosati et al . ( 1998 ) , in which the sample was complete down to a flux of @xmath18 and included clusters out a redshift of @xmath19 . in what follows we concentrate on complete large - area cluster surveys in the southern hemisphere . this has its roots in an agreement between the royal observatory in edinburgh ( roe ) , the naval research laboratory ( nrl ) and the max - planck - institut fr extraterrestrische physik ( mpe ) in garching , made early in the rosat program , to make the cosmos digitised uk schmidt iiia - j survey of the southern sky ( macgillivray & stobie 1985 , yentis et al . 1992 ) available for identification of x - ray sources . this data base , containing @xmath20 objects with a limiting magnitude of @xmath21 , was installed at mpe for use in x - ray source identification . in cluster studies it was used to derive unbiased lists of candidates through systematic correlations of x - ray source positions with concentrations of galaxies . a comprehensive search for clusters is being undertaken at mpe in the reflex southern hemisphere project ( bhringer et al . 2001 ) . the first results ( de grandi et al . 1999 ) described a sample of x - ray bright clusters , obtained by searching a region of area @xmath22 ( @xmath23 ) . the study reanalysed clusters identified in the rass-1 analysis and arrived at a sample of 130 clusters having a completeness of at least 90 percent and a limiting flux of @xmath24 in the @xmath1 band . the complete reflex study area covers @xmath25 in the southern sky , and the techniques used in constructing the cluster sample have been described by bhringer et al . ( 2001 ) . at the current stage of the study a sample of 452 clusters has been obtained , which is at least 90 percent complete , using a flux limit of @xmath26 . a power - law fit to the log n - log s distribution yielded an exponent of @xmath27 . collins et al . ( 2000 ) have analysed the spatial correlation function of clusters using a reflex sample of 452 clusters covering an area 4 times larger than the sgp , and derived a value of @xmath28 mpc for the correlation length , significantly greater than the value of @xmath29 mpc obtained by romer et al . however for the 109 reflex clusters in the sgp area collins et al . ( 2000 ) find a correlation length of @xmath30 mpc , consistent with the result of romer et al . the difference between the complete reflex and the sgp results is most likely due to sample fluctuations caused by cosmic variance ( see collins et al . a significant conclusion of the reflex study is that low - density cold dark matter ( cdm ) models ( @xmath31 ) provide a better fit to the observed correlation function , and that high - density cdm models ( @xmath32 ) yield a poorer fit because they provide insufficient power at large scales . schuecker et al . ( 2001 ) have analysed the power spectrum of the reflex sample , and the results are consistent with the results of collins et al . ( 2000 ) . a prominent aspect of the sgp and reflex survey projects has been optical observing programs to obtain cluster redshifts , which we describe briefly in section 2 . however , it should be borne in mind that while these two projects have a common root and have followed similar procedures for cluster selection , x - ray analysis and redshift determination , they do possess some differences in approach . following this introduction section 2 provides a summary of how the sgp survey was performed . section 3 describes the various procedures by which the cluster sample was extracted from the rass-2 data base and section 4 summarises how the x - ray energy flux and luminosity were calculated . section 5 presents the sgp cluster catalog in the form of a table and describes the archive of sgp cluster images in the form of overlays of the x - ray and optical data , which is accessible at mpe over the internet . section 6 examines the completeness of the cluster sample and section 7 discusses the large - scale structure traced by the clusters out to a redshift of @xmath33 . finally a summary of the results and conclusions of the paper are presented in section 8 . throughout the article we have assumed that the hubble constant @xmath34 , the deceleration parameter @xmath35 and the cosmological constant @xmath36 . this survey selected a region at high galactic latitude in the southern sky centered on the south galactic pole . rectangular in shape , it extends in right ascension from @xmath37 to @xmath38 and in declination from @xmath39 to @xmath40 ( see figure [ f18 ] ) , giving it an area of 1.013 steradian . the search for clusters in this region invoked no flux threshold , but intead worked to the limit of the rass data to maximise the completeness of the sample of cluster candidates . a flux limit is established subsequently to define a complete subsample . the first step in the search for clusters was to select all x - ray sources found in the region by the rass-2 analysis ( voges et al . 1999 ) . in this analysis the threshold values for the source detection algorithmns were lowered ( section 3.1 ) , with the aim of achieving as deep a survey as the rass would allow . therefore cluster searches using this data base have more sensitivity than has been available to previous southern hemisphere searches using the rass-1 data base ( romer 1995 , de grandi et al . 1999 ) . the rass-2 analysis was not equipped for detailed examination of the x - ray properties of extended sources such as clusters . to this end a number of techniques have been developed , including voronoi tesselation and percolation ( vtp , ebeling & wiedemann 1993 ) , the steepness ratio technique ( srt , degrandi et al . 1997 ) and the growth curve analysis ( gca , bhringer et al . the goal of these techniques is to obtain reliable fluxes for both extended and point sources , and to extract from the raw data as much information as possible about the x - ray source characteristics . in this study the gca technique has been used to analyse the list of candidates obtained from the rass-2 data base . cluster candidates were selected from this sample using the cosmos digitised iiia - j survey and a search procedure called csearch . this procedure , which is not designed to select clusters in an unambiguous manner , compares the galaxy density around the rass-2 position with the background galaxy density on the uk schmidt plate , and derives a probability that the result was a random coincidence . a sample of cluster candidates is obtained by setting a threshold to this probability , and this threshold is set low enough to ensure that few genuine clusters are lost and a high completeness is attained in the final sample . in our study a rass-2 list of 11981 sources is reduced by csearch to 3236 candidates ( section 3.3 ) . at this point much remains to be done to remove contaminating sources , which we estimate from internal statistical tests to comprise about half the candidates , in an objective process . at high redshifts ( @xmath41 ) the method reaches a limit of the cosmos data , when many cluster galaxies become so faint as to be classified by cosmos only as faint objects , which are not counteed by csearch . therefore some clusters at high redshift may be missed , which may be one source of incompleteness near the flux limit of the survey . the analysis of this sample of cluster candidates began by setting a limit of @xmath42 to the x - ray count rate in the hard band ( @xmath43 , section 3.1 , 3.4 ) , which reduced the number of candidates from 3236 to 477 . this was followed by the process of removing contaminating sources from this sub - sample , which proceeded through a number of stages . the first was the removal of cosmos artifacts , bright sources which to the eye were clearly not clusters , and redundant detections by rass-2 of the same source , particularly in extended sources such as clusters . this was done by examination of cosmos optical finding charts and of overlays of the optical and x - ray images . the second was to correlate rass-2 positions with the simbad catalog , in order to eliminate bright stars found in the rass-2 error circle , and then to use two x - ray properties of the source , hardness ratio and extent , as selection criteria . the hardness ratio was used to exclude those stars and active galactic nuclei ( agn ) whose spectra are sufficiently soft to exclude the possibility that the source is a cluster . the x - ray extent was used to classify candidates in three groups , namely clusters , agn candidates and a remainder of uncertain nature . the third and final stage , identifying the clusters and known agn among the remaining sample , proceeded along two parallel paths . the first comprised correlations of the candidate list with a variety of catalogs , in particular the nasa extragalactic data base ( ned ) , the simbad catalog , and the veron catalog of agn ( veron - cetty & veron 1998 ) . the second was an ongoing examination of the results of the supporting optical observations , described below , which identified agn in the sample . we summarise here briefly the sources of incompleteness in the final catalog obtained , which reduce the cluster count at low fluxes : + i ) the csearch analysis has a predetermined level of incompleteness in correlating + x - ray sources with galaxy concentrations ( section 3.3 ) . + ii ) the sharp count - rate limit ( @xmath42 ) rejects some sources at low energy + flux , due to the variation of interstellar column density in the field ( section 3.4 ) . + iii ) some distant clusters elude the tests applied in searching the list of candidates + yielded by csearch and the x - ray count - rate cut ( section 3.5 ) . + finally , a vital complementary part of the sgp survey and the reflex project has been a program of optical observations to obtain the redshifts of those clusters detected in the rosat survey , for which no redshifts were available in the literature . the sgp survey , which commenced not long after the launch of rosat in 1990 , was supported by three campaigns at the 4 m anglo - australian telescope ( aat ) and three at the 1.9 m telescope of the south african astronomical observatory . the results of these campaigns have been reported by romer ( 1995 ) . a larger program , which started in 1992 and was completed in 1995 , was installed as a key project at the european southern observatory ( eso ) . its purpose was to obtain redshifts for clusters in the reflex sample . an ongoing continuation of these observations is being pursued to support the extension of the reflex survey to lower x - ray flux limits . the rosat observatory was launched on 1990 june 1 , and saw first light on 1990 june 16 ( trmper et al . 1991 ) . after a two - month verification of the instrument operation it commenced its six - month long survey of the whole sky on 30 july 1990 . during the rass the observatory rotated about an axis through the sun , scanning on a given day a @xmath44 wide strip passing through the ecliptic poles the survey ended on 1991 february 18 . the first analysis of the survey ( rass-1 ) , which took place in 1991 - 1993 , used data collected into the same @xmath45 wide strips , and yielded a list of some 50,000 sources containing positions and such characteristics as count - rate , hardness ratio , angular extent and source detection likelihood . rass-1 was used to test the algorithms implemented in the analysis and to start scientific studies , and as a consequence a series of improvements was initiated ( voges et al . 1999 ) : + i ) sorting the data into 1378 sky fields of size @xmath46 , overlapping at least @xmath47 . + this improved the efficiency of source detection . + ii ) for operational reasons the rass-1 data base contained a few regions of low + exposure , which were filled in later by supplementary observations . this + later data was included in the rass-2 data base . + iii ) searching for sources in the soft @xmath48 , hard @xmath49 and broad + @xmath50 energy bands independently . + iv ) refining the pointing attitude solution and improving source positions . + v ) improvement of the spline - fit to the background , yielding more accurate source + count - rates . + vi ) the list of candidates selected by the sliding - window source detection algorithmns + was increased by lowering the detection threshold , and in the subsequent maximum + likelihood analysis the likelihood threshold for accepting sources was lowered + from 10 ( rass-1 analysis ) to 7 . + reanalysis of the survey data using the new software ( rass-2 ) was performed in 1994 - 1995 , yielding a list of 145060 sources . the sgp field contained 11981 sources above a likelihood threshold of 7 . the characteristics of the new software and results from the rass-2 analysis have been described by voges et al . the results described here relied solely on the rass-2 data base ( rass bright source catalog in voges et al . 1999 and rass faint source catalog in voges et al . 2000 ) . the @xmath45 wide rass scan strips overlap to an extent which increases with ecliptic latitude , and consequently the exposure time grows and reaches a maximum at the poles . this effect is visible in the map of the rass exposure time in the sgp survey region ( figure [ f1 ] ) , in which there is a systematic increase from the nw to the se corner . upon this change are superposed striations caused by periodic protective shutdown of the focal - plane detector as it traversed radiation belts . analysis of the exposure map of the sgp survey region yields the histogram shown in figure [ f2 ] , in which the mean exposure time is 300 seconds . there is an area of @xmath51 in the sgp field , in which the exposure time is less than @xmath52 . the roe cosmos digital data base contains the results of scanning 894 plates of the uk schmidt iiia - j ( blue ) southern hemisphere survey . the sgp study area contains 161 fields , each of size @xmath53 . cosmos analyses and classifies each object in a field , and records the @xmath54 magnitude , position ( r.a . and declination ( 2000 ) ) and object shape parameters . objects are classified primarily as stars , galaxies or faint objects , the last being objects whose low intensity on the plate defies a reliable determination of their class . heydon - dumbleton , collins , & macgillivray ( 1989 ) compared the cosmos galaxy number count - magnitude distribution with those derived from several independent optical surveys , and deduced that the galaxy sample was essentially complete ( @xmath55 percent completeness level ) to a magnitude limit @xmath56 . beyond this limit the star - galaxy discrimination gradually becomes less reliable as @xmath54 increases ( macgillivray & stobie 1985 , heydon - dumbleton et al . 1989 ) , leading to a significant incompleteness of the galaxy sample in the range @xmath57 . in the cosmos data base star / galaxy separation was performed down to a limit of @xmath58 . an example of the results is given in a figure , which shows the @xmath59 galaxies identified by cosmos in the sgp field ukj 411 ( this figure is in the published article , but was too large for this archived version ) . the galaxy magnitudes have been calibrated using photometric ccd observations of sequences of galaxies , which were used also in the calibration of the edinburgh / durham galaxy catalog ( heydon - dumbleton et al . 1989 ) . for any given plate cosmos yields a galaxy intensity , which depends on such factors as exposure time , emulsion characteristics and measures taken to keep the plate in a dry environment during exposure . this implies that the galaxy magnitudes for any given plate must be adjusted to match the surrounding plates . this is done by identifying galaxies in the overlapping border zone of adjacent plates , and deriving a shift which matches the galaxy magnitudes . no unique solution for the whole southern sky is possible , and instead a set of plate magnitude shifts is obtained by a fitting procedure which globally minimises the inconsistencies between adjacent plates . the resulting cosmos galaxy magnitudes are on an arbitrary machine scale , which now must be adjusted to match the ccd calibrations , in essence providing a global zero point solution to the matched plate data . this is done by fitting a correlation between the ccd and cosmos magnitudes of galaxies in the ccd observation sequences . to further ensure there were no calibration problems as a result of the iosophotal thresholding of the cosmos galaxies ( isophotal thresholding results in loss of light below the isophote level used and hence underestimation of the true brightness of a galaxy ) , we accepted for calibration purposes only galaxies in the cosmos data which were at least three magnitudes above the plate limit ( i.e. @xmath60 ) . this resulted in 207 sequence galaxies being accepted for calibration . the rms scatter of the correlation between the cosmos and the ccd magnitudes for these galaxies was found to be @xmath61 . as the intrinsic error in the ccd photometry is estimated to be @xmath62 , we estimate that the resulting error in the magnitudes derived for galaxies in the cosmos iiia - j southern hemisphere survey is @xmath63 . some independent corroboration of the cosmos galaxy photometry is provided by the study of caretta , maia , & willmer ( 2000 ) . they confirm after comparisons with their ccd data that the cosmos galaxy magnitudes are reliable at the @xmath63 level . their results indicate that galaxy samples are @xmath64 percent complete at magnitudes down to @xmath65 , in broad agreement with the cosmos estimates . however this agreement should be treated with caution , as caretta et al . ( 2000 ) studied a small area of sky having a lower galaxy count than is available in the uk schmidt southern hemisphere survey . the cluster search procedure ( csearch ) counts galaxies within 5 circles of radius @xmath66 = 1.5 , 3 , 5 , 7.5 and 10 arcmin ( circles 1 through 5 ) , centered at the rosat source position , and compares the results with those of random sampling in the ukj field . these radii correspond to the range of expected cluster core diameters ( e.g. at @xmath67 a radius of 300 kpc is equivalent to an angular scale of 1.1 arcmin and at @xmath68 to 11 arcmin ) . the random sampling was performed 1000 times for each radius , yielding for each field five histograms of the galaxy count . for each histogram a table is created , giving the probability @xmath69 that the galaxy count exceeds a value , @xmath70 : + @xmath71 where @xmath72 is the number of cases where the galaxy count was @xmath73 . selection of cluster candidates is made after setting a threshold @xmath74 , which implies that for each plate a threshold be set for @xmath70 . individual plate sensitivity and statistics make this galaxy count threshold vary from plate to plate and yield a significant uncertainty . the following argument describes how this is circumvented by analysing the ensemble of probability values @xmath69 for the whole sgp rass-2 source sample . the rosat source is accepted if the probability associated with one of the 5 galaxy counts ( circles 1 through 5 ) at the rosat position falls below a selected threshold @xmath75 . for convenience , we have defined and used the quantity @xmath76 , as this approaches a maximum value of 1 for cluster candidates . the range @xmath77 is divided into bins of equal width ( typically 400 in number ) , and for any given search radius @xmath66 we create a histogram of the source count ( @xmath78 ) per bin for all the rass-2 sources in the sgp field . if there were no correlation between the rosat position and the galaxy count , as would be close to the truth for an ensemble of stars and agn , the expectation value for the number of sources , @xmath78 , per bin would be independent of @xmath79 ( bhringer et al . figure [ f4 ] shows the histograms of the probability p obtained for the rosat sources detected in the sgp region . to save space we show results for @xmath66 = 1.5 , 3 , 5 and 10 arcmin only . the major characteristic of these histograms is indeed a flat distribution with statistical fluctuations , but in addition there is an enhancement at values of @xmath79 approaching @xmath80 , which is the signature of clusters . a comment on the data used to create these histograms is necessary . following the first csearch analysis of the sgp sources in the rass-2 data base , examination of images overlaying the x - ray and the optical fields revealed multiple detections by rass-2 of the more nearby clusters , due to their markedly extended x - ray emission ( e.g. a119 ) . this biases the csearch analysis , and therefore we removed 256 redundant multiple detections from the rass-2 sgp source list , and then repeated the analysis . this number is small compared to the total of 11981 sources , but is more significant in relation to the size of the eventual csearch sample of cluster candidates ( 3236 ) . the threshold @xmath81 must be chosen carefully in terms of the desired compromise between formal completeness @xmath82 and contamination @xmath83 , for inevitably the selected sample contains rosat sources having chance associations with concentrations of galaxies , which build the flat distribution , and so the price of higher completeness is an increased contamination . the probability threshold @xmath81 is related to @xmath82 and @xmath83 by the following equations : + @xmath84 @xmath85 where @xmath86 is the number of sources per bin in a specified flat portion of the probability histogram . before the results of csearch are discussed , two characteristics of the histograms shown in figure [ f4 ] should be clarified . first , the dip in @xmath79 for circle 1 in the range @xmath87 is a quantisation effect , due to the galaxy count jumping from 0 to 1 . the dip for circle 5 in the same region appears to be due to sampling limits , resulting from the circle diameter being a significant fraction of the field width . we have examined the number density of clusters found in our analyses as a function of position on the plate , and have found no sign of incompleteness near the cosmos digitisation boundaries on the plate . second , whereas the histograms for circles 1 and 2 decay rapidly from the peak at @xmath88 to a flat plateau , those for circles 3 and 5 decay even more rapidly to a minimum and then rise to a second enhancement in @xmath89 , covering the range @xmath90 for circle 3 and @xmath91 for circle 5 . to search for the cause of this second enhancement we examined how candidates found using circles 1 and 2 were registered using circles 3 through 5 . a typical result ( figure [ f5 ] ) shows how compact groupings of galaxies found by circle 2 , for which @xmath92 , are detected using circle 5 . the number of sources falls sharply from @xmath93 to a minimum , rises to a peak at @xmath94 and then falls again gradually , i.e. much of this second enhancement is due to relatively compact groups , for which the number of galaxies counted is not significantly enhanced when detected with circles 3 and 5 . as a consequence the effect of increasing the circle radius is to reduce @xmath79 . the five candidate lists ( @xmath66 = 1.5 , 3 , 5 , 7.5 and 10 arcmin ) are cross - correlated to remove redundant detections , and so this second enhancement is in principle not a problem . however , if for each circle @xmath81 is set so as to yield a high value of @xmath95 ( equation 2 ) , then for circles 3 through 5 the contamination @xmath83 is increased substantially , making the subsequent screening of the sample more difficult and time - consuming . therefore as a precaution , for circles * 4 and 5 only * we have set @xmath81 at @xmath96 , so as to include only the sharp peak . we show below that this is a reliable procedure . the sgp field contains 11981 rass-2 sources having a likelihood greater than or equal to @xmath97 . table 1 shows the results of analysing this sample with csearch , and summarises the assumptions made in selecting cluster candidates . the last entry @xmath98 gives the yield of new candidates as one starts with circle 1 , and then proceeds from circles 2 to 5 , * removing redundant detections * at each stage . the trends in these numbers confirm the expectation that the smaller circles should detect the majority of clusters . the decision to set @xmath81 to @xmath96 for circles 4 and 5 was supported later by a search of the final list of 186 cluster candidates , which revealed no case in which the csearch detection relied solely on circle 4 and/or circle 5 . the total sample extracted using csearch comprised 3236 candidates . table 1 indicates that the majority of clusters are detected using circles 1 , 2 and 3 , and we deduce from column 6 that the contamination of the sample should lie between 46 and 56 percent . + ccccccc 1 & 1.5 & 0.92 & @xmath99 & 0.925 & 0.46 & 1857 + 2 & 3 & 0.90 & @xmath99 & 0.900 & 0.50 & 884 + 3 & 5 & 0.90 & @xmath100 & 0.700 & 0.56 & 428 + 4 & 7.5 & 0.95 & @xmath101 & - & @xmath102 & 33 + 5 & 10 & 0.95 & @xmath101 & - & @xmath102 & 34 + & & & & & & + previous studies found that the energy flux from extended extended sources is underestimated by the standard rass source detection algorithm ( ebeling et al . 1996 , 1998 and de grandi et al . 1997 ) , and section 2 has stated the various analyses developed to correct this problem . the growth curve analysis ( gca ) method , described in detail by bhringer et al . ( 2000 ) , has been used in this study to analyse the rass-2 data base . the gca returns among other information the following important parameters : background subtracted source count - rate and its poisson ( photon statistical ) error , a locally redetermined position of the centroid of the source x - ray emission , the mean exposure for the source region , a significance of the source detection , the radius out to which source emission is significantly detected , a hardness ratio characterizing the source spectrum and its photon statistical error , a fitted source core radius , and the probability , obtained using a kolmogorov - smirnov test , that the object is consistent with being a point source . the source position is determined with an error which in most cases is less than @xmath103 , the exceptions being nearby clusters with irregular emission structure . errors in count - rate and hardness ratio are given in our catalog ( table 3 ) . taking a2726 as an example , the hard band count - rate is @xmath104 and the hardness ratio is @xmath105 . the basic parameters are derived from the photon distribution using the three energy bands ` hard ' ( 0.5 to 2.0 kev , channel 52 - 201 ) , ` broad ' ( 0.1 to 2.4 kev , channel 11 - 240 ) , and ` soft ' ( 0.1 to 0.4 kev , channel 11 - 40 ) . the band definitions are the same as those used in the standard analysis ( voges et al . most of the derived parameters are based on analyses of data in the hard energy band , as clusters are detected in this energy band with the highest signal - to - noise ratio . one of the exceptions is the hardness ratio , which requires also the results from the soft band . the gca method uses the photon event distributions in two circles ( figure [ f6 ] ) , of radii 20 and 41.3 arcmin , centered initially on the rass-2 source center . events within the inner circle are analysed to determine the source characteristics , and those in the outer annulus are used to determine the background level . in some cases special analyses were necessary , where either clusters had an unusually large angular extent , or where two or more sources lay in a confused region , requiring the use of deblending algorithms . the source count - rate is determined from the cumulative , radial source count rate profile ( growth curve ) after background subtraction . the construction of the growth curve is preceded by a redetermination of the source center and by the background measurement . a typical example for the resulting growth curve is displayed in figure [ f7 ] , showing the count - rate plotted as a function of integration radius ( solid curve ) . the dashed lines show the limits determined by applying the photon event statistical error ( including the error for the background subtraction ) . the count - rate is determined in two alternative ways . in the first a significant outer radius of the source is defined as the point where the increase in the @xmath106 error is larger than the increase of the source signal . the integrated count - rate is calculated using events within this cutoff radius . in the second method a horizontal level is fitted to the outer region of the plateau , and this value is adopted as the standard in quoting hard - band x - ray count - rates and in calculating cluster energy flux and luminosity ( section 4 ) . however , we use the first method also as a check on the count - rate and as a means of estimating systematic uncertainties in the count - rate determination , which must be combined with the pure photon statistical errors . careful examination of the overlays of x - ray and optical images ( section 3.5.1 ) of the cluster sample revealed 19 cases in which the source was very extended , or source confusion might cause errors in deriving the source count - rate . these 19 were subjected to the special analyses referred to above , in which the source count - rates were revised . six systems were identified as double clusters and not subjected to the deblending analysis . they are identified as double clusters in the survey catalog ( table 3 ) . the other two important source properties determined by the gca method are the spectral hardness ratio and the source extent . the hardness ratio , @xmath107 , is defined as @xmath108 where @xmath109 is the hard band and @xmath110 the soft band source count - rate ( both determined for the same outer radius limit ) . the source extent is investigated in two ways . in the first a @xmath111-model profile ( cavaliere & fusco - femiano 1976 ) convolved with the averaged survey psf is fitted to the differential count - rate profile ( using a fixed value of @xmath112 ) , and we derive from the results a core radius , @xmath113 . in the second a kolmogorov - smirnov ( ks ) test is used to estimate the probability that the object is a point source . it is taken as highly likely that the source is extended when the ks probability is less than 0.01 . tests with x - ray sources which have been identified with stars and agn showed that a false classification as an extended source occurred in less than 5% of the sample . the final step in the x - ray analysis was to establish a lower limit to the source count - rate , in order to set practical limits to source identification and redshift determination . the list of 3236 candidates from the csearch analysis ( section 3.3 ) was truncated using a threshold of @xmath42 for the cut - off radius count - rate in the hard band . this reduced the list of candidates to 477 . a sharp limit in count - rate corresponds to a range of energy flux , determined by the variation of interstellar column density @xmath114 in the field , which leads to some incompleteness at low fluxes . bhringer et al . ( 2001 ) estimate a range of @xmath115 in the band @xmath1 for @xmath116 . in the sgp field @xmath117 takes on relatively low values so that , for example , for only 8 percent of the clusters selected is @xmath118 . therefore we can be certain that this source of incompleteness is absent at fluxes greater than @xmath119 . the removal of contaminating sources from the list of 477 candidates was by necessity an iterative process , in which the decisions made were reviewed repeatedly to ensure that the criteria used had been applied consistently . the following sections describe the steps in this process . for each cluster candidate a finding chart @xmath120 in size was extracted from the cosmos data base . in addition an overlay was made of the x - ray surface brightness contours upon the cosmos optical field . for example figure [ f8 ] shows the overlay made for the abell cluster a3854 . examination of these finding charts and overlays during the selection process has been valuable for several reasons . first , visual scrutiny has removed cases in which csearch had been deceived by artifacts , for example satellite trails , the haloes and diffraction spikes of bright stars , and the attempt by cosmos to resolve the structure of bright nearby galaxies . these structures often were resolved by cosmos into objects identified as galaxies . second , the rass-2 analysis resulted in multiple detections of extended sources such as clusters , which may be removed after examination of the x - ray / optical overlays . as described in section 3.3 , a significant number of redundant detections was removed to minimise any bias of the csearch analysis . however this process was not perfect , and it was possible later to identify further cases by careful re - examination of the overlays . the latter are counted in the detection statistics shown later in table 2 . finally , scrutiny of the finding charts and overlays has identified occasional errors , which are simple to correct , where a csearch candidate is identified with a faint rass-2 x - ray source having a brighter x - ray neighbour . the gca technique then has preferred the latter and derived the higher count - rate . this problem was detected through visual examination of the overlays . the temperature and extent of the hot gas in clusters of galaxies make it possible to set limits to both the x - ray hardness ratio and the angular extent of a rass cluster source ( ebeling et al . 1996 , 1998 ) . this provides effective discriminants between clusters and the majority of contaminating sources in the csearch sample , made up of agn and stars . in this section we justify the following two criteria , which have been used in removing a large fraction of these contaminating sources : + 1 ) the * upper limit * of the x - ray hardness ratio ( section 3.4 ) is less than zero , + i.e. if @xmath121 , the source is too soft . + 2 ) the core radius @xmath113 obtained from the angular extent calculation is less than + @xmath122 , i.e. the object is a point source . + these criteria were applied independently in rejecting sources . in order to establish criteria using these parameters , we selected control samples describing the characteristics of clusters , agn and stars . the first contained @xmath123 abell clusters extracted from the total sample of @xmath124 candidates ( section 3.4 ) , using as a criterion a correlation between the x - ray position and the centroid of an abell cluster . conservatively we excluded from the sample clusters where the separation between the rosat and abell positions was greater than @xmath125 , where the survey exposure was low @xmath126 , or where x - ray maps revealed nearby components in the source region , one of which might not be a cluster . in each case the x - ray / optical overlay image ( section 3.5.1 ) and the cosmos finding chart were examined . the count - rates of these clusters covered the full range of the study reported here . the second and third control samples comprised agn and stars detected by the rosat survey and identified through spectroscopic optical observations . this project was a collaboration between mpe and eso , in which the error circles of all rass-1 sources in four relatively small regions of the southern hemisphere were searched using the cosmos data base to identify optical candidates . the nature of these candidates was examined through spectroscopic observations , after which the likely source of the x - ray emission was identified ( t. a. fleming 1994 , private communication ) . using three of these fields having a total area of @xmath127 , we selected samples of @xmath128 agn and @xmath129 stars , whose positions are within 20 arcsec of the rass-2 source positions . again , the x - ray count - rate of these sources covered the full range of the cluster survey reported here . the hardness ratio ( @xmath130 ) distributions for these samples are shown in figure [ f9 ] . whereas @xmath131 for @xmath132 percent of the abell clusters in our test sample , @xmath133 for @xmath134 and @xmath135 percent of the agn and star samples respectively . therefore a conservative threshold of zero for the * upper limit * to @xmath130 has been set , above which rass sources qualify for retention in the list of cluster candidates . this is an effective method of removing a large fraction of the stars and agn in our sample of 477 candidates . the angular extent , as measured by the derived core radius @xmath113 , provides an effective test for distinguishing clusters from agn , which however runs into two difficulties . the first is caused by the diminishing size of clusters with increasing redshift , an effect which is emphasised if the cluster contains a cooling flow . the second is the effect of source confusion in a rass-2 detection of an agn , when the gca analysis may yield a finite value for the source extent . these difficulties are constrained to a certain extent by the redshift limit ( @xmath67 ) and by the limit set to the x - ray count - rate in the hard band ( section 3.4 ) . the distributions of @xmath113 derived from the two samples are shown in figure [ f10 ] . the rise in these curves at a radius of @xmath136 is a result of setting a limit @xmath137 in the gca , and to a lesser extent of the influence of surrounding sources on the growth curve analysis ( section 3.4 ) . the important characteristic is that 58 percent of agns satisfy @xmath138 , whereas @xmath132 percent of abell clusters in our test sample satisfy the condition @xmath139 . therefore only those sources among the 477 csearch cluster candidates were retained which satisfied the condition @xmath139 . as discussed in section 3.5.5 , this results in some incompleteness near the survey limit , where the rass and cosmos data become marginal in the detection of distant clusters . we have examined four clusters in the abell control sample , which were rejected when these hardness ratio and core radius criteria were used . one , associated with a2800 , failed the hardness ratio test . this may be due to contamination , as it is in a confused region of x - ray emission . three sources , associated with a3866 , s181 and s1121 , failed the core radius test . in the case of s181 the galaxies are faint and displaced from the center of the x - ray source , which is comparatively strong ( @xmath140 in the hard band ) . the small rass-2 error circle contains a stellar object ( @xmath141 ) , and therefore we conclude that this source is an agn ( the emission from agn comes from a concentrated region , so that often cosmos classifies the object as a star ) . the source coincident with a3866 has a high count - rate ( @xmath142 ) yet shows only weak signs of extent . although the spectroscopic observations ( romer 1995 ) revealed evidence of an agn , overlaying an einstein hri image upon an optical dss image indicates that the central object is probably a cd galaxy and that the x - ray source is very compact but not a point source . thus the decision to reject this source is open to correction . s1121 appears to be an example of a distant cluster near the limit of the survey sensitivity . thus four clusters in the test sample of 118 abell clusters were rejected , from which we estimate that in applying the hardness ratio and core radius criteria to our cluster sample , the loss of clusters should be less than 4 percent . bright stars appear in the sample of cluster candidates either because of a chance association with a region of high galaxy density , or because cosmos has broken down the diffraction spikes and/or halo into objects classified as galaxies . they have been removed using two tests , one being the hardness ratio test described in section 3.5.2 and one being a statistical test . in the latter we have analysed @xmath143 regions of diameter @xmath144 in the sgp field , using the cosmos data to derive the mean density of stars as a function of @xmath54 magnitude . the following argument was used to reject rass-2 sources having stars brighter than @xmath145 in the error circle . the density of stars brighter than @xmath145 was found to be @xmath146 , which implies that the probability of finding such a star by chance in a rass-2 90-percent confidence error circle , which typically has a radius of about @xmath147 , is @xmath148 . therefore if we reject all candidates having a star brighter than @xmath145 within the error circle , the loss of clusters from the sample should be no more than @xmath149 . this procedure has been applied by correlating the candidate list with stars in the simbad and tycho ( hg et al . 1998 ) catalogs . a correlation of the @xmath150 magnitudes in these catalogs with the cosmos @xmath54 magnitudes shows that the latter are lower by about 0.5 magnitudes for stars fainter than @xmath151 , but that the difference widens steadily for brighter stars . therefore in examining bright simbad or tycho catalog stars in the rass-2 error circle we set a threshold of @xmath152 , below which the star was judged to be the cause of the x - ray emission . additional stars , not listed in the simbad and tycho catalogs , were identified later by inspection of the cosmos finding charts and the x - ray / optical overlay images for the 477 candidates . the x - ray sources removed by the hardness ratio @xmath153 and core radius @xmath154 tests have been reviewed carefully , using correlations of the list of 477 candidates with the veron catalog of active galaxies ( veron - cetty & veron 1998 ) and the nasa extragalactic data base ( ned ) to identify sources for which the 90 percent rass-2 error circle contained an identified agn , and examination of the x - ray / optical overlay images and the cosmos finding charts . however in the sample remaining after these tests , after removal of stars ( section 3.5.3 ) and after identification of detections resulting from cosmos artifacts , there remained a contaminant comprising 28 sources for which @xmath155 or @xmath139 , but whose identification as a cluster was suspect . these sources were reviewed using again the veron and ned correlations , overlay images and cosmos finding charts , but in addition using the results of the csearch analysis . of these 28 we rejected 9 as being agn and 1 as an x - ray source superposed on a nearby galaxy over 5 arcmin in extent . nine sources were excluded because the extent detection was due to source confusion , and 8 because csearch either made an identification at a large radius only , inconsistent with the small x - ray extent , or made an error due to cosmos artifacts . the one remaining source requires further study and was rejected tentatively . there is a possible association with a2813 , but the minimum found by gca in the core radius fit was very shallow and the evidence for a cluster in the x - ray / overlay image was indecisive . the total number of agn identified was 127 , and figure [ f11 ] shows the distribution of their @xmath54 magnitudes , as measured by cosmos . where the agn was not identified in the veron catalog or ned , @xmath54 for the brightest object inside the 90 percent error circle was used . the half - width of the distribution lies in the range @xmath156 . finally , we discovered five candidates showing large extended regions of x - ray emission @xmath157 in diameter , whose optical fields show no signs of a significant galaxy population and whose origin as the confusion of several sources is not obvious . after removal of rass-2 sources contaminating the sgp sample of 477 cluster candidates , using the procedures described above , there remained 186 candidates . the statistics of this selection process are summarised in table 2 . of the 61 redundant rass-2 detections in table 2 which eluded the earlier screening , 24 were associated with clusters . therefore of the 477 candidate sources @xmath158 are associated with clusters , which implies that the contamination of the sample of 477 was 0.56 . this is in marginal agreement with the results of the csearch analysis ( section 3.3 ) , which predicted @xmath159 . examination of table 1 shows that @xmath160 of the cluster candidates were identified by csearch using the smaller circles 1 and 2 , and consequently that formally the contamination of the sample should be 0.5 or less , and the completeness about 0.90 . therefore we can estimate that the sample of 477 candidates should contain @xmath161 clusters and that a complete sample obtained using csearch would have yielded @xmath162 clusters . therefore we can estimate that the overall completeness of the sample of selected clusters is @xmath163 . this incompleteness , which is the sum of the three effects summarised in section 2 , contributes to the flattening of the log n / log s distribution at fluxes below @xmath2 in the band @xmath1 ( section 5 ) . cc clusters of galaxies & 186 + active galactic nuclei ( agn ) & 127 + stars & 74 + extended x - ray emission regions & 5 + redundant rass-2 detections & 61 + sources below x - ray count - rate threshold & 24 + + total & 477 + the above argument implies that @xmath164 clusters have been missed in searching the 477 candidates . this has arisen in most cases while applying the criteria @xmath121 and @xmath138 to reject candidates , and in section 3.5.2 we have estimated that applying these criteria applied to a control sample of abell clusters should yield an incompleteness of @xmath165 . two examples of clusters with overly soft spectra are a2800 ( @xmath166 ) and the cluster @xmath167 ( @xmath168 ) discovered by tucker et al . ( 1995 ) in the emss . in neither case is the nature of the contaminating source clear . one limit of this study is that the criterion @xmath139 for selecting cluster candidates will inevitably discriminate against the more distant clusters of small angular extent , particularly those in which a cooling flow makes the x - ray source more compact . this loss is evident in the flattening of the log n - log s distribution at low flux ( section 5 ) . three aspects of the analysis bear out this trend . first , of the sources rejected because @xmath138 we identified at least 4 cases , where the x - ray / optical overlays showed suggestive but marginal evidence that they were clusters . second , there were 11 in our catalog of 186 clusters , for which the detection of angular extent appeared to be near its limits . the gca yielded a small but non - zero core radius , but simultaneously the ks - test yielded significant probabilities , between @xmath169 and @xmath170 percent , of the source being point - like . third , in the case of distant clusters accepted by csearch , cosmos classified some of the galaxies as faint objects , which the csearch analysis did not count . this is obvious in some of the x - ray / optical overlays of these objects , which show the dominant elliptical galaxy with halo surrounded by faint objects . therefore we conclude that some distant clusters may have been missed by csearch . in order to convert the rosat count - rate into an energy flux outside the galaxy , the spectrum of the hot intracluster medium ( icm ) has been convolved with the response functions of the interstellar medium ( ism ) and the instrument . the plasma emissivity was derived using the version of the raymond & smith ( 1977 ) code installed in 1994 in the mpe exsas software . the technique has been verified by comparing results with those obtained using the exspec analysis code ( bhringer et al . the cluster plasma was assumed to contain elements heavier than hydrogen having abundances 0.35 times the solar abundance . the hi column densities used in calculating the ism absorption have been taken from dickey & lockman ( 1990 ) . these calculations have been made over ranges of temperature between 0.05 and 10 kev ( @xmath171-@xmath172 k ) , and of hi column density between @xmath173 and @xmath174 . the resulting matrix was used to convert the count - rate in the hard band to an energy flux @xmath175 in units of @xmath176 in the 0.1 - 2.4 kev band . for each cluster whose redshift has been measured the x - ray luminosity has been calculated using the relations : + @xmath177 @xmath178 where @xmath179 is the luminosity in the 0.1 - 2.4 kev band , @xmath180 is the correction for redshifting of the radiation , and @xmath181 is the luminosity distance . k has been calculated for ranges of redshift between 0.025 and 0.5 , and of icm temperature between 0.05 and 10 kev . the resulting matrix was used to convert energy flux into luminosity . measured cluster temperatures have been taken from the compilation in table 1 of white et al . ( 1997 ) . where measured temperatures have not been found we have used the luminosity - temperature correlation derived by white et al . ( 1997 ) : + @xmath182 where @xmath183 is the x - ray luminosity in the 0.01 - 80 kev band and t is the temperature in kev . the assumed temperature was used to calculate a bolometric correction to the luminosity . for each source a cycle of iterations was performed in the flux and luminosity calculations , until the change in temperature fell below 1 percent . where neither redshift nor temperature has been measured , we have assumed a temperature of @xmath184 . the procedures described in sections 2 , 3 and 4 for searching the rass-2 data base for cluster candidates , selecting clusters from this sample and determining their x - ray characteristics , have resulted in the catalog of 186 clusters shown in table 3 . the following notes , listed by the column number in the table , augment the information given in the header to each column . + ( * 1 * ) rosat x - ray source name . + ( * 2 * ) right ascension ( h : m : s j2000 ) of the centroid ( gca ) of the cluster x - ray emission . + ( * 3 * ) declination ( d : m : s j2000 ) of the centroid ( gca ) of the cluster x - ray emission . + ( * 4 * ) count - rate in @xmath185 in the hard band ( @xmath43 ) , with error ( gca ) . + ( * 5 * ) x - ray hardness ratio , with error ( gca ) . + ( * 6 * ) total rosat survey exposure time for the cluster , in seconds . + ( * 7 * ) column density of interstellar atomic hydrogen at the cluster center , in units of @xmath186 . + ( * 8 * ) temperature of the cluster gas , derived by the procedure described in section 4 , in @xmath187 . the superscript @xmath188 indicates that no measurement was found , and the temperature was determined using an @xmath189 correlation . where no redshift was available a temperature of @xmath184 was assumed . + ( * 9 * ) energy flux from the cluster , with the absorption by galactic interstellar absorption removed . the units are @xmath190 in the @xmath1 band . + ( * 10 * ) x - ray luminosity of the cluster in its rest - frame , derived using the procedure described in section 4 . the units are @xmath191 in the energy band @xmath1 . we assume @xmath34 and @xmath35 . + ( * 11 * ) optical identification of the cluster . the meaning of the prefix to the identification and the appropriate reference are given in table 4 . for five sources identified as double clusters , the label dc has been appended to the identification . + ( * 12 * ) redshift of the cluster . + ( * 13 * ) label for the reference from which the redshift was obtained : + * 1 . * romer ( 1995 ) * 2 . * h. bhringer and l. guzzo 1999 , eso key project , private communication ) * 3 . * struble & rood ( 1999 ) * 4 . * katgert et al . ( 1996 ) * 5 . * quintana & ramirez ( 1995 ) * 6 . * fetisova et al . ( 1993 ) * 7 . * de grandi et al . ( 1999 ) * 8 . * h. andernach 1989,``a compilation of measured redshifts of abell clusters '' , unpublished , private communication * 9 . * crawford et al . ( 1995 ) * 10 . * huchra et al . ( 1999 ) * 11 . * shectman ( 1985 ) * 12 . * mazure et al . ( 1996 ) * 13 . * collins et al . ( 1995 ) * 14 . * dalton et al . ( 1997 ) * 15 . * de vaucouleurs et al . ( 1991 ) * 16 . * stocke et al . ( 1991 ) * 17 . * postman et al . ( 1992 ) * 18 . * abell , corwin , & olowin ( 1989 ) * 19 . * katgert et al . ( 1998 ) * 20 . * ebeling & maddox ( 1995 ) * 21 . * muriel , nicotra , & lambas ( 1995 ) * 22 . * shectman et al . ( 1996 ) * 23 . * lauberts & valentijn ( 1989 ) * 24 . * da costa et al . ( 1998 ) * 25 . * dalton , efstathiou , & sutherland ( 1994 ) * 26 . * di nella et al . ( 1996 ) * 27 . * ratcliffe et al . ( 1998 ) * 28 . * given in ned , reference uncertain . + ( * 14 * ) number of galaxy redshifts used in deriving the cluster redshift . + ccc a & abell catalog & abell et al . ( 1989 ) + s & abell catalog : southern supplementary clusters & abell et al . ( 1989 ) + am & cambridge apm catalog & ned : arp & madore ( 1987 ) + apmcc & cambridge apm cluster catalog & dalton et al . ( 1997 ) + cid & efar survey & barton et al . ( 1996 ) + edcc & edinburgh - durham cluster catalog & lumsden et al . ( 1992 ) + emss & einstein medium sensitivity survey & stocke et al . ( 1991 ) + eso & european southern observatory & no reference in ned + hcg & hickson compact group catalog & hickson ( 1982 ) , barton et al . ( 1996 ) + scg & catalog of cosmos southern compact groups & prandoni , iovino & macgillivray ( 1994 ) + ngc & cluster dominated by ngc galaxy & ned + sh & cluster catalog from shane - wirtanen galaxy counts & shectman ( 1985 ) + images of these clusters , showing overlays of the hard band x - ray contours upon optical images reconstructed from the cosmos digital data base , are accessible over the internet at ( http://wave.xray.mpe.mpg.de/publications/papers/2001/rass-sgp-clusters/ xray - optical - images ) . they comprise 134 abell clusters , 15 clusters found in other catalogs , and 37 newly discovered clusters . the number of abell clusters in the sgp survey area is 1215 , including 332 southern supplementary clusters , therefore the abell clusters in the sgp catalog ( table 3 ) comprise 11.0 percent of all abell clusters in the sgp survey area . the abell fraction of the sgp cluster sample ( 72% ) is in good agreement with that of the bcs ( ebeling et al . 1999 , 2000 ) . we have compared the catalog in table 3 with the reflex catalog of bright clusters published by de grandi et al . ( 1999 ) , which contains 54 clusters lying in the sgp region . applying their energy flux threshold of @xmath192 in the @xmath1 band to table 3 , we find all but one cluster from the list of de grandi et al . the exception is a3866 , which we have rejected tentatively as a source whose x - ray flux may be dominated by an agn ( section 3.5.2 ) . however table 3 contains 20 clusters above this threshold , which are not in the list of de grandi et al . the reason is that these sources were not in the rass-1 source list , but were found by the rass-2 analysis as a consequence of the improvements , in particular lowering source detection thresholds and searching for sources in the hard energy band , which are described in section 3.1 and by voges et al . these 20 sources are ngc 720 , ngc 1132 , five newly discovered clusters , sh518 , and the abell clusters 194 , 2410 , 2442 , 2474 , 2496 , 2554 , 2556 , 2566 , 3984 , 4010 , 4068 and s1136 . the sgp catalog contains 113 clusters having fluxes below the threshold of the de grandi et al . ( 1999 ) study . the log n - log s distribution for the sgp cluster sample is shown in figure [ f12 ] , where it is compared with the results of other surveys . to avoid any errors caused by converting hard band ( @xmath43 ) fluxes in these surveys to broad band ( @xmath1 ) fluxes , we have made this comparison in the hard band . the circular symbols ( @xmath193 ) are the result of the survey of clusters by de grandi et al . ( 1999 ) , who reanalysed clusters detected in the rass-1 data base over a @xmath23 area of the southern sky and made first use of redshifts obtained from the eso key project . figure [ f12 ] includes also one point from the heao-1 a-2 cluster survey ( piccinotti et al . 1982 ) , the fit to the log n - log s curve obtained by ebeling et al . ( 1998 ) from the rosat brightest cluster sample ( bcs ) , and a box setting limits at low fluxes , derived from an analysis of emss results ( rosati et al . 1995 ) . there are significant differences among these results at energy fluxes less than @xmath194 . there is a flattening of the sgp log n - log s distribution at energy fluxes less than about @xmath195 , equivalent to a flux of @xmath26 in the rosat broad band ( @xmath1 ) , which is the completeness limit derived in section 6 . applying this limit we obtain a sample of 112 clusters , of which 110 have measured redshifts . the sources of the incompleteness of the sgp catalog at low fluxes have been discussed in section 3.5.5 . in addition there is a steepening of the sgp log n - log s curve over a short range of hard band flux near @xmath196 , a characteristic evident to a lesser extent also in the results of de grandi et al . this results in the distribution being higher than the fit of ebeling et al . ( 1998 ) and being at the upper limit set by rosati et al . ( 1995 ) , within the range @xmath197 . this may be a statistical fluctuation , but as we show in section 7 , there is significant large - scale structure in the sgp field out to a redshift of @xmath33 , which may be one cause of this feature in the sgp log n - log s distribution ( figure [ f12 ] ) . the optical observing program and a search of the literature yielded redshifts for 157 of the 186 clusters in table 3 . the distribution of the clusters in redshift space is summarised in the wedge diagram shown in figure [ f13 ] , in which the clusters are projected onto a plane whose co - ordinates are cluster redshift and right ascension . in this figure the symbol size is a measure of the cluster x - ray luminosity in the @xmath1 band . table 3 contains five sources , labelled dc in column 11 , for which the x - ray and optical data provide compelling evidence that they are interacting double clusters . in contrast the pair of objects rxcj0108.8 - 1524 and rxcj0108.9 - 1537 , which in projection are close , do not constitute a double cluster as they are well separated in redshift space . the former is a151 , which has a redshift of 0.0537 ( h. andernach 1989 , unpublished , private communication ) , whereas spectroscopic observation of the latter yield two redshifts which tie the galaxies to a151 but nine having a mean of @xmath198 ( katgert et al . 1996 , 1998 ) . in section 3.5.5 it was estimated from a comparison of the number of clusters detected in the sgp with the csearch predictions , that the sample completeness was 0.79 , and it was concluded that the missing ones were mainly distant clusters of small angular extent . the effect is to flatten the log n - log s distribution at low fluxes ( figure [ f12 ] ) , and using the following arguments we have established that the sample is essentially complete at fluxes greater than @xmath2 . first , we compare our log n - log s distribution with the results of deeper surveys , and of the wider field reflex and bcs surveys . at @xmath2 in the sgp distribution ( figure [ f12 ] ) we obtain @xmath199 . for deeper surveys , which have smaller fields and therefore limited statistics at this flux , we obtain @xmath200 ( rosati et al . 1998 ) , @xmath201 ( vikhlinen et al . 1998 ) , @xmath202 ( henry et al . 2001 ) , and @xmath203 ( gioia et al . 2001 ) . for the reflex survey we obtain @xmath204 ( bhringer et al . 2001 ) and for the bcs survey @xmath205 ( ebeling et al . we find no evidence that the sgp result is inconsistent with these surveys at this flux threshold . second , we apply a standard measure of the formal completeness of a sample , namely the @xmath206 test ( e.g. , schmidt 1968 , avni & bahcall 1980 ) . for the computation of the maximum volume @xmath207 the effective area , the local variation of the x - ray flux limit ( both for a minimum of 10 source counts ) , and the @xmath180 correction ( see section 4 ) are taken into account . the test is performed for subsamples with different x - ray flux limits . the mean of the @xmath206 values and their @xmath208 standard deviations are plotted in figure [ f14 ] . note that the errors do not include cosmic variance . below @xmath209 the deviations from the ideal value of 0.5 are significant , indicating that the sample gets definitively incomplete below this flux limit . between @xmath210 and @xmath211 it is difficult to make a definite statement about significant deviations from the 0.5 line because the large scale inhomogeneities in the spatial distribution of the clusters ( see section 7 ) start to modulate the averaged @xmath206 values . note also that the measurements at different flux limits are not statistically independent . a good estimate of the flux completeness limit is @xmath212 , because for fluxes brighter than this limit the averaged @xmath206 values are all within the 90 percent confidence range ( note that cosmic variance increase the given formal errors ) . third , we examine the behavior of the radially averaged comoving cluster number density along the redshift direction . if we assume that within the tested redshift range no large evolutionary effects are present , one would expect an almost constant cluster number density profile with superposed large - scale inhomogeneities . figure [ f15 ] shows the superposition of a set of cluster number density profiles obtained from 10 volume - limited subsamples within the luminosity range @xmath213 to @xmath214 . outside this luminosity interval the sizes of the subsamples are smaller than 10 and are thus dominated by small - number statistics . the subsamples are restricted to clusters with at least 10 source counts and x - ray fluxes larger than @xmath215 , as obtained from the @xmath206 test mentioned above . the effective survey area covered by each subsample is taken into account . the individual density profiles are normalized using the average cluster number density estimated for each subsample . the bars in figure [ f15 ] represent the formal @xmath216 poisson errors , within which no systematic large - scale gradients of the number density are found for @xmath217 . the comparatively large high - density region found in the redshift interval @xmath218 is composed mainly of several superclusters . a more detailed analysis of the large - scale structure is given in section 7 . finally , we have verified that the variation of rass exposure time in the sgp field ( figure [ f1 ] ) has little effect on the effective area at a flux level of @xmath2 . at the detection threshold of the gca ( 10 source counts ) , this area is 98 percent of the total area of the field . the wedge diagram for the sgp cluster sample ( figure [ f13 ] ) reveals that the spatial density of clusters at redshifts greater than @xmath3 is too low for the purpose of tracing large - scale structure , and we concentrate the discussion on clusters within this limit . a detailed examination of large - scale structure in the southern sky was made by tully et al . ( 1992 ) , who analysed the abell , corwin & olowin ( 1989 ) catalog of clusters . in this study measured redshifts were complemented by estimates based on the magnitudes of the third and tenth brightest galaxies , having an accuracy of @xmath219 out to redshifts of about @xmath220 ( zamorani et al . we have condensed the results into a sky map of the centroids of superclusters identified in the sgp region ( figure [ f16 ] ) . another wedge diagram ( figure [ f17 ] ) shows the distribution of these superclusters in redshift space . . [ f16 ] the dotted line in figure [ f16 ] traces the supergalactic equator , defined by the plane of a disk - like structure , the local supercluster , identified by de vaucouleurs & de vaucouleurs ( 1964 ) using measurements of the radial velocities of a large sample of local galaxies . the maximum values of these velocities were around @xmath221 ( @xmath222 ) . tully et al . ( 1992 ) have presented persuasive evidence that the pisces - cetus supercluster is likewise a flattened structure extending along the supergalactic equator , which is an extension of the local supercluster out to redshifts of at least @xmath223 . this idea is supported by our x - ray survey of clusters in the sgp . figure [ f18 ] is a sky map of all clusters in the sgp catalog ( table 3 ) having a redshift of less than @xmath224 , which shows clearly a concentration of clusters around the supergalactic equator ( figure [ f16 ] ) . of the 36 clusters identified by tully ( 1987 ) as members of the pisces - cetus supercluster , the following seven appear in table 3 and have a redshift @xmath225 : a14 , a85 , a119 , a133 , a147 , a151 , and a168 . the six brightest clusters in figure [ f18 ] , having hard - band x - ray count rates greater then @xmath226 , lie close to the supergalactic equator . five of these clusters , a85 , a119 , a133 , a4059 and s1101 , have redshifts between @xmath227 and @xmath228 , consistent with their being members of the pisces - cetus supercluster . the fifth , a4038 , is nearer at a redshift of @xmath229 , placing it at the edge of the local supercluster . the pisces - cetus supercluster is evident also in the wedge diagram ( figure [ f13 ] ) , in which the group of x - ray clusters at @xmath230 and @xmath231 is close to the location of the supercluster center . more distant structures are difficult to identify in the x - ray survey . tully et al . ( 1992 ) identified a structure orthogonal to the supergalactic equator , which includes the superclusters aquarius a and b , aquarius - capricornus , and sculptor a and b. the location of this structure , which extends from @xmath231 to @xmath232 ( @xmath233 ) , may be traced in figures [ f16 ] and [ f17 ] . to search for evidence of this structure in the x - ray data , we have constructed a sky map ( figure [ f19 ] ) which includes only x - ray clusters in the range @xmath234 . the map shows an elongated feature in the nw quadrant , which is perpendicular to the supergalactic equator . the argument that this feature traces the structure evident in the findings of of tully et al . ( 1992 ) is supported by two observations . first , the group of x - ray clusters in the nw corner of figure [ f19 ] belongs to the aquarius - capricornus supercluster . a correlation in sky position is evident upon comparing figures [ f19 ] and [ f16 ] , and a correlation in redshift space is seen upon comparing figures [ f13 ] and [ f17 ] . second , the two bright clusters a2597 and a2670 , highlighted as black circles in the nw quadrant , were identified by tully ( 1987 ) as members of the aquarius supercluster . batuski et al . ( 1999 ) have performed redshift measurements recently which establish that aquarius a and b are essentially one filamentary structure almost aligned with the line - of - sight between @xmath235 and @xmath236 . this concentration of abell clusters on the sky near @xmath237 , @xmath238 ( batuski et al . 1999 ) is not evident in the x - ray data ( figure [ f19 ] ) . however , care is needed in interpreting this result as it may be biased by the relatively low exposure time in this part of the rass ( figure [ f1 ] ) . therefore we conclude that , at least for redshifts out to @xmath33 , there is significant large - scale structure in the sgp field . however , there is one part of this structure which is not strongly evident in the x - ray maps shown in figure [ f18 ] and figure [ f19 ] . this is the horologium - reticulum supercluster ( tully et al . 1992 ) , a rich association of clusters having its centroid at @xmath239 and located at the position shown in figure [ f16 ] . there is only one bright x - ray cluster in our survey , a3113 at ( @xmath240 , figure [ f19 ] ) and @xmath241 , which may be associated with this supercluster . we can not rule out the possibility that the supercluster is inconspicuous because of its location close to the boundary of the sgp survey region . a systematic search of the rosat all - sky survey for clusters of galaxies has been performed in an area of @xmath242 centered on the south galactic pole . the search was made by correlating the list of rass sources detected in the hard x - ray band ( @xmath43 ) with the projected surface density of galaxies derived from the cosmos digitised uk schmidt iiia - j survey of the southern sky . the analysis yielded a list of 3236 candidates , of which @xmath243 should be chance associations with non - cluster sources , mainly agn and stars , and which should be 90 percent complete in containing clusters . the study was bounded during x - ray analysis by applying a lower limit of @xmath42 to the count - rate in the hard band , yielding a list of 477 candidates . non - cluster sources were removed by applying tests using the source x - ray extent and hardness ratio , by scrutiny of cosmos finding charts and overlays of the rass and cosmos images , and by correlation with such catalogs as the nasa extragalactic ( ned ) , simbad , tycho and veron data bases . this was performed in as objective a manner as possible , requiring an iterative sequence of procedures . the search resulted in 186 clusters , comprising 134 abell clusters , 15 clusters found in other catalogs , and 37 newly discovered clusters . the catalog shown in table 3 , and the overlays of the rass hard x - ray and cosmos optical images , are accessible over the internet ( http://wave.xray.mpe.mpg.de / publications / papers/2001/rass - sgp - clusters / xray- optical - images ) . the catalog has a minimum energy flux of @xmath244 in the band @xmath1 . tests show that a sample of 112 clusters , obtained when a flux limit of @xmath2 is applied , is essentially complete . we estimate the completeness of the overall catalog at a flux limit of @xmath0 to be @xmath245 . in support of studies of cluster evolution and large - scale structure using the results of the survey , a series of optical observing campaigns at the aat and at saao was conducted to obtain cluster redshifts . in addition we have used a number of redshifts obtained in programs at eso in support of the reflex project . using these campaigns and an extensive search of the literature we have obtained redshifts for 157 of the clusters in the catalog ( table 3 ) . redshifts are available for 110 of the 112 clusters in the complete flux - limited sample . examination of the spatial distribution of the sgp x - ray cluster sample corroborates earlier studies of large - scale structure , which were based on optical observations ( de vaucouleurs & de vaucouleurs 1964 , tully et al . the basic features are a flattened structure in the supergalactic plane , reaching out to @xmath246 and embracing the pisces - cetus supercluster , and an orthogonal structure at @xmath247 which includes the aquarius , aquarius - capricornus and sculptor a and b superclusters . we thank prof . trmper and the rosat team for the unstinting support , the myriad valuable discussions , the access to the rass data base , the use of the rass and exsas software , and the use of the powerful computing system under joachim paul and his team , which have made this work possible . the rosat project is supported by the bildungsministerium fr bildung und forschung ( bmbf / dlr ) and the max - planck - gesellschaft ( mpg ) . in equal measure we thank the royal observatory edinburgh and the naval research laboratory ( nrl ) for the intensive effort that went into the creation of the cosmos digitised optical data base and its installation at mpe , which was an essential factor in identifying clusters in the rass data base . our gratitude goes also to the anglo - australian telescope , the south african astronomical observatory and the european southern observatory for the expert assistance provided to the optical observations crucial to this project . we thank the reflex project for its cooperation in exchanging redshifts , and in particular luigi guzzo of the osservatorio astronomica di brera for his analysis of those eso key project observations , which yielded 14 redshifts used in this study . finally , this research has been supported by the office of naval research and nrl , and by the national aeronautics and space administration through an astrophysics data program grant ( s57759-f ) . abell , g. o. , corwin jr . , h. g. , & olowin , r. p. 1989 , apjs , 70 , 1 + arp , h. c. , & madore , b. f. 1987 , `` a catalogue of peculiar galaxies and associations + ( cambridge : cambridge university press ) + avni , y. & bahcall , j.n . 1980 , apj , 235 , 694 + barton , e. , geller , m. j. , ramella , m. , marzke , r. o. , & da costa , l. n. 1996 , aj , 112 , + 871 + batuski , d. j. , miller , c. j. , slinglend , k. a. , balkowski , c. , maurogordato , s. , + cayatte , v. , felenbok , p. , & olowin , r. 1999 , apj , 520 , 491 + bhringer , h. , et al . 2000 , apjs , 129 , 435 + bhringer , h. , et al . 2001 , a&a , 369 , 826 + caretta , c. a. , maia , m. a. g. , & willmer , c. n. a. 2000 , aj , 119 , 524 + cavaliere , a. , & fusco - femiano , r. 1976 , a&a , 49 , 137 + collins , c. a. , guzzo , l. , nichol , r. c. , & lumsden , s l. 1995 , mnras , 274 , 1071 + collins , c. a. , burke , d. a. , romer , a.k . , sharples , r. m. , & nichol , r. c. 1997 , apj , + 479 , l117 + collins , c. a. , et al . 2000 , mnras , 319 , 939 + crawford , c. s. , edge a. c. , fabian , a. c. , allen , s. w. , boehringer , h. , ebeling , h. , + mcmahon , r. g. & voges , w. 1995 , mnras , 274 , 75 + da costa , l. n. , et al . 1998 , aj , 116 , 1 + dalton , g. b. , efstathiou , g. & sutherland , w. j. 1994 , mnras , 269 , 151 + dalton , g. b. , maddox , s. j. , sutherland , w. j. , & efstathiou , g. 1997 , mnras , 289 , 263 + de grandi , s. , molendi , s. , bhringer , h. , & voges , w. 1997 , apj , 486 , 738 + de grandi , s. , et al . 1999 , apj , 514 , 148 + de vaucouleurs , g. , & de vaucouleurs , a. 1964 , first reference catalogue of bright + galaxies ( austin , university of texas press ) + de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. , buta , r. , paturel , g. & fouque , p. + 1991 , third reference catalogue of bright galaxies ( springer - verlag : new york ) + di nella , h. , couch , w. j. , paturel , g. , & parker , q. a. 1996 , mnras , 283 , 367 + dickey , j. m. , & lockman , f. j. 1990 , ann . astrophys . , 28 , 215 + ebeling , h. , & wiedemann , g. 1993 , phys . e , 47 , 704 + ebeling , h. , & maddox , s. j. 1995 , mnras , 275 , 1155 + ebeling , h. , voges , w. , bhringer , h. , edge , a. c. , huchra , j. p. , & briel , u. g. 1996 , + mnras , 281 , 799 + ebeling , h. , edge , a. c. , bhringer , h. , allen , s. w. , crawford , c. s. , fabian , a. c. , + voges , w. , & huchra , j. p. 1998 , mnras , 301 , 881 + ebeling , h. , edge , a. c. , allen , s. w. , crawford , c. s. , fabian , a. c. , & huchra , j. r. + 2000 , mnras , 318 , 333 + edge , a. c. , stewart , g. c. , fabian , a. c. , & arnaud , k. a. 1990 , mnras , 245 , 559 + fetisova , t. s. , kuznetsov , d. yu . , lipovetski , v. a. , starobinski , a. a. , & olovin , r. p. + 1993 , astronomy letters , 19 , 198 + gioia , i. m. , henry , j. p. , mullis , c. r. , voges , w. , briel , u. g. , bhringer , h. & + huchra , j. p. 2001 , apj , 553 , l105 + henry , j.p . , gioia , i. m. , maccacaro , t. , morris , s. l. , stocke , j. t. , & wolter , a. 1992 , + apj , 386 , 408 + henry , j. p , gioia , i. m. , mullis , c. r. , voges , w. , briel , u. g. , bhringer , h. & + huchra , j. p. 2001 , apj , 553 , l109 + heydon - dumbleton , n. h. , collins , c. a. , & macgillivray , h. t. 1989 , mnras , 238 , 379 + hickson , p. 1982 , apj , 255 , 382 + hg , e. , kuzmin , a. , bastian , u. , fabricius , c. , kuimov , k. , lindegren , l. , + makarov , v. v. , & rser , s. 1998 , a&a , 335 , l65 + huchra , j. p. , vogeley , m. s. & geller , m. j. 1999 , apjs . 121 , 287 + katgert , p. , et al . 1996 , a&a , 310 , 8 + katgert , p. , mazure , a. , den hartog , r. , adami , c. , biviano , a. , & perea , j. 1998 , a&as , + 129 , 399 + kowalski , m. p. , ulmer , m. p. , cruddace , r. g. , & wood , k. s. 1984 , apjs , 286 , 822 + jones , l. r. , scharf , c. , ebeling , h. , perlman , e. , wegner , g. , malkan , m. , & horner , d. + 1998 , apj , 495 , 100 + lauberts , a. , & valentijn , e. 1989 , eso catalog , ' ' the surface photometry of the + eso - uppsala galaxies `` ( european southern observatory : garching ) + lumsden , s. l. , nichol , r. c. , collins , c. a. , & guzzo , l. 1992 , mnras , 258 , 1 + macgillivray , h. t. , & stobie , r. s. 1985 , vistas in astronomy , 27 , 433 + mazure , a. , et al . 1996 , a&a , 310 , 31 + mchardy , i. 1978 , mnras , 184 , 783 + mullis , c.r . , henry , j. p , gioia , i. m. , bhringer , h. , briel , u. g. , voges , w. + & huchra , j. p. 2000 , apj , 553 , l115 + muriel , h. , nicotra , m. a. , & lambas , d. g. 1995 , aj , 110 , 1032 + piccinotti , g. , mushotzky , r. f. , boldt , e. a. , holt , s. s. , marshall , f. e. , + serlemitsos , p. j. , & shafer , r. a. 1982 , apj , 253 , 485 + postman , m. , huchra , j. p. , & geller , m. j. 1992 , apj , 384 , 404 + prandoni , i. , iovino , a. , & macgillivray , h. t. 1994 , aj , 107 , 1235 + quintana , h. , & ramirez , a. 1995 , apjs , 96 , 343 + ratcliffe , a. , shanks , t. , parker , q. a. , broadbent , a. , watson , f. g. , oates , a. p. , + collins , c. a. , & fong , r. 1998 , mnras , 300 , 417 + raymond , j. c. , & smith , b. w. 1977 , apjs , 35 , 419 + romer , a. k. , collins , c. a. , bhringer , h. , cruddace , r. g. , ebeling , h. , + macgillivray , h. t. , & voges , w. 1994 , nature , 372 , 75 + romer , a. k. 1995 , doctoral thesis , university of edinburgh + romer , a. k. , et al . 2000 , apjs , 126 , 209 + rosati , p. , della ceca , r. , burg , r. , norman , c. , & giacconi , r. 1995 , apj , 445 , l11 + rosati , p. , della ceca , r. , norman , c. , & giacconi , r. 1998 , apj , 492 , l21 + schmidt , m. 1968 , apj , 151 , 393 + schwartz , d. a. 1978 , apj , 220 , 8 + shectman , s. a. 1985 , apjs , 57 , 77 + shectman , s. a. , landy , s. d. , oemler , a. , tucker , d. l. , lin , h. , kirschner , r. p. , + & schecter , l. p. 1996 , apj , 470,172 + schuecker , p. , et al . 2001 , a&a , 368 , 86 + stocke , j. t. , morris , s. m. , gioia , i. m. , maccacaro , t. , schild , r. , wolter , a. , + fleming , t. a. , & henry , j.p . 1991 , apjs , 76 , 813 + struble , m. f. , & rood , h. j. 1999 , apjs , 125 , 35 + trmper , j. 1983 , adv . space res . , 27 , 1404 + trmper , j. , et al . 1991 , nature , 349 , 579 + tucker , w. h. , tananbaum , h. , & remillard , r. a. 1995 , apj , 444 , 532 + tully , r. b. 1987 , apj , 323 , 1 + tully , r. b. , scaramella , r. , vettolani , g. , & zamorani , g. 1992 , apj , 388 , 9 + veron - cetty , m. p. , & veron , p. 1998 , eso scientific report 18 + vikhlinen , a. , mcnamara , b. r. , forman , w. , & jones , c. 1998 , apj , 502 , 558 + voges , w. , et al . 1999 , a&a , 349 , 389 + voges , w. , et al . 2000 , iau circular 7432 + voges , w. , henry , j. p , briel , u. g. , bhringer , h. , mullis , c. r. , gioia , i. m. & + huchra , j. p. 2001 , apj , 553 , l119 + white , d. a. , jones , c. , & forman , w. 1997 , mnras , 292 , 419 + yentis , d. j. , cruddace , r. g. , gursky , h. , stuart , b. v. , wallin , j. f. , + macgillivray , h. t. , & collins , c. a. 1992 , ' ' digitised optical sky surveys `` , ed . + h. t. macgillivray & e. b. thomson , ( netherlands : kluwer academic publishers ) , 67 + zamorani , g. , scaramella , r. , vettolani , g. , & chincarini , g. 1992 , in ' ' traces of primordial + structure in the universe , ed . h. bhringer & r. a. treumann , mpe report 227 + cccccccccccccc rxcj2201.8 - 2226 & 22 01 50.5 & -22 26 37 & 0.286@xmath2480.043 & 0.82@xmath2480.30 & 237 & 2.61 & 3.554@xmath249 & 5.75 & 1.193 & s987 & 0.0693 & 8 & 20 + rxcj2202.1 - 0949 & 22 02 07.2 & -09 49 48 & 0.280@xmath2480.038 & 1.15@xmath2480.31 & 307 & 4.20 & 4.039@xmath249 & 5.89 & 1.629 & a2410 & 0.0801 & 17 & 1 + rxcj2203.8 - 2130 & 22 03 50.6 & -21 30 31 & 0.139@xmath2480.030 & 0.11@xmath2480.19 & 227 & 2.45 & 2.809@xmath249 & 2.77 & 0.644 & & 0.0732 & 1 & 3 + rxcj2205.6 - 0535 & 22 05 39.6 & -05 35 29 & 0.674@xmath2480.109 & 0.79@xmath2480.12 & 191 & 4.68 & 3.800@xmath250 & 14.40 & 2.103 & a2415 & 0.0582 & 2 & 3 + rxcj2210.3 - 1210 & 22 10 20.0 & -12 10 33 & 0.746@xmath2480.056 & 1.04@xmath2480.08 & 278 & 3.92 & 6.000@xmath250 & 15.70 & 4.642 & a2420 & 0.0831 & 1 & 3 + rxcj2213.0 - 2753 & 22 13 05.0 & -27 53 56 & 0.425@xmath2480.034 & 0.14@xmath2480.17 & 205 & 1.40 & 3.760@xmath249 & 8.28 & 1.373 & & 0.0620 & 1 & 2 + rxcj2214.4 - 1701 & 22 14 26.3 & -17 01 24 & 0.198@xmath2480.020 & 0.79@xmath2480.24 & 305 & 2.61 & 5.845@xmath249 & 4.02 & 4.016 & a3847 & 0.1530 & 3 & 1 + rxcj2214.5 - 1022 & 22 14 32.4 & -10 22 12 & 0.594@xmath2480.052 & 0.95@xmath2480.09 & 280 & 3.84 & 6.578@xmath249 & 12.50 & 5.343 & a2426 & 0.1001 & 1 & 4 + rxcj2216.2 - 0920 & 22 16 15.2 & -09 20 18 & 0.405@xmath2480.041 & 0.87@xmath2480.09 & 274 & 5.24 & 4.869@xmath249 & 8.78 & 2.567 & a2428 & 0.0825 & 1 & 2 + rxcj2216.9 - 1725 & 22 16 56.4 & -17 25 30 & 0.425@xmath2480.034 & 0.35@xmath2480.12 & 277 & 2.28 & 6.969@xmath249 & 8.56 & 6.163 & & 0.1301 & 1 & 2 + rxcj2217.7 - 3543 & 22 17 43.7 & -35 43 32 & 0.329@xmath2480.035 & 0.32@xmath2480.11 & 319 & 1.10 & 6.865@xmath249 & 6.42 & 5.929 & a3854 & 0.1474 & 13 & 9 + rxcj2218.2 - 0350 & 22 18 17.1 & -03 50 01 & 0.427@xmath2480.047 & 0.92@xmath2480.14 & 273 & 5.73 & 5.375@xmath249 & 9.39 & 3.267 & ms2215.7 - 0404 dc & 0.0901 & 1 & 3 + rxcj2218.6 - 3853 & 22 18 40.3 & -38 53 48 & 0.425@xmath2480.034 & 0.27@xmath2480.09 & 314 & 1.33 & 7.246@xmath249 & 8.35 & 6.746 & eso 344-g 019 & 0.1379 & 28 & ? + rxcj2218.8 - 0258 & 22 18 49.0 & -02 58 07 & 0.425@xmath2480.034 & 0.90@xmath2480.19 & 268 & 5.84 & 5.350@xmath249 & 9.37 & 3.246 & & 0.0899 & 2 & 7 + rxcj2223.9 - 0137 & 22 23 56.7 & -01 37 44 & 0.454@xmath2480.049 & 1.00@xmath2480.12 & 224 & 5.34 & 9.000@xmath250 & 9.92 & 3.509 & a2440 & 0.0912 & 8 & 24 + rxcj2225.8 - 0636 & 22 25 50.8 & -06 36 09 & 0.239@xmath2480.034 & 0.68@xmath2480.15 & 251 & 5.14 & 4.196@xmath249 & 5.16 & 1.787 & a2442 & 0.0897 & 2 & 12 + rxcj2226.1 - 4745 & 22 26 10.6 & -47 45 01 & 0.091@xmath2480.020 & 0.31@xmath2480.23 & 329 & 1.44 & 3.289@xmath249 & 1.77 & 0.978 & a3876 & 0.1130 & 14 & ? + rxcj2227.4 - 4852 & 22 27 28.3 & -48 52 54 & 0.126@xmath2480.022 & 0.24@xmath2480.17 & 367 & 1.65 & 3.321@xmath249 & 2.47 & 1.004 & apmcc782 & 0.0970 & 14 & 3 + rxcj2227.8 - 3034 & 22 27 52.9 & -30 34 11 & 0.522@xmath2480.043 & 0.05@xmath2480.06 & 308 & 1.09 & 3.800@xmath250 & 10.10 & 1.415 & a3880 & 0.0570 & 13 & 11 + rxcj2231.2 - 3802 & 22 31 15.1 & -38 02 39 & 0.088@xmath2480.022 & 0.88@xmath2480.44 & 233 & 1.20 & 5.000@xmath249 & 1.71 & & & & & + rxcj2234.5 - 3744 & 22 34 30.4 & -37 44 05 & 0.573@xmath2480.051 & 0.09@xmath2480.07 & 241 & 1.22 & 7.900@xmath250 & 11.20 & 11.986 & a3888 & 0.1590 & 8 & 70 + rxcj2234.8 - 3033 & 22 34 53.5 & -30 33 56 & 0.119@xmath2480.027 & 0.64@xmath2480.28 & 305 & 1.11 & 6.983@xmath249 & 2.32 & 6.182 & a3889 & 0.2512 & 8 & 19 + rxcj2235.6 + 0128 & 22 35 40.6 & + 01 28 21 & 0.510@xmath2480.080 & 0.76@xmath2480.12 & 167 & 5.81 & 4.113@xmath249 & 11.20 & 1.702 & a2457 & 0.0594 & 3 & 18 + rxcj2239.2 - 1719 & 22 39 16.4 & -17 19 46 & 0.102@xmath2480.033 & 0.76@xmath2480.33 & 110 & 2.99 & 2.559@xmath249 & 2.05 & 0.503 & a2462 & 0.0751 & 17 & 1 + rxcj2241.9 - 4235 & 22 41 56.7 & -42 35 51 & 0.085@xmath2480.022 & 0.49@xmath2480.25 & 259 & 1.66 & 5.000@xmath249 & 1.68 & & a3902 & & & + rxcj2243.0 - 2010 & 22 43 05.1 & -20 10 34 & 0.246@xmath2480.098 & 0.71@xmath2480.20 & 185 & 2.60 & 5.892@xmath249 & 4.99 & 4.085 & a2474 & 0.1385 & 1 & 1 + rxcj2244.5 - 1744 & 22 44 34.9 & -17 44 21 & 0.128@xmath2480.035 & 0.38@xmath2480.29 & 151 & 2.79 & 4.361@xmath249 & 2.60 & 1.963 & a2478 & 0.1325 & 1 & 2 + rxcj2246.1 - 3600 & 22 46 11.8 & -36 00 22 & 0.100@xmath2480.022 & 0.40@xmath2480.25 & 260 & 1.14 & 2.290@xmath249 & 1.90 & 0.374 & & 0.0673 & 1 & 2 + rxcj2247.2 + 0204 & 22 47 15.2 & + 02 04 35 & 0.168@xmath2480.064 & 0.93@xmath2480.18 & 151 & 5.84 & 5.000@xmath249 & 3.70 & & & & & + rxcj2248.7 - 4431 & 22 48 43.4 & -44 31 41 & 0.431@xmath2480.047 & 0.47@xmath2480.10 & 208 & 1.79 & 11.946@xmath249 & 8.59 & 22.736 & s1063 & 0.2520 & 7 & ? + rxcj2249.2 - 3727 & 22 49 17.6 & -37 27 26 & 0.154@xmath2480.028 & 0.23@xmath2480.20 & 253 & 1.13 & 1.388@xmath249 & 2.66 & 0.096 & s1065 & 0.0289 & 23 & 2 + rxcj2250.8 - 4521 & 22 50 48.9 & -45 21 33 & 0.104@xmath2480.026 & 0.44@xmath2480.30 & 203 & 1.88 & 1.898@xmath249 & 1.99 & 0.226 & s1067 & 0.0511 & 21,22 & 6 + rxcj2251.0 - 1624 & 22 51 00.4 & -16 24 21 & 0.255@xmath2480.052 & 0.83@xmath2480.16 & 103 & 3.16 & 5.426@xmath249 & 5.24 & 3.342 & a2496 & 0.1221 & 1 & 3 + rxcj2253.5 - 3343 & 22 53 34.2 & -33 43 21 & 0.155@xmath2480.029 & 0.29@xmath2480.16 & 195 & 1.24 & 7.133@xmath249 & 3.04 & 6.504 & a3934 & 0.2250 & 8 & 6 + rxcj2305.5 - 4513 & 23 05 34.7 & -45 13 11 & 0.220@xmath2480.094 & 0.11@xmath2480.20 & 159 & 1.69 & 5.132@xmath249 & 4.34 & 2.919 & a3970 & 0.1253 & 1 & 3 + rxcj2306.8 - 1324 & 23 06 51.7 & -13 24 59 & 0.200@xmath2480.043 & 0.51@xmath2480.24 & 139 & 3.07 & 5.000@xmath249 & 4.10 & 0.765 & & 0.0659 & 2 & 2 + rxcj2307.2 - 1513 & 23 07 15.5 & -15 13 41 & 0.242@xmath2480.063 & 1.12@xmath2480.12 & 90 & 2.78 & 4.894@xmath249 & 4.92 & 2.599 & a2533 & 0.1110 & 5 & 1 + rxcj2308.3 - 0211 & 23 08 23.1 & -02 11 32 & 0.200@xmath2480.029 & 1.11@xmath2480.22 & 354 & 4.29 & 10.221@xmath249 & 4.26 & 15.544 & a2537 & 0.2966 & 1 & 2 + rxcj2312.2 - 2129 & 23 12 16.3 & -21 29 35 & 0.335@xmath2480.051 & 0.63@xmath2480.20 & 86 & 2.03 & 4.100@xmath250 & 6.65 & 3.532 & a2554 & 0.1111 & 3 & 28 + rxcj2313.0 - 2138 & 23 13 02.2 & -21 38 07 & 0.273@xmath2480.070 & 0.41@xmath2480.15 & 88 & 2.03 & 4.700@xmath250 & 5.43 & 1.746 & a2556 & 0.0865 & 3 & 2 + rxcj2313.9 - 4243 & 23 13 58.2 & -42 43 59 & 1.185@xmath2480.085 & 0.20@xmath2480.06 & 183 & 1.85 & 5.337@xmath249 & 23.50 & 3.211 & s1101 & 0.0564 & 1 & 4 + rxcj2315.7 - 3746 & 23 15 44.2 & -37 46 53 & 0.295@xmath2480.060 & 0.54@xmath2480.17 & 128 & 1.50 & 7.709@xmath249 & 5.82 & 7.852 & am 2312 - 380/a3984 & 0.1786 & 1 & 3 + rxcj2315.7 - 0222 & 23 15 46.0 & -02 22 38 & 0.425@xmath2480.034 & 0.80@xmath2480.18 & 337 & 4.18 & 2.024@xmath249 & 8.73 & 0.268 & ngc 7556 & 0.0267 & 1 & 2 + rxcj2316.1 - 2027 & 23 16 08.0 & -20 27 19 & 0.373@xmath2480.068 & 0.48@xmath2480.15 & 95 & 2.07 & 4.589@xmath249 & 7.43 & 2.222 & a2566 & 0.0834 & 1 & 2 + rxcj2321.4 - 2312 & 23 21 25.9 & -23 12 31 & 0.164@xmath2480.042 & 0.42@xmath2480.21 & 98 & 1.96 & 5.100@xmath250 & 3.26 & 4.883 & a2580 & 0.1870 & 16 & 1 + rxcj2321.5 - 4154 & 23 21 33.0 & -41 54 04 & 0.440@xmath2480.059 & 0.40@xmath2480.12 & 143 & 2.15 & 5.178@xmath249 & 8.80 & 2.983 & a3998 & 0.0889 & 1 & 2 + rxcj2325.3 - 1207 & 23 25 20.1 & -12 07 40 & 1.016@xmath2480.059 & 0.44@xmath2480.05 & 322 & 2.50 & 9.100@xmath250 & 20.60 & 6.365 & a2597 & 0.0852 & 8 & 3 + rxcj2326.2 - 2406 & 23 26 16.0 & -24 06 13 & 0.105@xmath2480.023 & 0.73@xmath2480.26 & 185 & 1.82 & 2.888@xmath249 & 2.06 & 0.692 & apmcc895 & 0.0880 & 14 & ? + rxcj2331.1 - 3630 & 23 31 11.3 & -36 30 22 & 0.562@xmath2480.124 & 0.54@xmath2480.17 & 51 & 1.44 & 6.005@xmath249 & 11.00 & 4.279 & a4010 & 0.0954 & 8 & 8 + rxcj2336.2 - 3135 & 23 36 14.9 & -31 35 55 & 0.448@xmath2480.057 & 0.16@xmath2480.19 & 125 & 1.18 & 3.712@xmath249 & 8.67 & 1.320 & s1136 & 0.0594 & 26,27 & 2 + rxcj2337.6 + 0016 & 23 37 40.9 & + 00 16 34 & 0.169@xmath2480.033 & 0.55@xmath2480.15 & 378 & 3.85 & 8.948@xmath249 & 3.56 & 11.264 & a2631 & 0.2753 & 6 & 1 + rxcj2338.2 - 4944 & 23 38 12.7 & -49 44 43 & 0.094@xmath2480.024 & 0.75@xmath2480.28 & 175 & 1.89 & 5.000@xmath249 & 1.87 & & & & & + rxcj2341.2 - 0901 & 23 41 17.3 & -09 01 34 & 0.168@xmath2480.044 & 0.95@xmath2480.16 & 346 & 2.55 & 8.182@xmath249 & 3.41 & 9.067 & a2645 & 0.2517 & 8 & 5 + rxcj2344.2 - 0422 & 23 44 16.2 & -04 22 07 & 0.425@xmath2480.034 & 0.69@xmath2480.07 & 335 & 3.54 & 4.691@xmath249 & 8.81 & 2.340 & & 0.0786 & 1 & 2 + rxcj2344.4 - 2153 & 23 44 28.9 & -21 53 34 & 0.115@xmath2480.028 & 0.61@xmath2480.23 & 198 & 1.88 & 3.603@xmath249 & 2.27 & 1.234 & a2655 & 0.1122 & 1 & 3 + rxcj2346.7 - 1028 & 23 46 42.4 & -10 28 11 & 0.129@xmath2480.024 & 0.38@xmath2480.20 & 340 & 2.78 & 5.872@xmath249 & 2.63 & 4.093 & a2661 & 0.1911 & 3 & 2 + rxcj2347.4 - 0218 & 23 47 24.5 & -02 18 49 & 0.180@xmath2480.070 & 1.22@xmath2480.23 & 343 & 3.62 & 1.197@xmath249 & 3.08 & 0.065 & hcg 97 & 0.0221 & 1 & 2 + rxcj2347.7 - 2808 & 23 47 43.2 & -28 08 32 & 2.546@xmath2480.097 & 0.35@xmath2480.03 & 310 & 1.55 & 4.250@xmath249 & 49.90 & 1.843 & a4038 & 0.0293 & 8 & 44 + rxcj2348.2 - 2849 & 23 48 13.0 & -28 49 01 & 0.103@xmath2480.021 & 0.09@xmath2480.16 & 309 & 1.45 & 6.077@xmath249 & 2.03 & 4.408 & a4041/s1151 dc & 0.2260 & 18 & ? + rxcj2350.1 - 0339 & 23 50 09.1 & -03 39 10 & 0.125@xmath2480.026 & 1.00@xmath2480.23 & 340 & 3.65 & 4.733@xmath249 & 2.60 & 2.399 & a2664 & 0.1466 & 6 & 2 + rxcj2351.6 - 2605 & 23 51 40.5 & -26 05 09 & 0.468@xmath2480.047 & 0.48@xmath2480.07 & 308 & 1.66 & 11.318@xmath249 & 9.29 & 19.901 & a2667 & 0.2264 & 1 & 1 + rxcj2354.2 - 1024 & 23 54 13.5 & -10 24 53 & 0.473@xmath2480.057 & 0.68@xmath2480.08 & 310 & 2.92 & 3.900@xmath250 & 9.62 & 2.403 & a2670 & 0.0761 & 8 & 220 + rxcj2355.1 - 2834 & 23 55 08.6 & -28 34 27 & 0.082@xmath2480.020 & -.09@xmath2480.16 & 300 & 1.49 & 5.000@xmath249 & 1.61 & & a4054 & & & + rxcj2356.0 - 0129 & 23 56 02.4 & -01 29 43 & 0.160@xmath2480.028 & -.01@xmath2480.13 & 367 & 3.58 & 5.000@xmath249 & 3.32 & & & & & + rxcj2357.0 - 3445 & 23 57 02.4 & -34 45 45 & 1.688@xmath2480.093 & 0.30@xmath2480.05 & 255 & 1.10 & 3.500@xmath250 & 32.60 & 3.404 & a4059 & 0.0492 & 1 & 2 + rxcj2359.9 - 3928 & 23 59 56.8 & -39 28 53 & 0.258@xmath2480.047 & 0.56@xmath2480.20 & 152 & 1.33 & 4.595@xmath249 & 5.03 & 2.229 & a4068 & 0.1016 & 8 & 2 + rxcj0000.4 - 0237 & 00 00 24.7 & -02 37 30 & 0.092@xmath2480.019 & 1.00@xmath2480.37 & 372 & 3.56 & 1.460@xmath249 & 1.77 & 0.110 & & 0.0379 & 24 & 1 + rxcj0001.6 - 1540 & 00 01 39.0 & -15 40 52 & 0.116@xmath2480.021 & 1.00@xmath2480.27 & 337 & 2.24 & 3.960@xmath249 & 2.32 & 1.552 & & 0.1246 & 1 & 3 + rxcj0003.1 - 3555 & 00 03 11.1 & -35 55 21 & 0.400@xmath2480.071 & 0.29@xmath2480.14 & 142 & 1.11 & 3.046@xmath249 & 7.70 & 0.799 & a2717 & 0.0490 & 3 & 56 + rxcj0003.2 - 0605 & 00 03 12.0 & -06 05 12 & 0.218@xmath2480.027 & 0.71@xmath2480.14 & 350 & 3.16 & 8.577@xmath249 & 4.50 & 10.167 & a2697 & 0.2320 & 2 & ? + rxcj0003.8 + 0203 & 00 03 50.9 & + 02 03 45 & 0.205@xmath2480.039 & 0.83@xmath2480.12 & 362 & 3.00 & 4.185@xmath249 & 4.18 & 1.776 & a2700 & 0.0994 & 1 & 2 + rxcj0006.0 - 3443 & 00 06 04.8 & -34 43 26 & 0.301@xmath2480.041 & 0.23@xmath2480.13 & 226 & 1.16 & 5.381@xmath249 & 5.86 & 3.276 & a2721 & 0.1143 & 8 & 72 + rxcj0007.4 - 2809 & 00 07 25.3 & -28 09 21 & 0.103@xmath2480.022 & 0.43@xmath2480.27 & 313 & 1.77 & 5.000@xmath249 & 2.04 & & a2726 & & & + rxcj0011.3 - 2851 & 00 11 19.5 & -28 51 41 & 0.612@xmath2480.055 & 0.39@xmath2480.09 & 254 & 1.84 & 4.426@xmath249 & 12.10 & 2.034 & a2734 & 0.0625 & 3 & 80 + rxcj0012.9 - 0853 & 00 12 54.7 & -08 53 03 & 0.090@xmath2480.020 & 0.32@xmath2480.17 & 336 & 3.32 & 5.000@xmath249 & 1.86 & & & & & + rxcj0013.6 - 1930 & 00 13 38.2 & -19 30 08 & 0.306@xmath2480.035 & 0.48@xmath2480.11 & 313 & 2.04 & 4.699@xmath249 & 6.09 & 2.355 & a13 & 0.0949 & 1 & 4 + rxcj0014.3 - 3022 & 00 14 20.1 & -30 22 33 & 0.250@xmath2480.047 & 0.23@xmath2480.15 & 122 & 1.65 & 11.184@xmath249 & 4.96 & 19.331 & a2744 & 0.3067 & 8 & 34 + rxcj0015.5 - 2346 & 00 15 31.8 & -23 46 13 & 0.175@xmath2480.032 & 0.36@xmath2480.19 & 277 & 2.45 & 2.814@xmath249 & 3.48 & 0.647 & a14 & 0.0655 & 3 & 9 + rxcj0016.3 - 3121 & 00 16 19.8 & -31 21 55 & 0.178@xmath2480.032 & 0.00@xmath2480.14 & 296 & 1.88 & 3.292@xmath249 & 3.50 & 0.981 & a2751 & 0.0805 & 20 & 1 + rxcj0017.5 - 3511 & 00 17 34.9 & -35 11 09 & 0.166@xmath2480.033 & 0.28@xmath2480.17 & 354 & 1.31 & 3.686@xmath249 & 3.22 & 1.304 & a2755 & 0.0969 & 8 & 18 + rxcj0019.0 - 2026 & 00 19 03.9 & -20 26 17 & 0.147@xmath2480.027 & 0.42@xmath2480.17 & 322 & 1.78 & 8.314@xmath249 & 2.93 & 9.432 & s26 & 0.2773 & 1 & 1 + rxcj0020.5 - 4913 & 00 20 34.1 & -49 13 40 & 0.124@xmath2480.030 & 0.77@xmath2480.25 & 151 & 2.12 & 2.620@xmath249 & 2.44 & 0.536 & a2764 & 0.0711 & 3 & 19 + rxcj0020.7 - 2542 & 00 20 43.3 & -25 42 53 & 0.289@xmath2480.035 & 0.48@xmath2480.12 & 281 & 2.26 & 6.360@xmath249 & 5.81 & 4.921 & a22 & 0.1410 & 3 & 3 + rxcj0024.0 - 1704 & 00 24 03.6 & -17 04 32 & 0.088@xmath2480.019 & 0.63@xmath2480.24 & 314 & 2.00 & 4.956@xmath249 & 1.75 & 2.680 & a2768 & 0.1890 & 1 & 2 + rxcj0025.5 - 3302 & 00 25 32.2 & -33 02 53 & 0.460@xmath2480.045 & 0.37@xmath2480.09 & 328 & 1.69 & 3.215@xmath249 & 9.00 & 0.921 & s41 & 0.0487 & 8 & 2 + rxcj0028.6 - 2338 & 00 28 39.5 & -23 38 27 & 0.245@xmath2480.033 & 0.55@xmath2480.13 & 294 & 1.82 & 4.933@xmath249 & 4.85 & 2.650 & a42 & 0.1129 & 8 & 5 + rxcj0030.6 - 2410 & 00 30 37.6 & -24 10 53 & 0.094@xmath2480.019 & 0.60@xmath2480.22 & 314 & 1.51 & 3.927@xmath249 & 1.84 & 1.516 & a47 & 0.1382 & 3 & 1 + rxcj0033.8 - 0750 & 00 33 51.6 & -07 50 29 & 0.087@xmath2480.021 & 0.56@xmath2480.24 & 270 & 3.50 & 5.000@xmath249 & 1.80 & & a56 & & & + rxcj0034.2 - 0204 & 00 34 13.9 & -02 04 31 & 0.412@xmath2480.033 & 0.67@xmath2480.11 & 597 & 2.82 & 4.763@xmath249 & 8.38 & 2.433 & sh518 & 0.0822 & 11 & 1 + rxcj0035.6 + 0138 & 00 35 36.3 & + 01 38 20 & 0.105@xmath2480.022 & 0.75@xmath2480.23 & 291 & 2.66 & 5.000@xmath249 & 2.13 & & & & & + rxcj0037.4 - 2831 & 00 37 27.0 & -28 31 52 & 0.113@xmath2480.023 & 0.39@xmath2480.21 & 326 & 1.73 & 3.387@xmath249 & 2.22 & 1.057 & a2798 & 0.1050 & 3 & 20 + rxcj0041.8 - 0918 & 00 41 50.0 & -09 18 12 & 3.554@xmath2480.114 & 0.63@xmath2480.03 & 328 & 3.58 & 6.200@xmath250 & 74.00 & 9.810 & a85 & 0.0556 & 12 & 116 + rxcj0042.1 - 2832 & 00 42 08.8 & -28 32 12 & 0.498@xmath2480.049 & 0.28@xmath2480.07 & 325 & 1.49 & 6.768@xmath249 & 9.82 & 5.726 & a2811 & 0.1170 & 8 & 32 + rxcj0043.5 - 0443 & 00 43 34.6 & -04 43 06 & 0.119@xmath2480.018 & 0.64@xmath2480.13 & 446 & 3.63 & 5.000@xmath249 & 2.48 & & & & & + rxcj0048.6 - 2114 & 00 48 38.0 & -21 14 50 & 0.217@xmath2480.032 & 0.09@xmath2480.13 & 339 & 1.52 & 2.759@xmath249 & 4.20 & 0.614 & a2824 & 0.0581 & 17 & 7 + rxcj0049.3 - 2931 & 00 49 23.8 & -29 31 27 & 0.265@xmath2480.042 & 0.40@xmath2480.13 & 323 & 1.80 & 5.172@xmath249 & 5.25 & 2.974 & s84 & 0.1150 & 8 & 20 + rxcj0051.1 - 4833 & 00 51 11.2 & -48 33 35 & 0.123@xmath2480.027 & 0.63@xmath2480.24 & 204 & 2.12 & 5.646@xmath249 & 2.46 & 3.684 & a2830 & 0.1873 & 1 & 2 + rxcj0051.3 - 2831 & 00 51 19.2 & -28 31 10 & 0.127@xmath2480.023 & 0.40@xmath2480.16 & 324 & 1.63 & 3.750@xmath249 & 2.49 & 1.359 & a2829 & 0.1125 & 13 & 6 + rxcj0056.0 - 3732 & 00 56 01.0 & -37 32 45 & 0.425@xmath2480.034 & 0.55@xmath2480.21 & 351 & 2.59 & 8.538@xmath249 & 8.65 & 10.102 & & 0.1663 & 2 & 11 + rxcj0056.3 - 0112 & 00 56 18.3 & -01 12 53 & 1.644@xmath2480.087 & 0.74@xmath2480.06 & 321 & 3.10 & 5.100@xmath250 & 33.70 & 2.855 & a119 & 0.0444 & 17 & 23 + rxcj0058.4 - 1425 & 00 58 28.2 & -14 25 39 & 0.108@xmath2480.018 & 0.17@xmath2480.15 & 492 & 1.87 & 3.105@xmath249 & 2.12 & 0.840 & a123 & 0.0957 & 5 & 2 + rxcj0102.7 - 2152 & 01 02 42.3 & -21 52 46 & 0.970@xmath2480.065 & 0.33@xmath2480.06 & 268 & 1.48 & 3.800@xmath250 & 18.90 & 2.574 & a133 & 0.0562 & 8 & 9 + rxcj0104.5 - 2400 & 01 04 35.0 & -24 00 00 & 0.119@xmath2480.021 & 0.64@xmath2480.19 & 335 & 1.61 & 4.672@xmath249 & 2.34 & 2.322 & a140 & 0.1520 & 8 & 12 + rxcj0105.5 - 2439 & 01 05 35.4 & -24 39 33 & 0.196@xmath2480.031 & 0.32@xmath2480.12 & 326 & 1.61 & 8.018@xmath249 & 3.88 & 8.635 & a141 & 0.2300 & 8 & 1 + rxcj0106.8 - 0229 & 01 06 51.5 & -02 29 18 & 0.152@xmath2480.032 & 0.91@xmath2480.23 & 423 & 4.10 & 6.386@xmath249 & 3.21 & 4.972 & a145 & 0.1909 & 5 & 1 + rxcj0107.8 - 3643 & 01 07 49.4 & -36 43 44 & 0.180@xmath2480.024 & 0.82@xmath2480.16 & 378 & 1.94 & 4.642@xmath249 & 3.57 & 2.285 & a2871 & 0.1221 & 2 & 11 + rxcj0108.2 + 0210 & 01 08 12.8 & + 02 10 48 & 0.230@xmath2480.037 & 0.59@xmath2480.15 & 415 & 3.01 & 2.342@xmath249 & 4.60 & 0.398 & a147 & 0.0447 & 3 & 11 + rxcj0108.5 - 4021 & 01 08 33.2 & -40 21 00 & 0.150@xmath2480.035 & 0.40@xmath2480.23 & 377 & 3.04 & 4.898@xmath249 & 3.07 & 2.610 & a2874 & 0.1408 & 13 & 6 + rxcj0108.8 - 1524 & 01 08 49.3 & -15 24 30 & 0.358@xmath2480.035 & 0.29@xmath2480.08 & 446 & 1.69 & 3.149@xmath249 & 7.00 & 0.872 & a151 & 0.0537 & 8 & 37 + rxcj0108.9 - 1537 & 01 08 55.5 & -15 37 42 & 0.189@xmath2480.029 & 0.46@xmath2480.09 & 437 & 1.69 & 3.954@xmath249 & 3.72 & 1.541 & & 0.0981 & 4,19 & 9 + rxcj0110.0 - 4555 & 01 10 00.7 & -45 55 24 & 0.756@xmath2480.060 & 0.26@xmath2480.08 & 301 & 2.10 & 3.500@xmath250 & 15.00 & 0.404 & a2877 & 0.0250 & 8 & 109 + rxcj0113.9 - 3145 & 01 13 54.2 & -31 45 07 & 0.080@xmath2480.017 & 0.58@xmath2480.24 & 333 & 2.40 & 1.000@xmath249 & 1.21 & 0.018 & s141 & 0.0184 & 8 & 18 + rxcj0114.7 - 2118 & 01 14 47.3 & -21 18 02 & 0.082@xmath2480.018 & 0.56@xmath2480.24 & 346 & 1.49 & 5.000@xmath249 & 1.61 & & & & & + rxcj0114.9 + 0024 & 01 14 57.5 & + 00 24 20 & 0.520@xmath2480.048 & 0.47@xmath2480.09 & 406 & 3.32 & 2.600@xmath250 & 10.60 & 1.043 & a168 & 0.0477 & 1 & 2 + rxcj0116.1 - 1555 & 01 16 11.8 & -15 55 24 & 0.130@xmath2480.023 & 0.23@xmath2480.17 & 394 & 1.69 & 1.862@xmath249 & 2.46 & 0.214 & scg 16 & 0.0448 & 1 & 2 + rxcj0118.1 - 2658 & 01 18 10.2 & -26 58 18 & 0.193@xmath2480.027 & 0.54@xmath2480.14 & 366 & 1.47 & 7.987@xmath249 & 3.81 & 8.554 & a2895 & 0.2310 & 1 & 1 + rxcj0120.9 - 1351 & 01 20 57.9 & -13 51 19 & 0.425@xmath2480.034 & 0.30@xmath2480.07 & 344 & 1.73 & 3.240@xmath249 & 8.33 & 0.939 & cid 10 & 0.0511 & 1 & 3 + rxcj0121.8 - 2022 & 01 21 48.1 & -20 22 34 & 0.110@xmath2480.020 & 0.70@xmath2480.28 & 465 & 1.45 & 5.000@xmath249 & 2.16 & & a2902 & & & + rxcj0122.2 - 2131 & 01 22 12.4 & -21 31 04 & 0.097@xmath2480.020 & 0.05@xmath2480.17 & 494 & 1.39 & 5.000@xmath249 & 1.90 & & a185 & & & + rxcj0122.9 - 1245 & 01 22 55.1 & -12 45 50 & 0.096@xmath2480.018 & 0.87@xmath2480.25 & 391 & 1.68 & 3.598@xmath249 & 1.88 & 1.229 & a188 & 0.1230 & 3 & 2 + rxcj0125.4 + 0145 & 01 25 29.9 & + 01 45 46 & 0.425@xmath2480.034 & 0.77@xmath2480.15 & 413 & 3.08 & 1.446@xmath249 & 8.03 & 0.107 & ngc 533 & 0.0176 & 2 & 4 + rxcj0125.6 - 0125 & 01 25 38.0 & -01 25 03 & 0.610@xmath2480.086 & 0.84@xmath2480.14 & 431 & 3.56 & 1.900@xmath250 & 12.20 & 0.176 & a194 & 0.0183 & 8 & 81 + rxcj0126.0 - 3758 & 01 26 04.6 & -37 58 36 & 0.105@xmath2480.019 & 0.76@xmath2480.23 & 351 & 1.68 & 2.703@xmath249 & 2.04 & 0.581 & a2911 & 0.0810 & 4 & 31 + rxcj0126.7 - 1810 & 01 26 44.4 & -18 10 20 & 0.103@xmath2480.025 & 0.32@xmath2480.21 & 328 & 1.62 & 5.000@xmath249 & 2.03 & & a197 & & & + rxcj0127.2 - 1746 & 01 27 13.9 & -17 46 17 & 0.109@xmath2480.021 & 0.28@xmath2480.19 & 351 & 1.52 & 4.358@xmath249 & 2.14 & 1.960 & a199 & 0.1459 & 1 & 2 + rxcj0131.7 + 0033 & 01 31 42.4 & + 00 33 43 & 0.088@xmath2480.017 & 0.26@xmath2480.18 & 432 & 2.92 & 2.535@xmath249 & 1.77 & 0.490 & a208 & 0.0798 & 3 & ? + rxcj0131.8 - 1336 & 01 31 53.5 & -13 36 27 & 0.277@xmath2480.029 & 0.33@xmath2480.10 & 453 & 2.27 & 8.513@xmath249 & 5.59 & 9.982 & a209 & 0.2060 & 3 & 2 + rxcj0137.2 - 0912 & 01 37 15.3 & -09 12 10 & 0.425@xmath2480.034 & 0.68@xmath2480.08 & 444 & 2.75 & 2.672@xmath249 & 8.50 & 0.564 & & 0.0392 & 1 & 3 + rxcj0137.4 - 1259 & 01 37 29.2 & -12 59 10 & 0.088@xmath2480.016 & 0.55@xmath2480.18 & 458 & 2.45 & 5.503@xmath249 & 1.78 & 3.462 & a222 & 0.2134 & 8 & 6 + rxcj0137.9 - 1248 & 01 37 56.4 & -12 48 01 & 0.115@xmath2480.018 & 0.60@xmath2480.20 & 454 & 1.84 & 5.935@xmath249 & 2.28 & 4.160 & a223 & 0.2070 & 8 & 4 + rxcj0139.1 - 1915 & 01 39 10.1 & -19 15 34 & 0.083@xmath2480.015 & 0.28@xmath2480.20 & 479 & 1.45 & 5.000@xmath249 & 1.63 & & & & & + rxcj0139.9 - 0555 & 01 39 59.7 & -05 55 23 & 0.084@xmath2480.018 & 0.43@xmath2480.20 & 441 & 2.85 & 5.000@xmath249 & 1.71 & & & & & + rxcj0143.4 - 4614 & 01 43 29.8 & -46 14 20 & 0.082@xmath2480.017 & 0.41@xmath2480.25 & 455 & 2.30 & 5.000@xmath249 & 1.65 & & a2937 & & & + rxcj0144.6 - 2213 & 01 44 41.8 & -22 13 43 & 0.097@xmath2480.015 & 0.25@xmath2480.13 & 494 & 1.18 & 6.991@xmath249 & 1.90 & 6.196 & a2938 & 0.2781 & 1 & 1 + rxcj0148.2 - 3155 & 01 48 14.8 & -31 55 08 & 0.105@xmath2480.016 & 0.37@xmath2480.17 & 521 & 1.53 & 3.400@xmath250 & 2.05 & & a2943 & & & + rxcj0148.3 - 0406 & 01 48 22.2 & -04 06 13 & 0.104@xmath2480.024 & 0.25@xmath2480.19 & 255 & 2.59 & 5.000@xmath249 & 2.10 & & & & & + rxcj0152.5 - 2853 & 01 52 32.8 & -28 53 32 & 0.084@xmath2480.016 & 0.16@xmath2480.16 & 391 & 1.51 & 5.000@xmath249 & 1.65 & & & & & + rxcj0152.7 + 0100 & 01 52 43.6 & + 01 00 58 & 0.151@xmath2480.022 & 0.42@xmath2480.13 & 407 & 2.84 & 7.309@xmath249 & 3.09 & 6.900 & a267 dc & 0.2300 & 9 & 1 + rxcj0152.9 - 1345 & 01 52 59.8 & -13 45 07 & 0.425@xmath2480.034 & 0.25@xmath2480.13 & 462 & 1.69 & 1.000@xmath249 & 6.30 & 0.009 & ngc 720 & 0.0057 & 15 & 1 + rxcj0153.5 - 0118 & 01 53 32.0 & -01 18 44 & 0.123@xmath2480.030 & 0.54@xmath2480.17 & 433 & 2.66 & 7.025@xmath249 & 2.50 & 6.277 & & 0.2438 & 1 & 2 + rxcj0157.5 - 0549 & 01 57 30.5 & -05 49 32 & 0.164@xmath2480.024 & 0.55@xmath2480.19 & 414 & 2.31 & 4.681@xmath249 & 3.29 & 2.332 & a281 dc & 0.1285 & 2 & 2 + rxcj0158.4 - 0146 & 01 58 28.4 & -01 46 51 & 0.116@xmath2480.020 & 0.48@xmath2480.17 & 406 & 2.55 & 4.948@xmath249 & 2.34 & 2.672 & a286 & 0.1632 & 1 & 1 + rxcj0201.7 - 0211 & 02 01 44.3 & -02 11 58 & 0.209@xmath2480.024 & 0.52@xmath2480.11 & 406 & 2.57 & 7.327@xmath249 & 4.25 & 6.941 & a291 & 0.1965 & 9 & 1 + rxcj0202.3 - 4447 & 02 02 18.8 & -44 47 37 & 0.124@xmath2480.020 & 0.31@xmath2480.18 & 465 & 2.70 & 2.359@xmath249 & 2.46 & 0.405 & & 0.0616 & 1 & 2 + rxcj0202.3 - 0107 & 02 02 20.6 & -01 07 06 & 0.191@xmath2480.026 & 0.22@xmath2480.13 & 405 & 2.58 & 2.103@xmath249 & 3.76 & 0.298 & a295 & 0.0428 & 5 & 3 + rxcj0206.4 - 1453 & 02 06 29.9 & -14 53 37 & 0.149@xmath2480.027 & 0.52@xmath2480.16 & 274 & 2.45 & 5.192@xmath249 & 3.00 & 3.003 & a305 & 0.1529 & 1 & 2 + rxcj0208.0 - 1537 & 02 08 00.2 & -15 37 05 & 0.085@xmath2480.019 & 0.92@xmath2480.24 & 286 & 2.38 & 5.000@xmath249 & 1.71 & & & & & + rxcj0211.4 - 4017 & 02 11 25.2 & -40 17 09 & 0.166@xmath2480.019 & 0.54@xmath2480.14 & 583 & 1.43 & 3.859@xmath249 & 3.24 & 1.457 & a2984 & 0.1022 & 8 & 6 + rxcj0212.8 - 4707 & 02 12 53.9 & -47 07 58 & 0.197@xmath2480.026 & 0.35@xmath2480.18 & 517 & 2.12 & 4.598@xmath249 & 3.93 & 2.232 & a2988 & 0.1150 & 25 & ? + rxcj0213.9 - 0253 & 02 13 56.6 & -02 53 53 & 0.101@xmath2480.022 & 0.46@xmath2480.20 & 350 & 2.16 & 5.000@xmath249 & 2.02 & & & & & + rxcj0214.6 - 0433 & 02 14 41.1 & -04 33 48 & 0.106@xmath2480.021 & 0.68@xmath2480.30 & 357 & 2.21 & 4.181@xmath249 & 2.12 & 1.772 & a329 & 0.1393 & 1 & 2 + rxcj0216.3 - 4816 & 02 16 19.1 & -48 16 23 & 0.140@xmath2480.019 & 0.85@xmath2480.23 & 542 & 2.95 & 5.573@xmath249 & 2.86 & 3.568 & a2998 & 0.1709 & 1 & 1 + rxcj0216.7 - 4749 & 02 16 43.3 & -47 49 31 & 0.180@xmath2480.030 & 0.89@xmath2480.28 & 523 & 2.95 & 2.809@xmath249 & 3.63 & 0.644 & s239 & 0.0640 & 23 & 1 + rxcj0218.3 - 3141 & 02 18 19.6 & -31 41 34 & 0.099@xmath2480.016 & 0.28@xmath2480.14 & 486 & 1.87 & 5.000@xmath249 & 1.96 & & & & & + rxcj0219.1 - 1724 & 02 19 08.3 & -17 24 40 & 0.085@xmath2480.016 & 0.82@xmath2480.23 & 390 & 2.99 & 5.000@xmath249 & 1.74 & & & & & + rxcj0220.9 - 3829 & 02 20 56.8 & -38 29 02 & 0.425@xmath2480.034 & 0.74@xmath2480.17 & 558 & 1.85 & 10.965@xmath249 & 8.48 & 18.532 & & 0.2287 & 2 & 6 + rxcj0225.1 - 2928 & 02 25 10.2 & -29 28 23 & 0.425@xmath2480.034 & 0.63@xmath2480.15 & 445 & 1.70 & 3.710@xmath249 & 8.35 & 1.327 & & 0.0607 & 1 & 2 + rxcj0225.8 - 4154 & 02 25 53.7 & -41 54 19 & 0.258@xmath2480.029 & 0.29@xmath2480.10 & 460 & 2.14 & 8.693@xmath249 & 5.19 & 10.503 & a3017 & 0.2195 & 1 & 1 + rxcj0227.2 - 2851 & 02 27 12.6 & -28 51 18 & 0.071@xmath2480.018 & 0.59@xmath2480.21 & 343 & 1.55 & 5.003@xmath249 & 1.40 & 2.743 & edcc651 & 0.2138 & 2 & ? + rxcj0229.3 - 3332 & 02 29 22.3 & -33 32 13 & 0.198@xmath2480.026 & 0.34@xmath2480.12 & 471 & 2.07 & 3.352@xmath249 & 3.92 & 1.029 & apmcc269 & 0.0779 & 1 & 2 + rxcj0229.9 - 1316 & 02 29 56.4 & -13 16 12 & 0.127@xmath2480.026 & 0.25@xmath2480.20 & 312 & 2.04 & 5.000@xmath249 & 2.53 & & & & & + rxcj0230.8 - 3305 & 02 30 51.2 & -33 05 49 & 0.166@xmath2480.026 & 0.71@xmath2480.24 & 456 & 1.96 & 3.116@xmath249 & 3.27 & 0.848 & a3027 & 0.0774 & 13 & 4 + rxcj0231.7 - 0451 & 02 31 47.3 & -04 51 21 & 0.124@xmath2480.033 & 0.55@xmath2480.24 & 125 & 2.55 & 5.619@xmath249 & 2.51 & 3.640 & a362 & 0.1843 & 1 & 3 + rxcj0231.9 + 0114 & 02 31 57.5 & + 01 14 38 & 0.425@xmath2480.034 & 0.97@xmath2480.26 & 227 & 2.56 & 1.714@xmath249 & 8.12 & 0.171 & & 0.0221 & 1 & 2 + rxcj0232.2 - 4420 & 02 32 17.2 & -44 20 43 & 0.425@xmath2480.034 & 0.14@xmath2480.09 & 369 & 2.61 & 13.183@xmath249 & 8.66 & 28.933 & & 0.2836 & 1 & 2 + rxcj0236.6 - 1923 & 02 36 39.6 & -19 23 05 & 0.171@xmath2480.025 & 0.57@xmath2480.19 & 349 & 2.65 & 4.100@xmath250 & 3.45 & 1.179 & a367 & 0.0891 & 17 & 3 + rxcj0237.4 - 2630 & 02 37 29.0 & -26 30 13 & 0.156@xmath2480.024 & 0.43@xmath2480.16 & 325 & 1.56 & 7.042@xmath249 & 3.08 & 6.305 & a368 & 0.2200 & 2 & 1 + rxcj0241.3 - 2839 & 02 41 20.9 & -28 39 27 & 0.130@xmath2480.021 & 0.10@xmath2480.13 & 423 & 1.56 & 4.300@xmath250 & 2.55 & 5.942 & a3041 & 0.2323 & 1 & 3 + rxcj0244.1 - 2611 & 02 44 06.3 & -26 11 07 & 0.152@xmath2480.027 & 0.79@xmath2480.25 & 286 & 1.74 & 4.726@xmath249 & 3.00 & 2.389 & & 0.1362 & 4,19 & 6 + rxcj0245.2 - 4627 & 02 45 12.6 & -46 27 14 & 0.114@xmath2480.017 & 0.66@xmath2480.22 & 539 & 3.15 & 2.986@xmath249 & 2.32 & 0.757 & a3047 & 0.0868 & 1 & 1 + rxcj0248.0 - 0332 & 02 48 02.6 & -03 32 09 & 0.229@xmath2480.042 & 0.49@xmath2480.16 & 224 & 4.12 & 7.521@xmath249 & 4.85 & 7.396 & a383 & 0.1899 & 1 & 1 + rxcj0248.2 - 0216 & 02 48 16.3 & -02 16 44 & 0.220@xmath2480.033 & 0.66@xmath2480.22 & 117 & 3.65 & 8.748@xmath249 & 4.60 & 10.743 & a384 & 0.2360 & 2 & ? + rxcj0249.6 - 3111 & 02 49 36.9 & -31 11 19 & 0.469@xmath2480.040 & 1.31@xmath2480.27 & 419 & 1.80 & 1.837@xmath249 & 8.89 & 0.206 & s301 & 0.0232 & 8 & 34 + rxcj0251.4 - 2456 & 02 51 28.8 & -24 56 58 & 0.144@xmath2480.023 & 0.04@xmath2480.12 & 365 & 1.90 & 4.500@xmath250 & 2.85 & 1.511 & a389 & 0.1111 & 8 & 6 + rxcj0251.7 - 4109 & 02 51 42.3 & -41 09 00 & 0.102@xmath2480.018 & 0.00@xmath2480.17 & 496 & 1.86 & 5.000@xmath249 & 2.02 & & & & & + rxcj0252.8 - 0116 & 02 52 50.1 & -01 16 28 & 0.425@xmath2480.034 & 1.34@xmath2480.23 & 135 & 5.25 & 1.833@xmath249 & 8.84 & 0.205 & ngc 1132 & 0.0232 & 10 & 1 + rxcj0258.2 - 2105 & 02 58 15.0 & -21 05 29 & 0.114@xmath2480.023 & 0.57@xmath2480.21 & 287 & 1.92 & 3.859@xmath249 & 2.25 & 1.458 & a3073 & 0.1226 & 1 & 1 + rxcj0303.2 - 2735 & 03 03 14.8 & -27 35 59 & 0.122@xmath2480.028 & 0.25@xmath2480.20 & 266 & 1.55 & 5.000@xmath249 & 2.40 & & a3082 & & & + rxcj0303.3 + 0155 & 03 03 21.3 & + 01 55 35 & 0.219@xmath2480.033 & 0.83@xmath2480.15 & 222 & 7.82 & 6.421@xmath249 & 5.08 & 5.062 & a409 & 0.1530 & 9 & 1 + rxcj0304.0 - 3656 & 03 04 06.0 & -36 56 32 & 0.151@xmath2480.023 & 0.76@xmath2480.23 & 384 & 1.95 & 6.958@xmath249 & 3.01 & 6.120 & a3084 & 0.2192 & 2 & ? + rxcj0307.0 - 2840 & 03 07 04.0 & -28 40 15 & 0.197@xmath2480.034 & 0.38@xmath2480.15 & 237 & 1.36 & 8.674@xmath249 & 3.88 & 10.434 & a3088 & 0.2534 & 7 & ? + rxcj0311.4 - 2654 & 03 11 25.3 & -26 54 00 & 0.154@xmath2480.078 & 0.52@xmath2480.22 & 210 & 1.56 & 2.722@xmath249 & 2.98 & 0.593 & a3094/edcc736 dc & 0.0677 & 3 & 67 + rxcj0313.6 - 3817 & 03 13 38.0 & -38 17 55 & 0.119@xmath2480.023 & 0.20@xmath2480.22 & 321 & 2.03 & 2.905@xmath249 & 2.34 & 0.704 & a3098 & 0.0833 & 3 & 6 + rxcj0314.3 - 4525 & 03 14 19.6 & -45 25 24 & 0.401@xmath2480.038 & 0.45@xmath2480.10 & 313 & 3.57 & 4.252@xmath249 & 8.31 & 1.845 & a3104 & 0.0718 & 2 & 4 + rxcj0317.7 - 4848 & 03 17 46.4 & -48 48 42 & 0.097@xmath2480.016 & 0.56@xmath2480.23 & 487 & 2.06 & 5.000@xmath249 & 1.93 & & a3113 & & & + rxcj0317.9 - 4414 & 03 17 58.2 & -44 14 16 & 1.447@xmath2480.056 & 0.30@xmath2480.04 & 481 & 2.53 & 4.100@xmath250 & 29.10 & 7.130 & a3112 & 0.0754 & 1 & 2 + Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: correlation and interference distinguish quantum from classical mechanics . the former is manifest in the measurement of many - body coincidences predicted by a quantum joint probability density function ( pdf ) , which are not observed classically @xcite . the latter is most familiar as a probability density function for an outcome that can be achieved in at least two indistinguishable ways . however , interference can also be generated by superposing many - body states in indistinguishable ways @xcite . experimental confirmation of quantum correlation has involved photons @xcite , atoms @xcite , and josephson phase qubits @xcite . however , there is little experimental evidence of correlated interference between massive particles . the boundary between the classical and quantum regimes continues to be an area of interest both in attempts to explain why quantum effects are difficult to measure on a macroscopic scale @xcite and in attempts to measure quantum effects on a macroscopic object @xcite . reflection measurement methods used in an attempt to extend this boundary involve light interacting with stationary @xcite and oscillating mirrors @xcite and with microwaves interacting with an oscillating mirror @xcite . reflection of a single photon from an oscillating mirror @xcite and using a bose einstein condensate as a mirror from which photons reflect @xcite have been studied . to extend this boundary to larger masses , we model correlation of a mesoscopic / macroscopic ` mirror ' with a non - zero rest mass microscopic ` particle ' after elastic reflection . motion is in free space along one dimension and all states are unbound . measurements of particle reflection , but not associated with correlated interference , have involved mirrors that reflect atoms @xcite and bose - einstein condensates @xcite , atoms reflecting from a solid surface @xcite , neutrons @xcite and atoms @xcite reflecting from vibrating mirrors , and atoms reflecting from a switchable mirror @xcite . if the particle and mirror are initially in eigenstates of energy then the following two states interfere : the particle - mirror state before the particle reflects from the mirror and the particle - mirror state after reflection ( interference is a consequence of the incident and reflected states being indistinguishable for a measurement of position ) . this is similar to the standing wave interference of a harmonic electromagnetic wave reflecting from a stationary mirror @xcite . classically , however , the incident and reflected waves interfere while the mirror experiences only a continuous force due to radiation pressure . the two - body quantum analogy involves solving the schrdinger equation with an interaction potential modeling the reflection . interference is expected between the incident and reflected particle substates _ along with _ interference of the mirror substates which have and have not reflected the particle . their correlation is perhaps not expected , being a consequence of the solution to the schrdinger equation from which a joint probability density function is constructed . this joint pdf then describes the correlations in the two - body interference which are manifest as coincidence rates , e.g. a correlation in the simultaneous measurement of particle _ and _ mirror positions . asynchronous measurement in such a system has also been discussed @xcite . any experimental realization of such an interferometer will involve wavegroups . while the coherence length for the particle currently can be as large as @xmath0 m for atoms released from a bose - einstein condensate , that for a mesoscopic / macroscopic mirror will be many orders of magnitude smaller . however , the coherence length of the two - body state involves that of both the particle and mirror substates . the two - body interference in reflection from a mirror is then limited spatially ( to the overlap region ) and temporally ( to the time of overlap ) in a manner similar to that of a pulse of light reflecting from a moving mirror . classical interference in reflection is more robust when the pulse of light reflects from a moving thin film for two main reasons : first the reflected pulses from each thin film surface maintain overlap as they both travel in the backward direction and second they have the same doppler shift ( since they reflect from interfaces moving at the same speed ) . interference is then limited by a pulse coherence length which at its minimum is of order the film thickness . to achieve such robust two - body quantum interference we consider the analogous two - body situation , an example of which is a neutron reflecting from each of the two surfaces of a moving aluminum ` slab . ' it is shown below that this two - body interference has the same ` non - local ' characteristics as does a pulse of light reflecting from a moving thin film . treating both the particle and reflector as quantum objects yields the expected results of a one - body treatment ( analogous to interference for a pulse of light reflecting from a moving thin film ) but only as a special case . in addition , this treatment predicts novel reflection regimes depending on the coherence lengths of the particle and reflector . quantum two - body reflection differs from that classically . for example , the classical kinematics , associated with reflection of a wavegroup , involves accelerated mirror motion while the kinematics associated with the expectation value of the position of either the particle or mirror , during the interaction , depends on the two - body wavegroup parameters such as coherence lengths . this is discussed in section [ sec : summary ] . also , under conditions of destructive interference , the particle and slab will never simultaneously be observed . classically , only the pulse of light will never be observed . a simultaneous measurement , however , is difficult to perform . a much simpler method involves only measuring the reflected particle . such a one - body measurement can reveal the quantum state of the reflector due to the effects of the reflector s coherence length on the two - body pdf , from which the one - body marginal pdf is derived . this manifestation of the macroscopic quantum state of the reflector on the interference measured only on the particle may be a practical way to verify that the mesoscopic / macroscopic reflector is indeed in a quantum state and thereby extend the quantum - classical boundary to larger masses . finding examples of mesoscopic / macroscopic quantum phenomena has been a topic of interest since early in the development of quantum mechanics , particularly with regard to validation issues . this continues to be a subject of interest . consider next some of the arguments used to explain why such effects are not observed with mesoscopic / macroscopic objects , in relation to the two - body treatment of particle reflection : ( 1 ) in an introductory quantum course , the lack of such evidence is often justified via the double slit experiment , where the fringe spacing becomes imperceptible as the mass increases . in the comments after eqn . [ eq : fringespacing ] , an explanation is given as to why the mirror fringe spacing does not shrink as its mass increases . ( 2 ) another concern is if quantum correlation is indeed generated . the initial particle - mirror separable system becomes correlated in reflection as a consequence of conservation of energy and momentum . the discussion in section [ sec : summary ] demonstrates why interferometric quantum correlation is a requisite in satisfying these conservation laws . ( 3 ) the difference in wavevectors of the interfering mesoscopic / macroscopic mirror substates , due to reflecting a microscopic particle , is much smaller than the spread in wavevectors of which the mirror wavegroup is composed , due to thermal motion . this effect is shown to not destroy interference in section [ sec : summary ] . ( 4 ) more recently , decoherence of the macroscopic superposition via interaction with the environment has been used to explain the lack of interference . mitigation of this and other decoherence mechanisms are discussed in section [ sec : decoher ] . this work examines the assumptions of these arguments in the context of reflection of a microscopic particle from a mesoscopic / macroscopic mirror . a synopsis of this work is as follows : first the schrdinger equation is solved for the two - body energy eigenstates in reflection from a moving mirror . wavegroups formed from these solutions are then used to illustrate interference in reflection with an emphasis on variations in coherence lengths of the mirror . coherence transfer from the particle to the mirror is also demonstrated and related to classical reflection . two models are next developed for two - body reflection of a particle from a slab , whose classical analog is that of reflection of a pulse of light from a moving thin film . the effects of the coherence lengths of both the particle and slab are then discussed . measurements of either the particle but not the slab and the slab but not the particle are then considered . finally , decoherence issues , as well as methods to measure a mesoscopic / macroscopic reflector in a superposition state , are addressed . the goal of this paper is to illustrate the unique interferometric properties of correlation interferometry in reflection . these properties are most easily understood using the example of a microscopic particle reflecting from a mesoscopic / macroscopic mirror . however , the focus on this example is not intended as a practical experimental proposal . observation of such correlated interference will most likely first occur with a microscopic particle reflecting from a `` microscopic '' mirror . before reflection , the solution to the schrdinger equation for the non - interacting particle - mirror state is @xmath1 , \label{eq : psiseparable}\end{aligned}\ ] ] where @xmath2 and @xmath3 are the particle and mirror positions along the x - axis while @xmath4 and @xmath5 are the wavevectors for the particle and mirror respectively ; @xmath6 and @xmath7 with masses @xmath8 , @xmath9 , and initial velocities @xmath10 and @xmath11 , respectively . a wavegroup constructed from such uncorrelated particle - mirror states then leads to predictions about the probability of simultaneously finding the particle at @xmath2 and mirror at @xmath3 . the particle - mirror interaction is modeled as a moving delta function potential where reflection is assumed to occur at the center of mass ( cm ) of the mirror with the schrdinger equation given by @xmath12 \notag + i\partial_{t})\psi=0 . \label{eq : schreqn}\end{aligned}\ ] ] where square brackets are used to indicate the argument of a function . the mirror reflectivity , related to @xmath13 , goes to infinity for a lossless mirror . the solution yields an energy eigenstates for the particle - mirror interaction ( ` harmonic wavefunctions ' ) , for which neither the particle nor the mirror is localized . a separable solution to the two - body schrdinger equation results from a transformation to the center of mass ( cm ) and relative ( rel ) system ( not to be confused with the cm of the particle or mirror ) . this does not change the total energy , @xmath14 . the transformed schrdinger equation becomes @xmath15 \notag + i\partial_{t})\psi[x_{cm},x_{rel},t]=0 \label{eq : scheqncm}\end{aligned}\ ] ] where @xmath16 , @xmath17 , @xmath18 , and @xmath19 . using @xmath20=\psi_{cm}\psi_{rel } \notag \\ = e^{-i e_{cm } t/\hbar } u[x_{cm } ] e^{-i e_{rel } t/\hbar}u[x_{rel } ] \notag , \label{eq : schtot}\end{aligned}\ ] ] reduces the schrdinger equation to two ordinary differential equations : @xmath21}{dx_{cm}^{2 } } = e_{cm } u[x_{cm } ] \label{eq : scheqode1}\ ] ] @xmath22}{dx_{rel}^{2}}+\beta \delta[x_{rel } ] = e_{rel } u[x_{rel } ] . \label{eq : scheqode2}\ ] ] the particle - mirror solution must vanish at @xmath23 to satisfy the boundary condition at the mirror and not exist for @xmath24 ( or @xmath25 ) since the particle can not move through the mirror ( for the uncorrelated incident state , however , there is no interaction and the particle can move past the mirror ) . in this transformed system , a solution is constructed from the superposition of incident and `` reflected '' wavefunctions , @xmath26 , \label{eq : superposition}\ ] ] where @xmath27 $ ] is the unit step function . the only difference between the arguments of the two exponentials is the sign of the relative wavevector @xmath28 which , due to reflection in the relative coordinate , is negative . that is , @xmath29 where the initial velocities must allow reflection to occur . the solution to eqn . [ eq : scheqode1 ] is given by @xmath30 the complete solution is then @xmath31=\psi_{cm}\psi_{rel}$ ] . in this separable system the probability density is @xmath32 . \label{eq : separableinterference}\end{aligned}\ ] ] the cm - rel transformation does not give a solution for a physically realizable system . there exists no particle with a reduced mass , for example . nevertheless , its utility lies in inverting the solution from cm and relative substates into particle and mirror substates . this yields a correlated particle - mirror state . an example of this procedure is found in the solution to the hydrogen atom where the schrdinger equation is first transformed from the laboratory to the cm - relative coordinates yielding uncorrelated substates . however , expressing this result in the electron and proton substates reveals a correlation between them @xcite . this change of partition is accomplished by the following substitutions in the separable solution given in eqn . [ eq : separableinterference ] : @xmath33 , @xmath34 , @xmath19 , @xmath18 , @xmath35 , and @xmath36 . these constraints transform @xmath37 in eqn . [ eq : superposition ] into the non - interacting particle - mirror solution of eqn . [ eq : psiseparable ] . transformation of the energy component of each reflected substate , given by @xmath38 with @xmath39 for the reflected particle and @xmath40 with @xmath41 for the mirror , is consistent with that of reflection classically . with this change in coordinates eqn . [ eq : separableinterference ] becomes @xmath42 . \label{eq : entangledinterference}\end{aligned}\ ] ] this is similar to gottfried s result for the interference obtained in the correlation between two particles produced in a momentum - conserving decay after each has traversed separate double slits @xcite . note also that the interference given in eqn . [ eq : entangledinterference ] couples particle and mirror variables , illustrating how many - body states interfere with themselves rather than each substate interfering only with itself @xcite . measurement of such a prediction involves preparation of the initial state , adjusting the instruments to measure _ both _ positions @xmath2 and @xmath3 simultaneously , making the measurement , and then repeating this procedure over an ensemble to build a distribution . to illustrate how this leads to a measurement of the mirror fringe spacing , first the apparatus is set to always measure the particle at fixed position @xmath2 while the mirrors cm is measured at different positions @xmath3 for different members of the ensemble . using the approximation @xmath43 in eqn . [ eq : entangledinterference ] leads to interference fringes _ for the mesoscopic / macroscopic mirrors cm _ which vary from maximum to minimum through a distance @xmath44 similarly , by always measuring the mirror at fixed @xmath3 while varying the measurement position of the particle @xmath2 for different members of the ensemble , this approximation leads to interference fringes _ for the microscopic particle s cm _ which vary from maximum to minimum through a distance @xmath45 . for a static mirror both the mirror and particle fringes are spaced at half the debroglie wavelength of the _ particle _ , which can be up to @xmath0 m for ultra cold atoms @xcite . one issue predicted here , fundamental to extending the quantum - classical boundary , is that of interferometric effects which do not become imperceptible in the limit of large mirror mass . such effects are surprising in comparison with the imperceptible fringe spacing for a massive particle traversing a double slit . to understand why fringe spacing in reflection is robust with respect to variations in @xmath9 consider simplifying the two - body state into two one - body states . the first consists of the wavefunctions of the incident and reflected particle . the second consists of the mirror wavefunctions before and after reflection of the particle , both of which travel in the same direction but with different momenta . in the case of a mirror whose speed is close to zero , superposition of these particle states results in a `` standing wave '' with nodes or fringes spaced at about half the particle debroglie wavelength . superposition of the moving mirror states , on the other hand , forms a fringe spacing commensurate with the difference in momentum between the incident and reflected _ particle _ states . the larger the mass of the mirror the smaller the change in velocity of the mirror upon reflection . yet the difference in momentum between the mirror before and after reflection remains the same , depending only on the change in momentum of the particle and _ not _ on the mass of the mirror . it is this difference in mirror momenta which leads to the phase difference @xmath46 when superposing these two mirror wavefunctions . that is , the mirror fringe spacing is only determined by the change in particle momentum since conservation of momentum in reflection requires @xmath47 . while this simplified model , reducing a two - body state into two one - body states , demonstrates why interference for a mesoscopic / macroscopic mirror mass does not become imperceptible as @xmath48 , it does not account for the correlation given by the exact solution in eqn . [ eq : entangledinterference ] , which is inherently a two - body effect . double slit interference with a massive particle , on the other hand , superposes two one - body states with the same momentum whose difference in phase , @xmath49 , is due to the difference in path lengths from each slit to the measurement point times an extremely large wavevector . this results in an imperceptible fringe spacing . the probability of measuring the particle at @xmath50 and the mirror at @xmath51 is given by @xmath52 dx_{1 } dx_{2}$ ] with the joint pdf determined by the solution of equation [ eq : schreqn ] as @xmath53 . using this equation , conservation of probability can then be expressed locally as , @xmath54}{\partial t}+ \frac{\partial j_{1}[x_{1},x_{2},t]}{\partial x_{1}}+\notag \\ \frac{\partial j_{2}[x_{1},x_{2},t]}{\partial x_{2}}=0 , \label{eq : consprob}\end{aligned}\ ] ] where @xmath55=\hbar ( \psi^ { * } \partial_{x_{1 } } \psi-\psi \partial_{x_{1 } } \psi^{*})/(2 i m)$ ] and @xmath56=\hbar ( \psi^ { * } \partial_{x_{2 } } \psi-\psi \partial_{x_{2 } } \psi^{*})/(2 i m)$ ] . while the expressions for these current densities appear similar to that for one particle systems there are subtle but important differences for a two - body system @xcite . multiplying equation [ eq : consprob ] by @xmath57 , integrating over the segment from @xmath58 to @xmath59 along the x - axis ( @xmath60 and @xmath61 ) , and then rearranging terms yields a solution to equation [ eq : consprob ] if @xmath62 dx_{1 } dx_{2 } + \notag \\ \int_{a}^{b } ( j_{1}[b , x_{2},t]-j_{1}[a , x_{2},t ] ) dx_{2 } \notag \\ + \int_{a}^{b } ( j_{2}[x_{1},b , t]-j_{2}[x_{1},a , t ] ) dx_{1}=0 \notag . \label{eq : cons2}\end{aligned}\ ] ] this is most easily interpreted using an @xmath63 by @xmath63 rectangular region aligned along the @xmath2 and @xmath3 axes . the time rate of change in probability within this region is determined by the change in flux of probability out along the @xmath2 axis ( @xmath64-j_{1}[a , x_{2},t]$ ] ) plus that along the @xmath3 axis @xmath65-j_{2}[x_{1},a , t])$ ] . since these fluxes in general vary spatially they have to be integrated over this variation along the rectangular boundaries . numerical integration of the wavegroup results presented below are consistent with conservation of probability . to better understand the practical consequences of these results , wavegroups are next formed from a superposition of the incident and reflected ` harmonic wavefunctions ' ( given in eqn . [ eq : superposition ] ) expressed in terms of the correlated particle and mirror substates . an analytic expression for such wavegroups can be obtained for a gaussian distribution in wavevector components @xmath4 and @xmath5 . for the mirror , this is proportional to @xmath66/(2 \delta k^{2})$ ] , where the peak of the distribution is at @xmath67 and @xmath68 is its width , while for the particle , this is proportional to @xmath69/(2 \delta k^{2})$ ] , where the peak of the distribution is at @xmath70 and @xmath71 is its width . the incident wavegroup propagates in the @xmath72 plane along a line whose slope is determined by a ratio of the group velocities of each substate and spreads due to dispersion independently in each direction . . the lower pdf waveform moves toward the diagonal white line , corresponding to @xmath23 , then reflects in the middle snapshot where the incident and reflected two - body wavefunctions ` overlap ' , and finally it moves away from the diagonal in the upper snapshot . the upper left inset is a schematic of the ` classical ' analog before reflection while the upper right inset is that after reflection with initial and final particle and mirror velocities @xmath10 , @xmath11 , @xmath73 , and @xmath74 respectively . there is no classical analog for the middle snapshot . ] in fig . [ fig : interference ] snapshots of the two - body probability density function are shown at three times for @xmath75 , @xmath76 , and @xmath77 . a slice of fig . [ fig : interference ] for @xmath78 along the @xmath3 coordinate is shown in fig . [ fig : fringespacing ] ( the solid line ) along with a slice of this fig . for @xmath79 along @xmath2 ( the dashed line ) for different bandwidth wavegroups . illustrates that the fringe spacing for the particle and mirror substates with narrow bandwidth wavegroups , are essentially the same for the particle and mirror substates , as discussed for the approximation @xmath80 following eqn . [ eq : fringespacing ] . which show the fringe spacing along the @xmath3 axis for @xmath78 ( dashed lines ) and along the @xmath2 axis for @xmath79 ( solid lines ) . the @xmath2 axis has been inverted to display both the dashed and solid lines together . although each graph has @xmath76 the value of @xmath68 increases sequentially by a factor of @xmath81 from the front to the back . ] verification of correlated interference requires simultaneous measurement of the particle and cm of the mirror with instruments having a spatial resolution smaller than the fringe spacing . for a static mirror this spacing is half the debroglie wavelength of the particle , which at @xmath82 for ultracold atoms @xcite satisfies this requirement while it is dubious at @xmath83 for slow neutrons @xcite . another constraint on the interference illustrated in fig . [ fig : interference ] is that the fringe visibility function must be non - zero . that is , the incident and reflected particle - mirror wavegroups must ` overlap . ' the interference fringes are then determined predominately by a superposition of ` harmonic wavefunctions ' @xcite . for example , the interference shown in fig . [ fig : interference ] is determined predominately by eqn . [ eq : entangledinterference ] when the wavegroups ` overlap ' in the center snapshot . the longitudinal coherence lengths for both the particle and mirror are greater than the fringe spacing in this figure . in the upper snapshot there is neither ` overlap ' nor such interference . the fringe visibility function is non - zero if each wavegroup substate ` overlaps ' within approximately a longitudinal coherence length @xcite , which is given by @xmath84 @xcite . for particle substates this can be @xmath85 for ultracold atoms @xcite or @xmath86 for slow neutrons @xcite . if the uncertainty in the mirror velocity is determined by its thermal equilibrium with the environment then @xmath87 yielding @xmath88 . this expression is consistent with results for ultra - cold atoms in a bose - einstein condensate @xcite . in the interference region , simultaneous measurement is approximately confined both to a spatial region determined by the two coherence lengths and a temporal region given by the time during which the wavegroups overlap . the former is small for mesoscopic / macroscopic mirror masses while the duration of the interference is essentially determined by the speed of the particle and its coherence length which is approximately equal to @xmath89 for a static mirror when @xmath90 . one method , discussed below , to reduce these coherence length limitations is with multiple such two - body states . first however , the effect of variations in mirror coherence lengths on correlated interference is illustrated . [ fig : spreadvelocity ] illustrates how variation in the longitudinal coherence length of the mirror substate affects the particle - mirror interference with fixed particle substate coherence length . part a shows a longer mirror coherence length than is used in fig . [ fig : interference ] while parts [ fig : spreadvelocity]b through [ fig : spreadvelocity]d progressively reduce the coherence length of only the mirror substate . one might expect that the small coherence length associated with a mesoscopic / macroscopic mirror mass would not allow for the interference shown in fig . [ fig : spreadvelocity]a or [ fig : spreadvelocity]b . in fig . [ fig : spreadvelocity ] d a small mirror coherence length along with a large mirror recoil prevents wavefunction overlap over a range of @xmath2 values ( @xmath91 ) where it was present before . nevertheless , a slice along the @xmath3 axis for measurement of the particle at these values of @xmath2 indicates a splitting of the mirror substate into two states which do not overlap and are therefore distinguishable . this splitting is a consequence of two ways that the particle could have reached @xmath2 . it could have come from the incident _ or _ reflected particle wavegroup substates since the position of the particle is not known to within its large coherence length . as the _ mirrors _ coherence length increases , the wavefunctions associated with these two ways overlap and generate correlated interference as shown in fig . [ fig : spreadvelocity]a . as the mirrors coherence length decreases , the position of the mirror before reflection is distinguishable from that after reflection due to mirror recoil , which results in _ no interference of either the particle or mirror _ , as shown in fig . [ fig : spreadvelocity]d for @xmath91 . that is , measurement of the particle at @xmath92 and mirror at @xmath79 is consistent with conservation of momentum for the particle and mirror not to have reflected while measurement of the particle at @xmath92 and mirror at @xmath93 is consistent with conservation of momentum for the particle and mirror to have reflected . the short coherence length of the mirror relative to the recoil distance then results in distinguishable two - body states . the condition for interference in a two - body system therefore involves the coherence of both bodies . but with different spreads in mirror velocities for a fixed spread in particle velocities while all other parameters are the same . @xmath94 , and @xmath95 for @xmath96 in parts a , b , c , and d respectively . note the change of scale in graph d. ] after reflection the spatial width of the mirror wavegroup substate is exchanged with that of the particle wavegroup substate when @xmath97 . this is most easily seen by constructing a particle - mirror wavefunction with different bandwidths for the particle and mirror wavegroup substates , shown in fig . [ fig : coherencetransfer ] which is a contour plot of joint pdfs similar to figure [ fig : interference ] but without the middle snapshot in the interference region . the solid and dashed contours correspond to @xmath98 and @xmath99 respectively with the spread in velocities given by @xmath100 . this result can be understood by comparing classical and quantum reflection . in a one - dimensional classical collision , conservation of energy and momentum require an exchange of particle - mirror velocities independent of either velocity for @xmath101 . this is manifest quantum mechanically in the exchange of commensurate wavefunction parameters @xmath4 and @xmath5 between the incident and reflected two - body wavefunctions . if an incident particle substate , consisting of only one harmonic component ( corresponding to speed @xmath10 ) reflects from a mirror substate with many velocity components , then each harmonic component of the mirror substate ( corresponding to different values of @xmath11 ) reflects the particle substate and therefore acquires velocity @xmath10 while the reflected particle substate acquires different velocity values for each reflected component of the mirror substate . this results in the reduction of the mirror bandwidth and an increase in the particle bandwidth , which is manifest in fig . [ fig : coherencetransfer ] as the exchange of incident and reflected wavegroup shapes . it also is responsible for the distortion of the reflected wavegroup shape in fig . [ fig : interference ] . but without the interference or `` overlap '' region snapshot , illustrating coherence transfer for two different particle - mirror mass ratios . the solid and dashed lines are for a spread in velocities given by @xmath100 with @xmath98 and @xmath99 , respectively . ] experimentally confirming coherence transfer involves reflection with disparate coherence lengths . on a microscopic scale , this could be observed between an atom which has been cooled to ultracold temperatures retro - reflecting from the same type of atom which has a shorter coherence length . the transfer of coherence might then be monitored not by a measurement of the pdf but rather by its coherence properties via interferometry . a more useful implementation for increasing the coherence length of larger objects could involve a group of bose - einstein particles ( the condensate ) , whose total mass is similar to that of the mirror , retro - reflecting`coherently ' from a mirror as ` one particle ' rather than a collection of individual particles . to mitigate both spatial and temporal fringe localization in reflection due to small particle and mirror coherence lengths , consider interference of multiple such two - body states : a one - dimensional example of which is the interference from the reflection of a neutron by each of the two surfaces of a moving aluminum ` slab ' . the two - body pdfs associated with two models of such reflection are calculated below . the first , presented in the next section , simply approximates each weak surface reflection as being due to a delta function potential which allows for both transmission and weak reflection . these two delta function potentials are fixed to and symmetrically offset from the cm of the slab ( of length @xmath102 ) by a distance @xmath103 , with multiple reflections neglected . the results from this simple model then corroborate those from a more accurate model , given in section [ sec : slab ] , involving a two - body treatment of a particle interacting with a finite barrier or finite well . the delta function ` beamsplitter ' model is included to focus attention on the physics associated with generating multiple two - body states . the more accurate finite barrier or well treatments conceal the fundamental physics but yield similar results . in essence , the latter incorporates multiple reflections along with the appropriately modified wavevectors between the surfaces of the slab while the former does not . the results of the simpler delta function model approach those of the finite barrier - well model as the reflections from the slab interfaces become weaker . for completeness , the results of the particle trapped in a moving infinite well are included in subsection [ sec : well ] . while the finite barrier or well examples discussed below deal only with unbound states , the infinite well is an example of the two - body formalism for a bound state . the classical analogy for the neutron reflecting from each of the two surfaces of a moving aluminum ` slab ' is that of a short pulse of light reflecting from a moving thin film . as the film s thickness , speed , or the average wavelength of the incident pulse changes , the reflected pulse intensity then varies as a consequence of the interference between the reflected waves from the two interfaces . classically , such thin film interference mitigates both the spatial and temporal fringe localization associated with reflection of a pulse of light from a moving mirror ( the analog of the two - body quantum treatment of a particle reflecting from a moving mirror ) . interference between the incident and reflected two - body pulse from the moving mirror is localized spatially by the coherence length of the pulse . these fringes are also temporally constrained by the pulse duration . however , interference of the two reflected light pulses , one from each surface of the thin film , dramatically reduces this fringe localization for two reasons : first the two reflected light pulses travel in the same direction thereby maintaining spatial overlap and second they experience the same doppler shift in reflecting from the two interfaces which move at the same speed resulting in fringes which are not time dependent ( this is not the case for interference of the incident and reflected pulses from a moving mirror ) . for similar reasons , interference of the two - body wavegroup states associated with reflection from each surface of the slab mitigates the fringe localization associated with reflection from a mirror however , the two - body interference is now between two wavegroups , each analogous to the one shown in the upper snapshot of fig . [ fig : interference ] and each of which corresponds to a reflection from only one surface of the slab and both of which move in the same direction thereby maintaining overlap . the classical treatment yields interference only for the pulse of light and not the mirror from which it reflects . the two - body quantum treatment presented below yields predictions about correlations in the measurements of both the neutron and slab . destructive interference of these wavegroups then corresponds to no probability of measuring the neutron and slab along the overlap region . a change in the slab s thickness , speed , or the average wavelength of the incident particle then results in a variation of the probability of simultaneously measuring the neutron and slab between zero and some maximum value , corresponding to constructive and destructive interference . a pulse of light reflecting from a thin film can also generate a standing wave within the film under appropriate conditions . similarly , two - body wavefunctions also generate such a `` fabry - prot '' resonance which is illustrated below . modifying @xmath13 to split rather than totally reflect the incident wave while including an offset in the delta function argument to model the surface which is offset from the cm of the slab yields the two - body reflected wavefunction . this eigenstate of energy or ` harmonic solution ' , for each surface reflection , is then superposed to form wavegroups associated with refection from that surface . rather than the waveform of the center snapshot in fig . [ fig : interference ] , a waveform analogous to that in the upper snapshot is the focus of this discussion . there are now two such reflected neutron - slab wavegroups , one for each slab surface , which interfere . since the neutron either reflects from the first surface or transmits through the first to the second , from which it then reflects , both reflected neutron substates have the same distribution of velocities . since in each case momentum is transferred to the slab only once for each of these two possible paths ( reflection from either the first _ or _ second surface of the slab ) , two associated slab substates are generated , each with the same velocity distribution . this results in two reflected neutron - slab wavegroups traveling at the same speed and direction but offset . interference then requires wavegroup ` overlap ' within a coherence length of both the neutron and slab substates . an example of a lack of overlap in the mirror substate , resulting in no two - body interference , was discussed in section [ sec : bandwidth ] and illustrated in fig . [ fig : spreadvelocity]d . this requirement , applied to the slab substate , is that the cm of the two reflected slab substates separate no further than their coherence lengths . this offset is estimated by the extra slab speed , beyond @xmath11 , acquired from the first reflection multiplied by the difference in particle reflection times between the two slab surfaces . assuming @xmath43 and @xmath104 this offset , @xmath105 , then must be less than @xmath106 , which if determined by thermal equilibrium requires @xmath107 next consider the constraint on @xmath108 . when a neutron wavegroup of spatial width @xmath109 ( where dispersion in reflection is neglected ) retro - reflects a distance @xmath110 from the cm of the slab , with @xmath110 much greater than both the wavegroup width @xmath109 and the slab length @xmath102 ( @xmath111 ) , ` overlap ' of the neutron wavegroups which have reflected from both surfaces occurs . for a near stationary slab the reflected neutron wavegroups are offset by about a distance @xmath112 , interference from which requires a neutron coherence length exceeding @xmath112 . an example which satisfies this constraint is a neutron of coherence length @xmath113 m @xcite reflecting from the two surfaces of an aluminum slab of thickness @xmath114 m ( there is little constraint on the area of this slab and therefore its mass ) . these parameters easily satisfy the overlap constraints for both the mesoscopic / macroscopic slab at room temperatures and the neutron . having achieved overlap , the resulting interference is then determined essentially by the harmonic wavefunctions yielding the _ constant _ two - body reflected joint probability density @xmath115 . \label{eq : slabinterference}\ ] ] assuming @xmath43 and @xmath104 yields @xmath116 . \label{eq : slabintapprox}\ ] ] the slab _ and _ reflected neutron will never be simultaneously observed when @xmath117 @xcite . such non - local two - body interference is similar to that of a pulse of light retro - reflecting from a thin film where destructive interference depends neither on the locations of the detector nor the thin film . indeed , in the limits just given these two cases are mathematically identical if the photon wavelength is replaced with the particle s debroglie wavelength . the pdf for reflection from the slab is illustrated for wavegroups in fig . [ fig : slab ] using parameters similar to those above : @xmath113 m , @xmath118 m , @xmath119 m/s , @xmath120 kg , and a slab coherence length determined by @xmath121 at @xmath122 . the neutron reflects from the first slab surface at @xmath79 , @xmath78 , and @xmath123 . to illustrate the non - local nature of the interference , three snapshots are chosen for times which allow the center of the wavegroup to change position by four orders of magnitude in both the neutron and slab coordinates . since the spatial size of the neutron - slab pdf is limited by small coherence lengths , a graph with a spatial region encompassing all of these points , while still showing the wavegroup structure , can not easily be generated . instead ` blow - ups ' in the @xmath72 plane of small regions around these points are shown . the snapshots in the upper row have a span in @xmath2 of @xmath124 m and a span in @xmath3 of @xmath125 m with the approximate centers of each graph located at @xmath126 m and @xmath127 m . the middle row of graphs have a span in @xmath2 of @xmath128 m and a span in @xmath3 of @xmath129 m with the approximate centers of each graph located at @xmath130 m and @xmath131 m . the lower row of graphs have a span in @xmath2 of @xmath132 m and a span in @xmath3 of @xmath133 m with the approximate centers of each graph located at @xmath134 m and @xmath135 m . the increase in span is needed since the wavegroup expands over these time scales . the vertical scale therefore decreases from top to bottom rows but is the same within each row so that pdf heights can only be compared within a row . the only parameter that differs between the graphs in the left and right columns of fig . [ fig : slab ] is the neutron velocity before reflection . it is @xmath136 m / s for the left and @xmath137 m / s for the right column respectively . these two neutron speeds result in either constructive or destructive interference ( left or right columns respectively ) of the neutron - slab wavegroups which are offset by @xmath138 due to reflection from the two slab surfaces . as this offset decreases ( e.g. by decreasing @xmath102 ) or @xmath108 increases , the pdf goes to zero approaching the destructive interference predicted in eqn . [ eq : slabintapprox ] ( since @xmath43 and @xmath104 ) . similar calculations also confirm the predictions that there is little effect on the interference from increasing @xmath9 while changes in @xmath10 have a significant effect . note also that the small size of the slab coherence length , @xmath139 m , is much larger than the offset due to reflection from the two slab surfaces , @xmath140 m , and therefore has little effect on the destructive interference . increasing the mass of the slab changes only the span in the @xmath3 coordinate of these graphs but not the interference . the parameters in fig . [ fig : slab ] are _ not _ chosen to illustrate effective wavegroup destructive interference but rather to show coherence length effects on the interference using experimentally realizable values . and @xmath2 respectively ) after reflection . the left column graphs differ from the right only in the speed of the incident neutron , illustrating constructive and destructive interference respectively . ] the model for a neutron traversing a slab of matter which incorporates multiple reflections and variation in the wavevector inside the slab is determined from the schrdinger equation given by @xmath141 \notag \\ + i\partial_{t})\psi[x_{1},x_{2},t]=0 , \label{eq : scheqn}\end{aligned}\ ] ] where @xmath142 $ ] is now the potential energy associated with either the well or barrier . of interest is a solution which yields an energy eigenstate for the particle - well or barrier interaction , for which neither the particle nor the well or barrier is localized . wavegroups are then constructed from these states . a separable solution to this two - body schrdinger equation is obtained in a manner similar to that outlined in section [ sec : theoryparticlemirror ] . the center of mass and relative coordinate transformation yields solutions which are matched at the boundaries and then transformed to the particle - well coordinates . the infinite well potential is presented to illustrate the application of the two - body formalism to a bound state . then in subsection [ subsec : barrier ] calculations are outlined for a particle interacting both with a finite barrier and a finite well in unbound states . these provide a more accurate model for a neutron traversing a slab of matter than that just discussed . the boundary condition for the infinite potential well , of width @xmath102 , is @xmath143\rightarrow 0 $ ] . the relative wavefunction does not exist outside the well for @xmath144 and @xmath145 . a solution to eqn . [ eq : scheqn ] for a particle which has reduced mass is then @xcite @xmath146~\theta[x_{rel},d ] , \label{eq : psiwell2}\ ] ] where @xmath147=(\theta [ x_{rel}+d]-\theta [ x_{rel}-d])$ ] has value one within the well and zero everywhere else with n being the number of nodes . the solution to the center of mass ode ( the same as eqn . [ eq : scheqode1 ] ) is @xmath148 the complete solution is then @xmath31 \propto \psi_{cm}\psi_{rel}$ ] . again , we are interested in measurements of the particle and well rather than measurements of two transformed `` objects , '' one with a reduced mass and the other with the total mass of the system . expanding the sine function in eqn . [ eq : psiwell2 ] into exponential form results in wavefunctions traveling in opposite directions . transforming back to the particle - well coordinates , the momenta and energies of the particle and well differ in magnitude in these two directions . the modes of the well , characterized by the number of nodes @xmath149 , are determined in the cm - rel coordinates by @xmath150 . in the particle - well coordinates this constrains the incident values of particle and well velocities . that is , given an initial well velocity @xmath11 only the following initial particle velocities are allowed , @xmath151 in the particle - well coordinates the wavefunction s phase , @xmath152 , needs to have its temporal part parsed into the kinetic energy of the particle and the kinetic energy of the well . this is done as in the previous calculations . the resulting kinetic energies of the particle and well differ as the particle moves in opposite directions . these energies and their associated momenta are those expected of a classical particle reflecting from the moving well walls . the finite barrier or well ( referred to as only the barrier ) has potential energy @xmath153 ( positive for the barrier and negative for the finite well ) and extends over a distance @xmath102 . this divides space into three regions : before the barrier or `` before '' , in the barrier region or `` barrier , '' and after the barrier or `` after . '' solutions are first obtained for these three regions in the cm and rel coordinates by solving eqns . [ eq : scheqode1 ] and [ eq : scheqode2 ] . the solution to eqn . [ eq : scheqode2 ] before the barrier consists of incident and reflected wavefunctions given by @xmath154 where @xmath155 . the solution in the barrier - finite well region also consists of incident and reflected wavefunctions given by @xmath156 where @xmath157 . the solution after the barrier consists only of a transmitted wavefunction given by @xmath158 where @xmath159 . the boundary conditions are continuity of the wavefunctions and their derivatives with respect to @xmath160 at @xmath161 . these then constrain the coefficients @xmath162 , @xmath163 , @xmath164 , and @xmath165 with @xmath166 . again , the solution to eqn . [ eq : scheqode1 ] is given by eqn . [ eq : psiwell1 ] . the complete solution is then @xmath31 \propto \psi_{cm}\psi_{rel}$ ] . since we are interested in predictions about measurements of the particle and barrier rather than measurements of the reduced mass and total mass `` objects , '' a transformation from the relative and center of mass to the particle - barrier coordinates is required . in the particle - barrier coordinates the wavefunction s phase , @xmath152 , needs to have its temporal part parsed into the kinetic energy of the particle and the kinetic energy of the barrier . this procedure is the same as it is for the particle - mirror and particle - infinite well two - body states as described above . however , the product of the potential energy part of the total energy with time @xmath167 in the barrier region must be added to the phase . such a term has no measurable effect as will be shown next . the wavefunction in the barrier , expressed in the particle - well coordinates , is then @xmath168 where @xmath169 and @xmath170 contain the spatial terms in the phase and are functions of @xmath171 and @xmath172 . the temporal terms contain the kinetic energy for the particle and barrier moving to the right and left , @xmath173 and the potential energy @xmath153 . the kinetic energy terms are functions of @xmath174 , and @xmath172 . the potential energy term @xmath175 is a common factor of both the incident and reflected wavefunctions in the barrier . since the pdf is generated from the wavefunction multiplied by its complex conjugate , such common factors have no effect on the pdf . there is then no need to associate the potential energy part of the total energy with either the particle or the barrier or as a separate term in the phase , @xmath176 . the potential energy part of the total energy has observable consequences only parametrically within the momenta and kinetic energies of the particle and barrier . this simple division of the momentum and energy of the particle in the barrier can be contrasted with that of the stress - energy tensor for an electromagnetic wave traversing a dielectric slab @xcite . wavegroups are next formed from a gaussian superposition of the two - body energy eigenstates for the particle and infinite well or finite particle and barrier - well described above . correlated interference is a consequence of any such superposition . however , we focus the following discussion on the subsets of correlated interference effects which deal with the superposition of two such wavegroups . in particular , emphasis is given to the two - body quantum analogy of a pulse of light reflecting from a thin film : interference from the two - body wavegroups which have reflected from the two surfaces of the slab . at the risk of distracting from this main topic , correlated interference from multiple reflections when such a wavegroup interacts with an infinite barrier are also discussed . the wavegroup for the particle and barrier or finite well is calculated using a gaussian superposition of the energy eigenstates given in subsection [ subsec : barrier ] . unfortunately , the coefficients @xmath162 , @xmath163 , @xmath164 , and @xmath165 of eqns . [ eq : psibefore ] , [ eq : psibarrier ] , and [ eq : psiafter ] depend in a non - trivial manner on the variables of integration . the resulting integrals can not be determined in closed form . to facilitate the calculation , the following sums will replace these integrals : @xmath177 \notag \label{eq : barrierwavegroup}\end{aligned}\ ] ] where the peak of the barrier velocity distribution is at @xmath178 , @xmath179 is its width , and the sum is from an initial barrier velocity @xmath180 to a final velocity @xmath74 . summing over the particle velocity distribution yields the wavefunction for the wavegroup given by @xmath181 where the peak of the particle velocity distribution is at @xmath182 , @xmath183 is its width , and the sum is from an initial particle velocity @xmath184 to a final velocity @xmath73 . two cases are now treated : the unbound two - body wavegroup state of a particle and finite well and a similar unbound state of a particle and finite barrier . [ [ sec : barriergreater ] ] particle traversing a finite well : @xmath185 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider next a particle interacting with a finite well whose sum of initial kinetic energies in the relative coordinate system is greater than the well pe . the size of the particle substate wavegroup is chosen to be a few times larger than the finite well width @xmath102 . [ fig : barrier_negenergy ] shows results of the pdfs for three sequential snapshots taken at equal time intervals progressing along the dashed line from the lower left to upper right and upper left . the analogous classical positions of the particle and finite well for particular snapshots are illustrated in the insets . the barrier boundaries occur at the diagonal white lines , corresponding to @xmath186 . the parameters used in fig . [ fig : barrier_negenergy ] are @xmath187 , @xmath188 , and @xmath189 while the @xmath190 using the average value of @xmath191 for the wavegroup particle and well distributions . one category of correlated interference , which we call type i , occurs when the incident and reflected two - body wavefunctions , traveling in opposite directions in the @xmath192 plane , ` overlap . ' this is illustrated in fig . [ fig : barrier_negenergy ] for weak reflection by the fringes of the middle snapshot just to the left of the barrier or line @xmath193 and also by the fringes in fig . [ fig : interference ] . these fringes are spaced by about half the debroglie wavelength of the _ particle _ for @xmath194 and @xmath195 as are the similar correlated interference fringes in the two - body reflection of a particle from a mirror given in eqn . [ eq : entangledinterference ] . however , the interaction generates another form of correlated interference when the reflected wavegroups , one from each barrier surface , interfere as they travel along the _ same _ direction in the @xmath192 plane . this new category of correlated interference , which is referred to as type ii , is illustrated in fig . [ fig : barrier_negenergy ] by the peak labeled @xmath59 . this is the same type of interference from reflections at the two surfaces of the slab modeled as two delta function in section [ sec : twodelta ] . it is similar to the classical interference of a pulse of light reflecting from a thin film . changing only the barrier spacing generates an oscillation in the pdf for peak @xmath59 analogous to that found in the interference of a pulse of light reflecting from a thin film whose thickness varies . that is , this peak goes through constructive and destructive interference from the two barrier reflections when the wavegroup size is much larger than that of the barrier and the spacing @xmath102 is varied . as time progresses peak @xmath59 maintains this interference as it travels in the @xmath192 plane . these are the same characteristics of the interference found in the simple two delta function model of reflection from a slab given in section [ sec : twodelta ] and shown in fig . [ fig : slab ] . this differs from the type i correlated interference associated with the fringes shown in the middle snapshot of fig . [ fig : barrier_negenergy ] which is localized to a small temporal and spatial region . this figure illustrates yet another form of correlated interference , referred to as type iii : that from multiple reflections from the two barrier edges , which is shown as the fringes within the well of the middle snapshot of fig . [ fig : barrier_negenergy ] . these are analogous to the standing wave formed in a thin film or optical cavity . . the first snapshot generates peak @xmath58 while peaks @xmath59 and @xmath196 comprise the last snapshot . the classical analogs of these peaks are shown in the insets . the pdf progresses temporally along the dashed line . ] [ [ sec : barriermiddle ] ] particle traversing a finite barrier : @xmath197 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider next wavegroups for which some fourier components of the particle and barrier substates have a total initial kinetic energy in the relative coordinate system which exceeds the barrier potential energy while other components have a total relative initial kinetic energy which is less than the barrier potential . to illustrate the resulting pdfs the size of the particle and barrier substate wavegroups are chosen to be slightly larger than the barrier width @xmath102 . [ fig : barrier_posenergy ] shows the pdfs from such an interaction using three sequential snapshots , progressing along the dashed line from the lower left to upper right . the speed of the particle and well are illustrated for a classical system of two such particles in the insets next to each snapshot . again the diagonal white lines correspond to @xmath186 . the parameters used in fig . [ fig : barrier_posenergy ] are @xmath187 , @xmath188 , @xmath189 while the @xmath198 for the average value of the @xmath191 for the wavegroup distributions . this figure also illustrates type iii correlated interference but it differs from fig . [ fig : barrier_negenergy ] in that only the first mode is excited . additionally , the mode decays slowly enough to be visible in the third snapshot labeled as peak @xmath199 ( located between both @xmath186 and peaks @xmath59 and @xmath196 in the third snapshot ) . this peak is analogous to the buildup and decay of electromagnetic energy in a optical cavity . later snapshots ( not shown ) illustrate its decay . the peaks labeled @xmath59 in fig . [ fig : barrier_posenergy ] and fig . [ fig : barrier_negenergy ] are both of type ii . for the particle traversing the barrier . the only difference between the parameters used here and in fig . [ fig : barrier_negenergy ] is the pe which forms a barrier . ] the position of peak @xmath196 can be compared between figs . [ fig : barrier_negenergy ] and [ fig : barrier_posenergy ] since all parameters are the same except the pe . the location of this peak indicates the effect of the interaction on the relative transit times for the particle and finite well or barrier wavegroup substates . the infinite well - particle calculation differs from that of the barrier due both to the particle and well velocities being constrained by the resonance condition given in eqn . [ eq : quantizationvv ] and by the lack of coefficients , such as the @xmath162 , @xmath163 , @xmath164 , and @xmath165 used in the previous section , which depend on the parameters of integration . a closed form expression for the well substate wavegroup can be obtained from a gaussian distribution of the energy eigenstates parameterized in terms of velocity components , @xmath11 , of the well , given by @xmath200dv \notag \label{eq : psiwellwavegroup}\end{aligned}\ ] ] where the peak of the distribution is at @xmath178 and @xmath179 is its width . this is then summed over integral values of @xmath149 ( the number of nodes ) using the gaussian distribution @xcite @xmath201 \psi_{well}^{wavegroup } , \notag \label{eq : psiparticlewavegroup}\end{aligned}\ ] ] where the peak of the distribution is at @xmath202 and @xmath203 is its width . using these relations , the pdf is plotted with particle and well substate wavegroups whose spatial widths are less than the well spacing @xmath102 . [ fig : wellsynchfinal ] shows such pdf results for six snapshots taken at equal time intervals progressing from the lower left to upper right along the dashed line . ` reflection ' occurs at the diagonal white lines , corresponding to @xmath186 . the particle and well `` reflect '' from each other twice , in the second and fourth snapshots . the classical analogs for the particle and well positions for some snapshots are shown in the insets , labeled by a , b , and c. these correspond to the snapshots of the wavegroups labeled with the respective letters . while the insets are schematics of the ` classical ' analog between the wave and particle pictures , there is nothing similar for the correlated interference snapshots . interference occurs when the incident and reflected two - body wavefunctions ` overlap ' and is shown in higher spatial resolution for the first reflection in the right side inset of fig . [ fig : wellsynchfinal ] . the fringes are spaced by about half the debroglie wavelength of the _ particle _ for @xmath194 and @xmath195 , as are similar correlated interference fringes in two - body reflection of a particle from a mirror in fig . [ fig : interference ] . the parameters used in fig . [ fig : wellsynchfinal ] are @xmath204 , @xmath205 , @xmath206 , and @xmath207 . for a two - body wavefunction whose wavegroup size is less than the infinite well spacing . the diagonal white lines correspond to @xmath186 . the trajectory is indicated by the dashed line . the right inset is a blow - up of the correlated interference of the second snapshot . ] correlation measurements are more difficult to perform than measuring just one of the bodies in a many body system . to predict one body effects , the two - body pdf is integrated ( traced ) over the coordinates of one of the bodies thereby converting it into a ` one - body ' or marginal pdf for the other body @xcite . for example , consider a measurement of the particle but not the mirror for the two - body pdfs shown in fig . [ fig : spreadvelocity ] . averaging over the mirror coordinate in fig [ fig : spreadvelocity]a essentially eliminates any interference when measuring only the particle . averaging over the particle coordinate eliminates any interference in a measurement of only the mirror . yet there is interference in a correlated measurement . gottfried describes a related result using a different two - body interferometer @xcite . such averaging over the mirror coordinate of fig . [ fig : spreadvelocity]c , however , does not ` wash out ' interference in the particle - only measurement . the resulting particle - only ` standing - wave ' interference pattern in reflection from a static mirror is of course expected . yet the integral of this two - body pdf over the particle coordinate yields no mirror - only interference . this matches the classically expectation that there is interference for the wave reflecting from a mirror but not for the mirror . however , if the coherence lengths are switched ( it is difficult to increase the coherence length of a mesoscopic / macroscopic mirror ) then a particle - only measurement will yield no interference while a mirror - only measurement will yield interference . the two - body solution , when used in this manner to determine a one - body pdf , then reveals the limitations of starting with a solution to the one - body schrdinger equation where the potential ( mirror ) is assumed to be a classical rather than a quantum object . for example , a one - body pdf for the particle , determined by the one - body schrdinger equation , is a consequence of the solution to the two - body schrdinger equation only for coherence lengths used in fig . [ fig : spreadvelocity]c . the reflected one - particle pdfs obtained when averaging over the mirror coordinates of fig . [ fig : spreadvelocity]a , [ fig : spreadvelocity]b , and [ fig : spreadvelocity]d , however , _ can not _ be obtained without treating the mirror as a quantum object . the particle - only measurements , determined from this two - body treatment , then act as a probe of the mirrors quantum state . two regimes in which the quantum nature of the mirror is revealed are illustrated in fig . [ fig : spreadvelocity]a and d using fixed particle bandwidth . the result in part [ fig : spreadvelocity]a was discussed above . the other regime , shown in fig . [ fig : spreadvelocity]d , requires such small mirror coherence lengths that the incident and reflected two - body states are distinguishable , thereby also eliminating any interference . consider a simple estimate of the condition for which interference of the particle reflecting from the mirror disappears in the two - body pdf of fig . [ fig : spreadvelocity]d . this is done by approximating the incident two - body pdf of fig . [ fig : spreadvelocity]d as a rectangle with height @xmath208 and width @xmath209 @xcite . assuming a near stationary mirror with @xmath43 and @xmath104 , reflection then tilts this rectangle to an angle @xmath13 given by @xmath210 where @xmath211 and @xmath212 are the velocities after reflection of the mirror and particle . the overlap of the incident and reflected rectangles occurs at a distance @xmath213 from the point of maximum overlap given by @xmath214 . beyond @xmath213 no overlap or interference occurs . using the coherence length associated with thermal equilibrium of the mirror yields the following condition for the particle fringes to disappear in the two - body solution of fig . [ fig : spreadvelocity]d @xmath215 this condition generates a regime in which the quantum nature of the mirror is manifest via elimination of the expected one - body interference of the particle in reflection , which is caused by the incident and reflected paths becoming distinguishable as the mirror coherence length becomes smaller ( see section [ sec : bandwidth ] ) . although there is a lack of interference in the particle - only measurement , the two - body state is still split . this is evident in fig . [ fig : spreadvelocity]d as a bimodal distribution for which there is a finite probability of finding states of both reflection _ and _ no reflection . the effect is enhanced with reflection of ultra - cold atoms which have both larger coherence lengths and masses than neutrons . a yet larger mass could involve a bose - einstein condensate itself if it reflects as ` one particle ' rather than a collection of individual particles from the mirror @xcite . such ` standing - wave ' interference on reflection of a bose - einstein condensate ( @xmath216 m ) is eliminated with a mirror of @xmath217 kg ( the average mass of a mosquito ) and @xmath218 k. reflection of both atoms @xcite and bose - einstein condensates @xcite from mirrors has been studied but neither in the regimes discussed here nor in an attempt to measure such ` standing - wave ' interference . next consider measuring only the particle after reflection . for destructive interference of the joint pdf illustrated in fig . [ fig : slab ] ( or destructive interference of peaks b in figs . [ fig : barrier_posenergy ] and [ fig : barrier_negenergy ] ) , such averaging yields a low probability of measuring only the neutron even if the slab is _ not _ measured . in this case , destructive interference is maintained in the one - body neutron pdf . the analogy between the measurement of a short pulse of light reflecting from a thin film and a measurement of only the neutron then has a similar physical interpretation . in both cases the detector need only be placed in the path of the moving neutron or the pulse of light , which leads to a one - body measurement that is independent of the wavegroup size . there is no requirement that the spatial resolution of the measuring instrument be smaller than the coherence length of the object being measured . as in the mirror case , this example of destructive interference in a one - body measurement ( which is derived from the two - body wavefunction ) is valid only for a range of coherence lengths for both the slab and particle . such coherence length dependence can be illustrated assuming a small reflection coefficient and using the mirror pdf of fig . [ fig : spreadvelocity]d for reflection from the first slab surface where the small slab coherence length is due to its mesoscopic / macroscopic mass . consider a similar figure to represent the `` mirror '' reflection from the second slab surface but offset by a slab thickness , @xmath102 , which is sufficient to prevent overlap of the reflected two - body wavefunctions from these two figures . this is a case where the particle does not experience interference even after having reflected from both slab surfaces and having a coherence length greater than the slab thickness . a simple estimate of this condition for the slab is given by again assuming that the incident two - body pdf of fig . [ fig : spreadvelocity]d is a rectangle of height @xmath208 and width @xmath209 . for a near sationary mirror with @xmath43 and @xmath104 , the rectangles reflected from each surface are tilted at an angle @xmath13 given by @xmath210 . the positions of the slab surfaces are essentially vertical in the @xmath219 plane . the two reflected rectangles just separate when @xmath220 . using the coherence length associated with thermal equilibrium of the slab with its environment yields the temperature condition for which there is no interference @xmath221 note the similarity with eqn . [ eq : slabtemp ] where the condition on @xmath222 is for interference rather than for lack thereof . such elimination of interference is due to the two particle - slab reflected states becoming distinguishable as the slab coherence length decreases in a manner similar to that discussed for the particle - mirror system in section [ sec : bandwidth ] and is only a consequence of treating both the particle and slab as quantum objects . if the slab is not treated as a quantum object then particle interference will always occur as long as the particle coherence length is greater than the slab thickness . a similar effect is found in the particle - well model . consider now the opposite extreme in the quantum character of the slab , that of small slab bandwidth or large uncertainty in the slab position . unlike the case of large mirror uncertainty where averaging over the mirror position in fig . [ fig : spreadvelocity]a ` washes out ' interference , uncertainty in the slab position does not eliminate the one - body - only measurement of the interference for the particle reflecting from both slab surfaces . this is a consequence of that interference not depending on the slab location . the inequality in eqn . [ eq : slabtemp2 ] is most easily satisfied by increasing the mass of the reflected particle ( increasing the recoil of the slab ) and/or that of the slabs spacing ( requiring an increase in the coherence length of the particle ) while decreasing the mass of the reflector ( increasing the slab recoil ) . reflection of a neutron traveling at @xmath223 m/s from a slab at @xmath224k , as in the example used above , yields no interference for @xmath225 kg . elimination of ` thin - film ' interference for a mesoscopic / macroscopic slab requires larger @xmath8 and @xmath102 ( with the size of the mesoscopic / macroscopic mass decreasing quadratically with these parameters ) . interference in a typical one - body interferometer ` washes - out ' as the fringe spacing becomes imperceptible with increasing mass . another mechanism that eliminates interference is a measurement to determine along which path the object traveled through the interferometer . the former issue was mitigated using two - body reflection of a microscopic particle from a mesoscopic / macroscopic object , as described above , while the latter is the first topic of this section . a method to determine path information and therefore destroy interference utilizes a ` heisenberg microscope , ' where an additional particle scatters from the object traversing the interferometer . if the wavelength of this particle , @xmath226 , is smaller than the spatial separation along the two paths , @xmath203 , then path information can be obtained in principle from the state of this scattered probe particle and interference is destroyed . this was verified using a three grating mach - zehnder interferometer traversed by an atom and using a photon as the probe to determine path information @xcite . the contrast of the interference , as a function of @xmath203 , was shown to drop dramatically for @xmath227 . generating such decoherence for particle - mirror interference involves using a probe particle to determine the mirrors position during interference . the ` two paths ' exist only during the time that the incident and reflected two - body wavegroup states overlap for reflection from a mirror . during this time there are two distances that the mirror could have moved : that associated with and that without reflection . if the probe particle can resolve the difference in these distances then the ` path ' associated with reflection is distinguishable from that without reflection . for a near stationary mirror with @xmath43 and @xmath104 , the mean difference in these mirror distances is @xmath228 , which when equal to the wavelength of a probe particle of mass @xmath229 then requires that the probe have velocity @xmath230 . this calculation can also be used to approximate the effect of such decoherence on the mirror from the collisions with gas atoms forming the thermal bath which surrounds the mirror if these gas atoms are considered to be the probe which determines path information and they scatter from only one of the superposed mirror states . the interference in reflection described above is then maintained if there is a small probability of these atoms having the velocity , @xmath231 . for a mesoscopic / macroscopic mirror and both a microscopic particle and a probe , @xmath203 is too small to be resolved by typical low temperature gas atoms . next this decoherence mechanism is applied to slab reflection . the two surface reflections cause a difference in slab displacement which was given in section [ sec : twodelta ] for a near sationary mirror with @xmath43 and @xmath104 , as @xmath232 . determining from which slab surface the particle reflects then requires that the probe have velocity @xmath233 . this decoherence mechanism can also be applied to the emission of photons by the slab between the times of particle reflection from each slab surface . if the photons wavelength is small enough to allow for the location of the slab associated with one of the reflections ( @xmath140 m in the neutron - slab example given above ) , interference is destroyed since it can then be determined at which slab surface the reflection occurred @xcite . obviously the probability of such a thermal photon emission is small . these calculations indicate that environmental decoherence associated with obtaining path information can potentially be mitigated . this is due in part to the displacements associated with the interfering reflector substates being proportional to @xmath234 . to measure such small displacements requires scattering particles with proportionally smaller wavelengths which are typically not found in the environment . it should be pointed out that the calculations presented above assume that states for the particle , mirror , and slab have a minimum spatial profile ( minimum uncertainty gaussian wavepackets ) . this will be referred to as a phased state . introducing a random phase into the fourier components of either one or both of the substates of such an incident two - body phased state affects the shape and size of the wavepacket while maintaining the same bandwidth and therefore the same coherence length . this is referred to as a dephased state . such random phase structure is introduced as an initial condition but could have been imposed via an earlier measurement which collapsed the reflector state into a dephased state . consider an incident phased particle substate and incident dephased mirror substate which then reflect . a one - body particle measurement after reflection can then act as a sensitive probe of this mirror substate . to illustrate this consider how the pdf of fig . [ fig : spreadvelocity]c changes for such a dephased state . first , the pdf is spread spatially along the @xmath3 axis . second , the magnitude of the interference part of the pdf is greatly reduced but not eliminated . finally , the pdf is not spatially uniform . apart from its magnitude and spatial uniformity , this pdf is more similar to fig . [ fig : spreadvelocity]a than [ fig : spreadvelocity]c . this is not surprising since the localization of the dephased mirror substate is similar to that of the mirror in fig . [ fig : spreadvelocity]a . results of a particle - only measurement are predicted from an integral ( trace ) over the mirror coordinate of this pdf which then ` washes - out ' this interference . if the mirror is in a dephased state , a reflecting particle - only measurement will not exhibit interference in the overlap region . however , such interference is expected both when treating the mirror as a classical potential and for a phased mirror substate with the coherence lengths similar to those shown in fig . [ fig : spreadvelocity]c . the spatial extent of such a dephased mirror substate can vary between being unlocalized to being the minimum size determined by the bandwidth . this size can then be probed by varying the wavelength of the incident particle in a reflected particle - only measurement , in which case interference either disappears or reappears as this wavelength becomes smaller or larger than the spatial size of the dephased mirror substate . yet another decoherence mechanism is through interaction with the environment . zurek predicted exponential decay of the off - diagonal density matrix element terms for a macroscopic object in a superposition state , @xmath235 $ ] , with time constant @xmath236 where @xmath237 is the thermal relaxation time , @xmath203 is the difference in distance between these two states , and @xmath238 is the thermal de broglie wavelength of the macroscopic object @xcite . the decay is into a mixed rather than a pure state . it is interesting to note that the lack of interference predicted in section [ subsec : onlyparticle ] also depends on @xmath203 but for very different reasons . applying this decoherence model to the macroscopic slab states after reflection of the neutron , as described above , yields @xmath239 . for these neutron parameters this ratio is large ( @xmath240 for @xmath241 kg at @xmath242k ) due to the very small offset introduced from the recoil ( @xmath243 ) . the parameter of interest , @xmath244 , is also affected by @xmath237 which can vary dramatically depending on environmental and material properties along with the dimensional values of the slab . mesoscopic masses can have short thermal relaxation times . nevertheless , it appears that environmental decoherence for reflection from a slab can easily be made negligible in this model of decoherence . applying this decoherence model to the mesoscopic / macroscopic mirror states during correlated interference in reflection allows a greater variation of parameters since atoms or molecules can more easily reflect from a surface than from both surfaces of a ` thin film . ' the result , @xmath245 , therefore allows environmental decoherence for the mirror to more easily be made either negligible or non - negligible . consider next the consequences of such decoherence on correlated interference where the particle and mirror are in superposition states of both having and not having reflected . if this mirror substate decoheres into a mixed state then it seems reasonable to assume that the commensurate particle substate must also decohere into a mixed state since the two substates are correlated via conservation of energy and momentum . measurement of only the particle , after the mirror has decohered , therefore yields no interference since the particle can not then be in a superposition state . applying this result to a measurement of only the neutron flux reflecting from the slab , while varying the incident neutron s wavelength , results in no interference if the slab decoheres during the time between reflection from both surfaces . the details of actually executing such a measurement and of choosing parameters which straddle the region where decoherence is and is not applicable are of course much more complicated and are beyond the scope of this work . braginsky and khalili also treat environmental decoherence @xcite . however , they predict that for a macroscopic oscillator to exhibit quantum effects both the mechanical and thermal relaxation times are important while the parameter @xmath203 is not incorporated in their model . they argue that quantum effects can indeed be exhibited for macroscopic high q resonators . it has been shown that correlated interference in reflection does not disappear with increasing mirror mass , is difficult to eliminate via a path measurement , and can be made insensitive to environmental decoherence . it should also be pointed out that the robust character of interference for objects with many degrees of freedom is reinforced by measurements which demonstrate that even if the size of the object is larger than both the coherence length and debroglie wavelength , interference can still be observed @xcite . the use of reflection to generate perceptible interference in a mesoscopic / macroscopic object involves a microscopic particle interacting with a mesoscopic / macroscopic reflector to create two reflector substates which differ only by the microscopic momentum exchanged . unlike the fringes for a particle traversing a double slit , the interference fringes for the reflector substates do not vanish in the limit of large reflector mass . in addition , particle and reflector interference is required for conservation of energy and momentum to be satisfied . for example , consider reflection of a neutron from a slab . ` thin - film ' interference yields peak b in the two - body pdf of fig . [ fig : barrier_negenergy ] . destructive interference of the reflected neutron substate , which is certainly expected in a one - body treatment , must be associated with destructive interference of the slab substate ( peak b must not exist ) . that is , if the neutron can not be measured in reflection then the slab can never be measured at the position associated with it either having or not having reflected the neutron . similarly , by only changing the neutron wavelength to generate constructive thin - film interference in reflection , the slab must then have a commensurate probability to be found in a position of having reflected the neutron . measurement of interference in the reflected particle then requires that the reflector be in a superposition state which interferes in a manner correlated with that of the particle . this conclusion also holds for the interference shown in fig . [ fig : interference ] and presumably for all two - body interferometers . existence of such interference might be disputed since the difference in wavevectors of the interfering mirror substates , due to reflecting a microscopic particle , is much smaller than the spread in wavevectors of which the mirror wavegroup is composed , due to its thermal motion . this can be expressed as the ratio @xmath246 . interference , however , is not determined by @xmath247 but rather by the relation between the interferometer path difference and the coherence lengths of the interfering wavegroups . for a neutron reflecting from a mirror of @xmath248 kg , @xmath249 m / s , and @xmath218 k , the ratio @xmath250 yields a number which is of the same order of magnitude as is this ratio for an optical pulse of duration @xmath251 s ( from a mode locked ti - sapphire laser ) traversing a mach - zehnder interferometer , with the mirror in one arm moving at speed @xmath252 m / s ( all other interferometer components are static ) . optical pulse trains exhibit interference in similar interferometers @xcite with smaller @xmath247 values than in the particle reflection example just given . correlated interference in reflection is an example of one of the simplest interferometers , utilizing neither division of amplitude nor division of wavefront methods to generate interference . in addition , path lengths need not be carefully matched for interference to be manifest . the complexity of this interferometer is revealed as the coherence lengths of the particle and mirror substates are varied . examples are : ( 1 ) when both are small , classical reflection rather than interferometry essentially occurs . ( 2 ) when both are large , compared with the particle s wavelength , reflection becomes an interferometer yielding correlated interference , as shown in fig . [ fig : spreadvelocity]a , while no interference is observed in a one - body - only measurement of either the particle or mirror . due to uncertainty in the location of both objects , a measurement of the particle at @xmath2 then interferometrically correlates with a range of position measurements of the mirror . ( 3 ) when the mirror s displacement , due to recoil in reflection , is less than the mirrors coherence length and the coherence length of the particle is larger than that of the mirror , then a measurement of only the particle yields the expected interference from a one - body schrdinger equation treatment ( an example is shown in fig . [ fig : spreadvelocity]c ) . ( 4 ) when the coherence length of the particle is much larger than that of the mirror and mirror recoil is larger than its coherence length , reflection becomes a two - body beamsplitter . this is illustrated in fig . [ fig : spreadvelocity]d for @xmath91 . here two states are generated but do not interfere , one being the particle - mirror state before reflection and the other that after reflection , similar to a beamsplitter producing photon states which traverse spatially separate and therefore distinguishable paths . for example , at fixed @xmath2 there is a bimodal pdf along the @xmath3 axis , each peak of which corresponds to one of these two states . there is no interference since the short coherence length of the mirror relative to the recoil distance results in distinguishable two - body states ( the mirror has reflected the particle in the upper but not the lower pdf peak ) . the kinematics in the correlated interference regime can be compared with that of an electromagnetic wavegroup reflecting classically from a mirror . the latter involves accelerated motion due to a continuous force delivered to the mirror throughout the reflection process . to determine the former , consider the interference region shown in fig . [ fig : interference ] with @xmath253 , @xmath254 , and equal coherence lengths . if only the mirror is measured ( integrating the pdf over the particle coordinate ) then the expectation value for the mirrors position accelerates throughout the interaction time , as expected classically . if the mirror is not measured but the particle is , then the expectation value for the mirrors position decelerates at the same rate . however , these expectation value kinematics vary with parameters such as particle and mirror substate coherence lengths and masses . finally , the issue of how to determine that the mirror is in a superposition of states is considered . one method is to perform a correlation measurement of the positions of the particle and cm of the mirror in the interference region . in a region of destructive interference neither will be found . however , correlation measurements are difficult to perform . another possibility is to use a third particle as a probe , interacting only with the mirror to determine if the mirror is in a superposition state , in the following way : let this microscopic probe particle reflect from both mirror states ( reflecting from the other side of the mirror ) resulting in one - body interference of the probe . the coherence length of the probe need only be larger than the very small displacement of the mirror states for this probe interference to be exhibited , while the magnitude of the phase shift is determined by the mass of the probe . this is a method to measure that the mesoscopic / macroscopic mirror is in a superposition state , without any direct measurement of the mirror . however , calculations of such three - body interactions , which follow easily from the formalism discussed above , are beyond the scope of this already lengthy foundational work . methods to measure quantum effects on a mesoscopic / macroscopic reflector using only the reflected particle have been a focus of this work , rather than the methods just described . the reflected particle flux is then used to reveal properties of the quantum state of the reflector . one example is of a neutron reflecting from a slab . as with interference in reflection of light , the one - body schrdinger equation predicts a variation in reflected neutron flux as a function of incident neutron wavelength whenever the neutron coherence length is longer than the slab thickness . the calculation presented here , treating both the particle and reflector as quantum objects , also predicts such interference but only as a special case . in addition , this treatment predicts the following regimes under which no reflected interference occurs _ for a measurement of only the particle _ : ( 1 ) the reflection process acts as a beamsplitter and not an interferometer , generating two mirror or slab states which do not overlap . ( 2 ) the fourier components of the mirror substate have random phases sufficient to extend its size beyond that of the particle s wavelength . ( 3 ) the slab and mirror have coherence lengths long enough for correlated interference but not interference when only the particle is measured . in each of these cases , consider varying the incident particle s wavelength and observing no interference in the reflected particle flux . this indicates that the mesoscopic / macroscopic reflector is in a particular quantum superposition state . by varying parameters appropriately , as discussed in section [ sec : particleonlymirror ] , interference in this reflected particle flux then reappears . ironically , it is this lack of interference that then reveals the quantum nature of this mesoscopic / macroscopic ` thin film ' or mirror . in this case , measuring only the reflected particle flux yields significant experimental advantages over a measurement of the correlation between the particle and mirror . another possible method to study the state of the mesoscopic / macroscopic mirror is via the expected standing wave pattern generated between incident and retro - reflected states of a bose - einstein condensate as it interacts with a mirror . if this process can be modeled as described above , with the mass of the condensate replacing that of the particle , then the point at which condensate interference disappears can be extended to larger mirror masses . although interference between split condensates has been observed @xcite , as far as we are aware , there is no evidence of standing wave interference due to reflection . reflection of atoms from mirrors has been previously studied but not in treating both the particle and reflector as quantum objects nor in the reflector mass regimes discussed here nor in an attempt to measure either the resulting ` standing - wave ' interference or the effect of decoherence on such interference . there exist few examples of quantum correlated interference for non - zero rest mass particles . verification of the above predictions , even in the microscopic regime for both particle and reflector , is therefore of fundamental interest . although far from being comprehensive , these results indicate a direction , heretofore unexplored , for further research in understanding quantum correlation , probing decoherence , and extending quantum measurements to larger masses . bell , physics * 1 * 195 ( 1964 ) . k. gottfried , am . * 68 * 143 ( 2000 ) . d. salart , et . al . , phys . * 100 * 220404 ( 2008 ) . m. a. rowe , et . nature , * 409 * 791 ( 2001 ) . m. ansmann , et . al . , nature , * 461 * 504 ( 2009 ) . a. hagar , phil . r. soc . * 370 * 4425 ( 2012 ) ; m.schlosshauer , rev . of mod 76 * 1267 ( 2004 ) . a. d. cronin , j. schmiedmayer , and d. e. pritchard , rev . phys . * 81 * , 1051 ( 2009 ) . t. juffmann et . al . , nature nanotech * 7 * , 297 ( 2012 ) . b. abbott , et . new j. of phys . * 11 * 073032 ( 2009 ) . m. aspelmeyer , p. meystre , and k. schwab , phys . today * 65 ( 7 ) * , 29 ( 2012 ) . a.d . oconnell et.al . , nature * 464 * , 697 - 703 ( 2010 ) . a. nunnenkamp , k. borkje , and s. m. girvin , phys . lett . * 107 * , 063602 ( 2011 ) . f. de martini et.al . * 104 * , 050403 ( 2010 ) . d. kouznetsov and h. oberst , phys . a * 72 * , 013617 - 1 ( 2005 ) . t. a. pasquini , m. saba , g .- b . jo , y. shin , w. ketterle , and d. e. pritchard , phys . 97 * , 093201 - 1 ( 2006 ) . f. shimizu , phys . lett . * 86 * , 987 ( 2001 ) . t. hils et . al . , phys . a * 58 * , 4784 ( 1998 ) . y. colombe , b. mercier , h. perrin , and v. lorent , phys . a * 72 * , 061601 ( 2005 ) . szriftgiser , p. , d. guery - odelin , m. arndt , and j. dalibard , phys . lett . * 77 * , 4 ( 1996 ) . h. rauch , contemp . * 2 * 345 ( 1986 ) . kowalski , f.v . and browne , arxiv:1405.0031 [ quant - ph ] p. tommasini et.al . , am . * 66 * , 881 ( 1998 ) . siverman , _ more than one mystery _ , p-96 ( springer verlag , new york , 1995 ) . in the particle - mirror coordinates the incident and reflected wavevectors for the particle ( or mirror ) are not necessarily equal and opposite as they are in the cm / rel coordinates ( nor are they equal and opposite for a one particle system in which the mirror is fixed ) . the probability current density to the left of the mirror then exhibits spatial and temporal interference due to the effect this difference in wavevectors has on the incident and reflected wavefunctions . this results in a non - additive current density , a consequence of the space to the left of the mirror increasing as it moves . the net flux must account for the build - up of probability within this `` new '' space . pushin et . al . , phys . 100 * , 250404 ( 2008 ) . hamilton , a.g . klein , and g.i . opat , phys . rev . a , * 28 * , 3149 ( 1983 ) . the coherence length does not change even if the size of the wavepacket changes due to dispersion . klein , g.i . opat , and w.a . hamilton , phys . * 50 * , 563 ( 1982 ) . m. ferrero and a. van der merwe , _ fundamental problems in quantum physics _ , ( kluwer academic publishing , dordrecht , 1995 ) . the pauli exclusion principle is another example of destructive interference in a two - particle system . see r. c. liu , b. odom , y. yamamoto , and s. tarucha , nature 391 , 263 - 265 ( 1998 ) . c.u.serge and j.d . sullivan , am . * 44 * 729 ( 1976 ) . the coherence length is used here rather than the wavepacket size since the former rather than the later determines interference , op . s.m . barnett and r. loudon , phil . r. soc.a * 368 * 927 ( 2010 ) . chapman , et.al . , * 75 * , 3783 , ( 1995 ) . w. h. zurek , rev . * 75 * , 715 , ( 2003 ) . v.b . braginsky and f.y . khalili , _ quantum measurement _ , ( cambridge university press 1992 ) . l. lepetit , g. cheriaux , and m. joffre , j. opt . b * 12 * , 2467 ( 1995 ) . c. kohstall , s. riedl , e. s. guajardo , l. sidorenkov , j. h. denschlag , r. grimm , new j. phys . * 13 * , 065027 ( 2011 ) m. witkowski , r. gartman , b. nagrny , m. piotrowski , m. podzie , k. sacha , j. szczepkowski , j. zachorowski , m. zawada , and w. gawlik , phys . a * 88 * , 025602 ( 2013 ) if the condensate particles reflect individually from a receding mirror then they each acquire a different energy which decreases in their order of reflection . this is due to the energy exchanged with the mirror from the particles which have already reflected and is manifest as a distorted reflected wavegroup . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the study of the formation of population iii stars in a cosmological context via high - resolution simulations is becoming a mature discipline , with multiple groups finding similar results ( bromm , coppi , & larson 2002 ; bromm & loeb 2004 ; abel , bryan , & norman 2002 ; oshea & norman 2007 ; yoshida et al . 2006b ; gao et al . however , all of these calculations operate under a common fundamental assumption namely , the absence of an ultraviolet background . given the abundance of halos in which population iii stars would form and the general consensus that these stars must be massive , one expects that , though only a small fraction of the volume of the universe would be ionized , a significant background of ultraviolet radiation in the lyman - werner ( lw ) band ( 11.18 - 13.6 ev ) , which is capable of photodissociating molecular hydrogen , would be present for the formation of the bulk of population iii stars ( omukai 2001 ; yoshida et al . 2003 ; machacek , bryan , & abel 2001 ; wise & abel 2005 ) . given that atomic hydrogen is optically thin to this radiation and that it easily destroys molecular hydrogen , this could have a significant impact on the formation of primordial stars . ( 2001 ) used cosmological amr simulations to study the formation of primordial stars in the presence of a soft uv background , and later including an x - ray background ( machacek , bryan , & abel 2003 ) . they found that the lw background delays the formation of population iii stars and shifts halo formation to higher masses . however , their calculations were not of sufficiently high resolution to give any significant information about the formation of the protostellar cloud at the core of each collapsing cosmological halo . ( 2003 ) performed a suite of large sph simulations of high redshift structure formation , including two calculations with a soft uv background with @xmath6 and @xmath7 , where @xmath8 is the mean intensity of the uv background in the lw band in units of @xmath9 erg s@xmath10 @xmath11 hz@xmath10 sr@xmath10 . they found that gas cooling is suppressed in the higher lw background case , and predicted that star formation would not occur . on the other hand , & haiman ( 2002 ) showed analytically that primordial gas in halos with t@xmath12 k can still collapse to high densities in the presence of a strong molecular hydrogen - dissociating background via atomic hydrogen line radiation , and can eventually form h@xmath13 . they referred to these objects as second generation objects " in the sense that although still of primordial composition , their formation pathway is different from the first stars . an issue that needs to be addressed concerns the mode of primordial star formation in halos with t@xmath5 k and above . it is apparent from recent studies using large numbers of simulations that smaller halos , with masses of @xmath14 m@xmath15 , appear to always form primordial stars in isolation ( gao et al . 2006 ; oshea & norman 2007 ) . is this still the case in halos whose masses are an order of magnitude or more larger ? does a single primordial star form , or several ? this is particularly relevant given that many semi - analytic models of structure formation that follow the chemical evolution of structures at high redshift ( scannapieco et al . 2005 ; schneider et al . 2006 ) suggest that halos of this scale which have not been enriched by metals ( and thus will form primordial stars ) exist up to at least @xmath16 . if this is true , there is expected to be a strong photodissociating background at that time ( yoshida et al . 2003 ; wise & abel 2005 ) . this paper is the second in a series . in & norman ( 2007 ) ( hereafter referred to as paper i ) , we examined several aspects of population iii star formation in a @xmath17cdm universe , in the absence of an ultraviolet background which photodissociates molecular hydrogen . in this paper , we study population iii star formation in a single cosmological realization varying the strength of the lw background , with the goal of investigating the effect that this background has on the evolution and properties of gas in the halo core ( at radii @xmath18 pc ) and the ultimate fate of the gas at the center of t@xmath5 k halos . we use a single cosmological realization , varying the strength of the photodissociating background over the range suggested by & abel ( 2005 ) , and look for the cutoff in population iii star formation suggested by ( 2003 ) . in agreement with the findings of ( machacek et al . 2001 ) , we find that the addition of a photodissociating background results in a delay of the collapse of high density gas at the center of the most massive halo in the simulation and , as a result , an increase in the virial mass of this halo at the onset of baryon collapse . we find that , contrary to the results of ( 2003 ) , star formation is not suppressed for j@xmath0 , but occurs even with backgrounds as high as j@xmath1 . we find that 2 cooling leads to collapse despite the depressed halo core molecular hydrogen fractions f@xmath19 by two multiplicative effects : ( 1 ) the elevated 2 cooling rates per molecule at @xmath2 k , and ( 2 ) time . we find that halo core collapse occurs in the usual way once the gas in the halo core has become dense enough that the cooling time becomes much less than the hubble time . we also observe a relationship between the strength of the photodissociating background and the rate of accretion onto the evolving protostellar cloud core , with higher lw background fluxes resulting in higher accretion rates . this is a simple consequence of the suppression of molecular hydrogen formation ( and thus suppression of cooling ) by the photodissociating background , as well as the higher virial temperatures of the more massive halos at the epoch of collapse . this may have implications for the range of population iii stellar masses . finally , we find that the collapsing halo cores in our simulations do not fragment at densities below @xmath3 @xmath4 regardless of the strength of the soft uv background , suggesting that population iii stars forming in halos with t@xmath5 k may still form in isolation . the organization of this paper is as follows . in section [ methodology ] we provide a description of enzo , the code used to perform the calculations in this paper and of the simulation setup . the results from our simulations are presented in sections [ results.meanprop ] through [ results.rephalos ] : section [ results.meanprop ] discusses some of the mean halo properties observed in the calculations , section [ results.evolution ] discusses the evolution of the halo core prior to collapse , section [ results.collapse ] compares spherically - averaged halo properties for all simulations at the epoch of collapse , section [ results.fixred ] discusses a variety of halo properties at a fixed redshift , and section [ results.rephalos ] compares the evolution of two representative simulations taken from our ensemble . in section [ sect.issues ] we discuss neglected physics and possible numerical issues , and in section [ discuss ] we discuss some of the results presented in this work and their implications . finally , we present a summary of the main results in section [ summary ] . ` enzo ' is a publicly available , extensively tested adaptive mesh refinement ( amr ) cosmology code developed by greg bryan and others ( bryan & norman 1997a , 1997b ; norman & bryan 1999 ; oshea et al . 2004 , 2005b ) . the specifics of the enzo code are described in detail in these papers ( and references therein ) , but we present a brief description here for clarity . the enzo code couples an n - body particle - mesh ( pm ) solver ( efstathiou et al . 1985 ; hockney & eastwood 1988 ) used to follow the evolution of a collisionless dark matter component with an eulerian amr method for ideal gas dynamics by & colella ( 1989 ) , which allows high dynamic range in gravitational physics and hydrodynamics in an expanding universe . this amr method ( referred to as _ structured _ amr ) utilizes an adaptive hierarchy of grid patches at varying levels of resolution . each rectangular grid patch ( referred to as a `` grid '' ) covers some region of space in its _ parent grid _ which requires higher resolution , and can itself become the parent grid to an even more highly resolved _ child grid_. enzo s implementation of structured amr places no fundamental restrictions on the number of grids at a given level of refinement , or on the number of levels of refinement . however , owing to limited computational resources it is practical to institute a maximum level of refinement , @xmath20 . additionally , the enzo amr implementation allows arbitrary integer ratios of parent and child grid resolution , though in general for cosmological simulations ( including the work described in this paper ) a refinement ratio of 2 is used . since the addition of more highly refined grids is adaptive , the conditions for refinement must be specified . in enzo , the criteria for refinement can be set by the user to be a combination of any or all of the following : baryon or dark matter overdensity threshold , minimum resolution of the local jeans length , local density gradients , local pressure gradients , local energy gradients , shocks , and cooling time . a cell reaching any or all of the user - specified criteria will then be flagged for refinement . once all cells of a given level have been examined , rectangular solid boundaries are determined which minimally encompass the flagged cells on that level . a refined grid patch is then introduced within each such bounding volume , and the results are interpolated to a higher level of resolution . in enzo , resolution of the equations being solved is adaptive in time as well as in space . the timestep in enzo is satisfied on a level - by - level basis by finding the largest timestep such that the courant condition ( and an analogous condition for the dark matter particles ) is satisfied by every cell on that level . all cells on a given level are advanced using the same timestep . once a level @xmath21 has been advanced in time @xmath22 , all grids at level @xmath23 are advanced , using the same criteria for timestep calculations described above , until they reach the same physical time as the grids at level @xmath21 . at this point grids at level @xmath23 exchange baryon flux information with their parent grids , providing a more accurate solution on level @xmath21 . cells at level @xmath23 are then examined to see if they should be refined or de - refined , and the entire grid hierarchy is rebuilt at that level ( including all more highly refined levels ) . the timestepping and hierarchy rebuilding processes are repeated recursively on every level to the maximum existing grid level in the simulation . two different hydrodynamic methods are implemented in enzo : the piecewise parabolic method ( ppm ) ( woodward & colella 1984 ) , which was extended to cosmology by ( 1995 ) , and the hydrodynamic method used in the zeus magnetohydrodynamics code ( stone & norman 1992a , 1992b ) . we direct the interested reader to the papers describing both of these methods for more information , and note that ppm is the preferred choice of hydro method since it is higher - order - accurate and is based on a technique that does not require artificial viscosity , which smoothes shocks and can smear out features in the hydrodynamic flow . the chemical and cooling properties of primordial ( metal - free ) gas are followed using the method of ( 1997 ) and ( 1997 ) . this method follows the non - equilibrium evolution of a gas of primordial composition with 9 total species : @xmath24 , @xmath25 , @xmath26 , @xmath27 , @xmath28 , @xmath29 , @xmath30 , @xmath31 , and @xmath32 . the code also calculates radiative heating and cooling following atomic line excitation , recombination , collisional excitation , free - free transitions , molecular line cooling , and compton scattering of the cosmic microwave background , as well as any of approximately a dozen different models for a metagalactic ultraviolet background that heat the gas via photoionization and/or photodissociation . we model the cooling processes detailed in ( 1997 ) , but use the & palla ( 1998 ) molecular hydrogen cooling function . the multispecies rate equations are solved out of equilibrium to properly model situations where , e.g. , the cooling time of the gas is much shorter than the hydrogen recombination time . a total of 9 kinetic equations are solved , including 29 kinetic and radiative processes , for the 9 species mentioned above . the chemical reaction equation network is technically challenging to solve due to the huge range of reaction time scales involved ; the characteristic creation and destruction time scales of the various species and reactions can differ by many orders of magnitude . as a result , the set of rate equations is extremely stiff , and an explicit scheme for integration of the rate equations can be costly if small enough timestep are taken to keep the network stable . this makes an implicit scheme preferable for such a set of equations , and enzo solves the rate equations using a method based on a backwards differencing formula ( bdf ) in order to provide a stable and accurate solution . it is important to note the regime in which this chemistry model is valid . according to ( 1997 ) and ( 1997 ) , the reaction network is valid for temperatures between @xmath33 k. the original model discussed in these two references is only valid up to n@xmath34 @xmath4 . however , addition of the 3-body h@xmath13 formation process allows correct modeling of the gas chemistry up until the point where collisionally induced emission from molecular hydrogen becomes an important cooling processes , which occurs at @xmath35 @xmath4 . we do not include heating by molecular hydrogen formation , which will be significant at densities of @xmath36 @xmath4 and above , and may affect temperature evolution at these high densities . a further concern is that the optically thin approximation for radiative cooling breaks down beginning at n@xmath37 @xmath4 . beyond this point , modifications to the cooling function that take into account the non - negligible opacity of the gas to line radiation from molecular hydrogen must be made , as discussed by & abel ( 2004 ) . even with these modifications , a more correct description of the cooling of gas of primordial composition at high densities will require some form of radiation transport , which will greatly increase the cost of the simulations . the simulations discussed in this paper are set up in a similar way to those in paper i. a dark matter - only calculation with @xmath38 particles in a three - dimensional simulation volume which is @xmath39 h@xmath10 mpc ( comoving ) on a side is set up at @xmath40 assuming a `` concordance '' cosmological model with no baryons : @xmath41 , @xmath42 , @xmath43 , @xmath44 ( in units of 100 km / s / mpc ) , @xmath45 , and using an eisenstein & hu power spectrum ( eisenstein & hu 1999 ) with a spectral index of @xmath46 . the cold dark matter cosmological model is assumed . this calculation is then evolved to @xmath47 using a maximum of four levels of adaptive mesh refinement , refining on a dark matter overdensity of 8.0 . at @xmath47 , the hop halo finding algorithm ( eisenstein & hut 1998 ) is used to find the most massive halo in the simulation . at this point , we generate a new set of initial conditions which contain the same large - scale power as the dark matter - only calculation , but include both dark matter and baryons such that the lagrangian volume in which the halo in the dark matter - only calculation formed is resolved at high spatial and mass resolution using a series of static nested grids , with a @xmath38 root grid and three static nested grids , for an overall effective root grid size of @xmath48 cells . the highest resolution grid is @xmath49 grid cells , and corresponds to a volume @xmath50 h@xmath10 comoving kpc on a side . the dark matter particles in the highest resolution grid are 14.48 h@xmath10 m@xmath15 and the spatial resolution of the highest resolution grid is 586 h@xmath10 parsecs ( comoving ) . previous work shows that this particle mass resolution is adequate to fully resolve the collapse of the halo ( abel et al . 2002 ; oshea & norman 2007 ) . all simulations are performed using the adaptive mesh cosmology code enzo , which is described in section [ enzocode ] . the simulations are started at @xmath40 and allowed to evolve until the collapse of the gas within the center of the most massive halo , assuming the presence of an unevolving soft uv background with intensities in the lyman - werner band of j@xmath51 and @xmath9 erg s@xmath10 @xmath11 hz@xmath10 sr@xmath10 . this range covers a much wider range of parameter space than the results described in ( 2001 ) , and completely encompasses the range of possible j@xmath52 values suggested by & abel ( 2005 ) , for a wide range of mean population iii stellar masses . note that many publications use f@xmath52 instead of j@xmath52 : f@xmath52 has units of erg s@xmath10 @xmath11 hz@xmath10 , and thus f@xmath53 j@xmath52 . we will use j@xmath52 throughout this paper , and for convenience express values in units of j@xmath54 , where @xmath8 is the mean intensity of the uv background in the lw band in units of @xmath9 erg s@xmath10 @xmath11 hz@xmath10 sr@xmath10 . the equations of hydrodynamics are solved using the ppm method with a dual energy formulation , which is required to adequately resolve the thermal properties of gas in high - mach flows . the nonequilibrium chemical evolution and optically thin radiative cooling of the primordial gas is modeled as described in section [ enzocode ] , following 9 separate species including molecular hydrogen ( but excluding deuterium ) , with an initial electron fraction of @xmath55 ( which is roughly consistent with ( 1968 ) for an @xmath56 , @xmath57 universe ) . note that the initial electron fraction in the calculation is relatively unimportant to the molecular hydrogen formation rates in halo cores , as the electron fraction at the center of a given halo is controlled primarily by mergers and the shock formed by accretion of gas onto the halo . adaptive mesh refinement is used throughout the innermost high resolution region such that cells are refined by factors of two along each axis , with a maximum of 22 total levels of refinement . this corresponds to a maximum spatial resolution of 115 h@xmath10 au ( comoving ) at the finest level of resolution , with an overall spatial dynamical range of @xmath58 . to avoid effects due to the finite size of the dark matter particles , the dark matter density is smoothed on a comoving scale of @xmath59 pc ( which corresponds to @xmath60 0.03 proper pc at @xmath61 ) . this is reasonable because at that scale in all of our calculations the gravitational potential in the halo of interest is completely dominated by the baryons . grid cells are adaptively refined based upon several criteria . we refine on baryon and dark matter overdensities in cells of 4.0 and 8.0 , respectively . this corresponds to a maximum mass of gas or dark matter per cell ( on the most highly refined static grid ) of m@xmath62 and @xmath63 m@xmath15 , respectively . in addition , the _ minimummassforrefinementlevelexponent _ parameter is set to @xmath64 for both the dark matter and baryon overdensity refinement criteria , meaning that the mass required to refine to a higher level decreases as a function of increasing level , as : @xmath65 where l is the current level of refinement . the negative exponent means that the mass resolution in the calculations is super - lagrangian for example , m@xmath66 m@xmath67 . in addition to refining on baryon and dark matter overdensity , these simulations include refinement criteria which ensure that shocks are always well - resolved , that the cooling time in a given cell is always longer than the sound crossing time of that cell , and that the jeans length is always resolved by at least 16 cells . this last criterion guarantees that the truelove criterion ( truelove et al . 1997 ) is always resolved by a factor of four more cells in each dimension than is strictly necessary , ensuring that no artificial fragmentation will take place . figures [ fig.image_jlw_0 ] and [ fig.image_jlw_1em21 ] show projections of baryon density and temperature for the j@xmath68 and j@xmath1 calculations , respectively , at the epoch at which each calculation collapses , defined as the redshift at which the baryon number density reaches @xmath69 @xmath4 . note that due to the rapid evolution of gas at high density , the `` collapse redshift '' depends very weakly on the exact choice of density threshold . the calculations are started from the same set of initial conditions , and the j@xmath1 calculation is clearly a later stage in the evolution of the j@xmath68 run . the satellite halos surrounding the primary halo which collapses at z=24.12 in the j@xmath68 run have merged by z=17.32 when the halo in the j@xmath1 case collapses ( cf . [ fig.merger ] ) . with virial masses of @xmath70 and @xmath71 , respectively , the latter halo is more than 20 times as massive as the former . though quite different in mass , both halos exhibit similar morphologies they are extremely centrally - concentrated , and only a single condensed object ( that is to say , a primordial protostellar cloud core ) is visible in the highest - resolution panel in each image . the existence of a single cloud core is common to all calculations discussed in this paper . simulation ( @xmath72 ; m@xmath73 m@xmath15 ) . top row : projected log baryon density . bottom row : projected , mass - weighted log baryon temperature . left column : region 2.13 kpc ( proper ) across and deep . middle column : region 133.3 pc ( proper ) across and deep . right column : region 0.130 pc ( proper ) across and deep . the middle and right column are zoomed images focusing on the high baryon density core which forms in the center of the halo . , scaledwidth=90.0% ] simulation ( @xmath74 ; m@xmath75 m@xmath15 ) . top row : projected log baryon density . bottom row : projected , mass - weighted log baryon temperature . left column : region 2.13 kpc ( proper ) across and deep . middle column : region 533 pc ( proper ) across and deep . right column : region 0.260 pc ( proper ) across and deep . the middle and right column are zoomed images focusing on the high baryon density core which forms in the center of the halo . , scaledwidth=90.0% ] figure [ fig.halovals1 ] shows several mean halo quantities for all simulations discussed in this paper , including halo collapse redshift as a function of j@xmath52 , virial mass as a function of j@xmath52 , virial temperature as a function of j@xmath52 , and the virial mass as a function of collapse redshift . simulations where the soft uv background is turned on are shown by solid squares , and the `` control '' j@xmath68 calculation is shown by an open square . in plots where j@xmath52 is shown on the x - axis , the j@xmath68 simulation is placed at log@xmath76 . figure [ fig.halovals1 ] shows that there is a clear relationship between the lw intensity and the collapse redshift and virial mass of the halo . a larger lw intensity results in a later collapse time and larger virial mass because the halo must be hotter in order to have a cooling time which is less than a hubble time despite the depressed 2 abundance in the high - density gas at the halo s center . this is discussed further in section [ results.evolution ] the final mass of the halos in the simulations with j@xmath1 is approximately a factor of 20 higher than that in the `` control '' simulation . this is in qualitative agreement with ( 2001 ) , who suggest that there is a `` minimum halo mass '' as a function of the strength of the lw background of the form @xmath77 this threshold mass is plotted in panel ( c ) of figure [ fig.halovals1 ] . [ eqn - mach - mthresh ] is only strictly valid over the range @xmath78 , because this was the range simulated by ( 2001 ) . we see that our points for @xmath79 parallel the threshold curve but at a mass approximately four times higher . this difference can be ascribed to two factors . first , the threshold mass is the lowest possible halo mass that can collapse , derived from a statistical sample . since we study only one realization which focuses on the most massive halo in the box , its mass is bound to be higher than the statistical minimum . second , ( 2001 ) s criterion for cooling catches halos at an earlier stage of evolution compared to our data points , which give the halos virial masses at the time of central baryon collapse ( the `` collapse redshift '' ) . examination of panels ( b ) and ( c ) of figure [ fig.halovals1 ] show that there is some sort of `` phase change '' between j@xmath80 and @xmath81 causing the halo mass at the time of collapse to increase steeply and non - monotonically . the collapse redshift steadily decreases as a function of uv increasing background strength ( i.e. collapse of gas at the center of the halo is delayed ) . however , panel ( c ) shows that the virial mass steadily increases with increasing uv background strength until j@xmath80 , at which point there is a jump in mass by more than a factor of four , above which the halo mass increases only slightly with increasing uv background strength . that something interesting should happen at these lw intensities is consistent with the findings of ( 2003 ) who found that 2 cooling is strongly suppressed at j@xmath82 . they predicted that star formation would be inhibited since the equilibrium 2 fraction is below the critical fraction for cooling derived by ( 1997 ) . contrary to these predictions , we find collapse not only at j@xmath82 , but also at values as large as j@xmath83 . we note that highest value considered by ( 2001 ) was j@xmath84 , and the highest simulated by ( 2003 ) was j@xmath82 . therefore our cases j@xmath85 have not been examined before , and certainly not at the resolution of our simulations . how is the gas in the center of these halos able to cool and collapse in such high radiation backgrounds ? this is analyzed in some detail in the next section . a hint is provided in ( d ) of figure [ fig.halovals1 ] , which plots the halo virial temperature as a function of j@xmath52 . at background strengths above j@xmath80 , the virial temperature is consistently approximately @xmath86 k , which is roughly the temperature at which atomic hydrogen cooling is effective . however , an examination of the radial temperature profiles shows that high - density gas in the center of the halo never reaches these temperatures , but is more typically 2000 k. atomic line cooling is unimportant at such low temperatures ; however the 2 cooling rate per particle is roughly 100 times as large at 2000k as at 500k . 2 cooling still operates in the halo centers despite low 2 abundance due to the higher cooling rates and long evolutionary timescales . in the rest of this section we merely present additional mean properties of the halo and its central region at collapse , and defer discussion of the relevant timescales to section [ results.evolution ] . vs. halo collapse redshift . panel ( b ) : halo virial mass vs. halo collapse redshift panel ( c ) : halo virial mass vs. j@xmath52 panel ( d ) : halo virial temperature vs. j@xmath52 the j@xmath68 `` control '' result are shown as an open square ( and is at log j@xmath87 in the panels which are a function of j@xmath52 ) . in the bottom left panel , the dashed line corresponds to the fitting function for threshold mass from ( 2001 ) , eqtn . 8.,scaledwidth=90.0% ] figure [ fig.halovals2 ] shows several properties of the halo core at the epoch of collapse , including the core temperature , molecular hydrogen fraction , and spherically - averaged accretion rate as a function of ultraviolet background strength , and the spherically - averaged accretion rate as a function of the molecular hydrogen fraction . all `` core '' values are spherically - averaged and measured at the mass shell where @xmath88 m@xmath15 of gas is enclosed . panel ( a ) shows that the core h@xmath13 fraction decreases significantly with increasing fuv flux , with a corresponding increase in the core temperature ( panel ( b ) ) . this relationship is similar to that noted in & norman ( 2007 ) , where the amount of molecular hydrogen at densities of @xmath89 @xmath4 varies between simulations , and correlates strongly with baryon temperature . given that the accretion of gas onto the protostellar cloud is subsonic , this results in a strong relationship between the soft uv background flux and accretion rate onto the protostellar cloud , as shown by panel ( c ) , with accretion rates varying by more than a factor of 30 between the j@xmath68 and @xmath90 cases . panel ( d ) shows the strength of the correlation between core h@xmath13 fraction and accretion rate . note that the values discussed above are insensitive to the exact definition of the halo `` core . '' core for several simulations with the same cosmic realization but a range of lw background intensities . panel ( a ) : j@xmath52 vs. baryon core temperature . panel ( b ) : j@xmath52 vs. baryon core @xmath31 fraction . panel ( c ) : j@xmath52 vs. instantaneous accretion rate . panel ( d ) : baryon core @xmath31 fraction vs. accretion rate . the j@xmath68 `` control '' result are shown as an open square ( and is at log j@xmath87 in the panels which are a function of j@xmath52 ) . all `` core '' values are spherically - averaged and measured at the mass shell where @xmath88 m@xmath15 of gas is enclosed . , scaledwidth=90.0% ] figure [ fig.halovals3 ] shows the halo core mass and mass fraction at the epoch of collapse ( note that the definition of `` core '' is somewhat different than in the previous figure ) . the halo `` core mass '' is defined as being all gas at a density of n@xmath91 @xmath4 or above , and the core mass fraction is defined as the core mass divided by the virial mass of the halo at the epoch of collapse . this choice of minimum density ensures that we have captured the entirety of the quasistatically contracting analog of a galactic molecular cloud core , and for practical purposes is comparable to the gas fraction of cold , dense gas discussed in ( 2001 ) . the size of the halo core increases in absolute terms as the strength of the photodissociating background is increased , from @xmath92 m@xmath15 until j@xmath93 , where it reaches @xmath94 m@xmath15 . above this value of j@xmath54 , the core mass then begins to decrease , reaching @xmath95 m@xmath15 at j@xmath1 . the core mass fraction generally decreases ( though with some noise in the relationship ) from @xmath96 at j@xmath68 to @xmath97 at j@xmath1 , though the total overall change is less than a factor of 2 . @xmath4 or above . the `` core mass fraction '' is defined as the `` core mass '' divided by the virial mass of the halo at the epoch of collapse . the j@xmath98 `` control '' result are shown as an open square ( and is at log j@xmath87 in the panels which are a function of j@xmath52 ) . , scaledwidth=90.0% ] we now examine in more detail the evolution of the conditions in the cloud center leading to collapse . in figure [ fig.evolution ] we plot the central density , temperature , entropy @xmath99 , and 2 fraction versus time for the calculations with @xmath100 ( solid line ) , @xmath101 ( short - dashed line ) , @xmath81 ( long - dashed line ) and @xmath102 ( dot - dashed line ) . in these plots , central values of temperature and density are defined as being the values of these quantities in the cell with the highest baryon density . it is most instructive to compare the two extreme cases , j@xmath103 , which very closely resembles the j@xmath104 case , and the j@xmath83 case , which according to ( 2003 ) should not collapse at all . in panel ( a ) we see the central density increase modestly from @xmath105 2 @xmath4 to @xmath105 20 @xmath4 over 40 myr , and then increase rapidly thereafter , reaching @xmath106 @xmath4 in a scant additional 15 myr . this evolution is driven by 2 cooling , as can be seen from the entropy evolution in panel ( c ) . as expected , runaway cooling occurs when the 2 fraction reaches @xmath107 ( panel d ) , consistent with the ( 1997 ) analysis . by contrast the j@xmath83 case requires 120 myr for the central density to increases from @xmath105 2 @xmath4 to @xmath105 20 @xmath4 . as shown in panel ( c ) , the first 50 myr of this contraction is adiabatic , followed by an increase in entropy due to some heating event ( mergers ) . the fact that central entropy is not decreasing for 90 myr is indicative of the fact that 2 cooling is unimportant over this interval due to extremely low equilibrium 2 fractions ( f@xmath108 ; ( panel d ) ) . however , something interesting happens in the j@xmath1 calculation at @xmath109 myr . prior to that time the central temperature has crept up to 5000 k due to the increased virial mass . the cooling rate per 2 molecule is about 3 orders of magnitude higher at 5000k than at 500k a typical temperature at the halo center in low uv background evolutions . we believe this high temperature is due to a combination of adiabatic heating as the potential well deepens , and merger - induced shock heating . as shown in figure [ fig.merger ] , three large halos merge between @xmath110 and @xmath111 , quadrupling the halo mass ( figure [ fig.halovals1 ] , panel ( b ) ) . the elevated cooling rates cool the gas at the halo center to @xmath112 , allowing the gas density to increase slightly . the two effects reduce the central entropy from @xmath113 700 k @xmath114 to @xmath115 250 k @xmath114 . this is followed by another heating event at t=190 myr , presumably due to another merger , followed by a second more catastrophic cooling event . at t=200 myr , the halo core collapses , driving the central density higher , temperature and entropy lower , and 2 fraction higher . why does this collapse occur ? this is analyzed in figure [ fig.timescales ] , where we plot the evolution of the important timescales in the center of the halo . in the j@xmath103 case , the 2-body 2 formation timescale is always roughly an order of magnitude smaller than the photodissociation timescale . consequently , the 2 abundance is out of equilibrium and grows steadily with time ( figure [ fig.evolution ] panel d ) . the 2 cooling time in the central zone ( @xmath116 myr ) is far less than the hubble time ( @xmath117 myr ) , and before too long runaway collapse occurs . at higher values of j@xmath52 , the 2-body 2 formation time always hovers around the photodissociation timescale , regardless of how long that timescale is . this tells us that 2 is in equilibrium : the 2-body formations balance the photo - destructions by the lw background . the equilibrium 2 fraction is given by @xmath118 where k@xmath119 is the rate coefficient for the formation of h@xmath120 ( the limiting reaction for h@xmath13 formation at the temperature and density range considered here ) , @xmath121 is the electron density , and k@xmath122 is the 2 photodissociation rate due to the solomon process ( abel et al . 1997 ; yoshida et al . 2003 ) . using t @xmath123 k and @xmath124 , we get @xmath125 , which is in good agreement with the 2 fractions we see in figure [ fig.evolution ] . since @xmath126/ n@xmath127 , the smaller @xmath128 is , the shorter its formation time . now let us consider the 2 cooling time evolution ( short - long - dashed line ) . it is far lass than a hubble time for j@xmath129 , ensuring that these halo cores will eventually cool and collapse . in the j@xmath83 case , the cooling time drops below the hubble time for the first time at t=170 myr , and again at t=190 myr where it begins a steady decrease toward @xmath130 yr . despite the low equilibrium 2 fraction , if we wait long enough the inexorable press of time eventually establishes the condition @xmath131 and we get runaway collapse , even for j@xmath83 . note that we get catastrophic cooling because we are in the density regime ( @xmath132 @xmath4 ) where 2 cooling is proportional to the square of the gas density , not linear in the gas density ( abel , bryan , & norman 2000 ; abel et al . 2002 ) . ( solid ) , @xmath101 ( short - dashed ) , @xmath81 ( long - dashed ) and @xmath90 ( dot - dashed ) . , scaledwidth=90.0% ] and @xmath90 ( panels ( a)(d ) , respectively ) : 2-body 2 formation time ( solid ) ; 3-body 2 formation time ( short - dashed ) ; 2 collisional dissociate time ( long - dashed ) ; 2 photodissociation time ( dot - dashed ) ; 2 cooling time ( short - long - dashed ) . , scaledwidth=90.0% ] , t@xmath133 k halo that collapses at z=17.32 with a lyman - werner background flux @xmath134 . in comic strip order , the redshifts are z=22 , 21 , 20 , 19 , 18 , and 17.32 . field of view is 53.571 comoving kpc ( 2.678 proper kpc at z = 19 ) . logarithm of the projected baryon density ( column density ) is displayed , with color table scaled to the maximum and minimum values in each image . , scaledwidth=90.0% ] figure [ fig.radprof_zcoll.1 ] shows radial profiles of baryon number density , temperature , enclosed mass , circular velocity , and rms mach number as a function of radius , as well as specific angular momentum as a function of enclosed mass , for all simulations discussed in this paper . to facilitate comparison , results from the output of each simulation where the peak baryon number density at the center of the halo is approximately @xmath135 @xmath4 are shown , in order to capture each halo at a similar evolutionary stage , rather than at a fixed point in time . panel ( a ) shows that all of the simulations have similar density profiles . the scatter in number density at a given radius is readily explained by variation in halo mass halos which collapse at later times ( due to higher uv background strength ) are more massive , and thus have higher overall baryon densities . this is shown in another way in panel ( c ) , in a plot of enclosed mass as a function of radius which shows that all halos have very similar profiles at @xmath136 pc , but a significant variation at larger radii which is related to halo mass . the plot of temperature as a function of radius in panel ( b ) shows an interesting trend as the lw background is increased , the overall halo temperature as well as the halo core ( where `` core '' is roughly defined as gas within @xmath137 pc of the halo center ) temperature go up . in the outskirts of the halo , where baryon densities are low and thus cooling times are long , this is due primarily to the increase in halo mass , with the peak temperature corresponding approximately to the halo virial temperature . in the halo core , however , this temperature is correlated more strongly with the h@xmath13 fraction , and thus the uv background , as suggested by figure [ fig.halovals2 ] . the plots of specific angular momentum as a function of enclosed mass and keplerian velocity fraction ( defined as the local circular velocity divided by the keplerian orbital velocity due to mass within that radius ) as a function of radius in panels ( d ) and ( e ) show some mild trends trends . it appears that halos in simulations with stronger uv backgrounds tend to have less specific angular momentum at a given mass shell , and also tend to have a lower keplerian velocity fraction at a given radius than gas in halos which form in the presence of a lower uv background indeed , the halo which forms in the j@xmath1 calculation has the least angular momentum of all of the simulations , and the lowest keplerian velocity fraction out to @xmath137 pc . we speculate that this is due to transport and segregation of angular momentum by turbulence halos whose collapse is delayed have more time for these processes to act within the halo core , resulting in less angular momentum overall when the halo ultimately collapses ( note that this will be discussed more fully in a later paper ) . the plot of rms mach number as a function of radius in panel ( f ) also shows no obvious trend , though at radii below @xmath138 pc , simulations with a smaller uv background generally have higher rms mach numbers . this is due primarily to the gas being colder , and not the gas velocities being higher . . panel ( a ) : @xmath139 as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : enclosed gas mass as a function of radius . panel ( d ) : specific angular momentum as a function of enclosed mass . panel ( e ) : keplerian velocity fraction as a function of radius . panel ( f ) : gas rms mach number as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to simulations , as follows . black solid line : j@xmath68 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : @xmath139 as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : enclosed gas mass as a function of radius . panel ( d ) : specific angular momentum as a function of enclosed mass . panel ( e ) : keplerian velocity fraction as a function of radius . panel ( f ) : gas rms mach number as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to simulations , as follows . black solid line : j@xmath68 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : @xmath139 as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : enclosed gas mass as a function of radius . panel ( d ) : specific angular momentum as a function of enclosed mass . panel ( e ) : keplerian velocity fraction as a function of radius . panel ( f ) : gas rms mach number as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to simulations , as follows . black solid line : j@xmath68 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : @xmath139 as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : enclosed gas mass as a function of radius . panel ( d ) : specific angular momentum as a function of enclosed mass . panel ( e ) : keplerian velocity fraction as a function of radius . panel ( f ) : gas rms mach number as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to simulations , as follows . black solid line : j@xmath68 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : @xmath139 as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : enclosed gas mass as a function of radius . panel ( d ) : specific angular momentum as a function of enclosed mass . panel ( e ) : keplerian velocity fraction as a function of radius . panel ( f ) : gas rms mach number as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to simulations , as follows . black solid line : j@xmath68 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : @xmath139 as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : enclosed gas mass as a function of radius . panel ( d ) : specific angular momentum as a function of enclosed mass . panel ( e ) : keplerian velocity fraction as a function of radius . panel ( f ) : gas rms mach number as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to simulations , as follows . black solid line : j@xmath68 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=30.0% ] figure [ fig.radprof_zcoll.2 ] shows the h@xmath13 fraction , electron fraction , h@xmath120 fraction , ratio of cooling time to sound crossing time , ratio of cooling time to dynamical time , and ratio of sound crossing time to dynamical time as a function of radius , for all simulations . outputs and line types correspond to those in figure [ fig.radprof_zcoll.1 ] . there is a clear relationship at all radii between h@xmath13 fraction and uv background strength in panel ( a ) as the fuv background is increased , the overall h@xmath13 fraction decreases . this difference is most noticeable at radii of @xmath146 pc , but is maintained at larger and smaller radii . there is a `` kink '' at approximately @xmath101 pc where the values are all quite similar . this corresponds to a baryon number density of @xmath36 @xmath4 in all simulations , which is where 3-body h@xmath13 formation begins to occur . panels ( b ) and ( c ) show that e@xmath120 and @xmath29 fractions track each other , which is to be expected the local electron fraction controls the amount of h@xmath120 which can be produced , which is the limiting reaction in the dominant mode of h@xmath13 formation for n @xmath147 @xmath4 and there is a general trend with increasing electron and h@xmath120 fractions with increasing uv background flux . the plots of the ratio of cooling time to sound crossing time , ratio of cooling time to dynamical time , and ratio of sound crossing time to dynamical time as a function of radius shown in panels ( d ) , ( e ) and ( f ) display some interesting trends . the three simulations with the lowest uv background strengths have a much longer cooling time than the other calculations , which are all grouped roughly together with no discernible pattern . this agrees well with the plot of the ratio of cooling time to dynamical time , where a similar trend is observed . the difference between the three simulations with the lowest uv background strengths and the others is due to the somewhat higher h@xmath13 fraction in the halo core of these calculations . the gas temperature is @xmath148 k at the center of the halo cores in these calculations , and the gas can not cool further , which results in a very long cooling time . the sound crossing time is also increased , but not as significantly . all simulations have approximately the same ratio of sound crossing to dynamical times at all radii . given that the cooling time is typically longer than both the sound crossing time and dynamical time , one can infer that the collapse of the halo is occurring quasi - statically for all simulations . . panel ( a ) : molecular hydrogen fraction as a function of radius . panel ( b ) : electron fraction as a function of radius . panel ( c ) : @xmath29 fraction as a function of radius . panel ( d ) : ratio of gas cooling time to sound crossing time as a function of radius . panel ( e ) : ratio of gas cooling time to dynamical time as a function of radius . panel ( f ) : ratio of gas sound crossing time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : molecular hydrogen fraction as a function of radius . panel ( b ) : electron fraction as a function of radius . panel ( c ) : @xmath29 fraction as a function of radius . panel ( d ) : ratio of gas cooling time to sound crossing time as a function of radius . panel ( e ) : ratio of gas cooling time to dynamical time as a function of radius . panel ( f ) : ratio of gas sound crossing time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : molecular hydrogen fraction as a function of radius . panel ( b ) : electron fraction as a function of radius . panel ( c ) : @xmath29 fraction as a function of radius . panel ( d ) : ratio of gas cooling time to sound crossing time as a function of radius . panel ( e ) : ratio of gas cooling time to dynamical time as a function of radius . panel ( f ) : ratio of gas sound crossing time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : molecular hydrogen fraction as a function of radius . panel ( b ) : electron fraction as a function of radius . panel ( c ) : @xmath29 fraction as a function of radius . panel ( d ) : ratio of gas cooling time to sound crossing time as a function of radius . panel ( e ) : ratio of gas cooling time to dynamical time as a function of radius . panel ( f ) : ratio of gas sound crossing time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : molecular hydrogen fraction as a function of radius . panel ( b ) : electron fraction as a function of radius . panel ( c ) : @xmath29 fraction as a function of radius . panel ( d ) : ratio of gas cooling time to sound crossing time as a function of radius . panel ( e ) : ratio of gas cooling time to dynamical time as a function of radius . panel ( f ) : ratio of gas sound crossing time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . , title="fig:",scaledwidth=30.0% ] . panel ( a ) : molecular hydrogen fraction as a function of radius . panel ( b ) : electron fraction as a function of radius . panel ( c ) : @xmath29 fraction as a function of radius . panel ( d ) : ratio of gas cooling time to sound crossing time as a function of radius . panel ( e ) : ratio of gas cooling time to dynamical time as a function of radius . panel ( f ) : ratio of gas sound crossing time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . , title="fig:",scaledwidth=30.0% ] figure [ fig.radprof_zcoll.3 ] shows radial velocity as a function of radius , accretion time ( defined as @xmath149 ) as a function of enclosed baryon mass @xmath150 as a function of enclosed baryon mass and @xmath150 as a function of time for all simulations . output time and lines correspond to those in figures [ fig.radprof_zcoll.1 ] and [ fig.radprof_zcoll.2 ] . panel ( a ) shows a clear relationship between the strength of the soft uv background and the radial velocity , with the calculations that have higher uv backgrounds also having higher infall velocities . this is easily understood by the quasistatic contraction of the gas , which takes place at or below the local sound speed in the gas , as discussed by ( 2002 ) and & norman ( 2007 ) . given that the halo temperatures are systematically higher in calculations where the uv background is stronger , the sound speed is higher , and thus the rate at which the gas at the center of the halo contracts is higher . this leads to lower accretion times ( panel ( b ) ) and higher accretion rates ( panel ( c ) ) , with the accretion rates for the simulations with the strongest uv backgrounds being higher than those with a low ( or no ) uv background . panel ( d ) shows that the time evolution of the accretion rate in all simulations is qualitatively similar , but varies strongly in absolute magnitude and in the time at which the accretion rate onto the evolving protostellar cloud peaks , suggesting that the stars forming in these simulations may have very different evolutionary histories . . panel ( a ) : radial velocity as a function of radius . panel ( b ) : accretion time as a function of enclosed baryon mass . panel ( c ) : instantaneous accretion rate as a function of enclosed baryon mass . panel ( d ) : estimated accretion rate as a function of time . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . the upper and lower light short - long - dashed curves which extend from the upper left corner of panel ( b ) correspond to the main sequence lifetime of a massive population iii star of that mass and the kelvin - helmholtz timescale of a population iii timescale with a given luminosity and radius . all values are taken from ( 2002 ) . the three light diagonal short - dashed lines which extend from bottom left to top right in panel ( b ) correspond to masses accreted using constant accretion rates of ( from top to bottom ) @xmath151 , @xmath101 , and @xmath81 m@xmath152yr . the light horizontal short - dashed line in panels ( c ) and ( d ) correspond to the `` critical '' accretion rate of & palla ( 2003 ) , @xmath153 m@xmath152yr . , title="fig:",scaledwidth=45.0% ] . panel ( a ) : radial velocity as a function of radius . panel ( b ) : accretion time as a function of enclosed baryon mass . panel ( c ) : instantaneous accretion rate as a function of enclosed baryon mass . panel ( d ) : estimated accretion rate as a function of time . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . the upper and lower light short - long - dashed curves which extend from the upper left corner of panel ( b ) correspond to the main sequence lifetime of a massive population iii star of that mass and the kelvin - helmholtz timescale of a population iii timescale with a given luminosity and radius . all values are taken from ( 2002 ) . the three light diagonal short - dashed lines which extend from bottom left to top right in panel ( b ) correspond to masses accreted using constant accretion rates of ( from top to bottom ) @xmath151 , @xmath101 , and @xmath81 m@xmath152yr . the light horizontal short - dashed line in panels ( c ) and ( d ) correspond to the `` critical '' accretion rate of & palla ( 2003 ) , @xmath153 m@xmath152yr . , title="fig:",scaledwidth=45.0% ] . panel ( a ) : radial velocity as a function of radius . panel ( b ) : accretion time as a function of enclosed baryon mass . panel ( c ) : instantaneous accretion rate as a function of enclosed baryon mass . panel ( d ) : estimated accretion rate as a function of time . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . the upper and lower light short - long - dashed curves which extend from the upper left corner of panel ( b ) correspond to the main sequence lifetime of a massive population iii star of that mass and the kelvin - helmholtz timescale of a population iii timescale with a given luminosity and radius . all values are taken from ( 2002 ) . the three light diagonal short - dashed lines which extend from bottom left to top right in panel ( b ) correspond to masses accreted using constant accretion rates of ( from top to bottom ) @xmath151 , @xmath101 , and @xmath81 m@xmath152yr . the light horizontal short - dashed line in panels ( c ) and ( d ) correspond to the `` critical '' accretion rate of & palla ( 2003 ) , @xmath153 m@xmath152yr . , title="fig:",scaledwidth=45.0% ] . panel ( a ) : radial velocity as a function of radius . panel ( b ) : accretion time as a function of enclosed baryon mass . panel ( c ) : instantaneous accretion rate as a function of enclosed baryon mass . panel ( d ) : estimated accretion rate as a function of time . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at the point where the maximum number density is approximately @xmath135 @xmath4 . line types and weights correspond to those in figure [ fig.radprof_zcoll.1 ] . the upper and lower light short - long - dashed curves which extend from the upper left corner of panel ( b ) correspond to the main sequence lifetime of a massive population iii star of that mass and the kelvin - helmholtz timescale of a population iii timescale with a given luminosity and radius . all values are taken from ( 2002 ) . the three light diagonal short - dashed lines which extend from bottom left to top right in panel ( b ) correspond to masses accreted using constant accretion rates of ( from top to bottom ) @xmath151 , @xmath101 , and @xmath81 m@xmath152yr . the light horizontal short - dashed line in panels ( c ) and ( d ) correspond to the `` critical '' accretion rate of & palla ( 2003 ) , @xmath153 m@xmath152yr . , title="fig:",scaledwidth=45.0% ] figure [ fig.radprof_fixred.1 ] shows the status of several spherically - averaged quantities for halos from all simulations at a fixed redshift , @xmath154 , which is shortly before the baryons in the most massive halo in the j@xmath68 simulation collapses to high density . this is at an early point in the evolution of the majority of these halos , and several conclusions can be drawn from the plots of number density , temperature , h@xmath13 fraction , and radial velocity as a function of radius . the molecular hydrogen fraction declines monotonically with increasing uv background strength , in agreement with an estimate for equilibrium h@xmath13 fraction at low ( @xmath155 @xmath4 ) densities : @xmath156 where k@xmath119 is the rate coefficient for the formation of h@xmath120 ( the limiting reaction for h@xmath13 formation at the temperature and density range considered here ) and k@xmath52 is the rate coefficient for h@xmath13 in the lw band ( abel et al . 1997 ; yoshida et al . halo cores at higher central densities are departing from the approximations used in the above estimate and thus have values of the h@xmath13 fraction that are not quite the equilibrium values . halo cores which have reached core h@xmath13 fractions higher than @xmath157 have been able to cool to significantly below the virial temperature ( since the cooling time is much less than the hubble time at that value of the h@xmath13 fraction ) , allowing the baryon density to increase . the effect of cooling can also be seen in the plot of radial velocity as a function of radius , where the core regions of halos that have cooled are contracting ( albeit very slowly ) , while those which are still at the virial temperature are not showing significant signs of contraction within @xmath158 pc . all halos have some inflow at large radii , due to the infall of gas from adjoining filaments . . panel ( a ) : number density as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : @xmath31 fraction as a function of radius . panel ( d ) : radial velocity as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at @xmath154 ( shortly before the j@xmath68 simulation collapses to high density . line types and weights correspond to simulations , as follows . black solid line : j@xmath159 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=45.0% ] . panel ( a ) : number density as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : @xmath31 fraction as a function of radius . panel ( d ) : radial velocity as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at @xmath154 ( shortly before the j@xmath68 simulation collapses to high density . line types and weights correspond to simulations , as follows . black solid line : j@xmath159 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=45.0% ] . panel ( a ) : number density as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : @xmath31 fraction as a function of radius . panel ( d ) : radial velocity as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at @xmath154 ( shortly before the j@xmath68 simulation collapses to high density . line types and weights correspond to simulations , as follows . black solid line : j@xmath159 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=45.0% ] . panel ( a ) : number density as a function of radius . panel ( b ) : temperature as a function of radius . panel ( c ) : @xmath31 fraction as a function of radius . panel ( d ) : radial velocity as a function of radius . all quantities except enclosed gas mass are mass - weighted , and all simulations are shown at @xmath154 ( shortly before the j@xmath68 simulation collapses to high density . line types and weights correspond to simulations , as follows . black solid line : j@xmath159 . black short - dashed line : j@xmath140 . black long - dashed line : j@xmath141 . black dot short - dashed line : j@xmath142 . black dot long - dashed line : j@xmath80 . black short dashed - long dashed line : j@xmath143 . red solid line : j@xmath144 . red short - dashed line : j@xmath145 . red long - dashed line : j@xmath1 . , title="fig:",scaledwidth=45.0% ] in this section we compare the evolution of two representative simulations . we choose the calculations which correspond to j@xmath68 and @xmath90 . these two simulations represent the extremes in halo evolution the j@xmath68 calculation is an example of the evolution of a `` standard '' halo , as discussed in previous literature , and the j@xmath1 simulation is the most extreme example of a population iii protostar forming in a t@xmath5 k halo available from our suite of calculations . as described in section [ sect.simsetup ] , both calculations start from the same initial conditions , with only the strength of the lw background being different . projections of the log baryon density and temperature for the j@xmath68 and @xmath90 simulations are shown in figures [ fig.image_jlw_0 ] and [ fig.image_jlw_1em21 ] , respectively . these projections are at the epoch of collapse , which is @xmath160 @xmath161 for the j@xmath68 @xmath162 simulation . the halo masses are significantly different : @xmath163 and @xmath164 m@xmath15 for the two calculations , reflecting the rapid pace of mergers at that redshift , and the amount of structure evolution that takes place over that relatively short time period ( @xmath165 years ) . this can clearly be seen in the left column of both figures , when the large amount of structure apparent in figure [ fig.image_jlw_0 ] has merged into the main halo by the time of protostellar cloud formation in the j@xmath1 case , as seen in figure [ fig.image_jlw_1em21 ] . despite the differences in evolutionary states , the halos themselves are quite similar in figures [ fig.image_jlw_0 ] and [ fig.image_jlw_1em21 ] . both calculations show halos that are strongly centrally concentrated , as shown by the center column in both figures . this column shows projections of the halos which are centered on the baryon peak and are approximately one virial radius across ( and thus are scaled differently in both images ) . no fragmentation of the halo core is visible . the rightmost panel in both calculations shows the center of the halo , where the protostellar cloud is evolving . there is still no evidence for fragmentation up to a central baryon density of @xmath3 @xmath4 . figures [ fig.radprof_jlw0.1 ] and [ fig.radprof_jlw1em21.1 ] show the temporal evolution of several spherically - averaged baryon quantities for the j@xmath68 and @xmath90 calculations , respectively . these plots show baryon number density , enclosed baryon mass , temperature , h@xmath13 fraction , ratio of cooling time to sound crossing time , and ratio of cooling time to dynamical time , as a function of radius , evolving from a number density of @xmath166 @xmath4 to @xmath69 @xmath4 in both calculations . some qualitative commonalities are obvious between the two calculations . both halos steadily grow in central density ( panel ( a ) ) , which is coupled to the growing molecular hydrogen fraction in the central region of the halo ( panel ( d ) ) . both halos collapse quasi - statically , as shown in panels ( e ) and ( f ) . also , the temperature profiles ( as shown in panel ( c ) ) are broadly similar , in that the gas in the halo central region cools to some minimum value at r @xmath167 pc ( corresponding to n @xmath168 @xmath4 in both simulations ) and then creeps steadily upward as gas collapses to higher densities . though there are qualitative similarities between the evolution of the two calculations , detailed examination shows some significant quantitative differences . the time that the halo in the j@xmath68 simulation takes to evolve from a central density of @xmath169 @xmath4 to @xmath135 @xmath4 is @xmath170 myr , with 21 myr of that being the time required to evolve from @xmath169 to n @xmath171 @xmath4 . the time required for the j@xmath1 calculation to evolve from @xmath169 @xmath4 to @xmath135 @xmath4 is slightly shorter , @xmath60 17 myrs , with 8 myrs required to evolve to n @xmath172 @xmath4 . the temperature evolution is also somewhat different the j@xmath68 simulation has a minimum temperature at r @xmath137 pc of 200 k , with n @xmath171@xmath173 and a molecular hydrogen fraction of @xmath174 , and roughly 1000 m@xmath15 of enclosed gas . as the gas within this radius evolves to higher densities , the temperature increases by a factor of a few , peaking at t @xmath175 k at n @xmath176 @xmath4 , with f@xmath177 . the gas cooling time within r @xmath137 pc in the j@xmath68 simulation is always larger than the sound crossing and dynamical times by at least a factor of two , implying a quasistatic gas contraction at all times . the j@xmath1 simulation evolves somewhat differently . the collapsing gas reaches a minimum temperature of @xmath178 k at r @xmath179 pc , with a baryon density of @xmath180 @xmath4 , an enclosed gas mass of roughly @xmath86 m@xmath15 , and a molecular hydrogen fraction of @xmath181 , two orders of magnitude below that in the j@xmath68 calculation at the equivalent temperature minimum . the low h@xmath13 fraction is due to the high lw background radiation flux . as the halo evolves to higher densities , the gas temperature and h@xmath13 fraction also creep upwards . the temperature of the gas when the peak reaches roughly @xmath135 @xmath4 is @xmath182 k , with a h@xmath13 fraction of @xmath183 . the cooling time at r @xmath179 pc is an order of magnitude higher than the sound crossing or dynamical times , implying an extremely slow contraction of the gas at that radius . as the central density increases , however , the cooling time becomes lower , with t@xmath184/t@xmath185 and t@xmath184/t@xmath186 dipping to @xmath187 at its minimum , but the latter increasing toward the central density peak to a ratio of @xmath188 . this implies a more rapid contraction of gas in the halo center in the simulation with a higher uv background , despite the lack of molecular hydrogen . it is also worth noticing that the j@xmath1 calculation has h@xmath13 fractions at n @xmath168 and @xmath106 @xmath4 of @xmath189 and @xmath190 , respectively , as compared to f@xmath191 and @xmath192 at the same densities in the j@xmath68 calculation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath193 , @xmath194 years . red dot long - dashed line : @xmath195 , @xmath196 years . red dot short - dashed line : @xmath197 , @xmath196 years . red long - dashed line : @xmath198 , @xmath199 years . red short - dashed line : @xmath200 , @xmath201 years . red solid line : @xmath202 , @xmath203 years . black short dashed - long dashed line : @xmath204 , @xmath205 years . black dot long - dashed line : @xmath206 , @xmath207 years . black dot short - dashed line : @xmath208 , @xmath209 years . black long - dashed line : @xmath210 , @xmath211 years . black short - dashed line : @xmath212 , @xmath213 years . black solid line : @xmath214 , @xmath215 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath193 , @xmath194 years . red dot long - dashed line : @xmath195 , @xmath196 years . red dot short - dashed line : @xmath197 , @xmath196 years . red long - dashed line : @xmath198 , @xmath199 years . red short - dashed line : @xmath200 , @xmath201 years . red solid line : @xmath202 , @xmath203 years . black short dashed - long dashed line : @xmath204 , @xmath205 years . black dot long - dashed line : @xmath206 , @xmath207 years . black dot short - dashed line : @xmath208 , @xmath209 years . black long - dashed line : @xmath210 , @xmath211 years . black short - dashed line : @xmath212 , @xmath213 years . black solid line : @xmath214 , @xmath215 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath193 , @xmath194 years . red dot long - dashed line : @xmath195 , @xmath196 years . red dot short - dashed line : @xmath197 , @xmath196 years . red long - dashed line : @xmath198 , @xmath199 years . red short - dashed line : @xmath200 , @xmath201 years . red solid line : @xmath202 , @xmath203 years . black short dashed - long dashed line : @xmath204 , @xmath205 years . black dot long - dashed line : @xmath206 , @xmath207 years . black dot short - dashed line : @xmath208 , @xmath209 years . black long - dashed line : @xmath210 , @xmath211 years . black short - dashed line : @xmath212 , @xmath213 years . black solid line : @xmath214 , @xmath215 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath193 , @xmath194 years . red dot long - dashed line : @xmath195 , @xmath196 years . red dot short - dashed line : @xmath197 , @xmath196 years . red long - dashed line : @xmath198 , @xmath199 years . red short - dashed line : @xmath200 , @xmath201 years . red solid line : @xmath202 , @xmath203 years . black short dashed - long dashed line : @xmath204 , @xmath205 years . black dot long - dashed line : @xmath206 , @xmath207 years . black dot short - dashed line : @xmath208 , @xmath209 years . black long - dashed line : @xmath210 , @xmath211 years . black short - dashed line : @xmath212 , @xmath213 years . black solid line : @xmath214 , @xmath215 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath193 , @xmath194 years . red dot long - dashed line : @xmath195 , @xmath196 years . red dot short - dashed line : @xmath197 , @xmath196 years . red long - dashed line : @xmath198 , @xmath199 years . red short - dashed line : @xmath200 , @xmath201 years . red solid line : @xmath202 , @xmath203 years . black short dashed - long dashed line : @xmath204 , @xmath205 years . black dot long - dashed line : @xmath206 , @xmath207 years . black dot short - dashed line : @xmath208 , @xmath209 years . black long - dashed line : @xmath210 , @xmath211 years . black short - dashed line : @xmath212 , @xmath213 years . black solid line : @xmath214 , @xmath215 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath193 , @xmath194 years . red dot long - dashed line : @xmath195 , @xmath196 years . red dot short - dashed line : @xmath197 , @xmath196 years . red long - dashed line : @xmath198 , @xmath199 years . red short - dashed line : @xmath200 , @xmath201 years . red solid line : @xmath202 , @xmath203 years . black short dashed - long dashed line : @xmath204 , @xmath205 years . black dot long - dashed line : @xmath206 , @xmath207 years . black dot short - dashed line : @xmath208 , @xmath209 years . black long - dashed line : @xmath210 , @xmath211 years . black short - dashed line : @xmath212 , @xmath213 years . black solid line : @xmath214 , @xmath215 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath216 , @xmath217 years . red dot long - dashed line : @xmath218 , @xmath219 years . red dot short - dashed line : @xmath220 , @xmath221 years . red long - dashed line : @xmath222 , @xmath223 years . red short - dashed line : @xmath224 , @xmath225 years . red solid line : @xmath226 , @xmath227 years . black short dashed - long dashed line : @xmath228 , @xmath229 years . black dot long - dashed line : @xmath230 , @xmath231 years . black dot short - dashed line : @xmath232 , @xmath233 years . black long - dashed line : @xmath234 , @xmath235 years . black short - dashed line : @xmath236 , @xmath237 years . black solid line : @xmath238 , @xmath239 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath216 , @xmath217 years . red dot long - dashed line : @xmath218 , @xmath219 years . red dot short - dashed line : @xmath220 , @xmath221 years . red long - dashed line : @xmath222 , @xmath223 years . red short - dashed line : @xmath224 , @xmath225 years . red solid line : @xmath226 , @xmath227 years . black short dashed - long dashed line : @xmath228 , @xmath229 years . black dot long - dashed line : @xmath230 , @xmath231 years . black dot short - dashed line : @xmath232 , @xmath233 years . black long - dashed line : @xmath234 , @xmath235 years . black short - dashed line : @xmath236 , @xmath237 years . black solid line : @xmath238 , @xmath239 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath216 , @xmath217 years . red dot long - dashed line : @xmath218 , @xmath219 years . red dot short - dashed line : @xmath220 , @xmath221 years . red long - dashed line : @xmath222 , @xmath223 years . red short - dashed line : @xmath224 , @xmath225 years . red solid line : @xmath226 , @xmath227 years . black short dashed - long dashed line : @xmath228 , @xmath229 years . black dot long - dashed line : @xmath230 , @xmath231 years . black dot short - dashed line : @xmath232 , @xmath233 years . black long - dashed line : @xmath234 , @xmath235 years . black short - dashed line : @xmath236 , @xmath237 years . black solid line : @xmath238 , @xmath239 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath216 , @xmath217 years . red dot long - dashed line : @xmath218 , @xmath219 years . red dot short - dashed line : @xmath220 , @xmath221 years . red long - dashed line : @xmath222 , @xmath223 years . red short - dashed line : @xmath224 , @xmath225 years . red solid line : @xmath226 , @xmath227 years . black short dashed - long dashed line : @xmath228 , @xmath229 years . black dot long - dashed line : @xmath230 , @xmath231 years . black dot short - dashed line : @xmath232 , @xmath233 years . black long - dashed line : @xmath234 , @xmath235 years . black short - dashed line : @xmath236 , @xmath237 years . black solid line : @xmath238 , @xmath239 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath216 , @xmath217 years . red dot long - dashed line : @xmath218 , @xmath219 years . red dot short - dashed line : @xmath220 , @xmath221 years . red long - dashed line : @xmath222 , @xmath223 years . red short - dashed line : @xmath224 , @xmath225 years . red solid line : @xmath226 , @xmath227 years . black short dashed - long dashed line : @xmath228 , @xmath229 years . black dot long - dashed line : @xmath230 , @xmath231 years . black dot short - dashed line : @xmath232 , @xmath233 years . black long - dashed line : @xmath234 , @xmath235 years . black short - dashed line : @xmath236 , @xmath237 years . black solid line : @xmath238 , @xmath239 years . , title="fig:",scaledwidth=30.0% ] simulation . panel ( a ) : number density as a function of radius . panel ( b ) : enclosed gas mass as a function of radius . panel ( c ) : temperature as a function of radius . panel ( d ) : @xmath31 fraction as a function of radius . panel ( e ) : ratio of cooling time to sound crossing time as a function of radius . panel ( f ) : ratio of cooling time to dynamical time as a function of radius . all quantities except enclosed gas mass are mass - weighted . line types and weights correspond to different times , as follows . red short dashed - long dashed line : @xmath216 , @xmath217 years . red dot long - dashed line : @xmath218 , @xmath219 years . red dot short - dashed line : @xmath220 , @xmath221 years . red long - dashed line : @xmath222 , @xmath223 years . red short - dashed line : @xmath224 , @xmath225 years . red solid line : @xmath226 , @xmath227 years . black short dashed - long dashed line : @xmath228 , @xmath229 years . black dot long - dashed line : @xmath230 , @xmath231 years . black dot short - dashed line : @xmath232 , @xmath233 years . black long - dashed line : @xmath234 , @xmath235 years . black short - dashed line : @xmath236 , @xmath237 years . black solid line : @xmath238 , @xmath239 years . , title="fig:",scaledwidth=30.0% ] in this paper we have examined aspects of the formation of population iii stars in the presence of a soft ultraviolet background , using an adaptive mesh refinement cosmological structure formation code . given the nature of the tool and current limitations in computing power , some physics were neglected , and possible numerical issues may arise . we discuss this here . the primordial chemistry model used in these calculations ignores the effects of deuterium , lithium , and the various molecules that form between these elements and ordinary hydrogen . deuterium and lithium have been shown to be unimportant in the temperature and density regimes that we have examined in this paper ( galli & palla 1998 , 2002 ; lipovka , nez - lpez , & avila - reese 2005 ) . however , it is possible that they may be relevant in other situations of importance to population iii star formation in particular , regions which have been ionized to very high electron fractions may experience significant cooling from the hd molecule , which due to its permanent dipole moment makes it more than 100 times more effective as a cooling agent than molecular hydrogen ( per molecule ) , and has the potential to cool gas down to approximately the temperature of the cosmic microwave background , which scales with redshift as @xmath240 k ( flower et al . 2000 ; galli & palla 2002 ; lipovka et al . this gives a minimum baryon temperature of approximately @xmath241 kelvin at @xmath110 and could reduce the minimum accretion rate onto a primordial protostar significantly . lithium , while in principle an effective coolant as lih , is safely ignored , since only a tiny fraction of lithium is converted into lih ( mizusawa , omukai , & nishi 2005 ) . a final chemical process that is omitted from our calculation is heating caused by molecular hydrogen formation at high densities ( n @xmath242 @xmath4 ) , the inclusion of which may result in differences in the temperature evolution of the gas in the highest density regimes considered here . this does not significantly affect the conclusions of our paper , as the significant differences we observe between simulations occur at densities much lower than @xmath36 @xmath4 . self - shielding of the photodissociating background by molecular hydrogen in the high - density gas is ignored in these calculations . though this effect could in principle be important , the actual column densities of molecular hydrogen are typically far too small to actually block the soft uv flux . according to ( 2001 ) ( and references therein ) , a column density of @xmath243 @xmath11 is enough for shielding to become important . however , this was derived for a static distribution of h@xmath13 , while the lw band consists of hundreds of individual lines whose width in this case is dominated by doppler broadening . it is useful to note in this case that the average line width is @xmath179 km / s and the rms baryon velocity in our calculations are @xmath244 km / s . in order for self - shielding to be important in the case of a turbulent medium , the column density must be much higher . typical maximum h@xmath13 column densities in our calculations are on the order of @xmath245 @xmath11 , but these occur late in the collapse of the halo core , and in the highest density regions the cooling and h@xmath13 production times are much shorter than the photodissociation time scale , at which point self - shielding becomes unimportant . it is worth noting that there are some regimes where self - shileding can be critical . ( 2007 ) finds that self - shielding can strongly affect the evolution of a collapsing primordial halo which is being illuminated with photodissociating flux by a neighboring star , though the situation is somewhat idealized . additionally , we do not consider the more complicated effects relating to population iii stars which form in halos in the cosmic neighborhoods where previous generations of stars have existed . this allows us to ignore complex radiative , chemical and dynamical effects that would vastly complicate our calculations ( oshea et al . 2005a ; johnson , greif , & bromm 2006 ; yoshida et al . 2006a ; greif et al . 2007 ; abel , wise , & bryan 2007 ; ahn & shapiro 2007 ) . a further effect that is ignored in this paper is @xmath29 photodetachment . this could in principle be a significant effect @xmath29 is the catalyst for molecular hydrogen formation in the dominant 2 formation channel at densities @xmath147 @xmath4 , and it can be detached by photons with energies @xmath246 ev . photons with this energy would be produced in great numbers by the same massive population iii stars that we assume are producing the molecular hydrogen photodissociating background . however , as shown by ( 2001 ) , the rate of photodetachment is orders of magnitude smaller than the rate of @xmath29 formation at densities comparable to that found in the centers of the halos examined in this work , and thus the process of @xmath29 photodetachment can be safely ignored . in this paper , we examine the effects of molecular hydrogen dissociating backgrounds which range in strength from j@xmath247 to @xmath9 ergs@xmath10 @xmath11 hz@xmath10 sr@xmath10 in the lw band ( @xmath248 ev ) . this range is in good agreement with the expected range of photodissociating backgrounds predicted by & abel ( 2005 ) , but it is still finite . examination of our simulations show that the case with the lowest uv background ( j@xmath140 ) is almost identical to the `` control '' ( j@xmath68 ) calculation , justifying our choice of minimum value . our upper value is consistent with wise & abel , and examination of figures [ fig.halovals1 ] and [ fig.radprof_zcoll.1][fig.radprof_zcoll.3 ] suggests that increases in the strength of the uv background ( within reasonable values ) will result in a further delay in halo collapse , but no major change in the mode of star formation observed . this is an effect of the halo properties a small amount of molecular hydrogen will always exist in these halos , allowing the gas at the center to cool and contract quasi - statically to higher densities . once the gas has collapsed to very high densities , rapid h@xmath13 formation via the 3-body process will occur , essentially independent of the strength of the soft uv background , and the gas will be able to cool down to @xmath249 k very quickly . the effects of magnetic fields are completely ignored in the simulations discussed in this work . magnetic fields are discussed in detail in paper i , but , to summarize , a fairly high seed magnetic field is needed to be dynamically significant at the relatively low densities we explore in this work . the possible importance of magnetic fields has been explored in analytic and semi - analytic work ( tan & mckee 2004 ; silk & langer 2006 ; maki & susa 2007 ) . we will examine the possible evolution of magnetic fields within the context of cosmological amr simulations in a later paper ( oshea & turk 2007 , in preparation ) . the simulations presented in this paper are generated assuming a cosmology that is somewhat different than the currently - favored wmap year iii `` best - fit '' model ( spergel et al . most importantly , our value of @xmath250 is 0.9 , which is significantly higher than the wmap value of 0.761 . the general effect of a higher @xmath250 is to cause structure formation to take place earlier , and thus at a given redshift one would expect significantly more halos in our chosen cosmological model than in the wmap year iii model . however , the evolution of any single cosmological halo , such as the one examined in this work , is not particularly affected by this parameter in the sense that we are not examining halo stastical properties . in addition to @xmath250 , the ratio of @xmath251 in our simulations is 0.1337 , while it is @xmath252 in the wmap year iii cosmology . this may result in some small quantitative differences in , e.g. , the redshift of halo collapse , but should not significantly affect our results . we direct the reader to paper i for a detailed discussion of other possible numerical issues , such as the generation of cosmological initial conditions , the assumption that the cold dark matter model is correct , the choice of halo in our simulations , and the size of the simulation volumes used . this paper explores the formation of population iii stars in simulations with a constant soft uv background . our results agree well qualitatively with that of ( 2001 ) ; we both find that a soft uv background can delay the formation of molecular hydogen , and thus the cooling and collapse , of small ( @xmath253 m@xmath15 ) cosmological halos in which population iii stars form . we also find that increasing the soft uv background increases the minimum halo mass required for a halo to collapse ( in a way similar to that of smoothing the dark matter power spectrum at small scales see & norman ( 2006 ) for a discussion ) . machacek et al . derived a mass threshold for collapse as a function of the lw background flux that agrees well with our simulations , though the halo masses in our calculation are significantly higher . this is due to the halo that we examine being a `` typical '' halo rather than at the threshold mass for star formation . presumably , if we performed these calculations using many halos in a range of cosmological realizations , we would find a minimum halo mass that is somewhat lower than the masses seen in our calculations . our work is a significant improvement upon that of machacek et al . in some respects , as our simulations are much more highly resolved and we examine the evolution of a single halo over a much wider range of soft uv background fluxes . one drawback of our work compared to machacek et al . is that we examine the evolution of a single halo , albeit with a broad range of uv backgrounds . our work is similar to the calculations in ( 2003 ) which examine population iii star formation in the presence of a photodissociating ultraviolet background . they find that for values of j@xmath52 above @xmath254 erg s@xmath10 @xmath11 hz@xmath10 sr@xmath10 , hydrogen molecules are rapidly dissociated and gas cooling is inefficient , implying that the collapse of gas in the halo center may be delayed until atomic line cooling dominates , in halos with t@xmath12 k. while we see delays in halo collapse , we do see that a tiny amount of 2 can form even at high fuv background strengths , eventually allowing gas to cool and collapse to high densities without ever reaching temperatures at which atomic line cooling would be effective ( t @xmath255 k ) . yoshida et al . make this statement based on a set of simulations using two different values of j@xmath52 ( @xmath254 and @xmath256 ) , which bracket the `` break '' in properties seen in our results . the simulations used to obtain this result had sph particles with m@xmath257 h@xmath10 m@xmath15 and thus an effective mass resolution comparable to the mass of gas within the quasistatically - collapsing central region of a halo . our mass resolution is more than two orders of magnitude higher in the center of the cosmological halos that we examine . the differences between our result and theirs is likely due to the significant differences in resolution and may in fact be consistent once resolution is taken into account . we see a strong relationship between accretion rates onto the protostellar cloud and the strength of the photodissociating background . calculations with higher ultraviolet background strengths typically have a larger spherically - averaged baryon accretion rate onto the evolving protostellar cloud . this is clearly due to variation in halo central temperature relating to the amount of molecular hydrogen existing in the center of the halo at the epoch of collapse . accretion onto the evolving protostellar cloud is subsonic , and thus regulated by the local sound speed ( which scales as t@xmath258 ) . this implies some relationship between the strength of the photodissociating background and the final stellar mass of the primordial star however , the details of this relationship are unclear , and depends on many factors . for example , & palla ( 2003 ) suggest that an increase in accretion rate above @xmath153 m@xmath15/yr will actually result in a _ decrease _ in the final mass of the star due to radiative feedback from the evolving protostar . however , their results use one - dimenstional , fairly idealized models , and geometrical effects may be important . this is explored in & mckee ( 2004 ) , who use a combination of analytic and semi - analytic models of the evolving system . they suggest that accretion onto the protostar is highly non - spherical , and is in fact mediated by an accretion disk . they argue that a larger accretion rate onto the disk will lead to a larger star , though with some limitations . finally , ( 2001 ) and & yoshii ( 2003 ) study the imf of stars that form in t@xmath12 k metal - free protogalaxies in the presence of an h@xmath13 photodissociating background . they conclude that photodissociation actually decreases the jeans mass of the gas at high densities , and thus reduces the fragmentation mass scale of the clouds and presumably the stellar mass . they use one - dimensional , spherically symmetric simulations , however . we can explore this issue in more detail in later calculations , when appropriate models for shielding , cooling via @xmath29 , and more advanced chemistry have been implemented into enzo . this will be examined in a later paper . it is worth noting that the range of accretion rates observed in this situation are within the range of rates seen in the @xmath17cdm simulations discussed in paper i. this implies that the mass ranges of the resulting stars will not be significantly different than in the `` standard '' population iii star formation scenario . we see no fragmentation of the high - density baryon core , up to a baryon number density of @xmath259 @xmath4 , for the entire range of simulations explored in this paper . this is due to a combination of effects , but predominantly the poor cooling properties of molecular hydrogen . the gas is relatively hot ( @xmath260 kelvin ) and thus has a high sound speed , which helps to damp out perturbations in the halo center which would otherwise result in multiple fragments . this effect is exacerbated in simulations with high ultraviolet background strengths , as the temperatures ( and thus the sound speeds ) are generally higher . this result implies that there is no fundamental change in the mode of population iii star formation as halos grow , and that more massive `` proto - galactic '' halos , with t@xmath261 k , will continue to form a single massive star per halo . these halos will have a much larger binding energy than the smaller , m @xmath14 m@xmath15 halos which have traditionally been examined by numerical simulations , and implies that multiple generations of star formation may be able to take place in a single halo . this leads to the possibility of `` self - enrichment , '' where a single population iii star enriches the high - density center of the halo to metallicities high enough to change the cooling properties of the gas . this could cause a strong change in the imf ( santoro & shull 2006 ; smith & sigurdsson 2007 ) , and these objects could make a significant contribution to the reionization history of the universe ( mackey , bromm , & hernquist 2003 ; greif & bromm 2006 ) . we will examine whether the high - density gas in the centers of these more massive halos fragments in a later work . the values of accretion rates onto the evolving protostellar cloud which are observed with increasing photodissociating background flux ( which are always comparable to those observed in paper i ) , in addition to the apparent lack of fragmentation of gas in the halo center up to densities of n@xmath262 @xmath4 , implies that there is no major change in the mass range of population iii stars as halo masses increase to t@xmath5 k. this is in mild disagreement with previous semianalytic results by & haiman ( 2002 ) and ( 2003 ) . this is apparently because previous authors assumed that 2 cooling in `` second generation '' ( t@xmath12 k ) objects is unimportant , while we observe that a small 2 fraction , which forms in equilibrium with the uv background , can cool gas effectively to temperatures of @xmath263 k. we do agree with these authors that the majority of primordial star formation will take place in objects with t@xmath5 k , the so - called `` second generation '' objects ( macintyre , santoro , & thomas 2006 ) our work appears to contradict the results of & bromm ( 2006 ) , who use analytic and semianalytic techniques to follow the collapse of gas within halos of t@xmath264 . they find that , in their model , gas cools to @xmath265 k via atomic line cooling , and then contracts nearly isothermally to high densities , allowing a molecular hydrogen fraction of f@xmath266 to build up independent of local density and temperature . the gas in the center of the halo then cools and allows the gas to fragment on scales of @xmath158 m@xmath15 . we do not see this mode of star formation , possibly because we never generate the significantly ionized halos with t@xmath267 k upon which their scenario depends . rather , we simulate halos that in the cases of j@xmath0 are built up to t@xmath5 k via mergers , and the gas is never ionized to a significant degree . one possible reason for the observed differences is that we ignore the formation of deuterium hydride and its effects on the cooling properties of primordial gas . it is possible that , in some contexts , the inclusion of hd may result in enhanced fragmentation . this will be examined in future studies . the scenario described by oh & haiman may still in fact be able to occur , and deserves further detailed numerical study to determine the fate of the collapsing gas and the implications of the ( presumably ) extremely high accretion rates onto the forming protostellar clouds . our results are similar to those shown by & abel ( 2007 ) , who perform similar simulations examining the collapse of gas in cosmological halos in the presence of a photodissociating background . they also find that , regardless of the strength of the uv background , the collapse of gas in the centers of the cosmological halos in question still occurs via h@xmath13 cooling . this holds even for quite extreme examples , such as when the original electron fraction is set to unphysically low levels ( which should suppress h@xmath13 formation , at least temporarily ) . in this paper we have performed a suite of high dynamical range ( l@xmath268 ) 3d adaptive mesh cosmological simulations of the formation of population iii stars in a @xmath17cdm universe in the presence of a molecular hydrogen photodissocating ( `` lyman - werner '' ) ultraviolet background . the purpose of these calculations is to understand the effect that the soft ultraviolet background has on the evolution of the gas in collapsing cosmological halos and to determine possible effects on the forming protostellar cloud . these simulations are all of a single cosmological realization , but with varied ultraviolet background strengths . our principal results are as follows : \1 . our calculations show that , as the flux of ultraviolet radiation in the lyman - werner band is increased , population iii star formation in a given cosmological halo is delayed to later times and , as a result , an increase in the virial mass of this halo at the onset of baryon collapse . this is in good agreement with previous work by ( 2001 ) and ( 2003 ) . we find that , contrary to previous work , the formation of primordial stars is never completely suppressed , regardless of the strength of the uv background . a small amount of molecular hydrogen always exists in cosmological halos , and allows cooling and halo collapse in gas which is bathed in a strong photodissociating background but has not reached @xmath86 k. the previously - suggested mode of star formation in `` second generation '' halos , where the collapse of gas to high density is completely suppressed until cooling can take place via atomic hydrogen line transitions , is never observed in our calculations , which have photodissociating background strengths up to j@xmath1 . \3 . though the molecular hydrogen fraction in the centers of halos which are bathed in strong ( j@xmath0 ) lyman - werner radiation is very small at early times ( f@xmath269 ) , the gas at the center of the halo can still collapse due to the efficient cooling of molecular hydrogen at @xmath270 k , and to the extended time these halos require to collapse compared to halos in the presence of much smaller uv backgrounds . we observe that the estimated accretion rate onto the forming protostellar cloud varies strongly as a function of j@xmath52 , with simulations that have a stronger ultraviolet background having higher accretion rates . this is a function of the suppression of molecular hydrogen formation ( and thus suppression of effective cooling ) by the photodissociating background and of the higher virial temperatures of these halos when the gas finally collapses . only a single clump is formed at the center of each collapsing cosmological halo , regardless of the strength of the photodissociating background , up to a baryon number density of @xmath3 @xmath4 . this implies that , as in the more commonly - studied mode of star formation , we will find only a single star per halo even in objects which are massive enough that t@xmath12 k. Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: i would like to revive , from a new viewpoint , a somewhat old fashioned discussion , which recently seems to be all but extinct . as goes without saying , every contribution , correction , criticism or suggestion is greatly appreciated . the discussion , which lasted for the major part of the early stages of atomic , nuclear and particle physics , as we know them today , dealt primarily with the scientific and logical implications of quantum theory . again revived by david bohm s contribution , the `` hidden variables '' , it gradually began to loose on impact , until today any fundamental question in this respect seems to be considered a lack of scientific soundness . from an epistemological point of view , the fact is not easily understood , since all the fundamental problems are still far from any solution . second quantization , on the contrary , provides an even more problematic conception by its inherent infinity problems @xcite , which , till now , have not been solved in a satisfying way . that all the original problems are still there , can be seen by a survey of late publications of bohm @xcite , de broglie @xcite , dirac @xcite , or schrdinger @xcite . from the viewpoint of common sense the most profound reason of uneasiness with quantum theory , despite its indisputable successes , has been expressed by richard feynman : _ you see , my physics students do nt understand it either . that is , because i do nt understand it . nobody does . _ @xcite that the discussion , mentioned above , is nevertheless non existent , might have its origins therefore not so much in a satisfying status quo , but rather in a lack of alternative concepts . from an epistemological point of view a well working theory , despite all its conflicts with logical reasoning , is better than no theory . the discussion then leads to the question , why so far no promising concept has been put forth . comparing with other sciences , the situation seems very unusual : seventy years of continuous research from schrdinger s first publications @xcite have neither altered the principles of quantum theory , nor have they removed the axiomatic qualities of its fundamental statements . this situation basically leaves two possibilities : either to accept , that we can not understand , i.e. that we _ have to believe _ , or to search for alternatives . the attempts so far , to remove the axiomatic qualities of quantum theory , which can be summarized by the _ copenhagen interpretation _ @xcite , mainly focussed on the statistical qualities attributed to fundamental physical events . it was thought , that a suitable theoretical framework could prove quantum events to be integral results of basically continuous processes . the first of these attempts can be attributed to erwin schrdinger @xcite . his wave mechanics , now a fundamental part of quantum theory , does not allow for an interpretation of wave functions as _ physical _ waves due to the contradiction a spreading wave packet presents to mass conservation . on a fundamental level , the contradiction originates in the difference between phase velocity and group velocity of de broglie waves @xcite . since schrdinger s equation is based on exactly these qualities of de broglie waves , it can not be retained without rejecting the interpretation of wave functions as physical waves . the problem was solved by born s interpretation @xcite of @xmath4 as a _ probability amplitude _ , an interpretation which constitutes an integral part of the copenhagen interpretation . this interpretation of the wave function is not without its inherent problems , as einstein , podolksky and rosen pointed out @xcite . if quantum theory contained all the physical information about a spin system , thus their line of reasoning , then information about a measurement process must be exchanged with a velocity faster than c , if spin conservation is to be retained . this leaves , as they point out , only two alternatives : either quantum theory contradicts the theory of relativity , or it does not contain all the information about the system : in this case , quantum theory is incomplete . the scientific controversy was subject to much discussion , but never conclusively solved . recently , its implications have been extended by the phenomenon of interaction free measurements @xcite , based on a thought experiment by renninger @xcite and realized by a mach zehnder interferometer . interaction free measurements have different implications in a classical or quantum context : while a measurement without interaction in classical physics does not alter the system , it leads to a reduction of the wave function in quantum theory . according to the solution , proposed by heisenberg ( see ref.@xcite ) only the reduction of the wave function constitutes physical reality : in this case the question is legitimate , why a wave function , obviously devoid of any physical meaning , plays such an important role in current theories . david bohm s famous theory of hidden variables @xcite was a far more elaborate concept along the same lines . he tried to prove , that the formalism of quantum theory allows for a sub layer of hidden variables , and that the framework therefore is , as einstein suspected , incomplete . the modification was rejected due to the work of bell , who devised a simple situation , where this assumption could be put to a test @xcite . the bell inequalities proved that , within the framework of quantum theory , such an assumption leads to logical contradictions . the theory of hidden variables was therefore rejected . the combined evidence of all these attempts allows the conclusion , that a modification of quantum theory within its fundamental axioms , yielding a framework of which the quantum aspect will be but a result , is not very promising . at best , it reproduces numerical results by a different mathematical structure , at worst , it not even accounts for basic experiments . there remain , therefore , three fundamental questions to be answered prior to any attempted modification of micro physics : * on what theoretical or experimental basis could an alternative formulation be based ? * which theoretical concept of physics , statistical , field theory , or mechanics , shall be employed ? * what constitutes a successful alternative to current formulations ? experimental evidence exists , that moving electrons exhibit wave like features . these qualities have been established beyond doubt by diffraction experiments of davisson and germer @xcite . quantum theory can not directly access wave like qualities , because , as already mentioned , the periodic wave function needs to be subjected to a complicated evaluation process in order to extract its physical relevance . if the initial problems of wave dispersion and velocity can be overcome , the experimental result of wave characteristics of moving mass and its mathematical formalization may well constitute the foundation of a different theoretical conception . the only other experimental evidence at hand with equivalent significance , the quantum phenomena , do not as easily fit into a physical interpretation , since the basic obstacle , why discrete values should be the only ones measured , has to be removed : quantum phenomena are therefore more remote in terms of basic concepts . concerning the theoretical context , it has to be considered , that quantum theory originally started as a modification of mechanics . most of the concepts of the theory are consequently still disguised mechanical concepts . field theoretical results are therefore only achieved by a statistical superstructure : the mechanical basis then leads to a statistical formulation of field theories , which allows for particle interactions , but not for internal particle structures . modifications of this methodology basically occur along two lines : either by a modification of interactions , which requires additional qualities of particles , or by postulating the existence of sub particles . physical progress then requires an ever refined toolkit of particles , sub particles and interactions . from the viewpoint of epistemology , the problem inherent to this approach is the limited degree of freedom . every modification requires a major change of particle properties , and since every new experiment provides data not yet accounted for , the development of physics becomes an ever increasing inventory of individual particles . the particle zoo , high energy physicists so frequently refer to , can , along these epistemological lines , be seen as a consequence of theoretical shortcomings . since the experimental basis the wave characteristics and thus the internal structure of particles suggest an approach based on field theory rather than mechanics , this essay will be based on field theory . if mechanical concepts are required , they always refer to continuous distributions of physical qualities , which yield discrete mass or charge values only by integration . if one is to try an alternative route of theoretical development , it seems important , not to get carried away and extend a concept beyond its limits of proper application . to this end , a goal has to be defined as well as careful checks of experimental evidence . a theory , which does not allow for falsification , remains useless , the same applies to a new theory , which does not deviate from the current one by at least one significant result . that a new theory can not and should not reproduce all results of an existing one , goes without saying personally , i think that the following results are sufficient to justify serious consideration of a new approach : * the reproduction of fundamental results of early quantum theory * the formalization of internal particle structures and the proof of physical significance of these structures * a model of ( hydrogen ) atoms consistent with the formalism developed and consistent with experimental evidence * the basics of a theory of particle interaction due to intrinsic characteristics * provision of at least one significant deviation which allows for an experimentum crucis the fundamental relations and concepts of this new approach have already been published or are in print in the _ speculations in science and technology _ ( chapman and hall , london ) @xcite . the accounts given in this and follow up papers are somewhat enhanced and thorougher versions , making use of the greater freedom of length . as already emphasized , any feedback is greatly appreciated and readers are welcomed to contact me directly via email at the email address : whofer@eapa04.iap.tuwien.ac.at to allow for internal characteristics of moving particles we base the following section on the experimental evidence of wave features . theoretically , the framework will be based on a suitable adaptation of the fundamental relations found by l. de broglie . originally , de broglie s formalization of waves @xcite was based on the framework of special relativity , the amplitude of a de broglie wave given by : @xmath5 the variables are four dimensional vectors , their components are described by the following relations : @xmath6 in a three dimensional and non relativistic ( @xmath7 = 1 ) adaptation of the original equation ( [ ws001 ] ) the amplitude can be expressed as a plane wave . this adaptation is quite common in quantum theory and therefore no deviation from the original framework : @xmath8 the first major deviation from the traditional framework is the interpretation of @xmath9 as a real function , as a physical wave . this conception is not compatible with quantum theory for the following reasons : as a moving particle can not consist of a specific plane wave ( particle mass in this case would be distributed over an infinite region of space ) , a moving particle in quantum theory _ must _ be formalized as a fourier integral over partial waves . an application of schrdinger s equation then leads to the , above mentioned , spreading of the wave packet , which contradicts experimental evidence . the interpretation of the wave function as a _ physical entity _ , furthermore , contradicts heisenberg s uncertainty relation @xcite , because it implies the possibility of measurements below quantum level . @xmath10 the reason , that this real formalization of de broglie remained so far unconsidered , is not only its inconsistency with quantum theory , but also a consequence of planck s relation : dispersion of de broglie waves is described by two independent relations , which are widely confirmed in experiments . the first is the wavelength of de broglie waves , established by diffraction experiments of davisson and germer @xcite , the second planck s energy relation @xcite : @xmath11 if particle energy is classical _ kinetic _ or relativistic energy , the phase velocity of a de broglie wave is not equivalent to particle velocity : @xmath12 the deduction of phase velocity , leading to the contradiction in eq . ( [ ws006 ] ) is , contrary to a first impression , not self evident . it contains , on thorougher analysis , a conjecture and a subsequent logical circle . firstly , it is not self evident , that on a micro level energy of a particle is restricted to _ kinetic _ energy of inertial mass : assuming , that a particle is , basically , a mechanical entity without any internal energy component , can not be inferred from experimental or theoretical results . it is therefore an _ assumption _ lacking a solid basis . based on this conjecture is the following logical circle : _ if energy is kinetic energy , phase velocity of a de broglie wave is not equal to mechanical velocity of a particle . if phase velocity does not equal mechanical velocity , a free particle can not consist of a single wave of specific frequency . if it does not consist of a single wave , then the wave features of a particle must be formalized as a fourier integral over infinitely many partial waves . in this case any partial wave can not be interpreted as a physical wave . if the interpretation as a physical wave is not justified , then internal wave features can not be related to physical qualities . if they can not be related to physical qualities , then internal processes must remain unconsidered . and if internal processes remain unconsidered , then energy of the particle is kinetic energy _ apart from the question , whether this assumption is necessary , which can only be answered a posteriori , it seems also legitimate to ask , whether it is the best possible assumption , especially in view of complications arising from fundamental particles formalized as fourier integrals over infinitely many partial waves . from the viewpoint of simplicity , the concept is necessary , if no alternative exists , it is unnecessary complicated , if the same theoretical result the wave features of single particles can be obtained by a simpler formalism . to this aim alternatives have to be analyzed . a direct way of formalization would be based on two basic statements : ( 1 ) the experimental evidence of wave features , and ( 2 ) the assumption , that wave features must be reflected by _ physical _ qualities of particles , density for example . mathematically , these statements are expressed by : @xmath13 @xmath14 in view of the dispersion relations eq . [ ws005 ] these relations seem contradictory : the standard evaluation procedure would , for a specific _ physical _ wave , lead to a difference between phase velocity and mechanical velocity in conflict with the energy principle . this contradiction depends , though , on the principal conjecture , that a moving particle is equivalent to inertial mass , and that all its energy is contained in the longitudinal motion of an inertial mass @xmath15 with velocity @xmath16 . as soon as wave features are assumed to describe the physical nature of particle motion , this conjecture can no longer be sustained : energy of the particle will then be contained in its _ mass oscillations_. since mass oscillations determine a periodic change of kinetic energy , the energy principle requires the existence of an _ intrinsic _ potential of particle motion , and averaging over kinetic and potential energy during particle motion then leads to a total energy of double the original kinetic value . using this total energy for planck s relation satisfies the dispersion relation of a monochromatic plane wave : based on the intrinsic properties of physical waves any monochromatic plane particle wave then satisfies the wave equation . averaging density of the particle wave over a full period we get initially : @xmath17 and the kinetic energy of particle motion is therefore ( @xmath18 denotes the volume of the particle ) : @xmath19 since kinetic energy is not constant but periodic , the result requires that an intrinsic potential exists , which yields an equivalent average of potential energy : @xmath20 and total energy of the particle wave is then : @xmath21 phase velocity of this particle wave equals , consistent with quantum theory , mechanical velocity of particle motion , although , contrary to quantum theory , no fourier integral over partial waves is required . therefore any monochromatic plane particle wave , consisting of oscillating mass and a complementary potential , describes motion of a _ free particle _ : @xmath22 from a theoretical point of view the existence of intrinsic potentials is a _ conjecture _ , it is more problematic than the original one based on inertial mass , since it implies the existence of physical variables not accounted for in the current framework of quantum theory . relating @xmath23 to total energy of the particle is , in this respect , not problematic , since experimentally energy of a particle can only be estimated by its state of motion or wavelength : and the relation between particle velocity @xmath24 , wavelength @xmath25 , and frequency @xmath26 is retained . with the dispersion relation given by eq . [ ws013 ] the wave function @xmath27 and density @xmath28 of free particles comply with a wave equation : @xmath29 the wave equations display intrinsic features of particle motion and possess , as will be seen in the following sections , the same dimensional level as electrodynamic formulations . the logical structure of physical statements is therefore modified : while in the standard procedure of quantum field theory the fundamental statements of quantum theory determine events on a micro level , and electromagnetic fields are logical superstructures , in the present context statements about the wave features of particles are different aspects of the same physical phenomena dealt within electrodynamics . this parallel development of electrodynamics and quantum theory from the same root the intrinsic wave features of particles must be understood in a literal sense , because the potentials deduced from mass oscillations will be related to electromagnetic fields . the schrdinger equation , in this context , is an expression of mechanical features i.e. potential and kinetic energy of particle motion . from the viewpoint of physical waves it is not without a certain degree of arbitrariness . to derive the relation we proceed from the periodic wave function eq . [ ws013 ] . then the time differential is given by : @xmath30 kinetic energy @xmath31 of particle motion will therefore be : @xmath32 and substituting eq . [ ws015 ] into [ ws016 ] we get for the kinetic energy of the particle : @xmath33 if the total energy of a particle is equal to kinetic energy and a potential @xmath34 : @xmath35 then the wave equation in the presence of an external potential is described by : @xmath36 the relation is equivalent to a time free schrdinger equation @xcite . it is important to consider , that this ( mechanical ) formulation of the wave equation explicitly relates physical potentials to kinetic energy and thus the wave length of a material wave ( as a result of the laplace operator acting on @xmath0 ) and that , therefore , in a region , where @xmath34 takes successively different values , every kinetic energy and thus wave length is generally possible . its interpretation in the context of material waves is not trivial , though . if the volume of a particle is finite , then wavelength and frequency become _ intrinsic _ variables of motion , which implies , due to energy conservation , that the wave function itself is a measure for the potential at an arbitrary point @xmath37 . if it is a measure for the physical conditions of the environment i.e. the intensity of @xmath34 , then it _ must _ have physical relevance . but if it has physical relevance , then the epr dilemma @xcite will be fully confirmed , and the result will favor einsteins s interpretation of quantum theory : quantum theory can not be complete under the condition that ( 1 ) particles do have finite dimensions , and ( 2 ) the schrdinger equation is a correct description of the variations of particle waves in the presence of external fields . the only alternative , which leaves quantum theory intact , is the assumption of particles with zero volume : an assumption which leads , in the context of electrodynamics , to the equally awkward result of infinite energy of particles . the first major result of this essay can therefore be formulated as follows : * the interpretation of particles as physical waves allows the conclusion , that quantum theory can not be , in a logical sense , a complete theoretical description of micro physical systems . there exists another problem usually hidden in the mathematical formalism of quantum theory . let an external potential be given by @xmath34 . transforming eq . [ ws019 ] into a moving coordinate system @xmath38 described by : @xmath39 we get the following equation : @xmath40 since the variable @xmath41 in this case is undefined , the relation contains an element of arbitrariness . it is not conceivable , from this deduction , how a time free schrdinger equation could be derived : if the time dependent part of @xmath0 is eliminated , the procedure is equivalent to a transformation into a moving coordinate system , but in this case the potential becomes a function of time . and if the wave equation is integrated in the system at rest , then the wave functions depend on time . from a physical point of view , the reduction of the schrdinger equation to its time free form actually eliminates all the time dependent interaction and adjustment processes bound to occur for _ every _ particle wave interacting with a changed environment . it may well be , that this disregard of development accounts for the commonly rather static concepts in quantum theory . within the framework of material waves , the arbitrariness of the schrdinger equation can be referred to the intrinsic potential of mass oscillations , not accounted for in quantum theory . and its consequence , as can equally be derived , is an uncertainty about the exact value of @xmath42 of the wave function @xmath43 . the uncertainty of @xmath42 together with the period of intrinsic potentials @xmath44 is then sufficient to derive heisenberg s uncertainty relations @xcite . to this aim we proceed from the schrdinger equation in the moving frame of reference eq . [ ws020 ] . setting the upper and lower limits of the potential @xmath45 to : @xmath46 and calculating the correlating @xmath47vectors of the wave function @xmath48 : @xmath49 we get the following relation : @xmath50 evaluating the relation in a one dimensional model and accounting for the yet unknown variation of the potential @xmath51 by : @xmath52 the uncertainty @xmath53 due to the undefined variable t will be : @xmath54 if the uncertainty originates from the neglected intrinsic potential @xmath55 ( see eq . [ ws011 ] ) , the uncertainty of the applied potentials is described by : @xmath56 and therefore the uncertainty of @xmath47 can be referred to the value of @xmath47 : @xmath57 the uncertainty does not result from wave features of particles , as heisenberg s initial interpretation suggested @xcite , nor is it an expression of a fundamental physical principle , as the copenhagen interpretation would have it @xcite . it is , on the contrary , an expression of the _ minimum error _ , inherent to _ any _ evaluation of the schrdinger equation , resulting from the fundamental assumption in quantum theory , i.e. the interpretation of particles as inertial mass aggregations . since the distance between two potential maxima is given by ( the value can deviate in two directions which yields total uncertainty of the @xmath58coordinate ) : @xmath59 the uncertainty may equally be written as : @xmath60 heisenberg actually derived @xcite the uncertainty relations @xmath61 . the fourier transform inherent to quantum theory suggests a correction of the constant eq . [ ws027 ] by a factor @xmath62 . in this case the deduction yields _ exactly _ the same result as the canonical formulation of quantum theory @xcite : @xmath63 the minimum error due to intrinsic potentials is therefore equal to the uncertainty of quantum theory , and the logical relation between schrdinger s equation and the uncertainty relations is therefore a relation of reason and consequence : _ because _ schrdinger s relation is not precise by the value of intrinsic potentials , the uncertainty relations _ must be _ generally valid . this is the second major result of the present essay : * _ because _ quantum theory neglects intrinsic characteristics of particle motion , the schrdinger equation is generally not precise by a value described by heisenberg s uncertainty relations . the result equally seems to settle the long standing controversy between the _ empirical _ and the _ axiomatic _ interpretation of this important relation : it is not empirical , since it does not depend on any measurement process ; but it is equally not a physical principle , because it is due to the fundamental assumptions of quantum theory . the solution to the problem of interpretation is therefore a somewhat wider frame of reference : although within the principles of quantum theory the relation is an _ axiom _ , it is nonetheless a _ result _ of fundamental theoretical shortcomings and not a physical principle . in the preceding section we could demonstrate , that quantum theory must not necessarily be a comprehensive representation of physical reality . it was shown , on the contrary , that the principal assumption of quantum theory , the assumption of inertial mass of particles , is a posteriori neither justified by the wave features of particles nor by the fundamental planck and de broglie relations . it is but _ one _ possible theoretical framework and , given the logical problems of a precise relation between measurements and reality arising from this concept , not even the best possible framework . on an analytical level the establishment of fundamental relations of quantum theory within a different framework and their significance that the uncertainty relations can also , and consistently , be interpreted as an expression of theoretical shortcomings due to a disregard of intrinsic properties served to establish the limitations of the current theoretical standard . they also served to define the logical _ location _ of quantum theory and to determine the regions of interest , where research is commendable if a theory _ outside _ the current framework is taken into consideration . from the viewpoint of theoretical synthesis we showed that the intrinsic properties of particles and their relations are a consequence of the experimental evidence of wave features of particles . the de broglie relations , in a non relativistic formulation , provided the essential laws by which these intrinsic variables are described . they allowed to establish ( 1 ) periodic mass oscillations within the particle , and ( 2 ) the existence of a periodic potential . these results were sufficient to deduce the fundamental relations of quantum theory _ and _ , at the same time , to show their inherent limitations . apart from showing the significance of the periodic potentials which will be done in the following sections the first two requirements mentioned in the introduction are therefore fulfilled . fundamental results of early quantum theory are reproduced ( schrdinger s relation as well as the uncertainty relations , while single experiments will be treated in the following sections ) , and the intrinsic wave features of particles were shown to be of _ physical _ significance , which strongly supports einstein s view in the epr paradox . classical electrodynamics is considered one of the most refined and powerful physical theories at hand . it served , during the last hundred years , ever again as a model for physical formalization , and its implications led to quite a few extensions of physical knowledge , not least albert einstein s theory of special relativity @xcite . a great deal of the fascination of classical electrodynamics results from its mathematical conciseness and brevity . the possibility to describe and account for most of the phenomena concerning interactions of charge by a set of basically four linear differential equations , the _ maxwell equations _ @xcite , must either seem a lucky coincidence or an expression of a deeply rooted analogy in physical reality . that classical electrodynamics can not be referred to mechanical conceptions , has been proved during the scientific controversy about the existence of the _ aether _ . this proposed medium of electromagnetic fields was never detected , its existence , furthermore , would contradict the experimental results of michelson and morley @xcite . from the theoretical side , the controversy was settled by einstein s theory of relativity . since the correctness of maxwell s equations is established beyond doubt , it remains an epistemological problem , why no attempt to refer them to fundamental physical axioms has succeeded so far @xcite . summarizing the current situation in two statements : ( 1 ) maxwell s equations are correct , and ( 2 ) they can not be referred to mechanical axioms , there exist equally two logical conclusions to the epistemological problem : ( 1 ) maxwell s equations are themselves of an axiomatic nature , and ( 2 ) the axioms , on which they are based , have not yet been detected . currently , the established opinion is ( 1 ) . basically , one could leave it at this state and employ the relations as correct algorithms . from the viewpoint of macro physics , the procedure is justified by the wealth of experimental verifications at hand . the same does not apply , though , to an application of electrodynamics in micro physical models . since the exact origin of the relations is obscure , it can not be guaranteed , that they do not , in some way , preclude characteristics of physical systems , which contradict other assumptions . obviously , there exist only two ways out of this logical dilemma : either it can be proved , that quantum theory is logically independent of electrodynamics a proof , which never has been given , or it can be established that electrodynamics itself is based on micro physical axioms : in this case the current procedure in qed @xcite is logically insufficient . the logical conclusion to our epistemological problem would then be : ( 2 ) the maxwell relations are an expression of intrinsic characteristics of particles . based on the intrinsic wave characteristics of free particles , developed in the previous section , the verification of ( 2 ) does not present an impossible problem . the maxwell equations , in material wave theory , are an expression of longitudinal variations of intrinsic variables of particles in motion . from the wave equation for longitudinal momentum @xmath64 : @xmath65 and the equation of continuity for momentum @xmath64 and density @xmath66 : @xmath67 it follows within a micro volume , where particle velocity @xmath24 is constant : @xmath68 the parameter @xmath69 is a dimensional constant employed to make mechanical units compatible with electromagnetic variables , its dimension is equal to density of charge . electric fields @xmath70 and magnetic fields @xmath71 are now defined in terms of momentum @xmath64 and the potential @xmath72 due to mass oscillations ( see the previous section ) : @xmath73 and substituting eq . [ ed003 ] into [ ed001 ] we get , with the help of [ ed004 ] and [ ed005 ] : @xmath74 and since the total intrinsic potential is a constant : @xmath75 the following relation holds in full generality within every micro volume : @xmath76 rotation of eq . [ ed004 ] yields , but for a constant , maxwell s first equation : @xmath77 to deduce the inhomogeneous equation we additionally have to employ the source equations of electrodynamic theory and the definition of @xmath78 fields : @xmath79 it is assumed that the material constants @xmath80 and @xmath81 also remain constant within the micro volume . substituting into eq . [ ed008 ] and computing the source of [ ed010 ] , the sum yields : @xmath82 and if the constant of integration is zero , the result will be maxwell s inhomogeneous equation : @xmath83 finally , by computing the time differentials of eqs . [ ed008 ] and [ ed009 ] , the wave equations for electromagnetic fields in a vacuum ( @xmath84 ) can be deduced : @xmath85 the results indicate , that electrodynamics and quantum theory do have a common root , the _ intrinsic _ properties of moving particles . from the viewpoint of material waves they are therefore different aspects of the same physical phenomenon , the wave features of mass , which is the third major result of this essay : * electrodynamics and wave mechanics are theoretical concepts based on the same physical phenomena , the intrinsic features of moving particles . logically speaking , the result means , that a method , which employs quantum theory to calculate physical variables by way of observables , and which identifies these observables measured variables as field variables in electrodynamics can not be correct . the energy principle of material waves can be identified as the classical lorentz standard by a combination of the continuity equation and the definition of electric fields including intrinsic potentials . since : @xmath86 the wave equation for linear motion @xmath87 leads to : @xmath88 and for uniform motion @xmath89 the potential can be related to momentum @xmath90 by : @xmath91 comparing with the expression in classical electrodynamics the vector potential @xmath92 , in classical electrodynamics an abstract conception , is proportional to momentum of particle motion : @xmath93 the wave features of photons , or the representation of electromagnetic fields by wave type field vectors , have long been the only aspect in consideration . the picture changed only after experimental research was directed towards the fundamental problems of interaction between matter and electromagnetic fields . due to planck s result on black body radiation and einstein s application of planck s formula to the problem of photo electricity @xcite , the energy of photons gained special attention , since it was ( 1 ) proportional to frequency and not , as field theory proposed , intensity , and ( 2 ) the energy quantum @xmath94 could no longer be distributed locally , because the energy changes in interactions required an exchange of a full number @xmath95 of energy quanta . in einstein s view the interaction suggested that energy of photons be distributed in single points within the electromagnetic fields of radiation . according to the copenhagen interpretation of quantum theory @xcite the two theoretical aspects wave features and energy quanta describe complementary facts of a complex physical reality not adaptable to any single theoretical framework . this interpretation , which focusses on the central notion of wave particle duality , states that all the qualities like polarization and interference , typical for a wave type theory , have to be treated by electrodynamics , while energy and interaction phenomena must be seen from the viewpoint of quantum theory . the logical justification for retaining the duality , which is in itself a logical contradiction if wave features and particle features were to prevail _ simultaneously _ , was seen by jordan @xcite and heisenberg @xcite in the impossibility of experiments i.e. physical processes where both aspects _ must _ simultaneously exist . this justification is not undisputed . renninger showed by a rather simple thought experiment @xcite , that an interference measurement involving two separate light paths does not allow for the probability interpretation of electromagnetic field vectors , since the result of the measurement can be altered in a deterministic manner by the insertion of optical devices into a single light path . the analysis of the experiment leads to the conclusion , that the wave features _ must have _ physical significance , which again seems to confirm the epr dilemma already quoted @xcite . based on this experiment renninger deduced the following statement about the features of photons ( translation by the author , emphasizing by renninger ) : _ every light quantum consists of an energy particle , which is guided by a wave without energy _ the important result , from our point of view , is the physical significance of wave features . the energy particle , which renninger refers to , is a direct result of einstein s interpretation of photo electricity , which equally determines the interpretation of the guiding wave as energy free . from the viewpoint of physical concepts , an energy free _ physical _ wave is not all too convincing . if a region of space is altered by a propagating wave , thus the obvious objection , then it can not be without a physical entity moving . but since all physical field variables are , in some way , related to energy , a physical entity propagating without any propagation of energy seems contradictory . on thorougher analysis the necessity to divide between two separate physical entities to account for energy as well as electromagnetic features of photons depends on the interpretation of the interaction process itself . since ( 1 ) the quantum of energy is fully transferred , and ( 2 ) the transfer interval is well below the level of measurements , the energy must be contained in an insignificantly small volume . the alternative being , as renninger pointed out , that the wave will be contracted with a velocity exceeding @xmath96 during every interaction process . it is this interpretation of the interaction process , which will be changed by material wave theory . to understand the decisive modification , we present here and without proof a result on the interactions of electrons and photons , which will be derived later on . assuming that the interaction interval equals , and setting @xmath97 for convenience equal to 1 , an intrinsic model of particles and fields will have to consider energy density at a specific point _ within _ the particle . if total energy density is constant and @xmath98 , then the time derivative of energy density equals : @xmath99 in this way the traditional features of photons are retained , while the meaning of the term _ energy quantum _ is modified : the energy quantum _ transferred _ in an interaction process equals @xmath100 , regardless of the actual volume of the photon , and equally regardless of the exact duration of the interaction process . as long as the total density of energy @xmath98 is constant , the relation holds for every single point of the underlying field and therefore for the whole region of the physical wave corresponding to a photon . now let us consider a case , where the initial photon is , by some optical device , split into two . if one of the partial volumes a half photon so to say then interacts with matter , the process is described by : @xmath101 the energy density transferred remains constant , but not total energy . the process therefore _ seems _ to be a confirmation of the principle , that all interactions occur in discrete energy quanta only , if discrete quantities of mass or charge are considered . only , in short , if the general outlook on processes is a mechanical one . it is no contradiction with einstein s calculation on photo electricity , to interpret it as an effect of _ energy densities _ and transfer rates . it only has to be considered , that the total volume affected by interactions will depend on the total volume of the fractional photon . from this result it can be concluded , that the term _ energy quantum _ is not appropriate for the physical process of interaction : since the total energy transferred depends on ( 1 ) the frequency of radiation , and ( 2 ) the volume of the photon , total energy transferred is the product of energy density proportional to frequency and volume which may be related to intensity . in an _ intrinsic _ formalism of interaction , furthermore , the total quantities transferred are only secondary , the more so , since electrodynamics already is a theory of intrinsic wave features , as deduced above . in material wave theory photons are distributions of mass and energy , and the guiding wave possesses the same dimensions as the energy distribution , while the guiding wave as well as mass movement contribute alike to total energy of the photon . to derive the essential features of photons we use the fundamental wave equation of intrinsic particle momentum . the exact dimensions as well as the problems of photon boundaries are shown to be of no effect for the results achieved . denoting an , initially , abstract momentum of a photon by @xmath102 , we proceed from the wave equation in a vacuum given by : @xmath103 with the identity for the laplace operator and the general relation between momentum @xmath102 and vector potential @xmath104 from eq . ( [ ed016 ] ) : @xmath105 the wave equation may be rewritten as : @xmath106 using the solution for the homogeneous maxwell relations ( the following deduction employs , for convenience , gaussian units ) : @xmath107 the wave equation yields the following result : @xmath108 we choose the specific solution by setting the term in brackets equal to zero and get therefore : @xmath109 the equation can be solved by any wave packet consisting of plain waves and complementary potentials given by : @xmath110 for a component @xmath111 of the wave packet the amplitudes must comply with : @xmath112 setting @xmath113 equal to @xmath114 and using the amplitudes of material waves as well as vacuum dispersion we recover einstein s energy relation @xcite : @xmath115 we define now a _ photon _ as any component of the wave packet . more specifically , we describe a photon by a plane material wave and its complementary electromagnetic potential . @xmath116 the _ energy density _ at a specific location of the photon is denoted by an electromagnetic potential and a potential due to the vector @xmath90 . if we identify @xmath90 as a longitudinal material wave , its energy density at any arbitrary point , the _ total potential _ , is double its value in classical mechanics . @xmath117 then the material aspect of a photon in motion is that of a material wave , while , due to additional electromagnetic characteristics , total energy density is constant and does not depend on location or time : @xmath118 the electromagnetic potential must correlate with electromagnetic field vectors . using electromagnetic potentials of a vacuum : @xmath119 and assuming for consistency with classical electrodynamics , that electromagnetic fields will be of transversal polarization , the transversal fields of a photon will be given by : @xmath120 @xmath121 transversal electromagnetic fields of a wave packet are therefore described by : @xmath122 the units of electromagnetic fields will be treated on a more fundamental basis in the following sections , presently they only signify a constant value of some suitable dimension . in the section on electrodynamics of his famous publication einstein @xcite referred the universal validity of maxwell s equations to their lorentz invariance . based on the theoretical framework of intrinsic features of variables , it is tempting to relate the lorentz invariance of maxwell s equations to the invariance of the wave equations under a lorentz transformation . if it can be established , that the lorentz transformation is also the transformation leaving intrinsic features of particles intact , then the kinematic section of the theory of relativity can also be seen as a necessary modification of mechanical concepts to account for electromagnetic phenomena . as the maxwell relations are not only valid for intrinsic features of photons but also electrons which derives from their independence of particle velocity u we proceed from the wave equation for the intrinsic momentum @xmath64 of a particle with arbitrary velocity @xmath123 in a moving frame of reference s with @xmath124 relative to a system s at rest . the wave equation in s then will be : @xmath125 with the standard lorentz transformation : @xmath126 the differentials in s are given by : @xmath127 density of mass and velocity in s are described by ( transformation of velocities according to the lorentz transformation , @xmath128 presently undefined ) : @xmath129 then the transformation of the wave equations yields two wave equations for momentum @xmath130 and density @xmath66 : @xmath131 energy measurements in the system s at rest must be measurements of the phase velocity @xmath132 of the particle waves . since , according to the transformation , the measurement in s yields : @xmath133 the total potential @xmath98 , measured in s will be : @xmath134 and if density @xmath135 transforms according to a relativistic mass effect with : @xmath136 then the total intrinsic potential measured in the system s will be higher than the potential measured in s : @xmath137 but the total intrinsic energy of a particle with volume @xmath18 will not be affected , since the x coordinate will be , again according to the lorentz transformation , contracted : @xmath138 the interpretation of this result exhibits some interesting features of the lorentz transformation applied to electrodynamic theory . it leaves the wave equations and their energy relations intact if , and only if energy _ quantities _ are evaluated , i.e. only in an integral evaluation of particle properties . in this case it seems therefore justified , to consider special relativity as a necessary adaptation of kinematic variables to the theory of electrodynamics . the same does not hold , though , for the intrinsic potentials of particle motion . any transformation into a moving reference frame in this case leads to an increase of the total potential @xmath98 , in the case of photons ( @xmath139 ) the intrinsic potential then will be infinite . the essential result of this analysis of special relativity may be expressed by : * the wave equations of intrinsic particle structures are lorentz invariant only for an integral evaluation of energy . the total intrinsic potentials increase for every transformation into a moving frame of reference . that maxwell s relations can be referred to intrinsic characteristics of free particles , proves two assumptions : ( 1 ) the relations are themselves not of an axiomatic nature , contrary to current opinions they can be referred to a sublayer of still more fundamental axioms . ( 2 ) quantum theory and electrodynamics are not logically independent , since both concepts are a formalization of wave features of particles . the features of photons in material wave theory equally seem to fit into the framework of classical electrodynamics : electromagnetic field vectors are transversal , while the longitudinal component is purely kinetic . both aspects together provide a model of photons , which is simultaneously an extension of electrodynamics ( since the variables of motion exist in addition to the electromagnetic field vectors ) , and an extension of quantum theory ( since the electromagnetic fields are not originally included in the kinetic model of quantum theory ) . relativistic electrodynamics only provides lorentz invariant formulations of intrinsic particle features , if energy quantities are evaluated . since the intrinsic potentials depend on the reference frame of evaluation and increase by a transformation into a moving coordinate system , the theory suggests that the infinity problems , inherent to relativistic quantum fields , may be related to the logical structure of qed . while the definition of magnetic fields eq . [ ed005 ] allows to deduce the maxwell equations and accounts for intrinsic properties of photons , it is incorrect , by a factor 2 , if magnetostatic interactions of electrons are considered . the definitions , describing the relations between the kinetic variables @xmath140 and the magnetic field @xmath141 , of photons and electrons correctly are written : @xmath142 the reason for this difference , which bears a superficial resemblance to the distinction between bosons and fermions ( s = @xmath143 ) , is that the two relations refer to different magnetic fields . on thorougher analysis , two different cases have to be considered : ( 1 ) the intrinsic magnetic fields , deriving from the kinetic properties of motion and oscillation , and ( 2 ) the external magnetic fields due to curvilinear motion of charge . while for photons the intrinsic electromagnetic fields have been known and dealt with in the framework of classical electrodynamics , for electrons these intrinsic fields can only be derived within material wave theory but not in quantum theory . and while the maxwell relations as an expression of intrinsic features are valid for both types of particles , the magnetostatic properties of charge apply only for electrons . the correction arises therefore from the treatment of different physical processes , and is easily understood as a reminiscence to the historical origins of the separate theories . it is tempting , at this stage , to refer the difference between fermions and bosons to the same historical origins , adopted for the concept of spin without further reflection . that the spin of particles can indeed be referred to magnetic fields , will be shown in section [ ep_spin ] . to account for magnetostatic interactions the definition of magnetic fields therefore has to be corrected . in _ homogeneous _ magnetic fields @xmath144 electron motion can be treated based on the changed definition of external magnetic fields eq . ( [ ep000 ] ) : @xmath145 the sources of @xmath146 vanish , since : @xmath147 the solution is equivalent to the traditional one : @xmath148 circular rotation in homogeneous magnetic fields is not subject to quantization , because the local component of de broglie s relation vanishes . particle density @xmath149 in any state of rotation is given by : @xmath150 then the relation between magnetic field @xmath71 and angular velocity @xmath26 will be , with eq . [ ep000 ] : @xmath151 and since @xmath152 it follows generally : @xmath153 considering the classical formulation for lorentz forces of charge in a magnetic field @xmath154 , we may write : @xmath155 which reduces in the average @xmath156 to : @xmath157 additionally , it has to be considered that rotation of mass in the rotating coordinate system gives rise to centrifugal forces described by : @xmath158 then the rotation of charge is , in the rotating coordinate system , free of forces , it is therefore its _ inertial _ state of motion in a magnetic field @xmath71 : @xmath159 but as centrifugal forces are inertial forces arising from the curvilinear path of motion , the same must apply to lorentz forces : a magnetic field determines rotation of charge , which experiences centrifugal and lorentz forces . which means , evidently , that lorentz forces must be inertial forces . the framework developed so far has some interesting consequences in electrostatic problems . defining a vector field @xmath160 by : @xmath161 and using the equation of continuity , which is equally valid for electron charge distributions : @xmath162 the vector field will be equivalent to : @xmath163 since momentum @xmath102 can be expressed in terms of @xmath164 , the wave equation also applies : @xmath165 if static problems are considered , @xmath166 will be independent of time , and using the definition of magnetic fields as well as ( [ ep002 ] ) , the statement can be rewritten in the following form : @xmath167 it is equivalent to _ ampere s law _ , if electrodynamic current density is defined by : @xmath168 in linear approximation , defining specific resistance @xmath169 by : @xmath170 and using the averages of density as well as neglecting boundary conditions , electromagnetic potentials inside a conductor will be due to sources of the electron wave vector @xmath42 : @xmath171 which means , otherwise , that the driving force of electrostatic current is the gradient of electron velocity . a possible way to show , how the system of physical units is affected by this , basically field theoretical approach to particle motion and interaction , is the following : by modeling a framework of particle motion consistent with classical electrodynamics , and comparing the energy expressions derived with energy of electron waves of hydrogen atoms , the unit of current and the electromagnetic coupling constants will be defined in terms of motion , they become _ mechanical _ units . if we consider the complementary fields of an electron or photon , we may describe their features by a vector field @xmath172 , which comprises , from an electrodynamic point of view , the dynamical situation : @xmath173 then electromagnetic fields are , similar to the formulations in classical electrodynamics , given by : @xmath174 the constants @xmath113 have to be added for reasons of consistency , since the traditional coupling of electromagnetic fields and dynamical variables can not be taken for granted . according to the definitions of @xmath175 and @xmath176 in the previous section , the @xmath113 will only be equal to unity , if @xmath177 $ ] is equal to @xmath178 $ ] . based on these definitions and the wave equation , maxwell s equations were accounted . if , on the other hand , @xmath177 $ ] is , like in classical electrodynamics , equal to energy density ( see eq . [ ed020 ] ) , then it equals @xmath179 , which has the dimensions of energy density but the direction of momentum . for reasons of consistency of the framework with classical electrodynamics the constants have to be set to unity and the vector field shall be equal to momentum per unit charge . @xmath180 then must be dimensionally equal to mechanical momentum . @xmath181 = [ p ] = \left[\frac{kg m}{s}\right]\ ] ] energy of the electromagnetic complement in terms of electromagnetic fields has to be modified accordingly . considering , that the time derivative of @xmath182 equals : @xmath183 and equally , that the energy of a particle is proportional to its frequency : @xmath184 energy density of electromagnetic complements , based on this definition of @xmath182 can be written : @xmath185 the correct relation between frequency and energy is then accounted for , while the constant @xmath180 may be estimated from hydrogen ground states . using the results of material wave theory for electron motion within a hydrogen atom @xcite and considering , that the amplitude of electromagnetic as well as kinetic energy must be double its kinetic value , @xmath180 is given by : @xmath186 \\ \hbar & = & \beta_{f } \sqrt{\frac{m_{e } } { 2 e^2}}\end{aligned}\ ] ] the unit ampere in this case is no longer free for definition , but must be expressed in terms of dynamical units m , kg , s. @xmath187 = \left[\frac{kg^{1/2}}{m s}\right]\ ] ] especially interesting is the meaning of coupling constants in this context . @xmath188 = \left[\frac{m^2}{s^2 } m^3\right ] = \left[u^2\ , v\right ] = \left[m u^2\ , \frac{v}{m}\right ] \nonumber\\ \left[\mu\right ] = \left[v\right ] \qquad \left[\alpha_{f}\right ] = \left[{\bf \epsilon}^{- 1}\right ] = \left[m u^2\ , \frac{v}{m}\right]\label{ep017}\end{aligned}\ ] ] where @xmath189 is sommerfeld s fine structure constant . the units signify , that the coupling constants actually describe the capacity of electron matter to store dynamical energy per unit mass : they give rise to the question , treated in the following sections , how quantum theory allows for energy or charge quantization . the field corresponding to electromagnetic fields @xmath190 and @xmath191 and in longitudinal direction will be defined by : @xmath192 it is not equal to intrinsic momentum , since it is harmonic in longitudinal direction . for the plane wave @xmath193 , the fields will be : @xmath194 on a more general basis we proceed from the lagrangian of electron photon interaction in the presence of an external field @xmath195 . the procedure is pretty standard : defining the lagrange density of a particle in motion , including external potentials and a presumed photon , a variation with fixed endpoints allows to calculate photon energy density in terms of the external potential and kinetic potential of the particle , as well as accelerations of the particle due to the potential applied . so far the result is the traditional one . but using a legendre transform for small velocity variations and calculating the hamiltonian density will show , that the interaction hamiltonian is equal to photon energy , which will allow , on a very general basis , to refer all electrostatic interactions to an exchange of photons . we define the lagrangian by the total energy of the moving electron , and include total energy of a presumed photon : @xmath196 @xmath197 using the hamiltonian principle for infinitesimal variations of the system with fixed endpoints : @xmath198 taking into account that @xmath199_{t_{1}}^{t_{2 } } } _ { = \,0 } - \int\limits_{t_{1}}^{t_{2 } } dt \frac{d}{d t } \left ( \frac{\partial \dot{x}_{i}^2}{\partial \dot{x}_{i } } \right ) \delta x_{i}\ ] ] the integral may be rewritten to : @xmath200 therefore the following statement is valid : @xmath201 the dynamic relation in the absence of a photon will be : @xmath202 the legendre transform of the electron photon system in an external field yields : @xmath203 for small variations of velocity the hamiltonian of the system is given by : @xmath204 the result seems paradoxical in view of kinetic energy of the moving electron , which does not enter into the hamiltonian . assuming , that an inertial particle is accelerated in an external field , its energy density after interaction with this field would only be altered according to its alteration of location . the contradiction with the energy principle is only superficial , though . since the particle will have been accelerated , its energy density _ must _ be changed . if this change does not affect its hamiltonian , the only possible conclusion is , that photon energy has equally been changed , and that the energy acquired by acceleration has simultaneously been emitted by photon emission . the initial system was therefore over determined , and the simultaneous existence of an external field _ and _ interaction photons is no physical solution to the interaction problem . the process of electron acceleration then has to be interpreted as a process of simultaneous photon emission : the acquired kinetic energy is balanced by photon radiation . a different way to describe the same result , would be saying that electrostatic interactions are accomplished by an exchange of photons : the potential of electrostatic fields then is not so much a function of location than a history of interactions . this can be shown by calculating the hamiltonian of electron photon interaction : @xmath205 since material wave theory is based on the intrinsic features of particles , it has to account for a problem , usually treated rather sloppily in quantum theory . we refer to the problem of particle scattering . if einstein s notion of energy as an intrinsic quality of motion is taken seriously - and the present paper seeks to accomplish the same for momentum : i.e. to establish momentum @xmath64 as an intrinsic quality of particle motion - then a thorough analysis of scattering processes raises problems neither quantum theory nor electrodynamics are qualified to solve . we assume two electrons of integral momenta @xmath206 and @xmath207 to interact ; during this process their path will be changed due to electrostatic interactions @xmath208 as well as their velocity . the decisive question , in this context , is the change of internal structures given by : @xmath209 where @xmath210 as well as @xmath211 are functions of time . electric and magnetic intensity @xmath212 and @xmath213 , and their relations were shown to be consequences of the internal structure of any single electron , electrodynamics in its traditional form therefore provides no answer to the question , by what means the alteration of internal structure is brought about . the argumentation of quantum theory in this respect is , strictly speaking , illegitimate , because it is not developed beyond the simple statement , that a specific momentum causes a specific wave packet . by what means this change is accomplished , remains unanswered . from the viewpoint of field theory , the medium of transmission can only be a field , whatever its qualities may be , and this field , by way of physical interactions , must cause the internal restructuring of the second particle from the definition of electric fields eq . [ ed004 ] and the equation of continuity for electron momentum and density , it may be concluded that ( in a one dimensional simplification ) : @xmath214 the time derivative , including the equation of continuity , then yields : @xmath215 identifying the external potential as the potential of a photon : @xmath216 the source term of the electron wave equation for any of the two particles can be determined : @xmath217 as quantization arises from the interaction of single points within the considered region ( see the following sections ) , conservation of electron momentum for the two particles and , consequently , for the two photons , must hold in a very general sense for an arbitrary location within the two photons . and using , again , the equation of continuity for intrinsic momenta and potentials of the photons , the sum of the two potentials must comply with : @xmath218 if the constant of integration equals zero , the total field of the two photons will vanish for every external observer : @xmath219 a specific feature of a component @xmath220 of the overall field will be , that it does not depend on the distance from its source @xmath221 : @xmath222 it is a consequence of ( [ ep031 ] ) , because if intensity would in some way depend on the distance @xmath223 , linear momentum could not remain constant for the system of two particles . if the external and static field @xmath224 is the field of an interacting particle , then emission and absorption will be balanced . the traditional hamiltonian , in this case , correctly describes dynamical variables of interaction , although only due to the - hidden - process of photon exchange . it is evident that neither emission , nor absorption of photons at this stage does show discrete energy levels : the alteration of velocity can be chosen arbitrarily small . the interesting question seems to be , how quantization fits into this picture of continuous processes . to understand this paradoxon , we consider the transfer of energy due to the infinitesimal interaction process . the differential of energy is given by : @xmath225 a denotes the cross section of interaction . setting @xmath226 equal to one period @xmath227 , and considering , that the energy transfer during this interval will be @xmath228 , we get for the transfer rate : @xmath229 since total energy density of the particle as well as the photon remains constant , the statement is generally valid . and therefore the transfer rate in interaction processes will be : @xmath230 the result was already quoted ( see section [ sec_ed ] ) . and it was equally emphasized , that the term energy quantum is , from the point of view taken in material wave theory , not quite appropriate . the total value of energy transferred depends , in this context , on the region subject to interaction processes , and equally , as will be seen further on , on the duration of the emission process . the view taken in quantum theory is therefore only a good approximation : a thorougher concept , of which material wave theory in its present form is only the outline , will have to account for _ every _ possible variation in the interaction processes . but even on this , limited , basis of understanding , planck s constant , commonly considered the fundamental value of energy quantization is the fundamental constant not of energy values , but of transfer rates in dynamic processes . constant transfer rates furthermore have the consequence , that volume and mass values of photons or electrons become irrelevant : the basic relations for energy and dispersion remain valid regardless of actual quantities . quantization then is , in short , a _ result of energy transfer _ and its characteristics . * if intrinsic features of particles and their interactions are considered , quantization has to be interpreted as a result of energy transfer processes and their specific properties . the photon interaction process leading to the acceleration of particles in an electrostatic field is accounted for by the field vector @xmath231 , of longitudinal orientation and including the effects of different field polarizations : @xmath232 elementary charge e shall be the flow through a surface of the field @xmath231 in unit time @xmath233 : @xmath234 integrating over a spherical surface with r = 1 and evaluating the time integral yields : @xmath235 the result is consistent with the source relation of electrostatic fields , since ( in si units ) : @xmath236 @xmath237 and the flow in unit time per unit volume is then : @xmath238 the calculation yields the numerically correct result . it is also dimensionally consistent , if it is considered , that it refers to unit volume and in unit time . the elementary quantum of charge is therefore not exactly a quantity of some physical variable , but , consistent with previous results , a measure for dynamical energy transfer rates per unit time . charge quantization is the consequence of _ photon interaction processes and their specific energy transfer rates_. * the elementary quantum of electron charge derives from energy transfer rates in electrostatic interactions , which are an expression of the interaction process accomplished by photons . using the result for @xmath180 in terms of e , planck s constant equally is not a fundamental constant but can be referred to dynamic interaction rates of particles and fields ( dimensions again corrected ) : @xmath239 comparing with the values for @xmath240 based on codata recommendation for a consistent system of fundamental constants @xcite , the result deviates by about 2.5 % : @xmath241 but considering , that the result was obtained by a completely different approach and , furthermore , that it is the first _ direct _ relation between the three fundamental constants @xmath242 without any additional parameter , the numerical value seems in acceptable accordance with current data . if the relation holds , then the consequences for the framework of quantum theory are significant . its essential conception , the interpretation of discrete frequency , mass and charge values as discrete quantities determining physical processes can , from this viewpoint , no longer be sustained . all processes analyzed so far , were found to be continuous in nature , and the appearance of discrete interaction quanta had to be referred to constant transfer rates applying to every single point of the underlying fields . * on the basis of electron photon interactions planck s constant in material wave theory only applies to interactions and transfer processes . a justification of discrete frequencies , mass or charge values from the viewpoint of allowed _ states _ can not be given in material wave theory . the result indicates , that the formalism of quantum theory lacks justification if it refers discrete energy levels to isolated particles . the result provides the logical reason for the deeply rooted epistemological problem of quantum theory expressed by heisenberg ( see ref . @xcite ) : if quantization is only appropriate for interactions , i.e. measurement processes , then the results of quantum theory can only hold for actual measurement processes . but since the formalism of quantum theory is based on single _ eigenstates _ , meaning states of isolated particles , this logical structure is not accounted for by the mathematical foundations of quantum theory . while , therefore , the mathematical formalism suggests a validity beyond any actual measurement , it can only be applied to specific measurements . what it amounts to , in short , is a logical inconsistency in the fundamental statements of quantum theory . the assumption of invariant charge , i.e. the conception , that charge remains invariant regardless of the interaction process or the surrounding , is not compatible with the results so far . if charge can only be determined by its photon interactions with other charge a result of the previous sections and if the term charge refers to the _ electromagnetic _ qualities of this interaction as described by the change of kinetic variables of charged particles then the _ charge _ of a particle must be referred to the dynamic qualities of interaction photons . as the dynamic qualities of a photon ( not accounting for transversal properties which will be treated in the next section ) are completely determined by ( 1 ) its frequency @xmath1 and ( 2 ) its momentum @xmath243 , the possible interactions of two particles of mass @xmath15 and charge @xmath244 e must be accomplished by a photon with linear momentum @xmath245 described by ( propagation assumed parallel to @xmath246 : @xmath247 the energy of the photon in every case equals @xmath23 , both forms comply with the wave equation for linear intrinsic momentum and yield identical transversal fields , the difference , though , is that the form @xmath248 can not be interpreted as a particle , because its momentum variable has the opposite direction of propagation . * the interpretation of electrostatic interactions as an exchange of photons , proper to material wave theory , has the consequence , that ( 1 ) the properties of single charges can only be referred to interactions with the environment , ( 2 ) photons can only be interpreted as particles if two identical charges interact , and ( 3 ) electrostatic potentials depend on the possibility of photon exchanges . while for two separate charges the third consequence is of no effect , it can not be assumed a priori , that aggregations of charge typical for nuclear environments will provide a mode of photon interactions . but if atomic nuclei do not provide a mode of photon interaction between individual nuclear particles , then electrostatic considerations do not apply to the problem of nuclear stability . so far the vector characteristics of electromagnetic fields have not been accounted for . the problem of electron photon interaction in this case requires a closer scrutiny of the energy transfer process itself . consistent with classical electrodynamics it may be postulated , that electromagnetic fields may be superimposed . in this case the vector field @xmath249 of an electron and interacting photon is given by : @xmath250 and consistent with the framework developed it may be stated , that total energy density at any location of the electron must be equal to : @xmath251 using ( [ ep014 ] ) and ( [ ep015 ] ) the energy relation yields : @xmath252 the energy is altered depending on the angle between electromagnetic field vectors of electron and interacting photon . the equation has the same structure as the energy relation in classical mechanics , it does not refer , though , to any longitudinal characteristic . even if longitudinal vectors are parallel , the difference in transversal fields must account for energy deviations . taking two identical @xmath47vectors and assuming , that every angle of deviation is equally probable , then the @xmath47vector after interaction will be in average : @xmath253 and if @xmath254 is considered as the undisturbed interaction energy , then the average deviations are given by : @xmath255 to estimate the importance of polarizations consider two electrons with arbitrary initial velocities and different orientation of intrinsic electromagnetic fields to interact . then the electromagnetic fields of the photons of interaction will show successively shifted angles and the kinetic characteristics of particle motion will be affected in a way not described by electrodynamics . it is this process , which must ultimately lead to a shift of scattering cross sections already predicted and confirmed by quantum theory . as a mathematical framework to calculate scattering processes on a dynamical basis complicated by the fact that the standard scattering approximations are bound to fail , since no potential can be defined for the angular dependency of photon interaction has not yet been developed , the exact changes compared to quantum theory can not currently be exhibited . david bohm devised a gedankenexperiment to differ between the theory of hidden variables and quantum theory @xcite , which was based on the assumption , that the direction of spin is a true but hidden variable . as bell formulated in his inequalities , the consequence of such an assumption leads to consequences , which should be , in principle , measurable @xcite . measurements of these consequences by aspect seemed to confirm , that a local theory of hidden variables is contradictory @xcite . any theory , thus the result , which assumes hidden variables of this type , is contradicted by measurements . this conclusion is not undisputed , as thompson recently showed by the chaotic ball model of this type of measurements @xcite . the usual interpretation , the author concluded , which assumes quantum theory to be superior to local hidden variable theories , contains a biased interpretation of experimental data . the theoretical framework of material waves , which can be seen as an extension of electrodynamics to intrinsic particle properties , is basically a local theory . particle spin is a local variable , and the theory is therefore a realistic and local theory of hidden variables , although , and this seems to be an important difference to bohm s concept , classical electrodynamics _ and _ quantum theory were shown to result from specific limitations . on these grounds the conflict of a measurement presumably excluding hidden variables in a local theory and a concept which explicitly goes beyond the limits of formalization of the quantum theory , on which evaluation of these measurements is based ( like the current framework ) , can only be clarified by exclusion . if these measurements are also in the current framework valid _ within _ the limits of quantum theory , then the current framework must be ruled out , since it contradicts experimental evidence of a highly consistent and successful theory . but if , on the other hand , these measurements do not yield valid results within the framework of quantum theory , then this experiment is not conclusive , because its interpretation presupposes theoretical formulations which have been violated . in this case an alternative interpretation of the measurements is possible , which is based on a different theory . although it can not currently be determined _ what _ was actually measured in aspect s experiments , it can be excluded that the measured result was obtained within the limits of quantum theory . and since bell s inequalities , which provided the theoretical background of interpretation , are based on the axioms of quantum theory , the interpretation of these results is questionable . we will return to these results in later publications . in the present framework spin is local and parallel to the polarization of the intrinsic magnetic field @xmath71 which can be shown as follows : the energy relations in classical electrodynamics yield , for a mass of magnetic moment @xmath256 in a magnetic field @xmath71 , the following energy : @xmath257 the following deduction is based on four assumptions : ( 1 ) the energy of an electron or photon is equal to the energy of the particle in quantum theory , ( 2 ) the magnetic field is the intrinsic ( photon ) or external ( electron ) magnetic field , ( 3 ) the frequency @xmath26 is also a frequency of rotation , and ( 4 ) the magnetic moment is described by the relations in quantum theory . with ( 1 ) the energy of photons ( total energy ) and electrons ( only kinetic energy ) will be : @xmath258 with ( 2 ) the magnetic fields of electrons and photons will be ( see eq . [ ep001 ] ) : @xmath259 and with ( 4 ) the magnetic moment will be ( @xmath260 denotes the gyromagnetic ratio ) : @xmath261 calculating the energy of interaction and considering , that the magnetic field in the current framework is @xmath96 times the magnetic field in classical electrodynamics ( see eq . [ ed009 ] ) , we get for the photon : @xmath262 with @xmath263 , which follows from the definition of magnetic fields ( see section [ ed_units ] ) , the relation can only hold for constant @xmath264 if : @xmath265 the energy relation of electrodynamics in this case is only consistent with the energy of photons of they possess intrinsic spin @xmath266 , a gyromagnetic ratio of 1 , and if the direction of spin polarization is equal to the direction of intrinsic magnetic fields . a consistent framework therefore requires ( 1 ) a localized spin , and ( 2 ) a spin @xmath266 of bosons . the same calculation for the properties of electron spin yields the following result : @xmath267 in principle the spin variable and the gyromagnetic ratio could be chosen like the values for photons . from the condition itself , the difference between bosons and fermions is not directly accessible . the difference can be accounted for , if the atomic model in quantum theory is considered : the theory of hydrogen fine structures by uhlenbeck and goudsmit @xcite requires two energy levels symmetric to the undisturbed one , thus a general multiplicity of 2 . the only solution , with this initial assumption , is a spin variable of @xmath268 and a gyromagnetic ratio of 2 . the direction of spin polarization is also for fermions equal to the direction of the magnetic field vector . generally the concept of spin seems questionable , because the frequency @xmath26 is not at all to be confused with rotation frequencies of a particle . consider , for example , a photon in uniform motion @xmath96 along a flat space time geometry . its intrinsic magnetic field does not determine any state of rotation , nor does the energy of a photon result from the interaction of a photon magnetic moment with a magnetic field . from the given deduction of spin properties of particles ( and the deduction seems highly appropriate , considering the effect of spin operators in the framework of quantum theory ) it seems , that the concept itself must be theoretically deficient . let us consider , on this basis , the result of any realistic measurement of photon spin in an epr like situation @xcite . the first problem , we are confronted with , is the exact definition of the spin value . from the given deduction it seems evident , that the local direction of the spin vector is either parallel or antiparallel to the intrinsic magnetic field . provided , we adopt the view of quantum theory for spin conservation of two photons , then spin up would mean , that the vector is antiparallel to @xmath71 , spin down means , it must be parallel . in this case we have two photons with an arbitrary angle @xmath269 of polarization in the state + 1 for photon one and -1 for photon two ( see fig . [ fig001 ] ) . = 1.0 the second problem originates from the periodic features of the intrinsic magnetic field . since the magnetic field vector is periodic with @xmath270 , the spin variable _ can not _ remain constant , but must oscillate from + 1 to - 1 . this oscillation requires , that we define the exact moment of measurement to obtain any valid result on the state of either particle . if the photon path remains constant in measurements , this requirement is fulfilled , but if , like in aspect s measurements @xcite , the measuring apparatus has to be rotated to account for different angles of observation , the effect on measurements seems not easy to estimate . this problem therefore leads to a problem of experimental methods , which is hidden in the usual argumentation of quantum theory . provided , the problem can be accounted for , our next problem is more substantial . since the spin variable oscillates , the measurement can only yield a definite result , if the variable is measured in the time interval @xmath271 , which requires , that the local resolution of the measuring apparatus is below @xmath272 . now as demonstrated in the deduction of the uncertainty relations , this interval is equal to the lowest limit of local resolution in the framework of quantum theory . thus we get the result , that a valid measurement of the spin variable must have a local precision which is explicitly beyond the fundamental uncertainty in quantum theory . if , therefore , the measurements yield any correlation related to physical processes , then the fundamental axioms of quantum theory must be violated . this result seems to turn the controversy about the epr paradox into a new direction . it can be established , that the mathematical framework of quantum theory , describing the state of either particle by a superposition of the two possible states @xmath244 1 is correct . but it can equally be established , that a significant correlation in spin measurements is not , as einstein suspected , a proof of hidden variables or a contradiction in terms of special relativity , but a proof of measurements below the fundamental level of uncertainty . if this sort of measurement is possible , then the uncertainty relations are violated . * if epr measurements are possible and yield a significant correlation for a pair of spin particles , then the uncertainty relations are violated . in section [ sec_ed ] it was established , that electromagnetic radiation consists of two distinct intrinsic variables : ( 1 ) a longitudinal momentum , and ( 2 ) a transversal electromagnetic field , both variables depending on the frequency of radiation . in this section we found , that the change of energy density due to the adsorption of a photon result in motion of the absorbing electron , the energy increase given by : @xmath273 if monochromatic x rays therefore interact with free electrons , i.e. electrons with a velocity small compared to @xmath274 , then the absorbing electrons should move in longitudinal direction with a velocity equal to : @xmath275 if this electron is decelerated , then it will emit electromagnetic radiation with a frequency described by : @xmath276 and , depending on the duration of the deceleration process , the radiation can possess arbitrarily low frequency . this _ primary _ emission therefore will not exhibit discrete frequency levels . but if _ secondary _ processes are considered , then the frequency change will depend on ( 1 ) the relative velocity between the source of radiation and the moving electron , and ( 2 ) the emission and absorption characteristics . assuming , that the emission and absorption processes are symmetric in terms of duration , then the frequency of secondary emission will depend on the original frequency as well as the angle of emission . the model presented is therefore a simplification , and it can not be excluded , that a precise and deterministic picture of the whole process will not yield a deviating result . but compared to the standard model in quantum theory @xcite , it is nevertheless an improvement , because it does not have to recur to mechanical concepts . if the electron moves in longitudinal direction and with a velocity equal to @xmath277 , the frequency @xmath278 of the source will be changed due to the electromagnetic doppler effect . the secondary absorption process therefore has to account for a frequency @xmath279 : @xmath280 an observer moving with the velocity @xmath277 of the electron will therefore measure changed frequencies and wavelengths of secondary emission . the wavelength @xmath281 will be : @xmath282 the second term , the shift of wavelength due to longitudinal motion of the electron relative to the source of radiation , is the _ compton wavelength _ of an electron @xcite . it is the general shift of wavelength for every electromagnetic field , and an expression of longitudinal motion . @xmath283 additionally it has to be considered , that the reference frame of observation is the same as the reference frame of the original radiation , thus a second frequency shift depends on the angle of observation @xmath269 . measured from the direction of field propagation this additional shift will be : @xmath284 and the overall shift of wavelength measured is therefore : @xmath285 the result is equivalent to compton s measurements . compton scattering , in our view , is therefore a result of ( 1 ) secondary absorption and emission processes , and of ( 2 ) successive doppler shifts due to longitudinal motion of the emitting electrons . that classical electrodynamics does not provide a theoretical model for compton scattering , can be understood as a consequence of neglecting longitudinal and kinetic features of photons . since electromagnetic radiation in electrodynamics is interpreted from transversal properties alone ( which are periodic ) , a uniform longitudinal motion due to absorption processes is not within its theoretical limits . the possibility to account for quantum effects with the modified and still continuous model of electrons is strongly supporting , in our view , the validity of the unification of electrodynamics and quantum theory by formalizing the intrinsic features of particles . as it turns out , retrospectively , the theory of material waves removes the two fundamental problems , which made a physical interpretation ( as opposed to the probability interpretation @xcite ) of intrinsic wave properties impossible . ( 1 ) the contradiction between phase velocity and mechanical velocity of a particle wave is removed by the discovery of intrinsic potentials . ( 2 ) the problem of intrinsic coulomb interactions is removed , because electrostatic interactions can be referred to an exchange of photons . a particle in uniform motion is therefore stable . to exhibit the full range of logical and physical implications of the theory developed , we repeat the essential results of this essay on material waves , displayed throughout the preceding sections of this paper : 1 . the interpretation of particles as physical waves allows the conclusion , that quantum theory can not be , in a logical sense , a complete theoretical description of micro physical systems . 2 . _ because _ quantum theory neglects intrinsic characteristics of particle motion , the schrdinger equation is generally not precise by a value described by heisenberg s uncertainty relations . electrodynamics and wave mechanics are theoretical concepts based on the same physical phenomena , the intrinsic features of moving particles . 4 . the wave equations of intrinsic particle structures are lorentz invariant only for an integral evaluation of energy . the total intrinsic potentials increase for every transformation into a moving frame of reference . if intrinsic features of particles and their interactions are considered , quantization has to be interpreted as a result of energy transfer processes and their specific properties . the elementary quantum of electron charge derives from energy transfer rates in electrostatic interactions , which are an expression of the interaction process accomplished by photons . 7 . on the basis of electron photon interactions planck s constant in material wave theory only applies to interactions and transfer processes . a justification of discrete frequencies , mass or charge values from the viewpoint of allowed _ states _ can not be given . the result indicates , that the formalism of quantum theory lacks justification if it refers discrete energy levels to isolated particles . the interpretation of electrostatic interactions as an exchange of photons , proper to material wave theory , has the consequence , that ( 1 ) the properties of single charges can only be referred to interactions with the environment , ( 2 ) photons can only be interpreted as particles if two identical charges interact , and ( 3 ) electrostatic potentials depend on the possibility of photon exchanges . 9 . if epr measurements are possible and yield a significant correlation for a pair of spin particles , then the uncertainty relations are violated . = 1.0 the theoretical framework thus provides definite answers to the epr paradox by rejecting the copenhagen interpretation of quantum theory @xcite , because ( 1 ) quantum theory is not a complete account of micro physical systems , since the wave function possesses physical significance , and ( 2 ) the theory is logically inconsistent , since it is applicable only to interaction processes , while its mathematical formulation suggests the results to derive from properties of individual particles . concerning renninger s objections on the quality of photons derived from thought experiments @xcite , it provides a model of photons where ( 1 ) the energy is homogeneously distributed over the whole region of electromagnetic fields , ( 2 ) the kinetic features of photons derive from a longitudinal component not accounted for in electrodynamics , and ( 3 ) the energy transfer in interactions applies for every single point of a given region , which yields the quantum effect of interaction . it is difficult to estimate either the impact or the subsequent changes due to a concept , which completely deviates from the mainstream of current physics . the aim of establishing and , to a limited extent , interpreting a physical reality by an alternative framework was to provide a means , by which experimental results can be seen from at least two different points of view . the fundamental idea , it seems , is not very different from the conceptions of einstein , schrdinger , bohm , or de broglie ( see the introduction ) , but the disregard of well established interpretations is considerably higher . this was not , initially , intended . it only turned out , that a theoretical concept limited to only minor changes will neither provide a scientifically satisfying solution since it still is based on some _ have to believe _ nor does it really offer any new insight into micro physical processes . it was not intended to recover all the results of current quantum physics , not least , because it would surely require many lifetimes to succeed . but it was intended , and , so i think , accomplished rather successfully , to extend the theoretical framework of micro physics beyond the uncertainty relations and the _ we can not know _ of quantum theory . whether the result presented it the best possible way out of the dilemma arising from a physics of quantities mechanics confronted with a world of intensities electrodynamics can not be currently estimated . to estimate the epistemological changes implied by the theoretical framework , fig . [ fig002 ] displays the logical structure of micro physics in material wave theory . to estimate , whether the new theory contradicts any experimental evidence within its range of formalization , it has to be considered , that it provides a meta theory combining the different frameworks of electrodynamics , quantum theory , and quantum electrodynamics in a single consistent framework . from the viewpoint of electrodynamics , it will reproduce any result the current framework provides , because the basic relations are not changed , merely interpreted in terms of electron and photon propagation . therefore an experiment , consistent with electrodynamics , will also be consistent with the new theory . in quantum theory , the only statements additional to the original framework are beyond the level of experimental validity , defined by the uncertainty relations . since quantum theory can not , in principle , contain results beyond that level , every experiment within the framework of quantum theory must necessarily be reproduced . that applies to the results of wave mechanics , which yields the eigenvalues of physical processes , as well as to the results of operator calculations , if von neumann s proof of the equivalence of schrdinger s equation and the commutator relations is valid . the only result , which could in this respect disprove the theory , is the spreading of a wave packet : but as this experimental result was the reason , that a _ physical _ interpretation of wave functions had to be discarded , the theory also in this respect seems to stand on firm ground . in quantum field theory , the same rule applies : the new theory only states , what is , in quantum theory , no part of a measurement result , therefore a contradiction can not exist . if , on the other hand , results in quantum electrodynamics exist , which are not yet verified by the new theory , then it is rather a problem of further development , but not of a contradiction with existing measurements . experimentally , the new theory can not be disproved by existing measurements , because its statements at once verify the existing formulations and extend the framework of theoretical calculations . the same does not hold , though , for the verification of existing theoretical schemes of calculation by the new framework . since the theory extends far beyond the level of current concepts , established theoretical results may well be subject to revision . the theory of material waves requires , that every valid solution of a microphysical problem can be referred to physical properties of particle waves . in cases , where the established theory provides a solution yielding , for example , correct eigenvalues but not wave functions which can be identified as _ physical _ waves , the result is , from this viewpoint , questionable . essentially , this is not limiting nor diminishing the validity of current results , since it extends the framework of micro physics only to regions , which so far have remained unconsidered . from the five tasks defined in the introduction as our personal criterium for a promising new approach , three have been accomplished by this publication . but as quantum theory , historically , was invented due to the puzzling results in atomic physics , the whole concept as such would be useless without a theory of atoms compatible with the framework developed . this theory has partly been published already ( see @xcite ) , partly it is still in the making although the essential results have been derived . it will be the aim of the next paper published in the internet , to give a thorough and up to date account of a model of hydrogen atoms in terms of material waves . schweber s. _ qed and the men who made it _ , princeton ( 1994 ) bohm d. _ causality and chance in modern physics _ , princeton ( 1957 ) de broglie l. in his foreword to @xcite dirac p.a.m . j. phys . _ * 5 * , 65 ( 1984 ) rseberg u. in gtschl j. ( ed . ) _ erwin schrdinger s world view _ , dordrecht ( 1992 ) p. 135 feynman r. _ qed _ , princeton ( 1988 ) p.9 schrdinger , e. _ ann . physik _ * 79 * , 361 ; 489 ( 1926 ) heisenberg w. `` remarks on the origin of uncertainty '' in price w. c. ( ed . ) _ the uncertainty principle and the foundations of quantum mechanics _ , wiley & sons , new york ( 1977 ) bohr n. `` discussion with einstein on epistemological problems in atomic physics '' , in _ a. einstein : philosopher scientist _ , schilpp p.a . ( ed . ) , harper and row , new york ( 1959 ) de broglie l. _ ann . phys . _ * 3 * , 22 ( 1925 ) born , m. _ z. physik _ * 37 * , 863 ( 1926 ) einstein a. , rosen n. , and podolsky b. _ phys . _ * 47 * , 180 ( 1935 ) kwiat p. , weinfurter h. , herzog t. , zeilinger a. , and kasevich m.a . lett . _ * 74 * , 4763 ( 1995 ) renninger m. _ zeitschr . f. phys . _ * 158 * , 417 ( 1960 ) bohm , d. _ phys . rev . _ * 85 * , 166 ; 180 ( 1952 ) bell j.s . _ physics _ * 1 * , 195 ( 1964 ) ; see also his remarks on hidden variable theories in _ rev . mod . phys . _ * 38 * , 447 ( 1966 ) davisson c. and germer l.h . rev . _ * 30 * , 705 ( 1927 ) hofer w. a. internal energy of free particles : de broglie waves in electrodynamics . tech . _ * 18 * , 157 ( 1995 ) hofer w. a. an inertial model of atoms : atomic spectra as a resonance problem . _ specs . _ , in print born m. , heisenberg w. , and jordan p. _ z. physik _ * 35 * , 557 ( 1926 ) planck m. _ ann . physik _ * 4 * , 553 ( 1901 ) heisenberg w. _ z. physik _ * 43 * , 172 ( 1927 ) cohen tannoudji c. , diu b. , and laloe f. _ quantum mechanics _ , wiley & sons , new york ( 1977 ) einstein a. _ ann . * 17 * , 891 ( 1905 ) sommerfeld a. _ elektrodynamik _ , akademische verlagsgesellschaft , reprint leipzig ( 1964 ) michelson , and morley . j. sci . _ * 34 * , 333 ( 1887 ) ; _ phil . mag . _ * 24 * , 449 ( 1887 ) jackson j. _ classical electrodynamics _ , wiley & sons , new york ( 1984 ) einstein a. _ ann . physik _ * 17 * , 132 ( 1905 ) jordan p. _ anschauliche quantentheorie _ , berlin ( 1936 ) p. 115 heisenberg w. _ die physikalischen grundlagen der quantentheorie _ , leipzig ( 1941 ) p. 7 and 107 renninger m. _ z. physik _ * 136 * , 251 ( 1953 ) a. einstein . * 20 * , 627 ( 1906 ) uipap document uip 20 . _ physica a _ * 93 * , 1 ( 1978 ) bohm d. _ quantum mechanics _ , prentice hall , eaglewood cliffs ( 1951 ) aspect a. _ trois tests experimenteaux des inegalites de bell par mesure de correlation de polarisation de photons _ , ph . d. thesis , universite de paris sud , centre dorsay ( 1983 ) thompson c. h. _ found . lett . _ * 9 * , 357 ( 1996 ) uhlenbeck g. e. and goudsmit s. a. _ naturwiss . _ * 13 * , 953 ( 1925 ) ; and _ nature _ * 117 * , 264 ( 1925 ) debye , p. _ phys . zeitschr . _ * 24 * , 161 ( 1923 ) compton , a. h. _ phys . * 22 * , 409 ( 1923 ) Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the motivation for the study reported in this paper comes from the celebrated and long - standing problem , originally posed by erds @xcite in 1946 , of obtaining a sharp lower bound for the number of distinct distances guaranteed to exist in any set @xmath0 of @xmath1 points in the plane . erds has shown that a section of the integer lattice determines only @xmath10 distinct distances , and conjectured this to be a lower bound for any planar point set . in spite of steady progress on this problem , reviewed next , erds s conjecture is still open . l. moser @xcite , chung @xcite , and chung @xcite proved that the number of distinct distances determined by @xmath1 points in the plane is @xmath11 , @xmath12 , and @xmath13 , respectively . szkely @xcite managed to get rid of the polylogarithmic factor , while solymosi and tth @xcite improved this bound to @xmath14 . this was a real breakthrough . their analysis was subsequently refined by tardos @xcite and then by katz and tardos @xcite , who obtained the current record of @xmath15 , for any @xmath16 , which is @xmath17 . in this paper we transform the problem of distinct distances in the plane to an incidence problem between points and a certain kind of curves ( helices or parabolas ) in three dimensions . as we show , sharp upper bounds on the number of such incidences translate back to sharp lower bounds on the number of distinct distances . incidence problems in three dimensions between points and curves have been studied in several recent works @xcite , and a major push in this direction has been made last year , with the breakthrough result of guth and katz @xcite , who have introduced methods from algebraic geometry for studying problems of this kind . this has been picked up by the authors @xcite , where worst - case tight bounds on the number of incidences between points and lines in three dimensions ( under certain restrictions ) have been obtained . the present paper serves two purposes . first , it studies in detail the connection between the distinct distances problem and the corresponding 3-dimensional incidence problem . as it turns out , there is a lot of interesting geometric structure behind this reduction , and the paper develops it in detail . we offer several conjectures on the number of incidences , and show how , if true , they yield the almost tight worst - case lower bound @xmath9 on the number of distinct distances . unfortunately , so far we have not succeeded in proving these conjectures . nevertheless , we have made considerable progress on the incidence problem itself , which is the second purpose of the study in this paper . we show how to adapt the algebraic machinery of @xcite to derive sharp bounds for the incidence problem . @xcite to derive sharp bounds for the incidence problem . these bounds are very similar to , and in fact even better than the bounds obtained in @xcite for point - line incidences , where they have been shown to be worst - case tight . however , they are not ( yet ) good enough to yield significant lower bounds for distinct distances . we believe that there is additional geometric structure in the particular problem studied here , which should enable one to further improve the bounds , but so far this remains elusive . the paper is organized as follows . we first describe the reduction from the planar distinct distances problem to the 3-dimensional incidence problem mentioned above . in doing so , we note and explore several additional geometric connections between the two problems ( as manifested , e.g. , in the analysis of _ special surfaces _ given below ) . we then present the tools from algebraic geometry that are needed to tackle the incidence problem ; they are variants of the tools used in @xcite , adapted to the specific curves that we need to handle . we then go on to bound the number of incidences . we first bound the number of rotations in terms of the number of parabolas , and then bound the number of incidences themselves . the latter task is achieved in two steps . we first use a `` purely algebraic '' analysis , akin to those in @xcite , to obtain a weaker bound , which we then refine in the second step , using more traditional space decomposition techniques . the final bound is still not as good as we would like it to be , but it shows that the case studied in this paper `` behaves better '' than its counterpart involving lines . [ [ distinct - distances - and - incidences - with - helices . ] ] distinct distances and incidences with helices . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we offer the following novel approach to the problem of distinct distances . [ [ h1-notation . ] ] ( h1 ) notation . + + + + + + + + + + + + + + let @xmath0 be a set of @xmath1 points in the plane with @xmath18 distinct distances . let @xmath19 denote the set of all quadruples @xmath20 , such that the pairs @xmath21 and @xmath22 are distinct ( although the points themselves need not be ) and @xmath23 . let @xmath24 denote the @xmath18 distinct distances in @xmath0 , and let @xmath25 . we have @xmath26 ^ 2 = \frac{\left[s(s-1)-x\right]^2}{x } .\ ] ] [ [ h2-rotations . ] ] ( h2 ) rotations . + + + + + + + + + + + + + + + we associate each @xmath27 with a ( unique ) _ rotation _ ( or , rather , a rigid , orientation - preserving transformation of the plane ) @xmath28 , which maps @xmath29 to @xmath30 and @xmath31 to @xmath32 . a rotation @xmath28 , in complex notation , can be written as the transformation @xmath33 , where @xmath34 and @xmath35 . putting @xmath36 , @xmath37 , we can represent @xmath28 by the point @xmath38 . in the planar context , @xmath39 is the counterclockwise angle of the rotation , and the center of rotation is @xmath40 , which is defined for @xmath41 ; for @xmath42 , @xmath28 is a pure translation . the _ multiplicity _ @xmath43 of a rotation @xmath28 ( with respect to @xmath0 ) is defined as @xmath44 the number of pairs @xmath45 such that @xmath46 . clearly , one always has @xmath47 , and we will mostly consider only rotations satisfying @xmath48 . as a matter of fact , the bulk of the paper will only consider rotations with multiplicity at least @xmath49 . rotations with multiplicity @xmath50 are harder to analyze . if @xmath51 then @xmath0 contains two congruent and equally oriented copies @xmath52 of some @xmath5-element set , such that @xmath53 . thus , studying multiplicities of rotations is closely related to analyzing repeated ( congruent and equally oriented ) patterns in a planar point set ; see @xcite for a review of many problems of this kind . [ [ anti - rotations . ] ] anti - rotations . + + + + + + + + + + + + + + + in this paper we will also consider _ anti - rotations _ , which are rigid , orientation - reversing transformations of the plane . any anti - rotation can be represented as a rotation , followed by a reflection about some fixed line , e.g. , the @xmath18-axis ( so , in complex notation , this can be written as @xmath54 ) . anti - rotations will be useful in certain steps of the analysis . [ [ h3-bounding - k . ] ] ( h3 ) bounding @xmath55 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + if @xmath51 then @xmath28 contributes @xmath56 quadruples to @xmath19 . let @xmath57 ( resp . , @xmath58 ) denote the number of rotations with multiplicity exactly @xmath5 ( resp . , at least @xmath5 ) , for @xmath8 . then @xmath59 [ [ h4-the - main - conjecture . ] ] ( h4 ) the main conjecture . + + + + + + + + + + + + + + + + + + + + + + + + + [ conj1 ] for any @xmath60 , we have @xmath61 suppose that the conjecture were true . then we would have @xmath62 ^ 2}{x } \le |k| = o(s^3 ) \cdot \left [ 1 + \sum_{k\ge 3 } \frac{1}{k } \right ] = o(s^3\log s ) , \ ] ] which would have implied that @xmath63 . this would have almost settled the problem of obtaining a tight bound for the minimum number of distinct distances guaranteed to exist in any set of @xmath1 points in the plane , since , as mentioned above , the upper bound for this quantity is @xmath10 @xcite . we note that conjecture [ conj1 ] is rather deep ; even the simple instance @xmath64 , asserting that there are only @xmath3 rotations which map ( at least ) two points of @xmath0 to two other points of @xmath0 ( at the same distance apart ) , seems quite difficult . in this paper we establish a variety of upper bounds on the number of rotations and on the sum of their multiplicities . in particular , these results provide a partial positive answer , showing that @xmath65 ; that is , the number of rotations which map a ( degenerate or non - degenerate ) triangle determined by @xmath0 to another congruent ( and equally oriented ) such triangle , is @xmath3 . bounding @xmath66 by @xmath3 is still an open problem . see section [ sec : impr ] for a simple proof of the weaker bound @xmath67 . [ [ lower - bound . ] ] lower bound . + + + + + + + + + + + + we next give a construction ( suggested by haim kaplan ) which shows : [ lem : lower ] there exist sets @xmath0 in the plane of arbitrarily large cardinality , which determine @xmath68 distinct rotations , each mapping a triple of points of @xmath0 to another triple of points of @xmath0 . * proof : * consider the set @xmath69 , where @xmath70 see figure [ lower ] . lower.pstex_t for each triple @xmath71 such that @xmath72 also belongs to @xmath73 , construct the rotation @xmath74 which maps @xmath75 to @xmath76 and @xmath77 to @xmath78 . since the distance between the two source points is equal to the distance between their images , @xmath74 is well ( and uniquely ) defined . moreover , @xmath74 maps the midpoint @xmath79 to the midpoint @xmath80 . we claim that the rotations @xmath74 are all distinct . indeed , suppose that two such rotations , @xmath74 and @xmath81 , for distinct triples @xmath82 , @xmath83 , coincide ; call the common rotation @xmath28 . we can represent @xmath28 as the rigid transformation which first translates the plane horizontally by distance @xmath84 , so that @xmath75 is mapped to @xmath76 , and then rotates it around @xmath76 by an appropriate angle @xmath85 , so that @xmath86 is mapped to @xmath78 . suppose first that @xmath87 . since @xmath88 , it maps @xmath89 to @xmath90 and then rotates this point by angle @xmath39 around @xmath76 , mapping it to a point outside the @xmath18-axis , contradicting the fact that @xmath81 maps @xmath89 to @xmath91 . if @xmath92 then we also must have @xmath93 , so @xmath94 . but then it is impossible to turn , around @xmath76 , the shifted point @xmath86 to @xmath78 and the shifted point @xmath95 to @xmath96 , by the same angle , a contradiction which shows that the two rotations are distinct . since there are @xmath97 triples @xmath82 with the above properties , the claim follows . @xmath98 * remarks . * * ( 1 ) * a `` weakness '' of this construction is that all the rotations @xmath74 map a _ collinear _ triple of points of @xmath0 to another collinear triple . ( in the terminology to follow , these will be called _ flat _ rotations . ) we do not know whether the number of rotations which map a _ non - collinear _ triple of points of @xmath0 to another non - collinear triple can be @xmath99 . we tend to conjecture that this is indeed the case . * ( 2 ) * we do not know whether conjecture [ conj1 ] is worst - case tight ( if true ) . that is , we do not know whether there exist sets @xmath0 , with @xmath100 arbitrarily large , so that there are @xmath101 distinct rotations , each mapping at least @xmath5 points of @xmath0 to @xmath5 other points of @xmath0 . [ [ h5-helices . ] ] ( h5 ) helices . + + + + + + + + + + + + + to estimate @xmath58 , we reduce the problem of analyzing rotations and their interaction with @xmath0 to an incidence problem in three dimensions , as follows . with each pair @xmath45 , we associate the curve @xmath102 , in a 3-dimensional space parametrized by @xmath103 , which is the locus of all rotations which map @xmath29 to @xmath31 . that is , the equation of @xmath102 is given by @xmath104 putting @xmath105 , @xmath106 , this becomes @xmath107 this is a _ helix _ in @xmath2 , having four degrees of freeedom , which we parametrize by @xmath108 . it extends from the plane @xmath42 to the plane @xmath109 ; its two endpoints lie vertically above each other , and it completes exactly one revolution between them . [ [ h6-helices - rotations - and - incidences . ] ] ( h6 ) helices , rotations , and incidences . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath110 be a set of rotations , represented by points in @xmath2 , as above , and let @xmath111 denote the set of all @xmath112 helices @xmath102 , for @xmath45 ( note that @xmath113 is permitted ) . let @xmath114 denote the number of incidences between @xmath110 and @xmath111 . then we have @xmath115 rotations @xmath28 with @xmath116 are not interesting , because each of them only contributes @xmath117 to the count @xmath114 , and we will mostly ignore them . for the same reason , rotations with @xmath118 are also not interesting for estimating @xmath114 , but they need to be included in the analysis of @xmath119 . unfortunately , as already noted , we do not yet have a good upper bound ( i.e. , cubic in @xmath1 ) on the number of such rotations . [ [ h7-incidences - and - the - second - conjecture . ] ] ( h7 ) incidences and the second conjecture . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + [ conj2 ] for any @xmath110 and @xmath111 as above , we have @xmath120 suppose that conjecture [ conj2 ] were true . let @xmath121 denote the set of all rotations with multiplicity at least @xmath5 ( with respect to @xmath0 ) . we then have @xmath122 from which we obtain @xmath123 thus establishing conjecture [ conj1 ] , and therefore also the lower bound for @xmath18 ( the number of distinct distances ) derived above from this conjecture . * remark . * conjecture [ conj2 ] can also be formulated for an _ arbitrary _ subset @xmath111 of all possible helices . note that two helices @xmath102 and @xmath124 intersect in at most one point this is the unique rotation which maps @xmath29 to @xmath31 and @xmath125 to @xmath126 ( if it exists at all , namely if @xmath127 ) . hence , combining this fact with a standard cutting - based decomposition technique , similar to what has been noted in @xcite , say , yields the weaker bound @xmath128 which , alas , only yields the much weaker bound @xmath129 , which is completely useless for deriving any lower bound on @xmath18 . ( we will use this bound , though , in section [ sec : conc ] . ) [ [ h8-from - helices - to - parabolas . ] ] ( h8 ) from helices to parabolas . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the helices @xmath102 are non - algebraic curves , because of the use of the angle @xmath39 as a parameter . this can be easily remedied , in the following standard manner . assume that @xmath39 ranges from @xmath130 to @xmath131 , and substitute , in the equations ( [ eq : helix ] ) , @xmath132 , @xmath133 , and @xmath134 , to obtain @xmath135 which are the equations of a _ planar parabola _ in the @xmath136-space . ( the parabola degenerates to a line if @xmath137 , a situation that we will rule out by choosing an appropriate generic coordinate frame in the original @xmath138-plane . ) we denote the parabola corresponding to the helix @xmath102 as @xmath139 , and refer to it as an _ @xmath140-parabola_. [ [ h9-joint - and - flat - rotations . ] ] ( h9 ) joint and flat rotations . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + a rotation @xmath141 is called a _ joint _ of @xmath111 if @xmath28 is incident to at least three helices of @xmath111 whose tangent lines at @xmath28 are non - coplanar . otherwise , still assuming that @xmath28 is incident to at least three helices of @xmath111 , @xmath28 is called _ let @xmath142 be a rotation , incident to three distinct helices @xmath102 , @xmath124 , @xmath143 . from their equations , as given in ( [ eq : helix ] ) , the directions of the tangents to these helices at @xmath28 are @xmath144 put @xmath145 and @xmath146 . then the three tangents are coplanar if and only if @xmath147 simplifying the determinant , and recalling that @xmath148 , the condition is equivalent to @xmath149 in other words , the three helices @xmath102 , @xmath124 , @xmath143 form a joint at @xmath28 if and only if the three points @xmath150 ( and thus also @xmath151 ) are non - collinear . that is , we have shown : [ jointri ] a rotation @xmath28 is a joint of @xmath111 if and only if @xmath28 maps a non - degenerate triangle determined by @xmath0 to another ( congruent and equally oriented ) non - degenerate triangle determined by @xmath0 . a rotation @xmath28 is a flat rotation if and only if @xmath28 maps at least three collinear points of @xmath0 to another collinear triple of points of @xmath0 , but does not map any point of @xmath0 outside the line containing the triple to another point of @xmath0 . * remarks : * * ( 1 ) * note that if @xmath28 is a flat rotation , it maps the entire line containing the three source points to the line containing their images . specifically ( see also below ) , we can respectively parametrize points on these lines as @xmath152 , @xmath153 , for @xmath154 , such that @xmath28 maps @xmath152 to @xmath153 for every @xmath155 . * ( 2 ) * for flat rotations , we also need to ensure , for technical reasons , that the three ( or more ) helices incident to a flat rotation @xmath28 are such that their tangents at @xmath28 are all distinct . this fortunately is always the case . indeed , the preceding analysis is easily seen to imply that if @xmath102 and @xmath124 meet at @xmath28 then their tangents at @xmath28 coincide if and only if @xmath156 . but then @xmath102 and @xmath157 can not have a common point ( rotation ) unless @xmath158 too , i.e. , they are the same helix ; otherwise the common rotation would have to map @xmath29 to the two distinct points @xmath31 and @xmath126 , an impossibility . [ [ h10-special - surfaces . ] ] ( h10 ) special surfaces . + + + + + + + + + + + + + + + + + + + + + + + in preparation for the forthcoming algebraic analysis , we need the following property of our helices . let @xmath28 be a flat rotation , with multiplicity @xmath4 , and let @xmath159 and @xmath160 be the corresponding lines in the plane , such that there exist @xmath5 points @xmath161 and @xmath5 points @xmath162 , such that @xmath28 maps @xmath163 to @xmath164 for each @xmath165 ( and in particular maps @xmath159 to @xmath160 ) . by definition , @xmath28 is incident to the @xmath5 helices @xmath166 , for @xmath167 . let @xmath168 and @xmath169 denote unit vectors in the direction of @xmath159 and @xmath160 , respectively . clearly , there exist two reference points @xmath170 and @xmath171 , such that for each @xmath165 there is a real number @xmath172 such that @xmath173 and @xmath174 . as a matter of fact , for each real @xmath155 , @xmath28 maps @xmath175 to @xmath176 , so it is incident to @xmath177 . note that @xmath29 and @xmath31 are not uniquely defined : we can take @xmath29 to be any point on @xmath159 , and shift @xmath31 accordingly along @xmath160 . let @xmath178 denote the set of these helices . since a pair of helices can meet in at most one point , all the helices in @xmath178 pass through @xmath28 but are otherwise pairwise disjoint . using the re - parametrization @xmath179 , we denote by @xmath180 the surface which is the union of all the @xmath140-parabolas that are the images of the helices in @xmath178 . we refer to such a surface @xmath181 as a _ special surface_. an important comment is that most of the ongoing analysis also applies when only two helices are incident to @xmath28 ; they suffice to determine the four parameters @xmath182 that define the surface @xmath181 . we also remark that , although we started the definition of @xmath183 with a flat rotation @xmath28 , the definition only depends on the parameters @xmath184 , and @xmath169 ( and even there we have , as just noted , one degree of freedom in choosing @xmath29 and @xmath31 ) . if @xmath28 is not flat it may determine many special surfaces , one for each line that contains two or more points of @xmath0 which @xmath28 maps to other ( also collinear ) points of @xmath0 . also , as we will shortly see , the same surface can be obtained from a different set ( in fact , many such sets ) of parameters @xmath185 , and @xmath186 ( or , alternatively , from different flat rotations @xmath187 ) . an `` intrinsic '' definition of special surfaces will be given shortly . the surface @xmath181 is a cubic algebraic surface , whose equation can be worked out as follows . the equation of the parabola @xmath188 corresponding to @xmath177 is @xmath189 we can view this as a parametrization of @xmath181 using @xmath155 and @xmath190 as parameters . we can simplify these equations as @xmath191 where @xmath192 are quadratic polynomials in @xmath190 . eliminating @xmath155 from these equations gives us the first version of the equation of @xmath181 , which is @xmath193 this is a quartic equation , although it is only linear in @xmath194 and @xmath195 . note also that the cross - section of @xmath181 by any plane @xmath196 is a line , so @xmath181 is a ruled surface . we next reduce ( [ sigma1 ] ) to a cubic equation , as follows . let @xmath197 denote the coordinates of @xmath28 in the @xmath198-frame . we note that @xmath199 . this can be worked out explicitly , or concluded by noting that @xmath197 is a common point of all our parabolas , so @xmath197 can not determine @xmath155 , meaning that the coefficients @xmath200 and @xmath201 in ( [ parsig ] ) must both be zero . hence , each of the three polynomials @xmath202 , @xmath203 , and @xmath204 , appearing in the left - hand side of ( [ sigma1 ] ) , vanishes at @xmath205 , and is therefore divisible by @xmath206 . factoring @xmath206 out , we get a reduced equation for @xmath181 , of the form @xmath207 where @xmath208 and @xmath209 are linear in @xmath190 . recalling that @xmath210 an explicit calculation yields : @xmath211 an additional explicit calculation shows that @xmath212 ( to see , say , the first equality , we need to show that @xmath213 . writing @xmath214 , @xmath215 , where @xmath39 is the angle of rotation , and recalling that @xmath216 , the claim follows by straightforward trigonometric manipulations . ) this allows us to rewrite @xmath217 hence , the `` free '' term in ( [ sigma ] ) is the cubic polynomial @xmath218 @xmath219 @xmath220 we refer to the cubic polynomial in the left - hand side of ( [ sigma ] ) as a _ special polynomial_. thus a special surface is the zero set of a special polynomial . [ [ h11-the - geometry - of - special - surfaces . ] ] ( h11 ) the geometry of special surfaces . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + special surfaces pose a technical challenge to the analysis . specifically , each special surface @xmath181 captures a certain underlying pattern in the ground set @xmath0 , which may result in many incidences between rotations and @xmath140-parabolas , all contained in @xmath181 . the next step of the analysis studies this pattern in detail . uvuv1.pstex_t consider first a simple instance of this situation , in which two special surfaces @xmath181 , @xmath221 , generated by two distinct flat rotations @xmath28 , @xmath187 , coincide . more precisely , there exist four parameters @xmath182 such that @xmath28 maps the line @xmath222 to the line @xmath223 ( so that points with the same parameter @xmath155 are mapped to one another ) , and four other parameters @xmath224 such that @xmath187 maps ( in a similar manner ) the line @xmath225 to the line @xmath226 , and @xmath227 . denote this common surface by @xmath181 . since the surfaces coincide , the coefficients @xmath228 , @xmath229 for @xmath230 must be proportional to the coefficients @xmath231 , @xmath232 for @xmath233 . that is , we must have @xmath234 and @xmath235 , for some real @xmath236 . in other words , since @xmath238 are unit vectors , the angle bisector between @xmath168 and @xmath169 must coincide with that between @xmath239 and @xmath186 , as depicted in figure [ uvuv1 ] . moreover , as is easily checked , if we let @xmath240 be the intersection point of @xmath241 and @xmath242 , and let @xmath243 be the intersection point of @xmath244 and @xmath245 , then both @xmath28 and @xmath187 map @xmath240 to @xmath243 , and @xmath246 is contained in @xmath181 . ( see figure [ uvuv2 ] . ) indeed , @xmath187 lies on some parabola @xmath247 through @xmath28 which is contained in @xmath181 , and @xmath28 lies on some parabola @xmath248 through @xmath187 which is also contained in @xmath181 . since a pair of distinct @xmath140-parabolas meet in at most one point , the two parabolas must coincide , so @xmath249 and @xmath250 . however , by construction , @xmath251 lies on @xmath241 and @xmath252 lies on @xmath242 , so this common point must be @xmath240 , and , similarly , @xmath253 , as claimed . uvuv2.pstex_t since the preceding analysis applies to any pair of distinct rotations on a common special surface @xmath181 , it follows that we can associate with @xmath181 a common direction @xmath254 and a common shift @xmath255 , so that for each @xmath256 there exist two lines @xmath159 , @xmath160 , where @xmath28 maps @xmath159 to @xmath160 , so that the angle bisector between these lines is in direction @xmath254 , and @xmath28 is the unique rigid motion , obtained by rotating @xmath159 to @xmath160 around their intersection point @xmath257 , and then shifting @xmath160 along itself by a distance whose projection in direction @xmath254 is @xmath255 . the fact that the shifts of any pair of rotations on @xmath181 have the same @xmath254-component follows from the fact that they both map the intersection point @xmath240 of their source lines to the intersection point @xmath243 of their target lines ; consult figure [ uvuv2 ] . let @xmath181 be a special surface , generated by @xmath178 ; that is , @xmath181 is the union of all parabolas of the form @xmath188 , for @xmath154 . let @xmath258 be the common rotation to all these parabolas , so it maps the line @xmath259 to the line @xmath260 , so that every point @xmath175 is mapped to @xmath176 . let @xmath261 be a parabola contained in @xmath181 but not passing through @xmath258 . take any pair of distinct rotations @xmath262 on @xmath261 . then there exist two respective real numbers @xmath263 , such that @xmath264 , for @xmath265 . thus @xmath266 is the unique rotation which maps @xmath125 to @xmath126 and @xmath173 to @xmath174 . in particular , we have @xmath267 . this in turn implies that the triangles @xmath268 and @xmath269 are congruent ; see figure [ uvuv3 ] . uvuv3.pstex_t given @xmath125 , this determines @xmath126 , up to a reflection about @xmath270 . we claim that @xmath126 has to be on the `` other side '' of @xmath270 , namely , be such that the triangles @xmath268 and @xmath269 are oppositely oriented . indeed , if they were equally oriented , then @xmath258 would have mapped @xmath125 to @xmath126 , and then @xmath261 would have passed through @xmath258 , contrary to assumption . now form the two sets @xmath271 the preceding discussion implies that @xmath272 and @xmath273 are congruent and oppositely oriented . to recap , each rotation @xmath256 , incident to @xmath8 parabolas contained in @xmath181 , corresponds to a pair of lines @xmath274 with the above properties , so that @xmath28 maps @xmath5 points of @xmath275 ( rather , of @xmath276 ) to @xmath5 points of @xmath277 ( that is , of @xmath278 ) . if @xmath28 is flat , its entire multiplicity comes from points of @xmath0 on @xmath159 ( these are the points of @xmath276 ) which are mapped by @xmath28 to points of @xmath0 on @xmath160 ( these are points of @xmath278 ) , and all the corresponding parabolas are contained in @xmath181 . if @xmath28 is a joint then , for any other point @xmath279 outside @xmath159 which is mapped by @xmath28 to a point @xmath280 outside @xmath160 , the parabola @xmath247 is not contained in @xmath181 , and crosses it transversally at the unique rotation @xmath28 . note also that any pair of parabolas @xmath281 and @xmath282 which are contained in @xmath181 intersect , necessarily at the unique rotation which maps @xmath283 to @xmath284 and @xmath285 to @xmath286 . this holds because @xmath287 , as follows from the preceding discussion . [ [ special - surfaces - are - anti - rotations . ] ] special surfaces are anti - rotations . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath181 be a special surface , and let @xmath52 be the subsets of @xmath0 associated with @xmath181 , as in ( [ aandb ] ) . then there exists a single _ anti - rotation _ which maps @xmath272 to @xmath273 . conversely , any anti - rotation can be associated with a unique special surface in this manner . however , the number of incidences within a special surface may be larger than the incidence count of the anti - rotation with the appropriate variants of the @xmath140-parabolas : the former counts incidences between the points of @xmath272 ( or of @xmath273 ) and the lines that they determine , while the latter only counts the size of @xmath272 ( or of @xmath273 ) . [ [ an - alternative - analysis . ] ] an alternative analysis . + + + + + + + + + + + + + + + + + + + + + + + + recall the equation ( [ sigma ] ) of @xmath181 @xmath288 where , writing @xmath289 and @xmath290 , @xmath291 now let @xmath139 be a parabola contained in @xmath181 . substituting the equations ( [ parabola ] ) of @xmath139 into the above equation , we get @xmath292 - ( \lambda z + \mu ) \biggl [ ( a_2+b_2)z^2 - 2a_1z + ( b_2-a_2 ) \biggr ] + k(z ) \equiv 0 , \ ] ] where @xmath293 is the `` free '' cubic term in the equation of @xmath181 . a straightforward algebraic simplification of this equation yields @xmath294 + k(z ) \equiv 0 .\ ] ] in particular ( an interesting observation in itself , albeit obvious from the definition of @xmath295 ) , @xmath296 must be divisible by @xmath297 , with the remainder being a linear function of @xmath190 . eliminating this factor , we get @xmath298 for appropriate reals numbers @xmath283 , @xmath285 . now , writing @xmath299 and @xmath300 , where @xmath39 is the angle of rotation , and observing that @xmath301 the containment of @xmath139 in @xmath181 is equivalent to the two conditions @xmath302 for appropriate parameters @xmath303 . the geometric interpretation of the first condition is that the midpoint of @xmath304 has to lie on a fixed line @xmath305 ( whose direction , @xmath306 , is parallel to the angle bisector between the lines @xmath307 ( see figure [ uvuv2 ] ) . the second condition means that @xmath84 has a fixed component in the direction of @xmath305 . in other words , @xmath139 is contained in @xmath181 if and only if @xmath308 , where @xmath309 is the anti - rotation obtained as a reflection about @xmath305 followed by a shift parallel to @xmath305 . this constitutes an alternative derivation of the characterization of @xmath181 given above . [ [ h12-special - surfaces - and - parabolas . ] ] ( h12 ) special surfaces and parabolas . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + finally , we study intersection patterns involving special surfaces . let @xmath181 be a special surface as above , and let @xmath310 be another @xmath311-linear surface of the form @xmath312 . then either @xmath310 coincides with @xmath181 , or there is at most one parabola contained in both of them . indeed , the intersection of @xmath310 and @xmath181 is the curve satisfying @xmath313 this is a linear system in @xmath194 and @xmath195 . suppose first that its determinant , @xmath314 , does not vanish identically . then , with the exception of finitely many values of @xmath190 , we get a unique solution of the form @xmath315 , @xmath316 , which can describe at most one parabola . if the determinant vanishes identically , then the equation of @xmath310 can be written as @xmath317 , for an appropriate rational algebraic function @xmath318 . if @xmath310 and @xmath181 do intersect in a parabola , then we must have @xmath319 , so @xmath310 and @xmath181 coincide . @xmath98 as a corollary , we have : [ only1 ] let @xmath310 be an @xmath311-linear surface of the above form , and let @xmath28 be a flat rotation contained in @xmath310 . then either @xmath310 contains at least two of the parabolas incident to @xmath28 , and then it must coincide with the corresponding special surface @xmath181 , or @xmath310 contains at most one of these parabolas , so at least two other parabolas cross @xmath310 at @xmath28 . [ sspec ] a special surface can contain at most @xmath1 @xmath140-parabolas . * proof : * let @xmath310 be the given special surface . we claim that for each @xmath320 there can be at most one point @xmath321 such that @xmath322 . indeed , suppose that there exist two such points @xmath323 . since any pair of @xmath140-parabolas on @xmath310 intersect , @xmath324 and @xmath325 meet at a rotation @xmath28 , which maps @xmath29 to both @xmath326 and @xmath327 , an impossibility which completes the proof . @xmath98 [ surfpar ] the number of containments between @xmath328 @xmath140-parabolas and @xmath329 special surfaces is @xmath330 * proof : * as argued above , a special surface @xmath181 is characterized by an anti - rotation @xmath331 in the plane , specified by a line @xmath159 and a shift @xmath255 , such that @xmath332 is the point obtained by reflecting @xmath29 about @xmath159 and then by shifting the reflected point parallel to @xmath159 by distance @xmath255 . thus @xmath181 has three degrees of freedom , and can be parametrized by @xmath333 , where @xmath334 is the equation of @xmath159 and @xmath255 is the shift . we write @xmath335 to denote the special surface parametrized by @xmath336 . by construction , a parabola @xmath139 is contained in @xmath181 if and only if @xmath337 . we use the following parametric setup . we represent each special surface @xmath181 by the corresponding triple @xmath333 , and regard it as a point in parametric 3-space . each parabola @xmath139 is mapped to the locus @xmath338 of all ( points representing ) special surfaces containing @xmath139 . this is a curve in the @xmath333-space , given by the pair of scalar equations @xmath339 . this is a low - degree algebraic curve , whose concrete equations can be worked out explicitly , but we skip over this step . we thus have a system of @xmath329 points and @xmath328 such curves in 3-space , and we wish to bound the number of incidences between them . we have the additional property , noted in lemma [ only1 ] , that two curves meet in at most one point . by projecting these points and curves onto some generic 2-plane , one can easily show that that the number of incidences , and thus the number of original containments , is at most @xmath340 , as claimed . * remark . * if we represent each special surface by its corresponding anti - rotation , lemma [ surfpar ] simply bounds the number of incidences between @xmath329 anti - rotations and @xmath328 ( appropriately transformed copies of ) @xmath140-parabolas , and the bound noted in ( [ weak23 ] ) holds here as well . we begin by reviewing and extending the basic tools from algebraic geometry which have been used in @xcite and in @xcite . however , we develop them here in the context of incidences between points and our @xmath140-parabolas , rather than the context of points and lines considered in the previous papers . so let @xmath341 be a set of @xmath342 @xmath140-parabolas in @xmath2 . for each @xmath343 , we denote the plane containing @xmath344 by @xmath345 and its equation as @xmath346 , where @xmath347 is a linear polynomial . we represent @xmath344 as the intersection curve of @xmath346 and @xmath348 , where @xmath349 is one of the quadratic equations in ( [ parabola ] ) defining @xmath344 , say the first one . note that all the parabolas of @xmath341 cross every plane of the form @xmath196 , each at a single point . recalling the definitions in ( h9 ) , and similar to the case of lines , we say that a point @xmath29 is a _ joint _ of @xmath341 if it is incident to three parabolas of @xmath341 whose tangents at @xmath29 are non - coplanar . let @xmath350 denote the set of joints of @xmath341 . we will also consider points @xmath29 that are incident to three or more parabolas of @xmath341 , so that the tangents to all these parabolas are coplanar , and refer to such points as _ flat _ points of @xmath341 . we recall ( see ( h9 ) ) that any pair of distinct @xmath140-parabolas which meet at a point have there distinct tangents . first , we note that a trivariate polynomial @xmath251 of degree @xmath126 which vanishes at @xmath351 points that lie on a common parabola @xmath343 must vanish identically on @xmath344 . indeed , these points are common roots of @xmath251 and @xmath349 , restricted to the plane @xmath345 . by bzout s theorem @xcite , either these restricted polynomials have a common factor , or they have at most @xmath352 roots . since @xmath349 is irreducible , it must divide the restricted @xmath251 , so @xmath251 must vanish identically on @xmath344 , as claimed . [ [ critical - points - and - parabolas . ] ] critical points and parabolas . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + a point @xmath29 is _ critical _ ( or _ singular _ ) for a trivariate polynomial @xmath251 if @xmath353 and @xmath354 ; any other point @xmath29 in the zero set of @xmath251 is called _ regular_. a parabola @xmath344 is _ critical _ if all its points are critical . the following proposition is adapted from @xcite . [ factor ] let @xmath355 and @xmath356 be two trivariate polynomials , of respective degrees @xmath5 and @xmath357 , so that there are @xmath358 parabolas of @xmath341 on which both @xmath359 and @xmath360 vanish identically . then @xmath359 and @xmath360 have a common factor . * proof . * assume that both @xmath355 and @xmath356 have a positive degree in @xmath18 ; this can always be enforced by an appropriate rotation of the coordinate frame . it is then an easy exercise to show that @xmath359 and @xmath360 have a common factor if and only if their resultant , when viewing them as polynomials in @xmath18 , is identically @xmath361 . recall that the resultant is a polynomial in @xmath362 and @xmath363 . ( the same holds when @xmath359 and @xmath360 have any number of variables , including @xmath18 , in which case the resultant is a polynomial in the remaining variables . ) for any fixed value @xmath364 of @xmath363 , @xmath365 and @xmath366 have at least @xmath358 common roots ( at the intersection points of the @xmath358 parabolas with @xmath367 ) , so , by bzout s theorem @xcite , they have a common factor . therefore , the resultant , with respect to @xmath18 , of @xmath365 and @xmath366 is identically @xmath361 ( as a polynomial in @xmath362 ) . since this is true for every value @xmath364 of @xmath363 , it follows that the resultant of @xmath355 and @xmath356 , with respect to @xmath18 , vanishes identically as a polynomial in @xmath362 and @xmath363 . therefore , @xmath355 and @xmath356 , as trivariate polynomials , have a common factor . @xmath98 [ prop1 ] let @xmath341 be as above . then any trivariate square - free polynomial @xmath251 of degree @xmath126 can have at most @xmath368 critical parabolas in @xmath341 . * proof : * we prove the claim by induction on the degree @xmath126 of @xmath251 . the claim holds trivially for @xmath369 , so assume that @xmath370 . assume first that @xmath251 is irreducible . apply proposition [ factor ] to @xmath251 and @xmath371 , say . both polynomials vanish identically on each critical parabola , and their respective degrees are @xmath126 and @xmath372 . if @xmath251 had more than @xmath368 critical parabolas then @xmath251 and @xmath371 would have a common factor , which is impossible since @xmath251 is irreducible . suppose next that @xmath251 is reducible ( but square - free ) , and write @xmath373 , so that @xmath359 and @xmath360 are non - constant polynomials which have no common factor ( since @xmath251 is square - free , this can always be done ) . denote the degrees of @xmath359 and @xmath360 by @xmath374 and @xmath375 , respectively ; we have @xmath376 . let @xmath344 be a critical parabola for @xmath251 . then either @xmath377 on @xmath344 or @xmath378 on @xmath344 ( or both ) . moreover , since @xmath379 on @xmath344 , it is easily checked that @xmath344 must satisfy ( at least ) one of the following properties : \(i ) @xmath380 on @xmath344 . \(ii ) @xmath344 is a critical parabola of @xmath359 . \(iii ) @xmath344 is a critical parabola of @xmath360 . indeed , if ( i ) does not hold , we have , without loss of generality , @xmath377 on @xmath344 , but @xmath360 vanishes only at finitely many points of @xmath344 . on any other point @xmath29 of @xmath344 we then must have @xmath381 , which implies that @xmath382 is identically zero on @xmath344 , so @xmath344 is critical for @xmath359 . this implies ( ii ) ; ( iii ) holds in the symmetric case where @xmath378 on @xmath344 but @xmath359 does not vanish identically on @xmath344 . by the induction hypothesis , the number of critical parabolas for @xmath359 is at most @xmath383 , and the number of critical parabolas for @xmath360 is at most @xmath384 . consider the parabolas that satisfy ( i ) and intersect all of them by any of the planes @xmath367 , as in the proof of proposition [ factor ] . all the intersection points are roots of @xmath385 and @xmath386 on this plane , and , as follows from the proof of proposition [ factor ] , these bivariate polynomials have no common factor ( or , more precisely , they can have a common factor only at finitely many values of @xmath363 ) . hence , by bzout s theorem , they have at most @xmath387 common roots . altogether , the number of critical parabolas for @xmath251 is at most @xmath388 @xmath98 [ prop2 ] let @xmath29 be a regular point of @xmath251 , so that @xmath389 on three parabolas of @xmath341 passing through @xmath29 . then these parabolas must have coplanar tangents at @xmath29 . * proof : * any such tangent line must be contained in the tangent plane to @xmath390 at @xmath29 . @xmath98 hence , a point @xmath29 incident to three parabolas of @xmath341 whose tangent lines at @xmath29 are non - coplanar , so that @xmath389 on each of these parabolas , must be a critical point of @xmath251 . [ prop4 ] given a set @xmath0 of @xmath357 points in 3-space , there exists a nontrivial trivariate polynomial @xmath391 which vanishes at all the points of @xmath0 , of degree @xmath126 , for any @xmath126 satisfying @xmath392 . * proof : * ( see @xcite . ) a trivariate polynomial of degree @xmath126 has @xmath393 monomials , and requiring it to vanish at @xmath357 points yields these many homogeneous equations in the coefficients of these monomials . such an underdetermined system always has a nontrivial solution . [ [ flat - points - and - parabolas . ] ] flat points and parabolas . + + + + + + + + + + + + + + + + + + + + + + + + + + call a regular point @xmath28 of a trivariate polynomial @xmath251 _ geometrically flat _ if it is incident to three distinct parabolas of @xmath341 ( with necessarily coplanar tangent lines at @xmath28 , no pair of which are collinear ) on which @xmath251 vanishes identically . let @xmath28 be a geometrically flat point of @xmath251 , and let @xmath394 be three incident parabolas on which @xmath251 vanishes . let @xmath395 denote the tangent line to @xmath396 at @xmath28 , and let @xmath397 denote a unit vector in the direction of @xmath395 , for @xmath398 . the second - order taylor expansion of @xmath251 at @xmath28 has the form @xmath399 for any vector @xmath254 , where @xmath400 is the _ hessian _ matrix of @xmath251 . @xmath401 is a quadratic polynomial ( in @xmath254 ) which approximates @xmath251 up to third order terms for sufficiently small values of @xmath402 . our goal is to construct , using this approximation and the fact that @xmath389 on three parabolas incident to @xmath28 , as above , a new polynomial , depending on @xmath251 , which vanishes at @xmath28 , and use this vanishing as a characterization of flat points . to do so , we need to make the analysis more specific , and taylor it to the special form of @xmath140-parabolas . let @xmath28 be a flat point , and let @xmath182 be the corresponding parameters in the @xmath138-plane ( so @xmath28 maps @xmath175 to @xmath176 for each @xmath154 ; cf . remark ( 1 ) at the end of ( h9 ) ) . let @xmath180 be the corresponding special surface spanned by the parabolas @xmath188 , for all @xmath155 ( here we vary @xmath155 continuously , but only finitely many corresponding parabolas belong to @xmath341 ) . since @xmath28 is flat , there exist at least three parabolas @xmath403 , @xmath398 ( all belonging to @xmath341 , contained in @xmath181 , and passing through @xmath28 ) , such that @xmath389 on each of them . let @xmath401 denote , as above , the quadratic polynomial which is the second - order taylor expansion of @xmath251 at @xmath28 . let @xmath404 be one of the above parabolas on which @xmath251 vanishes identically . for @xmath187 in the vicinity of @xmath28 , @xmath405 , so , for points @xmath187 near @xmath28 on @xmath344 , we have @xmath406 . let us continue to consider only points @xmath187 on @xmath344 . let @xmath197 ( resp . , @xmath136 ) be the coordinates of @xmath28 ( resp . , @xmath187 ) . the equations of @xmath344 ( see ( [ parabola ] ) ) are @xmath407 so we have @xmath408 which we can further rewrite as @xmath409 let us simplify these equations as @xmath410 where @xmath411 , @xmath412 , @xmath413 , and @xmath414 are all linear functions of @xmath155 . if we substitute these equations into the equation of @xmath401 , assume that @xmath190 is very close to @xmath205 , ignore terms which are at least cubic in @xmath206 , and use the fact that @xmath415 for any @xmath187 on @xmath344 sufficiently close to @xmath28 , we conclude that both the linear and the quadratic parts of @xmath416 ( in @xmath206 ) vanish identically . the linear part is @xmath417 and the quadratic part is @xmath418 hence we have @xmath419 note that both equations vanish for ( at least ) three distinct values of @xmath155 . since the first equation is linear in @xmath155 and the second is quadratic in @xmath155 , all the coefficients of both equations are identically zero . let us restrict ourselves to the coefficient of the linear term in the first equation and of the quadratic term in the second one . denote by @xmath420 ( resp . , @xmath421 ) the coefficient of @xmath155 in @xmath411 ( resp . , @xmath412 ) . then we have @xmath422 it is easily seen that @xmath420 and @xmath421 can not both be zero ( assuming a generic coordinate frame in the original @xmath138-plane ) , so , eliminating them gives @xmath423 which is the constraint we were after . in what follows , we refer to the left - hand side of ( [ sofsof ] ) as @xmath424 . that is , @xmath425 and this polynomial has to vanish at @xmath28 . we have thus shown : [ flat ] let @xmath251 be a trivariate polynomial . if @xmath28 is a regular geometrically flat point of @xmath251 ( with respect to three parabolas of @xmath341 ) then @xmath426 . * remark . * note that the left - hand side of ( [ sofsof ] ) is one of the three polynomials @xmath427 used in @xcite to analyze flat points in a 3-dimensional line arrangement . specifically , @xmath428 where @xmath429 is the unit vector in the @xmath363-direction ; the other two polynomials are defined analogously , using the other two coordinate vectors @xmath430 , @xmath431 . these polynomials form the _ second fundamental form _ of @xmath251 ; see @xcite for details . in particular , if the degree of @xmath251 is @xmath126 then the degree of @xmath424 is at most @xmath432 . in what follows , we call a point @xmath28 _ flat _ for @xmath251 if @xmath426 . we will need the following technical lemma . [ plinxy ] let @xmath251 be an irreducible trivariate polynomial , with the properties that ( i ) @xmath433 at each regular point @xmath28 of @xmath390 , and ( ii ) @xmath389 on at least two distinct intersecting @xmath140-parabolas of @xmath341 . then @xmath251 is a special polynomial . ( note that the converse of the lemma is trivial , because the second - order derivatives @xmath434 , @xmath435 , and @xmath436 are all identically zero for a special polynomial @xmath251 , and because of the way such polynomials are constructed . ) * proof : * fix @xmath437 and consider the restricted bivariate polynomial @xmath438 . clearly , @xmath439 on the plane @xmath440 . hence @xmath441 at each regular point @xmath442 of @xmath390 , and thus at each regular point of @xmath443 . ( note that a regular point of @xmath443 is also a regular point of @xmath251 , although the converse need not be true . ) note also that @xmath443 is an irreducible polynomial , except possibly for finitely many values of @xmath205 . as is well known @xcite , the curvature of the plane curve @xmath444 , at a regular point of @xmath443 , is given by @xmath445 hence this curve has zero curvature at every regular point of @xmath443 , and thus , being the zero set of an irreducible polynomial , it must be a single line . in other words , @xmath251 is linear in @xmath194 and @xmath195 for every fixed @xmath190 , except for finitely many values , implying that its equation is of the form @xmath446 , where @xmath447 , @xmath448 and @xmath449 are univariate polynomials . we now exploit assumption ( ii ) , denoting by @xmath181 the unique special surface determined by ( and containing ) the two given @xmath140-parabolas . the analysis in ( h12 ) then implies that @xmath181 coincides with the zero set of @xmath251 , so @xmath251 is indeed a special polynomial , as claimed . @xmath98 call an @xmath140-parabola @xmath343 _ flat _ for @xmath251 if all the points of @xmath344 are flat points of @xmath251 ( with the possible exception of a discrete subset ) . arguing as in the case of critical points , if @xmath344 contains more than @xmath450 flat points then @xmath344 is a flat parabola . as in @xcite , we next show that , in general , trivariate polynomials do not have too many flat parabolas . as before , we first establish this property for irreducible polynomials , and then extend the analysis to more general polynomials . [ lflat ] let @xmath251 be an irreducible trivariate polynomial of degree @xmath126 , which is not a special polynomial . then @xmath251 can have at most @xmath451 flat @xmath140-parabolas of @xmath341 . * proof : * suppose to the contrary that there are more than @xmath451 flat @xmath140-parabolas . as above , restrict @xmath251 and @xmath424 to a fixed plane @xmath452 of the form @xmath437 . the number of common roots of @xmath251 and @xmath424 on @xmath452 exceeds the product of their degrees . since this holds for every @xmath205 , proposition [ factor ] implies that they must have a common factor . since @xmath251 is irreducible , @xmath251 must be a factor of @xmath424 . this implies that all the ( regular ) points at which @xmath251 vanishes are flat . hence , by lemma [ plinxy ] , @xmath251 must be a special polynomial , a contradiction which completes the proof of the asserted bound . @xmath98 [ lflatx ] let @xmath251 be any trivariate square - free polynomial of degree @xmath126 with no special polynomial factors . then @xmath251 can have at most @xmath453 flat @xmath140-parabolas in @xmath341 . * proof : * if @xmath251 is irreducible , the claim holds by proposition [ lflat ] . otherwise , write @xmath373 where @xmath359 and @xmath360 are non - constant polynomials with no common factors ( and no special polynomial factors ) . let @xmath374 and @xmath375 denote their respective degrees , so @xmath376 . let @xmath28 be a regular flat point of @xmath251 . then either @xmath454 , or only exactly one of @xmath455 , @xmath456 vanishes . hence , if @xmath344 is a flat parabola for @xmath251 then either both @xmath359 and @xmath360 vanish identically on @xmath344 or exactly one of them vanishes identically on @xmath344 , while the other has only finitely many zeroes on @xmath344 . now , as already argued in the proof of proposition [ prop1 ] , there are at most @xmath387 parabolas of the former kind . to handle parabolas of the latter kind , consider a regular point @xmath28 of @xmath251 at which @xmath385 but @xmath360 is nonzero . a simple calculation yields : @xmath457 hence , at @xmath28 we have @xmath458 and therefore we have at @xmath28 , as is easily checked , @xmath459 that is , a regular flat point for @xmath251 , at which @xmath385 but @xmath360 is nonzero , is a regular flat point for @xmath359 , and a symmetric statement holds when @xmath386 but @xmath359 is nonzero . hence , any flat parabola of the latter kind is either a flat parabola for @xmath359 or a flat parabola for @xmath360 . arguing by induction on the degree , the number of flat parabolas for @xmath251 is thus at most @xmath460 and the lemma follows . in this section we extend the recent algebraic machinery of guth and katz @xcite , as further developed by elekes et al . @xcite , using the algebraic tools set forth in the preceding section , to establish the bound @xmath461 on the number of rotations with multiplicity at least @xmath49 in a collection of @xmath328 @xmath140-parabolas . [ thm : gk1 ] let @xmath341 be a set of at most @xmath328 @xmath140-parabolas in @xmath2 , and let @xmath110 be a set of @xmath357 rotations , each of which is incident to at least three parabolas of @xmath341 . suppose further that no special surface contains more than @xmath401 parabolas of @xmath341 . then @xmath462 . * remarks . * * ( 1 ) * the recent results of @xcite imply that the number of joints in a set of @xmath328 @xmath140-parabolas is @xmath463 . the proofs in @xcite are much simpler than the proof given below , but they do not apply to flat points ( rotations ) as does theorem [ thm : gk1 ] . since flat rotations are an integral part of the setup considered in this paper , we need to count them too , using the stronger theorem [ thm : gk1 ] . moreover , even if we were to consider only joint rotations , the analysis of their incidences with the @xmath140-parabolas will turn some of them into flat rotations ( by pruning some of the parabolas ) , so , as in @xcite , we will need to face flat rotations , no matter what . * ( 2 ) * by lemma [ sspec ] , we always have @xmath464 , and we also have @xmath465 , so the `` worst - case '' bound on @xmath357 is @xmath466 . * ( 3 ) * note that the parameter @xmath328 in the statement of the theorem is arbitrary , not necessarily the maximum number @xmath112 . when @xmath328 attains its maximum possible value @xmath112 , the bound becomes @xmath467 . the proof of theorem [ thm : gk1 ] uses the proof technique of @xcite , properly adapted to the present , somewhat more involved context of @xmath140-parabolas and rotations . * we first prove the theorem under the additional assumption that @xmath468 . the proof proceeds by induction on @xmath328 , and shows that @xmath469 , where @xmath272 is a sufficiently large constant whose choice will be dictated by the forthcoming analysis . the statement holds for all @xmath470 , for some constant @xmath471 , if we choose @xmath272 to be sufficiently large . fix @xmath472 , and suppose that the claim holds for all @xmath473 . let @xmath341 and @xmath110 be as in the statement of the theorem , with @xmath474 , and suppose to the contrary that @xmath475 . we first apply the following iterative pruning process to @xmath341 . as long as there exists a parabola @xmath343 incident to fewer than @xmath476 rotations of @xmath110 , for some constant @xmath477 that we will fix later , we remove @xmath344 from @xmath341 , remove its incident rotations from @xmath110 , and repeat this step with respect to the reduced set of rotations . in this process we delete at most @xmath478 rotations . we are thus left with a subset of at least @xmath479 of the original parabolas , each incident to at least @xmath476 surviving rotations , and each surviving rotation is incident to at least three surviving parabolas . for simplicity , continue to denote these sets as @xmath341 and @xmath110 . choose a random sample @xmath480 of parabolas from @xmath341 , by picking each parabola independently with probability @xmath155 , where @xmath155 is a small constant that we will fix later . the expected number of parabolas that we choose is @xmath481 , where @xmath482 is the number of parabolas remaining after the pruning . we have @xmath483 , because each surviving parabola is incident to at least @xmath476 surviving rotations , each incident to at least two other surviving parabolas ; since all these parabolas are distinct ( recall that a pair of parabolas can meet in at most one rotation point ) , we have @xmath484 . hence , using chernoff s bound , as in @xcite ( see , e.g. , @xcite ) , we obtain that , with positive probability , ( a ) @xmath485 . ( b ) each parabola @xmath343 contains at least @xmath486 rotations that lie on parabolas of @xmath480 . ( to see ( b ) , take a parabola @xmath343 and a rotation @xmath487 . note that @xmath28 will be incident to a parabola of @xmath480 with probability at least @xmath155 , so the expected number of rotations in @xmath488 which lie on parabolas of @xmath480 is at least @xmath489 . this , combined with chernoff s bound , implies ( b ) . ) we assume that @xmath480 does indeed satisfy ( a ) and ( b ) , and then ( recalling that @xmath490 ) choose @xmath491 arbitrary rotations on each parabola in @xmath480 , to obtain a set @xmath0 of at most @xmath492 rotations . applying proposition [ prop4 ] , we obtain a nontrivial trivariate polynomial @xmath493 which vanishes at all the rotations of @xmath0 , whose degree is at most the smallest integer @xmath126 satisfying @xmath494 , so @xmath495 for @xmath328 ( i.e. , @xmath471 ) sufficiently large . without loss of generality , we may assume that @xmath251 is square - free by removing repeated factors , we get a square - free polynomial which vanishes on the same set as the original @xmath251 , with the same upper bound on its degree . the polynomial @xmath251 vanishes on @xmath491 points on each parabola in @xmath480 . this number is larger than @xmath352 , if we choose @xmath155 sufficiently small so as to satisfy @xmath496 . hence @xmath251 vanishes identically on all these parabolas . any other parabola of @xmath341 meets at least @xmath486 parabolas of @xmath480 , at distinct points , and we can make this number also larger than @xmath352 , with an appropriate choice of @xmath155 and @xmath125 ( we need to ensure that @xmath497 ) . hence , @xmath251 vanishes identically on each parabola of @xmath341 . we will also later need the property that each parabola of @xmath341 contains at least @xmath498 points of @xmath110 ; that is , we require that @xmath499 , which will hold if @xmath500 . to recap , the preceding paragraphs impose several inequalities on @xmath125 and @xmath155 , and a couple of additional similar inequalities will be imposed later on . all these inequalities are easy to satisfy by choosing @xmath501 to be a sufficiently small positive constant , and @xmath125 a sufficiently large constant . ( these choices will also affect the choice of @xmath272see below . ) we note that @xmath251 can have at most @xmath502 special polynomial factors ( since each of them is a cubic polynomial ) ; i.e. , @xmath251 can vanish identically on at most @xmath502 respective special surfaces @xmath503 , for @xmath504 . we factor out all these special polynomial factors from @xmath251 , and let @xmath443 denote the resulting polynomial , which is a square - free polynomial without any special polynomial factors , of degree at most @xmath126 . consider one of the special surfaces @xmath505 , and let @xmath172 denote the number of parabolas contained in @xmath505 . then any rotation on @xmath505 is either an intersection point of ( at least ) two of these parabolas , or it lies on at most one of them . the number of rotations of the first kind is @xmath506 . any rotation @xmath28 of the second kind is incident to at least one parabola of @xmath341 which crosses @xmath505 transversally at @xmath28 . we note that each @xmath140-parabola @xmath344 can cross @xmath505 in at most three points . indeed , substituting the equations of @xmath344 into the equation @xmath507 of @xmath505 ( see ( [ sigma ] ) ) yields a cubic equation in @xmath190 , with at most three roots . hence , the number of rotations of the second kind is @xmath508 , and the overall number of rotations on @xmath505 is @xmath509 , since we have assumed in the present version of the proof that @xmath510 . summing the bounds over all surfaces @xmath505 , we conclude that altogether they contain @xmath511 rotations , which we bound by @xmath512 , for some absolute constant @xmath31 . we remove all these vanishing special surfaces , together with the rotations and the parabolas which are fully contained in them , and let @xmath513 and @xmath514 denote , respectively , the set of those parabolas of @xmath341 ( rotations of @xmath110 ) which are not contained in any of the vanishing surfaces @xmath505 . note that there are still at least three parabolas of @xmath515 incident to any remaining rotation in @xmath516 , since none of the rotations of @xmath516 lie in any surface @xmath505 , so all parabolas incident to such a rotation are still in @xmath515 . clearly , @xmath443 vanishes identically on every @xmath517 . furthermore , every @xmath518 contains at most @xmath126 points in the surfaces @xmath505 , because , as just argued , it crosses each surface @xmath505 in at most three points . note that this also holds for every parabola @xmath344 in @xmath519 , if we only count intersections of @xmath344 with surfaces @xmath505 which do not fully contain @xmath344 . hence , each @xmath517 contains at least @xmath520 rotations of @xmath516 . since each of these rotations is incident to at least three parabolas in @xmath515 , each of these rotations is either critical or geometrically flat for @xmath443 . consider a parabola @xmath518 . if @xmath344 contains more than @xmath352 critical rotations then @xmath344 is a critical parabola for @xmath443 . by proposition [ prop1 ] , the number of such parabolas is at most @xmath368 . any other parabola @xmath517 contains more than @xmath521 geometrically flat points and hence @xmath344 must be a flat parabola for @xmath443 . by proposition [ lflatx ] , the number of such parabolas is at most @xmath453 . summing up we obtain @xmath522 we require that @xmath523 ; that is , @xmath524 , which can be guaranteed by choosing @xmath155 sufficiently small . we next want to apply the induction hypothesis to @xmath515 , with the parameter @xmath525 ( which dominates the size of @xmath515 ) . for this , we first need to argue that each special surface contains at most @xmath526 parabolas of @xmath515 . indeed , let @xmath310 be a special surface . using ( [ sigma ] ) , eliminate , say , @xmath195 from the equation of @xmath310 and substitute the resulting expression into the equation of @xmath443 , to obtain a bivariate polynomial @xmath527 . let @xmath344 be a parabola of @xmath515 contained in @xmath310 . we represent @xmath344 by its @xmath194-equation of the form @xmath528 , and observe that @xmath527 vanishes on the zero set of @xmath529 . hence @xmath530 must be divisible by @xmath529 . note that , in a generic coordinate frame in the @xmath138-plane , two different parabolas can not have the same equation @xmath528 , because this equation uniquely determines @xmath531 , and @xmath532 , and then , in a generic frame , @xmath327 is also uniquely determined . note also that the degree of @xmath530 is at most @xmath533 , and that the degree of each factor @xmath529 is @xmath50 , implying that @xmath181 can contain at most @xmath534 parabolas of @xmath515 . an important observation , which we will use in the proof of general version of the theorem , is that the argument just given does not use the assumed bound on the number of @xmath140-parabolas contained in a special surface , but , rather , establishes this bound `` from scratch '' for the subproblem involving @xmath516 and @xmath515 . that is , even if the original problem does not satisfy the extra assumption in the restricted version , the subproblems that it generates always do satisfy it . hence , the maximum number of parabolas of @xmath515 contained in a special surface is at most @xmath535 , so , by the induction hypothesis , the number of points in @xmath516 is at most @xmath536 adding up the bounds on the number of points on parabolas removed during the pruning process and on the special surfaces @xmath505 ( which correspond to the special polynomial factors of @xmath251 ) , we obtain @xmath537 with an appropriate , final choice of @xmath155 , @xmath125 , and @xmath272 . this contradicts the assumption that @xmath538 , and thus establishes the induction step for @xmath328 , and , consequently , completes the proof of the restricted version of the theorem . [ [ proof - of - the - general - version ] ] proof of the general version : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the proof proceeds almost exactly as the proof of the restricted version , except for the analysis of the number of rotations on the special surfaces @xmath505 . as noted above , we encounter this difference only once , in handling the original problem : when we apply the induction step , we always fall into the restricted setup . by assumption , each special surface @xmath505 contains at most @xmath401 @xmath140-parabolas . we modify the preceding analysis , so that each parabola is considered only once . that is , we iterate over the special surfaces in some order . when handling a surface @xmath505 , we consider only those @xmath140-parabolas that are not contained in any previously processed surface , and bound the number of rotations that they contain . then we remove these parabolas and rotations from further considerations and go to the next surface . as argued above , a special surface @xmath505 containing @xmath172 ( surviving ) parabolas contains at most @xmath539 rotations which lie on these parabolas ( and on no previously processed parabola ) . summing these bounds over all special surfaces , and using the fact that @xmath540 for each @xmath165 , we get an overall bound @xmath541 , as asserted . @xmath98 we summarize the remarks following theorem [ thm : gk1 ] , combined with lemma [ lem : lower ] , in the following corollary . [ roth ] let @xmath0 be a set of @xmath1 points in the plane . then there are at most @xmath3 rotations which map some ( degenerate or non - degenerate ) triangle spanned by @xmath0 to another ( congruent and equally oriented ) such triangle . this bound is tight in the worst case . in the following section we will continue to adapt the analysis of @xcite to obtain bounds on the number of incidences between helices ( @xmath140-parabolas ) and rotations with multiplicity @xmath542 , and , consequently , obtain bounds on @xmath543 , for any @xmath4 . in this section we further adapt the machinery of @xcite to derive an upper bound on the number of incidences between @xmath357 rotations and @xmath328 @xmath140-parabolas in @xmath2 , where each rotation is incident to at least three parabolas ( i.e. , has multiplicity @xmath542 ) . [ inc - gen ] for an underlying ground set @xmath0 of @xmath1 points in the plane , let @xmath341 be a set of at most @xmath342 @xmath140-parabolas defined on @xmath0 , and let @xmath110 be a set of @xmath357 rotations with multiplicity at least @xmath49 ( with respect to @xmath0 ) . then @xmath544 * remark . * as easily checked , the first term dominates the second term when @xmath545 , and the second term dominates when @xmath546 . in particular , the first term dominates when @xmath547 , because we have @xmath548 * proof : * the proof of theorem [ inc - gen ] proceeds in two steps . we first establish a bound which is independent of @xmath357 , and then apply it to obtain the @xmath357-dependent bound asserted in the theorem . for the first step , we have : [ inc-32 ] let @xmath341 be a set of at most @xmath342 @xmath140-parabolas defined on @xmath0 , and let @xmath110 be a set of rotations with multiplicity at least @xmath49 with respect to @xmath0 , such that no special surface contains more than @xmath491 parabolas of @xmath341 . then the number of incidences between @xmath110 and @xmath341 is @xmath463 . * proof . * write @xmath549 for short , and put @xmath550 . we will establish the upper bound @xmath551 , for some sufficiently large absolute constant @xmath273 , whose specific choice will be dictated by the various steps of the proof . suppose then to the contrary that @xmath552 for the given @xmath341 and @xmath110 . for @xmath343 , let @xmath553 denote the number of rotations incident to @xmath344 . we refer to @xmath553 as the _ multiplicity _ of @xmath344 . we have @xmath554 . the average multiplicity of a parabola @xmath344 is @xmath555 . we begin by applying the following pruning process . put @xmath556 . as long as there exists a parabola @xmath343 whose multiplicity is smaller than @xmath557 , we remove @xmath344 from @xmath341 , but do not remove any rotation incident to @xmath344 . we keep repeating this step ( without changing @xmath557 ) , until each of the surviving parabolas has multiplicity at least @xmath557 . moreover , if , during the pruning process , some rotation @xmath28 loses @xmath558 incident parabolas , we remove @xmath28 from @xmath110 . this decreases the multiplicity of some parabolas , and we use the new multiplicities in the test for pruning further parabolas , but we keep using the original threshold @xmath557 . when we delete a parabola @xmath344 , we lose at most @xmath557 incidences with surviving rotations . when a rotation @xmath28 is removed , the number of current incidences with @xmath28 is smaller than or equal to twice the number of incidences with @xmath28 that have already been removed . hence , the total number of incidences that were lost during the pruning process is a most @xmath559 . thus , we are left with a subset @xmath516 of the rotations and with a subset @xmath515 of the parabolas , so that each @xmath517 is incident to at least @xmath560 rotations of @xmath516 , and each rotation @xmath561 is incident to at least three parabolas of @xmath515 ( the latter is an immediate consequence of the rule for pruning a rotation ) . moreover , we have @xmath562 . it therefore suffices to bound @xmath563 . let @xmath564 . since at least three parabolas in @xmath515 are incident to each rotation in @xmath516 , it follows that each parabola in @xmath515 is incident to at most @xmath565 rotations of @xmath516 , and therefore @xmath566 . combining this with the fact that @xmath562 , we get that @xmath567 . we fix the following parameters @xmath568 for an appropriate absolute constant @xmath569 , whose value will be fixed shortly . clearly , @xmath570 , and we can also ensure that @xmath571 , i.e. , that @xmath572 , by choosing @xmath573 . furthermore , since @xmath567 we have @xmath574 . we construct a random sample @xmath575 of parabolas of @xmath515 by choosing each parabola independently at random with probability @xmath155 ; the expected size of @xmath575 is @xmath576 . now take @xmath18 ( arbitrary ) rotations of @xmath516 on each parabola of @xmath575 ( which can always be done since @xmath577 ) , to form a sample @xmath0 of rotations in @xmath516 , of expected size at most @xmath578 . for any parabola @xmath517 , the expected number of rotations of @xmath579 which lie on parabolas of @xmath575 is at least @xmath580 ( each of the at least @xmath557 rotations @xmath581 is incident to at least one other parabola of @xmath515 , and the probability of this parabola to be chosen in @xmath575 is @xmath155 ) . we assume that @xmath273 is large enough so that @xmath582 is larger than @xmath583 ( it suffices to choose @xmath584 ) . since @xmath585 , and the expected size of @xmath586 is @xmath587 , we can use chernoff s bound , to show that there exists a sample @xmath575 such that ( i ) @xmath588 , and ( ii ) each parabola @xmath517 contains at least @xmath589 rotations of @xmath516 which lie on parabolas of @xmath575 . in what follows , we assume that @xmath575 satisfies these properties . in this case , we have @xmath590 . now construct , using proposition [ prop4 ] , a nontrivial suqare - free trivariate polynomial @xmath251 which vanishes on @xmath0 , of smallest degree @xmath126 satisfying @xmath591 , so @xmath592 for @xmath328 sufficiently large ( for small values of @xmath328 we ensure the bound by choosing @xmath273 sufficiently large , as before ) . we will choose @xmath593 , so @xmath594 . as above , and without loss of generality , we may assume that @xmath251 is square - free : factoring out repeated factors only lowers the degree of @xmath251 and does not change its zero set . the following properties hold : ( a ) since @xmath595 , @xmath251 vanishes at more than @xmath352 rotations on each parabola of @xmath575 , and therefore , as already argued , it vanishes identically on each of these parabolas . ( b ) each parabola @xmath517 contains at least @xmath596 rotations which lie on parabolas of @xmath575 . since , as just argued , @xmath251 vanishes at these rotations , it must vanish identically on @xmath344 . thus , @xmath389 on every parabola of @xmath515 . before proceeding , we enforce the inequality @xmath597 which will hold if we choose @xmath255 so that @xmath598 . similarly , an appropriate choice of @xmath255 ( or @xmath273 ) also ensures that @xmath599 . we next consider all the special polynomial factors of @xmath251 , and factor them out , to obtain a square - free polynomial @xmath443 , of degree at most @xmath126 , with no special polynomial factors . as in the previous analysis , @xmath251 can have at most @xmath502 special polynomial factors , so it can vanish identically on at most @xmath502 special surfaces @xmath503 , for @xmath504 . let @xmath600 denote the set of those parabolas of @xmath515 which are not contained in any of the vanishing surfaces @xmath505 . for each parabola @xmath601 , @xmath443 vanishes identically on @xmath344 , and ( as argued above ) at most @xmath126 rotations in @xmath579 lie in the surfaces @xmath505 . hence , @xmath344 contains at least @xmath520 remaining rotations , each of which is either critical or flat for @xmath443 , because each such point is incident to at least three parabolas ( necessarily of @xmath602 ) on which @xmath603 . hence , either at least @xmath352 of these rotations are critical , and then @xmath344 is a critical parabola for @xmath443 , or at least @xmath521 of these rotations are flat , and then @xmath344 is a flat parabola for @xmath443 . applying propositions [ prop1 ] and [ lflatx ] , the overall number of parabolas in @xmath602 is therefore at most @xmath604 on the other hand , by assumption , each vanishing special surface @xmath505 contains at most @xmath491 parabolas of @xmath341 , so the number of parabolas contained in the special vanishing surfaces is at most @xmath605 , with our choice of @xmath255 . hence , the overall number of parabolas in @xmath515 is smaller than @xmath606 , a contradiction that completes the proof of theorem [ inc-32 ] . @xmath98 * proof of theorem [ inc - gen ] . * write @xmath549 for short . set @xmath607 and @xmath608 , for some sufficiently large constant @xmath125 whose value will be determined later , and apply the following pruning process . as long as there exists a parabola @xmath343 whose multiplicity is smaller than @xmath557 , we remove @xmath344 from @xmath341 , but do not remove any rotation incident to @xmath344 . similarly , as long as there exists a rotation @xmath141 whose multiplicity is smaller than @xmath609 , we remove @xmath28 from @xmath110 . of course , these removals may reduce the multiplicity of some surviving rotations or parabolas , making additional rotations and parabolas eligible for removal . we keep repeating this step ( without changing the initial thresholds @xmath557 and @xmath609 ) , until each of the surviving parabolas has multiplicity at least @xmath557 and each of the surviving rotations has multiplicity at least @xmath609 . we may assume that @xmath610 , by choosing @xmath125 suficiently large and using theorem [ thm : gk1](i ) . when we delete a parabola @xmath344 , we lose at most @xmath557 incidences with surviving rotations . when a rotation @xmath28 is removed , we lose at most @xmath609 incidences with surviving parabolas . all in all , we lose at most @xmath611 incidences , and are left with a subset @xmath516 of @xmath110 and with a subset @xmath515 of @xmath341 , so that each parabola of @xmath515 is incident to at least @xmath557 rotations of @xmath516 , and each rotation of @xmath516 is incident to at least @xmath609 parabolas of @xmath515 ( these subsets might be empty ) . put @xmath564 and @xmath612 . we have @xmath613 , so it remains to bound @xmath563 , which we do as follows . we fix some sufficiently small positive parameter @xmath501 , and construct a random sample @xmath614 by choosing each rotation of @xmath516 independently with probability @xmath155 . the expected size of @xmath615 is @xmath616 , and the expected number of points of @xmath615 on any parabola of @xmath515 is at least @xmath617 . chernoff s bound implies that , with positive probability , @xmath618 , and @xmath619 for every @xmath517 . we can therefore assume that @xmath615 satisfies all these inequalities . ( for the bound to apply , @xmath620 ( and @xmath357 ) must be at least some sufficiently large constant ; if this is not the case , we turn the trivial bound @xmath621 ( or @xmath622 ) on @xmath623 into the bound @xmath624 ( or @xmath625 ) by choosing the constant of proportionality sufficiently large . ) construct , using proposition [ prop4 ] , a nontrivial square - free trivariate polynomial @xmath251 which vanishes on @xmath615 , whose degree is at most the smallest integer @xmath126 satisfying @xmath626 , so @xmath627 assuming ( as above ) that @xmath620 is sufficiently large . choosing @xmath125 to be large enough , we may assume that @xmath628 . ( this will hold if we ensure that @xmath629 . ) this implies that @xmath251 vanishes at more than @xmath498 points on each parabola @xmath517 , and therefore it vanishes identically on each of these parabolas . as in the previous analysis , we factor out the special polynomial factors of @xmath251 , obtaining a square - free polynomial @xmath443 , of degree at most @xmath126 , with no special polynomial factors . let @xmath503 denote the special surfaces on which @xmath251 vanishes identically ( the zero sets of the special polynomial factors of @xmath251 ) , for some @xmath504 . let @xmath600 ( resp . , @xmath630 ) denote the set of those parabolas of @xmath515 ( resp . , rotations of @xmath516 ) which are not contained in any of the vanishing surfaces @xmath505 . put @xmath631 and @xmath632 . for each parabola @xmath601 , @xmath443 vanishes identically on @xmath344 , and , as argued in the proof of theorem [ thm : gk1 ] , at most @xmath126 rotations of @xmath579 lie in the surfaces @xmath505 . hence , @xmath344 contains more than @xmath520 rotations of @xmath633 , and , arguing as in the preceding proof , each of these rotations is either critical or flat for @xmath443 . hence , either more than @xmath352 of these rotations are critical , and then @xmath344 is a critical parabola for @xmath443 , or more than @xmath521 of these rotations are flat , and then @xmath344 is a flat parabola for @xmath443 . applying propositions [ prop1 ] and [ lflatx ] , the overall number of parabolas in @xmath602 is therefore at most @xmath634 we now apply theorem [ inc-32 ] to @xmath602 and @xmath633 , with the bound @xmath525 on the size of @xmath602 . the conditions of this theorem hold for these sets : clearly , each rotation in @xmath633 is incident to at least three parabolas of @xmath602 . for the other condition , we argue exactly as in the proof of theorem [ thm : gk1 ] , to conclude that any special surface can contain at most @xmath534 parabolas of @xmath515 , establishing the second condition of theorem [ inc-32 ] . this theorem then implies that the number of incidences between @xmath633 and @xmath602 , which is also equal to the number of incidences between @xmath633 and @xmath515 , is @xmath635 moreover , since each parabola of @xmath602 contains at least eight times more rotations of @xmath633 than of @xmath636 , this bound also applies to the number of incidences between @xmath636 and @xmath602 . it therefore remains to bound the number of incidences between @xmath636 and @xmath637 , namely , between the rotations and parabolas contained in the vanishing special surfaces @xmath505 . to do so , we iterate over the surfaces , say , in the order @xmath503 . for each surface @xmath505 in turn , we process the rotations and parabolas contained in @xmath505 and then remove them from further processing on subsequent surfaces . let us then consider a special surface @xmath505 . let @xmath638 and @xmath639 denote respectively the number of rotations and parabolas contained in @xmath505 , which were not yet removed when processing previous surfaces . the number of incidences between these rotations and parabolas can be bounded by the classical szemerdi - trotter incidence bound @xcite ( see also ( [ weak23 ] ) ) , which is @xmath640 . summing these bounds over all the special surfaces @xmath505 , and using hlder s inequality and the fact , established in lemma [ sspec ] , that @xmath641 , we get an overall bound of @xmath642 @xmath643 where we use the facts that @xmath644 and @xmath645 , which follow since in this analysis each parabola and rotation is processed at most once . the two linear terms satisfy @xmath646 ( the bound obtained in the pruning process ) , and @xmath647 since @xmath648 ; see remark ( 2 ) following theorem [ thm : gk1 ] . we are not done yet , because each rotation of @xmath636 is processed only once , within the first surface @xmath505 containing it . this , however , can be handled as in @xcite . that is , let @xmath28 be a rotation which was processed within the first surface @xmath505 containing it . suppose that @xmath28 also lies on some later surface @xmath649 , with @xmath650 , and let @xmath344 be a parabola contained in @xmath649 , which has not been removed yet ; in particular , @xmath344 is not contained in @xmath505 , and thus meets it transversally , so the incidence between @xmath344 and @xmath28 can be regarded as one of the transversal incidences in @xmath505 , which we have been ignoring so far . to count them , we simply recall that each parabola , whether of @xmath637 or of @xmath602 , has at most three transversal intersections with a surface @xmath505 ( see the proof of theorem [ thm : gk1 ] ) , for a total of at most @xmath126 crossings with all the vanishing surfaces . since each of these parabolas contains at least @xmath498 rotations of @xmath516 , those `` transversal incidences '' are only a fraction of the total number of incidences , and we simply ignore them altogether . to recap , we obtain the following bound on the number of incidences between @xmath516 and @xmath515 : @xmath651 adding the bound @xmath652 on the incidences lost during the pruning process , we get the asserted bound . @xmath98 it is interesting to note that the proof technique also yields the following result . [ mgek ] let @xmath341 be a set of @xmath328 @xmath140-parabolas and @xmath110 a set of points in 3-space which satisfy the conditions of theorem [ inc - gen](i ) . then , for any @xmath653 , the number @xmath654 of points of @xmath110 incident to at least @xmath5 parabolas of @xmath341 satisfies @xmath655 * proof : * write @xmath656 for short . we clearly have @xmath657 . theorem [ inc - gen ] then implies @xmath658 , from which the first two bounds follow . if @xmath659 we use the other bound ( in ( [ weak23 ] ) ) , to obtain @xmath660 , which implies that @xmath661 ( which is in fact an equivalent statement of the classical szemerdi - trotter bound ) . @xmath98 in this section we further improve the bound in theorem [ inc - gen ] ( and corollary [ mgek ] ) using more standard space decomposition techniques . we show : [ thm : impr ] the number of incidences between @xmath357 arbitrary rotations and @xmath328 @xmath140-parabolas , defined for a planar ground set with @xmath1 points , is @xmath662 where the @xmath663 notation hides polylogarithmic factors . in particular , when all @xmath547 @xmath140-parabolas are considered , the bound is @xmath664 * proof : * we dualize the problem as follows . we map each parabola @xmath139 to the point @xmath665 in @xmath666 . each rotation @xmath28 is mapped to a 2-plane @xmath667 , which is the locus of all points @xmath668 such that @xmath28 is incident to @xmath344 . this is indeed a 2-plane , because the equations of @xmath28 , either ( [ eq : helix ] ) in the @xmath103-frame , or ( [ parabola ] ) in the @xmath136-frame , are a pair of linear ( independent ) equations in @xmath108 . so in this new setup we have @xmath328 points and @xmath357 2-planes in 4-space , and we wish to bound the number of incidences between these points and 2-planes . we note that any pair of these 2-planes intersect in at most one point . ( the corresponding statement in the primal setup is that two rotations can be incident to at most one common @xmath140-parabola . ) to bound the number of incidences , we first project the points and 2-planes onto the 3-space @xmath669 . we claim that , with a generic choice of the coordinate frame in the original @xmath138-plane , the projected points remain distinct . indeed , a point @xmath108 , dual to an @xmath140-parabola @xmath139 , is projected to the point @xmath670 , so the projected point uniquely determines @xmath29 , and also @xmath31 , because we may assume that no two points of @xmath0 have the same @xmath18-coordinate @xmath326 . hence the projected points are all distinct . this is not necessarily the case for the 2-planes . indeed , consider a 2-plane @xmath667 . its projection onto the @xmath671-space is the plane satisfying the first equation of ( [ parabola ] ) , say , namely @xmath672 it is easily checked that this equation uniquely determines the @xmath194 and @xmath190 components of @xmath28 , leaving @xmath195 ( i.e. , the shift along the @xmath362-direction that @xmath28 makes after its initial pure rotation ) undetermined . thus it is possible that several distinct rotations , all with the same @xmath194 and @xmath190 components , are projected to the same 2-plane . this has the potential danger that the projection loses incidences , when several 2-planes , incident to a common point @xmath667 , get projected into the same plane , so that , instead of several incidences with @xmath667 in 4-space , we get only one incidence in the projection . nevertheless , this bad situation can not arise . this follows from the easy observation that two distinct rotations with the same @xmath194 and @xmath190 components can not both map a point @xmath673 into the same point @xmath674 . to recap , after the projection we get @xmath328 points and at most @xmath357 planes in @xmath2 , and our goal is to bound the number of incidences between them . more precisely , we want to bound only the number of original incidences . we note that each such incidence appears as an incidence in the projection , but not necessarily the other way around . we recall that , in general , the number of incidences between @xmath328 points and @xmath357 planes in 3-space can be @xmath622 in the worst case , because of the possibility that many points lie on a common line and many planes pass through that line . this situation can also arise in our setup , but we will apply a careful analysis to show that the number of original incidences that project to such a degenerate configuration is much smaller . we proceed as follows . we fix a parameter @xmath675 , to be determined shortly , and construct the following decomposition of 3-space . first , we note that the projected points @xmath670 have only @xmath1 distinct @xmath676-coordinates , which are the @xmath18-coordinates of the points of @xmath0 . similarly , they have only @xmath1 distinct @xmath326-coordinates . we partition the 3-space by a set @xmath677 of @xmath675 planes orthogonal to the @xmath676-axis , so that within each resulting slab the projected points have at most @xmath678 distinct @xmath676-coordinates . we construct a similar collection @xmath679 of @xmath675 planes orthogonal to the @xmath326-axis , so that within each resulting slab the projected points have at most @xmath678 distinct @xmath326-coordinates . we then choose a random sample @xmath680 of @xmath675 of the projected planes . we take the set @xmath681 of @xmath682 planes , construct their arrangement , and decompose each of its cells into simplices . we obtain @xmath683 simplices , and the construction and the standard @xmath684-net theory @xcite imply that , with high probability , the following properties hold for every simplex @xmath685 of the partition : ( i ) @xmath685 is crossed by at most @xmath686 projected 2-planes ; ( ii ) the projected points that fall into @xmath685 have at most @xmath678 distinct @xmath676-coordinates and at most @xmath678 distinct @xmath326-coordinates . further refining the simplices , if necessary , we can also assume that ( iii ) each simplex contains at most @xmath687 projected points . property ( ii ) is crucial . it asserts that the number of points of @xmath0 which induce the parabolas whose dual points project into a fixed simplex is at most @xmath688 ; more precisely , there are only @xmath678 `` source '' points of @xmath0 and only @xmath678 `` target '' points , so that each of these parabolas is of the form @xmath139 , where @xmath29 is one of the @xmath678 source points and @xmath31 is one of the @xmath678 target points . ( note , by the way , that the number of parabolas , @xmath687 , in volved in a subproblem is much smaller than the maximum possible value @xmath689 , when @xmath690 . ) we now apply theorem [ inc - gen ] to each simplex @xmath685 ; that is , to the set @xmath691 of those parabolas whose ( projected ) dual points lie in @xmath685 , and to the set @xmath692 of those rotations whose ( projected ) dual 2-planes cross @xmath685 . put @xmath693 and @xmath694 . we note that some rotations in @xmath692 may be incident to no more than two parabolas in @xmath691 ; these rotations contribute @xmath695 to the overall incidence bound . by theorem [ inc - gen ] we thus have is subsumed by the other terms , because of the possibility that some of the @xmath696 rotations are incident to only one or two parabolas in a subproblem . ] @xmath697 summing these bounds over all cells @xmath685 , we get an overall bound of @xmath698 @xmath699 where , as above , @xmath663 hides polylogarithmic factors . we also have to add to the bound incidences involving points , which are projections dual to parabolas , which lie on the boundaries of the cells of the cutting . let @xmath700 , the projection of a ( unique ) point @xmath701 be such a point . let @xmath359 denote the face whose relative interior contains @xmath401 . if @xmath359 is a 2-face of some simplex @xmath685 , we can associate @xmath401 with @xmath685 : except for the single plane containing @xmath359 , any other plane incident to @xmath401 must cross @xmath685 , and we can count the incidence within the subproblem of @xmath685 . the uncounted incidences , at most one per parabola , add up to at most @xmath328 . if @xmath359 is a vertex ( so @xmath702 ) then any plane through @xmath359 either bounds or crosses some adjacent simplex , so the total number of such incidences is @xmath703 . the harder situation is when @xmath359 is an edge . again , if a plane _ crosses _ @xmath359 at @xmath401 , we can count this incidence within any adjacent simplex , arguing as in the case where @xmath359 is a 2-face . the difficult case is when the plane _ contains _ @xmath359 , and we handle it as follows . it is simpler to consider @xmath359 as a full line of intersection of two sampled planes , rather than a single edge . ( the decomposition , though , has also other edges , obtained in the decomposition of arrangement cells into simplices ; these edges require a slightly different treatment , given below . ) let @xmath704 be the projected dual points that lie on @xmath359 , and let @xmath705 denote the parabola corresponding to @xmath706 , for @xmath707 . consider the rotations @xmath28 whose dual 2-planes project to planes containing @xmath359 . rotations @xmath28 of this kind which are incident to just one of the parabolas @xmath705 are easy to handle , because the number of incidences involving these rotations is at most @xmath357 ( for the fixed line @xmath359 ) , for a total of @xmath708 . consider then those rotations @xmath28 which are incident to at least two of the parabolas @xmath705 . since the points @xmath709 lie on a common line , it follows that the points @xmath163 are also collinear in the original @xmath138-plane , lying on a common line @xmath305 . the points @xmath164 are not necessarily collinear , but they have the property that , for any pair of indices @xmath710 , the ratio @xmath711 is fixed . see figure [ abab ] . abab.pstex_t now if @xmath28 is incident to two parabolas @xmath705 , @xmath712 , then @xmath28 maps @xmath163 to @xmath164 and @xmath713 to @xmath714 . in particular , @xmath715 . this , and the fact that @xmath711 is fixed , imply that @xmath28 maps @xmath305 to the line through @xmath164 and @xmath714 , and that the slope of this line has a _ fixed absolute value _ @xmath716 . hence , considering , with no loss of generality , only lines of the latter kind with positive slope , we can partition @xmath717 into equivalence classes , so that , for each class , all the corresponding points @xmath164 lie on a common line of slope @xmath716 . moreover , there is at most one rotation that is incident to at least two parabolas from the same class ( and no rotation can be incident to two parabolas from different classes ) . thus the total number of incidences of this kind , for the fixed @xmath359 , is at most @xmath155 . summing over all lines @xmath359 , we get a total of @xmath508 such incidences . in the preceding analysis we considered only intersection lines between sampled planes , but , as noted , the cutting has additional edges , interior to cells of the arrangement . we handle such edges in almost the same way as above . that is , we consider such an edge @xmath718 , and argue , exactly as above , that the number of original incidences involving points on @xmath718 and planes that contain @xmath718 is proportional to the number @xmath719 of points on @xmath718 plus the number @xmath720 of planes containing @xmath718 . ( incidences involving planes that cross @xmath718 are also handled exactly as above , wih the same resulting bound . ) the sum @xmath721 is still at most @xmath328 . for the other sum @xmath722 , we note that the number of edges @xmath718 is @xmath683 ( instead of @xmath723 in the preceding analysis ) , but each edge @xmath718 can be contained in at most @xmath686 planes , as follows easily from the @xmath684-net theory ( this holds with high probability , but we may assume that our sample does indeed have this property ) . hence , we have @xmath724 , the same bound as above . altogether , the number of incidences is thus @xmath725 we now choose @xmath726 this choice of @xmath675 makes the first and third terms in the incidence bound equal to each other , and they both dominate the second term , as is easily verified , using the fact that @xmath342 . note also that @xmath727 when @xmath728 assume first that @xmath357 lies in this range . then the incidence bound becomes @xmath729 when @xmath730 , we use @xmath731 and get the bound @xmath732 since @xmath733 , the second term dominates the two other terms , and the bound is thus @xmath734 . finally , when @xmath735 , we use the szemerdi - trotter bound in ( [ weak23 ] ) , which is easily seen to yield the bound @xmath508 . adding all these bounds , the theorem follows . @xmath98 using this bound , we can strengthen corollary [ mgek ] , as follows . [ mgek1 ] let @xmath341 be a set of @xmath328 @xmath140-parabolas and @xmath110 a set of rotations , with respect to a planar ground set @xmath0 of @xmath1 points . then , for any @xmath4 , the number @xmath654 of rotations of @xmath110 incident to at least @xmath5 parabolas of @xmath341 satisfies @xmath736 for @xmath547 , the bound becomes @xmath737 * proof : * the proof is similar to the proof of corollary [ mgek ] , and we omit its routine details . @xmath98 in this paper we have reduced the problem of obtaining a near - linear lower bound for the number of distinct distances in the plane to a problem involving incidences between points and a special class of parabolas ( or helices ) in three dimensions . we have made significant progress in obtaining upper bounds for the number of such incidences , but we are still short of tightening these bounds to meet the conjectures on these bounds made in the introduction . to see how far we still have to go , consider the bound in corollary [ mgek1 ] , for the case @xmath547 , which then becomes @xmath738 . ( here @xmath654 coincides with @xmath58 as defined in ( h3 ) . ) moreover , we also have the szemerdi - trotter bound @xmath739 , which is smaller than the previous bound for @xmath740 . substituting these bounds in the analysis of ( h3 ) and ( h4 ) , we get @xmath62 ^ 2}{x } \le |k| = n_{\ge 2 } + \sum_{k\ge 3 } ( k-1 ) n_{\ge k } = \ ] ] @xmath741 = n_{\ge 2 } + o(s^{29/9 } ) .\ ] ] it is fairly easy to show that @xmath119 is @xmath742 , by noting that @xmath119 can be upper bounded by @xmath743 , where @xmath744 is as defined in ( h1 ) . using the upper bound @xmath745 @xcite , we get @xmath746 thus , at the moment , @xmath119 is the bottleneck in the above bound , and we only get the ( weak ) lower bound @xmath11 on the number of distinct distances . showing that @xmath747 too ( hopefully , a rather modest goal ) would improve the lower bound to @xmath748 , still a rather weak lower bound . \(ii ) it gave us a new , and considerably more involved setup in which the new algebraic technique of guth and katz could be applied . as such , the analysis in this paper might prove useful for obtaining improved incidence bounds for points and other classes of curves in three dimensions . the case of points and circles is an immediate next challenge . another comment is in order . our work can be regarded as a special variant of the complex version of the szemerdi - trotter theorem on point - line incidences @xcite . in the complex plane , the equation of a line ( in complex notation ) is @xmath749 . interpreting this equation as a transformation of the real plane , we get a _ homothetic map _ , i.e. , a rigid motion followed by a scaling . we can therefore rephrase the complex version of the szemerdi - trotter theorem as follows . we are given a set @xmath110 of @xmath357 pairs of points in the ( real ) plane , and a set @xmath750 of @xmath328 homothetic maps , and we seek an upper bound on the number of times a map @xmath751 and a pair @xmath752 `` coincide '' , in the sense that @xmath46 . in our work we only consider `` complex lines '' whose `` slope '' @xmath251 has absolute value @xmath117 ( these are our rotations ) , and the set @xmath110 is simply @xmath753 . \(a ) obtain a cubic upper bound for the number of rotations which map only two points of the given ground planar set @xmath0 to another pair of points of @xmath0 . any upper bound smaller than @xmath754 would already be a significant step towards improving the current lower bound of @xmath17 on distinct distances @xcite . the bulk of the paper was written after the passing away of gyrgy elekes in september 2008 . however , the initial infrastructure , including the transformation of the distinct distances problem to an incidence problem in three dimensions , and many other steps , is due to him . as a matter of fact , it was already discovered by elekes about 10 years ago , and lay dormant since then , mainly because of the lack of effective tools for tackling the incidence problem . these tools became available with the breakthrough result of guth and katz @xcite in december 2008 , and have made this paper possible . thanks are due to mrton elekes , who was a driving force in restarting the research on this problem . n. h. katz and g. tardos , a new entropy inequality for the erds distance problem , in _ towards a theory of geometric graphs _ , j. pach , ed . , _ contemporary math . 342 , amer . soc . press , providence , ri , 2004 , pp . 119126 . a. pressley , _ elementary differential geometry _ , springer undergraduate mathematics series , springer verlag , london , 2001 . r. quilodrn , the joints problem in @xmath755 , arxiv:0906.0555 , posted june 2 , 2009 . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: quantum hall effects are characteristic of two - dimensional ( 2d ) electron gases in strong magnetic fields . the integral quantum hall effect ( iqhe ) @xcite is explained in terms of weakly - interacting electrons filling quantized landau levels ( ll ) produced by application of a strong magnetic field perpendicular to the 2d gas @xcite . in contrast , the fractional quantum hall effect ( fqhe ) @xcite can occur only as a result of strong electronic correlations in partially - filled landau levels @xcite . because of its massless chiral charge carriers and atomic - monolayer honeycomb lattice exhibiting sublattice , valley , and spin degeneracies , quantum hall effects in graphene could be much richer than in the semiconductor 2d electron gas , where there is no crystal structure and the only degeneracies are those of the ( non - relativistic ) landau levels and spin . for graphene in strong magnetic fields an integral quantum hall effect @xcite and a fractional quantum hall effect @xcite have been observed , with anomalous filling factors that reflect the unique degeneracies of the graphene electronic structure and the dirac - like nature of its electrons . the valley isospin and spin degrees of freedom imply that graphene in a strong magnetic field is described well by a low - energy hamiltonian that commutes approximately with the generators of an su(4 ) lie algebra . this su(4 ) symmetry allows classification of states in graphene , and can serve as the basis for computing explicit breaking of the su(4 ) symmetry by small non - symmetric terms in the effective hamiltonian . however , there is growing evidence that many states observed in modern experiments can not be described by explicit breaking of su(4 ) . for example , the ground state of graphene in a magnetic field at low temperature exhibits a rapid increase of the longitudinal resistance @xmath0 above a critical magnetic field @xmath1 @xcite . the value of @xmath1 decreases for increasingly cleaner samples , indicating that the resistance is not caused by impurity scattering but instead is intrinsic to the state itself @xcite . in quantum hall systems the currents are carried by edge states , so this insulating ground state must correspond to an emergent state that does not support edge currents produced by _ spontaneous _ ( not explicit ) breaking of the su(4 ) symmetry . thus the approximate su(4 ) symmetry of graphene can suggest possible low - energy collective modes exhibiting spontaneously broken symmetry that are important for the properties of graphene in a magnetic field , but the su(4 ) symmetry alone can not determine which of these modes is the ground state . until now , those broken - symmetry modes have been addressed quantitatively by numerical simulations employing limited numbers of states and particles , or by effective low - energy field - theory approximations . these calculations find various possible low - energy collective states resulting from spontaneous breaking of the su(4 ) symmetry with very similar energies . thus they have been unable to give a definitive answer to the nature of the insulating ground state . let us note that other approaches to the problems addressed here have been proposed ( for example , refs . @xcite ) . however , the present discussion will concentrate on methods based on approximate su(4 ) symmetry of the hamiltonian . an alternative and potentially more powerful application of symmetries has been employed extensively in both nuclear structure and condensed matter physics @xcite . fermion dynamical symmetry method _ truncates the hilbert space to a tractable collective subspace by positing a highest symmetry associated with the physical operators for the system , and constructing effective hamiltonians from polynomials in the casimir invariants of the highest symmetry s subgroup chains . in this approach it is possible not only to classify low - energy collective modes , but to solve analytically for the properties of these modes and to determine which lie lowest in energy , either exactly in particular symmetry limits , or approximately using generalized coherent state methods . in an earlier letter the first application of fermion dynamical symmetry methods to graphene was introduced and applied to determining the ground state in strong magnetic fields @xcite . this paper develops the full dynamical symmetry formalism upon which ref . @xcite rests . it will be shown that the highest symmetry is so(8 ) , with its generators identifiable with particle hole and pairing degrees of freedom that have been discussed previously in the physics of graphene . this symmetry will be shown to be isomorphic to an so(8 ) symmetry used extensively in nuclear structure physics , which permits already - developed mathematics to be appropriated for the graphene problem , and suggests instructive physical analogies between two very different physical systems . a generalized coherent state approximation will be introduced that corresponds to a hartree fock - bogoliubov ( hfb ) formalism subject to so(8 ) symmetry constraints . this permits quantitative evaluation of energy surfaces associated with spontaneously - broken symmetry . the so(8 ) highest symmetry will be shown to have an su(4 ) subgroup that recovers the known physics of su(4 ) quantum hall ferromagnetism as a special case , but implies in the more general case new low - energy physics that transcends su(4 ) quantum hall ferromagnetism . hence , a solvable and physically - illuminating approach to the rich low - energy structure of undoped graphene in strong magnetic fields will be proposed that reproduces known physics , but also suggests testable new physics in this complex system . comprehensive reviews of graphene physics may be found in refs . the presentation here will recall only a select set of features that will be relevant for subsequent discussion . undoped graphene is a 2-dimensional semiconductor with zero bandgap . it has a bipartite honeycomb lattice structure illustrated in fig . [ fig : a - b_sublattices ] that corresponds to two interlocking triangular sublattices , labeled a and b. the two - fold degree of freedom specifying whether an electron is on the a sublattice or b sublattice is a spin - like quantum number termed the _ sublattice pseudospin_. the sublattice pseudospin behaves mathematically like the actual spin of the electron ( which will be introduced later ) , but it is a separate degree of freedom . the dispersion of energy with momentum for undoped graphene in the absence of a magnetic field is illustrated in fig . [ fig : dispersiongraphene ] . 0pt 0.44 electronic dispersion of graphene calculated in a tight - binding model with no magnetic field . two inequivalent points in the brillouin zone are labeled @xmath2 and @xmath3 . near these k - points the dispersion becomes linear , leading to _ dirac cones , _ as shown in the expanded view . for undoped graphene the fermi surface lies at the apex of the cones , where the level density vanishes and the effective electronic mass tends to zero . thus low - energy electrons are governed approximately by a dirac equation for massless electrons . the two inequivalent points @xmath2 and @xmath3 are not connected by reciprocal lattice vectors . the corresponding two - fold k degree of freedom may be viewed mathematically as a 2-valued spin - like quantity , commonly termed the _ valley isospin _ because of the valley - like structure in the dispersion of fig . [ fig : dispersiongraphene ] around the k - points . for brevity the valley isospin will sometimes be termed simply _ isospin . _ obviously this two - fold `` isospin '' degree of freedom is a pseudospin that should be distinguished physically from the actual spin of the electron , and from standard usage of isospin quantum numbers in nuclear and particle physics . near the _ dirac cones _ ( inset to fig . [ fig : dispersiongraphene ] ) the dispersion is linear , the density of electronic states tends to zero , and the electrons are described by a massless dirac equation in which the fermi velocity plays the role that the speed of light would play in an actual relativistic system . thus the low - energy electrons for undoped graphene in zero magnetic field behave to good approximation as _ massless chiral fermions , _ with the chirality representing the projection of the sublattice pseudospin on the direction of motion , not the projection of the actual spin . the vanishing of the density of states at the dirac point ( fermi surface ) implies that the transport properties of graphene are different from either a metal or a semiconductor . our interest will be primarily low - energy states in a strong magnetic field . for non - interacting electrons , the quantized levels may be found by solving the dirac equation for massless fermions with a vector potential appropriate to the applied magnetic field . the dispersion of energy with magnetic field strength for massless dirac electrons is illustrated in fig . [ fig : llgraphene_n0](a ) . consider the @xmath4 state , which corresponds to half filling of the fourfold - degenerate @xmath5 ll in graphene , as illustrated in fig . [ fig : llgraphene_n0](b ) . the graphene honeycomb lattice is bipartite , as illustrated in fig . [ fig : a - b_sublattices ] . the @xmath5 ll is located exactly at the dirac point corresponding to @xmath6 . for low - energy excitations in each valley labeled by @xmath2 or @xmath3 , the inter - valley tunneling may be ignored and the electrons in the valley reside entirely on either the a or b sublattice . hence , for the @xmath5 ll labeling with the valley isospin ( indicating whether the electron is in a @xmath2 or @xmath3 valley ) is equivalent to labeling with the sublattice pseudospin ( indicating whether the electron is on the a or b sublattice ) . this is reminiscent of a nel state with spins on two different sublattices , with a nel order defined by the difference in spins on the a and b sublattices . a quantum hall effect is signaled by a plateau in the hall conductance @xmath7 having quantized values @xmath8 where the _ filling factor _ @xmath9 is defined by @xmath10 with @xmath11 the charge - carrier density , @xmath12 the magnetic field strength , and @xmath13 the magnetic flux density in units of the fundamental flux quantum @xmath14 . these plateaus indicate formation of an _ incompressible quantum liquid . _ this is a compact way to say that the ground state is separated from excited states by an _ energy gap , _ which inhibits compression because of the energy required for excitation across the gap . the integral quantum hall effect ( iqhe ) in graphene is similar to the integral quantum hall effect for non - relativistic electrons in that it corresponds to the formation of incompressible states resulting from the complete filling of landau levels by weakly - interacting electrons . however , there are two important differences between the iqhe states observed in graphene and those observed in conventional 2d semiconductor heterostructures : ( 1 ) in addition to the two - fold spin degeneracy ( in the absence of zeeman splitting ) , there is a two - fold valley degeneracy associated with the distinct @xmath2 and @xmath15 points in the first brillouin zone . thus the filling factor changes in steps of four between plateaus in the hall resistance for graphene . ( 2 ) for graphene the filling factor @xmath9 defined in eq . ( [ iqhegraph1.1 ] ) vanishes at the dirac point for particle hole symmetric half filling of the graphene lattice , since the electron density @xmath11 tends to zero there . hence , in the absence of a zeeman effect or strong electronic correlations , there is no integral quantum hall effect in graphene for @xmath16 . ( a ) relativistic dispersion for massless dirac electrons as a function of magnetic field strength @xmath12 . states are labeled by a principle quantum number @xmath17 and a quantum number @xmath18 indicating particle - like ( @xmath19 ) and hole - like ( @xmath20 ) states . each landau level labeled by @xmath17 has a high orbital degeneracy @xmath21 ( see eq . ( [ degen1.1 ] ) ) , and an additional 4-fold degeneracy associated with spin and valley isospin . ( b ) one configuration for occupation of the @xmath5 landau level in monolayer graphene . the splitting and occupation are schematic only . the ground state would be a superposition of such configurations and in the su(4 ) symmetry limit ( which obtains for coulomb - only interactions ) the 4 levels shown labeled by valley @xmath2 or @xmath3 and spin up or down would be degenerate . in realistic cases the splitting often is small , suggesting approximate su(4 ) symmetry . ] in graphene the analog of the integral quantum hall effect was first observed at filling factors @xmath22 by sweeping the field or the carrier density through the landau levels @xcite . this implies hall resistance quantization for filling factors in the sequence @xmath23 where @xmath11 is the charge carrier density and @xmath17 is the landau level index @xcite . this sequence is quite different from the integral quantum hall effect sequence observed in other 2d electron gases . however , it likely results from the same basic physics as the normal iqhe , modified by the 4-fold spin valley degeneracies of non - interacting , massless dirac electrons . the period @xmath24 in eq . ( [ iqhegraph1.2 ] ) is is a consequence of the approximate four - fold degeneracy of the graphene landau level , and the added @xmath25 ( which is not present in non - relativistic 2d systems ) is a berry phase effect that results from the special status of the @xmath5 state for massless dirac fermions : a quantum phase arises at the band degeneracy point associated with precession of the pseudospinor describing the 2-fold sublattice degree of freedom , which modifies the quantization condition for electronic orbits @xcite . the four - fold near - degeneracy of the landau level follows because the zeeman energy is small compared with the interaction energy , and the pseudospin degree of freedom representing the two inequivalent dirac cones at the corners of the brillouin zone ( k and k@xmath26 ) does not couple to external fields if the two sublattices are equivalent . because of the dirac - cone dispersion with the fermi level located at the apex of the cones ( see fig . [ fig : dispersiongraphene ] ) , low - energy excitations in graphene occur in regions of reduced electron density , which disfavors electron correlations . but by placing a strong perpendicular magnetic field on the system the resulting landau quantization ( corresponding semiclassically to requiring that an integral number of debroglie wavelengths wrap around a cyclotron orbit , and implying that an integral number of magnetic flux quanta pass through the area bounded by the orbit ) leads to a bunching of levels into regions of locally high degeneracy separated by gaps . these regions of locally high level density may exhibit conditions more favorable for the development of strong correlations between electrons . landau levels ( ll ) become strongly correlated when inter - ll excitations are of sufficiently high energy that they may be neglected and the low - energy excitations involve only intra - ll transitions . then the kinetic energy of the ll is a constant that can be omitted . this limit of strong electronic correlations has two important physical implications : ( 1 ) the approximate 4-fold spin valley degeneracy of the graphene landau levels leads to quantum hall ferromagnetic states that will be discussed further below . ( 2 ) the strong correlations can produce incompressible states at partial filling of the ll that are reminiscent of the fractional quantum hall effect ( fqhe ) in semiconductor devices . after the discovery of the integral quantum hall effect in graphene , experiments performed at higher magnetic field strengths observed fragile quantum hall states at filling factors @xmath27 . the @xmath28 states are thought to be the result of single - particle zeeman splitting of the landau levels but the states at filling factors @xmath29 are thought to be caused by electron electron interactions breaking degeneracies of the @xmath5 landau level @xcite . quantum hall incompressible states having actual fractional filling were later observed at filling factors such as @xmath30 @xcite . the fractional states follow the standard composite fermion model @xcite sequence for filling factors @xmath31 , but only even - numerator fractions are seen for @xmath32 . these sequences , and the energy gaps for the corresponding incompressible states , are thought to reflect the interplay of strongly - correlated chiral electrons and the characteristic internal symmetries of graphene that will be discussed further below @xcite . as has been seen , in the quantum hall effect for 2d electron gases produced in semiconductor devices , two fundamental classes of incompressible states are found : 1 . those where the requisite gaps are produced by complete filling of landau levels ( fermi energy lying between landau levels ) that are explained by weakly - interacting electrons subject to impurity - scattering localization . 2 . those where the gaps are produced by strong electron correlations within a partially - filled landau level . for the normal semiconductor quantum hall effect , the first case is commonly termed the integral quantum hall effect ( iqhe ) , because it leads uniquely to quantum hall plateaus at integral filling factors , and the second case is termed the fractional quantum hall effect ( fqhe ) , because it is characterized uniquely by quantum hall plateaus at fractional values of the filling factor . in graphene , the anomalous counting implied by the four - fold spin valley degeneracy for massless chiral fermions modifies this correspondence . one may again assume incompressible states divided into those that have gaps produced by weakly - interacting electrons filling landau levels completely and those that have gaps produced by correlations within a partially filled level . however ( 1 ) states of the first class occur at integral fillings but the integers of eq . ( [ iqhegraph1.2 ] ) are not those of the normal iqhe because of the anomalous counting implied by graphene s internal structure . ( 2 ) states of the second class may have either integral or fractional filling factors because of the anomalous counting . thus iqhe and fqhe may be used as shorthand labels in graphene , but it should be understood that what is meant by `` iqhe '' in graphene is the formation of incompressible states by weakly - interacting electrons completely filling landau levels , and by `` fqhe '' the formation of incompressible states by strongly - correlated electrons in a partially - filled level , irrespective of whether the observed filling factors are integers or rational fractions . our primary interest in this paper lies in those incompressible states formed by electron electron and electron lattice correlations in partially - filled landau levels , and thus in graphene `` fqhe '' states . in the normal two - dimensional electron gas produced in semiconductor devices the landau levels can contain @xmath33 states , where @xmath34 is the electronic charge and @xmath12 is the magnetic field . as has been seen , in graphene there is an additional 4-fold degeneracy associated with the spin and valley degrees of freedom . it is common to unite these four degrees of freedom in terms of an su(4 ) symmetry that is termed _ quantum hall ferromagnetism ( qhfm ) . _ su(4 ) symmetry for graphene in a strong magnetic field is expected when all four spin and valley levels are degenerate . two conditions must be satisfied for this condition to be fulfilled . 1 . landau - level mixing caused by inter - ll electronic transitions must be negligible . perturbations within a single ll that break the 4-fold spin isospin symmetry must be small . the resulting theory predicts quantum hall states that have no analog in semiconductor heterostructures . let us study these states by introducing an effective low - energy hamiltonian exhibiting approximate su(4 ) symmetry . the two - dimensional electronic spin degree of freedom and the two - dimensional electronic valley ( k ) degree of freedom are most elegantly expressed in terms of independent spin and valley isospin states . for the spin , introduce the pauli matrix vector @xmath35 , with the standard representation in terms of the @xmath36 matrices @xmath37 which obey the su(2 ) lie algebra @xmath38 } = i\epsilon_{ijk } \frac{\sigma_k}{2 } , \label{su2comm}\ ] ] where @xmath39 is the completely antisymmetric 3rd - rank tensor . the matrices ( [ algebra1.5 ] ) are assumed to operate on a spinor basis of spin - up and spin - down electrons @xmath40 a set of equations in the valley isospin space completely analogous to eqs . ( [ algebra1.5])([algebra1.6 ] ) for the spin space results if one defines the su(2 ) pauli - matrix representation for the valley isospin operators @xmath41 , @xmath42 which operate on the valley isospinor basis @xmath43 the operators @xmath44 and @xmath45 may now be used to define an effective low - energy hamiltonian having landau - level and internal spin and valley isospin degrees of freedom . the two largest energy scales for graphene in a strong magnetic field are the landau - level separation and the coulomb energy . at the charge neutral point ( fermi energy for undoped graphene ) the ll separation is approximately three times larger than the coulomb energy , which is in turn considerably larger than any additional terms in the interaction . therefore , a strategy is adopted here of ignoring excitations between landau levels and projecting onto the @xmath5 ll . at a quantitative level the landau level mixing can not be ignored ( see the discussion in ref . @xcite ) , but such an approximation gives the correct qualitative physics and the effect of excluded landau levels can be included to some degree by parameter renormalization , which will be sufficient for our purposes . within this single landau level the hamiltonian is assumed to be dominated by a long - range coulomb interaction that is su(4 ) symmetric , with shorter - range interactions in spin and valley degrees of freedom ( originating in both screened electron electron interactions and electron phonon interactions ) causing su(4 ) symmetry breaking . to implement this a graphene hamiltonian projected onto the @xmath5 landau level is adopted that was proposed in ref . @xcite and employed further in ref . @xcite,@xmath46 where the valley - independent coulomb interaction @xmath47 may be expressed as @xmath48 @xmath49 is the short - range , valley - dependent interaction , @xmath50 \delta(\bm r_i - \bm r_j ) , \label{so5_1.1c}\ ] ] where the pauli matrices @xmath51 operate on the valley isospin and @xmath52 and @xmath53 are coupling constants , and the zeeman energy @xmath54 is given by @xmath55 where @xmath56 is the bohr magneton , @xmath12 is the magnetic field strength , the pauli matrices @xmath57 operate on the electronic spin degrees of freedom , and the @xmath58 direction for the spin space is chosen to be aligned with @xmath12 . letting @xmath59 and @xmath60 , the set of 15 operators @xmath61 [ algebra1.4 ] is closed under commutation , defining an su(4 ) lie algebra that commutes with the coulomb interaction @xmath47 @xcite . thus , if @xmath49 and @xmath54 are small compared with @xmath47 in eq . ( [ so5_1.1a ] ) , the hamiltonian will have an approximate su(4 ) invariance . the operator @xmath62 represents the total spin and the operator @xmath63 represents the total valley pseudospin . in the @xmath5 landau level for graphene there is an equivalence between valley and sublattice degrees of freedom , so @xmath64 can be viewed as a nel vector in the @xmath5 landau level measuring the difference in spins on the a and b sublattices . the operators @xmath65 coupling spin and valley isospin will be discussed further below . the occurrence of su(4 ) symmetry and its explicit symmetry - breaking pattern depend on the values of the effective coupling parameters @xmath52 and @xmath53 . for the lattice spacings found in graphene , each can be estimated to be considerably smaller than the su(4)-symmetric coulomb term , so one may expect su(4 ) to be broken only weakly by explicit terms in realistic systems . four basic explicit symmetry - breaking patterns have been identified @xcite : 1 . for arbitrary non - zero values of @xmath52 and @xmath53 , the symmetry is broken to @xmath66 where @xmath67 is associated with global conservation of spin and @xmath68 with conservation of its @xmath58 component , and @xmath69 is associated with conservation of the @xmath70 component of the valley isospin . ( conservation of @xmath70 implies physically that the difference in electronic densities between the @xmath2 and @xmath3 sites is invariant , which might be expected to be true for low - energy states having minimal scattering between valleys . ) in the absence of zeeman splitting spin is conserved , but only the @xmath58 component of the valley isospin is conserved . the full hamiltonian ( [ so5_1.1a ] ) including the zeeman term conserves only the @xmath58 components of the spin and valley isospin . if @xmath71 but @xmath72 , the symmetry is broken to @xmath73 in the absence of zeeman splitting this corresponds to conserving independent spin rotations for the wavefunction in each valley labeled by @xmath2 and @xmath3 , but breaking of the valley isospin to u(1)@xmath74 . the full hamiltonian ( [ so5_1.1a ] ) including the zeeman term conserves only the @xmath58 components of the spin in each valley separately , and the @xmath58 component of the valley isospin . 3 . if @xmath75 , the symmetry is broken to @xmath76 corresponding to full spin and valley isospin rotational symmetry in the absence of zeeman splitting . the complete hamiltonian ( [ so5_1.1a ] ) including the zeeman term conserves the su(2 ) isospin symmetry but only the @xmath58 component of the spin . 4 . if @xmath77 , the hamiltonian commutes with @xmath65 , @xmath62 , and @xmath70 , and these 10 operators generate the lie group so(5 ) , so @xcite @xmath78 where the @xmath79 symmetry is generated by @xmath80 . thus , in the absence of zeeman splitting the system exhibits an so(5 ) symmetry involving both spin and valley isospin . the full hamiltonian ( [ so5_1.1a ] ) including the zeeman term conserves the @xmath58 component of spin and the @xmath79 symmetry . the so(5 ) subgroup plays the role of a transitional symmetry connecting the nel - like states associated with @xmath81 and the states associated with valley degrees of freedom @xmath82 and @xmath83 . the subgroup structure for these four patterns of explicit su(4 ) symmetry breaking is illustrated in fig . [ fig : dynamicalchains_qhf].dynamicalchains_qhf figures / dynamicalchains_qhf_vert_compact_nobkg.eps 0pt 0.40 explicit symmetry - breaking structure for su(4 ) quantum hall ferromagnetism described by the hamiltonian ( [ so5_1.1a ] ) . these symmetries and explicitly - broken symmetries have proven extremely useful in understanding the states of graphene in a strong magnetic field @xcite . let us now consider using symmetries in an even more powerful way than that discussed in the preceding section . specifically , let us attempt to describe the quantum hall ferromagnetic behavior of graphene using _ fermion dynamical symmetries _ of an effective hamiltonian operating in a highly - truncated collective subspace . this has the potential to prescribe dynamics as well as taxonomy and conservation laws , within a many - body model having analytical solutions that illuminate the physics of quantum hall states in graphene . for clarity of discussion a valence space will be assumed corresponding to a single landau level . to avoid cluttered notation the index @xmath17 labeling the landau level is suppressed and the fermion creation operator @xmath84 and the corresponding hermitian conjugate @xmath85 are introduced . the index @xmath86 takes one of two values ( @xmath87 ) labeling the valley isospin projections corresponding to valleys @xmath2 or @xmath3 , the electron spin polarization @xmath88 takes one of two values ( @xmath89 ) labeling spin - up or spin - down , and @xmath90 is a quantum number distinguishing degenerate states within a given landau level ( typically an angular momentum in symmetric gauge or a linear momentum in landau gauge ) . the fermionic operators @xmath91 and @xmath92 are assumed to obey @xmath93 where @xmath94 . by virtue of this anticommutator , the @xmath95 create a fermion in the state labeled by @xmath96 , the @xmath97 annihilate a fermion in the same state , and @xmath98 counts the number of fermions in the state labeled by @xmath96 . the four states representing possible combinations of @xmath86 and @xmath88 are displayed in table [ tb : quantumnumbermapping ] , ccccc ' '' '' valley & @xmath86 & @xmath88 & @xmath99 & @xmath100 + ' '' '' @xmath2 & @xmath19 & @xmath101 & @xmath102 & @xmath103 + [ ] @xmath2 & @xmath19 & @xmath104 & @xmath105 & @xmath106 + [ ] @xmath3 & @xmath20 & @xmath101 & @xmath107 & @xmath108 + [ ] @xmath3 & @xmath20 & @xmath104 & @xmath109 & @xmath110 + and correspond physically to the four possible combinations of the electron being in either the @xmath2 or @xmath3 valley with either spin up or spin down ( see fig . [ fig : llgraphene_n0]b ) . for brevity the label @xmath111 displayed in the last column of the table will often be used to distinguish these states . the four basis states labeled by @xmath100 are illustrated graphically in fig . [ fig : graphenebasis ] . figures / graphenebasis.eps 0pt 0.16 the four isospin spin basis vectors @xmath112 of table [ tb : quantumnumbermapping ] . table [ tb : quantumnumbermapping ] also displays a unique mapping of these four states to a label @xmath99 that takes the four possible projection quantum numbers @xmath113 of a fictitious angular momentum @xmath114 ; the motivation for this mapping will become apparent later . let us now introduce an operator @xmath115 that creates a pair of electrons , one in the @xmath116 level and one in the @xmath117 level , with the total @xmath90 of each pair coupled to zero term by term , @xmath118 and its hermitian conjugate @xmath119 , which annihilates a corresponding electron pair . each index @xmath100 or @xmath120 ranges over four values , implying 16 components in eq . ( [ algebra1.1 ] ) . however , the pair wavefunction must satisfy the pauli principle which , upon expanding the indices @xmath100 and @xmath120 using table [ tb : quantumnumbermapping ] , eliminates the four diagonal @xmath121 possibilities . furthermore , because of the antisymmetry requirement the pair creation operators are constrained by @xmath122 implying that only half of the remaining 12 operators @xmath123 are independent . thus eq . ( [ algebra1.1 ] ) defines six independent operators @xmath123 , with six independent hermitian conjugates @xmath124 . let us introduce in addition to these pairing operators the 16 particle hole operators @xmath125 through @xmath126 where @xmath127 is the kronecker delta and @xmath128 is the total degeneracy the single landau level ( see eq . ( [ degen1.2 ] ) ) . the commutators for the 28 operators @xmath124 , @xmath123 , and @xmath12 are found to be @xcite @xmath129 } & = -b_{db } \delta_{ac } - b_{ca}\delta_{bd } + b_{cb } \delta_{ad } + b_{da } \delta_{bc } \label{algebra1.3a } \\ { \left[b_{ab } , b_{cd}\right ] } & = \delta_{bc } b_{ad } - \delta_{ad}b_{cb } \label{algebra1.3b } \\ { \left[b_{ab } , a^\dagger_{cd}\right ] } & = \delta_{bc } a^\dagger_{ad } + \delta_{bd } a^\dagger_{ca } \label{algebra1.3c } \\ { \left[b_{ab } , a_{cd}\right ] } & = -\delta_{ac } a_{bd } - \delta_{ad } a _ { cb } , \label{algebra1.3d}\end{aligned}\ ] ] [ algebra1.3 ] which is isomorphic to an so(8 ) lie algebra . thus , the 28 members of the operator set @xmath130 exhibit so(8 ) symmetry under commutation . the ultimate justification for introducing this so(8 ) lie algebra will lie in the results to which it will lead . however , a general discussion of why the known physics of graphene suggests the efficacy of such an algebra in understanding the possible collective modes of the system may be found in appendix [ h : appendixa ] . the su(4 ) generators ( [ algebra1.4 ] ) may be expressed in terms of the operators ( [ algebra1.2 ] ) by employing eqs . ( [ algebra1.5])([algebra1.7e ] ) in eq . ( [ algebra1.4 ] ) and utilizing the equivalences in table [ tb : quantumnumbermapping ] to expand the indices for @xmath125 from eq . ( [ algebra1.2 ] ) . for example , consider the spin operator @xmath131 . from eq . ( [ algebra1.4a ] ) one may write @xmath132 & \left . \qquad\qquad\qquad -i c^\dagger_{-\uparrow m_k } c_{-\downarrow m_k}{^{\vphantom{\dagger}}}+ i c^\dagger_{-\downarrow m_k } c_{-\uparrow m_k}{^{\vphantom{\dagger}}}\right ) \\ & = -ib_{12 } + i b_{21 } -i b_{34 } + i b_{43}.\end{aligned}\ ] ] by such methods one finds that the spin operators ( [ algebra1.4a ] ) may be expressed in terms of the @xmath125 generators of so(8 ) as @xmath133 [ algebra1.8 ] the valley isospin operators ( [ algebra1.4b ] ) as @xmath134 [ algebra1.9 ] the nel vector of eq . ( [ algebra1.4c ] ) as @xmath135 [ algebra1.10 ] and the operators @xmath65 of eq . ( [ algebra1.4d ] ) as @xmath136 [ algebra1.11 ] the inverse transformations expressing the @xmath125 in terms of the @xmath137 are given in appendix [ h : appendixb ] . hence the su(4 ) algebra generated by the operators in eq . ( [ algebra1.4 ] ) is a subalgebra of the so(8 ) algebra , with its generators corresponding to particular linear combinations of the subset of so(8 ) generators defined by the particle hole operators @xmath125 in eq . ( [ algebra1.2 ] ) . in fact , the present formalism also could be constructed by starting with the su(4 ) algebra generated by the @xmath125 operators and expanding that to the so(8 ) algebra of eq . ( [ algebra1.3 ] ) by adding pairing operators , using the motivation discussed in appendix [ h : appendixa ] . one wishes to investigate possible collective modes for graphene electrons undergoing strong correlations within a single landau level . pairs of fermions often afford a convenient basis for discussing collective states , so let us consider some possible configurations involving pairs of electrons in graphene . let us first consider the degeneracies of undoped graphene placed in a strong magnetic field , confining the discussion to the case of a single landau level for simplicity . the single - particle states within the landau level will be assumed labeled by the quantum numbers @xmath138 , where @xmath17 is the principle quantum number labeling the landau level and @xmath139 is a quantum number distinguishing the degenerate states within the landau level . in the absence of spin and valley degrees of freedom , the states @xmath138 of the landau level are assumed to hold a maximum of @xmath140 electrons ( for consistency with earlier applications of the so(8 ) algebra in nuclear physics , @xmath141 will be defined to be the electron pair degeneracy , so that @xmath140 is the electron degeneracy ) . from the solution of the dirac equation in a magnetic field @xmath142 where @xmath12 is the strength of the magnetic field , @xmath143 is the area of the two - dimensional sample and @xmath144 defines the magnetic flux quantum . but graphene has in addition @xmath145 internal degrees of freedom associated with the @xmath146 space . thus there are four copies of each landau level in graphene and the total electron degeneracy @xmath147 is given by @xmath148 some pair degeneracies calculated from eq . ( [ degen1.2 ] ) as a function of domain size and magnetic field strength are displayed in table [ tb : degeneracytable ] , cccc ' '' '' domain size & @xmath149 & @xmath150 & @xmath151 + ' '' '' @xmath152 & @xmath153 & @xmath154 & @xmath155 + [ ] @xmath156 & @xmath157 & @xmath158 & @xmath159 + where it is assumed that the collective wavefunction is delocalized over the entire domain size . the _ fractional occupation _ @xmath160 of the single landau level [ not to be confused with the filling factor @xmath9 given in eqs . ( [ iqhegraph1.1 ] ) and ( [ fillingfactor ] ) ] may be defined as @xmath161 where @xmath17 is the number of electrons and @xmath162 is the number of electron pairs . for half - filling of the @xmath5 landau level located at the fermi surface ( corresponding to the ground state of undoped graphene ) the electron number @xmath163 is then @xmath164 these degeneracies and occupation numbers are just the standard results for relativistic landau levels in a 2d electron gas subject to a strong perpendicular magnetic field , but modified by the graphene spin and valley degeneracies . graphene exhibits both integral and fractional quantum hall effects but the filling factors are anomalous relative to those for standard quantum hall effects in semiconductor heterostructures . this is because of 1 . the 4-fold degeneracy associated with the spin and valley degrees of freedom , which introduces factors of four in the counting . 2 . the nature of the dirac solution , illustrated in fig . [ fig : llgraphene_n0 ] , for which the negative - energy solutions may be interpreted as electron holes , the positive - energy solutions as electrons , and the @xmath5 level is unique , being half - filled in the neutral ground state ( equivalently , it may be thought of as being shared equally by particles and holes ) . because of the particle - hole symmetry , the charge carriers change sign near the dirac points and the hall conductivity vanishes at charge neutrality ( the electron number density tends to zero at a dirac point ) . for this reason , the filling factor for graphene must be defined _ relative to the charge - neutral state . _ at charge neutrality the @xmath5 landau level is half filled and when the @xmath5 ll is completely full the filling factor is @xmath165 , from eq . ( [ iqhegraph1.2 ] ) . the quantum hall filling factor @xmath9 may be related to the landau level fractional occupation @xmath160 employed in the present formalism by @xmath166 therefore , half filling of the @xmath5 landau level corresponds to a fractional occupation @xmath167 but to a filling factor @xmath4 , @xmath168 corresponds to @xmath169 ( completely empty ) , @xmath170 corresponds to @xmath171 filling , @xmath172 corresponds to @xmath173 filling , and @xmath174 corresponds to @xmath175 ( completely full ) . consider the states created by repeated application of the pair creation operator @xmath115 defined by eq . ( [ algebra1.1 ] ) to the pair vacuum . it is useful to classify states according to a seniority - like quantum number @xmath176 defined to be the number of particles in the system not coupled to one of the pairs defined in eq . ( [ algebra1.1 ] ) . the @xmath177 subspace will be of particular interest since it will contain states of maximal collectivity with respect to the pairs ( [ algebra1.1 ] ) . an @xmath178-pair state in the @xmath177 subspace is given by @xmath179 where the total pair number @xmath178 is @xmath180 with @xmath17 giving the total number of electrons and @xmath181 giving the number of electron pairs created by @xmath182 operating on the vacuum state . for our discussion here it will always be assumed that one is dealing with @xmath177 states , corresponding physically to no broken pairs . from eq . ( [ algebra1.3 ] ) , the 16 operators @xmath125 are closed under commutation and form a @xmath184 subalgebra of the so(8 ) algebra ( [ algebra1.3 ] ) . let us now investigate the irreducible representations ( irreps ) that are associated with the @xmath185 subgroup chain in the @xmath177 representations . for @xmath177 at half filling , the number of pairs is @xmath186 and the highest - weight u(4 ) representation is given by @xmath187 . let us define a highest - weight ( hw ) state in the @xmath177 space , and choose it to correspond to the pair state with maximal value of @xmath99 from table [ tb : quantumnumbermapping ] , which results from placing one electron in the @xmath188 state and one electron in the @xmath189 state . thus , for @xmath190 pairs the highest weight state is given by @xmath191 where the sum runs over the @xmath192 states in the landau level labeled by the @xmath193 quantum number . the other states of the irreducible representation may then be created by the cartan dynkin algorithm , which consists of using raising and lowering operators in the weight space to construct successively all the other states beginning with the highest weight state @xcite . the state in eq . ( [ hwstate ] ) appears to have a quite complex form , involving a sum with number of terms equal to the pair degeneracy of the landau level raised to a power equal to the pair degeneracy ( with the pair degeneracy typically a large number ) . however , the actual structure of this state is considerably simpler than eq . ( [ hwstate ] ) would suggest because of the pauli principle . as an illustrative example of this assertion , let s construct explicitly the highest - weight state for the case @xmath194 , corresponding to @xmath195 pairs in a single landau level . writing the sum over @xmath196 in eq . ( [ hwstate ] ) out term by term gives @xmath197 where in raising the sum of operators inside the parentheses to the @xmath198 power , all products containing two or more creation operators with the same index vanish because of the pauli principle . similar considerations apply for arbitrary values of @xmath199 and in general the highest - weight state is given by @xmath200 where the simplification in going from the second to the third line is a consequence of the antisymmetry of the fermion creation operators ( the pauli principle ) implied by eq . ( [ anticommutator ] ) . thus the highest - weight state is a product state of pairs , one pair for each of the @xmath190 levels labeled by @xmath90 in the landau level . by the cartan dynkin algorithm , other states in the @xmath177 subspace can be constructed by applying successively to the highest - weight state appropriate lowering and raising operators . these will be functions of the generators @xmath125 , so for an arbitrary state @xmath201 in the weight space one has schematically @xmath202 where the function @xmath203 is specified by the cartan dynkin procedure . as an example , consider the action of the valley isospin lowering operator @xmath204 on the highest - weight state . from eqs . ( [ algebra1.9 ] ) and ( [ algebra1.2 ] ) , @xmath205 thus the state @xmath201 created by applying @xmath204 to @xmath206 is @xmath207 c_{{\scriptscriptstyle k\uparrow } m_k}^\dagger c_{{\scriptscriptstyle k\downarrow } m_k}^\dagger \ , { |0\rangle } \nonumber \\ & = \prod_{m_k } \left ( c_{3 m_k}^\dagger c_{2 m_k}^\dagger + c_{4 m_k}^\dagger c_{1 m_k}^\dagger \right ) { |0\rangle } , \label{u0pairs1.7}\end{aligned}\ ] ] where the simplifications are because the only terms that survive correspond to those where an annihilation operator in a factor inside the square brackets is exactly balanced by a creation operator from the factor outside the square brackets . likewise , the other states of the @xmath177 representation can be constructed by using successive applications of raising and lowering operators fashioned from the generators defined in in eqs . ( [ algebra1.8])([algebra1.11 ] ) . from the preceding discussion , for @xmath208 the states may be written as @xmath209 { |0\rangle } , \label{upairs1.8}\end{aligned}\ ] ] where @xmath210 denote valley isospin projection quantum numbers and @xmath211 denote spin projection quantum numbers . this is the same form as the most general collective pair state used by kharitonov @xcite in his classification of possible broken symmetry states for the @xmath5 landau level in graphene ( see eq . ( [ basis1.0 ] ) in the appendix ) . thus , for undoped graphene the general pairing wavefunction ( [ u0pairs1.0 ] ) characteristic of the so(8 ) @xmath183 su(4 ) dynamical symmetry is in fact _ equivalent to the product form ( [ upairs1.8 ] ) _ employed in standard discussion of quantum hall ferromagnetism , for which the summations are over the internal @xmath212 rather than landau @xmath213 degrees of freedom . the equivalence of eqs . ( [ upairs1.8 ] ) and ( [ u0pairs1.0 ] ) , despite their superficially very different forms , is a fundamental consequence of the pauli principle acting in the collective fermionic pair subspace , which greatly restricts allowed pair configurations . the equivalence established above implies that the present so(8 ) formalism can be used to derive the framework used in ref . @xcite to describe possible collective states in graphene , which establishes an essential connection between the standard discussion of su(4 ) quantum hall magnetism in graphene and the present more general so(8 ) dynamical symmetry formalism . as will be demonstrated further below , this permits the so(8 ) formalism to ( 1 ) encompass the established physics of su(4 ) quantum hall ferromagnetism , ( 2 ) extend quantum hall ferromagnetism to suggest possible additional collective graphene physics beyond the su(4 ) limit , and ( 3 ) provide _ analytical solutions _ for the states suggested by spontaneous breaking of su(4 ) symmetry that can not be obtained using su(4 ) symmetry and must be addressed numerically , or with effective field theory approximations , within standard su(4 ) quantum hall ferromagnetism . furthermore , it will be seen below that the equivalence of product and paired forms for the wavefunction implies a deep formal connection between the collective states resulting from strong electron correlations in graphene landau levels and the collective states produced by strong nucleon correlations in nuclear structure physics , and a suggestive formal connection to the properties of the strongly - correlated electrons responsible for high - temperature superconductivity . the preceding discussion has established that the fermion dynamical symmetry method applied to undoped graphene in a strong magnetic field has one dynamical symmetry chain @xmath185 that recovers exactly su(4)-symmetric quantum hall ferromagnetism . since the @xmath125 operators introduced in eq . ( [ algebra1.2 ] ) form an su(4 ) subgroup of so(8 ) that is in one - to - one correspondence with the operators used to formulate the effective low - energy hamiltonian ( [ so5_1.1a ] ) , [ h : fdsgraphene ] implies that all of the physics associated with this effective hamiltonian that has been discussed in the prior literature ( see @xcite and references cited therein ) is implicit in the present formalism . furthermore , the discussion of [ h : pairrepresentations ] shows that the pair basis ( [ u0pairs1.0 ] ) of the truncated collective subspace for the so(8 ) fermion dynamical symmetry is in fact identical to the most general wavefunction ( [ upairs1.8 ] ) that has been proposed @xcite for collective states breaking the su(4 ) symmetry , despite its superficially very different form . thus the hilbert - space truncation implied by the collectively - paired so(8 ) subspace ( [ u0pairs1.0 ] ) recovers the understanding in the existing literature of the classes of states to be expected from spontaneous breaking of the su(4 ) symmetry by valley - dependent correlations . however , the existing discussions of these collective states in terms of broken su(4 ) symmetry have been largely qualitative , and have turned to numerical simulations to discuss the actual structure and energy of the states . it will now be demonstrated that the present formalism is capable not only of classifying , but also of addressing the quantitative nature of those collective states _ in analytical fashion . _ furthermore , it will be shown that the so(8 ) highest symmetry implies subgroup chains in addition to @xmath214 that are associated with spontaneous breaking of the symmetry by correlations and have not been discussed in the previous literature and that may play a role in graphene . let us begin that discussion by first transforming to a more convenient representation of the so(8 ) generators . this new representation will be physically equivalent to the original representation , but will offer some advantages in interpretation , and will expose an unexpected relationship between graphene physics and that of a very different field , nuclear structure physics . for the pair creation operators defined in eq . ( [ algebra1.1 ] ) , each electron creation operator @xmath91 carries both spin and valley isospin ; hence the products @xmath215 correspond to a clebsch gordan series representing sums of terms having different values of total spin and total isospin . likewise , in the particle hole operators of eq . ( [ algebra1.2 ] ) each creation operator @xmath91 and annihilation operator @xmath92 carries spin and isospin , so the product @xmath216 in eq . ( [ algebra1.2 ] ) represents a superposition of states carrying different total spin and total valley isospin . these representations with indefinite spin and isospin will be termed _ uncoupled representations . _ on physical grounds , the spin is expected to be conserved ( if the zeeman term in the hamiltonian is neglected ) and the valley isospin is expected to be approximately conserved for low - energy excitations . thus , it is desirable to use the uncoupled representation of the pairing and particle hole operators to construct new _ coupled representations _ that have good total spin and good total valley isospin quantum numbers for bilinear operators . using standard angular momentum coupling theory @xcite , an electron pair creation operator coupled to good spin and valley isospin may be defined by @xmath217 where @xmath143 is the total spin of the pair with @xmath218 its projection , @xmath219 is the total valley isospin of the pair with @xmath220 its projection , and @xmath221 are clebsch gordan coefficients for the angular momentum sum @xmath222 that couple the pair to good total spin or total valley isospin . antisymmetry implies that the the pair wavefunction can have only @xmath223 , @xmath224 ; or @xmath225 , @xmath226 ( spin - triplet , isospin - singlet ; or spin - singlet , isospin - triplet pairs ) . explicitly the possibilities are @xmath227 { a_{\scriptscriptstyle 00}^{\raisebox{0.2em}{$\scriptstyle\dagger$ } { \scriptscriptstyle 0 1 } } } = a_{14}^\dagger + a_{23}^\dagger \quad { a_{\scriptscriptstyle 01}^{\raisebox{0.2em}{$\scriptstyle\dagger$ } { \scriptscriptstyle 0 1 } } } = \sqrt 2 a_{13}^\dagger \qquad { a_{\scriptscriptstyle 0 - 1}^{\raisebox{0.2em}{$\scriptstyle\dagger$ } { \scriptscriptstyle 0 1 } } } = \sqrt2 a_{24}^\dagger , \end{array } \label{coupled1.3}\ ] ] with the hermitian conjugates of eq . ( [ coupled1.3 ] ) giving the six corresponding pair annihilation operators in coupled representation . it is useful to define an alternative set of six coupled pairing operators @xmath228 and @xmath229 according to @xmath230 d_0^\dagger = \tfrac{1}{\sqrt2 } \,{a_{\scriptscriptstyle 00}^{\raisebox{0.2em}{$\scriptstyle\dagger$ } { \scriptscriptstyle 0 1 } } } = \tfrac{1}{\sqrt2 } \left ( a_{14}^\dagger + a_{23}^\dagger \right ) \\[7pt ] d_1^\dagger = \tfrac{1}{\sqrt2 } \ , { a_{\scriptscriptstyle 01}^{\raisebox{0.2em}{$\scriptstyle\dagger$ } { \scriptscriptstyle 0 1 } } } = a_{13}^\dagger \quad d_{-1}^\dagger = \tfrac{1}{\sqrt2 } \,{a_{\scriptscriptstyle 0 - 1}^{\raisebox{0.2em}{$\scriptstyle\dagger$ } { \scriptscriptstyle 0 1 } } } = a_{24}^\dagger \\[7pt ] d_2^\dagger = \tfrac{1}{\sqrt2 } \ , { a_{\scriptscriptstyle 10}^{\raisebox{0.2em}{$\scriptstyle\dagger$ } { \scriptscriptstyle 1 0 } } } = a^\dagger_{12 } \quad d_{-2}^\dagger = \tfrac{1}{\sqrt2 } \,{a_{\scriptscriptstyle -10}^{\raisebox{0.2em}{$\scriptstyle\dagger$ } { \scriptscriptstyle 1 0 } } } = a_{34}^\dagger \end{array } \label{coupled1.4}\ ] ] and the six corresponding hermitian conjugates @xmath143 and @xmath231 . the physical meaning of these pairs may be deduced by constructing the corresponding electronic configurations . consider @xmath232 . from eqs . ( [ coupled1.4 ] ) and ( [ algebra1.1 ] ) @xmath233 where in the last step the correspondence between the index @xmath234 and the valley ( @xmath2 or @xmath3 ) and spin ( @xmath235 ) labels in table [ tb : quantumnumbermapping ] has been invoked . this implies that @xmath236 creates a state with one spin - up and one spin - down electron on each equivalent site @xmath2 in the brillouin zone , as illustrated schematically in fig . [ fig : actionddagger2 ] . figures / actionddagger2.eps 0pt 0.35 the action of the pair creation operator @xmath237 on the vacuum state is to create a charge density wave with a spin - singlet pair on each site @xmath2 and no electrons on the @xmath3 sites . this is a component of a lattice - scale charge density wave , since the charge differs by two electronic units between adjacent sites . likewise , one finds that @xmath238 creates a charge density wave as in fig . [ fig : actionddagger2 ] , but with the spin - singlet pairs on the @xmath15 sites . the pair configurations produced by all generators of eq . ( [ coupled1.4 ] ) operating on the pair vacuum @xmath239 are summarized in fig . [ fig : s_d_pairs_brillouin ] . figures / s_d_pairs_brillouin_withq.eps 0pt 0.30 configurations created by the @xmath228 and @xmath240 pair creation operators of eq . ( [ coupled1.4 ] ) , and for the linear combinations @xmath241 of eq . ( [ qdef ] ) , operating on the pair vacuum @xmath239 . the upper left box illustrates the valley isospin labeling . a total of six valence electrons as assumed to be distributed on sites in the brillouin zone . in each pair configuration , location of the dots ( @xmath2 or @xmath3 site ) indicates the valley isospin , spin - up electrons are indicated by up arrows , and spin - down electrons are indicated by down arrows . also shown are the configurations generated by the linear combinations @xmath242 which will be useful in later discussion . kharitonov has given a general classification of low - lying collective modes for the @xmath5 landau level of graphene in terms of collective pairs @xcite . the collective pairs created by the so(8 ) pair generators in fig . [ fig : s_d_pairs_brillouin ] are similar physically to the pairs identified by kharitonov , as will now be described . ( 1 ) the configuration generated by @xmath243 is to the difference of two terms , each with alternating spin - up and spin - down on adjacent sites , implying that all spins on the a sublattice ( identified with valley @xmath2 ) point in one direction and all spins on the b sublattice ( identified with valley @xmath15 ) point in the opposite direction . each term corresponds to a spin density wave ( af order ) , with a nel vector defined by the difference in total spins on the two sublattices serving as an order parameter , but because of the difference of the two terms the net af order for this configuration is zero ( see [ sh : orderparameters ] and table [ tb : orderparms ] ) . ( 2 ) the configurations generated by @xmath244 are spin - singlet charge density waves , with alternating charges of two and zero units on adjacent sites . an appropriate order parameter is the difference in charge between the a and b sublattices . ( 3 ) the configurations generated by @xmath245 have one spin on each site , all pointing in the same direction ; this is a ferromagnetic state , with the net spin as an order parameter . ( 4 ) the configuration generated by @xmath246 is the same as that generated by @xmath243 , except for a positive sign for the second term . this also implies alternating spins on adjacent sites and af order for each term , but the total af order vanishes because of the contribution of the two terms . ( 5 ) the configurations corresponding to @xmath247 are states with af order characterized by the difference in spins on the two sublattices labeled by @xmath2 and @xmath3 . thus the coupled - representation pairs carrying good spin and valley isospin quantum numbers in eq . ( [ coupled1.4 ] ) represent physical degrees of freedom already discussed in the literature as candidate collective modes representing spontaneous breaking of the su(4 ) graphene symmetry by interactions in a single partially - filled landau level . it is desirable to express the particle hole generators of eq . ( [ algebra1.2 ] ) in coupled representation . let us begin by introducing a set of operators @xmath248 with the definition @xmath249 where @xmath250 and @xmath251 take the values of the fictitious angular momentum projection @xmath99 in table [ tb : quantumnumbermapping ] , providing a labeling equivalent to that of @xmath100 and @xmath120 in @xmath125 , with @xmath250 or @xmath251 values @xmath252 mapping to @xmath100 or @xmath120 values @xmath253 , respectively . for example , from table [ tb : quantumnumbermapping ] , @xmath254 and @xmath255 label the same quantity , which is defined in eq . ( [ algebra1.2 ] ) . from the selection rules for angular momentum coupling in eq . ( [ multipole1.1 ] ) , the index @xmath256 can take the values @xmath257 , with @xmath258 projections @xmath259 for each possibility , which gives a total of 16 operators @xmath260 . by inserting the explicit values of the clebsch gordan coefficients the 16 independent @xmath261 may be evaluated in terms of the @xmath125 . @xmath262 p^1_0 & = \sqrt{\tfrac{9}{20}}(b_{11 } - b_{44 } ) + \sqrt{\tfrac{1}{20}}(b_{22 } -b_{33 } ) \nonumber \\ & = \sqrt{\tfrac{9}{20}}(n_1 - n_4 ) + \sqrt{\tfrac{1}{20}}(n_2 -n_3 ) \nonumber \\[2pt ] p^1_1 & = -\sqrt{\tfrac{3}{10 } } \,b_{12 } - \sqrt{\tfrac{4}{10 } } \,b_{23 } -\sqrt{\tfrac{3}{10}}\ , b_{34 } \nonumber \\[2pt ] p^1_{-1 } & = \sqrt{\tfrac{3}{10 } } \,b_{21 } + \sqrt{\tfrac{4}{10 } } \,b_{32 } + \sqrt{\tfrac{3}{10}}\ , b_{43 } \nonumber \\[2pt ] p^2_0 & = \tfrac{1}{2 } ( b_{11 } - b_{22 } + b_{44 } - b_{33 } ) \nonumber \\ & = \tfrac12(n_1 - n_2 + n_4 - n_3 ) \nonumber \\[2pt ] p^2_1 & = \tfrac{1}{\sqrt{2 } } ( b_{34 } - b_{12 } ) \qquad p^2_{-1 } = \tfrac{1}{\sqrt2 } ( b_{21 } - b_{43 } ) \label{pofb } \\[2pt ] p^2_2 & = -\tfrac{1}{\sqrt{2 } } ( b_{13 } + b_{24 } ) \qquad p^2_{-2 } = -\tfrac{1}{\sqrt2 } ( b_{31 } + b_{42 } ) \nonumber \\[2pt ] p^3_0 & = \sqrt{\tfrac{1}{20 } } ( b_{11 } - b_{44 } ) + \sqrt{\tfrac{9}{20 } } ( b_{33}-b_{22 } ) \nonumber \\ & = \sqrt{\tfrac{1}{20 } } ( n_1 - n_4 ) - \sqrt{\tfrac{9}{20 } } ( n_2 - n_3 ) \nonumber \\[2pt ] p^3_1 & = -\sqrt{\tfrac15 } b_{12 } + \sqrt{\tfrac35 } b_{23 } - \sqrt{\tfrac15 } b_{34 } \nonumber \\[2pt ] p^3_{-1 } & = \sqrt{\tfrac15 } b_{21 } - \sqrt{\tfrac35 } b_{32 } + \sqrt{\tfrac15 } b_{43 } \nonumber \\[2pt ] p^3_2 & = \sqrt{\tfrac12 } ( b_{24 } - b_{13 } ) \qquad p^3_{-2}= \sqrt{\tfrac12 } ( b_{42 } - b_{31 } ) \nonumber \\[2pt ] p^3_3 & = -b_{14 } \qquad p^3_{-3 } = b_{41 } , \nonumber\end{aligned}\ ] ] where the @xmath263 are number operators for each of the four states and the total particle number @xmath17 is the sum over the four states labeled by @xmath100 in table [ tb : quantumnumbermapping ] , @xmath264 it will be convenient notationally to sometimes replace the operator @xmath265 with the operator @xmath266 , according to the relationship @xmath267 where @xmath147 is the degeneracy of the space for the particles that participate in the so(8 ) symmetry . physically @xmath268 is one half the particle number measured from half filling ( which corresponds to @xmath269 ) . because the six operators defined by eq . ( [ coupled1.4 ] ) , their six hermitian conjugates , and the 16 operators defined by eq . ( [ multipole1.1 ] ) are independent linear combinations of the so(8 ) generators defined in eqs . ( [ algebra1.1 ] ) and ( [ algebra1.2 ] ) , the 28 operators @xmath270 also close an so(8 ) algebra under commutation . the so(8 ) commutation relations for the coupled representation @xmath271 are given explicitly by @xcite @xmath272 } & = -2s_0 \label{commcoupled1.1 } \\ { \left[d_{\mu ' } , d^\dagger_{\mu}\right ] } & = -2\delta_{\mu\mu ' } s_0 + \sum_{t { \rm\ odd } } ( -1)^{\mu ' } \nonumber \\ & \qquad \times \ { c_{2 , \scriptscriptstyle-\mu ' 2 \mu}^{\,t,\scriptscriptstyle\mu-\mu ' } } { \left\ { \begin{matrix } 2&2&t\\#4&\tfrac32&\tfrac32 \end{matrix}\right\ } } p_{\scriptscriptstyle\mu,-\mu'}^t \label{commcoupled1.2 } \\ { \left[d_\mu^\dagger , s\right ] } & = p_\mu^2 \label{commcoupled1.3 } \\ { \left[p^r_\mu , s^\dagger\right ] } & = 2 \delta_{r2}d^\dagger_\mu + 2 \delta_{r0 } \delta_{\mu0 } s^\dagger \label{commcoupled1.4 } \\ { \left[p_{\mu'}^r , d_\mu^\dagger\right ] } & = 2(-1)^{\mu ' } \delta_{r2 } \delta_{-\mu\mu ' } -4\sqrt{5(2r+1 ) } \nonumber \\ & \qquad\times\ { c_{r \scriptscriptstyle\mu ' 2 \mu}^{2,\scriptscriptstyle\mu+\mu ' } } { \left\ { \begin{matrix } 2&2&r\\#4&\tfrac32&\tfrac32 \end{matrix}\right\ } } d_{\scriptscriptstyle\mu+\mu'}^\dagger \label{commcoupled1.5 } \\ { \left[p^r_{\mu ' } , p^s_\mu\right ] } & = 2 ( -1)^{r+s } \sqrt{(2r+1)(2s+1 ) } \,\,\sum_t { c_{r { \scriptscriptstyle\mu ' } s { \scriptscriptstyle\mu}}^{\,t , { \scriptscriptstyle \mu+\mu ' } } } \nonumber \\ & \times \left[1-(-1)^{r+s+t}\right ] { \left\ { \begin{matrix } r&s&t\\#4&\tfrac32&\tfrac32 \end{matrix}\right\ } } p^{\,t}_{\scriptscriptstyle\mu+\mu ' } \label{commcoupled1.6}\end{aligned}\ ] ] [ commcoupled ] where @xmath266 is defined in eq . ( [ p0n ] ) and @xmath273 denotes the wigner 6-@xmath274 symbol @xcite for the recoupling of three angular momenta to good total angular momentum . the action of the so(8 ) pair creation operators on the pair vacuum @xmath178 times creates a @xmath275-particle state @xcite , @xmath276 where the total number of pairs is @xmath277 . the portion of the full hilbert space that is spanned by the states ( [ collective1.1 ] ) will be termed the _ collective subspace . _ it will play an important role in subsequent discussion where it will be shown that the so(8 ) symmetry may be used to construct effective hamiltonians that are diagonal in this space , and that the generators of so(8 ) do not couple the subspace to the remainder of the space . the reason for our alternative labeling of the states in table [ tb : quantumnumbermapping ] in terms of the index @xmath99 , and our particular choices of phases and normalizations in equations , can now be made clear . with these labelings and choices the six coupled particle particle operators @xmath228 and @xmath278 defined in eq . ( [ coupled1.4 ] ) , their six hermitian conjugates @xmath143 and @xmath231 , and the 16 coupled particle hole operators @xmath279 defined in eq . ( [ multipole1.1 ] ) , are mathematically in one - to - one correspondence with the 28 generators for the ginocchio so(8 ) model @xcite and the so(8 ) fermion dynamical symmetry model @xcite . these have found broad application in nuclear structure physics @xcite and may be viewed as a microscopic justification for the interacting boson model ( ibm ) @xcite , which is one of the most commonly used phenomenological models in nuclear structure physics . this correspondence has three important implications : ( 1 ) mathematically , the group - theoretical methodology obtained for so(8 ) already in nuclear physics applications may be appropriated for use in the graphene problem . ( 2 ) physically the nature of the generators for the nuclear physics and graphene so(8 ) symmetries are fundamentally different , but analogs of physical interpretations applied already for nuclear physics so(8 ) symmetries may shed light on the graphene problem . ( 3 ) philosophically , the so(8 ) correspondence between graphene and nuclear structure physics implies a satisfying convergence of mathematical reasoning and physical abstraction in two completely different scientific subfields . this convergence will be elaborated further in [ h : analogy_su4 ] . the so(8 ) group has various subgroups ( subsets of generators closed under commutation ) and these in turn may have other subgroups . these sequences of subgroups define subgroup chains . these chains will be discussed first in terms of the nuclear physics basis @xmath280 , and then in terms of a new basis that is mathematically equivalent but is physically better suited to describing the physics of graphene . in the nuclear physics basis @xmath281 is the particle number and generates a group @xmath282 , while @xmath283 is proportional to the total angular momentum and generates a group @xmath284 . in the nuclear physics context the total angular momentum and the particle number are expected to be conserved exactly for all physical states . thus one seeks subgroup chains of so(8 ) that end in the subgroup @xmath285 corresponding to charge and angular momentum conservation . three so(8 ) subgroup chains satisfy these conditions . from eq . ( [ commcoupled1.1 ] ) the quasispin generators @xmath288 close an @xmath289 algebra that is a subalgebra of so(8 ) , and from eq . ( [ commcoupled1.4 ] ) the operators @xmath260 with @xmath290 close an so(5 ) algebra and commute with these su(2 ) quasispin generators . thus , one subgroup of so(8 ) is @xmath291 where so(5 ) is generated by the 10 operators @xmath292 , where @xmath259 takes the @xmath258 values @xmath293 and the _ quasispin group _ @xmath289 is generated by @xmath294 , where @xmath295 . furthermore , the three generators @xmath296 are components of the total angular momentum @xmath297 and generate an @xmath284 subgroup of so(5 ) , and @xmath266 generates a @xmath282 subgroup of @xmath289 corresponding to conservation of charge . thus one subgroup chain is @xmath298 & \supset \underset{\scriptscriptstyle\ { p^1\ } } { { { \rm so(3)}}_l}\times\underset{\scriptscriptstyle \ { s_0 \ } } { { { { \rm u(1)}}_{{\scriptstyle\rm c } } } } , \label{chains1.2n}\end{aligned}\ ] ] where the generators of each subgroup are indicated in brackets below the subgroup and the product group on the last line corresponds to conservation of total angular momentum and particle number . the groups su(4 ) and so(6 ) share the same lie algebra . in nuclear physics it is more common to refer to this group as so(6 ) , but to maintain a parallel with the ensuing treatment of graphene it will be labeled su(4 ) in the present discussion . from eq . ( [ commcoupled1.6 ] ) , the 16 operators @xmath300 are closed under commutation , corresponding to @xmath301 where the generator of @xmath282 is @xmath302 and the 15 operators @xmath303 are the generators of su(4 ) . furthermore , the subset of @xmath304 with odd @xmath256 are generators of the so(5 ) symmetry discussed above and so generate an so(5 ) subgroup of this su(4 ) group . hence a second subgroup chain is @xmath305 & \supset \underset{\scriptscriptstyle \ { p^1 , \ , p^3\}}{{{\rm so(5 ) } } } \times \underset{\scriptscriptstyle \ { s_0\}}{{{{\rm u(1)}}_{{\scriptstyle\rm c } } } } \supset \underset{\scriptscriptstyle\ { p^1\ } } { { { \rm so(3)}}_l}\times\underset{\scriptscriptstyle \ { s_0 \ } } { { { { \rm u(1)}}_{{\scriptstyle\rm c } } } } , \label{chains1.3n}\end{aligned}\ ] ] where eq . ( [ p0n ] ) has been used to replace @xmath306 with @xmath266 . from eqs . ( [ commcoupled1.2 ] ) , ( [ commcoupled1.5 ] ) , and ( [ commcoupled1.6 ] ) , the 21 operators @xmath307 close an @xmath308 subalgebra of so(8 ) and the subset @xmath309 closes an @xmath310 subalgebra of so(7 ) . thus a third subgroup chain is given by @xmath311 the relationships of the nuclear analog so(8 ) subgroup chains described above are summarized in fig . [ fig : dynamicalchains ] . nuclear analog so(8 ) subgroup chains with generators in the coupled representation @xmath312 given by eqs . ( [ multipole1.1 ] ) and ( [ coupled1.4 ] ) and obeying the lie algebra of eq . ( [ commcoupled ] ) . generators are indicated below each group factor and @xmath266 and @xmath313 may be interchanged using eq . ( [ p0n ] ) . the subgroup structure expressed in this basis is in one - to - one correspondence with the so(8 ) fermion dynamical symmetry model @xcite of nuclear structure physics . however , for the description of graphene it is more useful to transform this basis to the new one employed in fig . [ fig : dynamicalchains_graphene ] using eqs . ( [ ngx ] ) , which gives a more direct physical interpretation of quantities important in graphene physics . ] the subgroup chains in [ h : subgroupchainsnuclear ] were expressed in terms of the nuclear physics basis @xmath314 . this basis demonstrates the deep connection between graphene quantum hall physics and nuclear structure physics , and is suitable mathematically to describe graphene quantum hall effects . however , it is not well suited physically to interpreting the graphene quantum hall effects for three reasons . ( 1 ) the relationship between the generators of the @xmath185 subgroup in the nuclear physics basis and those of the su(4 ) quantum hall ferromagnetism basis defined in eq . ( [ algebra1.4 ] ) is not clear , which hinders interpretation of the present results in terms of preceding results found in the graphene literature . ( 2 ) charge and electronic spin are expected to be conserved in graphene ( if the zeeman term is neglected ) , but none of the generators @xmath315 can be interpreted as spin in the application to graphene ( instead , spin is a linear combination of these generators ) . for physical applications it is desirable to employ a basis in which the relevant conservation laws are manifest . ( 3 ) in addition to the exact conservation laws expected for charge and spin in graphene , it is expected on physical grounds that for low - energy excitations the scattering between valleys is strongly suppressed and the difference in electron densities between neighboring valleys should be very nearly conserved . this difference is expressed by the @xmath58 component of the valley isospin @xmath70 , and the corresponding approximate invariance is reflected in a @xmath316 symmetry generated by @xmath70 . but @xmath70 is not proportional to any of the @xmath304 generators ( it is a linear combination of these generators ) , so this approximate invariance is not manifest in the nuclear su(4 ) basis . thus a new basis will be employed for the so(8 ) generators in application to graphene for which the particle number ( charge ) operator @xmath17 or @xmath266 and the 12 pairing operators @xmath317 are retained , but the 15 su(4 ) generators @xmath318 in the nuclear representation are replaced with the 15 su(4 ) generators @xmath319 with @xmath320 defined in the graphene representation given in eq . ( [ algebra1.4 ] ) , @xmath321 \ { { \mathscr{s}}_\alpha,\ , t_\alpha,\ , n_\alpha,\ , { \pi_{\alphax } } , \,{\pi_{\alphay } } , \ , s_0 , \ , s , \,s^\dagger , \,d{^{\vphantom{\dagger}}}_{\mu},\ , d^\dagger_\mu \ } { _ { \mbox{\scriptsizegraphene so(8 ) } } } . \end{array}\ ] ] the transformation from the @xmath322 generators to the @xmath323 generators is given in appendix [ h : appendixb ] . in the new basis it will be convenient to take as order parameters @xmath324 [ orderp1.1 ] where @xmath325 is the number operator counting particles in basis state @xmath326 and the expectation value is taken with respect to the collective wavefunction . physically ( 1 ) the net spin is measured by @xmath327 , which characterizes _ ferromagnetic _ order . ( 2 ) the difference in charge between the a and b sublattices is measured by @xmath328 , which characterizes _ charge density wave _ order . ( 3 ) the difference in spins between the a and b sublattices is measured by @xmath329 , which characterizes _ antiferromagnetic _ ( nel or spin density wave ) order . the order parameters evaluated for the states of fig . [ fig : s_d_pairs_brillouin ] are displayed in table [ tb : orderparms ] . lccc ' '' '' state & @xmath327 & @xmath328 & @xmath329 + ' '' '' @xmath330 & @xmath331 & @xmath331 & @xmath331 + [ ] @xmath332 & @xmath331 & @xmath333 & @xmath331 + [ ] @xmath334 & @xmath331 & @xmath103 & @xmath331 + [ ] @xmath335 & @xmath333 & @xmath331 & @xmath331 + [ ] @xmath336 & @xmath103 & @xmath331 & @xmath331 + [ ] @xmath337 & @xmath331 & @xmath331 & @xmath331 + [ ] @xmath338 & @xmath331 & @xmath331 & @xmath103 + [ ] @xmath339 & @xmath331 & @xmath331 & @xmath333 + in the new basis it will be assumed that both charge and spin are exactly conserved for the physical states of the model in the absence of the zeeman term @xmath54 , that the charge and the @xmath58-component of spin are exactly conserved if the zeeman term is included in the hamiltonian , and that @xmath70 is conserved , where appropriate . neglecting the zeeman term , the spin charge symmetry corresponds to a group structure @xmath340 , where @xmath341 is generated by the spin operators and @xmath282 is generated by the particle number operator . thus one seeks subgroup chains of so(8 ) that end in the subgroup @xmath340 corresponding to charge and spin conservation , and in some of these chains a u(1)@xmath74 subgroup implying conservation of @xmath70 will also be required . the group and subgroup structure in the new basis is illustrated in fig . [ fig : dynamicalchains_graphene ] , where seven nontrivial subgroup chains may be identified that begin with so(8 ) and end with the symmetry @xmath342 corresponding to conservation of charge and @xmath58-component of the spin in the magnetic field . @xmath343 [ groupchains ] where for brevity all u(1 ) factors are dropped in the notation and su(2 ) means su(2)@xmath344 corresponding to conservation of spin . each of these corresponds to a different dynamical symmetry that is realized for particular choices of parameters for the so(8 ) hamiltonian , and that yields _ exact manybody solutions _ using the dynamical symmetry methodology . let us now discuss in more detail three subgroup chains of so(8 ) . the quasispin generators @xmath345 close an @xmath289 algebra that is a subalgebra of so(8 ) , and the operators @xmath346 close an so(5 ) algebra and commute with these su(2 ) quasispin generators . putting this together , one subgroup of so(8 ) is @xmath347 furthermore , the three generators @xmath348 are components of the total spin and generate an @xmath341 subgroup of so(5 ) , and @xmath266 generates a @xmath282 subgroup of @xmath289 corresponding to conservation of charge . thus one subgroup chain is @xmath349 & \supset { { { \rm su(2)}}_{\sigma}}\times{{{\rm u(1)}}_{{\scriptstyle\rm c } } } , \label{chains1.2}\end{aligned}\ ] ] where the product group on the last line corresponds to conservation of spin and charge . this subgroup chain with its corresponding generators is illustrated in fig . [ fig : dynamicalchains_graphene ] . alternatively , so(5 ) may be broken according to the pattern @xmath350 & \supset { { { \rm su(2)}}_{\sigma}}\times { { { \rm u(1)}}_{{\scriptstyle\rm c } } } , \label{chains1.2b}\end{aligned}\ ] ] which also conserves spin and charge , and is illustrated in fig . [ fig : dynamicalchains_graphene ] . a @xmath351 subgroup of so(8 ) may be obtained by removing the 12 pairing operators from the so(8 ) generator set . the @xmath352 subgroup is generated by the particle number ( charge ) and the su(4 ) subgroup is generated by the 15 remaining operators , which are defined in eq . ( [ algebra1.4 ] ) in the current basis . there are several options for chains corresponding to further subgroups . ( 1 ) the subset @xmath353 defines generators of the so(5 ) symmetry discussed above and so forms an so(5 ) subgroup of this su(4 ) group . hence one su(4 ) subgroup chain is @xmath354 which is displayed in fig . [ fig : dynamicalchains_graphene ] . ( 2 ) physically , the total spin is conserved . if there is little inter - valley scattering one may also assume the spin within each @xmath2 valley and each @xmath3 valley to be separately conserved , corresponding to a @xmath355 symmetry . thus a second su(4 ) subgroup chain corresponds to @xmath356 & \supset { { { \rm su(2)}}_{\sigma}}\times { { { \rm u(1)}}_{{\scriptstyle\rm c } } } , \label{chainy1.1}\end{aligned}\ ] ] where @xmath316 is generated by @xmath70 . this chain also is displayed in fig . [ fig : dynamicalchains_graphene ] . ( 3 ) finally , one can imagine that su(4 ) is broken into subgroups corresponding to simultaneous conservation of both spin and valley isospin , giving a third su(4 ) subgroup chain @xmath357 as illustrated in fig . [ fig : dynamicalchains_graphene ] . comparing fig . [ fig : dynamicalchains_graphene ] with fig . [ fig : dynamicalchains_qhf ] , it is apparent that the three @xmath358 subgroup chains defined in this section correspond to the three symmetry - breaking patterns described in [ h : symmetriesfqhegraphene ] and discussed in refs . @xcite for the su(4 ) quantum hall ferromagnetism model . the portion of the so(8 ) subgroup structure leading to su(4 ) quantum hall ferromagnetism is indicated by the shaded boxes with dashed outer boundaries in fig . [ fig : dynamicalchains_graphene ] . thus it is seen explicitly that the special case corresponding to the @xmath358 subgroup chains of the present model imply the results of refs . @xcite . the 21 operators @xmath359 close an @xmath360 subalgebra of so(8 ) , and the subset of generators @xmath361 close an @xmath362 subalgebra of so(7 ) . thus a third subgroup chain is given by @xmath363 as illustrated in fig . [ fig : dynamicalchains_graphene ] . this subgroup chain is of particular interest because it will define a _ critical dynamical symmetry _ that represents an entire phase exhibiting critical behavior and interpolating between two other phases . let us use the subgroup structure of the preceding section to obtain exact solutions of the correlated many - body problem in these dynamical symmetry limits . the basic idea is to use the casimir invariants of the subgroup chains like those described in [ h : subgroupchainsgraphene ] and illustrated in fig . [ fig : dynamicalchains_graphene ] to label states . then model hamiltonians constructed only from the casimir invariants of a single subgroup chain permit analytical solution of the effective schrdinger equation in that symmetry limit @xcite . specifically , if a hamiltonian @xmath364 can be expressed as a function of the casimir invariants of some subgroup chain @xmath365 , where the @xmath366 represent casimir operators for the groups @xmath367 , then the system is said to possess a dynamical symmetry associated with the subgroup chain . the discussion will be simplified by restricting to the lowest - order casimir invariant for each group , which corresponds physically to omitting @xmath17-body interactions with @xmath368 . elementary properties of lie groups then permit the eigenvalues @xmath369 and eigenfunctions @xmath370 of this hamiltonian to be expressed in closed form as @xmath371 where the @xmath372 stand for the quantum numbers required to specify the irreducible representations ( irreps ) of the groups @xmath367 . the physical properties of the corresponding states can then be elucidated by using the methods of lie groups and lie algebras to evaluate matrix elements for observables . in this way , one generally finds that the dynamical symmetries associated with subgroup chains of some highest symmetry define collective ( emergent ) states that correspond to particular patterns of spontaneous symmetry breaking . in terms of the generators ( [ coupled1.4 ] ) , ( [ multipole1.1 ] ) , and ( [ algebra1.4 ] ) , the quadratic casimir operator @xmath373 for the so(5 ) subgroup is @xmath374 where @xmath375 . the corresponding casimir operator for the @xmath376 subgroup is @xmath377 the quadratic casimir operator for the so(7 ) subgroup is @xmath378 the casimirs for the su(2 ) subgroups are @xmath379 and for the u(1 ) charge subgroup the invariant operator is trivially the single generator , @xmath380 or some power of it . finally , the quadratic casimir operator for the full so(8 ) group may be expressed as @xmath381 the casimir operators that appear in each of these subgroups chains and the relevant quantum numbers labeling the states for each dynamical symmetry are summarized in table [ tb : so8subgroups ] . llllll ' '' '' group & dim & generators & quantum numbers@xmath382 & casimir operators @xmath383 & casimir eigenvalues@xmath384 + ' '' '' @xmath385 & @xmath386 & @xmath387 & @xmath388 & @xmath389 & @xmath390 + [ ] @xmath391 & @xmath392 & @xmath393 or & @xmath394 & @xmath395 & @xmath396 + [ ] & & @xmath397 & & & + [ ] @xmath398 & @xmath399 & @xmath400 or & @xmath401 & @xmath402 & @xmath403 + [ ] & & @xmath404 & & @xmath405 & + [ ] u(4 ) & @xmath406 & @xmath407 & @xmath408 & @xmath409 & @xmath410 + [ ] @xmath411 & @xmath412 & @xmath413 or & @xmath414 & @xmath415 & @xmath416 + [ ] & & @xmath417 & & @xmath418 & + [ ] @xmath289 & @xmath108 & @xmath419 & @xmath9 & @xmath420 & @xmath421 + [ ] @xmath341 & @xmath108 & @xmath422 & @xmath423 & @xmath424 & @xmath425 + [ ] @xmath426 & @xmath108 & @xmath63 & @xmath86 & @xmath427 & @xmath428 + [ ] @xmath282 & @xmath103 & @xmath266 & @xmath17 & @xmath266 & @xmath429 + + + + + + + + in the next section these results for the casimir operators will be used to construct the most general hamiltonian permitted in the truncated collective space for specific dynamical symmetries . as has been seen , in a particular dynamical symmetry limit the most general hamiltonian can be constructed from a sum of casimir invariants for the groups contained in the corresponding subgroup chain . for so(8 ) dynamical symmetries the most general hamiltonian in the absence of the zeeman term is represented by the linear combination @xmath430 where @xmath431 is assumed constant in the symmetry limit , the casimir operators @xmath373 have been summarized in the preceding section , and @xmath432 does not appear explicitly because it has been eliminated by the constraint @xcite : @xmath433 hamiltonians representing specific dynamical symmetry limits then correspond to particular choices of the coefficients @xmath434 in this general expression . it may be shown that the most general hamiltonian can also be expressed in the compact form ( see eqs . ( 4.1 ) of ref . @xcite ) @xmath435 where @xmath436 is assumed constant in a symmetry limit and where the different dynamical symmetry limits correspond to specific choices for the values of the parameters @xmath437 , @xmath438 , and @xmath439 . the last term is expressed in terms of the @xmath304 from the nuclear basis . it can be converted to the graphene basis by inverting eqs . ( [ ngx ] ) of appendix [ h : appendixb ] to solve for the @xmath304 . let us now discuss each of the so(8 ) dynamical symmetries and their physical interpretations . for brevity , let us refer to 1 . the dynamical symmetry structure associated with eqs . ( [ chains1.2 ] ) and ( [ chains1.2b ] ) as the _ @xmath440 dynamical symmetry , _ 2 . the dynamical symmetry structure associated with eqs . ( [ chains1.3])([chainy1.2 ] ) as the _ su(4 ) dynamical symmetry , _ and 3 . the dynamical symmetry structure associated with eq . ( [ chains1.4 ] ) as the _ so(7 ) dynamical symmetry . _ initially the role of the zeeman term ( which would break the full spin symmetry down to conservation of its @xmath58 component ) will be ignored and it will be assumed that the chains end in the subgroup @xmath441 corresponding to the physical requirement that spin and charge be conserved exactly . the dynamical symmetry chains given in eqs . ( [ chains1.2])([chains1.2b ] ) and illustrated in fig . [ fig : dynamicalchains_graphene ] correspond to two alternative ways of choosing subgroups of @xmath442 : @xmath443{45}{$\supset$}}}$ } \ { { { \rm su(2)}}_{\sigma}}\times { { \rm su(2)}}{_{\mbox{\scriptsizep}}}\ \raisebox{-0.8ex}{${\mathbin{\rotatebox[origin = c]{-45}{$\supset$}}}$ } \\ { { \rm so(5)}}\times{{\rm su(2)}}{_{\mbox{\scriptsizep } } } \hphantom{{\mathbin{\rotatebox[origin = c]{45}{$\supset$}}}\ { { { \rm su(2)}}_{\sigma}}\times{{{\rm su(2)}}_{\tau}}{\mathbin{\rotatebox[origin = c]{-45}{$\supset$ } } } } \ \ { { { \rm su(2)}}_{\sigma}}\times { { { \rm u(1)}}_{{\scriptstyle\rm c}}}\\ \hphantom{{{\rm so(5)}}\times{{{\rm su(2)}}_{\tau}}\ } \raisebox{1.0ex}{${\mathbin{\rotatebox[origin = c]{-45}{$\supset$}}}$ } \makebox[\widthof{$\ { { { \rm su(2)}}_{\sigma}}\times{{{\rm su(2)}}_{\tau}}\ $ } ] [ c]{${{\rm so(5)}}\ \times \ { { { \rm u(1)}}_{{\scriptstyle\rm c}}}$ } \raisebox{1.0ex}{${\mathbin{\rotatebox[origin = c]{45}{$\supset$}}}$ } . \end{array } \label{so5dyn1.1}\ ] ] in the upper branch of the middle step the so(5 ) symmetry is broken to its @xmath444 spin subgroup , with @xmath289 unbroken . physically this corresponds to conservation of the spin associated with the @xmath445 angular momentum algebra and the pseudospin associated with the @xmath446 pair algebra , but not the full so(5 ) symmetry . in the lower branch of the middle step , the so(5 ) symmetry remains intact and @xmath143-pair pseudospin @xmath444 is broken to @xmath447 charge . in the final subgroup of both chains , only the spin and charge remain as conserved quantities . the hamiltonian in the @xmath448 dynamical symmetry limit corresponds to eq . ( [ mostgeneralham ] ) with the restriction that @xmath449 , @xmath450 where terms that are constant within a given representation have been omitted . the most general so(8 ) state in the @xmath177 collective subspace is given by eq . ( [ collective1.1 ] ) and corresponds to a superposition of @xmath143 and @xmath231 pairs . schematically , @xmath451 with @xmath228 and @xmath452 defined in eq . ( [ coupled1.4 ] ) and @xmath453 , where @xmath178 is the total pair number . on the other hand , the most general states corresponding to the various subgroup chains illustrated in fig . [ fig : dynamicalchains_graphene ] correspond to pair superpositions having specific constraints on the relative contribution of @xmath143 and @xmath454 pairs . the collective subspace for the @xmath455 subgroup of so(8 ) is of the form ( see eq . ( 9.13 ) of ref . @xcite ) @xmath456 implying that it is a superposition of @xmath143 pairs . conceptually , the wavefunction of the @xmath455 subgroup chain for @xmath177 is obtained from the most general state in the collective subspace by converting all of its @xmath454 pairs to @xmath143 pairs . as was seen in [ sh : coupledpairrep ] and fig . [ fig : s_d_pairs_brillouin ] , the @xmath143 and @xmath454 pairs correspond to coherent superpositions of particular electronic distributions in spin and valley pseudospin . thus , specific @xmath143 and @xmath454 pair content for so(8 ) dynamical symmetry subgroup chains implies specific collective modes associated with coherent distribution of the electrons in spin and valley space . it has been noted above that in the @xmath457 dynamical symmetry limit the ground states correspond to a superposition of @xmath143 pairs . the nature of this collective state may be inferred from fig . [ fig : s_d_pairs_brillouin ] and is illustrated in fig . [ fig : s_d_pairs_brillouin_so5 ] . the collective state corresponding to the @xmath458 ground state . the state is an antisymmetric combination of a spin density wave with spin - up on the a lattice sites and spin - down on the b sites , and a spin density wave with spin - up on the b sites and spin - down on the a sites . ] the su(4 ) dynamical symmetry corresponds to the three so(8 ) subgroup chains @xmath459 that were introduced in eqs . ( [ chains1.3])([chainy1.2 ] ) and fig . [ fig : dynamicalchains_graphene ] . as already noted , these three dynamical symmetry chains are in one to one correspondence with the explicit symmetry breaking patterns that have been identified for su(4 ) quantum hall ferromagnetism . the most general su(4 ) wavefunction for @xmath178 pairs in the @xmath177 collective so(8 ) subspace is given by @xcite @xmath460^p { |0\rangle } , \label{su4pairstate}\ ] ] which corresponds physically to a restriction of the general so(8 ) wavefunction ( [ so8pairstate ] ) to a specific superposition of @xmath143 and @xmath454 pairs . the wavefunction for the @xmath362 subgroup of su(4 ) is given by eq . ( [ so5pairstate ] ) . the wavefunction of the parent su(4 ) group is a superposition of @xmath143 and @xmath454 pairs but the @xmath448 subgroup has a wavefunction containing only @xmath143 pairs . the so(7 ) dynamical symmetry corresponds to the so(8 ) subgroup chain @xmath461 that was introduced in eq . ( [ chains1.4 ] ) and displayed in fig . [ fig : dynamicalchains_graphene ] . the hamiltonian in the so(7 ) dynamical symmetry limit corresponds to eq . ( [ mostgeneralham ] ) with the restriction that @xmath462 , @xmath463 where terms have been dropped that are constant within a given representation . from the nuclear physics analog so(8 ) symmetry @xcite , one may surmise that so(7 ) will play the role of a _ critical dynamical symmetry _ interpolating smoothly between the collective states corresponding to the @xmath411 dynamical symmetry and the collective states corresponding to @xmath398 dynamical symmetry . such critical dynamical symmetries have been discussed previously in both nuclear physics @xcite and for the strongly - correlated electrons leading to cuprate and iron - based high - temperature superconductivity @xcite . they may be viewed as the generalization of a quantum critical point to an entire _ quantum critical phase , _ and may represent a fundamental organizing principle for quantum critical behavior . the physical implications of this so(7 ) critical dynamical symmetry for graphene quantum hall physics will be discussed further below . the dynamical symmetry limits discussed above represent special solutions resulting from particular choices of the coupling parameters appearing in the hamiltonian . for arbitrary choices of the coupling parameters the solutions will correspond generally to superpositions of the different symmetry - limit solutions and will not have exact analytical forms . in this more general case it is quite feasible to obtain solutions numerically , since the collective subspace is highly truncated relative to the full hilbert space . however , there is a powerful alternative approach : the _ generalized coherent state approximation , _ which permits _ analytical solutions _ for arbitrary choices of the coupling parameters in the hamiltonian . for the so(8 ) lie algebra introduced in this paper for graphene the gilmore perelomov algorithm @xcite may be implemented to obtain solutions in terms of a set of generalized coherent states . these solutions represent the most general hartree fock bogoliubov theory that can be formulated in the space , subject to a dynamical symmetry constraint @xcite . the solutions of this symmetry - constrained hartree fock bogoliubov ( schfb ) theory correspond to determining the stable points of energy surfaces , which represent the coherent - state expectation values of the effective hamiltonian on the coset space . thus the coherent state solutions also represent a microscopically - derived implementation of ginzburg landau theory . these coherent state solutions are uniquely well suited to study the interplay of competing spontaneous symmetry breaking in determining the ground state of the system and its properties . the coherent states associated with the full set of subgroup chains in fig . [ fig : dynamicalchains_graphene ] will be discussed in future work . in this paper the power of the method will be illustrated succinctly by restricting to the coherent states associated with the subgroup chains of so(8 ) that contain the so(5 ) subgroup , as illustrated in fig . [ fig : dynamicalchains_coherentso5 ] . 0pt 0.20 subgroup chains included in the present coherent state analysis . thus the corresponding coherent state solutions will represent a superposition of the symmetry - limit solutions for the @xmath464 { { \rm so(8)}}\supset{{\rm su(4)}}\supset{{\rm so(5)}}\qquad { { \rm so(8)}}\supset{{\rm so(7)}}\supset{{\rm so(5)}}\end{array}\ ] ] dynamical symmetries . these coherent state solutions will be seen to have the following properties : 1 . the @xmath391 dynamical symmetry will play the role of a critical dynamical symmetry interpolating between @xmath411 and @xmath398 symmetry - limit solutions . because of fundamental symmetries obeyed by the wavefunction , the coherent state solutions may be parameterized in terms of a single collective parameter @xmath465 that governs the mixture of the @xmath143 and @xmath454 pairs defined in eq . ( [ coupled1.4 ] ) contributing in the ground state . the collective parameter @xmath465 may also be interpreted physically in terms of the pair configurations displayed in fig . [ fig : s_d_pairs_brillouin ] . the so(8 ) coherent states corresponding to the symmetry structure in fig . [ fig : dynamicalchains_coherentso5 ] have been developed previously in ref . @xcite for nuclear physics applications and will be adapted extensively to development of the present formalism . let us now consider the energy surfaces that may be computed from the coherent states , which link the so(8 ) solutions to ginzburg landau theory . within the coherent state formalism the ground state energy may be determined through the variational requirement @xmath466 where @xmath467 is the coherent state , @xmath468 is the so(8 ) hamiltonian , @xmath469 and the coefficients @xmath437 and @xmath470 are functions of the effective interaction . for the dynamical symmetry chains having so(5 ) as a subgroup the energies take the general form @xcite @xmath471 , \label{generalenergyformula}\ ] ] where the group - dependent parameters @xmath472 , @xmath473 , @xmath474 , @xmath475 , and @xmath476 are given in table [ tb : stiffnessparameters ] . cccccc ' '' '' g & @xmath472 & @xmath124 & @xmath477 & @xmath478 & @xmath479 + ' '' '' su(2 ) & @xmath437 & @xmath480 & @xmath481 & @xmath482 & @xmath483 + [ ] so(5 ) & @xmath484 & @xmath485 & @xmath486 & @xmath487 & @xmath488 + [ ] su(4 ) & @xmath489 & @xmath490 & @xmath491 & @xmath492 & @xmath331 + [ ] so(7 ) & @xmath438 & @xmath493 & @xmath494 & @xmath495 & @xmath496 + our primary interest in this discussion is in the ground state properties of graphene in a strong magnetic field . the ground states in the coherent state approximation at fixed @xmath497 will be given by those values of @xmath498 that correspond to minima of the energy surface @xmath499 . these are determined by the values of @xmath465 satisfying @xmath500 evaluating these constraints for the energy surfaces ( [ generalenergyformula ] ) , one finds that the minima @xmath501 are given by @xcite @xmath502 the coherent state energy surfaces for the @xmath448 , so(7 ) , and su(4 ) symmetry limits computed from eq . ( [ generalenergyformula ] ) using the entries in table [ tb : stiffnessparameters ] are shown as functions of @xmath465 for several values of the fractional occupation @xmath160 in fig . [ fig : coherentenergysurfaces_graphene ] . coherent state energy surfaces as a function of the order parameter @xmath465 for three of the so(8 ) dynamical symmetry limits . the curves are labeled by the fractional occupation @xmath503 defined in eq . ( [ fillingfraction ] ) , where @xmath17 is the particle number and @xmath147 is the maximum number of particles that can be accommodated in the landau level . the formalism is particle hole symmetric so curves for fractional occupations with @xmath504 are equivalent to those shown above , but with the fractional occupation counted in terms of number of holes . for example , the curves for @xmath505 and @xmath506 are equivalent . for the direct product group @xmath507 the energy surface is the sum of contributions to eq . ( [ generalenergyformula ] ) from su(2 ) and so(5 ) . as will be demonstrated later , the su(2 ) contribution typically dominates that of so(5 ) . ] there one sees that indeed the minima for the @xmath448 and so(7 ) limits are at @xmath508 , and the minima for the su(4 ) limit are at @xmath509 . although the minimum energies for both the @xmath448 and so(7 ) limits are consistent with @xmath510 , fig . [ fig : coherentenergysurfaces_graphene ] shows that these symmetries differ fundamentally in the localization of the minimum . for @xmath448 the energy surface has a deep minimum at @xmath510 but for so(7 ) the energy surface is very flat around @xmath508 , with a broad range of @xmath465 giving essentially the same ground state energy . this highly - degenerate so(7 ) ground state has significant physical implications that will be discussed further below . the so(8 ) generalized coherent state is equivalent to the hartree fock bogoliubov ( hfb ) approximation subject to a symmetry constraint . since hfb is a bcs - type approximation married to a hartree fock mean field , its solutions correspond to symmetry - breaking intrinsic states . in particular , the bcs - like state conserves the physical particle number only on average , and the hartree fock mean field may break both translational and rotational invariance . let us address these issues for the so(8 ) coherent state . the fractional uncertainty in electron number @xmath511 for the so(8 ) coherent state is given by @xcite @xmath512 where @xmath513 is the value of @xmath465 at the minimum energy , given by eq . ( [ betaminima ] ) in the symmetry limits . expressing eq . ( [ particlenumberfluctuation ] ) in terms of the fractional occupation @xmath514 , in the @xmath448 and su(4 ) limits , respectively , one obtains @xmath515_{{\rm so}_5\times{{\rm su(2 ) } } } \hspace{-5pt } = \sqrt{\frac{1-{f}}{{f}\omega } } \qquad \left[\frac{\delta n}{n}\right]_{{\rm su}_4 } \hspace{-5pt } = \sqrt{\frac{1 - 2{f}}{{f}\omega } } , \label{particlenumberfluctuation3}\ ] ] from these results one may notice two important things . 1 . the fluctuation in particle number is large at low degeneracy @xmath128 but decreases with increasing @xmath128 . if su(4 ) symmetry is realized @xmath516 decreases with increasing @xmath160 and _ vanishes identically _ at @xmath167 for any @xmath128 , which corresponds to the fractional occupation for the ground state of undoped graphene . thus it is expected that the current theory applied to graphene has negligible particle number fluctuation @xmath516 in the su(4 ) limit . in the @xmath448 limit the particle number fluctuation @xmath516 remains finite for all @xmath160 but it becomes very small as @xmath128 becomes large , particularly near @xmath517 . thus it too may be neglected in the large-@xmath128 limit . comparison with table [ tb : degeneracytable ] suggests that graphene quantum hall experiments involve sufficient degeneracy that particle number fluctuation in the coherent state solution is not significant . the coherent state approximation represents a mean field localized in spatial position and orientation , so it violates translational and rotational invariance . however , because the crystal is generally macroscopic , the net violation of these symmetries may be expected to be negligible . one concludes that for applications of coherent state methods to graphene , violations of particle number conservation , rotational invariance , and translational invariance are negligible in realistic systems . the generalized coherent state method has been used above to calculate total energy surfaces for quantum hall states in graphene , but one also may use the coherent state wavefunctions and appropriate operators to calculate matrix elements of other relevant observables . this section addresses the nature of the wavefunction and the matrix elements that can serve as order parameters . a significant consequence of the so(8 ) dynamical symmetry structure displayed in fig . [ fig : dynamicalchains_coherentso5 ] is that the phases may be distinguished in terms of a single order parameter and its fluctuations , which may be taken to be @xmath465 . let us now characterize in more depth the physical meaning of this order parameter . in [ sh : orderparameters ] an antiferromagnetic order parameter @xmath329 was defined . in the coherent state approximation the onset of af order is signaled by an energy - surface minimum at a finite value of @xmath465 . because @xmath518 [ see eq . ( [ ngx9 ] ) ] , the antiferromagnetic order parameter @xmath329 is related to the coherent state af order parameter @xmath465 by the intrinsic state matrix element of @xmath519 @xcite @xmath520 where @xmath489 is the coupling strength for the @xmath521 term in the hamiltonian . each value of @xmath465 corresponds to a unique value of @xmath329 , so @xmath465 is a measure of antiferromagnetic order . the location of the maxima may be obtained by setting the derivative with respect to @xmath465 of eq . ( [ intrinsic1.1 ] ) equal to zero , which yields that @xmath522 for a given @xmath17 occurs at a @xmath465 of @xmath523 where eq . ( [ betaminima ] ) was used to make the identification in the last step . thus , if @xmath524 the maximum value of @xmath329 maps to a value of @xmath465 that corresponds to a minimum of the energy surface ( a ground state ) in the su(4 ) limit . substituting eq . ( [ betaminima ] ) for @xmath465 in eq . ( [ intrinsic1.1 ] ) , for @xmath524 ground states the af order parameter @xmath329 depends on the electron number @xmath17 as @xmath525 the so(8 ) model is particle hole symmetric so @xmath17 or @xmath526 count electrons up to half filling and holes for greater than half filling . hence the maximum af collectivity occurs for half filling of the single valence landau level . as was discussed in [ h : conservationlaws ] , the coherent state wavefunction corresponding physically to @xmath527 pairs conserves particle number only on average and so is a superposition of terms having different numbers of pairs . in eq . ( 5.27 ) of ref . @xcite the so(8 ) coherent state is decomposed into terms of definite pair number @xmath528 according to @xmath529 where @xmath530 denotes an intrinsic state with order parameter @xmath465 and closed forms for @xmath531 and @xmath532 are given in ref . @xcite . according to eq . ( 5.28 ) of ref . @xcite , the values of @xmath533 that correspond to the minima of the potential energy surface at @xmath534 for the @xmath448 limit and @xmath535 for the @xmath411 limit [ see eq . ( [ betaminima ] ) ] , respectively , are @xmath536 so the so(8 ) coherent state wavefunction ( [ coherentfn1.1 ] ) in the @xmath448 and su(4 ) limits , respectively , becomes @xmath537 \displaystyle { |{\rm su}_4\rangle } = \sum _ { p } { c_p}\left ( s^\dagger \pm d_0^\dagger \right)^p { |0\rangle } = 2\sum _ { p } { c_p}\left ( q_\pm^\dagger \right)^p { |0\rangle } , \end{array } \label{coherentfn1.4}\ ] ] where eq . ( [ qdef ] ) was used . as discussed in [ h : conservationlaws ] , fluctuations in particle number are negligible in the large-@xmath128 limit for so(8 ) coherent states , implying that the summations in eqs . ( [ coherentfn1.4 ] ) become dominated by terms with @xmath538 . thus for large @xmath128 the coherent state wavefunctions are well approximated up to a normalization by @xmath539 [ largeomegwf ] as seen from table [ tb : orderparms ] , the su(4 ) state of eq . ( [ largeomegwf.b ] ) is a coherent superposition of @xmath540 or @xmath541 pairs , each contributing vanishing ferromagnetic order @xmath327 and charge density wave order @xmath328 , but non - zero af order @xmath329 . conversely , the @xmath448 state of eq . ( [ largeomegwf.a ] ) is a coherent superposition of @xmath143 pairs , each with vanishing @xmath327 , @xmath328 , and @xmath329 . thus the @xmath448 and su(4 ) limits of the so(8 ) symmetry are distinguished by the order parameter @xmath329 , which is zero in the @xmath448 state and is non - zero in the su(4 ) state . equivalently , the coherent state order parameter @xmath465 vanishes in the pure @xmath448 limit and is equal to @xmath542 in the pure su(4 ) limit [ see eq . ( [ betaminima ] ) and fig . [ fig : coherentenergysurfaces_graphene ] ] . equation ( [ generalenergyformula ] ) depends only on even powers of @xmath465 so the sign for the two possible spontaneously broken symmetry solutions does not affect the energy . for undoped graphene the fermi surface corresponds to the @xmath543 curves of fig . [ fig : coherentenergysurfaces_graphene ] . these are shown in fig . [ fig : coherentenergysurfaces_graphenegs ] for the @xmath448 , so(7 ) , and su(4 ) limits . ground - state energy surfaces in coherent state approximation for three of the so(8 ) dynamical symmetry limits . ( a ) the @xmath448 limit . ( b ) the so(7 ) limit . ( c ) the su(4 ) limit . the diagrams indicate schematically the corresponding wavefunctions , as suggested by eqs . ( [ coherentfn1.1])([largeomegwf ] ) and fig . [ fig : s_d_pairs_brillouin ] . ] * the @xmath448 limit in fig . [ fig : coherentenergysurfaces_graphenegs](a ) has its minimum @xmath513 at @xmath544 . it corresponds to states of the form ( [ largeomegwf.a ] ) , with vanishing expectation values of @xmath545 , @xmath70 , and @xmath546 . * the su(4 ) limit of fig . [ fig : coherentenergysurfaces_graphenegs](c ) corresponds to states of the form ( [ largeomegwf.b ] ) , with minima located at @xmath547 when the symmetry is broken spontaneously by choosing one of these possibilities ( they are equivalent since the energy depends only on even powers of @xmath465 ) , the resulting state has @xmath548 ( no spin or isospin order ) , but @xmath549 ( spin density wave or af order ) . * the so(7 ) limit of fig . [ fig : coherentenergysurfaces_graphenegs](b ) corresponds to a critical dynamical symmetry that interpolates between the @xmath448 and su(4 ) states through critical fluctuations in the antiferromagnetic order . thus the so(8 ) dynamical symmetry limits illustrated in fig . [ fig : coherentenergysurfaces_graphenegs ] represent a rich set of collective states that can be distinguished by the expectation value and fluctuations associated with the order parameter @xmath465 . the so(8 ) coherent state solution can be used to study transitions among the phases defined in fig . [ fig : coherentenergysurfaces_graphenegs ] . for the @xmath177 space ( no broken pairs ) assumed here , @xmath550 yields a constant that is neglected and eq . ( [ coherent1.12 ] ) may be expressed as @xmath551 the last two terms yield constants when evaluated in a given representation , and @xmath552 is found to contribute negligibly to the total energy compared with @xmath553 . therefore , it will be instructive to set @xmath554 and study the approximate so(8 ) hamiltonian @xmath555 from eqs . ( [ casimir_so5])([casimir_so8 ] ) , one finds that @xmath556 { \langles^\dagger s\rangle } + { \langle p^2{\hspace{-0.1em}\cdot\hspace{-0.1em}}p^2\rangle } \sim { \langle{c_{\scriptscriptstyle{{\rm so(7)}}}}\rangle } \end{array}\ ] ] if constants are neglected . thus the model hamiltonian ( [ modelhamiltonian ] ) may be tuned to favor the @xmath448 , so(7 ) , or su(4 ) phases by varying the ratio of the coupling parameters @xmath437 and @xmath489 . to study the quantum phase transitions of the so(8 ) model with the approximate hamiltonian ( [ modelhamiltonian ] ) , it is convenient to define a parameter @xmath557 and to rewrite eq . ( [ modelhamiltonian ] ) as @xmath558 thus the value of @xmath559 tunes the hamiltonian ( [ modelhamiltonian2 ] ) between @xmath507 and su(4 ) phases via an intermediate so(7 ) phase exhibiting quantum critical behavior . 1 . if @xmath560 the ground - state energy surface is approximated by fig . [ fig : coherentenergysurfaces_graphenegs](a ) , with a minimum at @xmath544 , no antiferromagnetic order , and @xmath507 symmetry . 2 . if @xmath561 the ground - state energy surface is approximated by fig . [ fig : coherentenergysurfaces_graphenegs](c ) , with an energy minimum at @xmath562 implying @xmath411 symmetry and antiferromagnetic order . 3 . if @xmath563 , the ground - state energy surface is approximated by fig . [ fig : coherentenergysurfaces_graphenegs](b ) and the system exhibits so(7 ) critical dynamical symmetry , with large fluctuations in the af order parameter @xmath465 . let us now use the hamiltonian ( [ modelhamiltonian2 ] ) to study the quantum phase transitions and spontaneously broken symmetry of the so(8 ) model . if terms involving @xmath564 and @xmath565 are ignored ( as justified above ) , eq . ( [ generalenergyformula ] ) with the parameters in table [ tb : stiffnessparameters ] imply that @xmath566 hence the energy surfaces corresponding to the hamiltonian ( [ modelhamiltonian2 ] ) may be expressed as @xmath567 the variation of the energy surface computed from eq . ( [ coherent1.13d ] ) with the control parameter @xmath568 for half filling ( ground state for undoped graphene ) is shown in fig . [ fig : coherentenergysurfaces_control2d ] ; quantum phase transitions with coupling strength as control parameter : coherent state energy surface as a function of the control parameter @xmath569 for a degeneracy parameter @xmath570 and a fractional occupation @xmath571 . the solid green curves from @xmath572 correspond to approximate @xmath448 symmetry , the solid blue curves for @xmath573 correspond to approximate su(4 ) symmetry . the dashed red curves near @xmath563 correspond to an approximate so(7 ) symmetry mediating the quantum phase transition from @xmath448 to su(4 ) symmetry . ] by tuning the control parameter from @xmath574 to @xmath561 , one sees that the system undergoes a quantum phase transition near @xmath575 from an approximate @xmath457 state with energy minimum at @xmath534 [ see fig . [ fig : coherentenergysurfaces_graphenegs](a ) ] to an approximate su(4 ) state having energy minima at @xmath576 [ see fig . [ fig : coherentenergysurfaces_graphenegs](c ) and eq . ( [ betaminima ] ) ] . for @xmath563 the system has an approximate so(7 ) dynamical symmetry [ see fig . [ fig : coherentenergysurfaces_graphenegs](b ) ] , with no well - defined minimum for the energy as a function of @xmath465 , but with large fluctuations in @xmath465 implied by a highly - degenerate ground state . for fixed values of the coupling parameters @xmath437 and @xmath489 , phase transitions may be mediated by changing the particle occupancy . [ fig : coherentenergysurfaces_vs_n_2d ] illustrates for different values of @xmath497 at fixed @xmath577 . one sees that as the particle number is increased the system makes a transition from approximate @xmath448 symmetry with @xmath534 to su(4 ) symmetry with @xmath524 through a critical so(7 ) symmetry for which the energy is highly degenerate in @xmath465 . in the coherent state approximation the dynamical symmetry structure of fig . [ fig : dynamicalchains_coherentso5 ] has been examined and not the full group structure given in fig . [ fig : dynamicalchains_graphene ] . for the group chains that contain the so(5 ) subgroup , the only physical implication is to omit the physical effect of zeeman splitting from dynamical symmetry hamiltonian [ which would break the so(5 ) subgroup down into a u(1 ) subgroup generated by the @xmath58 component of the physical spin ] . our primary concern in this discussion is the structure associated with the @xmath5 landau level , for which the effect of the zeeman term is expected to be small ( see [ h : applicationsgraphene ] ) . thus , one may view the effect of the zeeman term as a perturbation on the results obtained thus far that will act only on the spin part of the wavefunction . as kharitonov @xcite has already discussed , the competition of the zeeman term with the valley interactions will convert the antiferromagnetic solution into a canted antiferromagnetic solution . quantum phase transitions with particle number as control parameter : coherent state energy surfaces for different filling fraction @xmath578 at fixed values for @xmath437 and @xmath489 with @xmath579 . the solid green curves for @xmath580 correspond to approximate @xmath448 symmetry , the solid blue curves for @xmath581 correspond to approximate su(4 ) symmetry . curves near @xmath582 ( dashed red ) correspond to an approximate so(7 ) symmetry mediating the quantum phase transition from @xmath448 to su(4 ) symmetry . ] placing a strong magnetic field on graphene leads to high level degeneracy at energies corresponding to the quantized landau levels . in general , interacting electronic systems with high level degeneracy near the fermi surface can produce ( even for relatively weak interactions ) a rich variety of collective states that differ qualitatively from the non - interacting ground state . these states correspond to a spontaneous breaking of the symmetry and generally can not be obtained through small perturbations of the weakly - interacting ground state since they are emergent in nature . the su(4 ) symmetry of quantum hall ferromagnetism gives rise to a ground state symmetry reflected in the su(4)-symmetric hamiltonian ( [ so5_1.1a ] ) and a possible symmetry breaking structure that has been outlined in [ h : symmetriesfqhegraphene ] . however , the symmetry - breaking patterns illustrated in fig . [ fig : dynamicalchains_qhf ] represents _ perturbations around the symmetric ground state ( explicit symmetry breaking ) . _ they can not capture the nature of these possible spontaneously - broken symmetry states , since the broken - symmetry states may differ fundamentally from the possible states in fig . [ fig : dynamicalchains_qhf ] : the states corresponding to the most general linear combination ( [ upairs1.8 ] ) represent a complex superposition of many su(4)-symmetric components and generally can not be classified by pure or any simple linear combination of su(4 ) irreducible representations . since the nature of the broken symmetry states can not be studied directly within the su(4 ) framework because they are unlikely to be anywhere near eigenstates of an su(4)-symmetric hamiltonian , the broken - symmetry states have typically been studied numerically , or by effective field theory methods . however , as has been shown , the kinds of collective configurations that have been proposed as candidates for low - lying broken - symmetry states in graphene ( see fig . [ fig : s_d_pairs_brillouin ] and @xcite ) bear strong resemblance to _ eigenstates _ of so(8 ) dynamical symmetry chains . thus , the present so(8 ) symmetry holds the promise of providing _ analytical _ solutions for possible broken - symmetry states in graphene . this is the most important result of the present paper . at specific filling factors the ground state of graphene will be determined by the competition among the su(4 ) symmetry breaking terms . the most obvious su(4)-anisotropic effect is the zeeman term , which favors a spin - polarized state , but the graphene sublattice structure introduces additional interactions that favor ground states without spin polarization that are characterized by spin density wave or charge density wave order at the lattice scale . the competition between zeeman - term spin polarization and the lattice - scale polarizations can be studied by changing the in - plane component of the magnetic field relative to the perpendicular component , since this changes the zeeman energy but not the orbital energies @xcite . such studies indicate that for the higher - energy landau levels the zeeman term is dominant , producing spin ferromagnets that have skyrmionic excitations at half filling , but in the @xmath5 landau level the lattice - scale interactions dominate the zeeman interaction and drive the system into a spin - unpolarized state @xcite . the remainder of this discussion will concentrate on these spin - unpolarized collective states that are candidates for the ground state in the @xmath5 landau level for charge - neutral graphene , with the zeeman interaction viewed as a perturbation on a collective structure that is dominated by lattice - scale interactions . consideration in this paper will be restricted further to those states that can arise from the dynamical symmetries of fig . [ fig : dynamicalchains_graphene ] that contain the so(5 ) subgroup ( those displayed in fig . [ fig : dynamicalchains_coherentso5 ] ) . it has been shown that so(8 ) describes analytically a number of spontaneously - broken - su(4 ) candidates for the states observed in modern experiments such as those described in refs . these solutions provide a spectrum of excited states as well as ground states . the excited states will not be discussed here , except to note that all ground state solutions have a gap to electronic and collective excitations and so correspond to incompressible states . the general theory to be discussed in forthcoming papers can accommodate fm , cdw , and af states , but for the dynamical symmetries containing so(5 ) that were the focus here , all solutions may be classified by a _ single parameter _ @xmath465 measuring af order : su(4 ) states have finite af order but no cdw or fm order , @xmath448 states have no af , cdw , or fm order , and so(7 ) states correspond to a critical dynamical symmetry interpolating between su(4 ) and @xmath448 with large af fluctuations but no static af order , and with no cdw or fm order . in a strong magnetic field the zero - energy state of graphene has four - fold spin and valley degeneracy per landau level , and ( neglecting the lattice - scale interactions ) near the sample boundary one might expect the zero landau level to split into one positively - dispersing ( electron - like ) and one negatively dispersing ( hole - like ) mode for each spin projection . this would suggest a ground state having a bulk energy gap at charge neutrality but with electron - like and hole - like states of opposite spin polarization crossing at the edge of the sample ( with the edge - state structure being analogous to the quantum spin hall effect ) @xcite . however , experiments indicate that the ground state of charge - neutral graphene becomes _ strongly insulating _ at high magnetic fields @xcite . the detailed nature of this state remains uncertain , but it is generally expected to correspond to a spontaneously broken symmetry caused by the strong coulomb interactions among the electrons in the zero landau level . transport properties are not manifest in the algebraic solutions presented here but the coherent state approximation is equivalent to symmetry - constrained hartree fock bogoliubov ( hfb ) theory @xcite , suggesting that so(8 ) theory can be mapped onto hartree fock ( hf ) transport calculations . hf calculations for armchair nanoribbons found that af and cdw states similar to ours have no edge currents @xcite . it may be speculated that our af states also are insulating and thus strong candidates for the high - field ground state , but confirmation requires more work . solutions depend on @xmath437 and @xmath489 in eq . ( [ modelhamiltonian ] ) , which define effective interactions in the truncated space [ highly renormalized relative to parameters in eq . ( [ so5_1.1a ] ) ] . they may be fixed by systematic comparison with data , enabling a robust prediction for the nature of the ground and other low - energy states . one expects modest impurity levels to modify the effective interaction parameters but to leave dynamical symmetries intact . a unified model of conventional superconductivity and high - temperature superconductors having an @xmath385 highest symmetry has been proposed @xcite . for sufficiently strong on - site coulomb repulsion ( true for the cuprates and approximately true for the fe - based compounds ) , the most general so(8 ) symmetry is reduced to its su(4 ) subgroup @xcite , which forms the basis of an su(4 ) dynamical symmetry model of high - temperature superconductivity . this su(4 ) model has three dynamical symmetry subchains ending in an @xmath583 subgroup representing conservation of spin and charge . physically , these su(4 ) subgroup chains represent 1 . an su(2 ) pseudospin subgroup chain that describes a @xmath584-wave singlet superconductor ( sc ) . an so(4 ) subgroup chain that describes an antiferromagnetic mott insulator ( af ) . an so(5 ) subgroup chain representing a critical dynamical symmetry that is soft with respect to af and sc fluctuations and interpolates between the af and sc collective modes . thus , in the su(4 ) model of high-@xmath585 superconductivity the so(5 ) subgroup chain plays a similar physical role as the so(7 ) subgroup chain of the graphene so(8 ) dynamical symmetry model and the so(7 ) subgroup chain of the nuclear so(8 ) model . in all three cases the subgroup chains represent the generalization of a quantum critical point to an entire quantum critical phase that exhibits large fluctuations ( in order - parameter space ) connecting collective modes defined by other dynamical symmetries of the problem . in the high-@xmath585 case the modes connected by the @xmath586 critical dynamical symmetry are antiferromagnetism and superconductivity , in graphene the critical @xmath587 dynamical symmetry connects states with and without to nel ( af ) order , and in the nuclear structure case the critical @xmath587 dynamical symmetry connects collective states that differ in the relative contributions of nucleon pairs carrying total angular momentum 0 and total angular momentum 2 . thus physically these three applications of dynamical symmetry have little in common , but one sees that mathematically they have a deep similarity . although the microscopic physics differs fundamentally , at the level of the observed emergent collective modes in the system one sees that graphene quantum hall states , high - temperature superconductors , and broad classes of nuclear collective states have a unified description in terms of dynamical symmetries associated with very similar compact lie algebras . this remarkable similarity is illustrated in fig . [ fig : nuc_sc_graphene_chains ] the well - known quantum hall ferromagnetic su(4 ) symmetry of graphene in strong magnetic fields has been extended by adding to the particle hole operators that generate su(4 ) a set of six creation and six annihilation operators that create or destroy fermion pairs in either a total valley isospin triplet , total spin singlet state , or a total valley isospin singlet , total spin triplet state ( the only possibilities allowed by the pauli principle ) . this extended set of operators is shown to close an so(8 ) algebra under commutation , which is formally analogous to the so(8 ) algebra of the fermion dynamical symmetry model of nuclear structure physics . this permits immediate adaptation of mathematical tools developed in nuclear physics to the graphene problem . the previously - known su(4 ) quantum hall ferromagnetism symmetry is recovered as one subgroup , but one finds a richer set of low - energy collective modes associated with the full subgroup structure of so(8 ) . by exploiting the established methodology of fermion dynamical symmetries , it was possible to decouple a collective - pair subspace from the full hilbert space of the problem , permitting exact , analytical , many - body solutions to be obtained in several physically - interesting limits . in addition to exact solutions in specific dynamical symmetry limits , a generalized so(8 ) coherent state approximation has been introduced that permits a broad range of solutions to be obtained even when not in the dynamical symmetry limits . the pairs spanning the collective subspace are shown to be analogous to pairs that have already been discussed at a qualitative level in the graphene literature as defining the possible broken symmetry ground states in the presence of strong electron electron and electron phonon correlations in the @xmath5 landau level . the development here places these pairs on a firm , unified mathematical footing and permits analytical solutions to be developed that explore the possible collective states that previously required numerical simulation for their quantitative description . finally , it has been shown that there are uncanny dynamical symmetry analogies among broken symmetry states for graphene in a strong magnetic field , high temperature superconductors , and strongly collective states in atomic nuclei . on the one hand this has the practical utility of allowing technology already developed in one field to be adapted easily to another . on the other hand , it implies a deep and intriguing mathematical affinity among physical problems that are not usually viewed as having more than a superficial connection . _ acknowledgments._* * this work is supported by the basque government ( grant no . it472 - 10 ) , the spanish micinn ( project no . fis2012 - 36673-c03 - 03 ) , and the basque country university ufi ( project 11/55 - 01 - 2013 ) . partial support was provided by lightcone interactive llc . the formalism discussed in this paper was introduced by postulating a set of physical operators that were shown explicitly to generate an so(8 ) lie algebra under commutation . it is illuminating to consider a somewhat different perspective on the motivation for introducing an so(8 ) symmetry for the graphene problem . it is well known that in a fermionic space having @xmath275 degrees of freedom the most general set of bilinear products @xmath588 of creation annihilation operators and their hermitian conjugates generates an su(@xmath275 ) lie algebra under commutation @xcite . ( physically , the restriction to bilinear products corresponds to limiting consideration to two - body interactions . ) likewise , it is well known that adding to this particle hole operator set the most general pair creation and annihilation operators @xmath589 and @xmath590 extends the su(@xmath275 ) algebra to so(@xmath591 ) . this extension is useful because sometimes more is less . the advantage of expanding the space from su(@xmath275 ) , with @xmath592 generators , to so(@xmath591 ) , with @xmath593 generators , is that the added pair operators permit the definition of a ( collective ) subspace of the full hilbert space spanned by the products of pair creation operators acting on the pair vacuum . if an effective hamiltonian is then constructed by writing the most general polynomial in the casimir invariants of all groups in the subgroup chains of the highest symmetry so(@xmath591 ) ( restricted to quadratic casimir invariants if one considers only two - body interactions ) , it will correspond to the most general hamiltonian that can be written in the collective subspace , and will be diagonal in the collective subspace basis for specific dynamical symmetry subgroup chains . thus , the schrdinger equation can be solved analytically in the symmetry limits defined by each dynamical symmetry subgroup chain , and even away from the symmetry limits it can be solved analytically in coherent - state approximation . the su(@xmath275 ) particle hole algebra alone can be used to construct a hamiltonian that commutes with its generators , and the corresponding schrdinger equation can be solved analytically . but when one considers the realistic case of adding symmetry - breaking terms to the hamiltonian that do not commute with the generators of su(@xmath275 ) , the best that can be done analytically is to assume that these terms are small and that the physical solutions can be treated as small perturbations around the symmetric solution . for the non - perturbative case where the added terms lead to spontaneously broken symmetry and new possible ground states that are not connected perturbatively to the symmetric ground state , one has no systematic way to construct the new ground state from the symmetric one except to guess it . but since the su(@xmath275)-symmetric solution is not connected analytically to non - perturbative broken symmetry solutions , the symmetries of the unperturbed ground state could be a poor guide to guessing the nature of the broken - symmetry states and one must rely on numerical solutions or other approximations and not symmetry to determine their properties . let us now apply this general discussion to quantum hall magnetism in graphene . as discussed in [ h : symmetriesfqhegraphene ] , it is widely accepted that an approximate su(@xmath275 ) particle hole symmetry with @xmath594 is relevant in the @xmath5 landau level because the dominant long - range coulomb interaction is su(4)-symmetric . however , the short - range terms that break this symmetry in the effective hamiltonian prevent the su(4 ) symmetry from providing a solution for the broken symmetry ground state , unless it is assumed that those terms only perturb the su(4)-symmetric solution ( small explicit symmetry breaking ) . but experimental evidence suggests that the true ground state of graphene in a strong magnetic field at low temperature breaks su(4 ) symmetry _ spontaneously , not explicitly _ ( see the introduction ) , and therefore is produced by a non - perturbative effect that can not be explained in terms of small fluctuations around the su(4)-symmetric solution . the possible ( spontaneously ) broken - symmetry states for the @xmath5 landau level have been described in terms of the most general sets of electron pairs occupying the two valley isospin and two spin degrees of freedom @xcite , @xmath595 { |0\rangle } , \label{basis1.0}\ ] ] where the vacuum state @xmath239 corresponds to completely filled landau levels for @xmath596 and completely empty landau levels for @xmath597 . each factor in the product @xmath598 creates a pair of electrons in the state @xmath599 at orbital @xmath600 of the @xmath5 ll , with @xmath601 equal to sublattice a or b , @xmath211 equal to spin up or down , and with the valley isospin and sublattice pseudospin identified : @xmath602 and @xmath603 . but the pair creation operators @xmath215 in eq . ( [ basis1.0 ] ) and their hermitian conjugates are not generators of su(4 ) and do not commute with the su(4)-symmetric hamiltonian , and the collective states of the form ( [ basis1.0 ] ) that are of interest in the present context are unlikely to represent small fluctuations around the su(4)-symmetric solution . thus the nature of these collective states is not determined by the su(4 ) symmetry and had to be investigated by numerical calculations using a small basis in prior work @xcite . on the other hand , the so(8 ) pair generators introduced in eq . ( [ coupled1.4 ] ) are included in the most general collective pairs generated by the @xmath604 operators in eq . ( [ basis1.0 ] ) , and include the collective degrees of freedom discussed in ref . @xcite ( see [ sh : otherstates ] and fig . [ fig : s_d_pairs_brillouin ] ) . thus , the so(8 ) fermion dynamical symmetry permits the nature of possible spontaneously broken symmetries to be investigated in terms of _ symmetry properties _ that permit analytical solutions for the broken - symmetry states . hence , the formalism described in this paper may also be introduced by the following logic . the su(4 ) particle hole symmetry generated by the operators @xmath125 defined in eq . ( [ algebra1.2 ] ) is known to provide a good starting point for graphene quantum hall states dominated by the long - range coulomb interaction , but does not describe quantitatively the broken - symmetry modes discussed by kharitonov @xcite in terms of collective pairs resulting from short - range correlations . motivated by the preceding discussion in this appendix , the su(4 ) generator set may be extended to include the possible pair creation and pair annihilation operators operating in the space corresponding to the indices in the su(4 ) basis . by the general su(@xmath275 ) @xmath605 so(@xmath591 ) extension discussed above , this gives the so(8 ) lie algebra of eq . ( [ algebra1.3 ] ) . hence , expanding the algebra from su(4 ) to so(8 ) introduces the capability to explore _ analytically _ the possible collective states following from perturbation of the su(4 ) quantum hall ferromagnet by short - range interactions that break su(4 ) both explicitly and spontaneously . this appendix collects some useful transformations among the several bases that have been employed in this paper . for brevity , in the following @xmath606 will be termed the nuclear so(8 ) basis and @xmath607 will be termed the graphene so(8 ) basis . in transforming from the nuclear so(8 ) basis to the graphene so(8 ) basis the particle number ( charge ) operator @xmath17 or @xmath266 and the 12 pairing operators @xmath608 are retained , but the 15 su(4 ) generators @xmath609 in the nuclear representation are replaced with the 15 su(4 ) generators @xmath610 defined in the graphene representation of eq . ( [ algebra1.4 ] ) . the explicit transformation from the @xmath611 generators to the @xmath612 generators is given by @xmath613 \label{ngx10 } \\ { \pi_{yx } } & = \tfrac{i}{2}\left [ \sqrt{\tfrac{2}{5 } } p^3_{-3 } + p^3_3 \right . \nonumber \\ & \left . + \sqrt{\tfrac{2}{5}}\left(p^1_{-1 } + p^1_1\right ) - \sqrt{\tfrac{3}{5}}\left(p^3_1 + p^3_{-1}\right)\right ] \label{ngx11 } \\ { \pi_{zx } } & = -\tfrac{1}{\sqrt2 } \left ( p^3_2 + p^3_{-2}\right ) \label{ngx12 } \\ { \pi_{xy } } & = \tfrac{i}{2}\left[\sqrt{\tfrac{2}{5 } } p^3_{-3 } + p^3_3 \right . \nonumber \\ & \left.- \sqrt{\tfrac{2}{5}}\left(p^1_{-1 } + p^1_1\right ) + \sqrt{\tfrac{3}{5}}\left(p^3_1 + p^3_{-1}\right)\right ] \label{ngx13 } \\ { \pi_{yy } } & = \tfrac{1}{2}\left [ \vphantom{\sqrt{\tfrac{2}{5 } } } -p^3_{-3 } + p^3_3 \right . \nonumber \\ & \left.- \sqrt{\tfrac{2}{5}}\left(p^1_1 - p^1_{-1}\right ) + \sqrt{\tfrac{3}{5}}\left(p^3_1 - p^3_{-1}\right)\right ] \label{ngx14 } \\ { \pi_{zy } } & = -\tfrac{i}{\sqrt2 } \left ( p^3_2 - p^3_{-2}\right ) \label{ngx15}\end{aligned}\ ] ] in eqs . ( [ algebra1.8])([algebra1.11 ] ) the graphene basis @xmath137 has been expressed in terms of the generators @xmath125 defined in eq . ( [ algebra1.2 ] ) . the inverse transformations giving the @xmath125 in terms of the @xmath614 are where the unlisted operators may be obtained from @xmath616 and the diagonal operators have been assumed to obey the u(4 ) constraint @xmath617 with @xmath618 the total particle number and @xmath128 the total pair degeneracy given by eq . ( [ degen1.2 ] ) . since from eq . ( [ orderp1.1 ] ) the order parameters for the quantum hall ground states are functions of the expectation values for the number operators @xmath619 specifying the population of the four basis states in table [ tb : quantumnumbermapping ] and fig . [ fig : graphenebasis ] labeled by the index @xmath100 , it is useful to have explicit expressions for them in terms of the @xmath260 operators . these are @xmath620 n_2 & = \tfrac14 n + \tfrac{\sqrt5}{10}p^1_0 - \tfrac12 p^2_0 - \tfrac{3\sqrt5}{10}p^3_0 \label{ptoi2 } \\[2pt ] n_3 & = \tfrac14 n - \tfrac{\sqrt5}{10}p^1_0 -\tfrac12 p^2_0 + \tfrac{3\sqrt5}{10}p^3_0 \label{ptoi3 } \\[2pt ] n_4 & = \tfrac14 n - \tfrac{3\sqrt5}{10}p^1_0 + \tfrac12 p^2_0 -\tfrac{\sqrt5}{10}p^3_0\end{aligned}\ ] ] f. t. arecchi , e courtens , r. gilmore , and h. thomas , phys . * a6 * , 2211 ( 1972 ) . r. gilmore , ann . phys . * 74 * , 391 ( 1972 ) . a. m. perelomov , commun . * 26 * , 222 ( 1972 ) . r. gilmore , rev . de fisica * 23 * , 143 ( 1974 ) . zhang , d. h. feng , and r. gilmore , _ rev . mod.phys._ * 62 * , 867 ( 1990 ) . a. f. young , c. r. dean , l. wang , h. ren , p. cadden - zimansky , k. watanabe , t. taniguchi , j. hone , k. l. shepard and p. kim nature physics * 8 * , 550556 ( 2012 ) . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: mixture models are common tools in statistical pattern recognition @xcite . they offer a mathematical basis to explain data in fields as diverse as astronomy , biology , ecology , engineering , and economics , amongst many others @xcite . a mixture model is composed of component probabilistic models ; a component may variously correspond to a subtype , kind , species , or subpopulation of the observed data . these models aid in the identification of hidden patterns in the data through sound probabilistic formalisms . mixture models have been extensively used in machine learning tasks such as classification and unsupervised learning @xcite . formally , mixture modelling involves representing a distribution of data as a weighted sum of individual probability distributions . specifically , the problem we consider here is to model the observed data using a mixture @xmath0 of probability distributions of the form : @xmath1 where @xmath2 is a @xmath3-dimensional datum , @xmath4 is the number of mixture components , @xmath5 and @xmath6 are the weight and probability density of the @xmath7 component respectively ; the weights are positive and sum to one . the problem of modelling some observed data using a mixture distribution involves determining the number of components , @xmath4 , and estimating the mixture parameters . inferring the optimal number of mixture components involves the difficult problem of balancing the trade - off between two conflicting objectives : low _ hypothesis complexity _ as determined by the number of components _ and _ their respective parameters , versus good quality of _ fit _ to the observed data . generally a hypothesis with more free parameters can fit observed data better than a hypothesis with fewer free parameters . a number of strategies have been used to control this balance as discussed in section [ sec : mixture_existing_methods ] . these methods provide varied formulations to assess the mixture components and their ability to explain the data . methods using the minimum message length criterion @xcite , a bayesian method of inductive inference , have been proved to be effective in achieving a reliable balance between these conflicting aims @xcite . although much of the literature concerns the theory and application of gaussian mixtures @xcite , mixture modelling using other probability distributions has been widely used . some examples are : poisson models @xcite , exponential mixtures @xcite , laplace @xcite , t - distribution @xcite , weibull @xcite , kent @xcite , von mises - fisher @xcite , and many more . @xcite provide a comprehensive summary of finite mixture models and their variations . the use of gaussian mixtures in several research disciplines has been partly motivated by its computational tractability @xcite . for datasets where the _ direction _ of the constituent vectors is important , gaussian mixtures are inappropriate and distributions such as the von mises - fisher may be used @xcite . in any case , whatever the kind of distribution used for an individual component , one needs to estimate the parameters of the mixture , and provide a sound justification for selecting the appropriate number of mixture components . software for mixture modelling relies on the following elements : 1 . an _ estimator _ of the parameters of each component of a mixture , 2 . an _ objective function _ , that is a cost or score , that can be used to compare two hypothetical mixtures and decide which is better , and 3 . a _ search strategy _ for the best number of components and their weights . traditionally , parameter estimation is done by strategies such as maximum likelihood ( ml ) or bayesian maximum _ a posteriori _ probability ( map ) estimation . in this work , we use the bayesian minimum message length ( mml ) principle . unlike map , mml estimators are invariant under non - linear transformations of the data @xcite , and unlike ml , mml considers the number and precision of a model s parameters . it has been used in the inference of several probability distributions @xcite . mml - based inference operates by considering the problem as encoding first the parameter estimates and then the data given those estimates . the parameter values that result in the least _ overall _ message length to explain the whole data are taken as the mml estimates for an inference problem . the mml scheme thus incorporates the cost of stating parameters into model selection . it is self evident that a continuous parameter value can only be stated to some finite precision ; the cost of encoding a parameter is determined by its prior and the precision . ml estimation ignores the cost of stating a parameter and map based estimation uses the probability _ density _ of a parameter instead of its probability measure . in contrast , the mml inference process calculates the optimal precision to which parameters should be stated and a probability value of the corresponding parameters is then computed . this is used in the computation of the message length corresponding to the parameter estimates . thus , models with varying parameters are evaluated based on their resultant total message lengths . we use this characteristic property of mml to evaluate mixtures containing different numbers of components . although there have been several attempts to address the challenges of mixture modelling , the existing methods are shown to have some limitations in their formalisms ( see section [ sec : mixture_existing_methods ] ) . in particular , some methods based on mml are incomplete in their formulation . we aim to rectify these drawbacks by proposing a comprehensive mml formulation and develop a search heuristic that selects the number of mixture components based on the proposed formulation . to demonstrate the effectiveness of the proposed search and parameter estimation , we first consider modelling problems using gaussian mixtures and extend the work to include relevant discussion on mixture modelling of directional data using von mises - fisher distributions . the importance of gaussian mixtures in practical applications is well established . for a _ given _ number of components , the conventional method of estimating the parameters of a mixture relies on the expectation - maximization ( em ) algorithm @xcite . the standard em is a local optimization method , is sensitive to initialization , and in certain cases may converge to the boundary of parameter space @xcite . previous attempts to infer gaussian mixtures based on the mml framework have been undertaken using simplifying assumptions , such as the covariance matrices being diagonal @xcite , or coarsely approximating the probabilities of mixture parameters @xcite . further , the search heuristic adopted in some of these methods is to run the em several times for different numbers of components , @xmath4 , and select the @xmath4 with the best em outcome @xcite . a search method based on iteratively _ deleting _ components has been proposed by @xcite . it begins by assuming a very large number of components and selectively eliminates components deemed redundant ; there is no provision for recovering from deleting a component in error . in this work , we propose a search method which selectively _ splits , deletes _ , or _ merges _ components depending on improvement to the mml objective function . the operations , combined with em steps , result in a sensible redistribution of data between the mixture components . as an example , a component may be split into two children , and at a later stage , one of the children may be merged with another component . unlike the method of @xcite , our method starts with a one - component mixture and alters the number of components in subsequent iterations . this avoids the overhead of dealing with a large number of components unless required . the proposed search heuristic can be used with probability distributions for which the mml expressions to calculate message lengths for estimates and for data given those estimates are known . as an instance of this , section [ sec : search_method ] discusses modelling directional data using the von mises - fisher distributions . directional statistics have garnered support recently in real data mining problems where the _ direction _ , not magnitude , of vectors is significant . examples of such scenarios are found in earth sciences , meteorology , physics , biology , and other fields @xcite . the statistical properties of directional data have been studied using several types of distributions @xcite , often described on surfaces of compact manifolds , such as the sphere , ellipsoid , torus _ etc_. the most fundamental of these is the von mises - fisher ( vmf ) distribution which is analogous to a symmetric multivariate gaussian distribution , wrapped around a unit hypersphere @xcite . the probability density function of a vmf distribution with parameters @xmath8 ( mean direction , concentration parameter ) for a random unit vector @xmath9 on a @xmath10- dimensional hypersphere @xmath11 is given by : @xmath12 where @xmath13 is the normalization constant and @xmath14 is a modified bessel function of the first kind and order @xmath15 . the estimation of the parameters of the vmf distribution is often done using maximum likelihood . however , the complex nature of the mathematical form presents difficulty in estimating the concentration parameter @xmath16 . this has lead to researchers using many different approximations , as discussed in section [ sec : vmf_existing_methods ] . most of these methods perform well when the amount of data is large . at smaller sample sizes , they result in inaccurate estimates of @xmath16 , and are thus unreliable . we demonstrate this by the experiments conducted on a range of sample sizes . the problem is particularly evident when the dimensionality of the data is large , also affecting the application in which it is used , such as mixture modelling . we aim to rectify this issue by using mml estimates for @xmath16 . our experiments section demonstrates that the mml estimate of @xmath16 provides a more reliable answer and is an improvement on the current state of the art . these mml estimates are subsequently used in mixture modelling of vmf distributions ( see sections [ sec : search_method ] and [ sec : vmf_applications ] ) . previous studies have established the importance of von mises circular ( two - dimensional ) and von mises - fisher ( three - dimensional and higher ) mixtures , and demonstrated applications to clustering of protein dihedral angles @xcite , large - scale text clustering @xcite , and gene expression analyses @xcite . the merit of using cosine based similarity metrics , which are closely related to the vmf , for clustering high dimensional text data has been investigated in @xcite . for text clustering , there is evidence that vmf mixture models have a superior performance compared to other statistical distributions such as normal , multinomial , and bernoulli @xcite . movmf is a widely used package to perform clustering using vmf distributions @xcite . + * contributions : * the main contributions of this paper are as follows : * we derive the analytical estimates of the parameters of a multivariate gaussian distribution with full covariance matrix , using the mml principle @xcite . * we derive the expression to infer the concentration parameter @xmath16 of a generic @xmath3-dimensional vmf distribution using mml - based estimation . we demonstrate , through a series of experiments , that this estimate outperforms the previous ones , therefore making it a reliable candidate to be used in mixture modelling . * a generalized mml - based search heuristic is proposed to infer the optimal number of mixture components that would best explain the observed data ; it is based on the search used in various versions of the snob classification program @xcite . we compare it with the work of @xcite and demonstrate its effectiveness . * the search implements a generic approach to mixture modelling and allows , in this instance , the use of @xmath3-dimensional gaussian and vmf distributions under the mml framework . it infers the optimal number of mixture components , and their corresponding parameters . * further , we demonstrate the effectiveness of mml mixture modelling through its application to high dimensional text clustering and clustering of directional data that arises out of protein conformations . the rest of the paper is organized as follows : sections [ sec : gaussian_estimates ] and [ sec : vmf_existing_methods ] describe the respective estimators of gaussian and vmf distributions that are commonly used . section [ sec : mml_framework ] introduces the mml framework for parameter estimation . section [ sec : mml_est_derivations ] outlines the derivation of the mml parameter estimates of multivariate gaussian and vmf distributions . section [ sec : mml_mixture_modelling ] describes the formulation of a mixture model using mml and the estimation of the mixture parameters under the framework . section [ sec : mixture_existing_methods ] reviews the existing methods for selecting the mixture components . section [ sec : search_method ] describes our proposed approach to determine the number of mixture components . section [ sec : gaussian_experiments ] depicts the competitive performance of the proposed mml - based search through experiments conducted with gaussian mixtures . section [ sec : vmf_experiments ] presents the results for mml - based vmf parameter estimation and mixture modelling followed by results supporting its applications to text clustering and protein structural data in section [ sec : vmf_applications ] . the probability density function @xmath17 of a @xmath3-variate gaussian distribution is given as @xmath18 where @xmath19 , @xmath20 are the respective mean , ( symmetric ) covariance matrix of the distribution , and @xmath21 is the determinant of the covariance matrix . the traditional method to estimate the parameters of a gaussian distribution is by maximum likelihood . given data @xmath22 , where @xmath23 , the log - likelihood @xmath24 is given by @xmath25 to compute the maximum likelihood estimates , equation needs to be _ maximized_. this is achieved by computing the gradient of the log - likelihood function with respect to the parameters and solving the resultant equations . the _ gradient vector _ of @xmath24 with respect to @xmath19 and the _ gradient matrix _ of @xmath24 with respect to @xmath20 are given below . @xmath26 the maximum likelihood estimates are then computed by solving @xmath27 and @xmath28 and are given as : @xmath29 @xmath30 is known to be a biased estimate of the covariance matrix @xcite and issues related with its use in mixture modelling have been documented in @xcite and @xcite . an unbiased estimator of @xmath20 was proposed by @xcite and is given below . @xmath31 in addition to the maximum likelihood estimates , bayesian inference of gaussian parameters involving conjugate priors over the parameters has also been dealt with in the literature @xcite . however , the unbiased estimate of the covariance matrix , as determined by the sample covariance ( equation ) , is typically used in the analysis of gaussian distributions . for a von mises - fisher ( vmf ) distribution @xmath17 characterized by equation , and given data @xmath22 , such that @xmath32 , the log - likelihood @xmath24 is given by @xmath33 where @xmath34 is the sample size and @xmath35 ( the vector sum ) . let @xmath36 denote the magnitude of the resultant vector @xmath37 and let @xmath38 and @xmath39 be the maximum likelihood estimators of @xmath40 and @xmath16 respectively . under the condition that @xmath38 is a unit vector , the maximum likelihood estimates are obtained by maximizing @xmath24 as follows : @xmath41 solving the non - linear equation : @xmath42 yields the corresponding maximum likelihood estimate where @xmath43 represents the ratio of bessel functions . because of the difficulties in analytically solving equation ( [ eqn : ml_estimates ] ) , there have been several approaches to approximating @xmath39 @xcite . each of these methods is an improvement over their respective predecessors . @xcite is an improvement over the estimate proposed by @xcite . @xcite is an improvement over @xcite and @xcite fares better when compared to @xcite . the methods are summarized below . the approximation given by equation ( [ eqn : banerjee_approx ] ) is due to @xcite and provides an easy to use expression for @xmath39 . the formula is very appealing as it eliminates the need to evaluate complex bessel functions . @xcite demonstrated that this approximation yields better results compared to the ones suggested in @xcite . it is an empirical approximation which can be used as a starting point in estimating the root of equation ( [ eqn : ml_estimates ] ) . @xmath44 the approximation given by equation ( [ eqn : tanabe_approx ] ) is due to @xcite . the method utilizes the properties of bessel functions to determine the lower and upper bounds for @xmath39 and uses a fixed point iteration function in conjunction with linear interpolation to approximate @xmath39 . the bounds for @xmath39 are given by @xmath45 @xcite proposed to use a fixed point iteration function defined as @xmath46 and used this to approximate @xmath39 as @xmath47 this a heuristic approximation provided by @xcite . it involves refining the approximation given by @xcite ( equation ( [ eqn : banerjee_approx ] ) ) by performing two iterations of newton s method . @xcite demonstrate that this approximation fares well when compared to the approximation proposed by @xcite . the following two iterations result in @xmath48 , the approximation proposed by @xcite : @xmath49 @xmath50 this approximation provided by @xcite uses halley s method which is the second order expansion of taylor s series of a given function @xmath51 . the higher order approximation results in a more accurate estimate as demonstrated by @xcite . the iterative halley s method is truncated after iterating through two steps of the root finding algorithm ( similar to that done by @xcite ) . the following two iterations result in @xmath52 , the approximation proposed by @xcite : @xmath53 @xmath54 the common theme in all these methods is that they try to approximate the maximum likelihood estimate governed by equation ( [ eqn : ml_estimates ] ) . it is to be noted that the maximum likelihood estimators ( of concentration parameter @xmath16 ) have considerable bias @xcite . to counter this effect , we explore the minimum message length based estimation procedure . this bayesian method of estimation not only results in an unbiased estimate but also provides a framework to choose from several competing models @xcite . through a series of empirical tests , we demonstrate that the mml estimate is more reliable than any of the contemporary methods . @xcite have demonstrated the superior performance of the mml estimate for a three - dimensional vmf distribution . we extend their work to derive the mml estimators for a generic @xmath3-dimensional vmf distribution and compare its performance with the existing methods . @xcite developed the first practical criterion for model selection to be based on information theory . the resulting framework provides a rigorous means to objectively compare two competing hypotheses and , hence , choose the best one . as per bayes s theorem , @xmath55 where @xmath56 denotes some observed data , and @xmath57 some hypothesis about that data . further , @xmath58 is the joint probability of data @xmath56 and hypothesis @xmath57 , @xmath59 is the prior probability of hypothesis @xmath57 , @xmath60 is the prior probability of data @xmath56 , @xmath61 is the posterior probability of @xmath57 given @xmath56 , and @xmath62 is the likelihood . mml uses the following result from information theory @xcite : given an event or outcome @xmath63 whose probability is @xmath64 , the length of the optimal lossless code @xmath65 to represent that event requires @xmath66 bits . applying shannon s insight to bayes s theorem , @xcite got the following relationship between conditional probabilities in terms of optimal message lengths : @xmath67 as a result , given two competing hypotheses @xmath57 and @xmath68 , @xmath69 @xmath70 gives the posterior log - odds ratio between the two competing hypotheses . equation ( [ eqn : shannon_bayes_msg ] ) can be intrepreted as the _ total _ cost to encode a message comprising the hypothesis @xmath57 and data @xmath56 . this message is composed over into two parts : 1 . _ first part : _ the hypothesis @xmath57 , which takes @xmath71 bits , 2 . _ second part : _ the observed data @xmath56 using knowledge of @xmath57 , which takes @xmath72 bits . clearly , the message length can vary depending on the complexity of @xmath57 and how well it can explain @xmath56 . a more complex @xmath57 may fit ( i.e. , explain ) @xmath56 better but take more bits to be stated itself . the trade - off comes from the fact that ( hypothetically ) transmitting the message requires the encoding of both the hypothesis and the data given the hypothesis , that is , the model complexity @xmath71 and the goodness of fit @xmath72 . our proposed method of parameter estimation uses the mml inference paradigm . it is a bayesian method which has been applied to infer the parameters of several statistical distributions @xcite . we apply it to infer the parameter estimates of multivariate gaussian and vmf distributions . @xcite introduced a generalized scheme to estimate a vector of parameters @xmath73 of any distribution @xmath17 given data @xmath56 . the method involves choosing a reasonable prior @xmath74 on the hypothesis and evaluating the _ determinant _ of the fisher information matrix @xmath75 of the _ expected _ second - order partial derivatives of the negative log - likelihood function , @xmath76 . the parameter vector @xmath73 that minimizes the message length expression ( equation ( [ eqn : two_part_msg ] ) ) is the mml estimate according to @xcite . @xmath77 where @xmath78 is the number of free parameters in the model , and @xmath79 is the lattice quantization constant @xcite in @xmath78-dimensional space . the total message length @xmath80 in mml framework is composed of two parts : 1 . the statement cost of encoding the parameters , @xmath81 and 2 . the cost of encoding the data given the parameters , @xmath82 . a concise description of the mml method is presented in @xcite . we note here a few key differences between mml and ml / map based estimation methods . in maximum likelihood estimation , the statement cost of parameters is ignored , in effect considered constant , and minimizing the message length corresponds to minimizing the negative log - likelihood of the data ( the second part ) . in map based estimation , a probability _ density _ rather than the probability measure is used . continuous parameters can necessarily only be stated only to finite precision . mml incorporates this in the framework by determining the region of uncertainty in which the parameter is located . the value of @xmath83 gives a measure of the volume of the region of uncertainty in which the parameter @xmath73 is centered . this multiplied by the probability density @xmath74 gives the _ probability _ of a particular @xmath73 and is _ proportional _ to @xmath84 . this probability is used to compute the message length associated with encoding the continuous valued parameters ( to a finite precision ) . based on the mml inference process discussed in section [ sec : mml_framework ] , we now proceed to formulate the message length expressions and derive the parameter estimates of gaussian and von mises - fisher distributions . the mml framework requires the statement of parameters to a finite precision . the optimal precision is related to the fisher information and in conjunction with a reasonable prior , the probability of parameters is computed . a flat prior is usually chosen on each of the @xmath3 dimensions of @xmath19 @xcite and a conjugate inverted wishart prior is chosen for the covariance matrix @xmath20 @xcite . the joint prior density of the parameters is then given as @xmath85 the computation of the fisher information requires the evaluation of the second order partial derivatives of @xmath86 . let @xmath87 represent the determinant of the fisher information matrix . this is approximated as the product of @xmath88 and @xmath89 @xcite , where @xmath88 and @xmath89 are the respective determinants of fisher information matrices due to the parameters @xmath19 and @xmath20 . differentiating the gradient vector in equation with respect to @xmath19 , we have : @xmath90 to compute @xmath89 , @xcite derived an analytical expression using the theory of matrix derivatives based on matrix vectorization @xcite . let @xmath91\ , \forall 1 \leq i , j \leq d$ ] where @xmath92 denotes the element corresponding to the @xmath93 row and @xmath94 column of the covariance matrix . let @xmath95 be the vector containing the @xmath96 free parameters that completely describe the symmetric matrix @xmath20 . then , the fisher information due to the vector of parameters @xmath97 is equal to @xmath89 and is given by equation @xcite . @xmath98 multiplying equations and , we have @xmath99 to derive the message length expression to encode data using a certain @xmath100 , substitute equations , , and , in equation using the number of free parameters of the distribution as @xmath101 . hence , @xmath102 to obtain the mml estimates of @xmath19 and @xmath20 , equation needs to be minimized . the mml estimate of @xmath19 is same as the maximum likelihood estimate ( given in equation ) . to compute the mml estimate of @xmath20 , we need to compute the gradient matrix of @xmath103 with respect to @xmath20 and is given by equation @xmath104 the mml estimate of @xmath20 is obtained by solving @xmath105 ( given in equation ) . @xmath106 we observe that the mml estimate @xmath107 is same as the _ unbiased _ estimate of the covariance matrix @xmath20 , thus , lending credibility for its preference over the traditional ml estimate ( equation ) . parameter estimates for two and three - dimensional vmf have been explored previously @xcite . mml estimators of three - dimensional vmf were explored in @xcite , where they demonstrate that the mml - based inference compares favourably against the traditional ml and map based estimation methods . we use the @xcite method to formulate the objective function ( equation ) corresponding to a generic vmf distribution . regarding choosing a reasonable prior ( in the absence of any supporting evidence ) for the parameters @xmath108 of a vmf distribution , @xcite and @xcite suggest the use of the following _ colourless _ prior that is uniform in direction , normalizable and locally uniform at the cartesian origin in @xmath16 " : @xmath109 regarding evaluating the fisher information , @xcite argue that in the general @xmath3-dimensional case , @xmath110 where @xmath111 and @xmath112 are described by equations and respectively . substituting equations , and in equation with number of free parameters @xmath113 , we have the net message length expression : @xmath114 to obtain the mml estimates of @xmath19 and @xmath16 , equation needs to be minimized . the estimate for @xmath19 is same as the maximum likelihood estimate ( equation ) . the resultant equation in @xmath16 that needs to be minimized is then given by : @xmath115 to obtain the mml estimate of @xmath16 , we need to differentiate equation ( [ eqn : i_kappa ] ) and set it to zero . @xmath116 the non - linear equation : @xmath117 does not have a closed form solution . we try both the newton and halley s method to find an approximate solution . we discuss both variants and comment on the effects of the two approximations in the experimental results . to be fair and consistent with @xcite and @xcite , we use the initial guess of the root as @xmath118 ( equation ( [ eqn : banerjee_approx ] ) ) and iterate twice to obtain the mml estimate . 1 . _ approximation using newton s method : _ @xmath119 2 . _ approximation using halley s method : _ @xmath120 the details of evaluating @xmath121 and @xmath122 are discussed in appendix [ subsec : appendix_derivations ] . equation ( [ eqn : mml_newton_approx ] ) gives the mml estimate ( @xmath123 ) using newton s method and equation ( [ eqn : mml_halley_approx ] ) gives the mml estimate ( @xmath124 ) using halley s method . we used these values of mml estimates in mixture modelling using vmf distributions . [ sec : mml_gaussian_est ] mixture modelling involves representing an observed distribution of data as a weighted sum of individual probability density functions . specifically , the problem we consider here is to model the mixture distribution @xmath0 as defined in equation . for some observed data @xmath125 ( @xmath34 is the sample size ) , and a mixture @xmath0 , the log - likelihood using the mixture distribution is as follows : @xmath126 where @xmath127 , @xmath5 and @xmath6 are the weight and probability density of the @xmath7 component respectively . for a fixed @xmath4 , the mixture parameters @xmath128 are traditionally estimated using a standard _ expectation - maximization_(em ) algorithm @xcite . this is briefly discussed below . the standard em algorithm is based on maximizing the log - likelihood function of the data ( equation ) . the maximum likelihood estimates are then given as @xmath129 . because of the absence of a closed form solution for @xmath130 , a gradient descent method is employed where the parameter estimates are iteratively updated until convergence to some local optimum is achieved @xcite . the em method consists of two steps : * _ e - step _ : each datum @xmath131 has fractional membership to each of the mixture components . these partial memberships of the data points to each of the components are defined using the _ responsibility matrix _ @xmath132 where @xmath133 denotes the conditional probability of a datum @xmath131 belonging to the @xmath7 component . the effective membership associated with each component is then given by @xmath134 * _ m - step _ : assuming @xmath135 be the estimates at some iteration @xmath136 , the expectation of the log - likelihood using @xmath135 and the partial memberships is then _ maximized _ which is tantamount to computing @xmath137 , the updated maximum likelihood estimates for the next iteration @xmath138 . the weights are updated as @xmath139 . the above sequence of steps are repeated until a certain convergence criterion is satisfied . at some intermediate iteration @xmath136 , the mixture parameters are updated using the corresponding ml estimates and are given below . * _ gaussian : _ the ml updates of the mean and covariance matrix are @xmath140 * _ von mises - fisher : _ the resultant vector sum is updated as @xmath141 . if @xmath142 represents the magnitude of vector @xmath143 , then the updated mean and concentration parameter are @xmath144 we will first describe the methodology involved in formulating the mml - based objective function . we will then discuss how em is applied in this context . we refer to the discussion in @xcite to briefly describe the intuition behind mixture modelling using mml . encoding of a message using mml requires the encoding of ( 1 ) the model parameters and then ( 2 ) the data using the parameters . the statement costs for encoding the mixture model and the data can be decomposed into : 1 . encoding the _ number of components _ @xmath4 : in order to encode the message losslessly , it is required to initially state the number of components . in the absence of background knowledge , one would like to model the prior belief in such a way that the probability decreases for increasing number of components . if @xmath145 , then @xmath146 . the prior reflects that there is a difference of one bit in encoding the _ numbers _ @xmath4 and @xmath147 . alternatively , one could assume a uniform prior over @xmath4 within some predefined range . the chosen prior has little effect as its contribution is minimal when compared to the magnitude of the total message length @xcite . 2 . encoding the _ weights _ @xmath148 which are treated as parameters of a multinomial distribution with sample size @xmath149 . the length of encoding all the weights is then given by the expression @xcite : + @xmath150 3 . encoding each of the _ component parameters _ @xmath151 as given by @xmath152 ( discussed in section [ subsec : mml_parameter_estimation ] ) . 4 . encoding the _ data _ : each datum @xmath131 can be stated to a finite precision which is dictated by the accuracy of measurement is a constant value and has no effect on the overall inference process . it is used in order to maintain the theoretical validity when making the distinction between _ probability _ and _ probability density_. ] . if the precision to which each element of a @xmath3-dimensional vector can be stated is @xmath153 , then the _ probability _ of a datum @xmath154 is given as @xmath155 where @xmath156 is the _ probability density _ given by equation . hence , the _ total _ length of its encoding is given by @xmath157 the entire data @xmath56 can now be encoded as : @xmath158 thus , the total message length of a @xmath4 component mixture is given by equation . @xmath159 note that the _ constant _ term includes the lattice quantization constant ( resulting from stating all the model parameters ) in a @xmath78-dimensional space , where @xmath78 is equal to the number of free parameters in the mixture model . the parameters of the mixture model are those that _ minimize _ equation . to achieve this we use the standard em algorithm ( section [ subsec : em_ml ] ) , where , iteratively , the parameters are updated using their respective _ mml estimates_. the component weights are obtained by differentiating equation with respect to @xmath5 under the constraint @xmath160 . the derivation of the mml updates of the weights is shown in appendix [ subsec : wts_mml ] and are given as : @xmath161 the parameters of the @xmath7 component are updated using @xmath162 and @xmath163 ( equations , ) , the partial memberships assigned to the @xmath7 component at some intermediate iteration @xmath136 and and are given below . * _ gaussian : _ the mml updates of the mean and covariance matrix are @xmath164 * _ von mises - fisher : _ the resultant vector sum is updated as @xmath165 . if @xmath142 represents the magnitude of vector @xmath143 , then the updated mean is given by equation . @xmath166 the mml update of the concentration parameter @xmath167 is obtained by solving @xmath168 after substituting @xmath169 and @xmath170 in equation . the em is terminated when the change in the total message length ( improvement rate ) between successive iterations is less than some predefined threshold . the difference between the two variants of standard em discussed above is firstly the objective function that is being optimized . in section [ subsec : em_ml ] , the log - likelihood function is _ maximized _ which corresponds to @xmath171 term in section [ subsec : em_mml ] . equation ( [ eqn : mixture_msglen ] ) includes additional terms that correspond to the cost associated with stating the mixture parameters . secondly , in the m - step , in section [ subsec : em_ml ] , the components are updated using their ml estimates whereas in section [ subsec : em_mml ] , the components are updated using their mml estimates . the standard em algorithms outlined above can be used only when the number of mixture components @xmath4 is fixed or known _ a priori_. even when the number of components are fixed , em has potential pitfalls . the method is sensitive to the initialization conditions . to overcome this , some reasonable start state for the em may be determined by initially clustering the data @xcite . another strategy is to run the em a few times and choose the best amongst all the trials . @xcite point out that , in the case of gaussian mixture modelling , em can converge to the boundary of the parameter space when the corresponding covariance matrix is nearly singular or when there are few initial members assigned to that component . inferring the right " number of mixture components for unlabelled data has proven to be a thorny issue @xcite and there have been numerous approaches proposed that attempt to tackle this problem @xcite . given some observed data , there are infinitely many mixtures that one can fit to the data . any method that aims to selectively determine the optimal number of components should be able to factor the cost associated with the mixture parameters . to this end , several methods based on information theory have been proposed where there is some form of penalty associated with choosing a certain parameter value @xcite . we briefly review some of these methods and discuss the state of the art and then proceed to explain our proposed method . aic in the simplest form adds the _ number _ of free parameters @xmath78 to the negative log - likelihood expression . there are some variants of aic suggested @xcite . however , these variants introduce the same penalty constants for each additional parameter : @xmath172 bic , similar to aic , adds a constant multiple of @xmath173 ( @xmath34 being the sample size ) , for each free parameter in the model . @xcite formulated minimum description length ( mdl ) which formally coincides with bic @xcite . aic and bic / mdl serve as scoring functions to evaluate a model and its corresponding fit to the data . the formulations suggest that the parameter cost associated with adopting a model is dependent only on the number of free parameters and _ not _ on the parameter values themselves . in other words , the criteria consider all models of a particular type ( of probability distribution ) to have the same statement cost associated with the parameters . for example , a generic @xmath3-dimensional gaussian distribution has @xmath175 free parameters . all such distributions will have the same parameter costs regardless of their characterizing means and covariance matrices , which is an oversimplifying assumption which can hinder proper inference . the criteria can be interpreted under the mml framework wherein the first part of the message is a constant multiplied by the number of free parameters . aic and bic formulations can be obtained as approximations to the two - part mml formulation governed by equation @xcite . it has been argued that for tasks such as mixture modelling , where the number of free parameters potentially grows in proportion to the data , mml is known in theory to give consistent results as compared to aic and bic @xcite . to determine the optimal number of mixture components @xmath4 , the aic or bic scores are computed for mixtures with varying values of @xmath4 . the mixture model with the least score is selected as per these criteria . a @xmath3-variate gaussian mixture with @xmath4 number of components has @xmath176 free parameters . all mixtures with a set number of components have the same cost associated with their parameters using these criteria . the mixture complexity is therefore treated as independent of the constituent mixture parameters . in contrast , the mml formulation incorporates the statement cost of losslessly encoding mixture parameters by calculating their relevant probabilities as discussed in section [ sec : mml_mixture_modelling ] . a mml - based scoring function akin to the one shown in equation was used to model gaussian mixtures . however , the authors only consider the specific case of gaussians with diagonal covariance matrices , and fail to provide a general method dealing with full covariance matrices . a rigorous treatment on the selection of number of mixture components @xmath4 is lacking . @xcite experiment with different values of @xmath4 and choose the one which results in the minimum message length . for each @xmath4 , the standard em algorithm ( section [ subsec : em_ml ] ) was used to attain local convergence . the method , also referred to as _ laplace - empirical criterion _ ( lec ) @xcite , uses a scoring function derived using bayesian inference and serves to provide a tradeoff between model complexity and the quality of fit . the parameter estimates @xmath128 are those that result in the minimum value of the following scoring function . @xmath177 where @xmath56 is the dataset , @xmath178 is the negative log - likelihood given the mixture parameters @xmath128 , @xmath4 is the number of mixture components , @xmath3 the dimensionality of the data , @xmath179 are hyperparameters ( which are set to 1 in their experiments ) , @xmath180 is a pre - defined constant or is pre - computed using the entire data , @xmath181 is the hessian matrix which is equivalent to the empirical fisher matrix for the set of component parameters , and @xmath78 is the number of free parameters in the model . the formulation in equation can be obtained as an approximation to the message length expression in equation by identifying the following related terms in both equations . 1 . @xmath182 2 . for a @xmath3-variate gaussian with mean @xmath19 and covariance matrix @xmath20 , the joint prior @xmath183 is calculated as follows : * _ prior on @xmath19 : _ each of the @xmath3 parameters of the mean direction are assumed to be have uniform priors in the range @xmath184 , so that the prior density of the mean is @xmath185 . * _ prior on @xmath20 : _ it is assumed that the prior density is dependent only on the diagonal elements in @xmath20 . each diagonal covariance element is assumed to have a prior in the range @xmath186 so that the prior on @xmath20 is considered to be @xmath187 + the joint prior , is therefore , assumed to be @xmath188 . + thus , @xmath189 + 3 . @xmath190 + 4 . @xmath191 although the formulation is an improvement over the previously discussed methods , there are some limitations due to the assumptions made while proposing the scoring function : * while computing the prior density of the covariance matrix , the off - diagonal elements are ignored . * the computation of the determinant of the fisher matrix is approximated by computing the hessain @xmath192 . it is to be noted that while the hessian is the _ observed information _ ( data dependent ) , the fisher information is the _ expectation _ of the observed information . mml formulation requires the use of the expected value . * further , the approximated hessian was derived for gaussians with diagonal covariances . for gaussians with full covariance matrices , the hessian was approximated by replacing the diagonal elements with the corresponding eigen values in the hessian expression . the empirical fisher computed in this form does not guarantee the charactersitic invariance property of the classic mml method @xcite . the search method used to select the optimal number of components is rudimentary . the optimal number of mixture components is chosen by running the em 10 times for every value of @xmath4 within a given range . an optimal @xmath4 is selected as the one for which the best of the 10 trials results in the least value of the scoring function . the icl criterion _ maximizes _ the _ complete log - likelihood _ ( cl ) given by @xmath193 where @xmath194 is the log - likelihood ( equation ) , @xmath133 is the responsibility term ( equation ) , and @xmath195 if @xmath131 arises from component @xmath196 and zero otherwise . the term @xmath197 is explained as the estimated mean entropy . the icl criterion is then defined as : @xmath198where @xmath78 is the number of free parameters in the model . we observe that similar to bic , the icl scoring function penalizes each free parameter by a constant value and does not account for the model parameters . the search method adopted in this work is similar to the one used by @xcite . the em algorithm is initiated 20 times for each value of @xmath4 with random starting points and the best amongst those is chosen . the method uses the mml criterion to formulate the scoring function given by equation . the formulation can be intrepreted as a two - part message for encoding the model parameters and the observed data . @xmath199 where @xmath200 is the _ number _ of free parameters per component and @xmath5 is the component weight . the scoring function is derived from equation by assuming the prior density of the component parameters to be a jeffreys prior . if @xmath151 is the vector of parameters describing the @xmath7 component , then the prior density @xmath201 @xcite . similarly , a prior for weights would result in @xmath202 . these assumptions are used in the encoding of the parameters which correspond to the first part of the message . we note that the scoring function is consistent with the mml scheme of encoding parameters and the data using those parameters . however , the formulation can be improved by amending the assumptions as detailed in in section [ sec : mml_est_derivations ] . further , the assumptions made in @xcite have the following side effects : * the value of @xmath203 gives the cost of encoding the component parameters . by assuming @xmath201 , the message length associated with using any vector of parameters @xmath151 is essentially treated the same . to avoid this , the use of independent uniform priors over non - informative jeffreys s priors was advocated previously @xcite . the use of jeffreys prior offers certain advantages , for example , not having to compute the fisher information @xcite . however , this is crucial and can not be ignored as it dictates the _ precision of encoding the parameter vector_. @xcite states that jeffreys , while noting the interesting properties of the prior formulation did not advocate its use as a genuine expression of prior knowledge . " + by making this assumption , @xcite _ sidestep _ " the difficulty associated with explicitly computing the fisher information associated with the component parameters . hence , for encoding the parameters of the entire mixture , _ only _ the cost associated with encoding the component weights is considered . * the code length to state each @xmath151 is , therefore , greatly simplified as @xmath204 ( notice the sole dependence on weight @xmath5 ) . @xcite interpret this as being similar to a mdl formulation because @xmath205 gives the expected number of data points generated by the @xmath7 component . this is equivalent to the bic criterion discussed earlier . we note that mdl / bic are highly simplified versions of mml formulation and therefore , equation does not capture the entire essence of complexity and goodness of fit accurately . the method begins by assuming a large number of components and updates the weights iteratively in the em steps as @xmath206 where @xmath207 is the effective membership of data points in @xmath7 component ( equation ) . a component is annihilated when its weight becomes zero and consequently the number of mixture components decreases . we note that the search method proposed by @xcite using the mml criterion is an improvement over the methods they compare against . however , we make the following remarks about their search method . * the method updates the weights as given by equation . during any iteration , if the amount of data allocated to a component is less than @xmath208 , its weight is updated as zero and this component is ignored in subsequent iterations . this imposes a lower bound on the amount of data that can be assigned to each component . as an example , for a gaussian mixture in 10-dimensions , the number of free parameters per component is @xmath209 , and hence the lower bound is 33 . hence , in this exampe , if a component has @xmath210 data , the mixture size is reduced and these data are assigned to some other component(s ) . consider a scenario where there are 50 observed 10 dimensional data points originally generated by a mixture with two components with equal mixing proportions . the method would always infer that there is only one component regardless of the separation between the two components . this is clearly a wrong inference ! ( see section [ subsec : fj_weight_updates_exp2b ] for the relevant experiments ) . * once a component is discarded , the mixture size decreases by one , and it can not be recovered . because the memberships @xmath207 are updated iteratively using an em algorithm and because em might not always lead to global optimum , it is conceivable that the updated values need not always be optimal . this might lead to situations where a component is deleted owing to its low prominence . there is no provision to increase the mixture size in the subsequent stages of the algorithm to account for such behaviour . * the method assumes a large number of initial components in an attempt to be robust with respect to em initialization . however , this places a significant overhead on the computation due to handling several components . _ summary : _ we observe that while all these methods ( and many more ) work well within their defined scope , they are incomplete in achieving the true objective that is to rigorously score models and their ability to fit the data . the methods discussed above can be seen as different approximations to the mml framework . they adopted various simplifying assumptions and approximations . to avoid such limitations , we developed a classic mml formulation , giving the complete message length formulations for gaussian and von mises - fisher distributions in section [ sec : mml_est_derivations ] . secondly , in most of these methods , the search for the optimal number of mixture components is achieved by selecting the mixture that results in the best em outcome out of many trials @xcite . this is not an elegant solution and @xcite proposed a search heuristic which integrates estimation and model selection . a comparative study of these methods is presented in @xcite . their analysis suggested the superior performance of icl @xcite and lec @xcite . later , @xcite demonstrated that their proposed method outperforms the contemporary methods based on icl and lec and is regarded as the current state of the art . we , therefore , compare our method against that of @xcite and demonstrate its effectiveness . with this background , we formulate an alternate search heuristic to infer the optimal number of mixture components which aims to address the above limitations . the space of candidate mixture models to explain the given data is infinitely large . as per the mml criterion ( equation ) , the goal is to search for the mixture that has the smallest overall message length . we have seen in section [ subsec : em_mml ] that if the number of mixture components are fixed , then the em algorithm can be used to estimate the mixture parameters , namely the component weights and the parameters of each component . however , here it is required to search for the optimal _ number _ of mixture components along with the corresponding mixture parameters . our proposed search heuristic extends the mml - based snob program @xcite for unsupervised learning . we define three operations , namely _ split , delete , _ and _ merge _ that can be applied to any component in the mixture . [ algm ] @xmath211 the pseudocode of our search method is presented in algorithm [ algm ] . the basic idea behind the search strategy is to _ perturb _ a mixture from its current suboptimal state to obtain a new state ( if the perturbed mixture results in a smaller message length ) . in general , if a ( current ) mixture has @xmath4 components , it is perturbed using a series of _ split , delete _ , and _ merge _ operations to check for improvement . each component is split and the new @xmath212-component mixture is re - estimated . if there is an improvement ( _ i.e. , _ if there is a decrease in message length with respect to the current mixture ) , the new @xmath212-component mixture is retained . there are @xmath4 splits possible and the one that results in the greatest improvement is recorded ( lines 5 7 in algorithm [ algm ] ) . a component is first split into two sub - components ( children ) which are locally optimized by the em algorithm on the data that belongs to that sole component . the child components are then integrated with the others and the mixture is then optimized to generate a @xmath147 component mixture . the reason for this is , rather than use random initial values for the em , it is better if we start from some already optimized state to reach to a better state . similarly , each of the components is then deleted , one after the other , and the @xmath213-component mixture is compared against the current mixture . there are @xmath4 possible deletions and the best amongst these is recorded ( lines 8 11 in algorithm [ algm ] ) . finally , the components in the current mixture are merged with their closest matches ( determined by calculating the kl - divergence ) and each of the resultant @xmath213-component mixtures are evaluated against the @xmath4 component mixture . the best among these merged mixtures is then retained ( lines 12 15 in algorithm [ algm ] ) . we initially start by assuming a one component mixture . this component is split into two children which are locally optimized . if the split results in a better model , it is retained . for any given @xmath4-component mixture , there might be improvement due to splitting , deleting and/or merging its components . we select the perturbation that best improves the current mixture . this process is repeated until there is no further improvement possible . and the algorithm is continued . the notion of _ best _ or improved mixture is based on the amount of reduction of message length that the perturbed mixture provides . in the current state , the observed data have partial memberships in each of the @xmath4 components . before the execution of each operation , these memberships need to be adjusted and a em is subsequently carried out to achieve an optimum with a different number of components . we will now examine each operation in detail and see how the memberships are affected after each operation . let @xmath214 $ ] be the @xmath215 responsibility ( membership ) matrix and @xmath5 be the weight of @xmath7 component in mixture @xmath0 . 1 . _ split ( line 6 in algorithm [ algm ] ) : _ as an example , assume a component with index @xmath216 and weight @xmath217 in the current mixture @xmath0 is split to generate two child components . the goal is to find two distinct clusters amongst the data associated with component @xmath218 . it is to be noted that the data have fractional memberships to component @xmath218 . the em is therefore , carried out _ within _ the component @xmath218 assuming a _ two - component sub - mixture _ with the data weighted as per their current memberships @xmath219 . the remaining @xmath220 components are untouched . an em is carried out to optimize the two - component sub - mixture . the initial state and the subsequent updates in the maximization - step are described below . + _ parameter initialization of the two - component sub - mixture : _ the goal is to identify two distinct clusters within the component @xmath218 . for _ gaussian _ mixtures , to provide a reasonable starting point , we compute the direction of maximum variance of the parent component and locate two points which are one standard deviation away on either side of its mean ( along this direction ) . these points serve as the initial means for the two children generated due to splitting the parent component . selecting the initial means in this manner ensures they are reasonably apart from each other and serves as a good starting point for optimizing the two - component sub - mixture . the memberships are initialized by allocating the data points to the closest of the two means . once the means and the memberships are initialized , the covariance matrices of the two child components are computed . + there are conceivably several variations to how the two - component sub - mixture can be initialized . these include random initialization , selecting two data points as the initial component means , and many others . however , the reason for selecting the direction of maximum variance is to utilize the available characteristic of data , _ i.e. , _ the distribution within the component @xmath218 . for _ von mises - fisher _ mixtures , the maximum variance strategy ( as for gaussian mixtures ) can not be easily adopted , as the data is distributed on the hypershpere . hence , in this work , we randomly allocate data memberships and compute the components ( initial ) parameters . + once the parameters of the sub - mixture are initialized , an em algorithm is carried out ( just for the sub - mixture ) with the following maximization - step updates . let @xmath221 $ ] be the @xmath222 responsibility matrix for the two - component sub - mixture . for @xmath223 , let @xmath224 be the effective memberships of data belonging to the two child components , let @xmath225 be the weights of the child components within the sub - mixture , and let @xmath226 be the parameters describing the child components . * the effective memberships are updated as given by equation . @xmath227 * as the sub - mixture comprises of two child components , substitute @xmath228 in equation to obtain the updates for the weights . these are given by equation . @xmath229 * for _ gaussian _ mixtures , the component parameters @xmath230 are updated as follows : @xmath231 * for _ von mises - fisher _ mixtures , the component parameters @xmath232 are updated as follows : @xmath233 @xmath234 represents the magnitude of vector @xmath235 . the update of the concentration parameter @xmath236 is obtained by solving @xmath237 after substituting @xmath238 and @xmath239 in equation . + the difference between the em updates in equations , and equations , is the presence of the coefficient @xmath240 with each @xmath131 . since we are considering the sub - mixture , the original responsibility @xmath241 is multiplied by the responsibility within the sub - mixture @xmath242 to quantify the influence of datum @xmath131 to each of the child components . + after the sub - mixture is locally optimized , it is integrated with the untouched @xmath220 components of @xmath0 to result in a @xmath147 component mixture @xmath243 . an em is finally carried out on the combined @xmath147 components to estimate the parameters of @xmath243 and result in an optimized @xmath212-component mixture as follows . + _ em initialization for @xmath243 : _ usually , the em is started by a random initialization of the members . however , because the two - component sub - mixture is now optimal and the @xmath220 components in @xmath0 are also in an optimal state , we exploit this situation to initialize the em ( for @xmath243 ) with a reasonable starting point . as mentioned above , the component with index @xmath218 with component weight @xmath217 is split . upon integration , the ( child ) components that replaced component @xmath218 will now correspond to indices @xmath244 and @xmath245 in the new mixture @xmath243 . let @xmath246 \,\forall 1 \leq i \leq n , 1 \leq j \leq m+1 $ ] be the responsibility matrix for the new mixture @xmath243 and let @xmath247 be the component weights in @xmath243 . * _ component weights : _ the weights are initialized as follows : @xmath248 * _ memberships : _ the responsibility matrix @xmath249 is initialized for all data @xmath250 as follows : @xmath251 where @xmath252 are the effective memberships of the components in @xmath243 . + with these starting points , the parameters of @xmath243 are estimated using the traditional em algorithm with updates in the maximization - step given by equations , , and . the em results in local convergence of the @xmath212-component mixture . if the resultant message length of encoding data using @xmath243 is lower than that due to @xmath0 , that means the perturbation of @xmath0 because of splitting component @xmath218 resulted in a new mixture @xmath243 that compresses the data better , and hence , is a better mixture model to explain the data . 2 . _ delete ( line 10 in algorithm [ algm ] ) : _ the goal here is to remove a component from the current mixture and check whether it results in a better mixture model to explain the observed data . assume the component with index @xmath218 and the corresponding weight @xmath217 is to be deleted from @xmath0 to generate a @xmath220 component mixture @xmath243 . once deleted , the data memberships of the component need to be redistributed between the remaining components . the redistribution of data results in a good starting point to employ the em algorithm to estimate the parameters of @xmath243 as follows . + _ em initialization for @xmath243 : _ let @xmath246 $ ] be the @xmath253 responsibility matrix for the new mixture @xmath243 and let @xmath247 be the weight of @xmath7 component in @xmath243 . * _ component weights : _ the weights are initialized as follows : @xmath254 it is to be noted that @xmath255 because the mml update expression in the m - step for the component weights always ensures non - zero weights during every iteration of the em algorithm ( see equation ) . * _ memberships : _ the responsibility matrix @xmath249 is initialized for all data @xmath250 as follows : @xmath256 where @xmath252 are the effective memberships of the components in @xmath243 . it is possible for a datum @xmath131 to have complete membership in component @xmath218 ( _ i.e. , _ @xmath257 ) , in which case , its membership is equally distributed among the other @xmath220 components ( _ i.e. , _ @xmath258 ) . + with these readjusted weights and memberships , and the constituent @xmath220 components , the traditional em algorithm is used to estimate the parameters of the new mixture @xmath243 . if the resultant message length of encoding data using @xmath243 is lower than that due to @xmath0 , that means the perturbation of @xmath0 because of deleting component @xmath218 resulted in a new mixture @xmath243 with better explanatory power , which is an improvement over the current mixture . 3 . _ merge ( line 14 in algorithm [ algm ] ) : _ the idea is to join a pair of components of @xmath0 and determine whether the resulting @xmath213-component mixture @xmath243 is any better than the current mixture @xmath0 . one strategy to identify an improved mixture model would be to consider merging all possible pairs of components and choose the one which results in the greatest improvement . this would , however , lead to a runtime complexity of @xmath259 , which could be significant for large values of @xmath4 . another strategy is to consider merging components which are close " to each other . for a given component , we identify its _ closest _ component by computing the kullback - leibler ( kl ) distance with all others and selecting the one with the least value . this would result in a linear runtime complexity of @xmath260 as computation of kl - divergence is a constant time operation . for every component in @xmath0 , its closest match is identified and they are merged to obtain a @xmath220 component mixture @xmath243 . merging the pair involves reassigning the component weights and the memberships . an em algorithm is then employed to optimize @xmath243 . + assume components with indices @xmath218 and @xmath261 are merged . let their weights be @xmath217 and @xmath262 ; and their responsibility terms be @xmath219 and @xmath263 respectively . the component that is formed by merging the pair is determined first . it is then integrated with the @xmath264 remaining components of @xmath0 to produce a @xmath213-component mixture @xmath243 . + _ em initialization for @xmath243 : _ let @xmath265 and @xmath266 be the weight and responsibility vector of the merged component @xmath267 respectively . they are given as follows : @xmath268 the parameters of this merged component are estimated as follows : * _ gaussian : _ the parameters @xmath269 are : @xmath270 * _ von mises - fisher : _ the parameters @xmath271 are : @xmath272 the concentration parameter @xmath273 is obtained by solving @xmath274 after substituting @xmath275 and @xmath276 in equation . + the merged component @xmath267 with weight @xmath265 , responsibility vector @xmath277 , and parameters @xmath278 is then integrated with the @xmath264 components . the merged component and its associated memberships along with the @xmath264 other components serve as the starting point for optimizing the new mixture @xmath243 . if @xmath243 results in a lower message length compared to @xmath0 that means the perturbation of @xmath0 because of merging the pair of components resulted in an improvement to the current mixture . r0.35 we explain the proposed inference of mixture components through the following example that was also considered by @xcite . consider a bivariate gaussian mixture shown in fig . [ fig : mix1 ] . the mixture has three components with equal weights of 1/3 each and their means at ( -2,0 ) , ( 0,0 ) , and ( 2,0 ) . the covariance matrices of the three components are the same and are equal to @xmath279 . we simulate 900 data points from this mixture ( as done by @xcite ) and employ the proposed search strategy . the progression of the search method using various operations is detailed below . _ search for the optimal mixture model : _ the method begins by inferring a one - component mixture @xmath280 ( see fig . [ fig : mix1_iter_1_splits](a ) ) . it then splits this component ( as described in _ split _ step of section [ subsec : search_operations ] ) and checks whether there is an improvement in explanation . the red ellipse in fig [ fig : mix1_iter_1_splits](b ) depicts the component being split . the direction of maximum variance ( dotted black line ) is first identified , and the means ( shown by black dots at the end of the dotted line ) are initialized . an em algorithm is then used to optimize the two children and this results in a mixture @xmath281 shown in fig [ fig : mix1_iter_1_splits](c ) . since the new mixture has a lower message length , the current is updated as @xmath281 . in the second iteration , each component in @xmath281 is iteratively split , deleted , and merged . [ fig : mix1_iter_2_splits ] shows the splitting ( red ) of the first component . on splitting , the new mixture @xmath282 results in a lower message length . deletion of the first component is shown in fig . [ fig : mix1_iter_2_deletions ] . before merging the first component , we identify its closest component ( the one with the least kl - divergence ) ( see fig . [ fig : mix1_iter_2_merges ] ) . deletion and merging operations , in this case , do not result in an improvement . these two operations have different intermediate em initializations ( figures [ fig : mix1_iter_2_deletions](b ) and [ fig : mix1_iter_2_merges](b ) ) but result in the same optimized one - component mixture . the same set of operations are performed on the second component in @xmath281 . in this particular case , splitting results in an improved mixture ( same as @xmath282 ) . @xmath282 is updated as the new parent and the series of split , delete , and merge operations are carried out on all components in @xmath282 . [ fig : mix1_iter_3 ] shows these operations on the first component . we see that splitting the first component in @xmath282 results in @xmath283 ( see fig . [ fig : mix1_iter_3](c ) ) . however , @xmath283 is not an improvement over @xmath282 as seen by the message lengths and is , therfore , discarded . similarly , deletion and merging of the components do not yield improvements to @xmath282 . the operations are carried out on the remaining two components in @xmath282 ( not shown in the figure ) too . these perturbations do not produce improved mixtures in terms of the total message length . since the third iteration does not result in any further improvement , the search terminates and the parent @xmath282 is considered to be the best mixture . + + in different stages of the search method , we have different intermediate mixtures . em is a gradient descent technique and it can get trapped in a local optimum . by employing the suggested search , we are exhaustively considering the possible options , and aiming to reduce the possibility of the em getting stuck in a local optimum . the proposed method infers a mixture by balancing the tradeoff due to model complexity and the fit to the data . this is particularly useful when there is no prior knowledge pertaining to the nature of the data . in such a case , this method provides an objective way to infer a mixture with suitable components that best explains the data through lossless compression . another example is shown in appendix [ subsec : appendix_mix2 ] , where the evolution of the inferred mixture is explored in the case of a mixture with overlapping components . + _ variation of the two - part message length : _ the search method infers three components and terminates . in order to demonstrate that the inferred number of components is the optimum number , we infer mixtures with increasing number of components ( until @xmath284 as an example ) and plot their resultant message lengths . for each @xmath285 , the standard em algorithm ( section [ subsec : em_mml ] ) is employed to infer the mixture parameters . r0.5 fig . [ fig : individual_msglens_gaussian ] shows the total message lengths to which the em algorithm converges for varying number of components @xmath4 . as expected , the total message length ( green curve ) drastically decreases initially until @xmath286 components are inferred . starting from @xmath287 , the total message length gradually increases , clearly suggesting that the inferred models are over - fitting the data with increasing statement cost to encode the additional parameters of these ( more complex ) models . we further elaborate on the reason for the initial decrease and subsequent increase in the total message length . as per mml evaluation criterion , the message length comprises of two parts statement cost for the parameters and the cost for stating the data using those parameters . the model complexity ( which corresponds to the mixture parameters ) increases with increasing @xmath4 . therefore , the first part of the message to encode parameters increases with an increase in the number of parameters . this behaviour is illustrated by the red curve in fig . [ fig : individual_msglens_gaussian ] . the first part message lengths are shown in red on the right side y - axis in the figure . as the mixture model becomes increasingly more complex , the error of fitting the data decreases . this corresponds to the second part of the message in the mml encoding framework . this behaviour is consistent with what is observed in fig . [ fig : individual_msglens_gaussian ] ( blue curve ) . there is a sharp fall until @xmath286 ; then onwards increasing the model complexity does not lower the error significantly . the error saturates and there is minimal gain with regards to encoding the data ( the case of overfitting ) . however , the model complexity dominates after @xmath286 . the optimal balance is achieved when @xmath286 . in summary , the message length at @xmath286 components was rightly observed to be the optimum for this example . we note that for a fixed number of mixture components , the em algorithm for the mml metric is monotonically decreasing . however , while searching for the number of components , mml continues to decrease until some optimum is found and then steadily increases as illustrated through this example . we compare our proposed inference methodology against the widely cited method of @xcite . the performance of their method is compared against that of bayesian information criterion ( bic ) , integrated complete likelihood ( icl ) , and approximate bayesian ( lec ) methods ( discussed in section [ sec : mixture_existing_methods ] ) . it was shown that the method of @xcite was far superior than bic , icl and lec ( using gaussian mixtures ) . in the following sections , we demonstrate through a series of experiments that our proposed approach to infer mixtures fares better when compared to that of @xcite . the experimental setup is as follows : we use a gaussian mixture @xmath288 ( true distribution ) , generate a random sample from it , and infer the mixture using the data . this is repeated 50 times and we compare the performance of our method against that of @xcite . as part of our analysis , we compare the number of inferred mixture components as well as the quality of mixtures . _ comparing message lengths : _ the mml framework allows us to objectively compare mixture models by computing the total message length used to encode the data . the difference in message lengths gives the log - odds posterior ratio of any two mixtures ( equation ) . given some observed data , and any two mixtures , one can determine which of the two best explains the data . our search methodology uses the scoring function ( @xmath289 ) defined in equation . as elaborated in section [ subsec : fj ] , @xcite use an approximated mml - like scoring function ( @xmath290 ) given by equation . we employ our search method and the method of @xcite to infer the mixtures using the same data ; let the inferred mixtures be @xmath291 and @xmath292 respectively . we compute two quantities : @xmath293 we use the two different scoring functions to compute the differences in message lengths of the resulting mixtures @xmath292 and @xmath291 . since the search method used to obtain @xmath291 optimizes the scoring function @xmath289 , it is expected that @xmath294 and consequently @xmath295 . this implies that our method is performing better using our defined objective function . however , if @xmath296 , this indicates that our inferred mixture @xmath291 results in a lower value of the scoring function that is defined by @xcite . such an evaluation not only demonstrates the superior performance of our search ( leading to @xmath291 ) using our defined scoring function but also proves it is better using the scoring function as defined by @xcite . _ kullback leibler ( kl ) divergence : _ in addition to using message length based evaluation criterion , we also compare the mixtures using kl - divergence @xcite . the metric gives a measure of the similarity between two distributions ( the lower the value , the more similar the distributions ) . for a mixture probability distribution , there is no analytical form to compute the metric . however , one can calculate its empirical value ( which asymptotically converges to the kl - divergence ) . in experiments relating to mixture simulations , we know the true mixture @xmath288 from which the data @xmath297 is being sampled . the kl - divergence is given by the following expression : @xmath298 \approx \frac{1}{n } \sum_{i=1}^n\log\frac{\pr(\mathbf{x}_i,{\mathcal{m}}^t)}{\pr(\mathbf{x}_i,{\mathcal{m}})}\end{aligned}\ ] ] where @xmath0 is a mixture distribution ( @xmath291 or @xmath292 ) whose _ closeness _ to the true mixture @xmath288 is to be determined . an experiment conducted by @xcite was to randomly generate @xmath299 data points from a two - component ( with equal mixing proportions ) bivariate mixture @xmath288 whose means are at @xmath300 and @xmath301 , and equal covariance matrices : @xmath302 ( the identity matrix ) , and compare the number of inferred components . we repeat the same experiment here and compare with the results of @xcite . the separation @xmath303 between the means is gradually increased and the percentage of the correct selections ( over 50 simulations ) as determined by the two search methods is plotted . [ fig : gaussian_2d_mixture_inference](a ) shows the results of this experiment . as the separation between the component means is increased , the number of correctly inferred components increases . we conducted another experiment where we fix the separation between the two components and increase the amount of data being sampled from the mixture . [ fig : gaussian_2d_mixture_inference](b ) illustrates the results for a separation of @xmath304 . as expected , increasing the sample size results in an increase in the number of correct selections . both the search methods eventually infer the true number of components at sample size @xmath305 . we note that in both these cases , the differences in message lengths @xmath306 and @xmath307 are close to zero . the kl - divergences for the mixtures inferred by the two search methods are also the same . therefore , for this experimental setup , the performance of both the methods is roughly similar . as the difference between the two search methods is not apparent from these experiments , we wanted to investigate the behaviour of the methods with smaller sample sizes . we repeated the experiment similar to that shown in fig . [ fig : gaussian_2d_mixture_inference](a ) but with a sample size of @xmath308 . our search method results in a mean value close to 1 for different values of @xmath303 ( see table [ tab : gaussian_2d_mixture_inference_boxplot ] ) . the mean value of the number of inferred components using the search method of @xcite fluctuates between 2 and 3 . however , there is significant variance in the number of inferred components ( see table [ tab : gaussian_2d_mixture_inference_boxplot ] ) . these results are also depicted through a boxplot ( fig . [ fig : gaussian_2d_mixture_inference_boxplot ] ) . there are many instances where the number of inferred components is more than 3 . the results indicate that the search method ( fj ) is overfitting the data . r0.5 further , we evaluate the correctness of the mixtures inferred by the two search methods by comparisons using the message length formulations and kl-divergence.fig . [ fig : gaussian_2d_msglen_comparisons ] shows the boxplot of the difference in message lengths of the mixtures @xmath291 inferred using our proposed search method and the mixtures @xmath292 inferred using that of @xcite . @xmath295 across all values of @xmath303 for the 50 simulations . as per equation , we have @xmath294 . this implies that @xmath291 has a lower message length compared to @xmath292 when evaluated using our scoring function . similarly , we have @xmath309 , _ i.e. , _ @xmath310 . this implies that @xmath292 has a lower message length compared to @xmath291 when evaluated using fj s scoring function . these results are not surprising as @xmath291 and @xmath292 are obtained using the search methods which optimize the respective mml and mml - like scoring functions . we then analyzed the kl - divergence of @xmath291 and @xmath292 with respect to the true bivariate mixture @xmath288 over all 50 simulations and across all values of @xmath303 . ideally , the kl - divergence should be close to zero . [ fig : gaussian_2d_kldiv_comparisons](a ) shows the kl - divergence of the mixtures inferred using the two search methods with respect to @xmath288 when the separation is @xmath304 . the proposed search method infers mixtures whose kl - divergence ( denoted by red lines ) is close to zero , and more importantly less than the kl - divergence of mixtures inferred by the search method of @xcite ( denoted by blue lines ) . the same type of behaviour is noticed with other values of @xmath303 . [ fig : gaussian_2d_kldiv_comparisons](b ) compares the kl - divergence for varying values of @xmath303 . the median value of the kl - divergence due to the proposed search method is close to zero with not much variation . the search method of @xcite always result in kl - divergence higher than that of ours . the results suggest that , in this case , mixtures @xmath292 inferred by employing the search method of @xcite deviate significantly from the true mixture distribution @xmath288 . this can also be explained by the fact that there is a wide spectrum of the number of inferred components ( fig . [ fig : gaussian_2d_mixture_inference_boxplot ] ) . this suggests that the mml - like scoring function is failing to control the tradeoff between complexity and quality of fit , and hence , is selecting overly complex mixture models . along the lines of the previous experiment , @xcite conducted another experiment for a 10-variate two - component mixture @xmath288 with equal mixing proportions . the means are at @xmath311 and @xmath312 , so that the euclidean distance between them is @xmath313 . the covariances of the two components are @xmath302 ( the identity matrix ) . random samples of size @xmath299 were generated from the mixture and the number of inferred components are plotted . the experiment is repeated for different values of @xmath303 and over 50 simulations . [ fig : gaussian_10d_mixture_inference](a ) shows the number of inferred components using the two search methods . at lower values of @xmath303 , the components are close to each other , and hence , it is relatively more difficult to correctly infer the true number of components . we observe that our proposed method performs clearly better than that of @xcite across all values of @xmath303 . we also compared the quality of these inferred mixtures by calculating the difference in message lengths using the two scoring functions and the kl - divergence with respect to @xmath288 . for all values of @xmath303 , @xmath295 , _ i.e. , _ our inferred mixtures @xmath291 have a lower message length compared to @xmath292 when evaluated using our scoring function . more interestingly , we also note that @xmath314 ( see fig . [ fig : gaussian_10d_comparisons](a ) ) . this reflects that @xmath291 have a lower message length compared to @xmath292 when evaluated using the scoring function of @xcite . this suggests that their search method results in a sub - optimal mixture @xmath292 and fails to infer the better @xmath291 . in addition to the message lengths , we analyze the mixtures using kl - divergence . similar to the bivariate example in fig . [ fig : gaussian_2d_kldiv_comparisons](a ) , the kl - divergence of our inferred mixtures @xmath291 is lower than @xmath292 , the mixtures inferred by @xcite . [ fig : gaussian_10d_comparisons](b ) shows the boxplot of kl - divergence of the inferred mixtures @xmath291 and @xmath292 . at higher values of @xmath315 , the median value of kl - divergence is close to zero , as the number of correctly inferred components ( fig . [ fig : gaussian_10d_mixture_inference](a ) ) is more than 90% . however , our method always infers mixtures @xmath291 with lower kl - divergence compared to @xmath292 . these experimental results demonstrate the superior performance of our proposed search method . another experiment was carried out where the @xmath316 was held constant ( extremely close components ) , gradually increased the sample size @xmath34 , and plotted the average number of inferred components by running 50 simulations for each @xmath34 . [ fig : gaussian_10d_mixture_inference](b ) shows the results for the average number of inferred components as the amount of data increases . our search method , on average , infers the true mixture when the sample size is @xmath317 . however , the search method of @xcite requires larger amounts of data ; even with a sample size of 2000 , the average number of inferred components is @xmath318 . in fig . [ fig : gaussian_10d_mixture_inference](b ) , the red curve reaches the true number of 2 and saturates more rapidly than the blue curve . one of the drawbacks associated with the search method of @xcite is due to the form of the updating expression for the component weights ( equation ) . as discussed in section [ subsubsec : fj_search_drawbacks ] , a particular instance of wrong inference is bound to happen when the net membership of a ( valid ) component is less than @xmath208 , where @xmath200 is the number of free parameters per component . in such a case , the component weight is updated as zero , and it is eliminated , effectively reducing the mixture size by one . we conducted the following experiment to demonstrate this behaviour : we considered the two - component 10-variate mixture @xmath288 as before and randomly generate samples of size 50 from the mixture . since the constituent components of @xmath288 have equal weights , on average , each component has a membership of 25 . we used @xmath319 , so that the two components are well apart from each other . for each @xmath303 , we run 50 simulations and analyze the number of inferred components . as expected , the search method of @xcite always infer a mixture with one component regardless of the separation @xmath303 . our method always infers the correct number of components . in order to test the validity of mixtures inferred by our proposed method , we analyze the resultant mixtures by comparing the message lengths as discussed in section [ subsec : comparison_mixtures ] . [ fig : gaussian_10d_msglens_comparisons_exp2b](a ) shows the difference in message lengths @xmath306 given in equation . we observe that @xmath295 for all @xmath303 . this demonstrates that our search based mixtures @xmath291 have lower message lengths compared to mixtures @xmath292 using our scoring function . the same phenomenon is observed when using the mml - like scoring function of @xcite . in fig . [ fig : gaussian_10d_msglens_comparisons_exp2b](b ) , we observe that @xmath314 , which means our search based mixtures @xmath291 have lower message lengths compared to mixtures @xmath292 when evaluated using their scoring function . r0.45 this demonstrates that @xmath291 is a better mixture as compared to @xmath292 and their search method is unable to infer it . we also note that the differences in message lengths increases with increasing @xmath303 . this is because for the one - component inferred mixture @xmath292 , the second part of the message ( equation ) which corresponds to the negative log - likelihood term increases because of poorer fit to the data . the two modes in the data become increasingly pronounced as the separation between constituent components of the true mixture increases , and hence , modelling such a distribution using a one - component mixture results in a poorer fit . this is clearly an incorrect inference . we further strengthen our case by comparing the kl - divergence of the inferred mixtures @xmath291 and @xmath292 with respect to the true mixture . [ fig : gaussian_10d_kldiv_comparisons_exp2b ] illustrates the results . as @xmath303 increases , the blue coloured plots shift higher . these correspond to mixtures @xmath292 inferred by @xcite . our search method , however , infers mixtures @xmath291 which have lower kl - divergence . the figure indicates that the inferred mixtures @xmath291 are more similar to the true distribution as compared to mixtures @xmath292 . these experiments demonstrate the ability of our search method to perform better than the widely used method of @xcite . we compared the resulting mixtures using our proposed mml formulation and the mml - like formulation of @xcite , showing the advantages of the former over the latter . we also used a neutral metric , kl - divergence , to establish the closeness of our inferred mixtures to the true distributions . we will now illustrate the behaviour of our search method on two real world datasets . at any intermediate stage of the search procedure , current _ mixture with @xmath4 components requires @xmath4 number of split , delete , and merge operations before it is updated . each of the perturbations involve performing an em to optimize the corresponding mixture parameters . to determine the convergence of em , we used a threshold of @xmath320 which was the same as used by @xcite . fj s method also requires to start from an initial large number of components . we used 25 as an initial number based on what was suggested in @xcite . we investigate the number of times the em routine is called and compare it with that of @xcite . we examine with respect to the simulations that were carried out previously . for the bivariate mixture discussed in section [ subsec : bivariate_mix_simulation ] , the number of resulting em iterations when the sample sizes are @xmath299 and @xmath308 are compared in fig . [ fig : em_iterations](a ) , ( b ) respectively . as per the discussion in section [ subsec : bivariate_mix_simulation ] , at @xmath299 , the average number of components inferred by the two methods are about the same ( fig . [ fig : gaussian_2d_mixture_inference](a ) ) . however , the number of em iterations required by fj s method is greater than 200 across all values of @xmath303 ( fig . [ fig : em_iterations](a ) ) . in contrast , the proposed method , on average , requires fewer than 50 iterations . in this case , both methods produce a similar result with fj s method requiring more number of em iterations . when the bivariate mixture simulation is carried out using @xmath308 , the number of em iterations required by fj s method , on average , is greater than 100 , while the proposed method requires fewer than 40 iterations ( fig . [ fig : em_iterations](b ) ) . in this case , the proposed method not only infers better mixtures ( as discussed in section [ subsec : bivariate_mix_simulation ] ) but is also conservative with respect to computational cost . for the simulation results corresponding to the 10-variate mixtures in section [ subsec:10d_mix_simulation ] , the proposed method requires close to 50 iterations on average , while fj s method requires about 20 ( fig . [ fig : em_iterations](c ) ) . however , the mixtures inferred by the proposed method fare better when compared to that of fj ( figs . [ fig : gaussian_10d_mixture_inference ] , [ fig : gaussian_10d_comparisons ] ) . furthermore , for the simulation results explained in section [ subsec : fj_weight_updates_exp2b ] , fj s method stops after 3 em iterations . this is because their program does not accommodate components when the memberships are less than @xmath208 . the proposed method requires 18 em iterations on average and infers the correct mixture components . in these two cases , our method infers better quality mixtures , with no significant overhead with regard to the computational cost . the first example is the univariate _ acidity _ data set which contains 155 points . our proposed search method infers a mixture @xmath291 with 2 components whereas the search method of @xcite infers a mixture @xmath292 with 3 components . the inferred mixtures are shown in fig . [ fig : acidity ] and their corresponding parameter estimates are given in table [ tab : acidity_mixtures_inference ] . in order to compare the mixtures inferred by the two search methods , we compute the message lengths of the inferred mixtures using our complete mml and the approximated mml - like scoring functions . when evaluated using our mml scoring function , our inferred mixture results in a gain of @xmath321 bits ( see table [ tab : acidity_comparison ] ) . based on the mml framework , our two - component mixture @xmath291 is @xmath322 times more likely than the three - component mixture @xmath292 ( as per equation ) . furthermore , when the inferred mixtures are evaluated as per the mml - like scoring function , @xmath291 is still considered better ( @xmath323 bits ) than @xmath292 ( @xmath324 bits ) . thus , using both forms of scoring function , @xmath291 is the better mixture model of this data set . [ tab : acidity_mixtures_inference ] .message lengths ( measured in bits ) of the mixtures ( in fig . [ fig : acidity ] ) as evaluated using the mml and mml - like scoring functions . [ cols="^,^,^ " , ] [ tab : null_models_comparison ] the per residue statistic is calculated by dividing the total message length by the sample size ( the number of @xmath325 pairs ) . this statistic shows that close to 2.5 bits can be saved ( on average ) if the protein data is encoded using the vmf null model . the vmf null model thus supercedes the naive model of encoding . this can potentially improve the accuracy of statistical inference that is central to the various protein modelling tasks briefly introduced above . we presented a statistically robust approach for inferring mixtures of ( i ) multivariate gaussian distributions , and ( ii ) von mises - fisher distributions for @xmath3-dimensional directional data . it is based on the general information - theoretic framework of minimum message length inference . this provides an objective tradeoff between the hypothesis complexity and the quality of fit to the data . an associated search procedure for an optimal mixture model of given data chooses the number of component distributions , @xmath4 , by minimizing the total message length . we established the better performance of the proposed search algorithm by comparing with a popularly used search method @xcite . we demonstrated the effectiveness of our approach through extensive experimentation and validation of our results . we also applied our method to real - world high dimensonal text data and to directional data that arises from protein chain conformations . the experimental results demonstrate that our proposed method fares better when compared with the current state of the art techniques . the authors would like to thank arun konagurthu for discussions pertaining to protein data clustering , and maria garcia de la banda for stimulating discussions and providing interesting insights . the authors would also like to acknowledge wray buntine s inputs with regard to text clustering . 91 [ 1]#1 [ 1]#1 urlstyle [ 1]doi # 1 [ 2][]#2 agusta y , dowe dl ( 2003 ) unsupervised learning of correlated multivariate gaussian mixture models using mml . in : ai 2003 : advances in artificial intelligence , springer , berlin , heidelberg , pp 477489 akaike h ( 1974 ) a new look at the statistical model identification . ieee transactions on automatic control , 19(6):716723 anderson e ( 1935 ) the irises of the gasp peninsula . bulletin of the american iris society 59:25 banerjee a , dhillon i , ghosh j , sra s ( 2003 ) generative model - based clustering of directional data . in : proceedings of the 9th international conference on knowledge discovery and data mining , acm , new york , pp 1928 banerjee a , dhillon is , ghosh j , sra s ( 2005 ) clustering on the unit hypersphere using von mises - fisher distributions . journal of machine learning research 6:13451382 barton de ( 1961 ) unbiased estimation of a set of probabilities . biometrika 48(1 - 2):227229 basu ap ( 1964 ) estimates of reliability for some distributions useful in life testing . technometrics 6(2):215219 best d , fisher n ( 1981 ) the bias of the maximum likelihood estimators of the von mises - fisher concentration parameters . communications in statistics - simulation and computation 10(5):493502 biernacki c , celeux g , govaert g ( 2000 ) assessing a mixture model for clustering with the integrated completed likelihood . ieee transactions on pattern analysis and machine intelligence 22(7):719725 bishop cm ( 2006 ) pattern recognition and machine learning , vol 1 . springer , new york boulton d , wallace c ( 1969 ) the information content of a multistate distribution . journal of theoretical biology 23:269278 bozdogan h ( 1983 ) determining the number of component clusters in the standard multivariate normal mixture model using model - selection criteria . tech . rep . , dtic document bozdogan h ( 1990 ) on the information - based measure of covariance complexity and its application to the evaluation of multivariate linear models . communications in statistics - theory and methods 19(1):221278 bozdogan h ( 1993 ) choosing the number of component clusters in the mixture - model using a new informational complexity criterion of the inverse - fisher information matrix . springer , berlin , heidelberg burnham kp , anderson dr ( 2002 ) model selection and multimodel inference : a practical information - theoretic approach . springer , new york collier jh , allison l , lesk am , de la banda mg , konagurthu as ( 2014 ) a new statistical framework to assess structural alignment quality using information compression . bioinformatics 30(17):i512i518 conway jh , sloane nja ( 1984 ) on the voronoi regions of certain lattices . siam journal on algebraic and discrete methods 5:294305 cordeiro gm , vasconcellos kl ( 1999 ) theory & methods : second - order biases of the maximum likelihood estimates in von mises regression models . australian & new zealand journal of statistics 41(2):189198 dempster ap , laird nm , rubin db ( 1977 ) maximum likelihood from incomplete data via the em algorithm . journal of the royal statistical society : series b ( methodological ) 39(1):138 dowe dl , allison l , dix ti , hunter l , wallace cs , edgoose t ( 1996 ) circular clustering of protein dihedral angles by minimum message length . in : pacific symposium on biocomputing , vol 96 , pp 242255 dowe dl , oliver jj , baxter ra , wallace cs ( 1996 ) bayesian estimation of the von mises concentration parameter . in : maximum entropy and bayesian methods , springer , netherlands , pp 5160 dowe dl , oliver jj , wallace cs ( 1996 ) mml estimation of the parameters of the spherical fisher distribution . in : algorithmic learning theory , springer , berlin , heidelberg , pp 213227 drton m , sturmfels b , sullivant s ( 2009 ) lectures on algebraic statistics . in : oberwolfach seminars ( 39 ) , birkhuser , basel dwyer ps ( 1967 ) some applications of matrix derivatives in multivariate analysis . journal of the american statistical association 62(318):607625 eaton ml , morris cn ( 1970 ) the application of invariance to unbiased estimation . the annals of mathematical statistics 41(5):17081716 figueiredo ams ( 2012 ) goodness - of - fit for a concentrated von mises - fisher distribution . computational statistics 27(1):6982 figueiredo ma , jain ak ( 2002 ) unsupervised learning of finite mixture models . ieee transactions on pattern analysis and machine intelligence 24(3):381396 fisher ni ( 1993 ) statistical analysis of spherical data . cambridge university press , cambridge fisher r ( 1953 ) dispersion on a sphere . proceedings of the royal society of london : series a ( mathematical and physical sciences ) 217(1130):295305 fisher ra ( 1936 ) the use of multiple measurements in taxonomic problems . annals of eugenics 7(2):179188 gauvain j , lee ch ( 1994 ) maximum a posteriori estimation for multivariate gaussian mixture observations of markov chains . ieee transactions on speech and audio processing 2(2):291298 gray g ( 1994 ) bias in misspecified mixtures . biometrics 50(2):457470 hornik k , grn b ( 2013 ) movmf : an r package for fitting mixtures of von mises - fisher distributions . r package version 01 - 2 jain ak , dubes rc ( 1988 ) algorithms for clustering data . prentice - hall , inc . , upper saddle river , nj , usa jain ak , duin rpw , mao j ( 2000 ) statistical pattern recognition : a review . ieee transactions on pattern analysis and machine intelligence 22(1):437 jeffreys h ( 1946 ) an invariant form for the prior probability in estimation problems . proceedings of the royal society of london : series a ( mathematical and physical sciences ) 186(1007):453461 jones p , mclachlan g ( 1990 ) laplace - normal mixtures fitted to wind shear data . journal of applied statistics 17(2):271276 jorgensen ma , mclachlan gj ( 2008 ) wallace s approach to unsupervised learning : the snob program . the computer journal 51(5):571578 kent jt ( 1982 ) the fisher - bingham distribution on the sphere . journal of the royal statistical society : series b ( methodological ) 44(1):7180 konagurthu as , lesk am , allison l ( 2012 ) minimum message length inference of secondary structure from protein coordinate data . bioinformatics 28(12):i97i105 konagurthu as , allison l , abramson d , stuckey pj , lesk am ( 2013 ) statistical inference of protein lego bricks " . in : 2013 ieee 13th international conference on data mining ( icdm ) , ieee , pp 10911096 krishnan t , mclachlan g ( 1997 ) the em algorithm and extensions . wiley , new york kullback s , leibler ra ( 1951 ) on information and sufficiency . the annals of mathematical statistics pp 7986 lee p ( 1997 ) bayesian statistics : an introduction . arnold , london . lo y ( 2011 ) bias from misspecification of the component variances in a normal mixture . computational statistics and data anaysis 55(9):27392747 magnus jr , neudecker h ( 1988 ) matrix differential calculus with applications in statistics and econometrics . wiley , new york mardia k , jupp p ( 2000 ) directional statistics . wiley , hoboken , nj , usa mardia k , holmes d , kent j ( 1984 ) a goodness - of - fit test for the von mises - fisher distribution . journal of the royal statistical society : series b ( methodological ) 46(1):7278 mardia kv , kent jt , bibby jm ( 1979 ) multivariate analysis . academic press , london mardia kv , taylor cc , subramaniam gk ( 2007 ) protein bioinformatics and mixtures of bivariate von mises distributions for angular data . biometrics 63(2):505512 mclachlan g , peel d ( 1997 ) contribution to the discussion of paper by s. richardson and p.j . journal of the royal statistical society b 59:779780 mclachlan g , peel d ( 2000 ) finite mixture models . wiley , new york mclachlan gj , basford ke ( 1988 ) mixture models : inference and applications to clustering ( statistics : textbooks and monographs ) . dekker , new york murzin a , brenner s , hubbard t , chothia c , et al . ( 1995 ) scop : a structural classification of proteins database for the investigation of sequences and structures . journal of molecular biology 247(4):536540 oliver j , baxter r ( 1994 ) mml and bayesianism : similarities and differences . dept comput sci monash univ , clayton , victoria , australia , tech rep 206 oliver jj , baxter ra , wallace cs ( 1996 ) unsupervised learning using mml . in : machine learning : proceedings of the 13th international conference , pp 364372 patra k , dey dk ( 1999 ) a multivariate mixture of weibull distributions in reliability modeling . statistics & probability letters 45(3):225235 peel d , mclachlan gj ( 2000 ) robust mixture modelling using the t - distribution . statistics and computing 10(4):339348 peel d , whiten wj , mclachlan gj ( 2001 ) fitting mixtures of kent distributions to aid in joint set identification . journal of the american statistical association 96(453):5663 rao cr ( 1973 ) linear statistical inference and its applications . wiley , new york richardson s , green pj ( 1997 ) on bayesian analysis of mixtures with an unknown number of components . journal of the royal statistical society : series b ( methodological ) 59(4):731792 rissanen j ( 1978 ) modeling by shortest data description . automatica 14(5):465471 rissanen j ( 1989 ) stochastic complexity in statistical inquiry theory . world scientific publishing co. , inc . , river edge , nj , usa roberts s , husmeier d , rezek i , penny w ( 1998 ) bayesian approaches to gaussian mixture modeling . ieee transactions on pattern analysis and machine intelligence 20(11):11331142 robertson s , zaragoza h ( 2009 ) the probabilistic relevance framework : bm25 and beyond . now publishers inc . , hanover , ma , usa salton g , buckley c ( 1988 ) term - weighting approaches in automatic text retrieval . information processing & management 24(5):513523 salton g , mcgill mj ( 1986 ) introduction to modern information retrieval . mcgraw - hill , inc . , new york , ny , usa schou g ( 1978 ) estimation of the concentration parameter in von mises fisher distributions . biometrika 65(2):369377 schwarz g , et al . ( 1978 ) estimating the dimension of a model . the annals of statistics 6(2):461464 seidel w , mosler k , alker m ( 2000 ) a cautionary note on likelihood ratio tests in mixture models . annals of the institute of statistical mathematics 52(3):481487 shannon ce ( 1948 ) a mathematical theory of communication . the bell system technical journal 27:379423 song h , liu j , wang g ( 2012 ) high - order parameter approximation for von mises fisher distributions . applied mathematics and computation 218(24):11,88011,890 sra s ( 2012 ) a short note on parameter approximation for von mises - fisher distributions : and a fast implementation of @xmath326 . computational statistics 27(1):177190 strehl a , ghosh j , mooney r ( 2000 ) impact of similarity measures on web - page clustering . in : workshop on artificial intelligence for web search ( aaai 2000 ) , pp 5864 taboga m ( 2012 ) lectures on probability theory and mathematical statistics . createspace independent pub . tanabe a , fukumizu k , oba s , takenouchi t , ishii s ( 2007 ) parameter estimation for von mises fisher distributions . computational statistics 22(1):145157 titterington dm , smith af , makov ue , et al . ( 1985 ) statistical analysis of finite mixture distributions . wiley , new york wallace c ( 1986 ) an improved program for classification . in : proceedings of the 9th australian computer science conference , pp 357366 wallace c ( 1990 ) classification by minimum - message - length inference . lecture notes in computer science , vol 468 , springer berlin heidelberg , pp 7281 wallace c , dowe d ( 1994 ) estimation of the von mises concentration parameter using minimum message length . in : proceedings of the 12th australian statistical society conference , monash university , australia wallace cs ( 2005 ) statistical and inductive inference using minimum message length . information science and statistics , springer - verlag , secaucus , nj , usa wallace cs , boulton dm ( 1968 ) an information measure for classification . computer journal 11(2):185194 wallace cs , dowe dl ( 1999 ) minimum message length and kolmogorov complexity . computer journal 42:270283 wallace cs , freeman pr ( 1987 ) estimation and inference by compact coding . journal of the royal statistical society : series b ( methodological ) 49(3):240265 wang p , puterman ml , cockburn i , le n ( 1996 ) mixed poisson regression models with covariate dependent rates . biometrics 52(2):381400 watson g , williams e ( 1956 ) on the construction of significance tests on the circle and the sphere . biometrika 43(3 - 4):344352 wedel m , desarbo ws , bult jr , ramaswamy v ( 1993 ) a latent class poisson regression model for heterogeneous count data . journal of applied econometrics 8(4):397411 white h ( 1982 ) maximum likelihood estimation of misspecified models . econometrica 50(1):125 wood at ( 1994 ) simulation of the von mises fisher distribution . communications in statistics - simulation and computation 23(1):157164 xu l , jordan mi ( 1996 ) on convergence properties of the em algorithm for gaussian mixtures . neural computation 8(1):129151 zhong s , ghosh j ( 2003 ) a comparative study of generative models for document clustering . in : proceedings of the workshop on clustering high dimensional data and its applications in siam data mining conference for brevity , we represent @xmath327 and @xmath328 as @xmath329 and @xmath330 respectively . expressions to evaluate @xmath331 and @xmath332 are given in equations ( [ eqn : ratio_bessels ] ) , ( [ eqn : ratio_first_derivative ] ) , and ( [ eqn : ratio_second_derivative ] ) respectively . we require @xmath330 for its use in the remainder of the derivation and we provide its expression below : @xmath333 now we discuss the derivation of @xmath121 and @xmath122 that are required for computing the mml estimate @xmath334 ( equations ( [ eqn : mml_newton_approx ] ) and ( [ eqn : mml_halley_approx ] ) ) . differentiating equation [ eqn : i_first_derivative ] , we have @xmath335 @xmath336 equations ( [ eqn : tmp1 ] ) and ( [ eqn : tmp2 ] ) can be used to evaluate @xmath121 and @xmath122 which can then be used to approximate the mml estimates @xmath123 and @xmath124 ( equations ( [ eqn : mml_newton_approx ] ) and ( [ eqn : mml_halley_approx ] ) respectively ) . to obtain the optimal weights under the constraint @xmath160 , the above equation is optimized using the _ lagrangian _ objective function defined below using some _ lagrangian multiplier _ @xmath341 . @xmath342 for some @xmath343 , the equation resulting from computing the partial derivative of @xmath344 with respect to @xmath345 and equating it to zero gives the optimal weight @xmath345 . @xmath346 @xmath347 where @xmath348 and @xmath349 are the responsibility and effective membership terms given as per equations and respectively . substituting the above value in equation , we have @xmath350 there are @xmath4 equations similar to equation for values of @xmath343 . adding all these equations together , we get @xmath351 substituting the above value of @xmath341 in equation , we get @xmath352 [ [ subsec : vmf_kldiv ] ] derivation of the kullback - leibler ( kl ) distance between two von mises - fisher distributions ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ the closed form expression to calculate the kl divergence between two vmf distributions is presented below . let @xmath353 and @xmath354 be two von mises - fisher distributions with mean directions @xmath355 and concentration parameters @xmath356 . the kl distance between any two distributions is given by @xmath357 \label{eqn : kldiv}\ ] ] where @xmath358 $ ] is the expectation of the quantity @xmath359 $ ] using the probability density function @xmath17 . @xmath360 using the fact that @xmath361 = a_d(\kappa_1)\boldsymbol{\mu_1}$ ] @xcite , we have the following expression : @xmath362 & = \log\frac{c_d(\kappa_1)}{c_d(\kappa_2 ) } + ( \kappa_1\boldsymbol{\mu_1 } -\kappa_2\boldsymbol{\mu_2})^t a_d(\kappa_1 ) \boldsymbol{\mu_1 } \notag\\ d_{kl}(f||g ) & = \log\frac{c_d(\kappa_1)}{c_d(\kappa_2 ) } + a_d(\kappa_1 ) ( \kappa_1 -\kappa_2\boldsymbol{\mu_1}^t\boldsymbol{\mu_2 } ) \label{eqn : vmf_kldiv}\end{aligned}\ ] ] we consider an example of a mixture with overlapping components and employ the proposed search method to determine the number of mixture components . @xcite considered the mixture shown in fig . [ fig : mix2 ] which is described as follows : let @xmath363 be the weight , mean , and covariance matrix of the @xmath364 component respectively . then , the mixture parameters are given as @xmath365 0.5 & 1 \end{bmatrix } , \mathbf{c}_2 = \begin{bmatrix } 6 & -2 \\[0.3em ] -2 & 6 \end{bmatrix},\ , \mathbf{c}_3 = \begin{bmatrix } 2 & -1 \\[0.3em ] -1 & 2 \end{bmatrix},\ , \mathbf{c}_4 = \begin{bmatrix } 0.125 & 0 \\[0.3em ] 0 & 0.125 \end{bmatrix}\end{gathered}\ ] ] we generate a random sample of size 1000 ( similar to @xcite ) and infer mixtures using the proposed heuristic . below are diagrams which portray the splitting of the individual components . we show a selection of operations which result in improved mixtures . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: there are many routes to minkowski space . but the most physical one still seems to me via the law of inertia . and even along these lines alternative approaches exist . many papers were published in physics and mathematics journals over the last 100 years in which incremental progress was reported as regards the minimal set of hypotheses from which the structure of minkowski space could be deduced . one could imagine a hesse - diagram - like picture in which all these contributions ( being the nodes ) together with their logical dependencies ( being the directed links ) were depicted . it would look surprisingly complex . from a general - relativistic point of view , minkowski space just models an _ empty _ spacetime , that is , a spacetime devoid of any material content . it is worth keeping in mind , that this was not minkowski s view . close to the beginning of _ raum und zeit _ he stated : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ in order to not leave a yawning void , we wish to imagine that at every place and at every time something perceivable exists . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this already touches upon a critical point . our modern theoretical view of spacetime is much inspired by the typical hierarchical thinking of mathematics of the late 19th and first half of the 20th century , in which the _ set _ comes first , and then we add various structures on it . we first think of spacetime as a set and then structure it according to various physical inputs . but what are the elements of this set ? recall how georg cantor , in his first article on transfinite set - theory , defined a set : von bestimmten wohlunterschiedenen objecten @xmath0 unserer anschauung oder unseres denkens ( welche die ` elemente ' von @xmath1 genannt werden ) zu einem ganzen . '' ( @xcite , p.481 ) ] _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ by a ` set ' we understand any gathering - together @xmath1 of determined well - distinguished objects @xmath0 of our intuition or of our thinking ( which are called the ` elements ' of @xmath1 ) into a whole . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ do we think of spacetime points as `` determined well - distinguished objects of our intuition or of our thinking '' ? i think minkowski felt a need to do so , as his statement quoted above indicates , and also saw the problematic side of it : if we mentally individuate the points ( elements ) of spacetime , we as physicists have no other means to do so than to fill up spacetime with actual matter , hoping that this could be done in such a diluted fashion that this matter will not dynamically affect the processes that we are going to describe . in other words : the whole concept of a rigid background spacetime is , from its very beginning , based on an assumption of at best approximate validity . it is important to realise that this does not necessarily refer to general relativity : even if the need to incorporate gravity by a variable and matter - dependent spacetime geometry did not exist would the concept of a rigid background spacetime be of approximate nature , _ provided we think of spacetime points as individuated by actual physical events_. it is true that modern set theory regards cantor s original definition as too nave , and that for good reasons . it allows too many `` gatherings - together '' with self - contradictory properties , as exemplified by the infamous _ antinomies _ of classical set theory . also , modern set theory deliberately stands back from any characterisation of elements in order to not confuse the axioms themselves with their possible _ interpretations_. however , applications to physics require _ interpreted _ axioms , where it remains true that elements of sets are thought of as definite as in cantors original definition . modern textbooks on special relativity have little to say about this , though an increasing unease seems to raise its voice from certain directions in the philosophy - of - science community ; see , e.g. , @xcite@xcite . physicists sometimes tend to address points of spacetime as _ potential events _ , but that always seemed to me like poetry , begging the question how a mere potentiality is actually used for individuation . to me the right attitude seems to admit that the operational justification of the notion of spacetime events is only approximately possible , but nevertheless allow it as primitive element of theorising . the only thing to keep in mind is to not take mathematical rigour for ultimate physical validity . the purpose of mathematical rigour is rather to establish the tightest possible bonds between basic assumptions ( axioms ) and decidable consequences . only then can we in principle learn anything through falsification . the last remark opens another general issue , which is implicit in much of theoretical research , namely how to balance between attempted rigour in drawing consequences and attempted closeness to reality when formulating once starting platform ( at the expense of rigour when drawing consequences ) . as the mathematical physicists glance & wightman once formulated it in a different context ( that of superselection rules in quantum mechanics ) : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the theoretical results currently available fall into two categories : rigorous results on approximate models and approximate results in realistic models . _ ( @xcite , p.204 ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ to me this seems to be the generic situation in theoretical physics . in that respect , minkowski space is certainly an approximate model , but to a very good approximation indeed : as global model of spacetime if gravity plays no dynamical rle , and as local model of spacetime in far more general situations . this justifies looking at some of its rich mathematical structures in detail . some mathematical background material is provided in the appendices . consider first the general situation where one is given a set @xmath2 . without any further structure being specified , the _ automorphisms _ group of @xmath2 would be the group of bijections of @xmath2 , i.e. maps @xmath3 which are injective ( into ) and surjective ( onto ) . it is called @xmath4 , where ` perm ' stands for ` permutations ' . now endow @xmath2 with some structure @xmath5 ; for example , it could be an equivalence relation on @xmath2 , that is , a partition of @xmath2 into an exhaustive set of mutually disjoint subsets ( cf . sect.[sec : groupactions ] ) . the automorphism group of @xmath6 is then the subgroup of @xmath7 that preserves @xmath5 . note that @xmath8 contains only those maps @xmath9 preserving @xmath5 whose inverse , @xmath10 , also preserve @xmath5 . now consider another structure , @xmath11 , and form @xmath12 . one way in which the two structures @xmath5 and @xmath11 may be compared is to compare their automorphism groups @xmath8 and @xmath12 . comparing the latter means , in particular , to see whether one is contained in the other . containedness clearly defines a partial order relation on the set of subgroups of @xmath4 , which we can use to define a partial order on the set of structures . one structure , @xmath5 , is said to be strictly stronger than ( or equally strong as ) another structure , @xmath11 , in symbols @xmath13 , iff the automorphism group of the former is properly contained in ( or is equal to ) the automorphism group of the latter . here . ] in symbols : @xmath14 . note that in this way of speaking a substructure ( i.e. one being defined by a subset of conditions , relations , objects , etc . ) of a given structure is said to be weaker than the latter . this way of thinking of structures in terms of their automorphism group is adopted from felix klein s _ erlanger programm _ @xcite in which this strategy is used in an attempt to classify and compare geometries . this general procedure can be applied to minkowski space , endowed with its usual structure ( see below ) . we can than ask whether the automorphism group of minkowski space , which we know is the inhomogeneous lorentz group @xmath15 , also called the poincar group , is already the automorphism group of a proper substructure . if this were the case we would say that the original structure is redundant . it would then be of interest to try and find a minimal set of structures that already imply the poincar group . this can be done by trial and error : one starts with some more or less obvious substructure , determine its automorphism group , and compare it to the poincar group . generically it will turn out larger , i.e. to properly contain @xmath15 . the obvious questions to ask then are : how much larger ? and : what would be a minimal extra condition that eliminates the difference ? these questions have been asked in connection with various substructures of minkowski space , whose definition is as follows : [ def : minkowskispace ] * minkowski space * of @xmath16 dimensions , denoted by @xmath17 , is a real @xmath18-dimensional affine space , whose associated real @xmath18-dimensional vector space @xmath19 is endowed with a non - degenerate symmetric bilinear form @xmath20 of signature @xmath21 ( i.e. there exists a basis @xmath22 of @xmath19 such that @xmath23 ) . @xmath17 is also endowed with the standard differentiable structure of @xmath24 . we refer to appendix[sec : affinespaces ] for the definition of affine spaces . note also that the last statement concerning differentiable structures is put in in view of the strange fact that just for the physically most interesting case , @xmath25 , there exist many inequivalent differentiable structures of @xmath26 . finally we stress that , at this point , we did not endow minkowski space with an orientation or time orientation . [ def : poincaregroup ] the * poincar group * in @xmath16 dimensions , which is the same as the * inhomogeneous lorentz group * in @xmath16 dimensions and therefore will be denoted by @xmath27 , is that subgroup of the general affine group of real @xmath18-dimensional affine space , for which the uniquely associated linear maps @xmath28 are elements of the lorentz group @xmath29 , that is , preserve @xmath30 in the sense that @xmath31 for all @xmath32 . see appendix[sec : affinemaps ] for the definition of affine maps and the general affine group . again we stress that since we did not endow minkowski space with any orientation , the poincar group as defined here would not respect any such structure . as explained in [ sec : affineframes ] , any choice of an affine frame allows us to identify the general affine group in @xmath18 dimensions with the semi - direct product @xmath33 . that identification clearly depends on the choice of the frame . if we restrict the bases to those where @xmath23 , then @xmath27 can be identified with @xmath34 . we can further endow minkowski space with an _ orientation _ and , independently , a _ time orientation_. an orientation of an affine space is equivalent to an orientation of its associated vector space @xmath19 . a time orientation is also defined trough a time orientation of @xmath19 , which is explained below . the subgroup of the poincar group preserving the overall orientation is denoted by @xmath35 ( proper poincar group ) , the one preserving time orientation by @xmath36 ( orthochronous poincar group ) , and @xmath37 denotes the subgroup preserving both ( proper orthochronous poincar group ) . upon the choice of a basis we may identify @xmath35 with @xmath38 and @xmath37 with @xmath39 , where @xmath40 is the component of the identity of @xmath41 . let us add a few more comments about the elementary geometry of minkowski space . we introduce the following notations : @xmath42 we shall also simply write @xmath43 for @xmath44 . a vector @xmath45 is called _ timelike , lightlike _ , or _ spacelike _ according to @xmath43 being @xmath46 , @xmath47 , or @xmath48 respectively . non - spacelike vectors are also called _ causal _ and their set , @xmath49 , is called the _ causal - doublecone_. its interior , @xmath50 , is called the _ chronological - doublecone _ and its boundary , @xmath51 , the _ light - doublecone _ : [ eq : vecdc ] @xmath52 a linear subspace @xmath53 is called timelike , lightlike , or spacelike according to @xmath54 being indefinite , negative semi - definite but not negative definite , or negative definite respectively . instead of the usual cauchy - schwarz - inequality we have [ eq : cauchyschwarz ] @xmath55 given a set @xmath56 ( not necessarily a subspace ) , its @xmath30-orthogonal complement is the subspace @xmath57 if @xmath45 is lightlike then @xmath58 . in fact , @xmath59 is the unique lightlike hyperplane ( cf . sect.[sec : affinespaces ] ) containing @xmath60 . in this case the hyperplane @xmath59 is called degenerate because the restriction of @xmath30 to @xmath59 is degenerate . on the other hand , if @xmath60 is timelike / spacelike @xmath59 is spacelike / timelike and @xmath61 . now the hyperplane @xmath59 is called non - degenerate because the restriction of @xmath30 to @xmath59 is non - degenerate . given any subset @xmath56 , we can attach it to a point @xmath62 in @xmath63 : @xmath64 in particular , the causal- , chronological- , and light - doublecones at @xmath65 are given by : [ eq : affdc ] @xmath66 if @xmath67 is a subspace of @xmath19 then @xmath68 is an affine subspace of @xmath63 over @xmath67 . if @xmath67 is time- , light- , or spacelike then @xmath68 is also called time- , light- , or spacelike . of particular interest are the hyperplanes @xmath69 which are timelike , lightlike , or spacelike according to @xmath60 being spacelike , lightlike , or timelike respectively . two points @xmath70 are said to be timelike- , lightlike- , or spacelike separated if the line joining them ( equivalently : the vector @xmath71 ) is timelike , lightlike , or spacelike respectively . non - spacelike separated points are also called causally separated and the line though them is called a causal line . it is easy to show that the relation @xmath72 defines an equivalence relation ( cf . [ sec : groupactions ] ) on the set of timelike vectors . ( only transitivity is non - trivial , i.e. if @xmath73 and @xmath74 then @xmath75 . to show this , decompose @xmath76 and @xmath77 into their components parallel and perpendicular to @xmath60 . ) each of the two equivalence classes is a _ cone _ in @xmath19 , that is , a subset closed under addition and multiplication with positive numbers . vectors in the same class are said to have the same time orientation . in the same fashion , the relation @xmath78 defines an equivalence relation on the set of causal vectors , with both equivalence classes being again cones . the existence of these equivalence relations is expressed by saying that @xmath63 is _ time orientable_. picking one of the two possible time orientations is then equivalent to specifying a single timelike reference vector , @xmath79 , whose equivalence class of directions may be called the _ future_. this being done we can speak of the future ( or forward , indicated by a superscript @xmath80 ) and past ( or backward , indicated by a superscript @xmath81 ) cones : [ eq : vecc ] @xmath82 note that @xmath83 and @xmath84 . usually @xmath85 is called the future and @xmath86 the past lightcone . mathematically speaking this is an abuse of language since , in contrast to @xmath87 and @xmath88 , they are not cones : they are each invariant ( as sets ) under multiplication with positive real numbers , but adding to vectors in @xmath89 will result in a vector in @xmath88 unless the vectors were parallel . as before , these cones can be attached to the points in @xmath63 . we write in a straightforward manner : [ eq : affc ] @xmath90 the cauchy - schwarz inequalities ( [ eq : cauchyschwarz ] ) result in various generalised triangle - inequalities . clearly , for spacelike vectors , one just has the ordinary triangle inequality . but for causal or timelike vectors one has to distinguish the cases according to the relative time orientations . for example , for timelike vectors of equal time orientation , one obtains the reversed triangle inequality : @xmath91 with equality iff @xmath60 and @xmath77 are parallel . it expresses the geometry behind the ` twin paradox ' . sometimes a minkowski ` distance function ' @xmath92 is introduced through @xmath93 clearly this is not a distance function in the ordinary sense , since it is neither true that @xmath94 nor that @xmath95 for all @xmath96 . in this section we consider general isometries of minkowski space . by this we mean general bijections @xmath97 ( no requirement like continuity or even linearity is made ) which preserve the minkowski distance ( [ eq : def : minkowskidistance ] ) as well as the time or spacelike character ; hence @xmath98 poincar transformations form a special class of such isometries , namely those which are affine . are there non - affine isometries ? one might expect a whole pandora s box full of wild ( discontinuous ) ones . but , fortunately , they do not exist : any map @xmath28 satisfying @xmath99 for all @xmath60 must be linear . as a warm up , we show [ thm : nowildisometries ] let @xmath28 be a surjection ( no further conditions ) so that @xmath100 for all @xmath32 , then @xmath9 is linear . consider @xmath101 . surjectivity allows to write @xmath102 , so that @xmath103 , which vanishes for all @xmath104 . hence @xmath105 for all @xmath106 , which by non - degeneracy of @xmath30 implies the linearity of @xmath9 . this shows in particular that any bijection @xmath97 of minkowski space whose associated map @xmath28 , defined by @xmath107 for some chosen basepoint @xmath108 , preserves the minkowski metric must be a poincar transformation . as already indicated , this result can be considerably strengthened . but before going into this , we mention a special and important class of linear isometries of @xmath109 , namely reflections at non - degenerate hyperplanes . the reflection at @xmath59 is defined by @xmath110 their significance is due to the following let the dimension of @xmath19 be @xmath18 . any isometry of @xmath109 is the composition of at most @xmath18 reflections . comprehensive proofs may be found in @xcite or @xcite . the easier proof for at most @xmath111 reflections is as follows : let @xmath112 be a linear isometry and @xmath45 so that @xmath113 ( which certainly exists ) . let @xmath114 , then @xmath115 so that @xmath116 and @xmath117 can not simultaneously have zero squares . so let @xmath118 ( understood as alternatives ) , then @xmath119 and @xmath120 . hence @xmath60 is eigenvector with eigenvalue @xmath121 of the linear isometry given by @xmath122 consider now the linear isometry @xmath123 on @xmath59 with induced bilinear form @xmath124 , which is non - degenerated due to @xmath113 . we conclude by induction : at each dimension we need at most two reflections to reduce the problem by one dimension . after @xmath125 steps we have reduced the problem to one dimension , where we need at most one more reflection . hence we need at most @xmath126 reflections which upon composition with @xmath112 produce the identity . here we use that any linear isometry in @xmath59 can be canonically extended to @xmath127 by just letting it act trivially on @xmath128 . note that this proof does not make use of the signature of @xmath30 . in fact , the theorem is true for any signatures ; it only depends on @xmath30 being symmetric and non degenerate . as already mentioned , theorem[thm : nowildisometries ] can be improved upon , in the sense that the hypothesis for the map being an isometry is replaced by the hypothesis that it merely preserve some relation that derives form the metric structure , but is not equivalent to it . in fact , there are various such relations which we first have to introduce . the family of cones @xmath129 defines a partial - order relation ( cf . sect.[sec : groupactions ] ) , denoted by @xmath130 , on spacetime as follows : @xmath131 iff @xmath132 , i.e. iff @xmath71 is causal and future pointing . similarly , the family @xmath133 defines a strict partial order , denoted by @xmath134 , as follows : @xmath135 iff @xmath136 , i.e. if @xmath71 is timelike and future pointing . there is a third relation , called @xmath137 , defined as follows : @xmath138 iff @xmath139 , i.e. @xmath62 is on the future lightcone at @xmath140 . it is not a partial order due to the lack of transitivity , which , in turn , is due to the lack of the lightcone being a cone ( in the proper mathematical sense explained above ) . replacing the future ( @xmath80 ) with the past ( @xmath81 ) cones gives the relations @xmath141 , @xmath142 , and @xmath143 . it is obvious that the action of @xmath144 ( spatial reflections are permitted ) on @xmath63 maps each of the six families of cones ( [ eq : affc ] ) into itself and therefore leave each of the six relations invariant . for example : let @xmath135 and @xmath145 , then @xmath146 and @xmath71 future pointing , but also @xmath147 and @xmath148 future pointing , hence @xmath149 . another set of ` obvious ' transformations of @xmath63 leaving these relations invariant is given by all dilations : @xmath150 where @xmath151 is the constant dilation - factor and @xmath152 the centre . this follows from @xmath153 , @xmath154 , and the positivity of @xmath155 . since translations are already contained in @xmath144 , the group generated by @xmath144 and all @xmath156 is the same as the group generated by @xmath144 and all @xmath156 for fixed @xmath0 . a seemingly difficult question is this : what are the most general transformations of @xmath63 that preserve those relations ? here we understand ` transformation ' synonymously with ` bijective map ' , so that each transformation @xmath9 has in inverse @xmath10 . ` preserving the relation ' is taken to mean that @xmath9 _ and _ @xmath10 preserve the relation . then the somewhat surprising answer to the question just posed is that , in three or more spacetime dimensions , there are no other such transformations besides those already listed : let @xmath157 stand for any of the relations @xmath158 and let @xmath159 be a bijection of @xmath63 with @xmath160 , such that @xmath161 implies @xmath162 and @xmath163 . then @xmath159 is the composition of an lorentz transformation in @xmath144 with a dilation . these results were proven by a.d.alexandrov and independently by e.c . a good review of alexandrov s results is @xcite ; zeeman s paper is @xcite . the restriction to @xmath160 is indeed necessary , as for @xmath164 the following possibility exists : identify @xmath165 with @xmath166 and the bilinear form @xmath167 , where @xmath168 . set @xmath169 and @xmath170 and define @xmath171 by @xmath172 , where @xmath173 is any smooth function with @xmath174 . this defines an orientation preserving diffeomorphism of @xmath166 which transforms the set of lines @xmath175 const . and @xmath176 const . respectively into each other . hence it preserves the families of cones ( [ eq : def : affcausalc ] ) . since these transformations need not be affine linear they are not generated by dilations and lorentz transformations . these results may appear surprising since without a continuity requirement one might expect all sorts of wild behaviour to allow for more possibilities . however , a little closer inspection reveals a fairly obvious reason for why continuity is implied here . consider the case in which a transformation @xmath159 preserves the families @xmath133 and @xmath177 . the open diamond - shaped sets ( usually just called ` open diamonds ' ) , @xmath178 are obviously open in the standard topology of @xmath63 ( which is that of @xmath179 ) . note that at least one of the intersections in ( [ eq : opendiamonds ] ) is always empty . conversely , is is also easy to see that each open set of @xmath63 contains an open diamond . hence the topology that is defined by taking the @xmath180 as sub - base ( the basis being given by their finite intersections ) is equivalent to the standard topology of @xmath63 . but , by hypothesis , @xmath159 and @xmath181 preserves the cones @xmath182 and therefore open sets , so that @xmath159 must , in fact , be a homeomorphism . there is no such _ obvious _ continuity input if one makes the strictly weaker requirement that instead of the cones ( [ eq : affc ] ) one only preserves the doublecones ( [ eq : affdc ] ) . does that allow for more transformations , except for the obvious time reflection ? the answer is again in the negative . the following result was shown by alexandrov ( see his review @xcite ) and later , in a different fashion , by borchers and hegerfeld @xcite : [ thm : borchershegerfeld ] let @xmath183 denote any of the relations : @xmath184 iff @xmath185 , @xmath184 iff @xmath186 , or @xmath184 iff @xmath187 . let @xmath159 be a bijection of @xmath63 with @xmath160 , such that @xmath184 implies @xmath188 and @xmath189 . then @xmath159 is the composition of an lorentz transformation in @xmath15 with a dilation . all this shows that , up to dilations , lorentz transformations can be characterised by the causal structure of minkowski space . let us focus on a particular sub - case of theorem[thm : borchershegerfeld ] , which says that any bijection @xmath159 of @xmath63 with @xmath160 , which satisfies @xmath190 must be the composition of a dilation and a transformation in @xmath15 . this is sometimes referred to as _ alexandrov s theorem_. it gives a precise answer to the following physical question : to what extent does the principle of the constancy of a finite speed of light _ alone _ determine the relativity group ? the answer is , that it determines it to be a subgroup of the 11-parameter group of poincar transformations and constant rescalings , which is as close to the poincar group as possibly imaginable . alexandrov s theorem is , to my knowledge , the closest analog in minkowskian geometry to the famous theorem of beckman and quarles @xcite , which refers to euclidean geometry and reads as follows under slightly weaker hypotheses : they allowed the map @xmath9 to be ` many - valued ' , that is , to be a map @xmath191 , where @xmath192 is the set of non - empty subsets of @xmath179 , such that @xmath193 for any @xmath194 and any @xmath195 . however , given the statement of theorem[thm : beckmanquarles ] , it is immediate that such ` many - valued maps ' must necessarily be single - valued . to see this , assume that @xmath196 has the two image points @xmath197 and define @xmath198 for @xmath199 such that @xmath200 for all @xmath201 and @xmath202 . then , according to theorem[thm : beckmanquarles ] , @xmath203 must both be euclidean motions . since they are continuous and coincide for all @xmath201 , they must also coincide at @xmath204 . ] : [ thm : beckmanquarles ] let @xmath179 for @xmath16 be endowed with the standard euclidean inner product @xmath205 . the associated norm is given by @xmath206 . let @xmath207 be any fixed positive real number and @xmath208 any map such that @xmath209 ; then @xmath9 is a euclidean motion , i.e. @xmath210 . note that there are three obvious points which let the result of beckman and quarles in euclidean space appear somewhat stronger than the theorem of alexandrov in minkowski space : * the conclusion of theorem[thm : beckmanquarles ] holds for _ any _ @xmath211 , whereas alexandrov s theorem singles out lightlike distances . * in theorem[thm : beckmanquarles ] , @xmath164 is not excluded . * in theorem[thm : beckmanquarles ] , @xmath9 is not required to be a bijection , so that we did not assume the existence of an inverse map @xmath10 . correspondingly , there is no assumption that @xmath10 also preserves the distance @xmath207 . in this subsection we wish to discuss the extent to which the law of inertia already determines the automorphism group of spacetime . the law of inertia privileges a subset of paths in spacetime form among all paths ; it defines a so - called _ path structure _ @xcite@xcite . these privileged paths correspond to the motions of privileged objects called _ free particles_. the existence of such privileged objects is by no means obvious and must be taken as a contingent and particularly kind property of nature . it has been known for long @xcite@xcite@xcite how to operationally construct timescales and spatial reference frames relative to which free particles will move uniformly and on straight lines respectively all of them ! ( a summary of these papers is given in @xcite . ) these special timescales and spatial reference frames were termed _ inertial _ by ludwig lange @xcite . their existence must again be taken as a very particular and very kind feature of nature . note that ` uniform in time ' and ` spatially straight ' together translate to ` straight in spacetime ' . we also emphasise that ` straightness ' of ensembles of paths can be characterised intrinsically , e.g. , by the desargues property @xcite . all this is true if free particles are given . we do not discuss at this point whether and how one should characterise them independently ( cf . @xcite ) . the spacetime structure so defined is usually referred to as _ projective_. it it not quite that of an affine space , since the latter provides in addition each straight line with a distinguished two - parameter family of parametrisations , corresponding to a notion of _ uniformity _ with which the line is traced through . such a privileged parametrisation of spacetime paths is not provided by the law of inertia , which only provides privileged parametrisations of spatial paths , which we already took into account in the projective structure of spacetime . instead , an affine structure of spacetime may once more be motivated by another contingent property of nature , shown by the existence of elementary clocks ( atomic frequencies ) which do define the same uniformity structure on inertial world lines all of them ! once more this is a highly non - trivial and very kind feature of nature . in this way we would indeed arrive at the statement that spacetime is an affine space . however , as we shall discuss in this subsection , the affine group already emerges as automorphism group of inertial structures without the introduction of elementary clocks . first we recall the main theorem of affine geometry . for that we make the following [ def : collineation ] three points in an affine space are called * collinear * iff they are contained in a single line . a map between affine spaces is called a * collineation * iff it maps each triple of collinear points to collinear points . note that in this definition no other condition is required of the map , like , e.g. , injectivity . the main theorem now reads as follows : [ thm : collareaffine ] a bijective collineation of a real affine space of dimension @xmath16 is necessarily an affine map . a proof may be found in @xcite . that the theorem is non - trivial can , e.g. , be seen from the fact that it is not true for complex affine spaces . the crucial property of the real number field is that it does not allow for a non - trivial automorphisms ( as field ) . a particular consequence of theorem[thm : collareaffine ] is that bijective collineation are necessarily continuous ( in the natural topology of affine space ) . this is of interest for the applications we have in mind for the following reason : consider the set @xmath212 of all lines in some affine space @xmath2 . @xmath212 has a natural topology induced from @xmath2 . theorem[thm : collareaffine ] now implies that bijective collineations of @xmath2 act as homeomorphism of @xmath212 . consider an open subset @xmath213 and the subset of all collineations that fix @xmath214 ( as set , not necessarily its points ) . then these collineations also fix the boundary @xmath215 of @xmath214 in @xmath212 . for example , if @xmath214 is the set of all timelike lines in minkowski space , i.e. , with a slope less than some chosen value relative to some fixed direction , then it follows that the bijective collineations which together with their inverse map timelike lines to timelike lines also maps the lightcone to the lightcone . it immediately follows that it must be the composition of a poincar transformation as a constant dilation . note that this argument also works in two spacetime dimensions , where the alexandrov - zeeman result does not hold . the application we have in mind is to inertial motions , which are given by lines in affine space . in that respect theorem[thm : collareaffine ] is not quite appropriate . its hypotheses are weaker than needed , insofar as it would suffice to require straight lines to be mapped to straight lines . but , more importantly , the hypotheses are also stronger than what seems physically justifiable , insofar as not every line is realisable by an inertial motion . in particular , one would like to know whether theorem[thm : collareaffine ] can still be derived by restricting to _ slow collineations _ , which one may define by the property that the corresponding lines should have a slope less than some non - zero angle ( in whatever measure , as long as the set of slow lines is open in the set of all lines ) from a given ( time-)direction . this is indeed the case , as one may show from going through the proof of theorem[thm : collareaffine ] . slightly easier to prove is the following : [ thm : timelikecollareaffine ] let @xmath159 be a bijection of real @xmath18-dimensional affine space that maps slow lines to slow lines , then @xmath159 is an affine map . a proof may be found in @xcite . if ` slowness ' is defined via the lightcone of a minkowski metric @xmath30 , one immediately obtains the result that the affine maps must be composed from poincar transformations and dilations . the reason is [ thm : samekernelpropto ] let @xmath19 be a finite dimensional real vector space of dimension @xmath16 and @xmath30 be a non - degenerate symmetric bilinear form on @xmath19 of signature @xmath21 . let @xmath216 be any other symmetric bilinear form on @xmath19 . the ` light cones ' for both forms are defined by @xmath217 and @xmath218 . suppose @xmath219 , then @xmath220 for some @xmath221 . let @xmath222 be a basis of @xmath19 such that @xmath223 . then @xmath224 for @xmath225 implies ( we write @xmath226 ) : @xmath227 and @xmath228 . further , @xmath229 for @xmath230 then implies @xmath231 for @xmath232 . hence @xmath220 with @xmath233 . this can be applied as follows : if @xmath234 is affine and maps lightlike lines to lightlike lines , then the associated linear map @xmath28 maps lightlike vectors to lightlike vectors . hence @xmath235 vanishes if @xmath236 vanishes and therefore @xmath237 by lemma[thm : samekernelpropto ] . since @xmath238 is timelike if @xmath60 is timelike , @xmath239 is positive . hence we may define @xmath240 and have @xmath241 for all @xmath242 , saying that @xmath243 is a lorentz transformation . @xmath9 is the composition of a lorentz transformation and a dilation by @xmath244 . as is well known , the two main ingredients in special relativity are the principle of relativity ( henceforth abbreviated by pr ) and the principle of the constancy of light . we have seen above that , due to alexandrov s theorem , the latter almost suffices to arrive at the poincar group . in this section we wish to address the complementary question : under what conditions and to what extent can the rp _ alone _ justify the poincar group ? this question was first addressed by ignatowsky @xcite , who showed that under a certain set of technical assumptions ( not consistently spelled out by him ) the rp alone suffices to arrive at a spacetime symmetry group which is either the inhomogeneous galilei or the inhomogeneous lorentz group , the latter for some yet undetermined limiting velocity @xmath245 . more precisely , what is actually shown in this fashion is , as we will see , that the relativity group must contain either the proper orthochronous galilei or lorentz group , if the group is required to comprise at least spacetime translations , spatial rotations , and boosts ( velocity transformations ) . what we hence gain is the group - theoretic insight of how these transformations must combine into a common group , given that they form a group at all . we do not learn anything about other transformations , like spacetime reflections or dilations , whose existence we neither required nor ruled out at this level . the work of ignatowsky was put into a logically more coherent form by franck & rothe @xcite@xcite , who showed that some of the technical assumptions could be dropped . further formal simplifications were achieved by berzi & gorini @xcite . below we shall basically follow their line of reasoning , except that we do not impose the continuity of the transformations as a requirement , but conclude it from their preservation of the inertial structure plus bijectivity . see also @xcite for an alternative discussion on the level of lie algebras . for further determination of the automorphism group of spacetime we invoke the following principles : * homogeneity of spacetime . * isotropy of space . * galilean principle of relativity . we take st1 to mean that the sought - for group should include all translations and hence be a subgroup of the general affine group . with respect to some chosen basis , it must be of the form @xmath246 , where @xmath247 is a subgroup of @xmath248 . st2 is interpreted as saying that @xmath249 should include the set of all spatial rotations . if , with respect to some frame , we write the general element @xmath250 in a @xmath251 split form ( thinking of the first coordinate as time , the other three as space ) , we want @xmath247 to include all @xmath252 finally , st3 says that velocity transformations , henceforth called ` boosts ' , are also contained in @xmath247 . however , at this stage we do not know how boosts are to be represented mathematically . let us make the following assumptions : * boosts @xmath253 are labelled by a vector @xmath254 , where @xmath255 is the open ball in @xmath256 of radius @xmath245 . the physical interpretation of @xmath257 shall be that of the boost velocity , as measured in the system from which the transformation is carried out . we allow @xmath245 to be finite or infinite ( @xmath258 ) . @xmath259 corresponds to the identity transformation , i.e. @xmath260 . we also assume that @xmath257 , considered as coordinate function on the group , is continuous . * as part of st2 we require equivariance of boosts under rotations : @xmath261 the latter assumption allows us to restrict attention to boost in a fixed direction , say that of the positive @xmath262-axis . once their analytical form is determined as function of @xmath60 , where @xmath263 , we deduce the general expression for boosts using ( [ eq : equivboostsrot ] ) and ( [ eq : so3imbedding ] ) . we make no assumptions involving space reflections . ) explicitly , though they all use it ( implicitly ) , usually in statements to the effect that it is sufficient to consider boosts in one fixed direction . once this restriction is effected , a one - dimensional spatial reflection transformation is considered to relate a boost transformation to that with opposite velocity . this then gives the impression that reflection equivariance is also invoked , though this is not necessary in spacetime dimensions greater than two , for ( [ eq : equivboostsrot ] ) allows to invert one axis through a 180-degree rotation about a perpendicular one . ] we now restrict attention to @xmath263 . we wish to determine the most general form of @xmath253 compatible with all requirements put so far . we proceed in several steps : 1 . using an arbitrary rotation @xmath264 around the @xmath262-axis , so that @xmath265 , equation ( [ eq : equivboostsrot ] ) allows to prove that @xmath266 where here we wrote the @xmath267 matrix in a @xmath268 decomposed form . ( i.e. @xmath269 is a @xmath270 matrix and @xmath271 is the @xmath270 unit - matrix ) . applying ( [ eq : equivboostsrot ] ) once more , this time using a @xmath272-rotation about the @xmath273-axis , we learn that @xmath239 is an even function , i.e. @xmath274 below we will see that @xmath275 . 2 . let us now focus on @xmath276 , which defines the action of the boost in the @xmath277 plane . we write @xmath278 we refer to the system with coordinates @xmath279 as @xmath280 and that with coordinates @xmath281 as @xmath282 . from ( [ eq : formboost3 ] ) and the inverse ( which is elementary to compute ) one infers that the velocity @xmath60 of @xmath282 with respect to @xmath280 and the velocity @xmath283 of @xmath280 with respect to @xmath282 are given by + [ eq : formboost4 ] @xmath284 + since the transformation @xmath285 is the inverse of @xmath286 , the function @xmath287 obeys @xmath288 hence @xmath289 is a bijection of the open interval @xmath290 onto itself and obeys @xmath291 3 . next we determine @xmath289 . once more using ( [ eq : equivboostsrot ] ) , where @xmath264 is a @xmath272-rotation about the @xmath273-axis , shows that the functions @xmath292 and @xmath293 in ( [ eq : formboost1 ] ) are even and the functions @xmath294 and @xmath245 are odd . the definition ( [ eq : formboost4b ] ) of @xmath289 then implies that @xmath289 is odd . since we assumed @xmath257 to be a continuous coordinatisation of a topological group , the map @xmath289 must also be continuous ( since the inversion map , @xmath295 , is continuous in a topological group ) . a standard theorem now states that a continuous bijection of an interval of @xmath296 onto itself must be strictly monotonic . together with ( [ eq : formboost6 ] ) this implies that @xmath289 is either the identity or minus the identity map . to save notation , so that ( [ eq : formboost6 ] ) now reads @xmath297 . first assume that @xmath289 is strictly monotonically increasing , then @xmath298 implies @xmath299 , a contradiction , and @xmath300 implies @xmath301 , likewise a contradiction . hence @xmath302 in this case . next assume @xmath289 is strictly monotonically decreasing . then @xmath303 is a strictly monotonically increasing map of the interval @xmath290 to itself that obeys ( [ eq : formboost6 ] ) . hence , as just seen , @xmath304 , i.e. @xmath305 . ] if it is the identity map , evaluation of ( [ eq : formboost5 ] ) shows that either the determinant of @xmath269 must equals @xmath306 , or that @xmath269 is the identity for all @xmath257 . we exclude the second possibility straightaway and the first one on the grounds that we required @xmath269 be the identity for @xmath307 . also , in that case , ( [ eq : formboost5 ] ) implies @xmath308 for all @xmath309 . we conclude that @xmath305 , which implies that the relative velocity of @xmath280 with respect to @xmath282 is minus the relative velocity of @xmath282 with respect to @xmath280 . plausible as it might seem , there is no a priori reason why this should be so . and @xmath283 are measured with different sets of rods and clocks . ] . on the face of it , the rp only implies ( [ eq : formboost6 ] ) , not the stronger relation @xmath310 . this was first pointed out in @xcite . 4 . we briefly revisit ( [ eq : formboosts2 ] ) . since we have seen that @xmath311 is the inverse of @xmath312 , we must have @xmath313 , so that ( [ eq : formboosts2 ] ) implies @xmath314 . but only @xmath315 is compatible with our requirement that @xmath316 be the identity . now we return to the determination of @xmath269 . using ( [ eq : formboost4 ] ) and @xmath305 , we write @xmath317 and @xmath318\,.\ ] ] equation @xmath319 is now equivalent to + [ eq : formboost8 ] @xmath320 + since , as already seen , @xmath292 is an even and @xmath294 is an odd function , ( [ eq : formboost8 ] ) is equivalent to @xmath321 , i.e. the unimodularity of @xmath253 . equation ( [ eq : formboost7 ] ) then allows to express @xmath294 in terms of @xmath292 : @xmath322\,.\ ] ] 6 . our problem is now reduced to the determination of the single function @xmath292 . this we achieve by employing the requirement that the composition of two boosts in the same direction results again in a boost in that direction , i.e. @xmath323 according to ( [ eq : formboost6b ] ) each matrix @xmath269 has equal diagonal entries . applied to the product matrix on the left hand side of ( [ eq : formboost10 ] ) this implies that @xmath324 is independent of @xmath60 , i.e. equal to some constant @xmath325 whose physical dimension is that of an inverse velocity squared . hence we have @xmath326 where we have chosen the positive square root since we require @xmath327 . the other implications of ( [ eq : formboost10 ] ) are + [ eq : formboost12 ] @xmath328 + from which we deduce @xmath329 conversely , ( [ eq : formboost11 ] ) and ( [ eq : formboost13 ] ) imply ( [ eq : formboost12 ] ) . we conclude that ( [ eq : formboost10 ] ) is equivalent to ( [ eq : formboost11 ] ) and ( [ eq : formboost13 ] ) . so far a boost in @xmath262 direction has been shown to act non - trivially only in the @xmath277 plane , where its action is given by the matrix that results from inserting ( [ eq : formboost9 ] ) and ( [ eq : formboost11 ] ) into ( [ eq : formboost6b ] ) : @xmath330 * if @xmath331 we rescale @xmath332 and set @xmath333 . then ( [ eq : formboost14 ] ) is seen to be a euclidean rotation with angle @xmath239 in the @xmath334 plane . the velocity spectrum is the whole real line plus infinity , i.e. a circle , corresponding to @xmath335 $ ] , where @xmath336 and @xmath337 are identified . accordingly , the composition law ( [ eq : formboost13 ] ) is just ordinary addition for the angle @xmath239 . this causes several paradoxa when @xmath60 is interpreted as velocity . for example , composing two finite velocities @xmath338 which satisfy @xmath339 results in @xmath340 , and composing two finite and positive velocities , each of which is greater than @xmath341 , results in a finite but negative velocity . in this way the successive composition of finite positive velocities could also result in zero velocity . the group @xmath342 obtained in this fashion is , in fact , @xmath343 . this group may be uniquely characterised as the largest connected group of bijections of @xmath344 that preserves the euclidean distance measure . in particular , it treats time symmetrically with all space directions , so that no invariant notion of time - orientability can be given in this case . * for @xmath345 the transformations are just the ordinary boosts of the galilei group . the velocity spectrum is the whole real line ( i.e. @xmath60 is unbounded but finite ) and @xmath247 is the galilei group . the law for composing velocities is just ordinary vector addition . * finally , for @xmath346 , one infers from ( [ eq : formboost13 ] ) that @xmath347 is an upper bound for all velocities , in the sense that composing two velocities taken from the interval @xmath290 always results in a velocity from within that interval . writing @xmath348 , @xmath349 , and @xmath350 , the matrix ( [ eq : formboost14 ] ) is seen to be a _ lorentz boost _ or _ hyperbolic motion _ in the @xmath334 plane : @xmath351 the quantity @xmath352 is called _ _ rapidity__. if rewritten in terms of the corresponding rapidities the composition law ( [ eq : formboost13 ] ) reduces to ordinary addition : @xmath353 . this shows that only the galilei and the lorentz group survive as candidates for any symmetry group implementing the rp . once the lorentz group for velocity parameter @xmath245 is chosen , one may fully characterise it by its property to leave a certain symmetric bilinear form invariant . in this sense we geometric structure of minkowski space can be deduced . this closes the circle to where we started from in section[sec : frommetrictoaffine ] . in the previous sections we always understood an automorphisms of a structured set ( spacetime ) as a bijection . mathematically this seems an obvious requirement , but from a physical point of view this is less clear . the physical law of inertia provides us with distinguished motions _ locally _ in space and time . hence one may attempt to relax the condition for structure preserving maps , so as to only preserve inertial motions _ hence we ask the following question : what are the most general maps that _ locally _ map segments of straight lines to segments of straight lines ? this local approach has been pursued by @xcite . to answer this question completely , let us ( locally ) identify spacetime with @xmath179 where @xmath16 and assume the map to be @xmath354 , that is , three times continuously differentiable . so let @xmath355 be an open subset and determine all @xmath354 maps @xmath356 that map straight segments in @xmath357 into straight segments in @xmath179 . in coordinates we write @xmath358 and @xmath359 , so that @xmath360 . a straight segment in @xmath357 is a curve @xmath361 ( the open interval @xmath362 is usually taken to contain zero ) whose acceleration is pointwise proportional to its velocity . this is equivalent to saying that it can be parametrised so as to have zero acceleration , i.e. , @xmath363 for some @xmath364 . for the image path @xmath365 to be again straight its acceleration , @xmath366 , must be proportional to its velocity , @xmath367 , where the factor of proportionality , @xmath368 , depends on the point of the path and separately on @xmath292 . hence , in coordinates , we have @xmath369 for each @xmath294 this must be valid for all @xmath370 in a neighbourhood of zero in @xmath371 . taking the second derivatives with respect to @xmath292 , evaluation at @xmath372 , @xmath373 leads to [ eq : projmap2 ] @xmath374 where @xmath375 here we suppressed the remaining argument @xmath294 . equation ( [ eq : projmap2 ] ) is valid at each point in @xmath357 . integrability of ( [ eq : projmap2-a ] ) requires that its further differentiation is totally symmetric with respect to all lower indices ( here we use that the map @xmath9 is @xmath354 ) . this leads to @xmath376 inserting ( [ eq : projmap2-b ] ) one can show ( upon taking traces over @xmath377 and @xmath378 ) that the resulting equation is equivalent to @xmath379 in particular @xmath380 so that there is a local function @xmath381 ( if @xmath357 is simply connected , as we shall assume ) for which @xmath382 . equation ( [ eq : projmap4 ] ) is then equivalent to @xmath383 so that @xmath384 for some @xmath385 and @xmath386 . using @xmath387 and ( [ eq : projmap4 ] ) , equation ( [ eq : projmap2-a ] ) is equivalent to @xmath388=0 $ ] , which finally leads to the result that the most general solution for @xmath9 is given by @xmath389 here @xmath390 is a @xmath391 matrix , @xmath292 and @xmath140 vectors in @xmath179 , and @xmath386 . @xmath62 and @xmath140 must be such that @xmath357 does not intersect the hyperplane @xmath392 where @xmath9 becomes singular , but otherwise they are arbitrary . , i.e. iff @xmath394 , the transformations ( [ eq : projmap5 ] ) are not affine . in this case they are called proper projective . are there physical reasons to rule out such proper projective transformations ? a structural argument is that they do not leave any subset of @xmath179 invariant and that they hence can not be considered as automorphism group of any subdomain . a physical argument is that two separate points that move with the same velocity cease to do so if their worldlines are transformed by by a proper projective transformation . in particular , a rigid motion of an extended body ( undergoing inertial motion ) ceases to be rigid if so transformed ( cf.@xcite , p.16 ) . an illustrative example is the following : consider the one - parameter ( @xmath395 ) family of parallel lines @xmath396 ( where @xmath397 is the parameter along each line ) , and the proper projective map @xmath398 which becomes singular on the hyperplane @xmath399 . the one - parameter family of image lines @xmath400 have velocities @xmath401 whose directions are independent of @xmath397 , showing that they are indeed straight . however , the velocity directions now depend on @xmath395 , showing that they are not parallel anymore . let us , regardless of this , for the moment take seriously the transformations ( [ eq : projmap5 ] ) . one may reduce them to the following form of generalised boosts , discarding translations and rotations and using equivariance with respect to the latter ( we restrict to four spacetime dimensions from now on ) : [ eq : focklorentztrans1 ] @xmath402 where @xmath403 represents the boost velocity , @xmath404 its modulus , and all functions of @xmath60 are even . the subscripts @xmath405 and @xmath406 refer to the components parallel and perpendicular to @xmath257 . now one imposes the following conditions which allow to determine the eight functions @xmath407 , of which only seven are considered independent since common factors of the numerator and denominator cancel ( we essentially follow @xcite ) : 1 . the origin @xmath408 has velocity @xmath257 in the unprimed coordinates , leading to @xmath409 and thereby eliminating @xmath410 as independent function . the origin @xmath411 has velocity @xmath412 in the primed coordinates , leading to @xmath413 and thereby eliminating @xmath293 as independent function . reciprocity : the transformation parametrised by @xmath412 is the inverse of that parametrised by @xmath257 , leading to relations @xmath414 , @xmath415 , and @xmath416 , thereby eliminating @xmath417 as independent functions . of the remaining three functions @xmath418 an overall factor in the numerator and denominator can be split off so that two free functions remain . transitivity : the composition of two transformations of the type ( [ eq : focklorentztrans1 ] ) with parameters @xmath257 and @xmath419 must be again of this form with some parameter @xmath420 , which turns out to be the same function of the velocities @xmath257 and @xmath419 as in special relativity ( einstein s addition law ) , for reasons to become clear soon . this allows to determine the last two functions in terms of two constants @xmath245 and @xmath421 whose physical dimensions are that of a velocity and of a length respectively . writing , as usual , @xmath422 the final form is given by [ eq : focklorentztrans2 ] @xmath423 in the limit as @xmath424 this approaches an ordinary lorentz boost : @xmath425 moreover , for finite @xmath421 the map ( [ eq : focklorentztrans2 ] ) is conjugate to ( [ eq : lorentzfock3 ] ) with respect to a time dependent deformation . to see this , observe that the common denominator in ( [ eq : focklorentztrans2 ] ) is just @xmath426 , whereas the numerators correspond to ( [ eq : lorentzfock3 ] ) . hence , introducing the deformation map @xmath427 and denoting the map @xmath428 in ( [ eq : focklorentztrans2 ] ) by @xmath9 , we have @xmath429 note that @xmath112 is singular at the hyperplane @xmath430 and has no point of the hyperplane @xmath431 in its image . the latter hyperplane is the singularity set of @xmath432 . outside the hyperplanes @xmath433 the map @xmath112 relates the following time slabs in a diffeomorphic fashion : [ eq : difftimeslabs ] @xmath434 since boosts leave the upper - half spacetime , @xmath435 , invariant ( as set ) , ( [ eq : difftimeslabs - a ] ) shows that @xmath9 just squashes the linear action of boosts in @xmath436 into a non - linear action within @xmath437 , where @xmath421 now corresponds to an invariant scale . interestingly , this is the same deformation of boosts that have been recently considered in what is sometimes called _ doubly special relativity _ ( because there are now two , rather than just one , invariant scales , @xmath421 and @xmath245 ) , albeit there the deformation of boosts take place in momentum space where @xmath421 then corresponds to an invariant energy scale ; see @xcite and also @xcite . in this section we wish to discuss in more detail some of the non - trivial structures in minkowski . i have chosen them so as to emphasise the difference to the corresponding structures in galilean spacetime , and also because they do not seem to be much discussed in other standard sources . let us start right away by characterising those vectors for which we have an inverted cauchy - schwarz inequality : [ lemma : stricticsi ] let @xmath19 be of dimension @xmath438 and @xmath45 be some non - zero vector . the strict inverted cauchy - schwarz inequality , @xmath439 holds for all @xmath106 linearly independent of @xmath60 iff @xmath60 is timelike . obviously @xmath60 can not be spacelike , for then we would violate ( [ eq : strictinvcsi ] ) with any spacelike @xmath77 . if @xmath60 is lightlike then @xmath77 violates ( [ eq : strictinvcsi ] ) iff it is in the set @xmath440 , which is non - empty iff @xmath438 . hence @xmath60 can not be lightlike if @xmath438 . if @xmath60 is timelike we decompose @xmath441 with @xmath442 so that @xmath443 , with equality iff @xmath60 and @xmath77 are linearly dependent . hence @xmath444 with equality iff @xmath60 and @xmath77 are linearly dependent . the next lemma deals with the intersection of a causal line with a light cone , a situation depicted in fig.[fig : robb ] . [ lemma : intlinecone ] let @xmath445 be the light - doublecone with vertex @xmath62 and @xmath446 be a non - spacelike line , i.e. @xmath447 , through @xmath448 . if @xmath60 is timelike @xmath449 consists of two points . if @xmath60 is lightlike this intersection consists of one point if @xmath450 and is empty if @xmath451 . note that the latter two statements are independent of the choice of @xmath452as they must be , i.e. are invariant under @xmath453 , where @xmath454 . we have @xmath455 iff @xmath456 for @xmath60 timelike we have @xmath457 and ( [ eq : lemmaintlinecone1 ] ) has two solutions @xmath458 indeed , since @xmath448 , the vectors @xmath60 and @xmath459 can not be linearly dependent so that lemma[lemma : stricticsi ] implies the positivity of the expression under the square root . if @xmath60 is lightlike ( [ eq : lemmaintlinecone1 ] ) becomes a linear equation which is has one solution if @xmath460 and no solution if @xmath461 [ note that @xmath462 since @xmath463 by hypothesis ] . ( -12,155)@xmath464 ( -22,114)@xmath140 ( -35,56)@xmath465 ( -72,98)@xmath62 ( -60,0)@xmath466 ( -42,15)@xmath60 ( -75,146)@xmath467 ( -75,62)@xmath468 ( -13,195)@xmath469 [ prop : einsteinsynch ] let @xmath469 and @xmath445 as in lemma[lemma : intlinecone ] with @xmath60 timelike . let @xmath464 and @xmath465 be the two intersection points of @xmath469 with @xmath445 and @xmath470 a point between them . then @xmath471 moreover , @xmath472 iff @xmath71 is perpendicular to @xmath60 . the vectors @xmath473 and @xmath474 are lightlike , which gives ( note that @xmath475 is spacelike ) : [ eq : prop : einsteinsynch23 ] @xmath476 since @xmath477 and @xmath478 are parallel we have @xmath479 with @xmath480 so that @xmath481 and @xmath482 . now , multiplying ( [ eq : prop : einsteinsynch3 ] ) with @xmath155 and adding this to ( [ eq : prop : einsteinsynch2 ] ) immediately yields @xmath483 since @xmath484 this implies ( [ eq : prop : einsteinsynch1 ] ) . finally , since @xmath477 and @xmath485 are antiparallel , @xmath486 iff @xmath487 . equations ( [ eq : prop : einsteinsynch23 ] ) now show that this is the case iff @xmath488 , i.e. iff @xmath489 . hence we have shown @xmath490 in other words , @xmath140 is the midpoint of the segment @xmath491 iff the line through @xmath62 and @xmath140 is perpendicular ( wrt . @xmath30 ) to @xmath469 . the somewhat surprising feature of the first statement of this proposition is that ( [ eq : prop : einsteinsynch1 ] ) holds for _ any _ point of the segment @xmath491 , not just the midpoint , as it would have to be the case for the corresponding statement in euclidean geometry . the second statement of proposition[prop : einsteinsynch ] gives a convenient geometric characterisation of einstein - simultaneity . recall that an event @xmath140 on a timelike line @xmath469 ( representing an inertial observer ) is defined to be einstein - simultaneous with an event @xmath62 in spacetime iff @xmath140 bisects the segment @xmath491 between the intersection points @xmath492 of @xmath469 with the double - lightcone at @xmath62 . hence proposition[prop : einsteinsynch ] implies [ corol : einsteinsim ] einstein simultaneity with respect to a timelike line @xmath469 is an equivalence relation on spacetime , the equivalence classes of which are the spacelike hyperplanes orthogonal ( wrt . @xmath30 ) to @xmath469 . the first statement simply follows from the fact that the family of parallel hyperplanes orthogonal to @xmath469 form a partition ( cf . sect.[sec : groupactions ] ) of spacetime . from now on we shall use the terms ` timelike line ' and ` inertial observer ' synonymously . note that einstein simultaneity is only defined relative to an inertial observer . given two inertial observers , [ eq : twoobservers ] @xmath493 we call the corresponding einstein - simultaneity relations _ @xmath469-simultaneity _ and _ @xmath494-simultaneity_. obviously they coincide iff @xmath469 and @xmath494 are parallel ( @xmath60 and @xmath283 are linearly dependent ) . in this case @xmath495 is @xmath469-simultaneous to @xmath470 iff @xmath470 is @xmath494-simultaneous to @xmath495 . if @xmath469 and @xmath494are not parallel ( skew or intersecting in one point ) it is generally not true that if @xmath495 is @xmath469-simultaneous to @xmath470 then @xmath470 is also @xmath494-simultaneous to @xmath495 . in fact , we have [ prop : skewsim ] let @xmath469 and @xmath494 two non - parallel timelike likes . there exists a unique pair @xmath496 so that @xmath497 is @xmath469-simultaneous to @xmath140 and @xmath140 is @xmath494 simultaneous to @xmath497 . we parameterise @xmath469 and @xmath494 as in ( [ eq : twoobservers ] ) . the two conditions for @xmath497 being @xmath469-simultaneous to @xmath140 and @xmath140 being @xmath494-simultaneous to @xmath497 are @xmath498 . writing @xmath499 and @xmath500 this takes the form of the following matrix equation for the two unknowns @xmath155 and @xmath501 : @xmath502 this has a unique solution pair @xmath503 , since for linearly independent timelike vectors @xmath60 and @xmath283 lemma[lemma : stricticsi ] implies @xmath504 . note that if @xmath469 and @xmath494 intersect @xmath505 . clearly , einstein - simultaneity is conventional and physics proper should not depend on it . for example , the fringe - shift in the michelson - morley experiment is independent of how we choose to synchronise clocks . in fact , it does not even make use of any clock . so what is the general definition of a ` simultaneity structure ' ? it seems obvious that it should be a relation on spacetime that is at least symmetric ( each event should be simultaneous to itself ) . going from one - way simultaneity to the mutual synchronisation of two clocks , one might like to also require reflexivity ( if @xmath62 is simultaneous to @xmath140 then @xmath140 is simultaneous to @xmath62 ) , though this is not strictly required in order to one - way synchronise each clock in a set of clocks with one preferred ` master clock ' , which is sufficient for many applications . moreover , if we like to speak of the mutual simultaneity of sets of more than two events we need an equivalence relation on spacetime . the equivalence relation should be such that each inertial observer intersect each equivalence class precisely once . let us call such a simultaneity structure ` admissible ' . clearly there are zillions of such structures : just partition spacetime into any set of appropriate spacelike hypersurfaces ( there are more possibilities at this point , like families of forward or backward lightcones ) . an _ absolute _ admissible simultaneity structure would be one which is invariant ( cf . sect.[sec : groupactions ] ) under the automorphism group of spacetime . we have [ prop : abssim ] there exits precisely one admissible simultaneity structure which is invariant under the inhomogeneous proper orthochronous galilei group and none that is invariant under the inhomogeneous proper orthochronous lorentz group . a proof is given in @xcite . there is a group - theoretic reason that highlights this existential difference : [ prop : abssimmath ] let @xmath249 be a group with transitive action on a set @xmath2 . let @xmath506 be the stabiliser subgroup for @xmath507 ( due to transitivity all stabiliser subgroups are conjugate ) . then @xmath2 admits a @xmath249-invariant equivalence relation @xmath508 iff @xmath509 is _ not _ maximal , that is , iff @xmath509 is properly contained in a proper subgroup @xmath510 of @xmath249 : @xmath511 . a proof of this may be found in @xcite ( theorem1.12 ) . regarding the action of the inhomogeneous galilei and lorentz groups on spacetime , their stabilisers are the corresponding homogeneous groups . now , the homogeneous lorentz group is maximal in the inhomogeneous one , whereas the homogeneous galilei group is not maximal in the inhomogeneous one , since it can still be supplemented by time translations without the need to also invoke space translations . this , according to proposition[prop : abssimmath ] , is the group theoretic origin of the absence of any invariant simultaneity structure in the lorentzian case . however , one may ask whether there are simultaneity structures _ relative _ to some _ additional _ structure @xmath512 . as additional structure , @xmath512 , one could , for example , take an inertial reference frame , which is characterised by a foliation of spacetime by parallel timelike lines . the stabiliser subgroup of that structure within the proper orthochronous poincar group is given by the semidirect product of spacetime translations with all rotations in the hypersurfaces perpendicular to the lines in @xmath512 : @xmath513 here the @xmath514 only acts on the spatial translations , so that the group is also isomorphic to @xmath515 , where @xmath516 is the group of euclidean motions in 3-dimensions ( the hyperplanes perpendicular to the lines in @xmath512 ) . we can now ask : how many admissible @xmath517 invariant equivalence relations are there . the answer is [ prop : realsim ] there exits precisely one admissible simultaneity structure which is invariant under @xmath517 , where @xmath512 represents am inertial reference frame ( a foliation of spacetime by parallel timelike lines ) . it is given by einstein simultaneity , that is , the equivalence classes are the hyperplanes perpendicular to the lines in @xmath512 . the proof is given in @xcite . note again the connection to quoted group - theoretic result : the stabiliser subgroup of a point in @xmath517 is @xmath518 , which is clearly not maximal in @xmath517 since it is a proper subgroup of @xmath516 which , in turn , is a proper subgroup of @xmath517 . here we wish to briefly discuss another important structure associated with causality relations in minkowski space , which plays a fundamental rle in modern quantum field theory ( see e.g. @xcite ) . let @xmath519 and @xmath520 be subsets of @xmath63 . we say that @xmath519 and @xmath520 are _ causally disjoint _ or _ spacelike separated _ iff @xmath521 is spacelike , i.e. @xmath522 , for any @xmath523 and @xmath524 . note that because a point is not spacelike separated from itself , causally disjoint sets are necessarily disjoint in the ordinary set - theoretic sense the converse being of course not true . for any subset @xmath525 we denote by @xmath526 the largest subset of @xmath63 which is causally disjoint to @xmath2 . the set @xmath526 is called the _ causal complement _ of @xmath2 . the procedure of taking the causal complement can be iterated and we set @xmath527 etc . @xmath528 is called the _ causal completion _ of @xmath2 . it also follows straight from the definition that @xmath529 implies @xmath530 and also @xmath531 . if @xmath532 we call @xmath2 _ causally complete_. we note that the causal complement @xmath526 of any given @xmath2 is automatically causally complete . indeed , from @xmath531 we obtain @xmath533 , but the first inclusion applied to @xmath526 instead of @xmath2 leads to @xmath534 , showing @xmath535 . note also that for any subset @xmath2 its causal completion , @xmath528 , is the smallest causally complete subset containing @xmath2 , for if @xmath536 with @xmath537 , we derive from the first inclusion by taking @xmath538 that @xmath539 , so that the second inclusion yields @xmath540 . trivial examples of causally complete subsets of @xmath63 are the empty set , single points , and the total set @xmath63 . others are the open diamond - shaped regions ( [ eq : opendiamonds ] ) as well as their closed counterparts : @xmath541 we now focus attention to the set @xmath542 of causally complete subsets of @xmath63 , including the empty set , @xmath543 , and the total set , @xmath63 , which are mutually causally complementary . it is partially ordered by ordinary set - theoretic inclusion @xmath544 ( cf . sect.[sec : groupactions ] ) and carries the ` dashing operation ' @xmath545 of taking the causal complement . moreover , on @xmath542 we can define the operations of ` meet ' and ` join ' , denoted by @xmath546 and @xmath547 respectively , as follows : let @xmath548 where @xmath199 , then @xmath549 is the largest causally complete subset in the intersection @xmath550 and @xmath551 is the smallest causally complete set containing the union @xmath552 . the operations of @xmath546 and @xmath547 can be characterised in terms of the ordinary set - theoretic intersection @xmath553 together with the dashing - operation . to see this , consider two causally complete sets , @xmath554 where @xmath199 , and note that the set of points that are spacelike separated from @xmath519 and @xmath520 are obviously given by @xmath555 , but also by @xmath556 , so that [ eg : capcup ] @xmath557 here ( [ eq : capcup1 ] ) and ( [ eq : capcup2 ] ) are equivalent since any @xmath548 can be written as @xmath558 , namely @xmath559 . if @xmath554 runs through all sets in @xmath542 so does @xmath560 . hence any equation that holds generally for all @xmath548 remains valid if the @xmath554 are replaced by @xmath561 . equation ( [ eq : capcup2 ] ) immediately shows that @xmath550 is causally complete ( since it is the @xmath562 of something ) . taking the causal complement of ( [ eq : capcup1 ] ) we obtain the desired relation for @xmath563 . together we have [ eg : meetjoin ] @xmath564 from these we immediately derive [ eg : negating ] @xmath565 all what we have said so far for the set @xmath542 could be repeated verbatim for the set @xmath566 of _ chronologically complete _ subsets . we say that @xmath519 and @xmath520 are _ chronologically disjoint _ or _ non - timelike separated _ , iff @xmath567 and @xmath568 for any @xmath523 and @xmath524 . @xmath526 , the _ chronological complement _ of @xmath2 , is now the largest subset of @xmath63 which is chronologically disjoint to @xmath2 . the only difference between the causal and the chronological complement of @xmath2 is that the latter now contains lightlike separated points outside @xmath2 . a set @xmath2 is _ chronologically complete _ iff @xmath569 , where the dashing now denotes the operation of taking the chronological complement . again , for any set @xmath2 the set @xmath526 is automatically chronologically complete and @xmath528 is the smallest chronologically complete subset containing @xmath2 . single points are chronologically complete subsets . all the formal properties regarding @xmath570 , @xmath546 , and @xmath547 stated hitherto for @xmath542 are the same for @xmath566 . one major difference between @xmath542 and @xmath566 is that the types of diamond - shaped sets they contain are different . for example , the closed ones , ( [ eq : closeddiamonds ] ) , are members of both . the open ones , ( [ eq : opendiamonds ] ) , are contained in @xmath542 but _ not _ in @xmath566 . instead , @xmath566 contains the closed diamonds whose ` equator'sphere in which the forward and backward light - cones in ( [ eq : closeddiamonds ] ) intersect . in the two - dimensional drawings the ` equator ' is represented by just two points marking the right and left corners of the diamond - shaped set . ] have been removed . an essential structural difference between @xmath542 and @xmath566 will be stated below , after we have introduced the notion of a lattice to which we now turn . to put all these formal properties into the right frame we recall the definition of a lattice . let @xmath571 be a partially ordered set and @xmath572 any two elements in @xmath573 . synonymously with @xmath574 we also write @xmath575 and say that @xmath292 is smaller than @xmath294 , @xmath294 is bigger than @xmath292 , or @xmath294 majorises @xmath292 . we also write @xmath576 if @xmath574 and @xmath232 . if , with respect to @xmath141 , their greatest lower and least upper bound exist , they are denoted by @xmath577called the ` meet of @xmath292 and @xmath294'and @xmath578called the ` join of @xmath292 and @xmath294'respectively . a partially ordered set for which the greatest lower and least upper bound exist for any pair @xmath572 of elements from @xmath573 is called a _ lattice_. we now list some of the most relevant additional structural elements lattices can have : a lattice is called _ complete _ if greatest lower and least upper bound exist for any subset @xmath579 . if @xmath580 they are called @xmath336 ( the smallest element in the lattice ) and @xmath121 ( the biggest element in the lattice ) respectively . an _ atom _ in a lattice is an element @xmath292 which majorises only @xmath336 , i.e. @xmath581 and if @xmath582 then @xmath583 or @xmath584 . the lattice is called _ atomic _ if each of its elements different from @xmath336 majorises an atom . an atomic lattice is called _ atomistic _ if every element is the join of the atoms it majorises . an element @xmath245 is said to _ cover _ @xmath292 if @xmath585 and if @xmath586 either @xmath587 or @xmath588 . an atomic lattice is said to have the _ covering property _ if , for every element @xmath294 and every atom @xmath292 for which @xmath589 , the join @xmath578 covers @xmath294 . the subset @xmath590 is called a _ distributive triple _ if [ eq : disttriple ] @xmath591 a lattice is called _ distributive _ or _ boolean _ if every triple @xmath592 is distributive . it is called _ modular _ if every triple @xmath592 with @xmath574 is distributive . it is straightforward to check from ( [ eq : disttriple ] ) that modularity is equivalent to the following single condition : @xmath593 if in a lattice with smallest element @xmath336 and greatest element @xmath121 a map @xmath594 , @xmath595 , exist such that [ eq : orthocomplement ] @xmath596 the lattice is called _ orthocomplemented_. it follows that whenever the meet and join of a subset @xmath597 ( @xmath598 is some index set ) exist one has demorgan s laws ) is redundant , as each of its two statements follows from the other , due to @xmath599 . ] : [ eq : gendemorgan ] @xmath600 for orthocomplemented lattices there is a still weaker version of distributivity than modularity , which turns out to be physically relevant in various contexts : [ def : orthomodular ] an orthocomplemented lattice is called _ orthomodular _ if every triple @xmath592 with @xmath574 and @xmath601 is distributive . from ( [ eq : modularity ] ) and using that @xmath602 for @xmath603 one sees that this is equivalent to the single condition ( renaming @xmath245 to @xmath604 ) : [ eq : orthomodularity ] @xmath605 where the second line follows from the first by taking its orthocomplement and renaming @xmath606 to @xmath607 . it turns out that these conditions can still be simplified by making them independent of @xmath245 . in fact , ( [ eq : orthomodularity ] ) are equivalent to [ eq : altorthomodularity ] @xmath608 it is obvious that ( [ eq : orthomodularity ] ) implies ( [ eq : altorthomodularity ] ) ( set @xmath609 ) . but the converse is also true . to see this , take e.g. ( [ eq : altorthomodularity2 ] ) and choose any @xmath610 . then @xmath611 , @xmath612 ( by hypothesis ) , and @xmath613 ( trivially ) , so that @xmath614 . hence @xmath615 , which proves ( [ eq : orthomodularity2 ] ) . complete orthomodular atomic lattices are automatically atomistic . indeed , let @xmath294 be the join of all atoms majorised by @xmath616 . assume @xmath232 so that necessarily @xmath617 , then ( [ eq : altorthomodularity2 ] ) implies @xmath618 . then there exists an atom @xmath245 majorised by @xmath619 . this implies @xmath620 and @xmath601 , hence also @xmath621 . but this is a contradiction , since @xmath294 is by definition the join of all atoms majorised by @xmath292 . finally we mention the notion of _ compatibility _ or _ commutativity _ , which is a symmetric , reflexive , but generally not transitive relation @xmath421 on an orthomodular lattice ( cf . sec.[sec : groupactions ] ) . we write @xmath622 for @xmath623 and define : [ eq : defcompatibility ] @xmath624 the equivalence of these two lines , which shows that the relation of being compatible is indeed symmetric , can be demonstrated using orthomodularity as follows : suppose ( [ eq : defcompatibility1 ] ) holds ; then @xmath625 , where we used the orthocomplement of ( [ eq : defcompatibility1 ] ) to replace @xmath626 in the first expression and the trivial identity @xmath627 in the second step . now , applying ( [ eq : altorthomodularity2 ] ) to @xmath628 we get @xmath629=(b\wedge a)\vee ( b\wedge a')$ ] , i.e. ( [ eq : defcompatibility2 ] ) . the converse , @xmath630 , is of course entirely analogous . from ( [ eq : defcompatibility ] ) a few things are immediate : @xmath622 is equivalent to @xmath631 , @xmath622 is implied by @xmath574 or @xmath632 , and the elements @xmath336 and @xmath121 are compatible with all elements in the lattice . the _ centre _ of a lattice is the set of elements which are compatible with all elements in the lattice . in fact , the centre is a boolean sublattice . if the centre contains no other elements than @xmath336 and @xmath121 the lattice is said to be _ the other extreme is a boolean lattice , which is identical to its own centre . indeed , if @xmath633 is a distributive triple , one has @xmath634 . after these digression into elementary notions of lattice theory we come back to our examples of the sets @xmath542 @xmath566 . our statements above amount to saying that they are complete , atomic , and orthocomplemented lattices . the partial order relation @xmath141 is given by @xmath635 and the extreme elements @xmath336 and @xmath121 correspond to the empty set @xmath543 and the total set @xmath63 , the points of which are the atoms . neither the covering property nor modularity is shared by any of the two lattices , as can be checked by way of elementary counterexamples . and @xmath140 . then @xmath636 whereas @xmath637 is given by the closed diamond ( [ eq : closeddiamonds ] ) . note that this is true in @xmath542 _ and _ @xmath566 . but , clearly , @xmath637 does not cover either @xmath638 or @xmath639 . ] in particular , neither of them is boolean . however , in @xcite it was shown that @xmath566 is orthomodular ; see also @xcite which deals with more general spacetimes . note that by the argument given above this implies that @xmath566 is atomistic . in contrast , @xmath542 is definitely _ not _ orthomodular , as is e.g. seen by the counterexample given in fig.[fig : countex ] . regarding this point , there are some conflicting statements in the literature . the first edition of @xcite states orthomodularity of @xmath566 in proposition4.1.3 , which is removed in the second edition without further comment . the proof offered in the first edition uses ( [ eq : altorthomodularity1 ] ) as definition of orthomodularity , writing @xmath640 for @xmath292 and @xmath641 for b. the crucial step is the claim that any spacetime event in the set @xmath642 lies in @xmath641 and that any causal line through it must intersect either @xmath640 or @xmath643 . the last statement is , however , not correct since the join of two sets ( here @xmath640 and @xmath643 ) is generally larger than the domain of dependence of their ordinary set - theoretic union ; compare fig.[fig : countex ] . : ( generally , the domain of dependence of a subset @xmath2 of spacetime @xmath1 is the largest subset @xmath644 such that any inextensible causal curve that intersects @xmath645 also intersects @xmath2 . ) ] it is also not difficult to prove that @xmath566 is irreducible . , the failure of irreducibility of @xmath646 is directly related to the existence of closed timelike curves ; see @xcite . ] ( -365,104)@xmath469 ( -344.5,72)@xmath292 ( -280,50)@xmath647 ( -322,85)@xmath648 ( -152,85)@xmath648 ( -180,50)@xmath294 ( -30,50)@xmath294 ( -105,50)@xmath647 ( -158,54)@xmath292 it is well known that the lattices of propositions for classical systems are boolean , whereas those for quantum systems are merely orthomodular . in classical physics the elements of the lattice are measurable subsets of phase space , with @xmath141 being ordinary set - theoretic inclusion @xmath635 , and @xmath546 and @xmath547 being ordinary set - theoretic intersection @xmath553 and union @xmath649 respectively . the orthocomplement is the ordinary set - theoretic complement . in quantum mechanics the elements of the lattice are the closed subspaces of hilbert space , with @xmath141 being again ordinary inclusion , @xmath546 ordinary intersection , and @xmath547 is given by @xmath650 . the orthocomplement of a closed subset is the orthogonal complement in hilbert space . for comprehensive discussions see @xcite and @xcite . one of the main questions in the foundations of quantum mechanics is whether one could understand ( derive ) the usage of hilbert spaces and complex numbers from somehow more fundamental principles . even though it is not a priori clear what ones measure of fundamentality should be at this point , an interesting line of attack consists in deriving the mentioned structures from the properties of the lattice of propositions ( quantum logic ) . it can be shown that a lattice that is complete , atomic , irreducible , orthomodular , and that satisfies the covering property is isomorphic to the lattice of closed subspaces of a linear space with hermitean inner product . the complex numbers are selected if additional technical assumptions are added . for the precise statements of these reconstruction theorems see @xcite . it is now interesting to note that , on a formal level , there is a similar transition in going from galilei invariant to lorentz invariant causality relations . in fact , in galilean spacetime one can also define a chronological complement : two points are chronologically related if they are connected by a worldline of finite speed and , accordingly , two subsets in spacetime are chronologically disjoint if no point in one set is chronologically related to a point of the other . for example , the chronological complement of a point @xmath62 are all points simultaneous to , but different from , @xmath62 . more general , it is not hard to see that the chronologically complete sets are just the subsets of some @xmath651 hypersurface . the lattice of chronologically complete sets is then the continuous disjoint union of sublattices , each of which is isomorphic to the boolean lattice of subsets in @xmath256 . for details see @xcite . as we have seen above , @xmath652 is complete , atomic , irreducible , and orthomodular ( hence atomistic ) . the main difference to the lattice of propositions in quantum mechanics , as regards the formal aspects discussed here , is that @xmath652 does _ not _ satisfy the covering property . otherwise the formal similarities are intriguing and it is tempting to ask whether there is a deeper meaning to this . in this respect it would be interesting to know whether one could give a lattice - theoretic characterisation for @xmath653 ( @xmath1 some fixed spacetime ) , comparable to the characterisation of the lattices of closed subspaces in hilbert space alluded to above . even for @xmath654 such a characterisation seems , as far as i am aware , not to be known . as is well known , the notion of a rigid body , which proves so useful in newtonian mechanics , is incompatible with the existence of a universal finite upper bound for all signal velocities @xcite . as a result , the notion of a perfectly rigid body does not exist within the framework of sr . however , the notion of a _ rigid motion _ does exist . intuitively speaking , a body moves rigidly if , locally , the relative spatial distances of its material constituents are unchanging . the motion of an extended body is described by a normalised timelike vector field @xmath655 , where @xmath214 is an open subset of minkowski space , consisting of the events where the material body in question ` exists ' . we write @xmath656 for the minkowskian scalar product . being normalised now means that @xmath657 ( we do _ not _ choose units such that @xmath658 ) . the lie derivative with respect to @xmath76 is denoted by @xmath659 . for each material part of the body in motion its local rest space at the event @xmath660 can be identified with the hyperplane through @xmath62 orthogonal to @xmath661 : @xmath662 @xmath663 carries a euclidean inner product , @xmath664 , given by the restriction of @xmath665 to @xmath663 . generally we can write @xmath666 where @xmath667 is the one - form associated to @xmath76 . following @xcite the precise definition of ` rigid motion ' can now be given as follows : [ def : rigidmotion ] let @xmath76 be a normalised timelike vector field @xmath76 . the motion described by its flow is _ rigid _ if @xmath668 note that , in contrast to the killing equations @xmath669 , these equations are non linear due to the dependence of @xmath216 upon @xmath76 . we write @xmath670 for the tensor field over spacetime that pointwise projects vectors perpendicular to @xmath76 . it acts on one forms @xmath239 via @xmath671 and accordingly on all tensors . the so extended projection map will still be denoted by @xmath672 . then we e.g. have @xmath673 it is not difficult to derive the following two equations : equation ( [ eq : rigmo2 ] ) simply follows from @xmath674 , so that @xmath675 for all @xmath676 . in fact , @xmath677 , where @xmath678 is the spacetime - acceleration . this follows from @xmath679)= g(a , x)-g(u,\nabla_xu)=g(a , x)$ ] , where @xmath680 was used in the last step . ] @xmath681 where @xmath9 is any differentiable real - valued function on @xmath214 . equation ( [ eq : rigmo1 ] ) shows that the normalised vector field @xmath76 satisfies ( [ eq : def : rigmo ] ) iff any rescaling @xmath682 with a nowhere vanishing function @xmath9 does . hence the normalization condition for @xmath76 in ( [ eq : def : rigmo ] ) is really irrelevant . it is the geometry in spacetime of the flow lines and not their parameterisation which decide on whether motions ( all , i.e. for any parameterisation , or none ) along them are rigid . this has be the case because , generally speaking , there is no distinguished family of sections ( hypersurfaces ) across the bundle of flow lines that would represent ` the body in space ' , i.e. mutually simultaneous locations of the body s points . distinguished cases are those exceptional ones in which @xmath76 is hypersurface orthogonal . then the intersection of @xmath76 s flow lines with the orthogonal hypersurfaces consist of mutually _ einstein synchronous _ locations of the points of the body . an example is discussed below . equation ( [ eq : rigmo2 ] ) shows that the rigidity condition is equivalent to the ` spatially ' projected killing equation . we call the flow of the timelike normalised vector field @xmath76 a _ killing motion _ ( i.e. a spacetime isometry ) if there is a killing field @xmath280 such that @xmath683 . equation ( [ eq : rigmo2 ] ) immediately implies that killing motions are rigid . what about the converse ? are there rigid motions that are not killing ? this turns out to be a difficult question . its answer in minkowski space is : ` yes , many , but not as many as navely expected . ' before we explain this , let us give an illustrative example for a killing motion , namely that generated by the boost killing - field in minkowski space . we suppress all but one spatial directions and consider boosts in @xmath262 direction in two - dimensional minkowski space ( coordinates @xmath684 and @xmath262 ; metric @xmath685 ) . the killing field is denote the vector fields naturally defined by the coordinates @xmath686 . pointwise the dual basis to @xmath687 is @xmath688 . ] @xmath689 which is timelike in the region @xmath690 . we focus on the ` right wedge ' @xmath691 , which is now our region @xmath214 . consider a rod of length @xmath469 which at @xmath692 is represented by the interval @xmath693 , where @xmath694 . the flow of the normalised field @xmath683 is [ eq : boostkill2 ] @xmath695 where @xmath696 labels the elements of the rod at @xmath697 . we have @xmath698 , showing that the individual elements of the rod move on hyperbolae ( ` hyperbolic motion ' ) . @xmath699 is the proper time along each orbit , normalised so that the rod lies on the @xmath262 axis at @xmath697 . the combination @xmath700 is just the flow parameter for @xmath280 ( [ eq : boostkill1 ] ) , sometimes referred to as ` killing time ' ( though it is dimensionless ) . from ( [ eq : boostkill2 ] ) we can solve for @xmath155 and @xmath699 as functions of @xmath684 and @xmath262 : [ eq : lambdatau ] @xmath701 from which we infer that the hypersurfaces of constant @xmath155 are hyperplanes which all intersect at the origin . moreover , we also have @xmath702 ( @xmath293 is just the ordinary exterior differential ) so that the hyperplanes of constant @xmath155 intersect all orbits of @xmath76 ( and @xmath280 ) orthogonally . hence the hyperplanes of constant @xmath155 qualify as the equivalence classes of mutually einstein - simultaneous events in the region @xmath703 for a family of observers moving along the killing orbits . this does not hold for the hypersurfaces of constant @xmath699 , which are curved . the modulus of the spacetime - acceleration ( which is the same as the modulus of the spatial acceleration measured in the local rest frame ) of the material part of the rod labelled by @xmath704 is @xmath705 as an aside we generally infer from this that , given a timelike curve of local acceleration ( modulus ) @xmath239 , infinitesimally nearby orthogonal hyperplanes intersect at a spatial distance @xmath706 . this remark will become relevant in the discussion of part2 of the noether - herglotz theorem given below . in order to accelerate the rod to the uniform velocity @xmath60 without deforming it , its material point labelled by @xmath704 has to accelerate for the eigentime ( this follows from ( [ eq : boostkill2 ] ) ) @xmath707 which depends on @xmath704 . in contrast , the killing time is the same for all material points and just given by the final rapidity . in particular , judged from the local observers moving with the rod , a rigid acceleration requires accelerating the rod s trailing end harder but shorter than pulling its leading end . in terms of the coordinates @xmath708 , which are co - moving with the flow of @xmath280 , and @xmath709 , which are co - moving with the flow of @xmath76 , we just have @xmath710 and @xmath711 respectively . the spacetime metric @xmath30 and the projected metric @xmath216 in terms of these coordinates are : [ eq : metricscomovcoord ] @xmath712 note the simple form @xmath30 takes in terms of @xmath704 and @xmath155 , which are also called the ` rindler coordinates ' for the region @xmath713 of minkowski space . they are the analogs in lorentzian geometry to polar coordinates ( radius @xmath704 , angle @xmath155 ) in euclidean geometry . let us now return to the general case . we decompose the derivative of the velocity one - form @xmath714 as follows : @xmath715 where @xmath716 and @xmath717 are the projected symmetrised and antisymmetrised derivatives respectively ) by @xmath547 and @xmath546 respectively and the symmetrised and antisymmetrised ( covariant- ) derivative by @xmath718 and @xmath719 . for example , @xmath720 and @xmath721 . note that @xmath722 is the same as the ordinary exterior differential @xmath723 . everything we say in the sequel applies to curved spacetimes if @xmath724 is read as covariant derivative with respect to the levi - civita connection . ] [ eq : thetaomega ] @xmath725 the symmetric part , @xmath716 , is usually further decomposed into its traceless and pure trace part , called the _ shear _ and _ expansion _ of @xmath76 respectively . the antisymmetric part @xmath717 is called the _ vorticity _ of @xmath76 . now recall that the lie derivative of @xmath30 is just twice the symmetrised derivative , which in our notation reads : @xmath726 this implies in view of ( [ eq : def : rigmo ] ) , ( [ eq : rigmo2 ] ) , and ( [ eq : thetaomega1 ] ) let @xmath76 be a normalised timelike vector field @xmath76 . the motion described by its flow is rigid iff @xmath76 is of vanishing shear and expansion , i.e. iff @xmath727 . vector fields generating rigid motions are now classified according to whether or not they have a vanishing vorticity @xmath717 : if @xmath728 the flow is called _ irrotational _ , otherwise _ rotational_. the following theorem is due to herglotz @xcite and noether @xcite : [ thm : noetherherglotz1 ] a rotational rigid motion in minkowski space must be a killing motion . an example of such a rotational motion is given by the killing field we now use standard cylindrical coordinates @xmath729 , in terms of which @xmath730 . ] @xmath731 inside the region @xmath732 where @xmath280 is timelike . this motion corresponds to a rigid rotation with constant angular velocity @xmath733 which , without loss of generality , we take to be positive . using the co - moving angular coordinate @xmath734 , the split ( [ eq : hormetric ] ) is now furnished by [ eq : rotsplitmetric ] @xmath735 the metric @xmath216 is curved ( cf . lemma[lemma : noetherherglotz1 - 1 ] ) . but the rigidity condition ( [ eq : def : rigmo ] ) means that @xmath216 , and hence its curvature , can not change along the motion . therefore , even though we can keep a body in uniform rigid rotational motion , we can not put it into this state from rest by purely rigid motions , since this would imply a transition from a flat to a curved geometry of the body . this was first pointed out by ehrenfest @xcite . below we will give a concise analytical expression of this fact ( cf . equation ( [ eq : omegaliezero ] ) ) . all this is in contrast to the translational motion , as we will also see below . the proof of theorem[thm : noetherherglotz1 ] relies on arguments from differential geometry proper and is somewhat tricky . here we present the essential steps , basically following @xcite and @xcite in a slightly modernised notation . some straightforward calculational details will be skipped . the argument itself is best broken down into several lemmas . at the heart of the proof lies the following general construction : let @xmath1 be the spacetime manifold with metric @xmath30 and @xmath736 the open region in which the normalised vector field @xmath76 is defined . we take @xmath214 to be simply connected . the orbits of @xmath76 foliate @xmath214 and hence define an equivalence relation on @xmath214 given by @xmath184 iff @xmath62 and @xmath140 lie on the same orbit . the quotient space @xmath737 is itself a manifold . tensor fields on @xmath738 can be represented by ( i.e. are in bijective correspondence to ) tensor fields @xmath739 on @xmath214 which obey the two conditions : [ eq : projectibility ] @xmath740 tensor fields satisfying ( [ eq : projectibility1 ] ) are called _ horizontal _ , those satisfying both conditions ( [ eq : projectibility ] ) are called _ projectable_. the @xmath741-dimensional metric tensor @xmath216 , defined in ( [ eq : hormetric ] ) , is an example of a projectable tensor if @xmath76 generates a rigid motion , as assumed here . it turns @xmath742 into a @xmath741-dimensional riemannian manifold . the covariant derivative @xmath743 with respect to the levi - civita connection of @xmath216 is given by the following operation on projectable tensor fields : @xmath744 i.e. by first taking the covariant derivative @xmath724 ( levi - civita connection in @xmath745 ) in spacetime and then projecting the result horizontally . this results again in a projectable tensor , as a straightforward calculation shows . the horizontal projection of the spacetime curvature tensor can now be related to the curvature tensor of @xmath738 ( which is a projectable tensor field ) . without proof we state [ lemma : noetherherglotz1 - 1 ] let @xmath76 generate a rigid motion in spacetime . then the horizontal projection of the totally covariant ( i.e. all indices down ) curvature tensor @xmath421 of @xmath746 is related to the totally covariant curvature tensor @xmath747 of @xmath742 by the following equation appears with a minus sign on the right hand side of ( [ eq : projectedcurvaturerelation ] ) because the first index on the hatted curvature tensor is lowered with @xmath216 rather than @xmath30 . this induces a minus sign due to ( [ eq : hormetric ] ) , i.e. as a result of our ` mostly - minus'-convention for the signature of the spacetime metric . ] : @xmath748 where @xmath749 is the total antisymmetriser , which here projects tensors of rank four onto their totally antisymmetric part . formula ( [ eq : projectedcurvaturerelation ] ) is true in any spacetime dimension @xmath18 . note that the projector @xmath750 guarantees consistency with the first bianchi identities for @xmath421 and @xmath747 , which state that the total antisymmetrisation in their last three slots vanish identically . this is consistent with ( [ eq : projectedcurvaturerelation ] ) since for tensors of rank four with the symmetries of @xmath751 the total antisymmetrisation on tree slots is identical to @xmath749 , the symmetrisation on all four slots . the claim now simply follows from @xmath752 . we now restrict to spacetime dimensions of four or less , i.e. @xmath753 . in this case @xmath754 since @xmath672 makes the tensor effectively live over @xmath125 dimensions , and any totally antisymmetric four - tensor in three or less dimensions must vanish . applied to ( [ eq : projectedcurvaturerelation ] ) this means that @xmath755 , for horizontality of @xmath717 implies @xmath756 . hence the right hand side of ( [ eq : projectedcurvaturerelation ] ) just contains the pure tensor product @xmath757 . now , in our case @xmath758 since @xmath745 is flat minkowski space . this has two interesting consequences : first , @xmath742 is curved iff the motion is rotational , as exemplified above . second , since @xmath747 is projectable , its lie derivative with respect to @xmath76 vanishes . hence ( [ eq : projectedcurvaturerelation ] ) implies @xmath759 , which is equivalent to . ] @xmath760 this says that the vorticity can not change along a rigid motion in flat space . it is the precise expression for the remark above that you can not rigidly _ set _ a disk into rotation . note that it also provides the justification for the global classification of rigid motions into rotational and irrotational ones . a sharp and useful criterion for whether a rigid motion is killing or not is given by the following [ lemma : noetherherglotz1 - 2 ] let @xmath76 be a normalised timelike vector field on a region @xmath761 . the motion generated by @xmath76 is killing iff it is rigid and @xmath762 is exact on @xmath214 . that the motion generated by @xmath76 be killing is equivalent to the existence of a positive function @xmath763 such that @xmath764 , i.e. @xmath765 . in view of ( [ eq : thetaomega1 ] ) this is equivalent to @xmath766 which , in turn , is equivalent to @xmath727 and @xmath767 . this is true since @xmath716 is horizontal , @xmath768 , whereas the first term in ( [ eq : matteo1 ] ) vanishes upon applying @xmath672 . the result now follows from reading this equivalence both ways : 1 ) the killing condition for @xmath769 implies rigidity for @xmath76 and exactness of @xmath762 . 2 ) rigidity of @xmath76 and @xmath770 imply that @xmath769 is killing , where @xmath771 . we now return to the condition ( [ eq : omegaliezero ] ) and express @xmath772 in terms of @xmath723 . for this we recall that @xmath677 ( cf . footnote[foot : lievel ] ) and that lie derivatives on forms commute with exterior derivatives , where @xmath773 is the map of inserting @xmath76 in the first slot . ] . hence we have @xmath774 here we used the fact that the additional terms that result from the lie derivative of the projection tensor @xmath672 vanish , as a short calculation shows , and also that on forms the projection tensor @xmath672 can be written as @xmath775 , where @xmath773 denotes the map of insertion of @xmath76 in the first slot . now we prove [ lemma : noetherherglotz1 - 3 ] let @xmath76 generate a rigid motion in flat space such that @xmath776 , then @xmath777 equation ( [ eq : omegaliezero ] ) says that @xmath717 is projectable ( it is horizontal by definition ) . hence @xmath778 is projectable , which implies @xmath779 using ( [ eq : decveloneform ] ) with @xmath727 one has @xmath780 antisymmetrisation in the first two tensor slots makes the first term on the right vanish due to the flatness on @xmath724 . the antisymmetrised right hand side is hence equal to @xmath781 . taking the lie derivative of both sides makes the left hand side vanish due to ( [ eq : hatnablaomega1 ] ) , so that @xmath782 where we also used ( [ eq : omegaliezero ] ) . so we see that @xmath783 if @xmath776 . ) is generally not true if @xmath728 ; see equation ( [ eq : proof : noeterherglotz2i ] ) . ] the last three lemmas now constitute a proof for theorem[thm : noetherherglotz1 ] . indeed , using ( [ eq : lieacczero ] ) in ( [ eq : lieomega1 ] ) together with ( [ eq : omegaliezero ] ) shows @xmath784 , which , according to lemma[lemma : noetherherglotz1 - 2 ] , implies that the motion is killing . next we turn to the second part of the theorem of noether and herglotz , which reads as follows : [ thm : noetherherglotzp2 ] all irrotational rigid motions in minkowski space are given by the following construction : take a twice continuously differentiable curve @xmath785 in minkowski space , where w.l.o.g @xmath699 is the eigentime , so that @xmath786 . let @xmath787 be the hyperplane through @xmath788 intersecting the curve @xmath789 perpendicularly . let @xmath214 be a the tubular neighbourhood of @xmath789 in which no two hyperplanes @xmath790 intersect for any pair @xmath791 of points on the curve . in @xmath214 define @xmath76 as the unique ( once differentiable ) normalised timelike vector field perpendicular to all @xmath792 . the flow of @xmath76 is the sought - for rigid motion . we first show that the flow so defined is indeed rigid , even though this is more or less obvious from its very definition , since we just defined it by ` rigidly ' moving a hyperplane through spacetime . in any case , analytically we have , @xmath793 in @xmath214 any @xmath262 lies on exactly one such hyperplane , @xmath794 , which means that there is a function @xmath795 so that @xmath796 and hence @xmath797 . this implies @xmath798 . using the expression for @xmath9 from ( [ eq : proof : noeterherglotz2a ] ) this is equivalent to @xmath799\,,\ ] ] where ` @xmath800 ' denotes the ` identity vector - field ' , @xmath801 , in minkowski space . note that in @xmath214 we certainly have @xmath802 and hence @xmath803 . in @xmath214 we now define the normalised timelike vector field , @xmath804 . ] @xmath805 using ( [ eq : proof : noeterherglotz2b ] ) , its derivative is given by @xmath806/(n^2c^2)\,,\ ] ] where @xmath807 this immediately shows that @xmath808 ( since @xmath809 ) and therefore that @xmath810 . hence @xmath76 , as defined in ( [ eq : proof : noeterherglotz2c ] ) , generates an irrotational rigid motion . for the converse we need to prove that any irrotational rigid motion is obtained by such a construction . so suppose @xmath76 is a normalised timelike vector field such that @xmath810 . vanishing @xmath717 means @xmath811 . this is equivalent to @xmath812 , which according to the frobenius theorem in differential geometry is equivalent to the integrability of the distribution it denotes an assignment of a linear subspace @xmath813 in the tangent space @xmath814 to each point @xmath62 of @xmath1 . the distribution @xmath815 is defined by @xmath816 . a distribution is called ( locally ) integrable if ( in the neighbourhood of each point ) there is a submanifold @xmath817 of @xmath1 whose tangent space at any @xmath818 is just @xmath813 . ] @xmath815 , i.e. the hypersurface orthogonality of @xmath76 . we wish to show that the hypersurfaces orthogonal to @xmath76 are hyperplanes . to this end consider a spacelike curve @xmath819 , where @xmath397 is the proper length , running within one hypersurface perpendicular to @xmath76 . the component of its second @xmath397-derivative parallel to the hypersurface is given by ( to save notation we now simply write @xmath76 and @xmath820 instead of @xmath821 and @xmath822 ) @xmath823 where we made a partial differentiation in the second step and then used @xmath727 . geodesics in the hypersurface are curves whose second derivative with respect to proper length have vanishing components parallel to the hypersurface . now , ( [ eq : proof : noeterherglotz2f ] ) implies that geodesics in the hypersurface are geodesics in minkowski space ( the hypersurface is ` totally geodesic ' ) , i.e. given by straight lines . hence the hypersurfaces are hyperplanes . theorem[thm : noetherherglotzp2 ] precisely corresponds to the newtonian counterpart : the irrotational motion of a rigid body is determined by the worldline of any of its points , and any timelike worldline determines such a motion . we can rigidly put an extended body into any state of translational motion , as long as the size of the body is limited by @xmath706 , where @xmath239 is the modulus of its acceleration . this also shows that ( [ eq : lieacczero ] ) is generally not valid for irrotational rigid motions . in fact , the acceleration one - form field for ( [ eq : proof : noeterherglotz2c ] ) is @xmath824 from which one easily computes @xmath825 from this one sees , for example , that for _ constant acceleration _ , defined by @xmath826 ( constant acceleration in time as measured in the instantaneous rest frame ) , we have @xmath784 and hence a killing motion . clearly , this is just the motion ( [ eq : boostkill2 ] ) for the boost killing field ( [ eq : boostkill1 ] ) . the lie derivative of @xmath762 is now easily obtained : @xmath827 showing explicitly that it is not zero except for motions of constant acceleration , which were just seen to be killing motions . in contrast to the irrotational case just discussed , we have seen that we can not put a body rigidly into rotational motion . in the old days this was sometimes expressed by saying that the rigid body in sr has only three instead of six degrees of freedom . this was clearly thought to be paradoxical as long as one assumed that the notion of a perfectly rigid body should also make sense in the framework of sr . however , this hope was soon realized to be physically untenable @xcite . in this appendix we spell out in detail some of the mathematical notions that were used in the main text . given a set @xmath2 , recall that an _ equivalence relation _ is a subset @xmath508 such that for all @xmath828 the following conditions hold : 1 ) @xmath829 ( called ` reflexivity ' ) , 2 ) if @xmath830 then @xmath831 ( called ` symmetry ' ) , and 3 ) if @xmath830 and @xmath832 then @xmath833 ( called ` transitivity ' ) . once @xmath421 is given , one often conveniently writes @xmath184 instead of @xmath830 . given @xmath507 , its _ equivalence class _ , @xmath834\subseteq s$ ] , is given by all points @xmath421-related to @xmath62 , i.e. @xmath834:=\{q\in s\mid ( p , q)\in r\}$ ] . one easily shows that equivalence classes are either identical or disjoint . hence they form a _ partition _ of @xmath2 , that is , a covering by mutually disjoint subsets . conversely , given a partition of a set @xmath2 , it defines an equivalence relation by declaring two points as related iff they are members of the same cover set . hence there is a bijective correspondence between partitions of and equivalence relations on a set @xmath2 . the set of equivalence classes is denoted by @xmath835 or @xmath836 . there is a natural surjection @xmath837 , @xmath838 $ ] . if in the definition of equivalence relation we exchange symmetry for antisymmetry , i.e. @xmath830 and @xmath831 implies @xmath839 , the relation is called a _ partial order _ , usually written as @xmath131 for @xmath830 . if , instead , reflexivity is dropped and symmetry is replaced by asymmetry , i.e. @xmath830 implies @xmath840 , one obtains a relation called a _ strict partial order _ , usually denoted by @xmath841 for @xmath830 . an _ left action _ of a group @xmath249 on a set @xmath2 is a map @xmath842 , such that @xmath843 ( @xmath844 group identity ) and @xmath845 . if instead of the latter equation we have @xmath846 one speaks of a _ right action_. for left actions one sometimes conveniently writes @xmath847 , for right actions @xmath848 . an action is called _ transitive _ if for every pair @xmath849 there is a @xmath850 such that @xmath851 , and _ simply transitive _ if , in addition , @xmath852 determine @xmath30 uniquely , that is , @xmath853 for some @xmath397 implies @xmath854 . the action is called _ effective _ if @xmath855 for all @xmath397 implies @xmath856 ( ` every @xmath857 moves something ' ) and _ free _ if @xmath855 for some @xmath397 implies @xmath856 ( ` no @xmath857 has a fixed point ' ) . it is obvious that simple transitivity implies freeness and that , conversely , freeness and transitivity implies simple transitivity . moreover , for abelian groups , effectivity and transitivity suffice to imply simple transitivity . indeed , suppose @xmath858 holds for some @xmath859 , then we also have @xmath860 for all @xmath861 and hence @xmath862 by commutativity . this implies that @xmath858 holds , in fact , for all @xmath397 . for any @xmath859 we can consider the _ stabilizer subgroup _ @xmath863 if @xmath112 is transitive , any two stabilizer subgroups are conjugate : @xmath864 . by definition , if @xmath112 is free all stabilizer subgroups are trivial ( consist of the identity element only ) . in general , the intersection @xmath865 is the normal subgroup of elements acting trivially on @xmath2 . if @xmath112 is an action of @xmath249 on @xmath2 , then there is an effective action @xmath866 of @xmath867 on @xmath2 , defined by @xmath868,s):=\phi(g , s)$ ] , where @xmath869 $ ] denotes the @xmath870-coset of @xmath870 in @xmath249 . the _ orbit _ of @xmath397 in @xmath2 under the action @xmath112 of @xmath249 is the subset @xmath871 it is easy to see that group orbits are either disjoint or identical . hence they define a partition of @xmath2 , that is , an equivalence relation . a relation @xmath421 on @xmath2 is said to be invariant under the self map @xmath3 if @xmath872 . it is said to be invariant under the action @xmath112 of @xmath249 on @xmath2 if @xmath873 for all @xmath850 . if @xmath421 is such a @xmath249-invariant equivalence relation , there is an action @xmath874 of @xmath249 on the set @xmath835 of equivalence classes , defined by @xmath875):=[\phi(g , p)]$ ] . a general theorem states that invariant equivalence relations exist for transitive group actions , iff the stabilizer subgroups ( which in the transitive case are all conjugate ) are maximal ( e.g. theorem1.12 in @xcite ) . [ def : affinespace ] an @xmath18-dimensional * affine space * over the field @xmath876 ( usually @xmath296 or @xmath877 ) is a triple @xmath878 , where @xmath2 is a non - empty set , @xmath19 an @xmath18-dimensional vector space over @xmath876 , and @xmath879 an effective and transitive action @xmath880 of @xmath19 ( considered as abelian group with respect to addition of vectors ) on @xmath2 . we remark that an effective and transitive action of an abelian group is necessarily simply transitive . hence , without loss of generality , we could have required a simply transitive action in definition[def : affinespace ] straightaway . we also note that even though the action @xmath879 only refers to the abelian group structure of @xmath19 , it is nevertheless important for the definition of an affine space that @xmath19 is , in fact , a vector space ( see below ) . any ordered pair of points @xmath881 uniquely defines a vector @xmath60 , namely that for which @xmath882 . it can be thought of as the difference vector pointing from @xmath140 to @xmath62 . we write @xmath883 , where @xmath884 is a map which satisfies the conditions [ thm : altdefaffspace ] let @xmath2 be a non - empty set , @xmath19 an @xmath18-dimensional vector space over @xmath876 and @xmath884 a map satisfying conditions ( [ eq : affdiffmap ] ) . then @xmath2 is an @xmath18-dimensional affine space over @xmath876 with action @xmath886 . one usually writes @xmath887 , which defines what is meant by ` @xmath80 ' between an element of an affine space and an element of @xmath19 . note that addition of two points in affine space is not defined . the property of being an action now states @xmath888 and @xmath889 , so that in the latter case we may just write @xmath890 . similarly we write @xmath891 , defining what is meant by ` @xmath81 ' between two elements of affine space . the minus sign also makes sense between an element of affine space and an element of vector space if one defines @xmath892 . we may now write equations like @xmath893 the formal proof of which is again left to the reader . it implies that considered as abelian group , any linear subspace @xmath56 defines a subgroup . the orbit of that subgroup in @xmath2 through @xmath507 is an affine subspace , denoted by @xmath68 , i.e. @xmath894 which is an affine space over @xmath67 in its own right of dimension @xmath895 . one - dimensional affine subspaces are called _ ( straight ) lines _ , two - dimensional ones _ planes _ , and those of co - dimension one are called _ affine morphisms , or simply affine maps , are structure preserving maps between affine spaces . to define them in view of definition[def : affinespace ] we recall once more the significance of @xmath19 being a vector space and not just an abelian group . this enters the following definition in an essential way , since there are considerably more automorphisms of @xmath19 as abelian group , i.e. maps @xmath28 that satisfy @xmath896 for all @xmath32 , than automorphisms of @xmath19 as linear space which , in addition , need to satisfy @xmath897 for all @xmath45 and all @xmath898 ) . in fact , the difference is precisely that the latter are all continuous automorphisms of @xmath19 ( considered as topological abelian group ) , whereas there are plenty ( uncountably many ) discontinuous ones , see @xcite . , then it is easy to see that @xmath896 for all @xmath32 implies @xmath897 for all @xmath45 and all @xmath899 ( rational numbers ) . for continuous @xmath9 this implies the same for all @xmath900 . all discontinuous @xmath9 are obtained as follows : let @xmath901 be a ( necessarily uncountable ) basis of @xmath296 as vector space over @xmath902 ( ` hamel basis ' ) , prescribe any values @xmath903 , and extend @xmath9 linearly to all of @xmath296 . any value - prescription for which @xmath904 is not constant gives rise to a non @xmath296-linear and discontinuous @xmath9 . such @xmath9 are ` wildly ' discontinuous in the following sense : for _ any _ interval @xmath905 , @xmath906 is dense @xcite . ] in the convenient way of writing introduced above , this is equivalent to @xmath911 for all @xmath912 and all @xmath45 . ( note that the @xmath80 sign on the left refers to the action @xmath879 of @xmath19 on @xmath2 , whereas that on the right refers to the action @xmath913 of @xmath914 on @xmath526 . ) this shows that an affine map @xmath159 is determined once the linear map @xmath9 between the underlying vector spaces is given and the image @xmath497 of an arbitrary point @xmath140 is specified . equation ( [ eq : affinemap2 ] ) can be rephrased as follows : setting @xmath915 equation ( [ eq : affinemap2 ] ) is equivalent to @xmath916 for all @xmath917 . in view of the alternative definition of affine spaces suggested by proposition[thm : altdefaffspace ] , this shows that we could have defined affine maps alternatively to ( [ eq : affinemap1 ] ) by ( @xmath918 is the difference map in @xmath526 ) @xmath919 affine bijections of an affine space @xmath878 onto itself form a group , the _ affine group _ , denoted by @xmath920 . group multiplication is just given by composition of maps , that is @xmath921 . it is immediate that the composed maps again satisfy ( [ eq : affinemap1 ] ) . for any @xmath45 , the map @xmath922 is an affine bijection for which @xmath923 . note that in this case ( [ eq : affinemap1 ] ) simply turns into the requirement @xmath924 for all @xmath106 , which is clearly satisfied due to @xmath19 being a commutative group . hence there is a natural embedding @xmath925 , the image @xmath926of which is called the subgroup of _ translations_. the map @xmath927 defines a group homomorphism @xmath928 , since @xmath929 . we have just seen that the translations are in the kernel of this map . in fact , the kernel is equal to the subgroup @xmath926 of translations , as one easily infers from ( [ eq : affinemap3 ] ) with @xmath923 , which is equivalent to @xmath930 for all @xmath917 . hence there exists a @xmath45 such that for all @xmath507 we have @xmath931 . the quotient group @xmath932 is then clearly isomorphic to @xmath933 . there are also embeddings @xmath934 , but no canonical one : each one depends on the choice of a reference point @xmath935 , and is given by @xmath936 , where @xmath937 for all @xmath507 . this shows that @xmath920 is isomorphic to the semi - direct product @xmath938 , though the isomorphism depends on the choice of @xmath935 . the action of @xmath939 on @xmath507 is then defined by @xmath940 which is easily checked to define indeed an ( @xmath108 dependent ) action of @xmath938 on @xmath2 . [ def : linearframe ] a * linear frame * of the @xmath18-dimensional vector space @xmath19 over @xmath876 is a basis @xmath941 of @xmath19 , regarded as a linear isomorphism @xmath942 , given by @xmath943 . the set of linear frames of @xmath19 is denoted by @xmath944 . there is a natural right action of @xmath946 on @xmath944 , given by @xmath947 . it is immediate that this action is simply transitive . it is sometimes called the _ passive interpretation _ of the transformation group @xmath946 , presumably because it moves the frames associated to the observer and not the points of @xmath19 . on the other hand , any frame @xmath9 induces an isomorphism of algebras @xmath948 , given by @xmath949 . if @xmath950 , then @xmath951 , where @xmath941 . restricted to @xmath952 , this induces a group isomorphism @xmath953 and hence an @xmath9-dependent action of @xmath946 on @xmath19 by linear transformations , defined by @xmath954 , where @xmath955 . this is sometimes called the _ active interpretation _ of the transformation group @xmath946 , presumably because it really moves the points of @xmath19 . [ def : affineframe ] an * affine frame * of the @xmath18-dimensional affine space @xmath878 over @xmath876 is a tuple @xmath956 , where @xmath108 is a base point in @xmath2 and @xmath942 is a linear frame of @xmath19 . @xmath159 is regarded as a map @xmath957 , given by @xmath958 . we denote the set of affine frames by @xmath959 . if we regard @xmath945 as an affine space @xmath961 , it comes with a distinguished base point @xmath108 , the zero vector . the group @xmath962 is therefore naturally isomorphic to @xmath963 . the latter naturally acts on @xmath945 in the standard way , @xmath964 , where group multiplication is given by @xmath965 the group @xmath963 has a natural right action on @xmath959 , where @xmath966 . explicitly , for @xmath967 and @xmath968 , this action reads : @xmath969 it is easy to verify directly that this is an action which , moreover , is again simply transitive . it is referred to as the _ passive interpretation _ of the affine group @xmath963 . conversely , depending on the choice of an affine frame @xmath970 , there is a group isomorphism @xmath971 , given by @xmath972 , and hence an @xmath159 dependent action of @xmath963 by affine maps on @xmath878 . if @xmath968 and @xmath973 , the action reads @xmath974 this is called the _ active interpretation _ of the affine group @xmath963 . an affine frame @xmath975 with @xmath941 defines @xmath976 points @xmath977 , where @xmath978 and @xmath979 for @xmath980 . conversely , any @xmath976 points @xmath977 in affine space , for which @xmath981 are linearly independent , define an affine frame . note that this linear independence does not depend on the choice of @xmath982 as our base point , as one easily sees from the identity @xmath983 which holds for any set @xmath984 of @xmath985 points in affine space . to prove it one just needs ( [ eq : affdistributivity ] ) . hence we say that these points are _ affinely independent _ iff , e.g. , the set of @xmath0 vectors @xmath986 is linearly independent . therefore , an affine frame of @xmath18-dimensional affine space is equivalent to @xmath976 affinely independent points . such a set of points is also called an _ affine basis_. this motivates writing the sums on the right hand sides of ( [ eq : affbasisexp ] ) in a perfectly symmetric way without preference of any point @xmath995 : @xmath996 where the right hand side is defined by any of the expressions ( [ eq : affbasisexp ] ) . this defines certain _ linear combinations _ of affine points , namely those whose coefficients add up to one . accordingly , the _ affine span _ of points @xmath997 in affine space is defined by @xmath998 harvey brown and oliver pooley . minkowski space - time : a glorious non - entity . in dennis dieks , editor , _ the ontology of spacetime _ , volume 1 of _ philosophy and foundations of physics _ , pages 6792 , 2006 . andrzej trautman . foundations and current problems of general relativity . in andrzej trautman , felix pirani , and hermann bondi , editors , _ lectures on general relativity _ , volume 1 of _ brandeis summer institute in theoretical physics _ , pages 1248 . prentice - hall , inc . , englewood cliffs , new jersey , 1964 . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: any conceivable theory of structure formation in the universe predicts statistical properties of observable quantities . therefore , the analysis of the present inhomogeneity of the universe is inevitably statistical . however , what kind of statistics is useful is not obvious , since we can invent infinite number of statistics to be analyzed . whether we can adopt better statistical descriptions is of great importance in this sense . each statistic has both advantages and disadvantages . the power spectrum , for example , can fully characterize the random gaussian fields , while it does not contain any information on the non - gaussianity , which contains significant information in the gravitational instability theory . among various statistical quantities , there is a promising class of statistics which utilizes smoothed cosmic fields . the smoothed field has less noisy property than the actual ( unsmoothed , or raw ) cosmic fields , such as galaxy distributions , temperature fluctuations of cosmic microwave background ( cmb ) , shear fields of the gravitational lensing , and so forth . perhaps the simplest example of such statistics is the variance @xmath0 of smoothed density contrast @xmath1 , which is a function of smoothing length @xmath2 . similarly , higher - order cumulants @xmath3 ( @xmath4 ) are also simple statistics . the density probability distribution function ( pdf ) @xmath5 , which in principle can be constructed from the hierarchy of cumulants ( e.g. , * ? ? ? * ) , is an another example of popular statistics of smoothed cosmic fields . rather recently , more complex statistics of smoothed cosmic fields have become popular in cosmology , such as the genus statistic @xcite , density peak statistics @xcite , area , length , level - crossing statistics @xcite , minkowski functionals @xcite , etc . these statistics provide assuring characterizations of the clustering pattern that can not be perceived only by the hierarchy of cumulants or by the pdf . the genus statistic is a powerful measure of the morphology in the 3-dimensional @xcite , and 2-dimensional @xcite clustering of galaxies and clusters , and also of the pattern of temperature fluctuations of cmb radiation @xcite . the peak statistics of the 3-dimensional density field are frequently used in connection with the statistics of the collapsing object @xcite , while the peak statistics of the cmb fluctuations @xcite and of the weak lensing fields @xcite are suitable for constraining cosmological models . the area , length , and level - crossing statistics directly quantify the amount of contour surfaces @xcite . the minkowski functionals @xcite , which are closely related to the above statistics , are also applied to smoothed cosmic fields @xcite . these statistics of smoothed density fields are considered as powerful descriptors of the statistical information of the universe . it is therefore essential to establish theoretical predictions of the behavior of such statistics so that we can effectively and ideally analyze the data of our universe . recent developments of the perturbation theory ( for review , see * ? ? ? * ) in calculating the variance , the cumulants , and the pdf are remarkable . the perturbation theory becomes more and more useful because the recent developments of observations enable us to have widely covered sample volume of the universe , which can minimize undesirable strongly nonlinear effects which we do not understand well . the direct comparison of the theoretical predictions of the perturbation theory with the actual data is a promising field of research in this sense . in the case of top - hat smoothing function , @xcite and @xcite used the tree - level ( i.e. , lowest - order ) perturbation theory to obtain the third- and forth - order cumulants , i.e. , the skewness and the kurtosis . the same quantities with gaussian smoothing are calculated by @xcite , @xcite and @xcite . quite cleverly , @xcite took advantage of special properties of the top - hat smoothing function , and succeeded to obtain full hierarchy of higher order moments in the tree - level perturbation theory . he also obtained the pdf from this hierarchy of cumulants , which remarkably describes the nonlinear behavior of the gravitational instability in numerical simulations . beyond the tree - level calculation , the perturbation theory with loop - corrections has been also developed @xcite . because of the simplicity of the statistics , calculating the variance , the cumulants and the pdf were primarily the playground of perturbation - theorists . the evaluation of other statistics by the perturbation theory is less trivial . a quite useful technique is the edgeworth expansion , which was first applied to cosmology in calculating the pdf from seeds perturbations by @xcite , and from non - linear perturbation theory by @xcite , and @xcite . the analytic expression of the genus statistics in the perturbation theory is derived by @xcite , whose technique corresponds to multivariate version of the edgeworth expansion . in some literatures the edgeworth expansion is used in connecting the statistics and dynamics of the universe @xcite . the purpose of this paper is to give a comprehensive description of the formalism by which the perturbative evaluation is possible for wide range of non - trivial statistics of smoothed cosmic fields in general . since the number of spatial dimensions of our universe is three , the statistics of large - scale cosmic fields are defined in either one- , two- , or three - dimensional space . for example , the space of the density field in a redshift map of galaxies or quasars is three - dimensional . a projected galaxy map on the sky , a shear field of gravitational lensing , and temperature fluctuations of cmb on the sky are fields in two - dimensional space . the absorption lines of quasar spectra and pencil - beam surveys of galaxies define fields in one - dimensional space . thus it is useful to develop our statistical method in @xmath6-dimensions for generality . as illustrative examples of applications of our method , we calculate level - crossing statistic ( or equivalently length and area statistics ) , genus statistics , density - extrema statistics and minkowski functionals . as cosmic fields , we consider density and velocity fields in three dimensions , the projected density field of galaxies in two dimensions . the basic formalism and results of the second order theory are presented in this paper . we will give results of the third order theory in a future paper , which are technically more involved . the primary purpose of the present paper is to fully describe the basic formalism for the statistics of smoothed cosmic fields in perturbation theory . applications of the second order theory to popular statistics and cosmic fields are systematically presented . thus , this paper is in some way a mixture of the new results and comprehensive review of the old results . the major new results in this paper are : * explicit formulas of the lowest non - gaussian correction in various smoothed cosmic fields in @xmath6-dimensions , including all the minkowski functionals . they are presented as functions of both density threshold and rescaled threshold by volume - fraction . * relations among several statistics of the smoothed field are clarified . * derivatives of skewness parameter of the velocity field and projected 2d field are derived in second order perturbation theory of gravitational instability theory . especially , skewness parameters with gaussian smoothing are detailed . these quantities are particularly important to perturbatively evaluate the statistics of smoothed field . * genus curve against the scaled threshold in perturbation theory is discussed and a theory - motivated new asymmetry parameter in genus curve is introduced . this paper is structured in the following way . [ sec2 ] and [ sec3 ] are carrying out a mathematical exercise of expressing high - order terms in terms of skewness parameters ; the physics of gravity only enters once one actually calculates the skewness parameters in [ sec4 ] . thus [ sec2 ] and [ sec3 ] are a derivation of an extension of the edgeworth expansion , and the basic results are equations ( [ eq1 - 13 ] ) and ( [ eq1 - 17 ] ) . in [ sec3 ] , popular statistics of smoothed fields are examined . the second - order expressions in terms of the skewness parameters are given for pdf , level - crossing statistics , 2-dimensional and 3-dimensional genus statistics , 2-dimensional weighted extrema , and minkowski functionals in [ sec3 - 1][sec3 - 6 ] . these quantities are re - expressed as functions of the volume - fraction threshold in [ sec3 - 7 ] . the skewness parameters for several cosmic fields are calculated in [ sec4 ] , applying the second order perturbation theory . detailed calculations of the simple hierarchical model , the 3-dimensional density field , the velocity field , the 2-dimensional projected density field , and the weak lensing field are given in [ sec4 - 1][sec4 - 5 ] . the effect of biasing on the skewness parameters are discussed in [ sec4 - 6 ] . we discuss implications of our second - order results in we introduce a theory - motivated asymmetry factor to characterize the shift of the genus curve in this section . the conclusions are given in useful gaussian integrals including hermite polynomials are given in appendix [ app1 ] . the bispectrum in a two - dimensional projected field is reviewed in appendix [ app2 ] . the symbols in this paper are summarized in appendix [ app3 ] . we consider a cosmic random field @xmath7 which represents any field constructed from observable quantities of the universe , such as three - dimensional density field , velocity field , or two - dimensional projected density field , shear or convergence field of weak lensing , temperature fluctuations of cmb , etc . the coordinates @xmath8 can be either three- , two- , or one - dimensions . we assume the field @xmath9 is already smoothed by a smoothing function @xmath10 with smoothing length @xmath2 which cuts the high frequency fluctuations which suffer strongly nonlinear effects : @xmath11 where @xmath12 is the dimension of the space @xmath8 , and @xmath13 is a raw , unsmoothed field . two of the most popular 3d smoothing functions are the tophat smoothing function @xmath14 and the gaussian smoothing function @xmath15 we also assume that the mean value of the field @xmath9 is zero , and that the existence the variance @xmath16 : @xmath17 it is convenient to introduce a normalized field @xmath18 which has a unit variance as follows : @xmath19 the statistics of a smoothed cosmic field we are interested in are the functions of the field @xmath18 and its spatial derivatives as we will see in the following sections . we denote the series of spatial derivatives by a set of variables @xmath20 which is defined by , for example , in 3d case , @xmath21 where @xmath22 . for convenience , the index is denoted as @xmath23 , 1 , 2 , 3 , ( 11 ) , ( 22 ) , ( 33 ) , ( 12 ) , ( 13 ) , ( 23 ) , @xmath24 in such cases . in this example of equation ( [ eq1 - 4 ] ) , only the value of the field and of the derivatives on a single point is considered , but in general , more than two points can be considered . the set @xmath25 forms multivariate random fields , which is denoted as an @xmath26-dimensional vector @xmath27 in the following . the dimension @xmath26 is the total number of derivatives which appear in the definition of the statistics we are interested in . the statistical information is described by the multivariate pdf , @xmath28 . the fourier transform of the pdf is the partition function : @xmath29 at this point , the cumulant expansion theorem ( e.g. , * ? ? ? * ) is very useful . this theorem states that @xmath30 is the generating function of the cumulants , @xmath31 : @xmath32 it follows from @xmath33 that @xmath34 , and first several cumulants are given by @xmath35 and so forth . from equations ( [ eq1 - 6 ] ) and ( [ eq1 - 7a ] ) , the partition function is given by @xmath36 where @xmath37 is an @xmath38 matrix whose components are given by @xmath39 . on the other hand , the equation ( [ eq1 - 5 ] ) is inverted as @xmath40 substituting @xmath41 in the last term of equation ( [ eq1 - 8 ] ) , the distribution function of equation ( [ eq1 - 9 ] ) can be transformed in a form @xmath42 where @xmath43 is the multivariate gaussian distribution function characterized by the correlation matrix @xmath37 . any statistical quantity of a smoothed cosmic field is expressed by an average @xmath44 of a certain function @xmath45 as we will see in the following sections . thus , from equation ( [ eq1 - 10 ] ) , we obtain @xmath46 where @xmath47 denotes the averaging by the gaussian distribution function of equation ( [ eq1 - 11b ] ) . this form , equation ( [ eq1 - 12b ] ) , is useful when the deviation from the gaussian distribution is not large . in principle , this equation reduces the general statistical averaging procedure to gaussian integrations . however , it contains the infinite series , thus we have to truncate this expression by some criteria . in most cases of interest , the weakly nonlinear evolution of cosmic fields satisfy @xmath48 , as we will see in [ sec4 ] . when this relation holds , we can expand the distribution function to arbitrary order in @xmath49 . in the following , we assume this relation , and introduce the normalized cumulants : @xmath50 which are assumed to be of order one in terms of @xmath49 . in this case , the equation ( [ eq1 - 12b ] ) is expanded as , up to @xmath51 , @xmath52 { \sigma_0}^2 + { \cal o}({\sigma_0}^3 ) , \label{eq1 - 13}\end{aligned}\ ] ] where we introduce the notation , @xmath53 etc . the calculation of the factors @xmath54 is performed for individual statistic which gives the explicit form of the function @xmath55 . in the expansion of equation ( [ eq1 - 13 ] ) , we need to calculate the gaussian average of derivatives of the function @xmath55 : @xmath56 in most cases , the evaluation is analytically feasible because only gaussian integration is needed . for the evaluation of such integration , we need the correlation matrix @xmath37 . throughout this paper , we consider statistically homogeneous , and isotropic fields , in which case , the correlation matrix @xmath37 is simplified because of the symmetry . in fact , the correlation matrix takes the following forms : @xmath57 where @xmath58 etc . and we define the following quantities : @xmath59 in most cases of interest , the derivatives of order higher than two do not appear in the definition of the statistics , so we do not give explicit form of those correlations . in the following , we use the notation @xmath60 and @xmath61 following @xcite . as is often the case , when the second order derivatives @xmath62 appears in @xmath55 as simple polynomials , the following transform is particularly useful : @xmath63 this transform erase the correlation between @xmath18 and @xmath64 , and the non - zero correlations are only @xmath65 where @xmath66 the gaussian integration is straightforward if the function @xmath55 is expressed by polynomials of @xmath64 . if the function @xmath55 is more complicated in which @xmath62 are not simply given by polynomials , it is useful to completely diagonalize the correlation matrix @xmath37 of equations ( [ eq1 - 13 - 2a])([eq1 - 13 - 2a ] ) . we introduce the following transform , which is quite similar one in @xcite : @xmath67 if @xmath68 , we ignore the variables @xmath69 , and similarly , if @xmath70 , we ignore the variable @xmath71 in the above equations and in the following . the above transform is similar to @xcite but notice it is not identical . this transform completely diagonalize the correlation matrix : @xmath72 and all non - diagonal correlations are zero . for later convenience , we write the inverse relation of the transform of @xmath73 , @xmath74 , and @xmath71 : @xmath75 if @xmath76 does not appear in the function @xmath55 , the variable @xmath71 should be omitted in the above equations . if both @xmath77 and @xmath76 do not appear in the function @xmath55 , the variables @xmath74 and @xmath71 should be omitted . after expressing the function @xmath55 in terms of the diagonalized variables @xmath18 , @xmath78 , @xmath73 , @xmath74 , @xmath71 and @xmath79 , the calculation of the gaussian integration of equation ( [ eq1 - 13 - 1 ] ) is performed . in this paper , only first two terms in equation ( [ eq1 - 13 ] ) are considered . thus we evaluate @xmath80 here . when the spatial symmetry is taken into account , the quantity @xmath80 reduces to the following expressions : @xmath81 , \label{eq1 - 14e}\\ & & \langle\alpha_{,i}\alpha_{,j}\alpha_{,k}\rangle = 0 , \label{eq1 - 14f}\\ & & \langle\alpha_{,i}\alpha_{,j}\alpha_{,kl}\rangle = \frac{\langle(\nabla\alpha\cdot\nabla\alpha)\nabla^2\alpha\rangle } { d(d-1 ) } \left [ \delta_{ij}\delta_{kl } - \frac12 \left(\delta_{ik}\delta_{jl } + \delta_{il}\delta_{jk}\right ) \right ] , \label{eq1 - 14g}\\ & & \langle\alpha_{,i}\alpha_{,jk}\alpha_{,lm}\rangle = 0 , \label{eq1 - 14h}\end{aligned}\ ] ] etc . to prove the above equations , the following identities for isotropic fields are useful : @xmath82 although the above equations are valid for @xmath83 , the case @xmath68 is obtained by just ignoring the terms with @xmath84 . generally , more complicated quantities can be appeared for @xmath80 , but the above relations are sufficient for our applications in this paper . according to the spatial symmetry of the above equations , the third - order correlations @xmath85 are explicitly given . at this point , it is useful to define the following quantities : @xmath86 we call quantities @xmath87 skewness parameters . the first one @xmath88 is usually called as skewness , and the others are its derivatives . they are to be calculated from usual perturbation theories in [ sec4 ] . using these quantities , the third - order correlations are given by @xmath89 , \qquad ( i \ne j ) , \label{eq1 - 16i}\\ & & { \widehat{m}^{(3)}}_{0(ii)(jk ) } = 0 , \qquad ( j < k ) , \label{eq1 - 16j}\\ & & { \widehat{m}^{(3)}}_{0(ij)(ij ) } = - \frac{1}{d(d+2)}\frac{{\sigma_1}^4}{{\sigma_0}^4 } \left [ s^{(2 ) } - s^{(2)}_2 \right ] \qquad ( i < j ) , \label{eq1 - 16k}\\ & & { \widehat{m}^{(3)}}_{0(ij)(kl ) } = 0 , \qquad ( i < j , k < l , i \ne k ) , \label{eq1 - 16l}\\ & & { \widehat{m}^{(3)}}_{ijk } = 0 , \label{eq1 - 16m}\\ & & { \widehat{m}^{(3)}}_{ii(ii ) } = 0 , \label{eq1 - 16n}\\ & & { \widehat{m}^{(3)}}_{ii(jj ) } = - \frac{2}{3 d^2}\frac{{\sigma_1}^4}{{\sigma_0}^4 } s^{(2 ) } , \qquad ( i \ne j ) , \label{eq1 - 16o}\\ & & { \widehat{m}^{(3)}}_{ii(jk ) } = 0 , \qquad ( j < k ) , \label{eq1 - 16p}\\ & & { \widehat{m}^{(3)}}_{ij(kk ) } = 0 , \qquad ( i \ne j ) , \label{eq1 - 16q}\\ & & { \widehat{m}^{(3)}}_{ij(ij ) } = \frac{1}{3 d^2}\frac{{\sigma_1}^4}{{\sigma_0}^4 } s^{(2 ) } , \qquad ( i < j ) , \label{eq1 - 16r}\\ & & { \widehat{m}^{(3)}}_{ij(kl ) } = 0 , \qquad ( i < j , k < l , i \ne k ) , \label{eq1 - 16s}\\ & & { \widehat{m}^{(3)}}_{i(jk)(lm ) } = 0 , \label{eq1 - 16t}\end{aligned}\ ] ] and so forth , where repeated indices are _ not _ summed over in the above equations , and @xmath90 is symmetric under permutation of its indices . thus , denoting @xmath91 for simplicity , @xmath92 \nonumber\\ & & \quad\qquad + \ , \frac{3 s^{(2)}_2}{d(d+2 ) } \frac{{\sigma_1}^4}{{\sigma_0}^4 } \left ( 3 \sum_{i } f_{0(ii)(ii ) } + 2 \sum_{i < j } f_{0(ii)(jj ) } + \sum_{i < j } f_{0(ij)(ij ) } \right ) + \cdots . \label{eq1 - 17}\end{aligned}\ ] ] this equation gives the second - order correction term of the statistical quantity @xmath44 through equation ( [ eq1 - 13 ] ) . once the field @xmath9 is specified , the skewness parameters @xmath87 are calculated by dynamical perturbation theory of the field @xmath9 . the remaining factors in the above equation are the gaussian integrations of the derivatives of the function @xmath55 , i.e. , @xmath93 . these factors can be calculated once the function @xmath55 is given . in the next section , we calculate the latter factors for individual statistics . in this section we calculate the factor @xmath94 for each statistic . some of the results in this section have been previously presented . the edgeworth expansion of the pdf in [ sec3 - 1 ] is a familiar result . the result of 3d genus statistic in [ sec3 - 4 ] was already given by @xcite . we include these old results for completeness . other subsections present new results . perhaps the simplest yet non - trivial statistic is the pdf , @xmath95 . the perturbative expansion of the pdf is known as the edgeworth expansion @xcite . as the simplest example , we re - derive the known edgeworth expansion from the point of view of our general formalism above ( see also * ? ? ? * ) . since the pdf is simply given by @xmath96 , where @xmath97 is the dirac s delta function , the function @xmath55 in the previous section for pdf @xmath95 is given by @xmath98 since this form of @xmath55 does not depend on derivatives of @xmath18 , only @xmath99 survives in the equation ( [ eq1 - 17 ] ) . from equation ( [ eqa1 - 4a ] ) with @xmath100 and @xmath101 , @xmath102 , where @xmath103 . thus , the pdf is derived from the equation ( [ eq1 - 13 ] ) : @xmath104 , \label{eq2 - 0 - 2}\end{aligned}\ ] ] which reproduces the well - known result . there is less advantage of applying our formalism to this simple statistic which can be treated by standard methods . our formalism has advantages when more non - trivial statistics are considered as shown below . next three statistics we consider here are the level - crossing statistic @xmath105 , the length statistic @xmath106 and the area statistic @xmath107 . the level - crossing statistic is defined by the mean number of intersection of a straight line and threshold contours of the field . the length statistic is defined by the mean length of intersection of a 2d surface and the threshold contours of the field . the area statistic is defined by the mean area of the contour surface in a 3d space @xcite . the level - crossing statistic is defined for 1d , 2d , and 3d cosmic fields , the length statistic is defined for 2d and 3d cosmic fields , while the area statistic is defined only for 3d cosmic fields . for statistically isotropic fields , those three statistics are proportional to each other . in general , a statistic of the smoothed field @xmath9 is a function of the threshold @xmath108 , or of the normalized threshold @xmath109 . the explicit expressions of statistics @xmath105 , @xmath106 , and @xmath107 are given by @xcite @xmath110^{1/2 } \right\rangle , \label{eq2 - 1b}\\ & & n_3(\nu ) = \left\langle \delta(\alpha - \nu ) \left [ ( \eta_1)^2 + ( \eta_2)^2 + ( \eta_3)^2 \right]^{1/2 } \right\rangle . \label{eq2 - 1c}\end{aligned}\ ] ] for isotropic fields , these statistics are actually equivalent @xcite . in fact , the distribution function of @xmath111 for fixed @xmath112 is the function of only the magnitude @xmath113 . thus , using spherical coordinates for @xmath114 , one can see @xmath115 i.e. , @xmath116 thus we only need to consider @xmath105 which has the simplest expression and the rests of the statistics are automatically given by equation ( [ eq2 - 2 ] ) . however , if the field is anisotropic , such as the density field in redshift space @xcite , the equation ( [ eq2 - 2 ] ) no longer holds , and equations ( [ eq2 - 1a])([eq2 - 1c ] ) should be used for each statistic . the equation ( [ eq1 - 17 ] ) only holds for statistically isotropic fields , but the equation ( [ eq1 - 13 ] ) is applicable even for anisotropic fields . now we calculate the factor @xmath117 for the particular statistic @xmath105 . the indices @xmath118 only take @xmath119 and @xmath120 for @xmath105 statistic . let the number of @xmath119 be @xmath121 and the number of @xmath120 be @xmath122 . then the factor is given by @xmath123 since the variables @xmath18 and @xmath78 are uncorrelated in the gaussian averaging , we can use equations ( [ eqa1 - 4a ] ) and ( [ eqa1 - 5a ] ) in appendix [ app1 ] . thus the above gaussian integration results in @xmath124 where @xmath125 is given by equation ( [ eqa1 - 3 ] ) . now , calculation of equation ( [ eq1 - 17 ] ) is straightforward : @xmath126 . \label{eq2 - 5}\end{aligned}\ ] ] on the other hand , the gaussian contribution is simply given by @xmath127 thus , the perturbative expansion of equation ( [ eq1 - 13 ] ) up to second order is finally given by @xmath128 \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\}. \label{eq2 - 7}\end{aligned}\ ] ] the second order formulas for area and length statistics are given by equation ( [ eq2 - 2 ] ) with the above equation . to evaluate the above formula , the factor @xmath88 and @xmath129 should be known . those factors are evaluated by usual perturbation theory in the following section . the next statistics we consider are the genus statistics . the genus statistics have been attractive because it has a geometrical meaning of the clustering as well as the cosmological significance . the 2d genus statistic @xmath130 is defined in a two - dimensional plane @xmath131 in a @xmath6-dimensional space , so that @xmath132 is required . in this plane @xmath131 , there are contours corresponding to each threshold @xmath133 . the 2d genus statistic is defined by the number of contours surrounding regions higher than the threshold value minus the number of contours surrounding regions lower than the threshold value @xcite . this definition is intuitive , but an alternative , equivalent definition is more useful : first we set an arbitrary , fixed direction on the plane @xmath131 . then there are maxima and minima of contours according to that direction . these points are classified into upcrossing points and downcrossing points with respect to that chosen direction . the number of those points are used to define the 2d genus statistic as the following way : @xmath134 , \label{eq2 - 8}\end{aligned}\ ] ] per unit area of the surface . according to this definition , the explicit expression of the 2d genus statistic is given by @xmath135 for this statistic , the indices @xmath136 only take @xmath119 , @xmath120 , @xmath137 and @xmath138 . let the number of @xmath119 be @xmath121 , @xmath120 be @xmath139 , @xmath137 be @xmath140 , and @xmath138 be @xmath141 . then we need to calculate the following quantity : @xmath142 since the second derivative @xmath143 appears as a polynomial , we just use the transform of equation ( [ eq1 - 13 - 3 - 1 ] ) . then , from equations ( [ eqa1 - 4a ] ) , ( [ eqa1 - 5a ] ) and ( [ eqa1 - 5b ] ) , the above equation reduces to @xmath144 . \label{eq2 - 11}\end{aligned}\ ] ] thus , from equation ( [ eq1 - 17 ] ) , @xmath145 - 2 s^{(2 ) } \left ( \frac{\sigma_1}{\sqrt{d}\sigma_0 } \right)^4 r(0;0;2,1 ) \nonumber\\ & = & \frac{1}{(2\pi)^{3/2 } } \left ( \frac{\sigma_1}{\sqrt{d}\sigma_0 } \right)^2 e^{-\nu^2/2 } \left [ s^{(0 ) } h_4(\nu ) + 4 s^{(1 ) } h_2(\nu ) + 2 s^{(2 ) } \right ] . \label{eq2 - 12}\end{aligned}\ ] ] on the other hand , the gaussian contribution is simply given by @xmath146 thus , the perturbative expansion of equation ( [ eq1 - 13 ] ) up to second order is finally given by @xmath147 \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\}. \label{eq2 - 14}\end{aligned}\ ] ] the second order formula for 3d genus statistic was already derived by @xcite , but the detailed derivation was omitted . for completeness , we revisit the same quantity from our general point of view here . while the 2d genus statistic is defined by the number of contour lines in 2d surface , the 3d genus statistic @xcite is defined by the number of contour surfaces and the number of handles in 3d space . thus the 3d genus is defined only for cosmic fields of @xmath148 . the 3d genus statistic @xmath149 is defined by @xmath150 , \label{eq2 - 15}\end{aligned}\ ] ] per unit volume of the 3d space . this quantity is mathematically equivalent to @xmath151 times euler characteristic of the contour surfaces , and thus is proportional to the total surface integral of local curvature of contours from the gauss - bonnet theorem . although those definition is intuitive , an alternative , equivalent definition is more useful as in the 2d genus case . we set an arbitrary direction in the 3d space . then there are maxima , minima and saddle points according to that direction . from the number of these points , the 3d genus is defined by @xmath152 , \label{eq2 - 16}\end{aligned}\ ] ] per unit volume . according to this definition , the explicit expression of the 3d genus statistic is given by @xcite @xmath153 we need to calculate the following quantity : @xmath154 after applying the transform of equation ( [ eq1 - 13 - 3 - 1 ] ) , the gaussian integration of @xmath64 fixing @xmath18 is given by @xmath155 , \label{eq2 - 19}\end{aligned}\ ] ] where @xmath156 is defined in table [ tab1 ] . ccc|ccc @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 & @xmath162 + @xmath119 & @xmath119 & @xmath119 & @xmath120 & @xmath119 & @xmath163 + @xmath120 & @xmath119 & @xmath119 & @xmath119 & @xmath120 & @xmath163 + @xmath119 & @xmath120 & @xmath119 & @xmath119 & @xmath120 & @xmath163 + @xmath120 & @xmath120 & @xmath119 & @xmath119 & @xmath119 & @xmath164 + @xmath119 & @xmath119 & @xmath137 & @xmath119 & @xmath119 & @xmath165 + then , from equations ( [ eqa1 - 4a ] ) , ( [ eqa1 - 5a ] ) and ( [ eqa1 - 5b ] ) , the equation ( [ eq2 - 18 ] ) reduces to @xmath166 . \label{eq2 - 20}\end{aligned}\ ] ] since the 3d genus is only defined for @xmath167 and our universe has the spatial dimension 3 , only @xmath148 is meaningful for actual cosmic fields . nevertheless , we preserve the general dimension @xmath6 for some flavor of generality . thus , from equation ( [ eq1 - 17 ] ) , @xmath168 . \label{eq2 - 21}\end{aligned}\ ] ] the gaussian contribution is given by @xcite @xmath169 thus , the perturbative expansion of equation ( [ eq1 - 13 ] ) up to second order is finally given by @xmath170 \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\}. \label{eq2 - 23}\end{aligned}\ ] ] the above equation with @xmath148 agrees with @xcite , @xmath171 , and @xmath172 in @xcite are related to the notations here by @xmath173 , @xmath174 , @xmath175 . ] next , we consider the weighted extrema density above a threshold @xmath133 . the field extrema is defined to be points where all the first - order spatial derivatives of the field @xmath9 vanish : @xmath176 . the weight @xmath177 is associated to each extrema according to the number of negative eigenvalues of spatial second - order derivatives of the field . when the threshold is high enough , the weighted extrema are approximately identified with the field peaks . mathematically , this statistic is equivalent to the 2d genus statistics , according to the morse s theorem @xcite . therefore , we do not need to calculate separately to obtain the result on this statistic . however , we present the derivation of the extrema density as an alternative calculation to the 2d genus . the weighted extrema in 2d field is given by @xmath178 following the similar calculation of previous examples , and using equations in appendix [ app1 ] , we obtain @xmath179 , \label{eq2 - 25}\end{aligned}\ ] ] and the perturbative expansion of equation ( [ eq1 - 13 ] ) up to second order is finally given by @xmath180 \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\}. \label{eq2 - 26}\end{aligned}\ ] ] this result is in fact equivalent to the 2d genus statistics given by equation ( [ eq2 - 14 ] ) . most of the minkowski functionals are closely related to statistics discussed above . in this subsection , we comprehensively describe the exact relation between the minkowski functionals @xmath181 of a smoothed field and the statistical quantities considered above . the minkowski functional of @xmath182 is simply the volume fraction of the excursion set @xmath183 which is defined by high - density regions above a given threshold @xmath133 : @xmath184 the other functionals with @xmath185 are defined by the integral of the curvatures on isodensity surfaces of the threshold @xmath133 @xcite . in 3-dimensions , @xmath148 , they are evaluated by a surface integration averaged over whole system of volume @xmath186 @xcite , i.e. , @xmath187 where the local minkowski functionals , @xmath188 are defined by the principal curvatures @xmath189 and @xmath190 of the surface oriented toward lower density values . in 2-dimensions , @xmath70 , the minkowski functionals of @xmath191 are evaluated by a line integration averaged over whole system of 2d volume ( surface ) @xmath186 @xcite , i.e. , @xmath192 where the local minkowski functionals , @xmath193 are defined by the principal curvature @xmath189 of the line oriented toward lower density values . all the minkowski functionals for a gaussian random field are analytically derived by @xcite : @xmath194 where the factor @xmath195 is the volume of the unit ball in @xmath121 dimensions , so that @xmath196 , @xmath197 , @xmath198 , @xmath199 @xcite . it turns out that the minkowski functionals in 2- and 3-dimensions are identical to the statistics @xmath105 , @xmath130 and @xmath149 in each dimensions , except for normalization factors . in fact , from the crofton s formula @xcite , the @xmath121-th minkowski functional is given by @xmath200 in this formula for body @xmath183 in @xmath6 dimensions , we consider an arbitrary @xmath121-dimensional hypersurface @xmath201 and calculate the euler characteristic @xmath202 of the intersection @xmath203 in @xmath121 dimensions . this quantity is integrated over the space @xmath204 of all conceivable hypersurfaces . the integration measure @xmath205 is normalized to give @xmath206 . from this formula , we can see the statistics @xmath149 , @xmath130 , @xmath105 are proportional to the minkowski functionals of @xmath207 , @xmath208 , @xmath209 . in fact , @xmath210 is given by @xmath211 times the 3d genus ( or , equivalently , @xmath212 times the euler number of boundaries , @xmath213 ) , @xmath214 is identical to the 2d genus , and @xmath215 is just @xmath212 times the number of level - crossing points @xcite . thus , minkowski functionals are given by @xmath216 where the boundary of the body @xmath183 is identified with the isodensity contours of threshold @xmath133 and the statistics on right hand sides are defined in @xmath6-dimensions . therefore , we have already obtained the weakly non - gaussian expressions for minkowski functionals , i.e. , the minkowski functionals of @xmath217 are given by the above equations and equations ( [ eq2 - 7 ] ) , ( [ eq2 - 14 ] ) , and ( [ eq2 - 23 ] ) . the gaussian parts of the above equations exactly reproduce the tomita s formula ( [ eq2 - 54 ] ) . the remaining minkowski functional is the volume functional @xmath218 in this case , from equation ( [ eqa1 - 4a ] ) in appendix [ app1 ] , @xmath219 so that equation ( [ eq1 - 17 ] ) reduces to @xmath220 the equivalent form can be obtained by integrating the edgeworth expansion of pdf , equation ( [ eq2 - 0 - 2 ] ) . all the formulas of minkowski functionals derived above for @xmath221 fit into a single expression : @xmath222 \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\}. \label{eq2 - 60}\end{aligned}\ ] ] the density threshold @xmath133 so far is simply defined so that isodensity surface is identified by @xmath223 . however , the horizontal shift of the nonlinear genus curve etc . is considerably attributed to the nonlinear shift of probability distribution of the density field ( e.g. , * ? ? ? * ; * ? ? ? in order to cancel the latter shift , the threshold @xmath224 is defined so that the volume fraction @xmath225 on the high - density side of the isodensity surface equals to @xmath226 in fact , most of the work on genus analysis uses the genus curve plotted against the volume - fraction threshold @xmath224 . recently , @xcite re - expressed the weakly non - gaussian formula of genus curve @xcite in terms of @xmath224 , using perturbative expansion of the probability distribution function of the density field @xcite . we follow this method to re - express the weakly non - gaussian formulas of various statistical quantities derived above in terms of the volume - fraction threshold @xmath224 . the relation of @xmath133 and @xmath224 of weakly non - gaussian field is simply given by equating the two equations ( [ eq2 - 104 ] ) and ( [ eq2 - 105 ] ) . up to first order in @xmath49 , the relation reduces to @xmath227 it is straightforward to rewrite the various analytical formulas we derived above . the results for level - crossing statistic , 2d and 3d genus are respectively given by @xmath228 \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\ } , \label{eq2 - 207a}\\ g_2({{\widetilde{\nu } } } ) & = & \rho_{\rm e2}({{\widetilde{\nu } } } ) = \frac{1}{(2\pi)^{3/2 } } \left(\frac{\sigma_1}{\sqrt{d}\sigma_0}\right)^2 e^{-{{\widetilde{\nu}}}^2/2 } \nonumber\\ & & \qquad\qquad\quad\times\ , \left\ { h_1({{\widetilde{\nu } } } ) + \left [ \frac23 \left(s^{(1 ) } - s^{(0)}\right ) h_2({{\widetilde{\nu } } } ) + \frac13 \left(s^{(2 ) } - s^{(0)}\right ) \right ] \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\ } , \label{eq2 - 207b}\\ g_3({{\widetilde{\nu } } } ) & = & - \frac{1}{(2\pi)^2 } \left(\frac{\sigma_1}{\sqrt{d}\sigma_0}\right)^3 e^{-{{\widetilde{\nu}}}^2/2 } \nonumber\\ & & \qquad\qquad\times\ , \left\ { h_2({{\widetilde{\nu } } } ) + \left [ \left(s^{(1 ) } - s^{(0)}\right ) h_3({{\widetilde{\nu } } } ) + \left(s^{(2 ) } - s^{(0)}\right ) h_1({{\widetilde{\nu } } } ) \right ] \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\ } , \label{eq2 - 207c}\end{aligned}\ ] ] the results for minkowski functionals of @xmath217 are again given by these equations and equations ( [ eq2 - 101a])([eq2 - 101c ] ) . all the minkowski functionals for @xmath229 fit into a single expression : @xmath230 \sigma_0 + { \cal o}({\sigma_0}^2 ) \right\}. \label{eq2 - 208}\end{aligned}\ ] ] remarkably , the highest - order hermite polynomial in the non - gaussian correction terms vanishes in each case . in addition , the skewness parameters only appear as combinations of the form , @xmath231 , which makes the result simpler compared with the original form with the direct threshold @xmath133 . as we will see in the following sections , the numerical values of @xmath232 are quite close , or even identical in some special models . this means that the non - gaussian corrections of the above statistics are smaller with the rescaled threshold @xmath224 than with the original threshold @xmath133 . this is one of the central results in this paper . this tendency is in agreement with the analyses of numerical simulations . we need to know the skewness parameters , @xmath87 for the evaluation of the second - order perturbative terms of equation ( [ eq1 - 17 ] ) . these quantities can be calculated by usual perturbation theory , once we specify the cosmic field , @xmath9 . in each example of the previous section , the quantity @xmath233 does not appear so that we evaluate @xmath88 , @xmath129 , and @xmath234 in this section . the other kinds of skewness parameters like @xmath233 can be similarly evaluated without difficulty . before we explore actual cosmic fields , we consider a simple , phenomenological statistical model , i.e. , the hierarchical model of higher order correlation functions ( e.g. , * ? ? ? * ) . in this model , @xmath26-point correlation function is a sum of @xmath235 products of the two - point correlation function . specifically , the three - point correlation function is given by @xmath236 . \label{eq4 - 1}\end{aligned}\ ] ] we assume the field @xmath9 in the above equation is already smoothed . in this case , skewness parameters , equations ( [ eq1 - 15a])([eq1 - 15d ] ) are given by straightforward calculations @xcite : @xmath237 these values depend on a hierarchical amplitude , @xmath238 , which is a free parameter of this model . the relative amplitudes among @xmath87 are not freely adjusted in the above equation . in the case of volume - fraction threshold , the first non - gaussian correction of the various statistics considered in the previous section is absent , since they depend only on @xmath239 . next , we consider the three - dimensional density field . the skewness parameters of this field in perturbation theory are already calculated by @xcite and @xcite , using fourier transforms of the field . we comprehensively review this calculation here for completeness . there is an alternative way to calculate the skewness not depending on fourier transforms @xcite . the cosmic field @xmath9 is identified with the 3d density contrast , @xmath240 , where @xmath241 is the density field . the dimension this field is defined in is three , @xmath148 . the fourier transform of the field is useful in the following : @xmath242 in this notation , two- and three - point correlations in fourier space have the forms , @xmath243 the above forms are the consequence of the statistical homogeneity of the space , where the functions @xmath244 and @xmath245 are the power spectrum and the bispectrum , respectively . thus , the variance and its variants [ eqs . ( [ eq1 - 2 ] ) , ( [ eq1 - 13 - 3a ] ) , and ( [ eq1 - 13 - 3b ] ) ] , in their fourier representation , are given by @xmath246 and the skewness parameters of equations ( [ eq1 - 15a])([eq1 - 15d ] ) are given by @xmath247 for initial random gaussian density field , the second order perturbation theory predicts the power spectrum and the bispectrum as follows ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) : @xmath248 { p_{\rm lin}}(k_1 ) { p_{\rm lin}}(k_2 ) w(k_1r ) w(k_2r ) w(k_3r ) \nonumber\\ & & \qquad\qquad\qquad\quad + \ , { \rm cyc.(1,2,3 ) } + { \cal o}({\sigma_0}^6 ) , \label{eq4 - 7b}\end{aligned}\ ] ] where @xmath249 is the linear power spectrum , and @xmath201 is a weak function of cosmology @xcite . the field smoothing corresponds to the multiplication of the window function @xmath250 which is the three - dimensional fourier transform of the smoothing function @xmath10 . it is a good approximation to use the value of @xmath201 for an einstein - de sitter universe , @xmath251 , in most cases . the explicit form of the function @xmath201 in terms of cosmological parameters @xmath252 and @xmath253 is given by @xcite , which is accurately fitted by @xmath254 the perturbation theory is considered to be an expansion by a parameter @xmath49 . in this respect , the power spectrum @xmath244 and the bispectrum @xmath245 is of order @xmath16 and @xmath255 , respectively . substituting equation ( [ eq4 - 7a ] ) into equation ( [ eq4 - 5 ] ) , we obtain @xmath256 up to the lowest order . similarly , substituting equation ( [ eq4 - 7b ] ) into equations ( [ eq4 - 6a])([eq4 - 6c ] ) , and introducing new integration variables , @xmath257 , @xmath258 , and @xmath259 , we obtain @xmath260 where @xmath261 and @xmath262 \frac{w\left(\sqrt{{l_1}^2 + { l_2}^2 + 2 l_1 l_2 \mu } \right ) } { w(l_1 ) w(l_2 ) } \nonumber\\ & & \qquad\qquad\qquad \times\ , \left\ { \begin{array}{ll } \displaystyle 2 , & a=0,\\ { l_1}^2 + { l_2}^2 + l_1 l_2 \mu , & a=1,\\ 3\ , { l_1}^2 { l_2}^2 \left ( 1 - \mu^2 \right ) , & a=2 , \end{array } \right . \label{eq4 - 9}\end{aligned}\ ] ] so far the smoothing function is arbitrary . for a general smoothing function , the above equations can be numerically integrated to obtain the skewness parameters for each model of the power spectrum . for some smoothing functions , further analytical reductions of the above equations are possible . as a popular example , we consider the gaussian smoothing , @xmath263 , which is frequently adopted for practical purposes . we follow the similar technique of @xcite , in which they derived the skewness of the density field with gaussian smoothing . for the gaussian smoothing , @xmath264 in this case , the following formula of the modified bessel function @xmath265 is useful : @xmath266 where @xmath267 is the @xmath122-th legendre polynomial . from this formula , the angular integration of @xmath268 in equation ( [ eq4 - 9 ] ) can be analytically performed and the result is @xmath269 , \label{eq4 - 12a}\\ & & \widetilde{s}^{(1 ) } = \sqrt{2\pi l_1l_2 } \left\ { \frac{5 + 2e}{4 } \left(\frac{l_2}{l_1 } + \frac{l_1}{l_2}\right ) i_{1/2}(l_1l_2 ) - \left [ \frac{3(9 + e)}{10 } + \frac{{l_2}^2}{{l_1}^2 } + \frac{{l_1}^2}{{l_2}^2 } \right ] i_{3/2}(l_1l_2 ) \right . \nonumber\\ & & \qquad\qquad\qquad\qquad\qquad\qquad + \ , \left . \frac{2 - e}{2 } \left(\frac{l_2}{l_1 } + \frac{l_1}{l_2}\right ) i_{5/2}(l_1l_2 ) - \frac{3(1 - e)}{10 } i_{7/2}(l_1 l_2 ) \right\ } , \label{eq4 - 12b}\\ & & \widetilde{s}^{(2 ) } = \sqrt{2\pi}(l_1l_2)^{3/2 } \left [ \frac{3(3 + 2e)}{5 } i_{1/2}(l_1l_2 ) - \frac{9}{10 } \left(\frac{l_2}{l_1 } + \frac{l_1}{l_2}\right ) i_{3/2}(l_1l_2 ) \right . \nonumber\\ & & \qquad\qquad\qquad\qquad\qquad -\ , \left . \frac{3(3 + 4e)}{7 } i_{5/2}(l_1l_2 ) + \frac{9}{10 } \left(\frac{l_2}{l_1 } + \frac{l_1}{l_2}\right ) i_{7/2}(l_1l_2 ) - \frac{18(1 - e)}{35 } i_{9/2}(l_1l_2 ) \right ] . \label{eq4 - 12c}\end{aligned}\ ] ] at this point , it is useful to define the following quantity : @xmath270 in the above equation , variance parameters @xmath49 , @xmath271 are given by equation ( [ eq4 - 7 - 1 ] ) . the nonlinear correction for @xmath272 is not needed because our estimate of @xmath87 is only lowest order in @xmath49 . when the higher order corrections , e.g. , third - order perturbation corrections , are estimated , one should be sure that all the necessary nonlinear corrections are properly taken into account . with this quantity , the skewness parameters are given by @xmath273 , \label{eq4 - 14b}\\ & & s^{(2)}(r ) = 9 \left [ \frac{3 + 2e}{15 } s^{33}_0 - \frac15 s^{24}_1 - \frac{3 + 4e}{21 } s^{33}_2 + \frac15 s^{24}_3 - \frac{2(1-e)}{35 } s^{33}_4 \right ] . \label{eq4 - 14c}\end{aligned}\ ] ] thus the lowest order estimates of skewness parameters are given by the above equations ( [ eq4 - 13])([eq4 - 14c ] ) . for each given power spectrum , the integration of equation ( [ eq4 - 13 ] ) is straightforward . the resulting skewness parameters are independent on the amplitude of the power spectrum . when the power spectrum is given by a cdm - like model , @xmath274 , where @xmath275^{-1/4 } , \label{eq4 - 15}\end{aligned}\ ] ] is the cdm - like transfer function fitted by @xcite , and @xmath276 is the shape parameter of this model , then the skewness parameters are functions of @xmath277 . in table [ tab2 ] , we give the values of skewness parameters @xmath87 for cdm - like models for several values of @xmath277 . c|cccccc @xmath277 & @xmath278 & @xmath279 & @xmath280 & @xmath281 & @xmath282 & @xmath283 + @xmath88 & @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath288 & @xmath289 + @xmath129 & @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath294 & @xmath295 + @xmath234 & @xmath296 & @xmath297 & @xmath298 & @xmath299 & @xmath300 & @xmath301 + in this table , the value of @xmath201 is approximated by @xmath302 . since the skewness parameters are weak functions of @xmath277 as seen from the table , one can interpolate the values in this table to obtain the values of arbitrary scales for practical purposes . when the power spectrum is given by a power - law form , @xmath303 the integration of equation ( [ eq4 - 13 ] ) can be analytically performed . first , the simple gaussian integration gives @xmath304 second , we expand the modified bessel function as @xmath305 then the equation ( [ eq4 - 13 ] ) reduces to @xmath306 where @xmath307 , and @xmath55 is the gauss hypergeometric function : @xmath308 the equations ( [ eq4 - 14a])([eq4 - 14c ] ) and equation ( [ eq4 - 19 ] ) give the skewness parameters @xmath87 . the following recursion relations for hypergeometric function @xmath309 , \label{eq4 - 21b}\\ & & ( \gamma-\alpha)(\gamma-\beta)z f(\alpha,\beta,\gamma+1;z ) + \gamma[(\alpha+\beta-2\gamma+1)z+\gamma-1 ] f(\alpha,\beta,\gamma;z ) \nonumber\\ & & \qquad\qquad\qquad \ , + \gamma(\gamma-1)(z-1 ) f(\alpha,\beta,\gamma-1;z ) = 0 . \label{eq4 - 21c}\end{aligned}\ ] ] simplify the result : @xmath310 in this power - law case , skewness parameters do not depend on scales @xmath2 but only on power - law index , @xmath311 . incidentally , @xmath312 and @xmath313 are identical . this is just the coincidence and is not generally the case when the power spectrum is not given by power - law . several numerical values are shown in table [ tab3 ] , where @xmath251 is assumed . c|ccccccccc @xmath314 & @xmath315 & @xmath316 & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath321 & @xmath322 & @xmath278 + @xmath323 & @xmath324 & @xmath325 & @xmath326 & @xmath327 & @xmath328 & @xmath329 & @xmath330 & @xmath331 & @xmath332 + @xmath234 & @xmath333 & @xmath334 & @xmath335 & @xmath336 & @xmath337 & @xmath338 & @xmath339 & @xmath340 & @xmath341 + next , we consider the 3d velocity field as the cosmic field @xmath9 . since the rotational components of the velocity field are decaying modes of gravitational evolution in perturbation theory , we only consider the rotation - free component . the cosmic field @xmath9 is identified with the dimensionless scalar field , @xmath342 where @xmath343 is the hubble parameter . one can also consider other quantities like radial component of the velocity field , @xmath344 , where @xmath345 is the line - of - sight normal vector . those quantities are more complicated than a simple divergence . we illustrate only the simplest case in this paper . the second order perturbation theory predicts power spectrum and bispectrum of the velocity field as follows ( e.g. , * ? ? ? * ) : @xmath346 { p_{\rm lin}}(k_1 ) { p_{\rm lin}}(k_2 ) w(k_1r ) w(k_2r ) w(k_3r ) \nonumber\\ & & \qquad\qquad\qquad\quad + \ , { \rm cyc.(1,2,3 ) } + { \cal o}({\sigma_0}^6 ) . \label{eq4 - 24b}\end{aligned}\ ] ] in the above equation , the factor @xmath347 is a logarithmic derivative of the growth factor , given by @xmath348 @xcite and @xmath349 is the linear growth factor and @xmath350 is the expansion factor . the logarithmic derivative is evaluated at present . the factor @xmath351 is a weak function of cosmology . it is a good approximation to use the value for an einstein - de sitter universe , @xmath352 . the explicit form of the function @xmath351 in terms of @xmath353 and @xmath354 is given by @xcite , which is accurately fitted by @xmath355 the similarity of the equations ( [ eq4 - 24a ] ) and ( [ eq4 - 24b ] ) for velocity field to the equations ( [ eq4 - 7a ] ) , ( [ eq4 - 7b ] ) for density field is obvious . we can easily see the skewness parameters for the velocity field are obtained by similar equations as ( [ eq4 - 14a])([eq4 - 14c ] ) : @xmath356 , \label{eq4 - 26a}\\ & & s^{(1)}(r ) = - \frac{3}{2 { g_f } } \left [ \frac{5 + 2e_v}{3 } s^{13}_0 - \frac{9 + e_v}{5 } s^{22}_1 - s^{04}_1 + \frac{2(2-e_v)}{3 } s^{13}_2 - \frac{1-e_v}{5 } s^{22}_3 \right ] , \label{eq4 - 26b}\\ & & s^{(2)}(r ) = - \frac{9}{{g_f } } \left [ \frac{3 + 2e_v}{15 } s^{33}_0 - \frac15 s^{24}_1 - \frac{3 + 4e_v}{21 } s^{33}_2 + \frac15 s^{24}_3 - \frac{2(1-e_v)}{35 } s^{33}_4 \right ] , \label{eq4 - 26c}\end{aligned}\ ] ] where @xmath357 is given by equation ( [ eq4 - 13 ] ) without any modification . in table [ tab4 ] , we show the values of skewness parameters of the velocity field for the cdm - like models for several values of @xmath277 . c|cccccccc @xmath277 & @xmath278 & @xmath279 & @xmath280 & @xmath281 & @xmath282 & @xmath283 + @xmath358 & @xmath359 & @xmath360 & @xmath361 & @xmath362 & @xmath363 & @xmath364 + @xmath365 & @xmath366 & @xmath367 & @xmath368 & @xmath369 & @xmath370 & @xmath371 + @xmath372 & @xmath373 & @xmath374 & @xmath375 & @xmath376 & @xmath377 & @xmath378 + in this table , the value of @xmath351 is set as @xmath379 . for the power spectrum of the power - law form , @xmath380 , \label{eq4 - 27a}\\ & & s^{(2)}({{n_{\rm s } } } ) = -\frac{3}{{g_f } } \left [ f\left(\frac{{{n_{\rm s}}}+5}{2},\frac{{{n_{\rm s}}}+5}{2},\frac52;\frac14\right ) - \frac15 ( { { n_{\rm s}}}+ 4 - 4e_v ) f\left(\frac{{{n_{\rm s}}}+5}{2},\frac{{{n_{\rm s}}}+5}{2},\frac72;\frac14\right ) \right ] . \label{eq4 - 28c}\end{aligned}\ ] ] several numerical values are shown in table [ tab5 ] , where @xmath381 is assumed . c|ccccccccc @xmath314 & @xmath315 & @xmath316 & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath321 & @xmath322 & @xmath278 + @xmath382 & @xmath327 & @xmath383 & @xmath384 & @xmath385 & @xmath386 & @xmath387 & @xmath388 & @xmath389 & @xmath390 + @xmath372 & @xmath391 & @xmath392 & @xmath393 & @xmath394 & @xmath395 & @xmath396 & @xmath397 & @xmath398 & @xmath399 + the projection of the density field @xmath400 defines the 2d cosmic fields on the sky ( @xmath70 ) . here , we derive the skewness parameters for this field . the 2d projected density field with top - hat kernel is investigated by @xcite . here , we are interested in gaussian kernel for our purpose . in a friedman - lematre universe , the comoving angular diameter distance at a comoving distance @xmath401 is given by @xmath402 depending on the sign of the spatial curvature @xmath403 . thus , in spherical coordinates , projected density field in 2d is given by @xmath404 where @xmath405 is the selection function without volume factor , normalized as @xmath406 , and @xmath407 is the 3d comoving density field constant . ] . the present value of the conformal time @xmath408 is @xmath409 . the projected density contrast is defined by @xmath410 , where one can see @xmath411 . we identify the projected density contrast with the 2d ( @xmath70 ) field @xmath9 . since the smoothing angle @xmath412 is much smaller than @xmath413 in most of the interested cases , we consider the small patch of the sky of the vicinity of the polar axis , @xmath414 . with this approximation , we introduce the variables @xmath415 , and @xmath416 , which are considered as 2d euclidean coordinates , @xmath417 . therefore , the projection equation is given by @xmath418 where @xmath419 is the density contrast at comoving coordinates @xmath8 and conformal time @xmath420 . the power spectrum and the bispectrum for the above projected field are given by the limber s equations ( [ eqa2 - 2 ] ) and ( [ eqa2 - 7 ] ) in appendix [ app2 ] , with @xmath421 : @xmath422 where @xmath244 , @xmath245 are 2d projected power spectrum and bispectrum , respectively , of the field @xmath9 , and @xmath423 , @xmath424 are 3d power spectrum and bispectrum , respectively . the 3d power spectrum and the 3d bispectrum are evaluated by the second order perturbation theory . they are similar to equations ( [ eq4 - 7a ] ) and ( [ eq4 - 7b ] ) , but we have to take into account the time - dependence here . they are given by @xmath425 { p_{\rm lin}}(k_1 ) { p_{\rm lin}}(k_2 ) \nonumber\\ & & \qquad\qquad\qquad\qquad\qquad\quad + \ , { \rm cyc.(1,2,3 ) } , \label{eq4 - 34b}\end{aligned}\ ] ] where @xmath426 is the linear growth factor at conformal lookback time @xmath401 , ( i.e. , at conformal time @xmath427 ) , which is normalized as @xmath428 . the following fitting formula @xcite is useful : @xmath429 where @xmath353 and @xmath354 are time - dependent cosmological parameters at conformal lookback time @xmath401 . the variable @xmath430 is a weak function of time and cosmology , and for einstein - de sitter universe , @xmath251 . this quantity @xmath201 is the same we used in 3d density field , but here we also take into account the time - dependence . it is accurately approximated by @xmath431 the variance parameters of the smoohted projected field are given by @xmath432 , \label{eq4 - 36}\end{aligned}\ ] ] where @xmath433 the skewness parameters of the smoothed projected field are given by @xmath434 { \sigma_1}^{2a}\left[s_k(\chi){{\theta_{\rm f}}}\right ] c^{(a)}\left[s_k(\chi){{\theta_{\rm f}}}\right ] , \label{eq4 - 38}\end{aligned}\ ] ] where @xmath435 and @xmath436 \frac{w\left(\sqrt{l_1^{\,2 } + l_2^{\,2 } + 2 l_1 l_2 \mu } \right ) } { w(l_1 ) w(l_2 ) } \nonumber\\ & & \qquad\qquad\quad \times\ , \left\ { \begin{array}{ll } \displaystyle 2 , & a=0,\\ { l_1}^2 + { l_2}^2 + l_1 l_2 \mu , & a=1,\\ 4\ , { l_1}^2 { l_2}^2 \left ( 1 - \mu^2 \right ) , & a=2 , \end{array } \right . \label{eq4 - 40}\end{aligned}\ ] ] so far the smoothing function is arbitrary . for a general smoothing function , the above equations can be numerically integrated to obtain the skewness parameters for each model of the power spectrum . in the following , we adopt the gaussian smoothing function , @xmath437 . for this smoothing function , the equation ( [ eq4 - 10 ] ) holds even for this 2d case . in this case , the following integral representation of the modified bessel function @xmath265 for @xmath438 is useful : @xmath439 actually , the derivatives of the above equation , @xmath440 are sufficient to perform the angular integration in equation ( [ eq4 - 40 ] ) . moreover , one can use the property of the bessel function , @xmath441 , and @xmath442 to obtain formulas , @xmath443 from these formulas , equation ( [ eq4 - 40 ] ) reduce to @xmath444 , \label{eq4 - 45a}\\ & & \widetilde{c}^{(1 ) } = \frac34 \left\ { ( 4+e ) ( l_1^{\,2 } + l_2^{\,2 } ) i_0(l_1 l_2 ) - \left [ \frac{15 + e}{2 } l_1 l_2 + 2 \left ( \frac{l_1^{\,3}}{l_2 } + \frac{l_2^{\,3}}{l_1 } \right ) \right ] i_1(l_1 l_2 ) \right . \nonumber\\ & & \qquad\qquad\qquad \left . + \ ; ( 2-e ) ( l_1^{\,2 } + l_2^{\,2 } ) i_2(l_1 l_2 ) - \frac{1 - e}{2 } l_1 l_2 i_3(l_1 l_2 ) \right\ } , \label{eq4 - 45b}\\ & & \widetilde{c}^{(2 ) } = \frac34 \left [ ( 5 + 3e ) l_1^{\,2 } l_2^{\,2 } i_0(l_1 l_2 ) - 2 l_1 l_2(l_1^{\,2 } + l_2^{\,2 } ) i_1(l_1 l_2 ) - 4(1+e ) l_1^{\,2 } l_2^{\,2 } i_2(l_1 l_2 ) \right . \nonumber\\ & & \qquad\qquad \left . + \ ; 2 l_1 l_2 ( l_1^{\,2 } + l_2^{\,2 } ) i_3(l_1 l_2 ) - ( 1-e ) l_1^{\,2 } l_2^{\,2 } i_4(l_1 l_2 ) \right ] . \label{eq4 - 45c}\end{aligned}\ ] ] at this point , defining @xmath445 the equation ( [ eq4 - 39 ] ) reduces to @xmath446 , \label{eq4 - 47a}\\ & & c^{(1 ) } = \frac34 \left [ 2(4+e ) c^{13}_0 - \frac{15 + e}{2 } c^{22}_1 - 4 c^{04}_1 + 2(2-e ) c^{13}_2 - \frac{1 - e}{2 } c^{22}_3 \right ] , \label{eq4 - 47b}\\ & & c^{(2 ) } = \frac34 \left [ ( 5 + 3e ) c^{33}_0 - 4 c^{24}_1 - 4(1 + e ) c^{33}_2 + 4c^{24}_3 - ( 1 - e ) c^{33}_4 \right ] . \label{eq4 - 47c}\end{aligned}\ ] ] the two - dimensional integration of equation ( [ eq4 - 46 ] ) is performed only once as a function of @xmath2 . the result is stored as a table , and is used in one - dimensional numerical integration of equation ( [ eq4 - 38 ] ) for finally obtaining the skewness parameters in 2d projected density fields . functions @xmath447 for cdm - like models are given in table [ tab6 ] which are functions of @xmath448 . c|cccccccccccc @xmath277 & @xmath449 & @xmath322 & @xmath278 & @xmath279 & @xmath280 & @xmath281 & @xmath282 & @xmath283 & @xmath450 + @xmath451 & @xmath452 & @xmath453 & @xmath454 & @xmath455 & @xmath456 & @xmath457 & @xmath458 & @xmath459 & @xmath460 + @xmath461 & @xmath462 & @xmath463 & @xmath464 & @xmath465 & @xmath466 & @xmath467 & @xmath468 & @xmath469 & @xmath470 + @xmath471 & @xmath472 & @xmath473 & @xmath474 & @xmath475 & @xmath476 & @xmath477 & @xmath478 & @xmath479 & @xmath480 + when the 3d power spectrum is given by a power - law form of equation ( [ eq4 - 16 ] ) , the integration by wave length @xmath139 , @xmath140 can be analytically performed as in the 3d case . in fact , the parameter @xmath481 of equation ( [ eq4 - 37 ] ) with gaussian smoothing @xmath482 is given by @xmath483 and the variance parameters are given by @xmath484 with similar technique used in 3d density field , @xmath485 the equations ( [ eq4 - 47a])([eq4 - 47c ] ) and the above equation finally give the values of @xmath447 . the recursion relations of equations ( [ eq4 - 21a ] ) and ( [ eq4 - 21b ] ) simplify the result : @xmath486 these functions are independent on scale @xmath2 in the power - law case , but dependent on power - law index , @xmath311 . again , @xmath487 and @xmath488 are identical only in the power - law case . numerical values are given in table [ tab7 ] , where @xmath489 is assumed . c|ccccccccc @xmath311 & @xmath315 & @xmath316 & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath321 & @xmath322 & @xmath278 + @xmath490 & @xmath491 & @xmath492 & @xmath493 & @xmath494 & @xmath495 & @xmath496 & @xmath497 & @xmath498 & @xmath499 + @xmath471 & @xmath500 & @xmath501 & @xmath502 & @xmath503 & @xmath504 & @xmath505 & @xmath506 & @xmath475 & @xmath507 + for the power - law case , the skewness parameters of equation ( [ eq4 - 38 ] ) reduces to the following simple form : @xmath508^{2 - 2{{n_{\rm s } } } } n^3(\chi ) d^4(\chi ) } { \displaystyle \left\ { \int d\chi \left[s_k(\chi)\right]^{2-{{n_{\rm s } } } } n^2(\chi ) d^2(\chi ) \right\}^2}\ , c^{(a)}({{n_{\rm s } } } ) . \label{4 - 50 - 2}\end{aligned}\ ] ] it is interesting to compare the results with those of top - hat smoothing . according to @xcite , the top - hat smoothing gives @xmath509 for power - law case . comparing this expression with our table [ tab7 ] , they roughly agree with each other , and have similar behavior with spectral index . in detail , the gaussian smoothing gives more or less larger values . for example , top - hat smoothing gives @xmath510 for @xmath511 , respectively , while gaussian smoothing gives @xmath512 , respectively . it is reasonable that the @xmath451 of top - hat smoothing differs from that of gaussian smoothing for large spectral index , because the top - hat smoothing gathers more power on small scales than gaussian smoothing . the local convergence field of the weak lensing is commonly used for studying the large - scale structure of the universe ( e.g. , * ? ? ? * ; * ? ? ? assuming the situation where limber s equation ( see appendix [ app2 ] ) and also the born approximation @xcite hold , the following correspondence between the convergence field @xmath513 and the 3d density contrast @xmath514 is useful ( e.g. , * ? ? ? * ) : @xmath515 , \label{eq4 - 60}\end{aligned}\ ] ] where @xmath516 is the scale factor at conformal lookback time @xmath517 . the above equation is reduced to exactly the same form as the projected field of equation ( [ eq4 - 31 ] ) , but with the substitution , @xmath518 however , one should note that this substitution is valid only under the assumption of the born approximation . although the effect of the born approximation on the skewness is known to be weak @xcite , the validity of the born approximation in general situation has not been tested in detail @xcite . there is some subtlety which could arise when various combinations of skewness are considered . the above expressions of the skewness parameters are for unbiased fields . the skewness parameters for biased fields are non - trivial . they depend on the details of the biasing scheme in the real universe which is poorly known so far . however , the skewness parameters are well - defined quantities , so that they are calculated from the first principle once the biasing scheme is given . perhaps , one of the simplest , yet non - trivial case is the local , deterministic biasing . in this case , we can follow @xcite to obtain the perturbative expansion of the biasing : @xmath519 where @xmath520 and @xmath97 are the galaxy and mass density contrast , respectively . the spatial dimension is arbitrary , so that @xmath520 , @xmath97 can be functions in either 1d , 2d , or 3d space . the biased variance in lowest order is given by @xmath521 it is also straightforward to calculate skewness parameters . after some algebra , all skewness parameters are shown to transform in a same way : @xmath522 irrespective of spatial dimensions , @xmath6 , and of kinds of skewness parameters , @xmath523 . in this framework , the parameters @xmath524 and @xmath525 are needed . a possible way to determine these parameters is to measure the variance @xmath526 and the skewness @xmath527 from the observation . theoretical models predict the variance @xmath528 and the skewness @xmath529 of the mass distribution . we obtain biasing parameters by @xmath530 and @xmath531 . the derivatives of the skewness , @xmath532 , @xmath533 , are then obtained from equation ( [ eq4 - 71 ] ) . in case of level - crossing , and genus statistics as functions of scaled threshold @xmath224 , the second - order corrections in equation ( [ eq2 - 207a])([eq2 - 207c ] ) only depends on the difference of the skewness parameters times the variace . quite remarkably , this type of second - order correction term does not depend on bias parameter at all : @xmath534 \sigma_{\rm 0,g } = \left[s^{(a ) } - s^{(0)}\right ] \sigma_{\rm 0 } \label{eq4 - 72}\end{aligned}\ ] ] thus , the second - order nonlinear corrections for level - crossing and genus statistics of locally biased field are exactly the same as that of unbiased mass density field . this can be considered as an advantage of the scaled threshold @xmath224 in these statistics , because the biasing is one of the most embarrasing uncertainty in the analysis of galaxy distribution . however , one should note this result is derived under the local biasing scheme . so far the locality of the bias is not guaranteed in general . stochastic argument @xcite of the biasing is more complicated for derivative skewness @xmath129 and @xmath234 than for usual skewness @xmath88 , because they involve the correlation between field derivatives . the phenomenological nature of the stochastic biasing requires many parameters which is not calculable from first principles . therefore , the stochastic biasing sheme does not effectively work for our problem . more physical treatment of the biasing schemes of galaxy formation is needed , in which case the nonlocality of the bias could also be important @xcite . the perturbative calculations offer valuable aspects of weakly non - linear evolution of the various statistics without laborious parameter survey by numerical simulations . in this paper , we obtain the lowest nonlinear corrections to relatively popular statistics of smoothed cosmic fields . using these results , it is interesting to see how the weakly nonlinear effect tends to distort the gaussian prediction for those exemplified statistics . in figure [ fig1 ] , various statistics for 3d density field is shown . the amplitude of each statistic is appropriately normalized as we are interested in the deviation from the gaussian prediction . if we neglect the normalization , minkowski functionals of @xmath217 are equivalent to the statistics @xmath105 , @xmath130 , and @xmath149 , respectively , so that they degenerate in this figure ( we should note the sign of @xmath535 is inverted ) . the rms @xmath49 , which is considered as a weakly nonlinear parameter , is set @xmath536 . a limit @xmath537 corresponds to the prediction of the linear theory , which is given by thin solid lines in the figure . this linear prediction is equivalent to the gaussian fluctuations , because we assume the initial density field is random gaussian . in general , the curves of statistics plotted against the direct density threshold , @xmath133 ( dotted lines ) , exhibit considerable deviations from gaussian predictions . the overall tendency does not depend much on the shape of the spectrum we consider here , i.e. , power - law spectrum with index @xmath165 , @xmath211 , @xmath119 , and cdm model with smoothing length @xmath538 , where @xmath276 is the shape parameter of the cdm spectrum . they are consistent with the so - called meat - ball shift , which means that there are more isolated regions in a nonlinear field than in a gaussian field for a fixed threshold . in fact , @xmath105 , @xmath130 , @xmath539 of high value of threshold , e.g. , @xmath540 , virtually correspond to the number of isolated regions , and each figure shows that the number is indeed increased by weakly nonlinear evolution . the weakly nonlinear formula for the genus curve against the density threshold , @xmath541 , which was first derived by @xcite has been compared with numerical simulations in literatures . @xcite shows the good agreement of the analytic prediction with the simulations for various spectra . @xcite compared the prediction of the genus curve against direct @xmath133 with the simulated sdss data . unfortunately , in their published paper , they have transcribing errors , which incorrectly made the perturbation theory considerably disagree with their data . there are also ambiguity in their comparison on biasing which alter the values of skewness parameters . one can guess the biasing effect on skewness parameters by equation ( [ eq4 - 70 ] ) . they chose the peak particles as galaxies , and the linear biasing parameter is inferred as @xmath542 , but the nonlinear parameter @xmath525 is not obvious in their work . some literatures indicate the skewness @xmath88 of peaks are roughly given by 12 @xcite but for highly biased peaks @xmath543 . if we adopt @xmath544 , the skewness is given by @xmath545 which is not unreasonable . if it is the case for their simulation , the perturbative prediction and their data completely agree with each other . the @xmath546-value per degrees of freedom reduces to only 1.03 ( private communication with w. n. colley & d. h. weinberg ) . obviously , we have to further investigate the biasing effects in numerical simulations to obtain a conclusive result . most of the topological analyses of the previous work use the scaled threshold @xmath224 . the dashed lines in fig . [ fig1 ] shows the corresponding curves . the deviations from the linear theory is dramatically reduced . this fact is empirically known by the analyses of numerical simulations @xcite . the reason for this reduction is mathematically due to the closeness of the values of skewness parameters @xmath87 , ( @xmath523 ) , since all the terms of the nonlinear corrections in equations ( [ eq2 - 207a])([eq2 - 207c ] ) depend only on @xmath239 ( @xmath547 ) . for the hierarchical model of equation ( [ eq4 - 1 ] ) , they are exactly zero , that means there is not any ( second order ) nonlinear correction for the hierarchical model . since the hierarchical model is known to roughly approximate the nonlinear evolution , it is not surprising that more realistic fields have only small corrections of nonlinearity if they are plotted against the volume - fraction threshold , @xmath224 . for power - law models , @xmath88 and @xmath129 are exactly the same . that makes the nonlinear correction for @xmath105 or @xmath548 exactly vanishes . thus , the nonlinear corrections for other statistics are arisen by the difference between @xmath88 and @xmath234 . for the cdm model , there still is a difference between @xmath88 and @xmath129 , but it is relatively small as seen from the table [ tab2 ] . as for the topological statistics , @xmath130 , @xmath149 , @xmath549 and @xmath535 , deviations of curves against @xmath224 from the linear theory prediction depend on the underlying spectrum through differences of skewness parameters . the weakly nonlinear effect for redder spectrum of @xmath550 induces a sponge - like shift , that means the number of holes in isolated regions increases . on the other hand , the bluer spectrum of @xmath551 indicates a meat - ball shift . these tendencies are qualitatively in agreement with the numerical results ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? it is not trivial to estimate errors expected in observationally estimating statistics of smoothed cosmic fields . the observational errors mainly consists of the shot noise and the cosmic variace . the systematic comparison with simulations should be used for the error estimates . unfortunately , previous earlier simulations does not have enough resolution to be quantitatively compared in weakly nonlinear regime . @xcite use much larger simulations than those earlier work , and gives the genus curve in weakly nonlinear regime in their fig . 9 . relative depths of left and right troughs in the genus curve show slightly meat - ball shift , which is consistent with our prediction . although their variation of smoothing length is rather limited , their plots ensures the slight meat - ball shift predicted by the perturbation theory can definitely be observed . @xcite plot the 2d genus curve , but they use the thin slice which suffers strongly nonlinear effects so that the quantitative comparison is not appropriate with weakly nonlinear results . they report the meat - ball shift in 2d genus , which is the same direction of the weakly nonlinear correction besides amplitude . we find there is stll neccesity of proper , systematic comparison between the perturbation theory and numerical simulations in weakly nonlinear regime . the decisive factor between sponge - like shift and meat - ball shift is the sign and amplitude of @xmath552 , since @xmath553 is exactly or approximately zero for wide range of models like power - law , or cdm - like models . if the factor @xmath552 is positive , the meat - ball shift takes place . if this factor is negative , the sponge - like shift occurs . for power - law case , that factor is positive for @xmath554 , and is negative for @xmath555 . the amplitude of this factor times the amplitude of the nonlinearity parameter , @xmath556 determines the amplitude of the meat - ball shift in the genus curve . prominent shifts are seen around the two troughs of the genus curve at @xmath557 . at these troughs , the factor @xmath553 does not contribute to the genus curve since the accompanying factor @xmath558 vanishes . therefore , one can almost completely characterize the shifts by a factor @xmath559 , which we call the genus asymmetry parameter , around the troughs . we propose this genus asymmetry parameter as a theory - motivated parameter of the genus asymmetry . this factor is observationally determined by fitting the shape of the genus curve by a form , @xmath560 where @xmath245 is the genus asymmetry parameter . in this fitting , @xmath561 is much smaller if the non - gaussian features are from purely gravitational evolution of the initial gaussian field , and if the power spectrum is smooth enough as cdm models . the genus asymmetry parameter for cdm models with shape parameter @xmath276 , normalized by @xmath562 is listed for various smoothing length in table [ tab8 ] . cccccccccc @xmath563 & ... & @xmath564 & @xmath565 & @xmath566 & @xmath567 & @xmath568 & @xmath569 & @xmath570 & @xmath571 + @xmath572 & ... & @xmath573 & @xmath574 & @xmath575 & @xmath576 & @xmath577 & @xmath578 & @xmath579 & @xmath580 + @xmath581 & ... & @xmath582 & @xmath583 & @xmath584 & @xmath585 & @xmath586 & @xmath587 & @xmath588 & @xmath589 + @xmath590 & ... & @xmath591 & @xmath592 & @xmath593 & @xmath594 & @xmath595 & @xmath586 & @xmath596 & @xmath597 + @xmath322 & ... & @xmath598 & @xmath599 & @xmath600 & @xmath601 & @xmath602 & @xmath603 & @xmath604 & @xmath597 + since the factor @xmath552 increase with the spectral index , as seen from table [ tab3 ] , this factor also increase with the smoothing length for cdm model on scales of interest . on the other hand , the factor @xmath49 is a decreasing function of the smoothing length . as a result , the genus asymmetry parameter is not a monotonic function of the smoothing length . for example , the genus asymmetry parameter in the @xmath605 , @xmath606 cdm model is approximately @xmath607 on wide range of smoothing length of @xmath608 . from the second order formula for the genus of equation ( [ eq2 - 207c ] ) , the genus at the troughs is proportional to @xmath609 . therefore , the fraction of the deviation from the gaussian prediction at @xmath557 is @xmath610 . when the genus asymmetry parameter is @xmath607 , this fraction is about @xmath611 . thus , the perturbation theory predicts that the values of genus at positive and negative troughs differ @xmath612 for the @xmath613 , @xmath606 cdm model , and so on . this difference increases with the shape parameter @xmath276 . the degree of deviation is qualitatively consistent with the numerical result in fig . 9 of @xcite , although quantitative comparison is still difficult because of the noise in the simulation . the statistics for the velocity field is plotted in figure [ fig2 ] . since the the sign of the skewness parameters is negative in this case , the weakly nonlinear evolution of statistics against the density threshold @xmath133 indicates the sponge - like shift . in terms of the volume - fraction threshold @xmath224 , on the other hand , meat - ball shifts are observed even for relatively bluer spectrum with @xmath614 . the 2d projected galaxy statistics are dependent on the selection function of galaxies and cosmological models . as an example , we assume the apm luminosity function @xcite for galaxies with b band magnitude limit @xmath615 . the differential number count @xmath616 for this sample is plotted in figure [ fig3 ] . the resulting mean redshift is @xmath617 . the relation between the differential number count and the normalized mean number density @xmath405 in comoving coordinates are given by @xmath618 = \frac{\displaystyle h(z ) \frac{dn}{dz}(z ) } { \displaystyle \left\{s_k\left[\chi(z)\right]\right\}^2 \int_0^\infty \frac{dn}{dz } dz } , \label{eq5 - 1}\end{aligned}\ ] ] where @xmath619 are the hubble parameter and the comoving distance at redshift @xmath71 , respectively . once the selection function @xmath405 is given by equation ( [ eq5 - 1 ] ) , the integration of equations ( [ eq4 - 36])([eq4 - 38 ] ) , and the interpolation of tabulated values of @xmath447 in table [ tab6 ] give the skewness parameters . for power - law power spectra , the skewness parameters are given by the simpler integration of equation ( [ 4 - 50 - 2 ] ) and the values of @xmath447 in table [ tab7 ] . in figure [ fig4 ] , 2-dimensional statistics , @xmath105 , @xmath130 , @xmath620 and @xmath621 are plotted with assumed cosmological parameters , @xmath622 , @xmath623 , and apm luminosity function with limiting magnitude 19 . the nonlinear parameter is assumed as @xmath624 . for the cdm model , the shape parameter is assumed as @xmath625 , and we take the smoothing angle as @xmath626 . with this smoothing angle the value @xmath624 corresponds to the normalization @xmath627 . basic features for these 2-dimensional statistics are the same as for 3-dimensional density field , except that the smoothing scale in cdm model we adopt corresponds to smaller scale than in the example of 3-dimensional density field of the figure [ fig1 ] . in this work , we comprehensively presented a basic formalism to treat the statistics of smoothed cosmic fields in perturbation theory . this formalism provides a methodology on evaluating how various statistics deviate from the prediction of simple random gaussian fields . as long as the non - gaussianity is weak , the behavior of the statistics caused by non - gaussianity is predicted by our formalism , which enable us to quantitatively compare the statistical quantities and the source of the non - gaussianity . this method is considered as an extension of the edgeworth expansion , which has been proven to be useful in various fields of research , as long as the non - gaussianity is weak . in this paper , we derive useful formulas and relations focusing on application of the second - order perturbation theory to various cosmic fields . several examples of statistics of cosmic fields in second - order perturbation theory are investigated in datail , including level - crossing statistics , 2d and 3d genus statistics , 2d extrema statistics , and the minkowski functionals , which are extensively used in cosmology . more complicated statistics , such as 2d and 3d density peaks , can also be calculated , although they are more tedius . a particular interest in cosmology of our method is in the application to the cosmic fields . even if the cosmic field was random gaussian at the initial stage , the gravitational evolution induces the non - gaussianity . the gravitational instability is a well - defined process , so that we can evaluate the non - gaussianity without any ambiguity when the evolution remains in the quasi - linear regime provided that the biasing from the non - gravitational process is simple enough on large scales . therefore , we performed the perturbative analysis to obtain the necessary skewness parameters based on the gravitational instability theory . we considered the 3d density field , the 3d velocity field , the 2d projected density field . in the application of second order theory to various statistics of smoothed cosmic fields , three types of skewness parameters are commonly useful , i.e. , @xmath88 , @xmath129 , an @xmath234 . extensive calculations of these parameters for various cosmic fields are one of the new results of this paper . it would be true that other skewness parameters are needed when other complex statistics are considered . such other parameters , if needed , are similarly calculated by the method we outlined in this paper . we find the lowest order deviations from the gaussian predictions of various statistics of smoothed cosmic fields depend only on the differences of the skewness parameters when we use a threshold @xmath224 , which is rescaled by a volume - fraction of the smoothed field . this rescaling makes the lowest deviations much smaller than in the case of direct threshold @xmath133 . this is because three skewness parameters @xmath87 take similar values if it is arisen from the gravitational evolution . for the phenomenological hierarchical model , these parameters are identical . in this case , the weakly nonlinear correction of the statistical quantities in terms of the volume - fraction threshold vanishes . when evaluated by the second - order perturbation theory of density fluctuations , those three types of skewness parameter are still close to each other this fact explains the smallness of the deviations from gaussian predictions of statistical quantities like genus , level - crossing , or minkowski functionals when the rescaled threshold by volume - fraction is used . we discussed small , but detectable deviations from gaussian prediction of the 3d genus curve against rescaled threshold in detail . in the framework of the second - order perturbation theory , a prominent deviation of the genus curve occurs at the two troughs of the curve . relative depths of left and right troughs in the genus curve show slightly meat - ball shift for cdm - like models . we found the degree of this asymmetry is proportional to the combination @xmath559 . we call this factor as a genus asymmetry parameter , which we propose as theory - motivated parameter that characterize the asymmetry of the genus curve . the genus asymmetry parameter can observationally be obtained by fitting the genus curve by hermite polynomials as equation ( [ eq5 - 0 ] ) . qualitative comparison with the numerical simulations in literatures suggests that such asymmetry can actually be observable . we have not estimated over what dynamic range in nonlinearity parameter like @xmath49 , @xmath628 , or @xmath629 are each perturbative expressions valid . this estimation requires a systematic comparison with large n - body simulations , which is beyond the scope of this paper , and will be given in a subsequent paper of the series . in principle , any order in the perturbation theory can be calculated as further as one would like . although the computation of the higher - order theory becomes more and more tedious , the necessity of the comparison with large - scale cosmological observations is a good reason to perform such computation as further as we can . one of the spectacular example of the detailed comparison between perturbation theory and observations is the fine structure constant in quantum electrodynamics ( e.g. , * ? ? ? our analysis in this paper will be extended to the third - order perturbation theory in a subsequent paper of the series . the present time is in an unique decade when the observations of cosmic fields are in unforeseen progress , like large - scale redshift surveys , detailed mapping of cmb fluctuations , gravitational lensing surveys , and so forth . statistics of smoothed density field with higher - order perturbation theory will provide an unique method to analyze those high - precision data . the precision cosmology is undoubtedly providing clues to unlock the door to the origin of the universe . i would like to thank m. kerscher and b. jain for discussions . i wish to acknowledge support from jsps postdoctoral fellowships for research abroad , and from the ministry of education , culture , sports , science , and technology , grant - in - aid for encouragement of young scientists , 13740150 , 2001 . in this appendix , we give gaussian integrals which are useful in this paper . in the following , @xmath630 is the hermite polynomials , @xmath631 and we further employ the notation @xmath632 several hermite polynomials are @xmath633 the hermite polynomial at zero is given by @xmath634 we generalize the above definition of @xmath635 to the case @xmath636 by interpreting @xmath637 as the appropriate gamma function so that @xmath638 , @xmath639 , etc . for example , @xmath640 , @xmath641 , @xmath642 , @xmath643 , and so forth . in this appendix , we give useful gaussian averages @xmath644 of equation ( [ eq1 - 12 - 1 ] ) for normalized cosmic field @xmath18 defined by equation ( [ eq1 - 3 ] ) and its spatial derivatives @xmath645 . in the following , @xmath646 represents any one of the components @xmath78 . first , concerning @xmath18 , @xmath647 second , concerning @xmath646 , in the notation of equation ( [ eqa1 - 3 ] ) , @xmath648 the correlation functions on projected sky is expressible by the 3-dimensional correlation functions . the explicit relation for the two - point correlation function is given by limber s equation @xcite . the fourier - space version of limber s equation is given by kaiser @xcite and is somewhat simpler . his argument was generalized to higher - order correlation functions and their fourier transforms @xcite . since the higher - order limber s equation in fourier space was discussed in the context of weak lensing field in literatures , here we review the derivation in a way more useful in this paper . let 2d projected field @xmath9 be the projection of a time - dependent 3d field @xmath649 along a light - cone : @xmath650 where @xmath651 is some radial weighting function and @xmath401 is the radial comoving distance , and @xmath409 is the conformal time at the observer . the past light - cone of the observer is specified by the equation @xmath652 . the comoving angular distance @xmath653 is defined by equation ( [ eq4 - 29 ] ) . in the following , we explicitly derive the relation for 3-point correlation function , and bispectrum . from the limber s equation , power spectrum is already derived by @xcite : @xmath654 where @xmath655 and @xmath656 is the power spectrum of fields @xmath9 and @xmath55 , respectively . we generalize this equation to the one for 3-point statistics below . the generalization of the following derivation to higher - order statistics is straightforward . the angular 3-point correlation function @xmath657 of @xmath9 is @xmath658 where @xmath659 is the spatial 3-point correlation function of the field @xmath55 with the 3-point configuration @xmath660 at conformal time @xmath420 . according to the spirit of the limber s equation , we assume that @xmath661 is slowly varying compared to the scale of the fluctuations of interest and also that these fluctuations occur on a scale much smaller than the curvature scale . the equation ( [ eqa2 - 3 ] ) is the generalization of the limber s equation to higher - order correlation functions . now we transform equation ( [ eq2 - 13 ] ) to obtain the 2d bispectrum . we use the following convention of the fourier transforms @xmath662 the bispectrum @xmath663 of the 2d field @xmath9 , and @xmath664 of the 3d field @xmath55 are defined by @xmath665 where @xmath666 and @xmath667 . the dirac s delta function comes from the translational invariance of statistics . from the relations , @xmath668 the fourier transform of the equation ( [ eqa2 - 3 ] ) reduces to a simple equation , @xmath669 this is a bispectrum version of the limber s equation and the generalization of the kaiser s equation for power spectrum ( [ eqa2 - 2 ] ) . in table [ tab9 ] , the quantites used in this paper are listed . ccc @xmath670 & ensemble average & ( [ eq1 - 12a ] ) + @xmath644 & ensemble average by gaussian field & ( [ eq1 - 12 - 1 ] ) + @xmath481 & 2d spectral parameter & ( [ eq4 - 36 ] ) + @xmath353 & time - dependent density parameter & ( [ eq4 - 34 - 1 ] ) + @xmath252 & density parameter & ( [ eq4 - 34 - 1 ] ) + @xmath18 & normalized field of @xmath9 & ( [ eq1 - 3 ] ) + @xmath671 , etc . & @xmath672 , etc . & ( [ eq1 - 4 ] ) + @xmath673 & spectral parameter & ( [ eq1 - 13 - 5 ] ) + @xmath97 & mass density contrast & ( [ eq4 - 70 ] ) + @xmath674 & 3d density contrast & ( [ eq4 - 31 ] ) + @xmath520 & galaxy number density contrast & ( [ eq4 - 70 ] ) + @xmath62 & @xmath675 , tensor notation & + @xmath64 & linear combination of @xmath18 and @xmath62 & ( [ eq1 - 13 - 3 - 1 ] ) + @xmath78 & @xmath676 , vector notation & + @xmath417 & angular coordinates in a small patch of the sky & ( [ eq4 - 31 ] ) + @xmath513 & local convergence field & ( [ eq4 - 60 ] ) + @xmath354 & time - dependent scaled cosmological constant & ( [ eq4 - 34 - 1 ] ) + @xmath253 & scaled cosmological constant & ( [ eq4 - 34 - 1 ] ) + @xmath133 & threshold by variance & + @xmath677 & threshold by volume fraction & ( [ eq2 - 105 ] ) + @xmath241 & 3d comoving density field & ( [ eq4 - 30 ] ) + @xmath678 & 2d weighted extrema & ( [ eq2 - 24 ] ) + @xmath400 & 2d projected density field & ( [ eq4 - 30 ] ) + @xmath49 & variance of @xmath9 & ( [ eq1 - 2 ] ) + @xmath271 , @xmath679 & spectral parameters & ( [ eq1 - 13 - 3a ] ) , ( [ eq1 - 13 - 3b ] ) + @xmath409 & conformal time at the present & ( [ eq4 - 30 ] ) + @xmath401 & comoving distance & ( [ eq4 - 29 ] ) + @xmath680 & volume of the unit ball in @xmath121 dimensions & ( [ eq2 - 54 ] ) + @xmath681 & vector of spatial derivatives of @xmath18 & ( [ eq1 - 4 ] ) + @xmath682 & bispectrum at the present time & ( [ eq4 - 4b ] ) + @xmath683 & 2d bispectrum & ( [ eq4 - 33b ] ) + @xmath684 & 3d bispectrum at conformal time @xmath420 & ( [ eq4 - 33b ] ) + @xmath685 & integrand of the 2d skewness parameters & ( [ eq4 - 39 ] ) + @xmath686 & integrand of the 2d skewness parameters & ( [ eq4 - 40 ] ) + @xmath687 & contributions to the 2d skewness parameters & ( [ eq4 - 46 ] ) + @xmath349 & linear growth rate & ( [ eq4 - 34a ] ) + @xmath201 & parameter of the perturbation theory & ( [ eq2 - 21 - 1 ] ) + @xmath351 & parameter of the perturbation theory & ( [ eq4 - 24b ] ) + @xmath55 & cosmic field , in general & ( [ eq1 - 12a ] ) + @xmath688 , etc . & @xmath689 , etc . & ( [ eq1 - 13 ] ) + @xmath690 & gauss hypergeometric function @xmath691 & ( [ eq4 - 20 ] ) + @xmath130 & 2d genus statistic & ( [ eq2 - 8 ] ) , ( [ eq2 - 9 ] ) + @xmath149 & 3d genus statistic & ( [ eq2 - 15 ] ) , ( [ eq2 - 17 ] ) + @xmath343 & time - dependent hubble parameter & ( [ eq4 - 23 ] ) + @xmath692 & hubble constant & ( [ eq5 - 2 ] ) + @xmath693 & modified bessel function & ( [ eq4 - 11 ] ) + @xmath694 & fourier counterpart of @xmath25 & ( [ eq1 - 5 ] ) + @xmath695 & integers defined by table [ tab1 ] & ( [ eq2 - 19 ] ) + @xmath183 & spatial curvature & ( [ eq4 - 29 ] ) + @xmath696 & hermite polynomials & ( [ eqa1 - 1 ] ) + @xmath697 & extended hermite polynomial of order @xmath211 & ( [ eqa1 - 2 ] ) + @xmath698 & @xmath314-th order cumulants of @xmath25 & ( [ eq1 - 6])([eq1 - 7d ] ) + @xmath699 & normalized cumulants of @xmath25 & ( [ eq1 - 12 ] ) + @xmath37 & second - order cumulant of @xmath25 & ( [ eq1 - 8 ] ) + @xmath105 & level - crossing statistic & ( [ eq2 - 1a ] ) + @xmath106 & length statistic & ( [ eq2 - 1b ] ) + @xmath107 & area statistic & ( [ eq2 - 1c ] ) + @xmath700 & multivariate gaussian distribution function & ( [ eq1 - 5 ] ) + @xmath244 & multivariate probablitiy distribution function & ( [ eq1 - 5 ] ) + @xmath701 & power spectrum at present & ( [ eq4 - 4a ] ) + @xmath702 & 2d power spectrum & ( [ eq4 - 33a ] ) + @xmath703 & 3d power spectrum at conformal time @xmath420 & ( [ eq4 - 33a ] ) + @xmath249 & linear power spectrum & ( [ eq4 - 7a ] ) + @xmath238 & hierarchical amplitude of the 3-point function & ( [ eq4 - 1 ] ) + @xmath87 , @xmath704 & skewness parameters & ( [ eq1 - 15a])([eq1 - 15d ] ) + @xmath705 & integrand of the skewness parameters & ( [ eq4 - 9 ] ) + @xmath706 & contributions to the skewness parameters & ( [ eq4 - 13 ] ) + @xmath707 & galaxy skewness parameters & ( [ eq4 - 29 ] ) + @xmath708 & comoving angular diameter distance & ( [ eq4 - 29 ] ) + @xmath709 & cdm transfer function & ( [ eq4 - 15 ] ) + @xmath181 & minkowski functionals & ( [ eq2 - 50 - 0 ] ) , ( [ eq2 - 50 ] ) , ( [ eq2 - 52 ] ) , ( [ eq2 - 54 ] ) + @xmath710 & smoothing function in real space & ( [ eq1 - 1 ] ) + @xmath10 & smoothing function in real space & ( [ eq1 - 1 ] ) + @xmath711 & generating function & ( [ eq1 - 5 ] ) + @xmath524 & bias parameter & ( [ eq4 - 70 ] ) + @xmath525 & nonlinear bias parameter & ( [ eq4 - 70 ] ) + @xmath6 & dimension of the sample space & ( [ eq1 - 1 ] ) + @xmath712 & differential number count & ( [ eq5 - 1 ] ) + @xmath9 & smoothed cosmic field , in general & ( [ eq1 - 1 ] ) + @xmath225 & volume fraction on the high - density side & ( [ eq2 - 105 ] ) + @xmath347 & logarithmic derivative of the growth factor & ( [ eq4 - 25 ] ) + @xmath125 & hermite polynomial at zero , @xmath713 , extended to @xmath714 & ( [ eqa1 - 3 ] ) + @xmath314 & selection function per unit comoving volume & ( [ eq4 - 30 ] ) + @xmath311 & index for the power - law power spectrum & ( [ eq4 - 16 ] ) + @xmath73 , @xmath74 , @xmath71 & linear combinations of @xmath18 and @xmath62 s & ( [ eq1 - 13 - 4a])([eq1 - 13 - 4c ] ) + adler , r. j. 1981 , r. k. 1986 , the geometry of random fields ( chichester ; wiley ) balian , r. & schaeffer , r. 1989 , , 220 , 1 bardeen , j. m. , bond , j. r. , kaiser , n. , & szalay , a. s. , 1986 , , 304 , 15 bartelmann , m. & schneider , p. 1999 , to be submitted to physics reports ( astro - ph/9912508 ) baugh , c. m. & efstathiou , g. 1994 , , 270 , 183 beaky , m. m. , scherrer , r. j. & villumsen , j. v. 1992 , , 387 , 443 bernardeau , f. 1992 , , 392 , 1 bernardeau , f. 1994a , , 433 , 1 bernardeau , f. 1994b , , 291 , 697 bernardeau , f. 1995 , , 301 , 309 bernardeau , f. , juszkiewicz , r. , dekel , a. & bouchet , f. r. 1995 , , 274 , 20 bernardeau , f. & kofman , l. 1995 , , 443 , 479 bernardeau , f. , van waerbeke , l. & mellier , y. 1997 , , 322 , 1 bernardeau , f. , colombi , s. , gaztanaga , e. , & scoccimarro , r. 2002 , , 367 , 1 bharadwaj , s. , sahni , v. , sathyaprakash , b. s. , shandarin , s. f. , & yess , c. 2000 , , 528 , 21 bond , j. r. & efstathiou , g. 1987 , , 226 , 655 bouchet , f. r. , juszkiewicz , r. , colombi , s. & pellat , r. 1992 , , 394 , l5 buchalter , a. , kamionkowski , m. , & jaffe , a. h. 2000 , , 530 , 36 buchalter , a. & kamionkowski , m. 1999 , , 521 , 1 canavezes , a. et al . 1998 , , 297 , 777 carroll , s. m. , press , w. h. & turner , e. l. 1992 , , 30 , 499 cayon , l. & smoot , g. 1995 , , 452 , 487 cen , r. 1998 , , 509 , 494 chodorowski , m. j. & lokas , e. l. 1997 , , 287 , 591 coles , p. 1988 , , 234 , 509 coles , p. 1990 , , 243 , 171 coles , p. & barrow , j. d. 1987 , , 228 , 407 coles , p. & plionis , m. 1991 , , 250 , 75 coles , p. , moscardini , l. , plionis , m. , lucchin , f. , matarrese , s. & messina , a. 1993 , , 260 , 572 colley , w. n. 1997 , , 489 , 471 colley , w. n. , gott , j. r. i. , weinberg , d. h. , park , c. and berlind , a. a. 2000 , , 529 , 795 croft , r. a. c. & efstathiou , g. 1994 , , 267 , 390 crofton , m. w. 1868 , phil . lond . a , 158 , 181 davies , a. & coles , p. 1993 , , 260 , 553 dekel , a. & lahav , o. 1999 , , 520 , 24 doroshkevich , a. g. 1970 , astrophysica , 6 , 320 fabbri , r. & torres , s. 1996 , , 307 , 703 fry , j. n. 1984 , , 279 , 499 fry , j. n. & gaztanaga , e. 1993 , , 413 , 447 gabrielli , a. , labini , f. s. & durrer , r. 2000 , , 531 , l1 goroff , m. h. , grinstein , b. , rey , s .- j . & wise , m. b. 1986 , , 311 , 6 gott , j. r. , melott , a. l. & dickinson , m. 1986 , , 306 , 341 gott , j. r. , weinberg , d. h. & melott , a. l. 1987 , , 319 , 1 gott , j. r. , mao , s. , park , c. & lahav , o. 1992 , , 385 , 26 gott , j. r. i. et al . 1989 , , 340 , 625 gott , j. r. , park , c. , juszkiewicz , r. , bies , w. e. , bennett , d. p. , bouchet , f. r. & stebbins , a. 1990 , , 352 , 1 hamilton , a. j. s. , gott , j. r. & weinberg , d. 1986 , , 309 , 1 heavens , a. f. & sheth , r. k. 1999 , , 310 , 1062 jain , b. & bertschinger , e. 1994 , , 431 , 495 jain , b. & van waerbeke , l. 2000 , , 530 , l1 juszkiewicz , r. 1981 , , 197 , 931 juszkiewicz , r. , sonoda , d. h. & barrow , j. d. 1984 , , 209 , 139 juszkiewicz , r. , bouchet , f. r. & colombi , s. 1993 , , 412 , l9 juszkiewicz , r. , weinberg , d. h. amsterdamski , p. , chodorowski , m. & bouchet , f. r. 1995 , , 442 , 39 kaiser , n. 1984 , , 284 , l9 kaiser , n. 1998 , , 498 , 26 kerscher , m. , schmalzing , j. , buchert , t. & wagner , h. 1998 , , 333 , 1 kinoshita , t. 1996 , rep . phys . , 59 , 1459 kogut , a. , banday , a. j. , bennett , c. l. , hinshaw , g. , lubin , p. m. & smoot , g. f. 1995 , , 439 , l29 lahav , o. , lilje , p. b. , primack , j. r. & rees , m. j. 1991 , , 251 , 128 laveday , j. , peterson , b. a. , efstathiou , g. & maddox , s. j. 1992 , , 390 , 338 lightman , a. p. & schechter , p. l. 1990 , , 74 , 831 limber , d. n. 1954 , , 119 , 655 lokas , e. l. 1998 , , 296 , 491 okas , e. l. , juszkiewicz , r. , weinberg , d. h. & bouchet , f. r. 1995 , , 274,730 ma , s .- k . 1985 , statistical mechanics ( philadelphia ; world scientific ) makino , n. , sasaki , m. & suto , y. 1992 , , 46 , 585 & suto , y. 1992 , mann , r. g. , heavens , a. f. & peacock , j. a. 1993 , , 263 , 798 martel , h. 1991 , , 377 , 7 matsubara , t. 1994 , , 434 , l43 matsubara , t. 1995a , , 101 , 1 matsubara , t. 1995b , prog . ( lett . ) , 94 , 1151 matsubara , t. 1996 , , 457 , 13 matsubara , t. 1999 , , 525 , 543 matsubara , t. & suto , y. 1996 , , 460 , 51 matsubara , t. & yokoyama , j. 1996 , , 463 , 409 mecke , k. r. , buchert , t. & wagner , h. 1994 , , 288 , 697 mellier , y. 1999 , ara&a , 37 , 127 melott , a. l. , cohen , a. p. , hamilton , a. j. s. , gott , j. r. & weinberg , d. h. 1989 , , 345 , 618 moore , b. et al . 1992 , , 256 , 477 morse , m. & cairns , s. 1969 , critical point theory in global analysis and differential topology ( new york ; academic press ) naselsky , p. d. & novikov , d. i. 1998 , , 507 , 31 park , c. & gott , j. r. 1991 , , 378 , 457 park , c. , gott , j. r. & da costa , l. n. 1992 , , 392 , l51 park , c. , colley , w. n. , gott , j. r. i. , ratra , b. , spergel , d. n. and sugiyama , n. 1998 , , 506 , 473 peebles , p. j. e. 1980 , the large - scale structure of the universe ( princeton ; princeton university press ) plionis , m. , valdarnini , r. & coles , p. 1992 , , 258 , 114 plionis , m. & valdarnini , r. 1995 , , 272 , 869 protogeros , z. a. m. & weinberg , d. h. 1997 , , 489 , 457 rhoads , j. e. , gott , j. r. i. & postman , m. 1994 , , 421 , 1 ryden , b. s. 1988 , , 333 , l41 ryden , b. s. 1988 , , 100 , 1360 ryden , b. s. , melott , a. l. , craig , d. a. , gott , j. r. , iii , wenberg , d. h. , scherrer , r. j. , bhavsar , s. p. & miller , j. m. 1989 , , 340 , 647 sahni , v. , sathyaprakash , b. s. , & shandarin , s. f. 1998 , , 495 , l5 sahni , v. , sathyaprakash , b. s. , & shandarin , s. f. 1997 , , 476 , l1 sathyaprakash , b. s. , sahni , v. , & shandarin , s. 1998 , , 508 , 551 sazhin , m. v. 1985 , , 216 , 25p scherrer , r. j. & bertschinger , e. 1991 , , 381 , 349 schmalzing , j. & buchert , t. 1997 , , 482 , l1 schmalzing , j. & gorski , k. m. 1998 , , 297 , 355 schmalzing , j. , buchert , t. , melott , a. l. , sahni , v. , sathyaprakash , b. s. & shandarin , s. f. 1999 , , 526 , 568 schmalzing , j. and diaferio , a. 2000 , , 312 , 638 schneider , p. , ehlers , j. , falco , . e. e. 1992 , gravitational lenses , ( springer ) scoccimarro , r & frieman , j. a. 1996a , , 105 , 37 scoccimarro , r & frieman , j. a. 1996b , , 473 , 620 scoccimarro , r. ; . , zaldarriaga , m. , & hui , l. 1999 , , 527 , 1 seto , n. 2000 , , 537 , in press ( astro - ph/0002315 ) smoot , g. f. , tenorio , l. , banday , a. j. , kogut , a. , wright , e. l. , hinshaw , g. & bennett , c. l. 1994 , , 437 , 1 springel , v. , and 10 colleagues 1998 , , 298 , 1169 suto , y. & sasaki , m. 1991 , physical review letters , 66 , 264 taruya , a. & soda , j. 1999 , , 522 , 46 tomita , h. 1986 , progr . , 76 , 952 torres , s. 1994 , , 423 , l9 torres , s. , cayon , l. , martinez - gonzalez , e. & sanz , j. l. 1995 , , 274 , 853 van waerbeke , l. 2000 , , 313 , 524 vishniac , e. t. 1983 , , 203 , 345 vittorio , n. & juszkiewicz , r. 1987 , , 314 , l29 vogeley , m. s. , park , c. , geller , m. j. , huchra , j. p. & gott , j. r. i. 1994 , , 420 , 525 watanabe , t. , matsubara , t. & suto , y. 1994 , , 432 , 17 winitzki , s. & kosowsky , a. 1997 , new astronomy , 3 , 75 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the phase - space formulation of quantum mechanics has received a renewed attention in the last decades @xcite because it allows to employ notions and tools of classical physics in the quantum realm . the central object in this approach is _ wigner s function _ @xmath0 which is merely a phase - space @xmath1 representation of the density matrix @xmath2 . in a closed system , the dynamical equation for the wigner function is equivalent to _ schrdinger s ( or von neumann ) equation _ while , in the @xmath3 limit , it recovers _ liouville s equation _ for the phase - space distribution in classical mechanics . another remarkable property of the wignerian representation is the _ average property_. the expectation value of a quantum operator @xmath4 is obtained from the corresponding classical variable @xmath5 ( related with @xmath4 by weyl s rule ( * ? ? ? * ch . 1 ) ) by a prescription analogous to the classical one , namely @xmath6 thus , in spite of not being necessarily positive , the wigner function can be considered as a sort of quantum - mechanical `` distribution '' . besides , the marginal distributions of @xmath7 give the true quantum probabilities of @xmath8 , as @xmath9 , or of @xmath10 , as @xmath11 . therefore , the wigner formalism provides a natural quantum - classical connection . this has been specially exploited in the search for quantum analogues of various classical effects ( e.g. , chaos ) or in the problem of the fuzzy borders between the quantum and classical worlds @xcite . in contact with the surrounding medium , the ( open ) system experiences dissipation , fluctuations , and decoherence @xcite . typically one is interested in a system consisting of a few relevant degrees of freedom coupled to its environment , or bath , which has a very large number of degrees of freedom ( e.g. , phonons , photons , nuclear spins , etc . ) . the system is not necessarily microscopic , but it can be a mesoscopic system described by some collective variables which under appropriate conditions can behave quantum mechanically @xcite . because of the genericity of these situations and the fundamental issues raised ( e.g. , approach to thermal equilibrium , dissipation in tunnelling and coherence , measurement problem ) , the study of _ quantum dissipative systems _ is of interest in several areas of physics and chemistry @xcite . the dynamics of an open system can in many cases be formulated in terms of a _ quantum master equation _ for the ( reduced ) density matrix . in the wigner representation this can be compactly written as @xmath12 , with @xmath13 a certain evolution operator . taking the classical limit , the master equation for a particle of mass @xmath14 in a potential @xmath15 goes over the klein kramers equation @xcite @xmath16 \wf % % % % % % % % % % % % % % % % % % % \;,\ ] ] which is simply a _ planck equation _ in phase space . the first two terms ( poisson bracket ) generate the classical reversible evolution ( liouville equation ) . the last term ( `` collision operator '' ) accounts for irreversible effects due to the coupling to the environment ( dissipation and fluctuations ) . the damping parameter @xmath17 measures the strengh of the coupling to the bath , which is at temperature @xmath18 . it is worth recalling that phase - space dynamics also includes problems reducible to `` mechanical '' analogues : josephson junctions , certain electrical circuits , chemical reactions , etc . a suitable non - perturbative technique to solve classical fokker planck equations of systems with few degrees of freedom is the _ continued - fraction method _ @xcite . this is a special case of the expansion into complete sets ( grad s ) method to solve kinetic equations in statistical mechanics @xcite . in brief , the distribution is expanded into a series of orthogonal polynomials and , as a result , the kinetic equation is transformed into an _ infinite set of coupled equations _ for the expansion coefficients @xmath19 . approximate solutions are obtained by truncating the hierarchy of equations at various levels . to get manageable expressions , however , the truncation needs to be performed at a low level . in the continued - fraction method , instead of truncating directly , one seeks for a basis in which the range of index coupling is short ( ideally , the equation for @xmath19 involves @xmath20 , @xmath19 , and @xmath21 ) . then , the ( differential ) recurrence relations among the @xmath19 can be solved by iterating a simple algorithm , the structure of which is like that of a continued fraction ( [ app : rr - cf ] ) . for the klein kramers equation ( [ kk ] ) the tridiagonal chain of coupled equations for the @xmath19 is called the _ brinkman hierarchy _ it has been solved by continued - fraction methods to study classical particles subjected to dissipation and fluctuations @xcite . the method has been also extended to rotational brownian motion problems involving classical spins and dipoles ( vd . @xcite and references therein ) . to deal with quantum dissipative systems , however , is a more delicate task @xcite . to begin with , phenomenological quantisation of dissipative systems poses fundamental problems ( e.g. , with the uncertainty and superposition principles ) . a rigorous route is to model the bath in a simple way , quantise it together with the system , and eventually trace over the environmental variables . however , the resulting theoretical frameworks are technically involved . _ quantum master equations _ , except for simple cases , are difficult to solve . an alternative is provided by _ quantum state diffusion _ methods ( vd . @xcite ) , where stochastic evolution equations for state vectors in hilbert space are introduced as computational tools . nevertheless , its implementation seems to be restricted to systems with discrete spectrum ( e.g. , oscillators , 2-state systems , etc . ) . similarly , _ quantum langevin equations _ @xcite are of limited use beyond nearly harmonic systems . for an arbitrary system , there exist exact _ path - integral _ expressions for the evolution of the density matrix ( involving the so - called _ influence functional _ that incorporates environmental effects ) @xcite . however , those expressions are difficult to evaluate , even numerically , because the propagating function is highly oscillatory , rendering numerical methods unstable at long times . finally , _ quantum monte carlo _ simulations can in principle be used . nevertheless , in spite of ongoing progress ( vd . @xcite ) , they are computationally complex and suffer from the ( dynamical ) sign problem . this situation strongly motivates the development of _ alternative methods _ for quantum dissipative systems . inspired on their suitability for classical systems , continued - fraction techniques have been developed for a number of problems . shibata and co - workers @xcite applied them to solve a c - number quantum fokker planck equation for a spin in a dissipative environment . vogel and risken @xcite employed continued - fraction methods to solve master equations in quantum nonlinear optics ( vd.also @xcite ) . recently @xcite this method has been extended to study genuine phase - space problems within the wigner formalism . a generalisation of the brinkman hierarchy for quantum master equations of the caldeira leggett type was presented . it was shown that the continued - fraction method for the classical problem can , in principle , be adapted to solve this hierarchy , yielding a promising technique to study several classes of nonlinear quantum systems subjected to environmental effects . in this article we first present the details necessary for the derivation of the quantum hierarchies and discuss their solution by continued fractions . the approach is then implemented for the problem of quantum brownian motion in periodic potentials ( a demanding problem with ( partly ) continuous spectrum ) . both the cosine and simple ratchet potentials ( lacking inversion symmetry ) are considered . for the former a number of previous results are recovered ( so testing the method ) and extended , while the physical interpretations are revisited within the wigner formalism . for particles in ratchet potentials we study the effects of finite damping ( inertia ) on the rectified velocities . the phenomenology is interpreted in terms of the interplay of thermal hopping , overbarrier wave reflection , and tunnelling . results for non - equilibrium dynamics under oscillating forcing are also discussed , which show the combined effect of quantum phenomena , thermal activation , and dissipation in a nonlinear system . a number of technical issues are consigned to the appendices . we start from the following generic form for the quantum master equation in the wigner representation @xcite : @xmath22 \wf % % % % % % % % % % % % % % % % % % \;. \;\ ] ] the first three terms , identifying @xmath23 , give the classical klein kramers equation ( [ kk ] ) . the mixed diffusion term @xmath24 is heuristically related with the colour of the quantum noise @xcite . the series of derivatives @xmath25 ( wigner moyal term ) gives the quantum contribution to the unitary evolution of the closed system . conditions under which this series can be truncated are sometimes discussed , to avoid dealing with an infinite - order partial differential equation . however , we shall fully keep this term because it can be specially important in nonlinear systems @xcite . conditions of validity for the master - equation description are discussed in @xcite . equations of the type of ( [ wkk : dxp ] ) can be derived modelling the particle surroundings as a _ bath of oscillators _ representing the normal modes of the environment . however , a number of assumptions are typically involved like semiclassical or high - temperature bath ( rendering the kernels of the influence functional local in time ) , weak system - bath coupling ( born markov approximations ) , etc . then , one would heuristically expect equation ( [ wkk : dxp ] ) being valid for small enough @xmath26 . notwithstanding this , the _ structure _ of the equation seems to be quite generic . thus , a quantum master equation recently derived for strong coupling @xcite , and valid for all @xmath26 , involves terms similar to those in ( [ wkk : dxp ] ) [ with @xmath8-dependent coefficients and renormalised @xmath15 ] . besides , the phase - space representation of the celebrated lindblad master equation is obtained by simply adding to ( [ wkk : dxp ] ) terms of the form @xmath28 and @xmath29 @xcite . most of these terms can be readily incorporated in the treatment below ( indeed , the term @xmath24 , absent in the original caldeira leggett equation , is included to illustrate this ) . the same , in principle , would apply to possible extensions of the quantum master equation ( [ wkk : dxp ] ) . first it is convenient to introduce appropriate scaled units . these are based on a characteristic length @xmath30 ( e.g. , the distance between the minima in a double - well oscillator , the period in a periodic potential ) and a characteristic energy @xmath31 ( e.g. , a barrier height ) . then one scales frequencies by @xmath32 , time by @xmath33 , forces by @xmath34 , the action by @xmath35 , etc . on the other hand , @xmath36 ( @xmath37 in a high-@xmath18 bath ) is handled as an effective temperature @xmath38 and we introduce some convenient thermal rescalings via @xmath39 . thus , in the equations @xmath10 is scaled by @xmath40 , the potential and @xmath41 appear divided by @xmath42 , @xmath17 enters in the combination @xmath43 , time is multiplied by @xmath44 , while the thermal de broglie wave length @xmath45 enters divided by @xmath30 . omitting any marks for the scaled quantities ( but keeping in mind specially the thermal rescaling of @xmath46 , @xmath15 and @xmath10 ) , the master equation is simply written as @xmath47 \wf % % % % % % % % % % % % % % % % % % % \ ; , \quad\ ] ] with the coefficient in the quantum sum given by @xmath48 planck s constant @xmath49 is introduced in terms of the characteristic action @xmath35 via the quantum parameter @xmath50 ( denoted @xmath51 in @xcite ) @xmath52 the second relation gives @xmath53 in terms of the kondo parameter @xmath54 , related by the third one with our @xmath50 . the classical limit is approached as @xmath55 , i.e. , letting @xmath56 . let us now introduce a splitting of the evolution operator @xmath13 that will facilitate the calculation of the matrix elements required when expanding @xmath57 into complete sets . on inspecting equations ( [ wkk : scaled ] ) and ( [ kappa : s ] ) , we see that extending the quantum sum down to @xmath58 , we obtain the part @xmath59 of the classical term . thus , we can decompose @xmath13 into the irreversible , kinetic , and potential parts : @xmath60 % % % % % % % % % % % % % % % % % % \lkin & = & -\big(\p-\bdxp\pp\big)\ , \px \\[0.5ex ] % % % % % % % % % % % % % % % % % % \lv & = & \sumiqo \qcoef^{(\iq ) } \ , \bv^{(2\iq+1 ) } \ , \pp^{(2\iq+1 ) } \end{array } \right . % % % % % % % % % % % % % % % % % % \;.\ ] ] this natural splitting has the advantage of dealing with @xmath15 ( the most problem dependent part ) in a unified way . besides , we have grouped the term @xmath61 with the kinetic term @xmath62 because they are structurally similar . in the expansion into complete sets approach to solve kinetic equations , the distribution is expanded in an orthonormal basis @xmath63 and the coupled equations for the coefficients @xmath64 derived @xcite . for the klein kramers equation ( [ kk ] ) one uses hermite functions of @xmath10 as basis ( * ? ? ? * sec 4.4 ) @xmath65^{-1/2 } % % % % % % % % % % % % % % % % % % \;.\ ] ] one starts with the expansion in the momentum because , while the @xmath10 dependence of the hamiltonian is fully specified ( kinetic energy @xmath66 ) , the potential part @xmath15 depends on the problem . thus , explicit manipulations can be done on the parts involving @xmath10 , which are valid for any system . on the other hand , the hermite basis has a number of advantages @xcite ; not the least is the handling of the derivatives @xmath67 in the dynamical equation by means of the associated creation and annihilation operators @xmath68 and @xmath69 ( this will prove very convenient in the quantum case ) . for the klein kramers equation the resulting equations for the expansion coefficients @xmath64 are called the _ brinkman hierarchy _ @xcite . this plays an important rle in classical systems , both for analytical treatments derivations of @xmath70 expansions , etc. and to develope efficient non - perturbative numerical methods . in fact , it has the structure of a @xmath71-term recurrence relation which , after expansion in the position basis , can be solved by continued fractions . in what follows we shall proceed in the quantum case following as closely as possible the steps of the classical limit . we start expanding the wigner function as @xmath72 where we have extracted the boltzmann - type factor @xmath73 . this involves the auxiliary parameter @xmath74 and potential @xmath75 , which is frequently chosen proportional to @xmath76 , i.e. , @xmath77 . the results should not depend on the @xmath78 or @xmath79 used , but these may ( i ) put parts of the evolution operator @xmath13 in self - adjoint form and/or ( ii ) improve the stability and convergence of the eventual numerical implementation . from the orthonormality of the @xmath80 , the expansion `` coefficients '' can be written as @xmath81 . then , differentiating with respect to @xmath46 and using the evolution equation @xmath12 , we get the dynamical equations for @xmath64 @xmath82 to get the matrix elements @xmath83 , which are still operators on the @xmath8-dependence of @xmath84 , one needs the @xmath85 transform of @xmath13 . this can be split as @xmath86 , with @xmath87 , @xmath88 , and @xmath89 given by ( [ wkk : lir : lkin : lv ] ) . the transformation of these operators is done in [ app : lb ] , taking advantage of results for _ normal ordering _ of @xmath90 and @xmath91 ( which we use to express the @xmath67 and @xmath10 s in @xmath13 ) . then , with the @xmath92 , @xmath93 , and @xmath94 obtained we compute their matrix elements between the @xmath95 , getting the @xmath83 . the details of this part of the calculation are given in [ app : qnm ] . introducing the results obtained for @xmath83 into @xmath96 one gets the following _ quantum ( brinkman ) hierarchy _ @xmath97 } \ , \big [ \g_{\ip}^{\iq,- } \ , \bv^{(2\iq+1 ) } \big ] \ , \ec_{\ip-(2\iq+1 ) } + \sqrt{(\ip-1)\ip}\ ; \gammamm \ , \ec_{\ip-2 } \nonumber\\ % % % % % % % % % % % % % % % % % % & & { } + \sqrt{\ip}\ ; \dm \ec_{\ip-1 } + \gammao_{\ip } \ , \ec_{\ip } + \sqrt{\ip+1}\ ; \dp \ec_{\ip+1 } \\ % % % % % % % % % % % % % % % % % % & & { } + \sqrt{(\ip+1)(\ip+2)}\ ; \gammapp \ , \ec_{\ip+2 } + \sum_{\iq=0}^{\infty } \big [ \g_{\ip+(2\iq+1)}^{\iq,+ } \ , \bv^{(2\iq+1 ) } \big ] \ , \ec_{\ip+(2\iq+1 ) } \nonumber % % % % % % % % % % % % % % % % % % \;.\end{aligned}\ ] ] the auxiliary ( damping ) parameters introduced read @xmath98 % [ ( \etap-\etapro)+2\ip(\etab-\etapro ) ] % % % % % % % % % % % % % % % % % % % \;,\ ] ] which involve the following @xmath78-related parameters ( note the sign exchange ) @xmath99 in equation ( [ qbh ] ) , the operators on the @xmath8 dependences are @xmath100 here @xmath101 , @xmath102 is the confluent hypergeometric ( kummer ) function ( * ? ? ? * ch . 13.6 ) , and @xmath103 is given by equation ( [ kappa : s ] ) . in is important to note that the @xmath104 act only on @xmath15 not on the @xmath105 ( [ app : lb ] ) . the quantum hierarchy ( [ qbh ] ) is equivalent to the caldeira leggett master equation ( [ wkk : dxp ] ) . previously , several hierarchies had been derived @xcite , but the treatment was for closed ( hamiltonian ) systems @xcite or involved unsuitable bases like @xmath106 @xcite . in contrast , our hierarchy of _ hermitian moments _ constitutes a direct quantum generalisation of the brinkman hierarchy . to verify this , note that in the classical case only the @xmath58 terms survive in the sums of ( [ qbh ] ) ( [ app : qnm ] ) . then , setting the usual @xmath107 , equation ( [ qbh ] ) reduces to @xmath108 % \;,\ ] ] which is indeed the celebrated hierarchy associated to the klein kramers equation . for potentials obeying @xmath109 , @xmath110 , the quantum hierarchy ( [ qbh ] ) is a recurrence relation with coupling to a finite number of terms . examples include the harmonic oscillator @xmath111 and the double - well ( duffing ) oscillator @xmath112 . finite - coupling recurrences are always reducible to @xmath71-term vector recurrences ( [ app : rr - cf ] ) and solvable by continued fractions , as in the classical case . however , for non - polynomial @xmath15 this approach is prevented by the infinite coupling range in the index @xmath113 , consequence of the infinite series of @xmath10-derivatives @xmath114 in the wigner moyal term . in what follows we shall exploit the expansion in the position basis to show how this problem can be circumvented . recall that the coefficients @xmath64 are still functions of @xmath8 . inserting their expansion in an orthonormal basis @xmath115 into the quantum hierarchy ( [ dcdt : qtot : lbar ] ) , we get the generic form @xmath116 with matrix elements @xmath117 . to put ( [ qbh ] ) into the above form we just need to ( i ) substitute the operators ( and @xmath8-dependent coefficients ) by their matrix elements @xmath118 , ( ii ) attach position superscripts to the @xmath64 , and , finally , ( iii ) sum over them . on so doing we get the _ expanded quantum hierarchy _ @xmath119_{\ix\ixp}\ , \ec_{\ip-(2\iq+1)}^{\ixp } + \sqrt{(\ip-1)\ip } \sum_{\ixp } \gammamm \ , \delta_{\ix\ixp } \ , \ec_{\ip-2}^{\ixp } \nonumber\\ % % % % % % % % % % % % % % % % % % \fl & { } + \sqrt{\ip } \sum_{\ixp } \dm_{\ix\ixp}\ , \ec_{\ip-1}^{\ixp } + \sum_{\ixp } \gammao_{\ip}\ , \delta_{\ix\ixp } \ , \ec_{\ip}^{\ixp } + \sqrt{\ip+1 } \sum_{\ixp } \dp_{\ix\ixp}\ , \ec_{\ip+1}^{\ixp } \\ % % % % % % % % % % % % % % % % % % \fl & { } + \sqrt{(\ip+1)(\ip+2 ) } \sum_{\ixp } \gammapp \ , \delta_{\ix\ixp } \ , \ec_{\ip+2}^{\ixp } + \sumiqo \sum_{\ixp } \big [ \g_{\ip+(2\iq+1)}^{\iq,+ } \bv^{(2\iq+1 ) } \big]_{\ix\ixp}\ , \ec_{\ip+(2\iq+1)}^{\ixp } \nonumber % % % % % % % % % % % % % % % % % % \;.\end{aligned}\ ] ] considering that @xmath120 and @xmath77 , we see that to get all matrix elements involved we just need those of @xmath121 and of @xmath122 , namely @xmath123 because the @xmath104 only increase the order of the potential derivatives . the expanded hierarchy has the form of a system of ordinary differential equations , @xmath124 . if @xmath125 and @xmath126 are some large truncation indices for the bases @xmath127 and @xmath115 , we have a problem with @xmath128 equations . one may be tempted to try direct runge kutta integration or numerical diagonalization of the associated @xmath129 matrix . however , the dimensions involved are typically very large , so that it is worth to pursue a continued - fraction treatment . as mentioned before the @xmath10-recurrence has a finite coupling range for polynomial potentials . for other potentials , with an appropriate choice of the basis @xmath115 , the matrix elements ( [ matr - elem : der : v ] ) may vanish when the second index @xmath130 lays at a certain `` distance '' of the first @xmath131 . then , the expanded hierarchy would be a recurrence relation in the position indices with a finite coupling range , also amenable for a continued - fraction treatment . to proceed in these cases , we just need to transform the two - index extended hierarchies into ordinary recurrence problems ( i.e. , with one index ) . in order to apply the continued - fraction method when finite coupling range in one of the indices is attained , we need to convert first the two - index recurrences into ordinary recursions . in the resulting one - index recurrences the coefficients will be matrices acting on appropriate vectors formed with the @xmath132 . the generic expanded form ( [ dcdt : generic ] ) can be converted into an _ one - index _ recurrence relation in the momentum index introducing the following vectors and matrices @xmath133 where @xmath134 . that is , for a fixed @xmath113 one forms the column vector with all the @xmath132 with @xmath135 ( this index may take positive and negative values ) . similarly , for given @xmath136 we build the matrix @xmath137 with all indices @xmath138 . before giving explicit expressions for the matrix coefficients we introduce the following notation : @xmath139 , @xmath140 , @xmath141 , and @xmath142 . then , the matrix quantum hierarchy ( [ dcdt : vectors : matrices : p ] ) can be written as @xmath143 % % % % % % % % % % % % % % % % % % & & { } + \mq_{\ip}^{- } \mc_{\ip-1 } + \mq_{\ip } \mc_{\ip } + \mq_{\ip}^{+ } \mc_{\ip+1 } \\[+0.5ex ] % % % % % % % % % % % % % % % % % % & & { } + \mq_{\ip}^{++ } \mc_{\ip+2 } + { \textstyle\sum}_{\iq\ge1 } \mq_{\ip}^{+(2\iq+1 ) } \mc_{\ip+(2\iq+1 ) } \nonumber % % % % % % % % % % % % % % % % % % \;.\end{aligned}\ ] ] we have accounted for @xmath144 , @xmath145 [ vd . equation ( [ qbh : pre - matrix : i ] ) ] and incorporated the @xmath58 terms into the central @xmath146 ones . the matrix coefficients explicitly read @xmath147_{\ix\ixp}\ , \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ip}^{-- } \big)_{\ix\ixp } & = & -\sqrt{(\ip-1)\ip}\ ; \gammamm\ , \delta_{\ix\ixp } \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ip}^{- } \big)_{\ix\ixp } & = & - \sqrt{\ip}\ , \dm_{\ix\ixp } - \big [ \g_{\ip}^{0,- } \bv ' \big]_{\ix\ixp}\ , \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ip } \big)_{\ix\ixp } & = & - \gammao_{\ip}\,\delta_{\ix\ixp } \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ip}^{+ } \big)_{\ix\ixp } & = & - \sqrt{\ip+1}\ , \dp_{\ix\ixp } - \big [ \g_{\ip+1}^{0,+ } \bv ' \big]_{\ix\ixp}\ , \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ip}^{++ } \big)_{\ix\ixp } & = & -\sqrt{(\ip+1)(\ip+2)}\ ; \gammapp\ , \delta_{\ix\ixp } \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ip}^{+(2\iq+1 ) } \big)_{\ix\ixp } & = & - \big [ \g_{\ip+(2\iq+1)}^{\iq,+ } \bv^{(2\iq+1 ) } \big]_{\ix\ixp}\ , % % % % % % % % % % % % % % % % % % \end{array } \right . % \;.\ ] ] for finite coupling range in @xmath113 ( e.g. , in polynomial potentials ) , the hierarchy ( [ qbh : matrix : p ] ) is a recurrence relation to which continued - fraction methods can now be applied . this amounts to convert the underlying large @xmath129 problem into @xmath125 problems with associated @xmath148 matrices [ or @xmath149 . as the recursion `` coefficients '' are matrices , one would need _ matrix continued - fraction _ methods ( [ app : rr - cf ] ) . when the @xmath10-recurrence has an infinite coupling range ( in periodic potentials , the morse potential , etc . ) , we can convert the @xmath150-index hierarchy ( [ dcdt : generic ] ) into matrix form , but introducing recurrences in the position index @xmath131 @xmath151 that is , @xmath152 , with @xmath153 . thus , for a fixed @xmath131 one constructs the column vector with the @xmath132 for all @xmath154 , and for given @xmath155 , the matrix @xmath137 with all @xmath156 . comparing the matrix in ( [ dcdt : vectors : matrices : x ] ) with equation ( [ q : matr - elem : p : i ] ) for @xmath157 , we can construct @xmath158 diagonal by diagonal @xmath159_{\ix\ixp}\ , \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ix\ixp } \big)_{\ip,\ip-2 } & = & -\sqrt{(\ip-1)\ip}\ ; \gammamm\ , \delta_{\ix\ixp } \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ix\ixp } \big)_{\ip,\ip-1 } & = & - \sqrt{\ip}\ , \dm_{\ix\ixp } - \big [ \g_{\ip}^{0,- } \bv ' \big]_{\ix\ixp}\ , \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ix\ixp } \big)_{\ip,\ip } & = & - \gammao_{\ip}\,\delta_{\ix\ixp } \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ix\ixp } \big)_{\ip,\ip+1 } & = & - \sqrt{\ip+1}\ , \dp_{\ix\ixp } - \big [ \g_{\ip+1}^{0,+ } \bv ' \big]_{\ix\ixp}\ , \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ix\ixp } \big)_{\ip,\ip+2 } & = & -\sqrt{(\ip+1)(\ip+2)}\ ; \gammapp\ , \delta_{\ix\ixp } \\ % % % % % % % % % % % % % % % % % % \big ( \mq_{\ix\ixp } \big)_{\ip,\ip+(2\iq+1 ) } & = & - \big [ \g_{\ip+(2\iq+1)}^{\iq,+ } \bv^{(2\iq+1 ) } \big]_{\ix\ixp}\ , % % % % % % % % % % % % % % % % % % \end{array } \right . % \;.\ ] ] all other diagonals being zero @xmath160 , @xmath145 . to recognise better the structure of these matrices , we decompose them into the `` free '' and potential contributions @xmath161 . for @xmath162 , including the kinetic and irreversible terms , we have the pentadiagonal structure ( tridiagonal for @xmath107 ) @xmath163 the part due to @xmath15 , on the other hand , has the following alternate dense structure , reflecting the odd powers of @xmath67 in the wigner moyal term : @xmath164_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{3}^{1,+ } \bv^{(3)}]_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{5}^{2,+ } \bv^{(5)}]_{\ix\ixp } \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-1 [ \g_{1}^{0,- } \bv']_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{2}^{0,+ } \bv']_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{4}^{1,+ } \bv^{(3)}]_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-2 0 \!\!\!\!&\!\!\!\ ! [ \g_{2}^{0,- } \bv']_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{3}^{0,+ } \bv']_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{5}^{1,+ } \bv^{(3)}]_{\ix\ixp } \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-3 [ \g_{3}^{1,- } \bv^{(3)}]_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{3}^{0,- } \bv']_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{4}^{0,+ } \bv']_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-4 0 \!\!\!\!&\!\!\!\ ! [ \g_{4}^{1,- } \bv^{(3)}]_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{4}^{0,- } \bv']_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{5}^{0,+ } \bv']_{\ix\ixp } \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-5 [ \g_{5}^{2,- } \bv^{(5)}]_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{5}^{1,- } \bv^{(3)}]_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! [ \g_{5}^{0,- } \bv']_{\ix\ixp } \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-6 \ddots \!\!\!\!&\!\!\!\ ! \ddots \!\!\!\!&\!\!\!\ ! \ddots \!\!\!\!&\!\!\!\ ! \ddots \!\!\!\!&\!\!\!\ ! \ddots \!\!\!\!&\!\!\!\ ! \ddots \!\!\!\!&\!\ ! \ddots \end{array } \right ) % % % % % % % % % % % % % % % % % % \ ] ] with an appropriate choice of the basis @xmath115 the matrix elements @xmath118 may couple only a finite number of basis functions . then , the recurrence ( [ dcdt : vectors : matrices : x ] ) would be solvable by continued fractions when the @xmath113-recurrence is infinite and this approach is seemingly precluded . even for finite coupling in @xmath113 , the @xmath131-recurrence may be preferable in certain situations ( e.g. , at low @xmath18 ) , as indicated by risken for a cosine potential in the classical limit ( * ? ? ? 11.5.6 ) ) . in these cases , one has been able to reduce the underlying large @xmath129 problem to @xmath126 problems with associated @xmath165 matrices . once the @xmath166 or @xmath167 are solved , we can reconstruct the wigner function from its expansion coefficients @xmath132 in the double basis @xmath168 @xmath169 obtaining the _ full solution of the quantum master equation_. if preferred , one can switch to the familiar density matrix @xmath170 by the inverse of transformation ( [ wigner : def ] ) @xmath171 we can get any observable inserting the above @xmath57 in equation ( [ average ] ) for the averages . nevertheless , common observables can many times be extracted _ directly _ from the @xmath132 . let us illustrate this with transport observables , which are characterised by the averages @xmath172 . for an arbitrary function of @xmath10 only , we get from ( [ w : expansion : xp ] ) @xmath173 where we have used @xmath174 , with @xmath175 [ equations ( [ hermite ] ) and ( [ w : expansion ] ) ] . next we introduce the auxiliary integrals ( for their calculation vd.[app : auxint ] ) @xmath176 then , the moments ( case @xmath177 ) can be written in terms of the expansion coefficients as @xmath178 . taking into account that @xmath179 has the parity of @xmath113 and carrying this to @xmath180 , we get for the first two @xmath181 the first moment is the _ current _ and the second characterises _ fluctuations _ around it . the zeroth moment corresponds to the _ normalisation _ of @xmath7 , and imposses on the @xmath132 the condition @xmath182 . finally , the quantum probabilities of @xmath8 and @xmath10 , given by the marginal distributions @xmath183 and @xmath184 , can be expressed in a similar fashion @xmath185 henceforth we shall apply the method described to incorporate fluctuations and dissipation in the problem of _ quantum transport in periodic potentials_. the simplest model consists of a particle evolving in a cosine potential subjected to external forces ( tilted `` washboard '' potential ) , which is also a paradigm of _ quantum brownian motion _ others include potentials lacking inversion symmetry ( ratchets ) , which have been used to model rectification of current and _ directional motion _ in several systems @xcite . these problems are demanding because periodic potentials have ( partly ) continuous spectra @xcite , rendering inappropriate methods devised for systems with discrete levels . in this section we compute the matrix coefficients of our recurrences for arbitrary periodic @xmath15 . with them we can implement the continued - fraction method to solve the quantum master equation , which we first check by regaining the classical dissipative and quantum hamiltonian limits . in periodic potentials the @xmath10-recurrence can not be used because of its infinite coupling range ( the derivatives of @xmath15 neither vanish , nor decrease , as they are essentially the potential itself ) . however , one can guess that the @xmath8-recurrence may have a coupling range related with the number of harmonics in @xmath15 . to exploit this we introduce plane waves as basis functions and the fourier expansion of the potential derivative : @xmath186 to preserve the periodicity of @xmath7 , one extracts the external force @xmath187 from @xmath15 including it in a generalised @xmath188 ; cf . equation ( [ dmdp ] ) . then , the auxiliary potential @xmath189 is set proportional to the periodic part @xmath77 . to compute the matrix elements in @xmath158 , we only need to do the integrals ( [ matr - elem : der : v ] ) with plane waves for @xmath190 . using @xmath191 , the integral @xmath192 , and the expansion @xmath193 , we find @xmath194_{\ix\ixp}=\v_{\ix-\ixp}^{(\iq ) } % % % % % % % % % % % % % % % % % % \;.\ ] ] from these elements and @xmath195 we get @xmath196_{\ix\ixp } & = & 0 & \delta_{\ix\ixp } & - & \etapm^{2\iq+1 } \gs_{\ip}^{\iq}(-\etapro \qab^{2}\ldb^{2 } ) & \v_{\ix-\ixp } ' \\ % % % % % % % % % % % % % % % % % % \dpm_{\ix\ixp } & = & \big(\iu\,\dplmi\ix-\etapm\bf\big ) & \delta_{\ix\ixp } & - & \dplmi\epspot & \bv_{\ix-\ixp } ' % % % % % % % % % % % % % % % % % % \end{array } % % % % % % % % % % % % % % % % % % \ ] ] with @xmath197 , @xmath101 and [ cf . equation ( [ gamma : qcoeff ] ) ] @xmath198 that is , @xmath199 is the coefficient in ( [ gamma : qcoeff ] ) with @xmath200 . inserting equation ( [ dmdpvper : matr - elem ] ) into the general matrices @xmath158 of section [ sec : mqbh ] we explicitly get the matrix coefficients of our recurrences ( [ app : matrices ] ) . for the cosine potential @xmath201 the coupling range is @xmath202 , because @xmath203 for @xmath204 and we have a @xmath71-term recurrence @xmath205 . for a @xmath150-harmonic ratchet potential @xmath206 % % % % % % % % % % % % % % % % % % % \;,\ ] ] the range is @xmath150 ( @xmath207-term recurrence ) because @xmath203 for @xmath208 , and so on . as we restricted @xmath74 , we have @xmath209 , so that @xmath210 . thus , the argument of @xmath211 is positive and @xmath212 can act as a regularisation factor [ an advantage of allowing @xmath213 in ( [ w : expansion ] ) ] . this factor reduces the weight of the off - diagonal terms inside @xmath158 , enhancing the numerical stability when going into the deep quantum regime . then we typically use @xmath214@xmath215 , while for classical calculations @xmath107 performs better . concerning the auxiliary potential @xmath189 , for the problems considered below we use @xmath216 . convergence is achieved with @xmath217 hermite functions and @xmath218 plane waves [ then @xmath219 . finally , a word on the scaled quantities ( section [ sec : wkk : scaled ] ) . to get period @xmath220 , we have set @xmath221 with @xmath222 the original period . the characteristic energy @xmath31 is given by the potential amplitude @xmath223 ; then @xmath224 is scaled by @xmath225 and the other characteristic quantities are : action , @xmath226 , frequency , @xmath227 , and force , @xmath228 . ) for a particle in a tilted sinusoidal potential ( two periods are displayed along @xmath8 ) . here @xmath229 , and @xmath230 , while @xmath231 , @xmath232 , @xmath233 , and @xmath234 ( left to right , top to bottom ) . , title="fig : " ] ) for a particle in a tilted sinusoidal potential ( two periods are displayed along @xmath8 ) . here @xmath229 , and @xmath230 , while @xmath231 , @xmath232 , @xmath233 , and @xmath234 ( left to right , top to bottom ) . , title="fig : " ] ) for a particle in a tilted sinusoidal potential ( two periods are displayed along @xmath8 ) . here @xmath229 , and @xmath230 , while @xmath231 , @xmath232 , @xmath233 , and @xmath234 ( left to right , top to bottom ) . , title="fig : " ] ) for a particle in a tilted sinusoidal potential ( two periods are displayed along @xmath8 ) . here @xmath229 , and @xmath230 , while @xmath231 , @xmath232 , @xmath233 , and @xmath234 ( left to right , top to bottom ) . , title="fig : " ] before going into the quantum regime , let us check that for small enough @xmath235 we recover the classical results . recall that in the deterministic limit a particle in a periodic potential has two critical forces @xmath236 ( retrapping force ) and @xmath237 ( force at which the barrier disappears ) . for @xmath238 the attractors of the dynamical system are solutions `` locked '' around the potential minima , whereas for @xmath239 the `` running '' solution is an attractor globally stable . in the range @xmath240 the system exhibits bistability . finally , @xmath236 decreases with @xmath17 [ for the cosine potential @xmath241 , whereas @xmath236 tends to @xmath237 as the damping increases , narrowing the bistability range . thermal fluctuations are incorporated by the klein kramers equation , which for classical particles in periodic potentials was solved by continued fractions in @xcite ( using the @xmath10-recurrence ) . here we solve the caldeira leggett quantum master equation ( [ wkk : dxp ] ) using a large @xmath50 . as shown before , when @xmath242 we have a @xmath71-term recurrence which , for the static response , can be written as @xmath243 with @xmath244 , @xmath245 , and @xmath246 . solving it with the continued - fraction method we reobtain risken s impressive graphs for the classical distribution @xmath57 ( * ? ? ? some of them are displayed in figure [ fig : risken ] for weak damping and various external forces ( here @xmath247 and @xmath248 ) . at low @xmath249 the distribution , always periodic along @xmath8 , has maxima at the potential minima , while the profile in momenta is a maxwellian envelope @xmath250 . as @xmath249 is increased we see the evolution from these _ locked _ solutions ( @xmath251 ) to bistability between the locked and _ running _ solutions , and eventually purely running states @xmath252 $ ] are obtained at large forces . having checked the retrieval of the classical results , we now enter the quantum regime by decreasing @xmath253 . firstly , it will be useful to get a more quantitative feeling of the relation of our parameter @xmath50 and `` how quantal is the system '' . to this end we study the bands of the corresponding hamiltonian problem ( without force ) . inserting the _ bloch _ function @xmath254 in the schrdinger equation , we arrive at the following eigenvalue problem ( in scaled units ; section [ sec : wkk : scaled ] ) @xmath255 u(\x ) = e\ , u(\x ) % % % % % % % % % % % % % % % % % % \ ; , \quad % % % % % % % % % % % % % % % % % % -\half\leq \q\leq\half % % % % % % % % % % % % % % % % % % \;.\ ] ] expanding @xmath256 in plane waves ( [ vper ] ) and truncating @xmath257 at some large @xmath126 one gets a @xmath258 matrix which is diagonalised numerically . various bands @xmath259 pertaining to problems discussed below are displayed in figure [ fig : bands ] . we see that the rule of thumb is that @xmath260 is of the order of the number of bands laying below the barrier . finally , the quantum parameter used by kandemir @xcite is @xmath261 , while chen and lebowitz employed @xmath262 @xcite . besides , for a josephson junction problem we have @xmath263 , with @xmath264 and @xmath265 the josephson and coulomb energies . ( left ) and @xmath150 ( right ) . the `` islands '' correspond to zones of negative @xmath7 . , title="fig : " ] ( left ) and @xmath150 ( right ) . the `` islands '' correspond to zones of negative @xmath7 . , title="fig : " ] let us continue for a while in the hamiltonian case to familiarise ourselves with the structure of the schrdinger eigenstates in a phase - space representation . kandemir @xcite obtained analytical approximations for the eigenfunctions of a particle in a cosine potential in terms of mathieu functions . they were represented through the wigner function associated to the density matrix of pure states @xmath266 . ( in @xcite various dirac deltas occurring were regularised with gaussians @xmath267 and in the plots @xmath268 was used ; we attain the same setting @xmath230 since our @xmath10 is thermally rescaled . ) using a very small damping ( @xmath269 ) in the caldeira leggett master equation we reobtain kandemir s hamiltonian results for the ground state ( figure [ fig : kandemir ] ) . we see that at low @xmath50 the state is extended or delocalised , becoming more localised at the potential wells when @xmath270 is increased . according to @xcite , the regions of negative @xmath7 are associated to brillouin zone boundaries . we have represented a larger range of @xmath10 to see how these @xmath271 islands are repeated in the @xmath10 direction ( @xmath7 can be as small as @xmath272 bordering the limits of our numerical accuracy ) . to see how these structures arise , note that for small @xmath50 the potential can be treated perturbatively [ equation ( [ schrodinger : bloch ] ) ] . then , the wave functions have the form @xmath273 , with @xmath274 . the first term is a pure plane wave , and the rest first order corrections due to the sinusoidal @xmath15 . performing the wigner transform ( [ wigner : def ] ) of @xmath275 , we get for the ground state @xmath276 ( cf . @xcite ) @xmath277 \nonumber\\ % % % % % % % % % % % % % % % % % % & - 2|\lambda|^{2 } \cos(2\x)\ , \delta(\p ) + 2\iu\lambda \sin(\x ) \big [ \delta(\p-\half ) + \delta(\p+\half ) \big ] % % % % % % % % % % % % % % % % % % \;.\end{aligned}\ ] ] the first line gives the momentum localisation at @xmath278 ( unperturbed ) and at @xmath279 . the first term in the second line arises from the interference of @xmath280 and @xmath281 , and is responsible for the weak modulation of the height of the @xmath278 bell along @xmath8 . this modulation is more intense for larger @xmath282 ( i.e. , larger @xmath50 ) . the last term accounts for the interference between the plane wave centred at @xmath278 with those at @xmath279 , and can produce the negative islands . therefore , the above simple functional form captures most features of the wigner funtions of figure [ fig : kandemir ] . after checking the connection with the classical and hamiltonian quantum limits , we finally include dissipation and temperature in the quantum case . here examples of particles in cosine and ratchet potentials will be discussed . transport properties of weakly damped particles in a cosine potential were studied by chen and lebowitz @xcite . they started from the system - plus - bath density matrix , traced out the bath variables , and performed a perturbative calculation in the potential height ( followed by a resummation ) to get @xmath283 . for low forces a free - particle like behaviour @xmath284 was obtained . increasing @xmath249 , the wave - vector associated to @xmath10 approaches the first zone boundary ( figure [ fig : bands ] ) . there , while landau zener tunnelling can bring the particle to the next band , bragg scattering reduces the velocity ( @xmath285 ) . eventually , at larger forces , @xmath10 corresponding to states inside the next band become favoured , and the free - particle behaviour is progressively recovered . vs. @xmath249 for various dampings at @xmath286 and @xmath287 . right : marginal distribution of momenta @xmath288 for @xmath289 at @xmath290 , @xmath291 , @xmath292 , @xmath293 , and @xmath207 . the arrows mark the zone boundaries @xmath294 . bottom panels : two views of @xmath57 for @xmath289 and @xmath290 . , title="fig : " ] vs. @xmath249 for various dampings at @xmath286 and @xmath287 . right : marginal distribution of momenta @xmath288 for @xmath289 at @xmath290 , @xmath291 , @xmath292 , @xmath293 , and @xmath207 . the arrows mark the zone boundaries @xmath294 . bottom panels : two views of @xmath57 for @xmath289 and @xmath290 . , title="fig : " ] vs. @xmath249 for various dampings at @xmath286 and @xmath287 . right : marginal distribution of momenta @xmath288 for @xmath289 at @xmath290 , @xmath291 , @xmath292 , @xmath293 , and @xmath207 . the arrows mark the zone boundaries @xmath294 . bottom panels : two views of @xmath57 for @xmath289 and @xmath290 . , title="fig : " ] vs. @xmath249 for various dampings at @xmath286 and @xmath287 . right : marginal distribution of momenta @xmath288 for @xmath289 at @xmath290 , @xmath291 , @xmath292 , @xmath293 , and @xmath207 . the arrows mark the zone boundaries @xmath294 . bottom panels : two views of @xmath57 for @xmath289 and @xmath290 . , title="fig : " ] since high @xmath295 approximations were involved in their calculations , we set @xmath23 and @xmath296 in the master equation ( [ wkk : dxp ] ) . solving it with the continued - fraction method we reobtain the effect just described ( figure [ fig : cheleb92 ] ; here we work in the regime @xmath297 ) . note that this quantum slowing effect is reduced as @xmath17 increases , since the coupling to the bath `` broadens '' the effective energy levels . this broadening makes less relevant the presence of the band gap ( `` bridges '' it ) and brings the curves progressively closer to the free - damped - particle behaviour @xmath284 . with our method we get , in addition , the wigner function ( i.e. , the full density matrix ) . the distribution of velocities ( [ wp ] ) , @xmath184 , when the curves start to rise again ( @xmath298 ) , shows two peaks separated by a minimum corresponding to the wave vector of the first zone boundary . loosely speaking the system shows coexistence of _ two _ quantum `` running '' solutions . classical running solutions have a wiggling structure along @xmath8 , as the particle slows down near the potential maxima ( figure [ fig : risken ] ) . the full @xmath57 shows that the quantum running solutions have a nearly straight structure ( substrate insensitive ) . we also recognise the familiar crescent islands of negative @xmath7 at large @xmath10 ( there @xmath299 ) . around @xmath300 a complex structure is developed along @xmath8 ( `` locked part '' ) , with the maxima deformed and the minima becoming slightly negative ( @xmath301 ) . these features may reflect some interference between the locked and the nearby running solution . however , simply adding a plane wave @xmath302 to the @xmath303 used to analyse the wigner functions of the hamiltonial problem [ equation ( [ wigner : turkeys ] ) ] we have not been able to reproduce such structure . anyway , it is erased when integrating over @xmath8 to get @xmath304 . we now turn our attention to periodic potentials without spatial inversion symmetry . this feature combined with out - of - equilibrium conditions can give rise to directional motion with net unbiased driving ( _ ratchet effect _ ) . this directional motion has been invoked to explain the behaviour of molecular motors and , with the emergence of nanoscience , opens the way to build mesoscopic devices and engines based on it . the theoretical work has concentrated on the classical behaviour and in the high - damping regime @xcite . the few quantum studies also addressed strong system - bath coupling in two situations : semiclassical @xcite and a extreme quantum case disregarding thermal effects @xcite . an exception to the large @xmath17 studies is @xcite , but there the substrate potential ( and the force ) were treated perturbatively . here , we shall solve the quantum master equation ( [ wkk : dxp ] ) with the ( non - perturbative ) continued - fraction method for particles in ratchet potentials , taking into account finite damping ( @xmath305 non - negligible inertia ) . for the ratchet potential ( [ vrat ] ) , the fourier coefficients of @xmath306 are @xmath307 and @xmath308 ( recall the thermal scaling of @xmath309 ; section [ sec : wkk : scaled ] ) . we use @xmath310 , which smooths a small shoulder that this potential exhibits on the `` easy '' side ( figure [ fig : bands ] ) . as discussed before , the range of index coupling is @xmath150 , which gives a @xmath207-term recurrence . for the stationary response it can be written as [ cf.equation ( [ rr:3:cosine ] ) ] @xmath311 which can be folded onto a canonical @xmath71-term recurrence by introducing appropriate ( block ) vectors and matrices ( [ app : rr - cf ] ) . the potential considered is the minimal extension of the cosine potential lacking inversion symmetry , so that the average velocities can be different for positive and negative forces ( vd . figure [ fig : cl - ratchet ] ) . we excite the system with a square - wave force switching alternatively between @xmath312 , and compute the `` rectified '' current of particles @xmath313 here @xmath314 are the correponding stationary velocities since we consider adiabatic conditions ( @xmath315 ) . in what follows we briefly investigate the classical limit ( understudied for finite damping ) and then proceed to study quantum corrections . figure [ fig : cl - ratchet ] displays @xmath316 as a function of @xmath18 , showing the appearance of non - zero rectified velocities ( ratchet effect ) . the rectification is optimum at some intermediate temperature ( at too low @xmath18 there is hardly a response , and at too high @xmath18 the potential is irrelevant , so its asymmetry plays no rle in either case ) . at @xmath317 the results coindice with those obtainable from the analytical @xmath283 for overdamped classical particles ( * ? ? ? 11.3 ) ( vd . also @xcite ) @xmath318 with @xmath319/t$ ] . inertia , however , broadens the @xmath316 curves and shifts the maxima to higher @xmath18 ( for the lowest @xmath229 the maximum moves to a slightly lower @xmath18 ; vd . this broadening is accompanied by a slower decrease of @xmath316 with @xmath18 . this is consistent with the asymptotic behaviour of @xmath316 when @xmath320 @xcite . for overdamped particles it goes as @xmath321 , while for weak damping it decreases only with @xmath322 ( @xmath323 ) . we have checked that our results recover these dependences ( curves not shown due to the smallness of the asymptotic @xmath324 ) . ) with @xmath229 and @xmath230 . two forces are used @xmath325 ( left ) and @xmath326 ( right ) , to show the lack of symmetry in the response ( running part more developed to the easy side ) and the locked regions reflecting the potential profile . bottom left : rectified classical velocity vs. temperature for various values of the damping and force @xmath327 . right : @xmath328 vs. @xmath18 for the positive force . , title="fig : " ] ) with @xmath229 and @xmath230 . two forces are used @xmath325 ( left ) and @xmath326 ( right ) , to show the lack of symmetry in the response ( running part more developed to the easy side ) and the locked regions reflecting the potential profile . bottom left : rectified classical velocity vs. temperature for various values of the damping and force @xmath327 . right : @xmath328 vs. @xmath18 for the positive force . , title="fig : " ] in our case @xmath329 ( the `` retrapping '' force ; section [ sec : risken ] ) , so that at @xmath330 the attractors are locked solutions . then , the absolute @xmath328 vs. @xmath18 curves start from @xmath331 at @xmath330 , depart from zero as @xmath18 is increased and evolve towards @xmath332 for @xmath320 ( free damped particle ) . in the overdamped case this evolution is nearly a step , whereas the slope of the initial raising decreases with decreasing @xmath17 ( if @xmath238 ) . to understand this , let us assimilate @xmath283 to the escape rate @xmath333 ( modified by some mean free path ) , so that @xmath334 . kramers theory @xcite shows that @xmath335 increases monotonously with @xmath17 . therefore , the @xmath328 vs. @xmath18 curves for finite @xmath17 must lay ordered below the overdamped result . on the other hand , the maximum rectification occurs in the temperature region where @xmath328 transits from @xmath336 to @xmath249 . this intermediate region is narrower ( and the slopes steeper ) the higher the damping is ( figure [ fig : cl - ratchet ] ) , while the curves start to raise from zero at a lower @xmath18 . therefore , increasing the damping the @xmath337 curves will be narrower and the maxima shift to lower @xmath18 , as observed . eventually , this also allows to explain the anomalous curve showing the return ( @xmath229 ) . here @xmath338 , so that at @xmath330 the system starts close to the zone of bistability . then the particle can perform incursions into the running solution , increasing the initial slope of @xmath339 but now when the damping is lowered , shifting the maximum to a lower @xmath18 . after the mandatory exploration of the classical limit , we proceed to make the system quantal , by decreasing @xmath50 . figure [ fig : q - ratchet ] shows the rectified current with @xmath340 ( with @xmath341 bands below the barrier ; cf . figure [ fig : bands ] ) for various @xmath17 , together with the reference classical curves . we have set @xmath342 is all cases , to have roughly the same amount of locked and running components in the solutions . at high temperatures ( @xmath343 ) the classical and quantum curves coincide , as expected . however , decreasing @xmath18 down to @xmath344 , the ratchet effect gets reduced or enhanced , with respect to the classical result , depending on the dissipation . finally , at low temperatures ( @xmath345 ) the rectification is always reduced . with @xmath346 ( using @xmath342 ) for @xmath340 and the reference classical curve ( @xmath347 ) . the insets of @xmath348 and @xmath215 show enlargements of the peaks ( together with results for @xmath349 and @xmath350 ) . bottom right panel : kramers classical rate vs. @xmath17 at various forces ( from melnikov meshkov formula @xcite ) . the crosses mark the points with @xmath351 . , title="fig : " ] with @xmath346 ( using @xmath342 ) for @xmath340 and the reference classical curve ( @xmath347 ) . the insets of @xmath348 and @xmath215 show enlargements of the peaks ( together with results for @xmath349 and @xmath350 ) . bottom right panel : kramers classical rate vs. @xmath17 at various forces ( from melnikov meshkov formula @xcite ) . the crosses mark the points with @xmath351 . , title="fig : " ] with @xmath346 ( using @xmath342 ) for @xmath340 and the reference classical curve ( @xmath347 ) . the insets of @xmath348 and @xmath215 show enlargements of the peaks ( together with results for @xmath349 and @xmath350 ) . bottom right panel : kramers classical rate vs. @xmath17 at various forces ( from melnikov meshkov formula @xcite ) . the crosses mark the points with @xmath351 . , title="fig : " ] with @xmath346 ( using @xmath342 ) for @xmath340 and the reference classical curve ( @xmath347 ) . the insets of @xmath348 and @xmath215 show enlargements of the peaks ( together with results for @xmath349 and @xmath350 ) . bottom right panel : kramers classical rate vs. @xmath17 at various forces ( from melnikov meshkov formula @xcite ) . the crosses mark the points with @xmath351 . , title="fig : " ] to understand qualitatively the underlying physics we turn to some semiclassical results for various related problems . first , let us consider the _ reflection / transmission coefficient _ for an asymmetric saw - tooth potential @xcite . the results depend on the energy of the incident quantum particle . the reflection coefficient for energies higher than the barrier , @xmath352 , and the underbarrier transmission coefficient , @xmath353 , read @xmath354 % % % % % % % % % % % % % % % % % % % \;,\ ] ] where @xmath355 , @xmath356 are the slopes at both sides of the cusp , @xmath357 the particle energy , and @xmath358}$ ] . suppose now that we have an asymmetric potential , with an easy direction to the right , and we deform it with forces @xmath312 , computing the corresponding reflection coefficients @xmath359 and @xmath360 . for energies higher than the barrier , the above @xmath352 gives @xmath361 ( [ app : semicl ] ) . that is , _ overbarrier wave reflection _ is less intense when the slope of the potential is smaller . thus overbarrier transmission ( as thermal hopping ) is favoured along the easy direction for energies above the barrier @xcite . in contrast , using @xmath353 for energies below the barrier , we find @xmath362 , so that @xmath363 . that is , the reflection is more intense in the easy direction or , equivalently , tunnelling events are favoured in the hard direction @xcite . for a distribution of particle energies one would expect a competition between these two types of quantum effects . then , the enhancement of the rectified velocity under @xmath312 forcing for the @xmath348 curve follows from the increase of the thermal escape with @xmath17 in the intermediate - to - low damping range @xmath364 ( figure [ fig : q - ratchet ] ) . at that damping more escape events occur and the particles launched over the barrier are subjected to the discussed phenomenon of wave reflection , which favours net motion along the easy side and hence increases @xmath316 . lowering the damping , on the other hand , less escape events are produced and wave - reflection becomes progressively dominated by tunnel events ( favouring the hard direction ) . at @xmath344 one would expect the tunnel to be thermally assisted through the higher bands , which are wider . a simple estimation of channel probabilities comes from multiplying a band thermal population ( using kramers rate ) by the probability of tunnel ( given by the semiclassical transmission coefficient @xmath353 ) . here the exponential reduction of @xmath333 with @xmath357 compites with the exponential increase of @xmath353 for the upper bands ( [ app : semicl ] ) . at @xmath344 and for @xmath340 we find that tunnel is indeed favoured through the upper bands . to check that the behaviours found at the maxima are systematic , we increase @xmath50 to @xmath349 ( @xmath365 bands below the barrier ) and reduce it to @xmath366 ( @xmath291 bands ; here the temperatures around the peak are still inside the range of validity of the master equation , small @xmath26 ) . the results ( insets ) show clearly the tendency to amplify either the enhancement or the reduction when the system becomes more quantal . finally , at low temperatures ( @xmath367 ) thermal activation becomes ineficient for all dampings . then a reduction of the rectification is expected , due to dominance of below - barrier energies and hence of tunnel events ( through the lowest bands ) . note that the decrease of @xmath316 relative to the classical results is comparable in the curves shown . this suggests that we have similar crossover temperatures @xmath368 in this damping range . ( @xmath368 , below which quantum effects dominate , is useful as a measure of quantum corrections at a given @xmath18 ) . this quantity is known exactly in the problem of the escape from a `` quadratic - plus - cubic '' potential @xmath369 $ ] ( * ? ? ? 14 ) @xmath370 in our units @xmath371 and in the damping range studied the factor in brackets is close to one ( @xmath372 ) . this gives crossovers @xmath373 and hence comparable reductions of the ratchet effect , as observed . this reduction of the rectification is the precursor of the current reversals ( @xmath374 ) found in these systems when tunnelling completely dominates . although we find small @xmath374 in some cases , the results can not be fully trusted . for example , at @xmath348 , @xmath340 and @xmath375 we have @xmath376 , which is bordering the limits of validity of the quantum master equation employed here . we conclude with the study of the ( non - adiabatic ) dynamical response to oscillating forces @xmath377 . then , the coefficients of the expansion ( [ w : expansion : xp ] ) of the wigner function are periodic functions of @xmath46 , so they can be fourier expanded as follows : @xmath378 to lowest order in the probing field , the static part of the corresponding vectors [ equation ( [ dcdt : vectors : matrices : x ] ) ] obeys a recurrence relation of the type ( vd . [ app : pert : ac ] ) @xmath379 the equations for the harmonics have the form ( @xmath380 is the identity matrix ) @xmath381 % % % % % % % % % % % % % % % % % % % \;,\ ] ] with the `` forcing '' @xmath382 involving the previous order results . specifically , @xmath383 , with @xmath384 the part of @xmath385 corresponding to @xmath386 , namely @xmath387 % % % % % % % % % % % % % % % % % % line-1 \sqrt{1}\,\etam\delta f \!&\ ! 0 \!&\ ! \sqrt{2}\,\etap\delta f \!&\ ! 0 \!&\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-2 0 \!&\ ! \sqrt{2}\,\etam\delta f \!&\ ! 0 \!&\ ! \sqrt{3}\,\etap\delta f \!&\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-3 0 \!&\ ! 0 \!&\ ! \sqrt{3}\,\etam\delta f \!&\ ! 0 \!&\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-6 ( 4 ) \ddots \!&\ ! \ddots \!&\ ! \ddots \!&\ ! \ddots \!&\ ! \ddots \end{array } \right ) % % % % % % % % % % % % % % % % % % \ ] ] [ vd . [ app : matrices ] for the general @xmath158 ; the @xmath384 actually entering in @xmath382 is divided by @xmath386 , cf . equations ( [ eom : generic ] ) and ( [ coefficients : a ] ) ] . equations ( [ 3rr : statics ] ) and ( [ 3rr : harmonics ] ) can be solved sequentially with the continued - fraction method . figure [ fig : suscep - w ] shows the linear susceptibility ( @xmath388 ) computed in this way for a particle in a _ cosine _ potential as a function of @xmath389 . ( here we return to safe ground because @xmath390@xmath391 ; for related dielectric and kerr relaxation curves of a quantum rotator vd . @xcite and references therein . ) for the largest damping the results correspond to the classical limit . there the line - shape @xmath392 broadens and extends to frequencies lower than the oscillation frequency near the bottom of the potential wells ( @xmath393 in our units ) . for low damping , this classical _ spreading of oscillation frequencies _ has an important contribution from the amplitude dependence of the period of oscillation in anharmonic potentials @xcite . this non - dissipative contribution to @xmath392 is also known in classical spin problems , where the precession frequency depends on @xmath394 due to the magnetic anisotropy @xcite . and @xmath395 with @xmath289 ( @xmath305 classical ) , @xmath396 and @xmath397 ( less to more peaked curves ) . left panel : real part . right panel : imaginary part ( dashed line , @xmath398 but halving @xmath399 ) . vertical lines : loci of the transition frequencies of the associated quadratic - plus - quartic anharmonic oscillator [ equation ( [ delta - pert - fey ] ) ] . , title="fig : " ] and @xmath395 with @xmath289 ( @xmath305 classical ) , @xmath396 and @xmath397 ( less to more peaked curves ) . left panel : real part . right panel : imaginary part ( dashed line , @xmath398 but halving @xmath399 ) . vertical lines : loci of the transition frequencies of the associated quadratic - plus - quartic anharmonic oscillator [ equation ( [ delta - pert - fey ] ) ] . , title="fig : " ] the quantum regime is approached by decreasing the kondo parameter @xmath400 [ equation ( [ kondos ] ) ] . reducing @xmath17 with a fixed @xmath50 we find that the @xmath401 curves develop a multipeaked structure . these peaks could be vaguely seen as the mentioned nonlinear oscillations becoming quantised . to give content to this statement , we investigate perturbatively the effect of the anharmonic terms of a cosine potential ( those beyond @xmath402 ) on the energy levels of the harmonic oscillator part . since the main contribution should come from the quartic term , we consider the potential @xmath403 . first - order perturbation theory gives the eigenvalues ( * ? ? ? 6.3 ) @xmath404 expanding the cosine potential @xmath405 , we identify @xmath406 and @xmath407 ( @xmath408 , soft anharmonicity ) . then we can compute the level spacing @xmath409 % % % % % % % % % % % % % % % % % % % \;,\ ] ] which has acquired a dependence on the level index @xmath410 absent in the harmonic case . the associated transition frequencies @xmath411 are plotted in figure [ fig : suscep - w ] [ scaled as @xmath412 . they agree well with the location of the main peaks of the quantum @xmath392 curves , supporting their relation with the nonlinearity of the potential . furthermore , the contribution from transitions between the higher levels should be reduced when they are less populated . decreasing @xmath18 , we indeed see that the corresponding peaks ( located at lower @xmath389 ) are substantially reduced . in the suggestive language of nonlinear oscillations we would say that lowering the temperature the large amplitude oscillations ( having longer period ) become less probable . the applicability of the quantum nonlinear oscillator picture is grounded on the flattness of the lowest bands at the @xmath50 considered , which are well approximated by constant levels ( vd.figure [ fig : bands ] ) . we have computed the semiclassical width of the bands in a cosine potential in [ app : semicl ] . an upper bound for the width [ equation ( [ wb : approx : top ] ) ] reads in dimensionless units @xmath413 $ ] , with @xmath414 the band mean energy relative to the potential amplitude . for moderate @xmath50 this bound gives in fact very flat bands , specially the deepest ones ( @xmath415 ) . this also implies that the width of the @xmath392 peaks has a negligible contribution from the intrinsic level width , being dominated by the damping and thermal broadening . traditionally , the continued - fraction method has been employed to solve fokker planck equations for few - variable classical systems ( translational and rotational problems ) . when compared with numerical simulations of the langevin equation , the method has several _ shortcomings _ : ( i ) it is quite specific of the problem to be solved ( one needs to recalculate the recurrence coefficients for each potential ) , ( ii ) its convergence and stability depend on the parameters of the problem ( and can be poor in some ranges ) , and ( iii ) it does not return `` trajectories '' , which in the simulations provide helpful insight . nevertheless , when the method can be used its _ advantages _ are most valuable : ( i ) it is free from statistical errors , ( ii ) is essentially nonperturbative , ( iii ) specially apt to get stationary solutions ( static and dynamic ) , ( iv ) high efficiency ( allowing to explore parameter ranges out of the reach of the simulation ) , and ( v ) it gives the distribution @xmath7 , which certainly compensates for the lack of trajectories . for these reasons , and the lack of quantum langevin simulations , it was worth to develope continued - fraction methods for quantum dissipative systems . this had been done for problems of spins in a thermal bath and in quantum nonlinear optics . in this article we have discussed in detail the adaptation of the continued - fraction method to tackle quantum master equations in phase - space problems , taking advantage of the parallels with the classical case provided by the wigner formalism . we have seen that the quantum extension of the continued - fraction method is more problem dependent ( except for polynomial potentials ) , due to the necessity of using the @xmath8-recurrence of the quantum hierarchies and finding suitable bases . nevertheless , it inherits from the classical methods most of the abovementioned advantages , in particular , the obtaining of the wigner function , from which any observable can be computed . furthermore , the eigenvalue spectrum of the system , although helpful in understanding the physics , is not necessarily required . this is advantageous when dealing with non - bounded hamiltonians , continuous spectra , etc . finally , by changing the appropriate quantum parameter , the connection with the classical results is attained in a natural way . note that `` continued - fraction '' is a quite generic term ( like `` series expansion '' ) , which can be found in many different contexts , and in particular in works dealing with quantum dissipative systems ( see , e.g. , @xcite and references therein ) . however , in those cases one usually obtains some continued - fraction expression for a certain quantity ( e.g. , the linear susceptibility , memory functions , etc . ) , whereas we have obtained the _ complete solution of the quantum master equation _ by continued - fraction methods . the approach allows to study the interplay of quantum phenomena , nonlinearity , thermal fluctuations , and dissipation in several classes of systems . this is important , because methods optimised for one of the sides ( e.g. , nonlinearity or quantal behaviour ) tend to perform poorly for the others ( e.g. , fluctuations or dissipation ) . in addition , the visualization of @xmath57 plus the knowledge of the classical phase - space structure ( orbits , separatrices , attractors , etc . ) can provide valuable insight in difficult problems . finally , the main limitations of the approach are those that the starting master equations may have . we have considered the celebrated caldeira leggett quantum master equation , but the method can be applied to a class of equations of this type , and to possible generalisations . we have implemented the method in the canonical problem of quantum brownian motion in periodic potentials . we have considered both the cosine and ratchet potentials . for the former we have recovered and extended a number of classic results and interpreted them on the basis of the wigner formalism . results for the dynamics under oscillating forcing have also been obtained , illustrating the modification of the nonlinear effect of spreading of oscillation frequencies by quantum phenomena . for particles in ratchet potentials we have studied , under adiabatic conditions , the effects of finite damping on the rectified velocity . taking into account the competition of thermal hopping , overbarrier wave reflection and tunnel events , together with semiclassical analytical results , has allowed us to understand in great detail the physics of these systems . this work was supported by dges , project bfm2002 - 00113 , and dga , grant no . b059/2003 . we acknowledge f. falo , j. j. mazo , and a. p. piazzolla for useful discussions and valuable inspiration . here we briefly discuss the solution of recurrence relations by continued - fraction methods . we first consider @xmath71-term recurrences and then differential recurrences , vector - matrix cases , folding of larger - coupling recursions into @xmath71-term ones , and the problem of the intial conditions . ( for a brief description of the relation of continued fractions , series expansions , recursions , and orthogonal polynomials see @xcite . ) suppose we have the simplest case of a @xmath71-term recurrence relation of the form @xmath416 with @xmath417 and @xmath418 given quantities . to obtain the @xmath19 one inserts in ( [ rr ] ) the anstze @xcite ( upper and lower signs for @xmath419 and @xmath420 ) @xmath421 getting the following ladder coefficients @xmath422 and shifts @xmath423 @xmath424 if the recurrence is finite or the @xmath19 decrease quickly with @xmath425 ( e.g. , when they are coefficients of some judiciosly chosen expansion ) , we can truncate at some large @xmath426 , by setting @xmath427 and @xmath428 . next we _ iterate downwards _ in ( [ s : a ] ) getting @xmath422 and @xmath423 , down to @xmath429 . then , to get the @xmath19 from ( [ c : risken : com ] ) , we need @xmath430 , which is obtained from @xmath431 combination of equation ( [ rr ] ) at @xmath429 and @xmath432 . thus , starting from @xmath430 , we _ iterate @xmath433 upwards _ , getting the solution of the recursion . for @xmath434 ( homogeneous recurrences ) , we have @xmath435 and the solution reads @xmath436 . note that @xmath422 is given in terms of @xmath437 in the denominator , which can in turn be written as a fraction with @xmath438 in the denominator , and so on . this has the structure of a _ continued fraction _ , naming the method @xmath439 the iterative solution can be cast into explicit form as a series of products of continued fractions @xcite . the above form , however , has the virtue of being easy to implement in a computer . convergence is checked increasing @xmath440 and repeating the procedure . solving equation ( [ rr ] ) in this way requires a computational effort of order @xmath441 , instead of the order @xmath442 of an arbitrary linear algebra problem . this reduction arises from the tridiagonal structure of the matrix associated to ( [ rr ] ) . ( for analytical inversion of tridiagonal matrices see @xcite and references therein . ) we can solve analogously a differential recurrence of the type @xmath443 laplace transformation converts this equation into [ @xmath444 @xmath445 % % % % % % % % % % % % % % % % % % % \;,\ ] ] where @xmath446 has been used . then , introducing @xmath447 and @xmath448 , the above equation has the structure of the ordinary recurrence ( [ rr ] ) . the quantities involved in the above recurrences need not to be scalars , but they can be @xmath449-vectors ( @xmath19 and @xmath418 ) and the coefficients @xmath417 are @xmath450-matrices . then we speak of _ matrix continued fractions_. the only change in the above solution is that the fraction bars then mean matrix inversion ( `` from the left '' , @xmath451 ) . this is important , because a recurrence relation involving more than three coefficients can be `` folded '' into a @xmath71-term recurrence by introducing vector and matrix quantities . let us show how to do so for the following @xmath207-term recursion @xmath452 defining the @xmath150vectors @xmath453 and the @xmath454matrices @xmath455 it can be easily seen that equation ( [ rr:5 ] ) for @xmath456 and @xmath457 is equivalent to @xmath458 then , insertion of @xmath459 gives the matrix version of ( [ s : a ] ) . note that the @xmath19 and @xmath418 can already be vectors and the @xmath417 matrices . then , the above @xmath460 , @xmath461 , and @xmath462 are `` block '' vectors and matrices . finally , to iterate upwards @xmath459 , the `` initial condition '' @xmath463 is obtained from the matrix version of ( [ condini ] ) ( which we write @xmath464 ) . as in the absence of forcing ( @xmath465 ) the general solution involves an overall multiplicative constant , one adds to this system an extra equation to fix it ( e.g. , a normalisation contidion on the coefficients ; vd . section [ sec : obs ] ) . alternatively , we can fix arbitrarily one element of @xmath463 and rescale the solution at the end . in either case , one has an extra equation which added to @xmath466 yields the extended @xmath467 system @xmath468\hline\\[-2.ex ] % % % % % % % % % % % % % % % % % % \mbox{\,l.h.s.\ eq.\ , } \end{array } \right ) } \hspace{-0.5em } \begin{array}{c } \uparrow \\ { \scriptstyle ( j+1 ) } \\ \downarrow \end{array } % % % % % % % % % % % % % % % % % % \quad % % % % % % % % % % % % % % % % % % \mathbf{b}^{\ast } = \left ( \begin{array}{c } b_{1 } \\[-1.5ex ] \vdots \\[-1.5ex ] b_{j } % % % % % % % % % % % % % % % % % % \\[0.ex]\hline\\[-2.5ex ] % % % % % % % % % % % % % % % % % % \mathrm{\!\!r.h.s.\!\!\ ! } \end{array } \right ) % % % % % % % % % % % % % % % % % % \ ] ] this system can be solved by any method appropriate for cases with more equations than unknowns ( e.g. , using @xmath469- or svd - decomposition @xcite ) , getting @xmath463 . here we derive @xmath470 , the @xmath85 transform of the operator @xmath13 in the master equation with @xmath471 $ ] [ equation ( [ w : expansion ] ) ] . we split @xmath470 as @xmath86 , with @xmath87 , @xmath88 , and @xmath89 given by ( [ wkk : lir : lkin : lv ] ) . for the `` product '' of operators we use that the transformation of the product is equal to the product of the transformations : @xmath472 which follows by sandwiching `` identities '' @xmath473 between the factors . besides , the multiplication by functions of @xmath1 commutes with @xmath85 , so that only the differential terms need to be transformed . from these two properties we see that _ only @xmath474 and @xmath475 need to be calculated _ ( and then raised to some power or multiplied ) . _ calculation of @xmath474_. in order to transform @xmath121 , we apply @xmath476 , to an arbitrary function @xmath477 . the @xmath10 dependent part of @xmath73 cancels out and we obtain @xmath478 which has the form of a `` displaced '' differential operator . _ calculation of @xmath475_. applying @xmath479 to a function @xmath480 , we similarly get @xmath475 : @xmath481 this @xmath475 can be expressed in terms of the following creation and annihilation operators @xmath482 thus , using the shifted @xmath78 parameters @xmath483 [ equation ( [ etam : etap ] ) ] we arrive at @xmath484 for @xmath107 ( the choice in the classical case ) , @xmath485 and @xmath486 , so that @xmath487 . using the product property @xmath489 , we have @xmath490 . then , replacing @xmath67 by its `` bar '' form ( [ ppb ] ) and writing @xmath491 [ from ( [ bm - bp ] ) ] , we find @xmath492 here we have also used @xmath493 and the commutation rule @xmath494=1 $ ] to get a _ normally ordered _ form . if we introduce some compact notation for the coefficients @xmath495 we can finally write @xmath496 ( the choice @xmath107 gives the self - adjoint form @xmath497 @xcite . ) using equations ( [ pxb ] ) and ( [ ppb ] ) for @xmath474 and @xmath475 , as well as @xmath491 and the definition ( [ dmdp ] ) of @xmath499 we get for the kinetic part @xmath500 : @xmath501 the coefficients of @xmath90 and @xmath91 are @xmath502 and @xmath503 , also defined in ( [ dmdp ] ) , and we have @xmath504 to obtain @xmath506 [ vd . equation ( [ wkk : lir : lkin : lv ] ) ] , we use the product property ( [ bar : product ] ) to raise @xmath475 to @xmath507 , getting @xmath508 to get later the matrix elements of @xmath94 between hermite functions it is convenient to cast ( [ lvb ] ) in a normally ordered form . to this end we use pathak s theorems @xcite in the following generalised form @xmath509 ! } \ ; \frac{(\etapro/2)^{q}}{q ! } \;.\ ] ] here @xmath510 with @xmath511 $ ] ( @xmath512 is the result of moving the @xmath513 to the left as if they were scalars ) . the generalisation resides in letting @xmath514 and @xmath513 have a constant but different from one commutator . in our case @xmath515=\etapro % % % % % % % % % % % % % % % % % % % \;,\ ] ] follows from the commutator @xmath494=1 $ ] with @xmath516 . note that @xmath514 and @xmath513 are not the adjoint of each other , but the notation is symbolic . with help from equation ( [ pathak:2:gral : odd : alt : rep ] ) and @xmath517 [ equation ( [ kappa : s ] ) ] we can write ( [ lvb ] ) as follows @xmath518 where we have introduced the notations @xmath519 ! } \frac{(\etapro/2)^{q}}{q ! } \bv^{(2\iq+1 ) } % % % % % % % % % % % % % % % % % % \ ; , \qquad % % % % % % % % % % % % % % % % % % \hat{x}^{\ell } = \nol\big(\aplus+\aminus\big)^{\ell}\nor % % % % % % % % % % % % % % % % % % \;.\ ] ] for a given @xmath520 different powers of @xmath521 appear . we take a common factor @xmath522 and add coefficients , e.g. , @xmath523 with @xmath521 , @xmath524 , with @xmath525 , etc . @xmath526 for @xmath527 in @xmath528 we write @xmath529}=\px^{2q}\bv^{(2\ell+1)}$ ] and introduce the operator @xmath530 @xmath531 now the @xmath532 and @xmath533 dependences factorise , so we can sum the series of @xmath532 in @xmath534 , getting an exponential . then , recalling the definition of @xmath103 we finally obtain the normally ordered form for @xmath94 we were looking for @xmath535 the operator @xmath536 does not act on what could appear after @xmath15 , since @xmath530 was introduced as some abreviated notation for the high - order derivatives of the potential . here we calculate the matrix elements @xmath537 , with the transformed operator @xmath86 obtained in [ app : lb ] . we split in parts the calculation by introducing the corresponding decomposition : @xmath538 , @xmath539 , and @xmath540 , where we have used the custom `` bra''``ket '' notation . the calculation is simplified because we have all operators @xmath470 expressed as normally ordered forms of @xmath90 and @xmath91 . then we can take advantage of the ladder actions @xmath541 , @xmath542 , and the number property @xmath543 . from ( [ lirb ] ) for @xmath92 and the orthonormality of the @xmath545 , we find for @xmath546 @xmath547 then , we introduce the notation @xmath548 for the coefficient of the diagonal term [ cf.equation ( [ gammas ] ) ] , multiply by @xmath84 and sum over @xmath549 , getting @xmath550 using equation ( [ lkinb ] ) for @xmath93 and the `` ladder '' action of @xmath90 and @xmath91 , we have @xmath552 where we have used the kronecker delta to interchange @xmath113 and @xmath553 . then , multiplying by @xmath84 and summing over @xmath549 , one obtains @xmath554 this will take longer . first , the operator @xmath556 appearing in equation ( [ lvb : norord : el ] ) for @xmath94 can be put in explicit form by using the binomial formula @xmath557 the @xmath514 and @xmath513 are proportional to @xmath90 and @xmath91 [ equation ( [ am - ap ] ) ] , so the contribution of each summand to @xmath558 involves @xmath559 . from orthonormality only one term of the sum ( [ sum : pathak ] ) will contribute , namely @xmath560 , while @xmath561 will be odd , @xmath562 with @xmath563 ( reflecting the odd powers of @xmath67 in @xmath89 ) . then , in terms of @xmath520 , the index @xmath564 is restricted to @xmath565 [ [ relating - the - results - for - ippipmp2iq1 . ] ] relating the results for @xmath562 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the two cases can be easily related . first , the combinatorial coefficients are equal @xmath566 , because @xmath567 . second , the @xmath78 factors , coming from the proportionality of the @xmath568 , to @xmath569 , are simply obtained from each other by exchanging pluses and minuses ( @xmath516 ) : @xmath570 third and last , the matrix elements are also relatable . for @xmath101 , we have @xmath571 , while for @xmath572 @xmath573 which is formally identical to the former by replacing @xmath574 and @xmath575 . thus , from now on we consider only the case @xmath101 ( factor @xmath576 ) . the result for @xmath572 , will be readily obtained from this by replacing @xmath577 from all these considerations , we can write the matrix elements of @xmath556 as @xmath578 where we omit a kronecker delta ensuring @xmath101 . the rest of the calculation consists of multiplying this by @xmath579 and sum over all @xmath533 to get @xmath580 [ equation ( [ lvb : norord : el ] ) ] . [ [ restrictions - on - the - sum - on - ell . ] ] restrictions on the sum on @xmath533 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the contribution of the infinite sum ( [ lvb : norord : el ] ) to @xmath580 is fortunately cut - off . first , the lower index in @xmath581 should be positive and , introducing the shifted index @xmath582 , we obtain ( we rename @xmath583 at the end ) @xmath584 here we have already used that the sum is also cut - off from above at @xmath585 , since @xmath586 , for @xmath587 . on the other hand , for @xmath588 [ equation ( [ lvb : norord : el ] ) ] we have @xmath589 ! } ( -\ldb^{2}\px^{2})^{\ell } \frac{(-1)^{\iq}\ldb^{2\iq}}{(2\iq+1 ) ! } \bv^{(2\iq+1 ) } % % % % % % % % % % % % % % % % % % \;.\ ] ] combining this with @xmath590 , gathering @xmath591 with @xmath592 to form the operator @xmath530 [ equation ( [ aqlq ] ) ] , and recalling the definition ( [ kappa : s ] ) of @xmath103 , we obtain @xmath593 ! } \ ; \big\langle\ip\big| ( \bp)^{(2\iq+1)+\ell}\bd^{\ell } \big|\ipp\big\rangle \ ; \frac{(-\dlpxx)^{\ell}}{\ell ! } \ ; \bv^{(2\iq+1 ) } % % % % % % % % % % % % % % % % % % \;.\ ] ] note that the quotient of factorials equals @xmath594 , with the pochhammer symbol defined by ( * ? ? ? 13.5 ) @xmath595 [ [ computation - of - the - matrix - element . ] ] computation of the matrix element . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to conclude , we only need the matrix element @xmath596 in ( [ lvb : matele:1 ] ) . we first pass @xmath507 times the action of @xmath91 to @xmath597 by taking its adjoint @xmath90 and then we use repeteadly its downward ladder action , getting @xmath598 with @xmath101 . then , in terms of the pochhammer symbol ( [ pochhammer ] ) , we can write @xmath599 now , taking into account that the factor @xmath600 cancels that of @xmath601 and using the modified coefficient @xmath602 [ equation ( [ gamma : qcoeff ] ) ] , we arrive at @xmath603 \bv^{(2\iq+1 ) } % \;\delta_{\ip-(2\iq+1),\ipp } % % % % % % % % % % % % % % % % % % \;.\ ] ] comparing with the series definition of the confluent hypergeometric ( kummer ) function @xmath102 ( * ? ? ? 13.6 ) we get @xmath604 \bv^{(2\iq+1 ) } % % % % % % % % % % % % % % % % % % \quad % % % % % % % % % % % % % % % % % % \kum(a , c\,;z ) = \sum_{\ell=0}^{\infty } \frac{(a)_{\ell}}{(c)_{\ell } } \frac{z^{\ell}}{\ell ! } % % % % % % % % % % % % % % % % % % \;.\ ] ] the operator in front of @xmath76 is simply @xmath605 [ equation ( [ gamma : qcoeff ] ) ] , so that we can finally write @xmath606 note that @xmath607 is a polynomial because the first negative index cuts the upper pochhammer symbol . indeed the `` integer - plus - one '' second argument @xmath608 tells us that it is an associated laguerre polynomial ( * ? ? ? 13 ) @xmath609 in our case @xmath610 and @xmath611 , so that @xmath612 . [ [ final - expressions - for - sum_ippq_ipipprm - vec_ipp . ] ] final expressions for @xmath555 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + once we have computed @xmath613 , the matrix elements of @xmath94 , we multiply by @xmath84 and sum over @xmath549 ( i.e. , over @xmath520 ) , to get @xmath614 } \big [ \g_{\ip}^{\iq,- } \ , \bv^{(2\iq+1 ) } \big ] \ , \ec_{\ip-(2\iq+1 ) } % % % % % % % % % % % % % % % % % % \;.\ ] ] note that this @xmath520-sum is restricted because @xmath615 . this yields the upper limit @xmath616 $ ] , with @xmath617 $ ] the integer part of @xmath618 . the contribution from @xmath619 is easily obtained recalling the transformation rules ( [ rule ] ) ( here the sum is not restricted ) @xmath620 \ , \ec_{\ip+(2\iq+1 ) } % % % % % % % % % % % % % % % % % % % \;,\ ] ] with @xmath621 given by ( [ gamma : qcoeff ] ) . we have enclosed into square brackets the action of the operators @xmath622 to recall that they only act on the @xmath8-dependence of the potential . for @xmath107 , the choice in the classical case , @xmath485 , @xmath486 , and @xmath623 . then , @xmath624 and @xmath625 , so that only the @xmath626 terms survive . the chain of equations then adquires an unbalanced structure ( like in the old hierarchies @xcite ) , which may result in a poor stability when the quantum terms are important . in constrast , @xmath214 gives @xmath627 , and hence the weight of the terms at both sides of @xmath113 ( @xmath305 above and below the diagonals in the matrices @xmath158 ) is similar , improving the stability when going into the deep quantum regime . for special @xmath78 and @xmath189 , the integrals @xmath628 and @xmath180 [ equation ( [ ial : knl ] ) ] appearing in the expressions ( [ j : ial : knl ] ) for the observables , are easily done . for @xmath107 , we simply have @xmath629 , @xmath630 , and @xmath631 . if @xmath632 ( i.e. , @xmath216 ) , we get @xmath633 . then the first moments reduce to @xmath634 ( normalisation ) and to @xmath635 and @xmath636 . in the general case , one can derive recurrences relating the @xmath180 with lower order ( in @xmath533 ) ones . thus , only @xmath637 is needed , which can be found analytically . concerning the @xmath628 , we mostly use @xmath216 , and hence @xmath633 . anyway , for periodic potentials one can derive recurrence relations for them which can also be solved by continued fractions . to get the recurrence for the @xmath180 we start from their definition ( [ ial : knl ] ) with index @xmath638 , express the last @xmath10 as @xmath491 [ equation ( [ bm - bp ] ) ] , and then use @xmath639 , getting @xmath640 this recurrence can be used to get the @xmath641 from the @xmath180 . we can get analytical expressions for the first few @xmath180 with help from the tabulated integral ( * ? ? ? * equation ( 7.374 - 4 ) ) @xmath642 for @xmath643 , we use this result with @xmath644 , obtaining @xmath645 although @xmath646 and @xmath647 can now be obtained from ( [ knl : rr ] ) , they can also be done with help from equation ( [ hermite : func : int:1 ] ) , getting : @xmath648 , @xmath649 , and @xmath650 for a general periodic potential we can derive a recurrence relation for the integrals @xmath651\,u_{\ix}(\x)$ ] [ equation ( [ ial : knl ] ) ] . we integrate by parts the expression for @xmath652 and use @xmath653 with @xmath654 , getting @xmath655 for a finite number of harmonics ( @xmath656 , if @xmath657 ) , the recurrence ( [ rr : ial : vper ] ) has a finite coupling range . for instance , for the ratchet potential ( [ vrat ] ) , we have @xmath307 and @xmath308 and the above recurrence reduces to @xmath658 % % % % % % % % % % % % % % % % % % \ ; , \qquad % % % % % % % % % % % % % % % % % % \epspott=\epspot\,\vo/2\kt % % % % % % % % % % % % % % % % % % \;.\ ] ] with @xmath659 as input ( obtained numerically , e.g. , by simpson rule ( * ? ? ? * app . 2 ) ) , this recursion can be solved by continued fractions ( [ app : rr - cf ] ) . here we write expliticly the matrices @xmath158 for arbitrary periodic potentials . we shall give the results for a slightly generalised basis , obtained by shifting the origin of momenta by an amount @xmath660 [ cf.equation ( [ w : expansion : xp ] ) ] @xmath661 this momentum shift is convenient to `` catch '' solutions centred far from zero @xmath10 , where the ordinary basis would need many @xmath80 to reconstruct the distribution . ( a useful choice is , when generating a curve , to set @xmath662 of the precedent point . ) the momentum shift is handled as a change of variable @xmath663 . then , @xmath664 , but the term @xmath62 makes the force to enter in the combination ( scaled units ) @xmath665 as mentioned in section [ sec : vper ] , we extract the force @xmath187 from @xmath15 and set the auxiliary potential @xmath189 proportional to the periodic part @xmath77 . the @xmath158 are obtained by inserting the formulae for @xmath666 , and @xmath667_{\ix\ixp}$ ] [ equation ( [ dmdpvper : matr - elem ] ) ] , into the general matrices of section [ sec : mqbh ] . instead of giving a single expression , we write separately the matrices for @xmath668 and @xmath669 . the central matrix @xmath670 is determined by the terms involving @xmath671 , and reads @xmath672 % % % % % % % % % % % % % % % % % % line-1 \sqrt{1}(\iu\,\dmi\ix\!-\!\etam\bfsh ) \!\!\!\!&\!\!\!\ ! \gammao_{1 } \!\!+\!\ ! \pshift\,\iu\ix \!\!\!\!&\!\!\!\ ! \sqrt{2}(\iu\,\dpl\ix\!-\!\etap\bfsh ) \!\!\!\!&\!\!\!\ ! \sqrt{2\cdot3}\ , \gammapp \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-2 \sqrt{1\cdot2}\ , \gammamm \!\!\!\!&\!\!\!\ ! \sqrt{2}(\iu\,\dmi\ix\!-\!\etam\bfsh ) \!\!\!\!&\!\!\!\ ! \gammao_{2 } \!\!+\!\ ! \pshift\,\iu\ix \!\!\!\!&\!\!\!\ ! \sqrt{3}(\iu\,\dpl\ix\!-\!\etap\bfsh ) \!\!\!\!&\!\!\!\ ! \sqrt{3\cdot4}\ , \gammapp \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-3 0 \!\!\!\!&\!\!\!\ ! \sqrt{2\cdot3}\ , \gammamm \!\!\!\!&\!\!\!\ ! \sqrt{3}(\iu\,\dmi\ix\!-\!\etam\bfsh ) \!\!\!\!&\!\!\!\ ! \gammao_{3 } \!\!+\!\ ! \pshift\,\iu\ix \!\!\!\!&\!\!\!\ ! \sqrt{4}(\iu\,\dpl\ix\!-\!\etap\bfsh ) \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-4 0 \!\!\!\!&\!\!\!\ ! 0 \!\!\!\!&\!\!\!\ ! \sqrt{3\cdot4}\ , \gammamm \!\!\!\!&\!\!\!\ ! \sqrt{4}(\iu\,\dmi\ix\!-\!\etam\bfsh ) \!\!\!\!&\!\!\!\ ! \gammao_{4 } \!\!+\!\ ! \pshift\,\iu\ix \!\!\!\!&\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-6 ( 5 ) \ddots \!\!\!\!&\!\!\!\ ! \ddots \!\!\!\!&\!\!\!\ ! \ddots \!\!\!\!&\!\!\!\ ! \ddots \!\!\!\!&\!\!\!\ ! ! \ddots \end{array } \right ) % % % % % % % % % % % % % % % % % % \ ] ] with @xmath673 and @xmath674 given by ( [ gammas ] ) , @xmath675 , and @xmath676 . here the periodic potential does not appear ; only @xmath187 enters in the matrices @xmath670 . besides , they do depend on the index @xmath131 ( cf . below ) . the matrices @xmath677 are determined by the terms involving @xmath678 in equation ( [ dmdpvper : matr - elem ] ) and read @xmath679 % % % % % % % % % % % % % % % % % % line-1 \etam \gs_{1}^{0 } \!-\ ! \sqrt{1}\,\epspot \!\!\!&\!\!\ ! -\pshift\,\epspot \!\!\!&\!\!\ ! \etap \gs_{2}^{0 } \!-\ ! \sqrt{2}\,\epspot \!\!\!&\!\!\ ! 0 \!\!\!&\!\!\ ! \etap^{3 } \gs_{4}^{1 } \!\!\!&\!\!\ ! 0 \!\!\!&\!\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-2 0 \!\!\!&\!\!\ ! \etam \gs_{2}^{0 } \!-\ ! ! -\pshift\,\epspot \!\!\!&\!\!\ ! \etap \gs_{3}^{0 } \!-\ ! \sqrt{3}\,\epspot \!\!\!&\!\!\ ! 0 \!\!\!&\!\!\ ! \etap^{3 } \gs_{5}^{1 } \!\!\!&\!\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-3 \etam^{3 } \gs_{3}^{1 } \!\!\!&\!\!\ ! 0 \!\!\!&\!\!\ ! \etam \gs_{3}^{0 } \!-\ ! \sqrt{3}\,\epspot \!\!\!&\!\!\ ! -\pshift\,\epspot \!\!\!&\!\!\ ! \etap \gs_{4}^{0 } \!-\ ! \sqrt{4}\,\epspot \!\!\!&\!\!\ ! 0 \!\!\!&\!\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-4 0 \!\!\!&\!\!\ ! \etam^{3 } \gs_{4}^{1 } \!\!\!&\!\!\ ! 0 \!\!\!&\!\!\ ! \etam \gs_{4}^{0 } \!-\ ! \sqrt{4}\,\epspot \!\!\!&\!\!\ ! -\pshift\,\epspot \!\!\!&\!\!\ ! \etap \gs_{5}^{0 } \!-\ ! \sqrt{5}\,\epspot \!\!\!&\!\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-5 \etam^{5 } \gs_{5}^{2 } \!\!\!&\!\!\ ! 0 \!\!\!&\!\!\ ! \etam^{3 } \gs_{5}^{1 } \!\!\!&\!\!\ ! 0 \!\!\!&\!\!\ ! \etam \gs_{5}^{0 } \!-\ ! \sqrt{5}\,\epspot \!\!\!&\!\!\ ! -\pshift\,\epspot \!\!\!&\!\!\ ! \ddots \\[0.ex ] % % % % % % % % % % % % % % % % % % line-6 \ddots \!\!\!&\!\!\ ! \ddots \!\!\!&\!\!\ ! \ddots \!\!\!&\!\!\ ! \ddots \!\!\!&\!\!\ ! \ddots \!\!\!&\!\!\ ! ! \ddots \end{array } \right ) % % % % % % % % % % % % % % % % % % \ ] ] where @xmath211 is evaluated at @xmath680 , with @xmath516 . these matrices enjoy properties opposite to those of @xmath670 . the substrate potential does appear ( multiplying all elements in front of the matrices ) while the force is absent . besides , the matrices @xmath677 are independent of the index @xmath131 , which permits to write them simply as @xmath681 . the only technical complication with respect to the common choice @xmath107 is that we need to compute the kummer functions @xmath102 included in @xmath211 , which in our case are simple polynomials [ equation ( [ assoc : laguerre : poly : kummer ] ) ] . note that to compute the observables we must eventually undo the momentum shift @xmath682 . in this appendix we derive some semiclassical formulae discussed in the main text . in section [ ratqbm ] we invoked the current through a saw - tooth barrier deformed by @xmath683 ( figure [ fig : saw - width ] ) . ( b ) and by @xmath684 ( c ) . right panel : width of the energy bands @xmath685 in a cosine potential ( exact and various approximations ) . , title="fig : " ] ( b ) and by @xmath684 ( c ) . right panel : width of the energy bands @xmath685 in a cosine potential ( exact and various approximations ) . , title="fig : " ] we introduced the reflexion coefficients @xmath359 and @xmath360 for @xmath312 . one has to analyse separately the cases of incident particles with energies above or below the barrier . in the first case , the reflexion coefficient by a potential with a cusp ( with slopes @xmath355 and @xmath356 ) is given by the first equation ( [ rs ] ) @xcite . owing to the right slope in figure [ fig : saw - width ] is infinity , we make it finite and finally take the vertical limit , getting @xmath686 next we consider the case of tunnelling through the barrier . the semiclassical transmission coefficient is also written in ( [ rs ] ) . defining @xmath687 , we calculate @xmath688 and @xmath689 for positive and negative forces . after simple integrals , we find @xmath690 with @xmath357 the particle energy ( cf . reference @xcite ) . for moderate forces , @xmath691 , we get for the transmission coefficients @xmath692 $ ] @xmath693 thus , for @xmath357 below the barrier @xmath363 , while @xmath361 for @xmath357 above the barrier . in order to get the semiclassical transmision coefficient through a potential @xmath405 , we need to do the integral @xmath694 } \big| % % % % % % % % % % % % % % % % % % % \;,\ ] ] with turning points obeying @xmath695 . due to @xmath696 below the barrier , we change the sign inside the square root extracting an @xmath697 which is absorbed by the modulus . simple tranformations bring the integral into a tabulated form ( * ? ? ? * equation ( 2.576 ) ) , and we obtain @xmath698 % % % % % % % % % % % % % % % % % % \ ; , \qquad % % % % % % % % % % % % % % % % % % \varepsilon = e/\vo \ ; , \quad s=\sqrt{\half(1-\varepsilon ) } % % % % % % % % % % % % % % % % % % % \;,\ ] ] where @xmath699 while @xmath700 and @xmath701 the first and second complete elliptic integrals ( * ? ? ? insertion in @xmath692 $ ] gives the transmision coefficient through a cosine potential . this @xmath353 also gives the _ semiclassical energy bands _ @xcite @xmath702 with @xmath703 the oscillation frequency in the wells . the width of the bands is therefore @xmath704 \big\ } % % % % % % % % % % % % % % % % % % \ ; , \qquad % % % % % % % % % % % % % % % % % % \varepsilon = e/\vo % % % % % % % % % % % % % % % % % % \;.\ ] ] the parameter @xmath705 goes from @xmath202 when @xmath706 ( potential bottom ) to @xmath336 at the barrier top @xmath707 . for @xmath708 we can expand the elliptic integrals , getting @xmath709 \big\ } \nonumber\\ % % % % % % % % % % % % % % % % % % & + ( 1+\varepsilon)^{2 } \big\ { 1 + \case{1}{8 } \ln \big [ \case{1}{32 } ( 1+\varepsilon ) \big ] \big\ } \big ) \big ] % % % % % % % % % % % % % % % % % % \;.\end{aligned}\ ] ] at the potential bottom @xmath706 this gives @xmath710 $ ] . the band width increases exponentially with the energy . near the top , @xmath711 , one has @xmath712 , and we can use the series expansion of the elliptic integrals , obtaining @xmath713 % % % % % % % % % % % % % % % % % % \;.\ ] ] figure [ fig : saw - width ] shows that equations ( [ wb : approx : bottom ] ) and ( [ wb : approx : top ] ) approximate well the exact results in their respective ranges ( the later also serves as un upper bound ) . if we disregard the terms with @xmath714 in ( [ wb : approx : bottom ] ) , the result happens to fit remarkably @xmath685 in all the range . in this last appendix we shall solve the following generic differential equation @xmath715 perturbatively in the forcing @xmath716 . ( for cosine forcing @xmath717 . ) comparing with section [ sec : dynamics ] , we see that @xmath718 corresponds to the vectors @xmath719 , @xmath720 to the static part of the matrices @xmath158 , and the perturbation @xmath721 to @xmath722 , the part of @xmath670 involving @xmath386 . a periodic solution can be fourier expanded as @xmath723 with @xmath724 the unperturbed response , while @xmath725 for complex @xmath718 . separating terms oscillating with @xmath726 and @xmath727 , the left - hand side of ( [ eom : generic ] ) reads @xmath728 \e^{+\iu\omega_{\iw } t } \\ % % % % % % % % % % % % % % % % % % & + & \sum_{\iw=1}^{\infty } \dfm^{\iw}\ , \big [ -\iu(\omega_{\iw}\tau)\tilde{\ec}^{(\iw ) } + \rel\,\tilde{\ec}^{(\iw ) } \big ] \e^{-\iu\omega_{\iw } t } % % % % % % % % % % % % % % % % % % \;.\end{aligned}\ ] ] to obtain the right - hand side , we use @xmath729 , redefine the indices ( keeping the same names ) to get all the oscillating factors at @xmath730 , and introduce the definition @xmath731 , arriving at @xmath732 \e^{+\iu\omega_{\iw } t } \\ % % % % % % % % % % % % % % % % % % & + & \sum_{\iw=1}^{\infty } \dfm^{\iw } \big [ \vpert\,\tilde{\ec}^{(\iw-1 ) } + \left(\dfp\dfm\right ) \vpert\,\tilde{\ec}^{(\iw+1 ) } \big ] \e^{-\iu\omega_{\iw } t } % % % % % % % % % % % % % % % % % % \;.\end{aligned}\ ] ] equating terms with the same oscillating factors at both hand sides ( _ uniqueness _ of the fourier expansion ) , we obtain @xmath733 \\[1.ex ] % % % % % % % % % % % % % % % % % % \iu(\omega_{\iw}\tau)\ec^{(\iw ) } + \rel\,\ec^{(\iw ) } & = & \vpert\ , \big[\ ; \ec^{(\iw-1 ) } & + & \left(\dfp\dfm\right ) \ec^{(\iw+1 ) } \big ] % % % % % % % % % % % % % % % % % % \end{array } % \;,\ ] ] the @xmath734 are included as @xmath735 with @xmath736 . the contribution @xmath737 comes from products of rotating " terms @xmath738 . for instance , the oscillating factor @xmath739 in @xmath740 when multiplied by the part @xmath741 of the field raises the harmonic from @xmath742 to @xmath743 . the term @xmath744 comes from products of counter - rotating " terms @xmath745 . for example , the product of the oscillating factor @xmath746 and the @xmath747 part of the field lowers the order from @xmath748 to @xmath743 . this contribution ( the contamination " from higher harmonics ) has an order higher in @xmath716 than the contribution of @xmath740 , which is therefore the leading one . indeed , expanding the @xmath749 in powers of @xmath716 we get the leading terms at each harmonic @xmath750 the @xmath751 equation gives the static response [ corresponding to ( [ 3rr : statics ] ) ] , @xmath388 the linear dynamical response , etc . [ corresponding to equation ( [ 3rr : harmonics ] ) ] . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: one of the most important developments in the field of interstellar gas dynamics during the last half - century was the renewed perception that most processes and structures are strongly affected by turbulence . this is a paradigm shift unparalleled in many other fields of astronomy , comparable perhaps to the discovery of extrasolar planets and cosmological structure at high redshift . interstellar turbulence and its implications were commonly discussed in the 1950s , but without the range of observations that are available today , the theory of the interstellar medium ( ism ) drifted toward increasingly detailed models based largely on a preference for thermal and dynamical equilibria . standing apart were two subfields that continued to include turbulence as the primary driver of all observations : radio scintillation and cosmic ray transport . the connections between these subfields and the prevailing picture of the ism were mostly ignored because of the large discrepancy in scales . consequently , two other small - scale processes got lost in the equilibrium paradigm of the large - scale models : mixing of the elements and molecular chemistry . in this review , we discuss these four subfields in some detail . all have a well - established observational base going back several decades , but because of the complexity of turbulence and the infancy of the relevant theory , all of them are now in a state of rapid evolution . our previous review _ interstellar turbulence i _ ( this volume ) , emphasized observations in the dense neutral ism and discussed in detail the various theoretical approaches to this field . the degree to which turbulence and other transport processes mix or homogenize the gas in the face of repeated local pollutions by supernovae and other sources of fresh metals could provide an important constraint on the hydrodynamics of the ism . this general problem is called passive scalar turbulence because the particles being transported have no effect on the velocity field . a review of experiments involving incompressible turbulent scalar transport can be found in warhaft ( 2000 ) , who emphasized that passive scalar probability density functions ( pdfs ) usually have strong exponential tails . a review focussing on a particular class of theoretical approaches has been given by falkovich , gawedzki & vergassola ( 2001 ) , whereas reviews of methods to calculate the concentration pdf and its moments are in dopazo ( 1994 ) and obrien ( 1980 ) . briefer and more specialized treatments can be found in shraiman & siggia ( 2000 ) and pope ( 2000 , section 12.4 ) . here we summarize observations of abundance variations in stars and gas , and we review recent theoretical work on interstellar turbulent mixing . we do not discuss star - to - star variations in globular clusters that may be affected by stellar evolution ( yong et al . 2003 ) . recent observations of metallicity dispersions in stars of the same age and in interstellar gas suggest that the ism is very well mixed on average , with typical fluctuations about the mean of only 520% . this low value is a challenge to explain considering the spotty pollution of new elements from supernovae and other sources . in addition , there are much larger observed fluctuations in a few places , suggesting the overall distribution function for abundance has a long , non - gaussian tail . both the low dispersion and the long tail might be explained by interstellar turbulence . stars of a similar age have often shown large variations in metallicity ( a term usually referring to iron abundance relative to the sun ) . for example , edvardsson et al . ( 1993 ) and chen et al . ( 2000 ) found factor - of - four abundance variations in several elements for f and g main sequence stars . after correcting for selection effects , edvardsson et al . estimated that the dispersion at a given age is about 0.15 to 0.20 dex , i.e. , 4060% , whereas garnett & kobulnicky ( 2000 ) suggested it is smaller than 0.15 dex for the same sample . a large unbiased sample of over 550 hipparcos stars with homogeneous chromospheric age estimates and photometric metallicities showed a dispersion at a given age of only 0.13 dex ( rocha - pinto et al . reddy et al . ( 2003 ) studied 27 elements in 181 f and g stars and found a dispersion at a given age that is somewhat smaller than in edvardsson et al . ( 1993 ) and chen et al . feltzing & gonzales ( 2001 ) used an inhomogeneous sample of 5828 stars and found a larger scatter in the metallicity at a given age of 0.2 to 0.3 dex , which exceeds the observational errors . a major problem is that the ages of the stars used in these studies have large uncertainties . considering that the stellar orbits probably migrated and led to additional abundance variations from galactic radial gradients ( e.g. , wielen et al . 1996 ) , the stellar results are consistent with a dispersion in the gas of 2030% or less , the upper limit set by uncertainties inherent in field star samples . other evidence for homogeneity among field stars comes from elemental abundance ratios . reddy et al . ( 2003 ) estimated the intrinsic scatter in the ratio [ x / fe ] for element x by fitting a gaussian curve to the histogram of deviations between [ x / fe ] and the linear fit of [ x / fe ] versus [ fe / h ] . the standard deviations in the gaussian for 26 elements in 181 stars varied between 0.03 and 0.10 dex . these deviations are comparable to the errors , so the result is an upper limit to the intrinsic scatter . it does not depend on stellar age , as do the absolute abundances . reddy et al . noted that because the elements come in differing proportions from different sites of stellar nucleosynthesis , the lack of scatter implies all the ejecta were well - mixed in the gas before the stars formed . abundance fluctuations of 0.050.15 dex had also been found in studies of open clusters ( e.g. , friel & boesgaard 1992 , twarog et al . 1997 , carraro et al . 1998 ) , b stars in star - forming regions ( e.g. , cunha & lambert 1994 ) and clusters ( rolleston et al . 1994 ) , the magellanic clouds ( e.g. , olszewski et al . 1991 ) , the interstellar medium ( meyer et al . 1998 ) , galactic _ regions ( deharveng et al . 2000 ) , and _ h ii _ regions in disk ( see review in henry & worthey 1999 ) and dwarf irregular ( e.g. , thuan et al . 1995 ) galaxies . however , f stars in nearby open clusters ( friel & boesgaard 1992 and references therein ) show no evidence for star - to - star abundance variations at the level of detectability . cluster - to - cluster variations are also small , including clusters of all ages , and essentially zero ( less than 1020% ) for the four clusters younger than @xmath0 years . the latter result implies that the ism is mixed , at least to this level , for scales between @xmath11 pc ( the size of the clumps from which clusters form ) and @xmath1100 pc ( the distances between clusters ) on a timescale less than @xmath0 years . even more stringent constraints were found from the photometric metallicity study of 76 clusters by twarog et al . comparison of high - precision photometry with stellar evolution models suggested a metallicity spread less than @xmath10.03 dex in the hyades cluster ( quillen 2002 ) . paulson , sneden & cochran ( 2003 ) derived the same small [ fe / h ] spread , excluding three outlier stars , using spectroscopic abundance determinations for individual stars . these results imply fluctuations of less than @xmath17% in the gas from which the hyades cluster formed . interstellar absorption studies suggest equally low dispersions . meyer et al . ( 1998 ) , moos et al.(2002 ) , and andre et al . ( 2003 ) found oxygen abundance dispersions over distances of around one kpc that are only a few percent . a recent study of interstellar kr / h by cartledge , meyer & lauroesch ( 2003 ) found a spread about the mean of only 0.06 dex for lines of sight within the local orion spur ; this was less than the measurement error . we conclude that the most recent evidence , based on field stars of a given age , cluster stars , and the ism , suggests a metallicity dispersion that is very small , less than 20 to 30% and perhaps as small as a few percent . this small galactic dispersion is consistent with the study of 41 _ h ii _ regions in m101 by kennicutt & garnett ( 1996 ) . there are notable exceptions , though , including : ( _ a _ ) the existence of old super - metal - rich stars ( e.g. , feltzing & gonzalez 2001 ) , and a large range in abundance ratios among metal - poor stars ( fields et al . 2002 ) ; ( _ b _ ) the lack of an age - metallicity relation for five nearby clusters , which led boesgaard ( 1989 ) to conclude that local abundance fluctuations of @xmath10.20.3 dex must persist for at least @xmath2@xmath3 years ( but see friel & boesgaard 1992 for a revised result ) ; ( _ c _ ) the factor of 2 to 3 fluctuations in [ o / h ] among subgroups in orion ( cunha & lambert 1994 ) , which , however , may be a result of self - pollution ; ( _ d _ ) the study of b stars in several clusters at galactocentric distance @xmath113 kpc that indicates one cluster has [ fe / h ] larger than the rest by a factor of @xmath15 ( rolleston , dufton & fitzsimmons 1994 ) , ( _ e _ ) two stars in the carina nebula ( andre et al . 2003 ) that are separated by only a few hundred parsecs but have line - of - sight oxygen abundances different by about 40% , and ( _ f _ ) outlier stars in the cluster studies discussed earlier . these examples suggest a metallicity pdf with a small variance and possibly large higher moments , i.e. , a fat tail at large metallicities . such a pdf is similar to that found for the pdf of passive scalars in incompressible turbulence ( warhaft 2000 ) . the analogous pdf for a compressible , magnetic , and self - gravitating ism is unknown . theoretical investigations of abundance fluctuations have until recently nearly all been based on order of magnitude arguments involving characteristic spatial scales and timescales for various processes . two of the more thorough discussions are in roy & kunth ( 1995 ) and tenorio - tagle ( 1996 ) . regarding spatial scales , elmegreen ( 1998 ) pointed out that for a hierarchically structured ism with star formation operating on a local dynamical time , the size of an o and b star association is usually larger than the supernovae remnants it produces . unless there is some immediate mixing , as in hot ionized or evaporating flows , the new elements should be spotty and variable for subsequent generations of stars ( self - enrichment ) . the point about timescales was highlighted by reeves ( 1972 ) : the mean time for a gas element to get reincorporated into a star is shorter than the time for its dispersal by supernova mixing , so abundance fluctuations should be large . both results contradict the apparent uniformity of abundances inside and between clusters and in the general ism . estimates for mixing times have changed since reeves ( 1972 ) paper . edmunds ( 1975 ) found that the timescale for mixing of gas by turbulence and galactic shear is relatively short compared with the star formation time , provided the sources of metals do not cluster together making the dispersion more difficult ( kaufman 1975 ) . bateman & larson ( 1993 ) also derived a short mixing time , considering diffusion by cloud motions , as did roy & kunth ( 1995 ) , who applied the diffusion approximation to galactic shear , radial flows , turbulent diffusion , supernovae , and gas instabilities . similarly short timescales can be derived using structure functions by considering the separation of initially nearby lagrangian fluid particles ( frisch 1995 ) . this latter procedure is often used to derive the richardson 4/3 law of superdiffusive incompressible turbulence . the implication of these short mixing times is that the ism should be homogeneous . the expectation of large spatial fluctuations may be viewed in a different way too , by considering the number of overlapping regions of contamination . edmunds ( 1975 ) showed that @xmath4 supernovae in a local volume should lead to inhomogeneities of order @xmath5 if there is no subsequent mixing . adopting an average of @xmath6 , he suggested that fluctuations would be relatively small . more detailed models for overlapping contaminations were studied by roy & kunth ( 1995 ) , argast et al . ( 2000 ) and oey ( 2000 ) . oey ( 2003 ) used turbulent dispersal from bateman & larson ( 1993 ) to enlarge the effective sizes and reduce the average metallicities in the overlapping regions . low metallicity stars and galaxies should have large metallicity dispersions because the numbers of contaminating events are small ( audouze & silk 1995 ) . a problem with most of the timescale arguments is that they apply to turbulent transport and not homogenization . moving clouds around or shredding them with fluid instabilities does not homogenize the gas at the atomic level or change the frequency distribution of abundance concentrations ; viscosity and molecular diffusion are required for that . the same is true for incompressible turbulence imagined as a cascade of vortices . unless the advection process creates steep gradients on small scales , molecular diffusion can not occur in a reasonable time . the situation changes when there are large - scale gradients in the mean concentration ; then advection from distant regions can change the concentration distribution locally ( see equation [ eq : zave ] below ) . even so , homogenization requires molecular diffusion . thus , a small transport time in the ism does not imply the gas will homogenize enough to remove abundance differences before it forms stars and clusters . the size and placement of each protostellar core within the complex spatial pattern of abundance fluctuations will determine its metallicity . releasing metals into the interstellar turbulent gas is similar in basic respects to dropping a blob of ink into a turbulent fluid . figure [ fig : mix ] shows such a release into a two - dimensional ( 2d ) incompressible forced chaotic velocity field ( jullien et al . advection spreads the scalar out in space , while concentrating it locally through stretching and folding , shearing it into thin ribbons until the spatial gradients are large enough that molecular diffusion becomes significant in the third frame . by the final frame , diffusion has smeared out most of the striated structures . how does turbulence deliver gradients to small enough scales for molecular diffusion to cause homogenization in a reasonable time ? de avillez & mac low ( 2002 ) found a surprisingly long timescale for the decay of the variance in a passive scalar field that had an initial checkerboard pattern and was mixed by supernovae . their time scaled to a galaxy ( at least a few times @xmath2 years ) is larger than previous estimates and still is a lower limit because they did not include sources of new metals . de avillez & mac low also found that the mixing time is independent of the mixing length , unlike mixing modeled as diffusion , and that the mixing time decreases with increasing supernova rate as a result of the increased velocities . there may be ways other than turbulence to get gradients on small enough scales for molecular diffusion to homogenize . in tenorio - tagle s ( 1996 ) `` fountain with a spray '' model , metal - rich droplets falling onto the disk are subject to rayleigh - taylor instabilities that reduce their size even further ; diffusion will be very effective in such a situation . there are many methods for studying passive scalar turbulent transport . here we describe two approaches : prescribed artificial velocity fields and closure methods for studying restricted properties of turbulent transport . most of what is known about passive scalar turbulence comes from models using an idealized velocity field that is not a rigorous solution to the hydrodynamics equations ( see review by falkovich et al . an example would be a time - uncorrelated velocity field with gaussian statistics , a prescribed spatial correlation function , and a constraint of incompressibility . this is called a kraichnan velocity field ( e.g. , shraiman & siggia 2000 ) . the advantages of such a field are that the equations for the concentration field and its statistical properties can in some cases be solved analytically without a closure assumption , and the generic properties of the result might be valid for any velocity field that has the same variance and conservation constraints . however , there is a large danger in assuming a delta function for the time correlation function . large intermittency in the velocity gradient field means that a large fraction of the fluid contains large gradients compared with a gaussian . in incompressible turbulence ( and the ism see _ interstellar turbulence i _ ) , this fraction gets larger on smaller scales . this implies that intermittency should homogenize the gas faster than purely gaussian motions , because molecular diffusion depends on large velocity gradients at small scales . decamp & le bourlot ( 2002 ) used another model velocity field that was obtained from a wavelet reconstruction with statistical properties similar to the ism velocity field , including motions correlated in space and time . they solved the continuity equations and showed that the correlated velocity field has a faster dispersion and a larger standard deviation for passive scalars ( by factors of 2 to 4 ) than an uncorrelated gaussian velocity field . the spatial distributions for species undergoing chemical reactions were also different , suggesting a mechanism for molecular segregation in dense clouds ( section [ sect : chem ] ) . further work on turbulent mixing using synthetic velocity fields is in elliott & majda ( 1996 ) , fung & vassilicos ( 1998 ) , and boffetta et al . ( 1999 ) . for some purposes , it may be sufficient to understand how the mean , variance , and skewness of a metallicity distribution depend on the source terms and large - scale gradients . this is the moment approach , and it has the advantage that some of its results can be obtained analytically . the moment approach might explain , for example , why there is an inverse square root scaling on the supernova rate for the decay timescale of the abundance variance ( de avillez & mac low 2002 ) . the evolution equation for the one - point probability distribution of metallicities was derived rigorously for incompressible turbulence by dopazo , valino & fueyo ( 1997 ) , who also derived the moment equations for incompressible flow ( see also obrien 1980 ; chen & kollman 1994 ; dopazo 1994 ; pope 1994 , 2000 ) . perhaps the simplest question that can be asked about scalar turbulence concerns the way in which the average relative concentration @xmath7 of a local patch of contaminant ( relative to hydrogen , for example ) will spread in space as a function of time . the usual place to start is to examine the mean square distance , @xmath8 , traveled by a marked lagrangian fluid particle . this distance can be related to an integral over the lagrangian correlation function , assuming only statistically stationary and isotropic turbulence . without knowing the correlation function , however , this relation is of little use . an approximation is to use the fact that the correlation is unity at time zero and approaches zero for large separations . then for times much smaller than the correlation time @xmath9 , @xmath10 whereas at @xmath11 , @xmath12 . here , @xmath13 is the two - time correlation function at a given point in the flow for lag @xmath14 ; i.e. , @xmath15 for flow speed @xmath16 and the average @xmath17 is over all pairs of points separated by @xmath14 . normalization to the variance , @xmath18 , makes @xmath13 dimensionless with values between @xmath19 and @xmath20 . in this formulation , the particle behaves ballistically at short times and has an uncorrelated random walk at large times . these results are usually referred to as taylor s theorem ( taylor 1921 ; see mccomb 1990 ; pope 2000 ) . the result is important for illustrating the nature of turbulence as a correlated random walk and the analogy between turbulent and molecular diffusion , where in the latter case @xmath21 would be the mean collision time . klessen & lin ( 2003 ) recently showed that taylor s theorem is correct in 3d - compressible turbulence . mccomb ( 1990 ) , piterbarg & ostrovskii ( 1997 ) and others have pointed out that taylor s theorem does not solve the problem of turbulent dispersal and transport , even for the mean field because what is required is information in the lab frame , not following the particle . klessen & lin ( 2003 ) suggest that the result can be used to develop a phenomenological mixing - length or diffusion model for turbulent transport in the ism in which the integral correlation time is related to the mean time between shock passages . another fundamental relationship is richardson s law ( richardson 1926 ) for the rate at which pairs of points separate in space in the inertial range of turbulence . one form of this law states that the time dependence of the rms separation is @xmath22 ( falkovich et al . 2001 , boffetta & sokolov 2002 , nicolleau & vassilicos 2003 ) . work on pair dispersion using synthetic velocity fields has resulted in at least two important insights : ( _ a _ ) the dynamics of particle pairs is sensitive to initial conditions but their rms separation increases algebraically , not exponentially ( unlike low - dimensional chaotic systems ) . ( _ b _ ) particle pairs travel together for long times and then separate explosively when they encounter straining regions around hyperbolic points in the flow ( e.g. , nicolleau & vassilicos 2003 ) . a traditional approach that could be applied to the spreading of newly produced elements from sources in the ism is to ignore the detailed dynamics and use an equation for the evolution of the mean concentration field alone . in general , the concentration @xmath23 of an element evolves in a given velocity field @xmath24 as @xmath25 where @xmath26 is the mass diffusivity and @xmath27 is the rate at which sources inject the element . the second term on the left represents the effect of the turbulent velocity field in advecting the concentration field . for the interstellar medium this velocity field will consist of complex compressions and vortical motions . the first term on the right represents molecular diffusion , which operates only on small scales in regions where the gradient of the concentration field is large . this equation illustrates how turbulent advection of a passive scalar spreads it out in space while increasing the local gradients through stretching , compressing , and folding . molecular diffusion then operates where the gradients are high . timescales for molecular diffusion have been discussed recently by oey ( 2003 ) for a uniform gas . the result is sensitive to temperature , so most of the homogenization of elements in the ism may occur in the hot regions , like superbubbles ( silich et al . 2001 ) , as well as those with the steepest velocity gradients . to obtain the time evolution of the average concentration , one might choose to ignore all the details of the turbulence and spatially average equation [ eq : z ] . ignoring also the molecular diffusion and source terms , we get @xmath28 for mean concentration and velocity @xmath7 and @xmath29 and fluctuating parts @xmath30 and @xmath24 . the last term is zero for incompressible turbulence , leaving only @xmath31 , which contains the unknown ensemble average of the scalar flux , @xmath32 , representing interactions between the fluctuating parts of the velocity and concentration fields . this is a classic closure problem . we could derive a differential equation for the scalar flux , but it would contain more unknown correlations , including triple correlations , and further manipulation only leads to higher and higher order unknowns . in the incompressible case , the simplest treatment is to imagine that the turbulent transport behaves diffusively , like molecules in a gas . then the scalar flux is proportional to the mean concentration gradient and given by @xmath33 where @xmath34 is a scalar diffusivity given by some timescale times the square of some velocity . often these are taken to be the correlation time and the rms velocity , although a variety of other forms have been used to fit specific experimental results ( piterbarg & ostrovskii 1997 ) . using equation [ eq : flux ] , equation [ eq : zave ] becomes the usual diffusion equation , and the patch of concentration , if initially a point source , would spread according to a gaussian whose width grows @xmath35 . in fact , turbulent transport can not be described by classical diffusion because triple correlations are as important as quadratic correlations , or , more physically , turbulent motions are not random independent steps but correlated bulk transports over a wide range of scales . there have been many attempts , all unsuccessful to various degrees , to develop closure techniques for passive scalar turbulence . many of these are described in reviews by obrien ( 1980 ) and dopazo ( 1994 ) . perhaps the most popular in the physics community is the mapping closure suggested by chen , chen & kraichnan ( 1989 ) . a more complicated closure method is based on the eddy - damped quasinormal markovian approximation ( lesieur 1990 ) . an application of this closure to the passive scalar problem with reference to earlier work is in herr , wang & collins ( 1996 ) . a promising approach that is not too complex yet captures the importance of the triple correlations was proposed by blackman & field ( 2002 ) in connection with turbulent dynamo theory and then applied to the scalar turbulence problem ( blackman & field 2003 ) . the idea is to model the time derivative of the scalar flux as a normal diffusion term with an eddy viscosity plus a damping term @xmath36 , where @xmath14 is a relaxation time . this approach is similar to a closure method used in terrestrial atmospheric turbulent transport ( lewellen 1977 ) . when this is substituted in the equation for @xmath7 , a little manipulation results in a damped wave equation for the mean concentration . brandenburg , kapyla & mohammed ( 2004 ) did extensive comparisons with numerical simulations of passive scalars in mildly compressible turbulence ( rms mach number @xmath10.2 ) and found good agreement with this model for the time evolution of the spreading size and the non - gaussianity ( kurtosis ) of @xmath7 . the moment approach can derive the variance of the abundance distribution directly from equation [ eq : z ] by writing the velocity , abundance , and source fields as the sum of mean and fluctuating parts ( e.g. , @xmath37 ) , multiplying the resulting equation for @xmath38 by @xmath30 , and then ensemble averaging to obtain @xmath39 where @xmath40 is the fluctuating part of the source term ( @xmath41 ) . the first term on the right shows how the scalar flux ( @xmath42 ) interacts with any large - scale gradient in the abundance , increasing variance if @xmath42 is proportional to a mean gradient ( as in equation [ eq : flux ] ) . this is analogous to how stellar orbit diffusion in the presence of a radial gradient produces fluctuations in the stellar abundances . the second term indicates that any gradient in the mean square fluctuations will be advected with the mean flow ( e.g. , radial flows or infall ) . the third term in equation [ eq : z2 ] shows how turbulence can affect a concentration variance even with no gradient in @xmath7 , zero mean velocity , and without diffusion . writing this term as @xmath43 shows that it consists of the mean divergence of the flux of scalar variance , as might be expected , minus a term that is only nonzero in the presence of compressibility . differentiation and ensemble averaging commute , so the first term of these two is zero for a homogeneous fluid because ensemble averages do not depend on spatial position . the fourth term represents the net production of scalar variance by the fluctuating nucleosynthetic sources . `` diffusion '' in this context is shorthand for the fairly complex set of second - order derivatives involving mass diffusion at the molecular level . as a simple illustration , assume there are no large scale flows and the abundance is linear in galactocentric distance with constant of proportionality @xmath44 . for a general gradient term , we would have the triple correlation whose closure is discussed by blackman & field ( 2003 ) and brandenburg , kapyla & mohammed ( 2004 ) . then we can also derive an equation for the correlation @xmath42 by multiplying equation [ eq : z ] by @xmath24 and ensemble averaging . the result is , for @xmath45 , @xmath46 the first term on the right , which is @xmath47 for an isotropic velocity field , represents the production of correlations by interaction of the velocity fluctuations with the mean metallicity gradient . we can assume that we are given some steady - state isotropic turbulent velocity field so that @xmath48 is constant . the second term on the right , often called pressure scrambling , is a serious closure problem for compressible turbulence . the source term is important if the metals come from young stars whose formation is associated with the turbulence itself , but for type i supernovae or any other source that has time to decorrelate from the flow , @xmath49 . this indicates the importance of using a consistent model for the formation of metal - producers from the turbulent field . equations [ eq : z2 ] and [ eq : zu ] can now be solved to give the evolution of the standard deviation of the abundance distribution for a particular velocity and source field . if we neglect pressure scrambling in equation [ eq : zu ] and the compressibility term in equation [ eq : z2 ] , and also replace the triple correlation term with a simple gradient - diffusion closure , then the result would show that the concentration fluctuation variance increases @xmath50 , even in the absence of sources . the turbulence is transporting high - z gas into low - z gas ( and vice versa ) due to the existence of the gradient . obviously , after some time the large fluctuations imply steep gradients at very small scales , and so the microscopic diffusion term will cause the variance to level off . in addition , the gradient itself will be washed out by the turbulence unless there is some mechanism continually producing it . without a sustained mean gradient , such equations could be used to understand how the variance in elemental fluctuations should decrease with time under the action of only advection and diffusion , giving a simplified model to understand the numerical decay results of de avillez & mac low ( 2002 ) . with some assumed closure for the source terms , or with prescribed velocity field statistics , the solution of equation [ eq : z2 ] would show how the abundance variance depends on the source rate . the source terms in the transport equations depend on how the metals in expanding supernovae and superbubbles are released into the turbulent background . this is not a trivial problem , as discussed by tenorio - tagle ( 1996 ) . simulations of superbubbles ( korpi et al . 1999 ) and supernova remnants ( balsara et al . 2001 ) expanding into a turbulent ism may provide guidance . some metal - producing outflows should be jet - like , leading to the release of metals by shear flow along the walls of a chimney . begelman & fabian ( 1990 ) and slavin , shull & begelman ( 1993 ) proposed that such turbulent layers could mix hot and warm gas , but they could just as well mix freshly - produced metals . tenorio - tagle ( 1996 ) suggested that superbubbles blow out of the disk and rain down metal rich droplets that break apart by rayleigh - taylor instabilities . this process could disperse metals over several kpc in the disk with no assistance from disk turbulence . these processes are also relevant for the redistribution of elements ejected by galactic blowout into the intergalactic medium ( igm ) . several papers ( e.g. , ferrara et al . 2000 , aguirre et al . 2001 and references therein ) have noticed that it is difficult to mix metals on scales comparable to the mean distance between galaxies , requiring an additional ( unknown ) mechanism . turbulence in the igm could provide such a mechanism . the advection of the grains by turbulent gas can lead to spatial clustering relative to the gas , and to a spatially inhomogeneous dust - to - gas ratio . the degree of segregation depends on the grain size . local variations in grain - grain velocities affect grain growth and shattering . early estimates for the efficiency of this process were given by cameron ( 1973 ) , burke & silk ( 1976 ) , scalo ( 1977 ) , and draine ( 1985 ) , but the conception of how grain advection occurs differs greatly in these papers . more rigorous derivations , aimed at grain collisions in protostellar disks but also applicable to the ism , were given by vlk et al . ( 1980 ) and markiewicz , mizuno & vlk ( 1991 ) . recent work has been directed toward trapping grains in protoplanetary disks by anticyclonic vortices ( e.g. , bracco et al . 1999 ) , which can persist for many turnover times . large grain concentrations can result , accelerating the formation of planetesimals . grains might also become trapped in the general ism because of drag forces , as proposed by falgarone & puget ( 1995 ) . a review of experiments and simulations on turbulent gas - grain segregation was presented by fessler , kulick & eaton ( 1994 ) . more recent theoretical investigations ( e.g. , sigurgeirsson & stuart 2002 , lopez & puglisi 2003 ) have demonstrated several interesting properties of this clustering . all of this work relies on synthetic velocity fields because inertial particle advection in real turbulent flows would require resolution much finer than the dissipation scale . nevertheless , the essential physics of particle drag is becoming more clear . the particle motion is dissipative even if the advecting flow is incompressible , and hence , dissipationless ( balkovsky , falkovich & fouxon 2001 ) . this leads to simplified treatments in which the stokes drag effect is modeled as a small compressible component to the velocity field of the inertial particles . a detailed study of inertial particle dynamics in incompressible flows ( bec et al . 2003 ) shows that particles smaller than a critical size form fractal clusters whereas larger sizes fill space with a nonuniform density . an extension to highly compressible flows is desirable . the motion of charged dust grains in anisotropic magnetohydrodynamic ( mhd ) turbulence was considered by yan & lazarian ( 2002 ) , who made a distinction between the directions parallel and perpendicular to the field . grains with a gyration time around the field that is shorter than the gas drag time will follow the turbulent field fluctuations in the perpendicular direction . yan & lazarian ( 2003b ) calculated the acceleration of charged grains that resonate with mhd fluctuations . another reason to study stirring by interstellar turbulence is to understand the origin of galactic radial abundance gradients ( carraro et al . 1998 , hou et al . 2000 , rolleston et al . 2000 ) . models typically account for these gradients using a radially dependent timescale for infall and star formation ( e.g. , chiappini et al . 2001 ) . a problem is that turbulence should reduce large - scale gradients , so studies of this turbulent washout could give an estimate of the timescale on which the gradient must be replenished . order of magnitude estimates along the lines of those mentioned in section [ sect : howlarge ] indicate replenishment timescales larger than characteristic infall times , suggesting that contributions from a radially dependent stellar initial mass function ( imf ) or other effects are not required . a peculiar abundance gradient has been observed in the outer galaxy , where clusters ( twarog et al . 1997 ) and cepheids ( caputo et al . 2001 , andrievsky et al . 2002 , luck et al . 2003 ) suggest a sudden decrease in metal abundance by about 0.2 dex setting in at 10 kpc . this discontinuity appears over a radial distance less than 1 kpc . it is hard to see how such a feature could survive turbulent washout for more than @xmath2 years using timescale arguments of the type summarized earlier . either we are missing something fundamental about turbulent transport ( e.g. , maybe differential rotation or magnetism suppress turbulent motions in the radial direction ) , or this metallicity near - discontinuity does not really exist . the coupling of interstellar chemistry to the dynamics of interstellar gas has focused for a long time on collapsing or slowly contracting clouds ( e.g. , gerola & glassgold 1978 ; prasad , heere & tarafdar 1991 ; shematovich et al . 2003 ) and single shocks or mhd waves ( e.g. , hollenbach & mckee 1979 , 1989 ; draine & katz 1986 ; charnley 1998 ; flower & pineau des forts 1998 ) . the collapse models explained why early - time solutions for static chemistry often agreed better with observations than late - time solutions ( e.g. , prasad et al . 1991 ; nejad & wagenblast 1999 ; but see turner et al . the result that nondynamical time - dependent abundances agreed with observations for different ages is a strong motivation for the influence of turbulence because turbulent timescales are often smaller than the chemical equilibrium times . turbulent transport can prevent the chemistry from attaining a steady state ( e.g. , phillips & huggins 1981 ; xie , allen & langer 1995 ; willacy , langer & allen 2002 ) . turbulence can affect ism chemistry in three ways : ( _ a _ ) by continually transporting material between regions with different physical conditions , like ambient ultraviolet ( uv ) radiation flux , temperature , and density , ( _ b _ ) by creating localized heating where temperature - sensitive reactions , especially those involving endothermic reactions , will be enhanced , and ( _ c _ ) by magnetically forcing ions to move faster than the thermal speed where they can enhance the temperature - sensitive ion - neutral reactions . progress in this field is difficult because of limitations in the dynamical models and because of the large number of nondynamical chemical effects that may occur , like neutral - neutral reactions at low temperature , variable c and o depletion , three - body channels for dissociative recombination , the presence of cosmic - ray - induced uv photons , chemical phase transitions , and many more , as reviewed by herbst ( 1999 ) , hollenbach & tielens ( 1999 ) , van dishoeck & hogerheijde ( 1999 ) , and langer et al . ( 2000 ) . for more recent work on chemistry covering a variety of applications not necessarily related to turbulence , see roberts & herbst ( 2002 ) , doty et al . ( 2002 ) , stantcheva & herbst ( 2003 ) , shematovich et al . ( 2003 ) , and charnley & markwick ( 2003 ) . the classic observation that implied turbulent transport was the large amount of carbon in the gas phase of dense molecular cores , considering the short freeze - out time on dust . boland & dejong ( 1982 ) proposed that turbulent circulation could explain this discrepancy by bringing grains to the outer layers for uv desorption . modern observations ( kramer et al . 1999 ) show more depletion , however , and photodissociation region ( pdr ) models suggest carbon comes from clump surfaces at a range of depths ( howe et al . 2000 ) . turbulent transport has other applications in modern chemical models . pijpers ( 1997 ) considered a diffusive model of turbulent grain transport and obtained a long migration time . ragot ( 1998 ) pointed out that turbulent transport is probably neither ballistic nor diffusive , and used a model based on levy flights to return to the smaller timescales . this transport effect is important because the degree to which co is released from grains into the gas can affect the rest of the chemistry ( co deactivates much chemistry by depleting the h@xmath51 ion ; nejad & wagenblast 1999 ) . a detailed study of turbulent transport including 87 species and 1100 reactions was given by xie et al . ( 1995 ) , who assumed that radial flux is proportional to the spatial gradient of a species and used a mixing length model for turbulent diffusion . they found that most carbon - bearing species and several other important molecules were strongly affected by turbulence . the model was extended recently by willacy , langer & allen ( 2002 ) , who showed how the _ h i_/h@xmath52 ratio could provide a sensitive signature of turbulent transport . this occurs because h@xmath52 forms when h reacts on dust grains . the formation rate depends on the density of atomic h and the total dust cross section , whereas the destruction rate depends on the uv field . if turbulence cycles material between regions with different uv optical depths on a timescale comparable to the formation time , then h will penetrate deeper into the cloud and h@xmath52 will be closer to the surface . because many other simple molecules are formed from h@xmath52-based intermediaries ( especially h@xmath51 ) , turbulent transport could control the chemistry for many species . timescales for reaching chemical equilibrium depend on density , temperature , and ionization fraction , and are typically @xmath53@xmath54 years ( van dishoeck & hogerheijde 1999 ) . for the atomic part of a photodissociation region , the longest part of the cycle c@xmath55 ch@xmath56 co@xmath57 c @xmath58c@xmath59 is the formation time of ch@xmath60 through radiative association ( tielens & hollenbach 1985 ) with a timescale @xmath110@xmath61 years for @xmath62 particle density in units of @xmath63 @xmath64 . for turbulent transport , the fastest possible timescale is ballistic , @xmath65 , for turbulent speed @xmath66 and length @xmath67 . using the ism scaling relations for illustration ( see _ interstellar turbulence i _ ) , this minimum time is @xmath68 years . the chemical and transport timescales are comparable at @xmath63 @xmath64 , but the transport time becomes larger than the chemical time at higher density and then turbulence would seem to have little effect . because cosmic ray ionization drives the chemistry , and each ionized h@xmath52 molecule eventually forms co or some other important molecule , the chemical timescale in the cores of dense clouds is the ratio of the relative abundance of co , @xmath110@xmath69 , to the cosmic ray ionization rate , @xmath14@xmath70 s@xmath71 ; this gives @xmath110@xmath72 years ( d.j . hollenbach , private communication ) . another way in which turbulence may affect chemistry is through localized heating in shocks ( flower & pineau des for@xmath73ts 1998 ) , vortices ( joulain et al . 1998 ) , ambipolar diffusion ( padoan , zweibel & nordlund 2000 ) , and magnetic reconnection ( lazarian & vishniac 1999 ) . there are many reactions that seem to require higher than normal temperatures . for diffuse clouds , the large abundance of ch@xmath59 and molecules like hco@xmath59 that can form from ch@xmath59 , and the large abundance of oh ( gredel 1997 , lucas & liszt 1997 ) , all suggest that high temperature ( @xmath110@xmath74 k ) reactions take place , as in shock fronts ( elitzur & watson 1980 , draine & katz 1986 ) . however , the correlation between ch@xmath59 and the lower - temperature molecule c@xmath52 ( gredel 1999 ) , and the similar radial velocities of ch@xmath59 and ch rule out ch@xmath59 production in single large shocks or cloud surfaces . instead , ch@xmath59 could be made in numerous unresolved shocks that blend on a line of sight with cooler gas at the same average velocity ( gredel , pineau des for@xmath75ts & federman 2002 ) . part of the appeal of turbulence is its great range of scales that can yield such an effect . temperature fluctuations of several hundred degrees could play a major role in the chemistry of many species because of the inverse temperature dependence of radiative association and the range of activation energies for collisional dissociation , neutral - neutral reactions , and some ion - molecule reactions . neutral - neutral reactions may be important even in dark clouds ( bettens et al . prasad & huntress ( 1980 ) long ago pointed out the importance of temperature variations in radiative association reactions and some ion - molecule reactions . also , in dense shielded clouds , warm regions drive most of the o , oh , and o@xmath52 to h@xmath52o through reactions like o + h@xmath76 oh@xmath77h and oh+h@xmath76h@xmath52o+h ( e.g. , charnley 1998 ) , not only enhancing the h@xmath52o abundance , but also preventing the formation of molecules like so@xmath52 , which rely on the oxygen species for their formation . the cn abundance should be enhanced because its destruction by reaction with o and o@xmath52 is suppressed ; this can lead to enhanced production of hc@xmath78n through reaction of cn with c@xmath52h@xmath52 . there have been several attempts to model the effects of local turbulent heating on chemistry . black & van dishoeck ( 1991 ) pointed out that if turbulence extends down to the collisional mean free path , then the tail of the particle velocity distribution could be enhanced . reaction rates are averages over the particle relative velocity distribution , so energy - dependent cross - sections could have greatly enhanced rates . spaans ( 1996 ) noted that the distribution of velocity differences in turbulent flows at small scales has a large tail and suggested this could result in non - maxwellian relative speeds between molecules , especially those involved with ch@xmath59 production . this assumption is questionable because the reaction scales are the mean free paths , of order 1au@xmath79 , and these are smaller than the viscous dissipation scale of incompressible turbulence . the validity of the fluid approximation for neutrals is also questionable at these scales ( see _ interstellar turbulence i _ ) , although mhd turbulence in the ionic component could enhance the ion - neutral collision rate . another approach to the problem was presented in a series of papers by falgarone and coworkers . falgarone & puget ( 1995 ) , falgarone et al . ( 1995 ) , and joulain et al . ( 1998 ) considered the diffuse cloud abundances of ch@xmath59 and several other molecules such as oh and hco@xmath59 . they focused on the solenoidal part of the velocity field in which molecular viscosity on 10 au scales produces numerous local hot spots ( up to several times @xmath80k ) . falgarone et al . ( 1995 ) showed that the resulting chemical column densities were consistent with observations . joulain et al . ( 1998 ) used a burgers - vortex model for the dissipation to explain other features , such as the similarity of ch and ch@xmath59 line centroid velocities , when the number of dissipation regions on the line of sight is large . decamp & le bourlot ( 2002 ) approximated turbulence by a stochastic 1d velocity field correlated in space and time , and solved the continuity equation for the density of each species . for a small reaction network , the abundances developed different patterns on different scales , showing how species segregation can occur in a turbulent flow . species that do not normally peak at the same time for a nonturbulent model can coexist in a turbulent model , which means that the age of a region can not be inferred from the abundances . the issue of chemical differentiation is important because it is well - established and unexplained in a number of starless cores such as tmc-1 , l134n , and l1544 , and observed as 110 au variations in co , h@xmath52co , and oh relative to h@xmath52 toward extragalactic background sources ( marscher , moore & bania 1993 ; moore & marscher 1995 ; for more recent work , see pan , federman & welty 2001 ; rollinde et al . 2003 ; stanimirovic et al . although some of these variations may be attributed to freeze - out of molecules on dust grains ( bergin et al . 2002 ) , most are probably also affected by time - dependent chemistry ( langer et al . one important timescale is cloud contraction ( shematovich et al . 2003 ) , but another is turbulent cycling between different physical conditions over a range of timescales ( decamp & le bourlot 2002 ) . the observed variations at scales so small that gravity is unimportant seem to favor turbulence . the calculations by decamp & lebourlet also develop a bistable chemical pattern in which dual equilibria are possible . however , it is not known how many of the effects result from the imposed artificial velocity field , which is not able to react to the changing density field through pressure . the only truly hydrodynamical turbulence simulation that included a small but relevant chemical reaction subset was by pavlovski et al . ( 2002 ) , who generalized previous work by mac low and coworkers to include a large number of coolants and a wide range of possible temperatures and densities . the reaction network in pavlovski et al . included h , c , o , and five molecules formed from them , and it included h@xmath52 formation on grains . they followed the decay of hypersonic turbulence without self - gravity in a dense ( @xmath81 @xmath64 ) , small ( @xmath82 cm ) region . their surprising result was that after h@xmath52 dissociated in a shock , the molecules reformed in a pattern of filaments , clumps , and diffuse gas yet the abundances were fairly uniform after only 100 years . there are probably other processes that produce localized regions of hot chemistry . for example , ambipolar diffusion heating depends on a high - order derivative of the magnetic field and so can produce heating rates that exceed the average rate by orders of magnitude on very small scales ( padoan , zweibel , & nordlund 2000 ) . magnetic reconnection ( lazarian & vishniac 1999 ) is another possibility for localized hot gas . the chemistry induced by these processes has not yet been investigated . cosmic rays scatter off magnetic waves and mhd turbulence as the particles propagate along magnetic field lines , and they generate waves and turbulence if they stream much faster than the local alfvn speed . observations of cosmic rays at the earth and of radio and x - ray emissions produced by cosmic rays at their sources and throughout the galaxy suggest the ism is filled with irregularities on the scale of a particle gyroradius . this ranges from 1 au to 1 pc for protons with energies between a few gev and the knee of the cosmic ray energy distribution at @xmath83 ev . this section reviews the connection between cosmic rays and turbulence . it is not known if the turbulence that scatters cosmic rays has the same origin as the turbulence on larger scales , i.e. , whether it is part of an energy cascade from these larger scales . it could be generated locally , for example , by the cosmic rays themselves or in other small scale instabilities ( see also section 3 , _ interstellar turbulence i _ ) , and it could be more similar to scintillation turbulence , which has the same small scale ( section [ sect : scint ] ) than molecular cloud turbulence , which is presumably generated on much larger scales . other reviews of cosmic rays are in cesarsky ( 1980 ) , berezinskii et al . ( 1990 ) , drury et al . ( 2001 ) , jokipii ( 2001 ) , ptuskin ( 2001 ) , and schlickeiser ( 2002 ) . the important point about cosmic rays from the perspective of this review is that they scatter frequently , diffusing rather than streaming through the galaxy . their total path length determined from the ratio of secondary to primary nuclei is equivalent to @xmath84 galactic disk crossings at gev energies ( engelmann et al . 1990 , higdon & lingenfelter 2003 ) . their flight time determined from isotope ratios ranges from @xmath115 my for a leaky - box model with a small halo ( simpson & garcia - munoz 1988 , yanasak et al . 2001 ) to @xmath2 years for a diffusion model in a @xmath15 kpc halo ( ptuskin & soutoul 1998 ) . they scatter so frequently that their flux at the earth is isotropic to within @xmath85 ( cutler & groom 1991 , munakata et al . 1997 ) , even though there are only a handful of nearby sources ( i.e. , geminga , vela , lupus loop , loop iii see ptuskin & soutoul 1998 , jones et al . 2001 , taillet & maurin 2003 ) . the scattering distance , @xmath86 , is the step size in a random walk over the galactic thickness @xmath87 ( achterberg , blandford & reynolds 1994 ) . for @xmath88 random isotropic scatterings , @xmath89 , so for total distance travelled @xmath90 with @xmath91 particle speed , we get @xmath92 at gev energies . this is a very small distance on galactic scales , but still much larger than a gyroradius , @xmath93 for rest mass @xmath94 , relativistic @xmath95 , and @xmath96 , magnetic field strength @xmath97(gauss ) and charge @xmath98 with @xmath99 esu . in the expression for @xmath100 , the angle of particle motion with respect to the field line is @xmath101 . taking @xmath102 and @xmath103 , we get @xmath104 cm for hadron particle energy @xmath105 in gev and field strength @xmath106 in microgauss . thus , the mean scattering distance for gev energies is @xmath110@xmath72 gyroradii and the gyroradius has the same scale as the solar system . the relative amplitude of the magnetic irregularities @xmath107 that have the size @xmath100 may be determined from the random walk of angular deflections , @xmath108 , which build up in proportion to the square root of the number @xmath109 of gyrations as @xmath110 for small @xmath111 . the scattering frequency is @xmath112 , and the number of gyrations is @xmath113 for gyrofrequency @xmath114 . thus , @xmath115 and @xmath116 when @xmath1170.2 pc . evidently , magnetic fluctuations of less than one percent on 0.1 au scales can isotropize cosmic rays with gev energies ; in general , a wide spectrum of fluctuations is needed to scatter the full energy range . this level of irregularity is much smaller than observed on @xmath118 pc scales from pulsar dispersion measures , which suggest @xmath119 g and @xmath120 g ( rand & kulkarni 1989 ) . it is also smaller than the integrated fluctuations found by scintillation on scales less than @xmath121 pc , which give @xmath122 g and @xmath123 g ( minter & spangler 1997 ) , and smaller than @xmath124 from optical polarization ( heiles 1996 , fosalba et al . 2002 ) . a spectrum of magnetic energy fluctuations like the kolmogorov or kraichnan spectra will make @xmath111 much smaller on the scale of the gyroradius than these observed field irregularities . such a spectrum will also give the required rigidity dependence for the diffusion coefficient ( strong & moskalenko 1998 ) . the rigidity of a cosmic ray is @xmath125 . cosmic rays can also scatter off strong magnetic irregularities having a scale comparable to or larger than r@xmath126 ( jokipii 1987 ) . inverting the above expression and integrating over a spectrum of waves @xmath127 , the mean free path is @xmath128 ( schlickeiser 1989 ) . for @xmath124 and a particle gyroradius that fits inside the compressed region , @xmath129 , this gives @xmath130 . if the scattering arises in a network of strong shocks in a supersonically turbulent medium , then the mean free path for diffusion will be about the shock separation . because the gyroradius of gev protons is less than the collision mean free path for atoms and ions , collisional mhd models of compressible turbulence do not apply to these small scales ( see _ interstellar turbulence i _ and section [ sect : mirror ] below ) . the field line fluctuations that scatter cosmic rays could be coherent waves , stochastic turbulence , or shock fronts . one common view of waves is that they are weak fluctuations compared with the mean field , with wave travel times much longer than an oscillation period and a well - defined dispersion relation between frequency and wavenumber . particles in such a wave field interact successively with many crests , travelling rapidly along the field until their cumulative pitch angle deviation becomes large and they begin to interact with a different wave train . a turbulent medium is not so regular but particle interactions with a broad wave spectrum can be similar as long as the particle sees a magnetic perturbation with the right wavelength after each gyration . in both cases the field can be treated as static for the fast - moving particles . much of the history of cosmic ray theory is based on observations of the solar wind , which contains a mixture of weak and strong fluctuations , depending on frequency ( drge 1994 ; goldstein , roberts & matthaeus 1995 ) . perhaps @xmath115% of it is in the form of slab - like alfvn waves ( @xmath131 ) and @xmath185% is two - dimensional ( 2d ) magnetic turbulence ( @xmath132 , bieber , wanner & matthaeus 1996 ) . here , @xmath133 and @xmath134 are wavenumbers perpendicular and parallel to the mean field . the solar wind observations led to the quasilinear theory of cosmic ray propagation ( jokipii 1966 ) , which assumes the field is uniform in space and unchanging in time over a particle gyration . interstellar applications of this theory may be appropriate at high spatial frequencies where weak electron perturbations scatter radio waves ( section [ sect : scint ] ) , or near shock fronts where weak waves on either side scatter relativistic particles back and forth ( section [ sect : cracc ] ) , or for waves that are excited by the cosmic rays themselves ( chandran 2000a ) . the waves cause cosmic rays to diffuse in both momentum and space ( e.g. , schlickeiser 1994 ) . momentum diffusion means that velocities change and isotropize ; space diffusion means that cosmic rays spread out slowly to more uniform densities . the two diffusion coefficients are inversely related : higher momentum diffusion means lower spatial diffusion as the particles isotropize their velocities more easily . there are two types of waves that could be important for cosmic ray scattering . alfvn waves are transverse oscillations with a restoring force from field line tension and a frequency less than the ion gyro frequency . their dispersion relation between frequency @xmath135 and wavenumber @xmath26 is @xmath136 for angle @xmath137 between the wave propagation vector and the average field . their group and phase velocity parallel to the mean field is the alfvn speed @xmath138 for ion density @xmath139 . when thermal pressure with sound speed @xmath140 contributes to the restoring force there are also magnetosonic waves with the dispersion relation @xmath141^{1/2}\right).\ ] ] the positive and negative signs are for fast and slow waves . for oblique propagation of magnetosonic waves , when @xmath142 , the phase speed of the fast mode can be large along the field lines : @xmath143 in the low @xmath144 limit ( @xmath144 is the ratio of thermal to magnetic pressures , @xmath145 ) and @xmath146 in the high @xmath144 limit . this speed is important for cosmic rays because they also move quickly along the field lines and will resonate with the magnetic fluctuations in these waves . oblique magnetosonic waves also have a magnetic pressure that varies slightly along the field as a result of the successive convergences and divergences of the velocity and field , which are in phase for fast waves and out of phase for slow waves . these field line changes create magnetic mirrors that enhance the scattering ( schlickeiser & miller 1998 , ragot 2000 ) . the condition for resonance between a cosmic ray and a wave is ( jokipii 1966 , hall & sturrock 1967 , hasselmann & wibberenz 1968 ) @xmath147 this expression says that the doppler shifted frequency of the magnetic wave , as viewed by the particle moving parallel to the field at speed @xmath148 for pitch angle @xmath101 , equals an integer number , @xmath4 , of the gyration frequency . for an alfvn wave , @xmath149 , and for a fast mode at low @xmath144 , @xmath150 . the case @xmath151 is gyroresonance . for wave spectra that decrease with @xmath26 as a power law , only the lowest - order resonances ( @xmath152 ) are important ( cesarsky & kulsrud 1973 ) . magnetosonic waves have an additional resonance for @xmath153 ( lee & vlk 1975 , fisk 1976 , achterberg 1981 ) , when particles stay between two crests by moving along the field at the parallel wave speed . such a particle slows down when it hits the compressive part of the wave ahead of it ( by the mirror effect ) , and it speeds up if the compressive part hits the particle from behind . as a particle oscillates between crests , it can have a parallel velocity that resonates with another wave moving faster along the field , and then jump over to become trapped between two new crests . for a broad wave spectrum , the particle jumps from one wave to the next , gaining energy on average . this process is called transit time acceleration of particles or transit time damping ( ttd ) of waves ( miller 1997 , schlickeiser & miller 1998 ) . ttd increases the parallel component of cosmic ray momentum by mirror scattering in the wave frame . the mirror force depends on the square of the perpendicular component of the velocity ( section [ sect : mirror ] ) , so ttd must operate with the pitch - angle scattering of gyroresonance to maintain momentum isotropy ( miller 1997 ) . a particle that interacts with an alfvn wave can resonate only with the wavelength that is equal to the parallel gyrolength in the doppler shifted frame , @xmath154 . a particle that interacts with magnetosonic waves can respond to the whole wave spectrum because it can choose the wave direction @xmath137 where the parallel wave speed equals its own ( schlickeiser & miller 1998 , schlickeiser & vainio 1999 ) . in practice , such a wave field may not have the appropriate @xmath137 because wave damping is larger at higher @xmath137 . ttd was originally thought to be such an important energy loss for the waves that the fast mode would not be important for cosmic rays ( schlickeiser 1994 , minter & spangler 1997 , tsap 2000 ) . damping is large for high @xmath144 because the thermal particles resonate with the waves and only highly parallel modes exist ( holman et al . 1979 , foote & kulsrud 1979 ) . however , @xmath155 in the solar wind and in much of the cool dense phase of the ism , and observations of the solar wind show that oblique and fast modes dominate over plane - parallel alfvn waves ( tu , marsch & thieme 1989 ; bieber , wanner & matthaeus 1996 ; matthaeus , goldstein & roberts 1990 ) . thus , ttd particle acceleration could be important in parts of the ism where @xmath144 is small . the waves still have to be nearly isotropic to get effective scattering , though , and they should not be damped by other processes , such as ion - neutral or viscous friction . such damping is important in most of the neutral medium where @xmath156 is otherwise favorable for ttd ( kulsrud & pearce 1969 , felice & kulsrud 2001 ) . this leaves few places where wave damping is low enough that ttd might be important . alfvn wave scattering is questionable too because of the extreme anisotropy of the fluctuations that may cascade down from pc scales to the gyroradius . anisotropic turbulence produces tiny perpendicular irregularities that average out over a particle gyration ( chandran 2000a , 2001 ; lerche & schlickeiser 2001 ; yan & lazarian 2002 ; teufel , lerche & schlickeiser 2003 ) . the result is a scattering coefficient that is many orders of magnitude less than in the isotropic case . for this reason , yan & lazarian ( 2003a ) reconsidered ttd scattering by fast modes at moderate @xmath144 , emphasizing the isotropy of the fast waves and considering ism regions where wave damping at particular cosmic ray energies is relatively small . the resolution of this damping problem for wave scattering is far from clear . the energy of fast cosmic ray gyrations is adiabatically invariant during slow changes in the magnetic field amplitude , so an increasing field puts more energy into the gyromotion while removing it from the parallel motion . this change , combined with momentum conservation , causes fast particles with high pitch angles to bounce off converging field lines . the mirror force is @xmath157 for magnetic moment @xmath158 with particle mass @xmath94 . this type of scattering involves very long wavelengths parallel to the field ( ragot 1999 , 2000 ; felice & kulsrud 2001 ; lu et al . the mirror sources could also be molecular clouds and clumps ( chandran 2000b ) . electric fields can also scatter cosmic rays . the flow of plasma at speed @xmath159 transverse to a magnetic field generates an electric field , @xmath160 , that can be important in the parallel direction when the magnetic field has at least 2d structure , and it is convected around by turbulent motions ( fedorov et al . 1992 , le roux et al . 2002 ) . transport transverse to the mean field occurs as individual field lines wander ( jokipii & parker 1969 , bieber & matthaeus 1997 , michalek & ostrowski 1998 , giacalone & jokipii 1999 , kta & jokipii 2000 ) . this is important for radial diffusion in the solar wind , which has a spiral field ( chen & bieber 1993 ) , for diffusion in oblique shock fronts ( duffy et al . 1995 ) , and for diffusion in the vertical direction out of the galaxy . magnetic fields with irregularities smaller than the gyroradius cause particles to skip over field lines , giving anomalous diffusion ( parker 1964 ; chuvilgin & ptuskin 1993 ; casse , lemoine & pelletier 2002 ; otsuka & hada 2003 ; erlykin et al . braided field lines can lead to sub - diffusion ( getmantsev 1963 ) , when the mean squared particle position in the cross field direction increases as the square root of time , instead of directly with time as in normal diffusion . compound diffusion has both cross - field diffusion and diffusion along the field lines ( kirk et al . cross diffusion requires a lot of perpendicular structure to the field lines over a diffusion length for the particle motion parallel to the field ( qin , matthaeus & bieber 2002 ) . for ism turbulence , the cross - field diffusion coefficient is 0.10.2 of the parallel coefficient ( chandran & maron 2004a ) . shocks provide another site for cosmic ray scattering . blandford & ostriker ( 1980 ) considered an ism made of cleared hot cavities from supernovae and showed how the resulting shocks could scatter and accelerate cosmic rays . bykov & toptygin ( 1985 ) extended this model to include secondary shocks that arise in the turbulence caused by the supernovae . klepach , ptuskin & zirakashvili ( 2000 ) included stellar wind shocks . the composite spectrum of cosmic rays from an ensemble of shocks in a supersonically turbulent medium was determined by schneider ( 1993 ) . in such a medium , the summed energy distribution for cosmic rays from all of the shocks is close to a power law , but there can be flat parts or bumps , unlike the energy distribution from a single shock ( see also achterberg 1990 , bykov & toptygin 1993 ) . cosmic rays gain energy as they scatter off randomly moving parts of the ism ( fermi 1949 ) . this is momentum diffusion and called the second - order fermi mechanism because the energy gained per collision depends on the second power of the rms turbulent speed . it is reminiscent of the thermalization of a star cluster with the light particles gaining speed as they approach equilibrium with the heavy particles . scott & chevalier ( 1975 ) first applied this mechanism to supernova remnants , using random motions inside the remnant for scattering sites . supernovae are a likely source for cosmic rays because the cosmic ray energy density , @xmath11 ev @xmath64 , divided by a @xmath115 my lifetime in the galaxy , is about 5% of the total supernova power ( baade & zwicky 1934 , ginzburg & syrovatskii 1964 ) . cosmic ray acceleration by momentum diffusion occurs throughout a compressibly turbulent ism , not just in supernova remnants ( kulsrud & ferrari 1978 , ptuskin 1988 , dolginov & silantev 1990 , bykov & toptygin 1993 , webb et al . 2003 , chandran & maron 2004b ) . in a system of many shocks , particles that are trapped in the weak magnetic turbulence of the ambient medium also get accelerated every time these regions are compressed ( jokipii 1987 ) . multiple shock crossings at the edge of a supernova remnant also accelerate cosmic rays ( see review in blandford & eichler 1987 ) . if the field is parallel to the shock direction , then magnetic turbulence created ahead of the shock by the outward streaming particles ( wentzel 1974 , skilling 1975 ) scatters these particles back into the shock where they encounter more turbulence . the post - shock turbulence was formerly pre - shock turbulence that got compressed and amplified ( schlickeiser , campeanu & lerche 1993 ; vainio & schlickeiser 1998 ) . if the field is oblique to the shock direction , then gyromotions ( jokipii 1987 ) and field line wandering cycle particles through the front ( ragot 2001 ) . each time a particle cycles through the front it gains energy from the converging flow . this is called the first - order fermi mechanism ( fermi 1954 ) because the energy gained per crossing depends on the first power of the shock velocity . about @xmath161 of the incoming thermal particles are injected into cosmic rays ( pryadko & petrosian 1997 ; kang , jones & gieseler 2002 ; bamba et al . 2003 ) , with an efficiency depending on obliqueness ( ellison , baring & jones 1995 ; kobayakawa , honda , samura 2002 ) . generally , the first - order mechanism dominates in strong shocks ( axford et al . 1977 , krymsky 1977 , bell 1978 , ostrowski 1994 ) , although the second mechanism is more important downstream than upstream ( vainio & schlickeiser 1998 ) . shock acceleration explains the cosmic ray energy spectrum up to the `` knee '' at @xmath110@xmath162 ev . the particles that stay in the shock longest end up with the most energy . the spectrum is a distribution function in the number of shock crossings , considering the continuous loss of particles that are trapped in the downstream flow ( e.g. , kato & takahara 2003 ) . observations of the edge sharpness of supernova remnants suggest the amplitude of mhd waves near the shock is @xmath160 times the average ism value ( achterberg , blandford & reynolds 1994 ) . simonetti ( 1992 ) observed a factor of at least 10 in the magnetic wave amplitude from faraday rotation irregularities in a supernova remnant compared with the adjacent line of sight . the x - ray synchrotron emission from supernova remnants is direct evidence for acceleration of relativistic electrons ( koyama et al . 1995 , aschenbach & leahy 1999 ) . a collection of particles with a high enough density can not stream along a magnetic field much faster than the alfvn speed because they generate magnetic irregularities that scatter them ( lerche 1967 ; wentzel 1968a , 1969 ; kulsrud & pearce 1969 ; tademaru 1969 ) . the growth rate of the instability at wavenumber @xmath26 for isotropic wave generation is ( cesarsky 1980 ) @xmath163 the first term in the parenthesis represents the stabilizing effect of wave damping from ion gyromotions . the second term drives the instability with a rate proportional to the ratio of the streaming speed to the alfvn speed . the influence of collective effects is in the ratio of the density of those cosmic rays that resonate with wavenumber @xmath26 to the background density of thermal ions . the size of the magnetic irregularity produced is about @xmath164 times the particle gyroradius ( @xmath110@xmath165 cm for gev energy and @xmath166gauss field ) , so more energetic particles make larger - scale field distortions . these distortions are not at the bottom of a turbulent cascade , so they should not be anisotropic ( chandran 2000a ) . they could cascade to give the ism scintillation . the upper limit to the cosmic ray energy that can generate waves sufficient for their own scattering is around 100 to 1000 gev ( cesarsky 1980 , yan & lazarian 2003a ) . much higher energies are possible in a galactic wind model where the boundary between diffusion and advection depends on energy and the streaming instability landau damps in a highly ionized galactic halo ( ptuskin et al . 1997 ) . hall ( 1980 ) suggested that waves generated by cosmic rays would damp quickly and proposed instead that scintillation - scale structures result from mirror and firehose instabilities in the hot ism phase . both require pressure anisotropies , which hall notes should arise at the level of 0.01 - 0.1 if the hot medium is turbulent . the firehose instability requires @xmath167 for parallel and perpendicular pressures @xmath168 and @xmath169 ( lerche 1966 , wentzel 1968b ) , and the mirror instability requires @xmath170 . these conditions are satisfied for anisotropy @xmath1710.010.1 if the plasma @xmath172 is very large , as might be the case in the hot intercloud medium if supernovae continuously sweep it free of gas and field . one of the earliest indications that the ism is turbulent came from scintillation observations of electron density fluctuations on very small scales . these fluctuations cause diffraction and refraction of radio signals from pulsars and a few extragalactic sources . diffraction broadens pulsar images , spreads out the pulse arrival times , and narrows the frequency interval over which the pulses have a coherent behavior . the relative motion of the diffracting medium also modulates the pulsar intensity on a timescale of minutes . refraction from larger structures causes the images to split or wander in position and to vary in intensity on timescales of days to months . scintillation effects like these offer many diagnostics for electron density structures in the ionized interstellar medium , including _ regions , hot bubble edges and the ionized and hot intercloud media . still , the origin of this turbulence is not clear . because the scales are very small , typically @xmath83 cm or below , it could be part of a cascade from larger structures past the viscous length and down into the collisionless regime of mhd turbulence ( section 4.11 in _ interstellar turbulence i _ ) , or it could be generated locally by cosmic ray streaming or other small - scale instabilities or by low - mass stellar winds and wakes ( sections 3 and [ sect : cr ] in _ interstellar turbulence i _ ) . here we review ism scintillations . more extensive reviews may be found in rickett ( 1977 , 1990 ) , hewish ( 1992 ) , and cordes , rickett & backer ( 1988 ) . the index of refraction for radio waves is ( nicholson 1983 ) @xmath173 with electron density @xmath174 , classical electron radius @xmath175 cm , and wavenumber of the radio signal @xmath176 for wavelength @xmath86 . the phase change of a signal that passes through a clump of size @xmath177 with an excess index of refraction @xmath178 compared with a neighboring average region is @xmath179 . the transmitted signal adds constructively to the neighboring signal at a relative propagation angle @xmath180 if the difference in their path lengths , @xmath181 , multiplied by @xmath26 , equals the phase change . this gives @xmath182 for small @xmath180 . this phase change is random for each clump the signal meets , so after @xmath183 such clumps on a path length @xmath184 , the root mean square angular change of the signal from diffraction is @xmath185 . the most important clump size has a cumulative scattering angle @xmath186 equal to the clump diffraction angle , @xmath187 . evaluation of @xmath188 requires some knowledge of how @xmath189 depends on @xmath177 . for a turbulent medium , this relation comes from the power spectrum of electron density fluctuations , @xmath190 where @xmath191 and @xmath192 for 3d turbulence with a kolmogorov spectrum ( we use @xmath193 here for the wavenumber of electron density fluctuations to distinguish it from the wavenumber @xmath26 of the radio radiation used for the observation ) . the mean squared electron density fluctuation is obtained from the integral over phase space volume , @xmath194 . for logarithmic intervals , this is approximately @xmath195 . thus , we have @xmath196 . using the wave equation , cordes , pidwerbetsky & lovelace ( 1986 ) got essentially the same result with @xmath197 replacing @xmath198 . the full theory more properly accounts for differences in the cumulative effects of the line - of - sight and transverse density variations . a recent modification considers power - law rather than gaussian statistics for the clump size distribution ( boldyrev & gwinn 2003 ) . for a reference set of parameters , e.g. , ghz frequencies , kpc distances , and typical @xmath199 meters@xmath200 , the characteristic scale is @xmath201 cm . this small size implies that only pulsars and a few extragalactic radio sources can produce detectible diffraction effects at radio frequencies . the smallness compared with the collisional mean free path of electrons also implies that the fluctuations have to be collisionless ( see _ interstellar turbulence i _ ) . many observable properties of small radio sources give information , either directly or indirectly , about @xmath177 , which can be used to infer the strength and power index of the electron density fluctuation spectrum and its distribution on the line of sight . for example , the logarithm of the visibility of an interferometer is @xmath202 where @xmath40 is the baseline length and @xmath203 is the scattering measure . spangler & cordes ( 1998 ) found @xmath204 using this baseline dependence for 4 pulsars . very long baseline interferometry ( vlbi ) observations of pulsar image sizes measure angular broadening as a function of frequency ( lee & jokipii 1975a ; cordes , pidwerbetsky & lovelace 1986 ) and determine @xmath205 ( spangler et al . for the reference parameters , @xmath206 mas . angular broadening gives a spread in path lengths for the radio waves , and this corresponds to a spread in arrival times of pulses , @xmath207 ( lee & jokipii 1975b ) . such pulse broadening scales as @xmath208 , so the frequency dependence gives @xmath101 and the absolute value gives @xmath205 for an assumed screen depth @xmath184 . the flux density is correlated only over a small range of frequencies , @xmath209 ( salpeter 1969 , lee & jokipii 1975b ) . pulsar amplitudes vary on a timescale @xmath210 several minutes as a result of pulsar transverse motions @xmath211 km s@xmath71 ( cordes 1986 ) . using the relations above , this can be rewritten @xmath212 . if pulsar distances and proper motions are also known , along with the dispersion measure , then the distribution of scattering material on the line of sight can be modeled ( harrison & lyne 1993 , cordes & rickett 1998 ) . longer time variations ( sieber 1982 ) result from the changing refraction of radio signals in moving structures that have the diffraction angular size @xmath213 or larger ( rickett , coles & bourgois 1984 ; blandford & narayan 1985 ) . this angular size corresponds to a physical size for the refracting elements @xmath214 . for typical @xmath215 mas , @xmath216 cm at @xmath217 kpc . the corresponding refraction scintillation time is @xmath218 days . time variations over months are also observed from larger interstellar structures . the relative rms amplitude of the source is the modulation index , @xmath219 . for diffraction this can be 100% , but for refraction it is typically less than @xmath130% ( stinebring & condon 1990 , labrecque et al . 1994 , stinebring et al . the modulation index depends on the strength of scattering , which involves a length scale equal to the geometric mean of the scale for angular broadening and the dominant scale for electron density fluctuations , @xmath220 . this is the fresnel length , @xmath221 , which is the transverse size of some object at distance @xmath184 that is just small enough to show diffraction effects . when @xmath222 is high , there are many diffracting elements of size @xmath223 inside each refracting element of size @xmath224 , so diffraction is strong . for weak diffraction , the total phase change on the line of sight is small and the dominant clump size is the fresnel length itself ( lovelace et al . 1970 , lee & jokipii 1975c , rickett 1990 ) . in this case , the modulation index scales approximately with @xmath225 ( lovelace et al . the modulation index measures interstellar properties in this weak limit because then @xmath226 , giving a diagnostic for @xmath227 and @xmath101 ( rickett 1977 ) . small @xmath94 usually corresponds to high frequencies , where diffraction is relatively unimportant . time variations can be visualized with a dynamic spectrum , which is a gray scale plot of intensity on a coordinate system of time versus frequency ( figure [ fig : dynamspect ] ) . diffraction alone gives a random dynamic spectrum from the motion of unresolved objects of size @xmath177 , but larger refractive structures , which disperse their radio frequencies over larger angles , @xmath228 ( from the above expression @xmath229 ) , sweep a spectrum of frequencies past the observer ( hewish 1980 ) . the intensity peaks on a dynamic spectrum ( averaged over many pulses ) then appear as streaks with slope @xmath230 ( cordes , pidwerbetsky & lovelace 1986 ) ; this can be used as a diagnostic for @xmath231 . the right - hand part of figure [ fig : dynamspect ] is the 2d power spectrum of the dynamic spectrum . stinebring et al . ( 2001 ) and hill et al . ( 2003 ) suggest that the arcs arise from interference between a central image and a faint scattering halo 2030 times larger . strong refraction can result in multiple images that produce interference fringes on a dynamic spectrum ( cordes & wolszczan 1986 ; rickett , lyne & gupta 1997 ) . multiple images imply that refraction , which determines the image separation , bends light more than diffraction , which determines the size . because large - scale fluctuations dominate refraction , this observation implies either @xmath232 , so that small wavenumbers dominate ( cordes , pidwerbetsky & lovelace 1986 ; romani , narayan & blandford 1986 ) , or there is additional structure on @xmath233 cm scales that is not part of a power law power spectrum ( rickett , lyne & gupta 1997 ) . radio scintillation has been used to determine many properties of electron density fluctuations in the ism . diffraction and refraction come from the same electrons , so the ratio of their strengths is proportional to the ratio of the amplitudes of the electron density fluctuations on two different scales . from this ratio armstrong , rickett & spangler ( 1995 ) derived the power spectrum of the fluctuations spanning @xmath234 orders of magnitude in scale . they determined @xmath235 m@xmath200 and @xmath192 between @xmath81 cm and @xmath236 cm . they also suggested from rotation measures ( rm@xmath237 ) that the same power law extends up to @xmath238 or @xmath239 cm . cordes , weisberg & boriakoff ( 1985 ) found @xmath240 using the relation between @xmath241 and @xmath86 . they mapped the spatial distribution of @xmath227 from @xmath241 observations of 31 pulsars , suggesting the galaxy contains both thin and thick disk components . for the thin component , @xmath242 m@xmath200 assuming @xmath192 and for the thick component , with @xmath243 kpc , @xmath244 m@xmath200 . the rms level of electron density fluctuations at the dominant scale @xmath201 cm , which comes from the integral over the power spectrum ( see above ) , was found to be @xmath245 @xmath64 for the high latitude medium , for which @xmath246 @xmath64 from dispersion measures , and @xmath247 @xmath64 for the low latitudes . bhat , gupta & rao ( 1998 ) observed 20 pulsars for three years to average over refraction variations and determined @xmath227 in the local ism from @xmath241 . they observed an excess of scattering material at the edge of the local bubble , which also produced multiple images of one of these pulsars ( gupta , bhat & rao 1999 ) . the edge of the local bubble may also have been seen by rickett , kedziora - chudczer & jauncey ( 2002 ) as a source of scattering in a quasar . bhat & gupta ( 2002 ) found a similar enhancement at the edge of loop i , where @xmath248 pc m@xmath200 and the density enhancement is a factor of @xmath1100 over the surrounding gas . they also found excess scattering for more distant pulsars from the sagittarius spiral arm . lazio & cordes ( 1998a , b ) used the angular sizes of radio sources and other information on lines of sight to the galactic center and outer galaxy to suggest that the scattering material is associated with the surfaces of molecular clouds . ionized cloud edges were also suggested by rickett , lyne & gupta ( 1997 ) to explain multiple images . spangler & cordes ( 1998 ) observed @xmath249 from small sources behind six regions in the cygnus ob1 association and found an excess in sm that was correlated with the emission measure , indicating again that scattering is associated with _ h ii _ regions . this result is consistent with the high cooling rate required by the dissipation of this turbulence , which implies high temperatures ( zweibel , ferriere & shull 1988 ; spangler 1991 ; minter & spangler 1997 ) . the relation @xmath250 combined with pulsar distances and proper motions led gupta ( 1995 ) to determine a 1 kpc scale height for the scattering layer from long - term observations of 59 pulsars . cordes & rickett ( 1998 ) also used this method to find that scattering is rather uniformly distributed toward two pulsars , but for three lines of sight it was concentrated toward the pulsar , including vela where the supernova remnant is known to contribute to the scattering ( desai et al . 1992 ) ; six other pulsars in that study had significant scattering from either a foreground spiral arm or _ h ii _ regions . bhat , gupta & rao ( 1999 ) compared @xmath251 obtained from @xmath241 with the angular size of the refraction pattern , @xmath252 from dynamic spectra . they showed that @xmath253 for all 25 pulsars that had this data , implying that diffraction dominates refraction and therefore @xmath254 . they also compared @xmath227 from @xmath255 on the refractive scale with @xmath227 from @xmath241 on the diffractive scale to determine the slope of the power spectrum directly ( as did armstrong et al . 1995 ) . they found the kolmogorov value @xmath192 to within the accuracy in most cases , but six pulsars , mostly nearby , gave slightly higher @xmath256 . the longest - term variations had significantly more power than an extrapolation of the kolmogorov spectrum , however , and the modulation indices were large , leading them to suggest an additional component of scattering electrons at scales of @xmath257 to @xmath83 cm . the same data led bhat , rao & gupta ( 1999 ) to derive @xmath258 m@xmath200 and @xmath259 , indicating strong scattering . lambert & rickett ( 2000 ) looked at the correlation between the modulation index for long - term variations from refraction and the relative decorrelation bandwidth , @xmath260 , which comes from diffraction . the correlation at 610 mhz for 28 sources fit kolmogorov scaling ( @xmath192 ) better than a shock - dominated model ( @xmath261 ) model but at 100 mhz the kolmogorov fit was not as good . they suggested that the minimum turbulent length was large , @xmath262 to @xmath233 cm instead of @xmath263 cm ( armstrong , rickett & spangler 1995 ) . an excess of electron density structure at large scales could also explain the discrepancy . stinebring et al . ( 2000 ) monitored the modulation index of 21 pulsars for five years and found low values ( @xmath264% ) that bracket @xmath265 with no perceptible inner ( smallest ) scale for most of the pulsars . for the crab , vela , and four others , enhanced modulation indices were consistent with an inner scale of @xmath262 cm . rickett , lyne & gupta ( 1997 ) suggested this excess scattering could be from au - sized ionized clumps at cloud edges . an inner scale of @xmath266 cm was found by spangler & gwinn ( 1990 ) after noting that interferometer baselines shorter than this had scintillation with @xmath2674 whereas longer baselines had @xmath101 near the kolmogorov value , @xmath268 . they suggested the inner scale is caused by a lack of turbulence smaller than the ion gyroradius , @xmath269 for thermal speed @xmath270 , or by the ion inertial length , @xmath271 , whichever is larger . this value of the inner scale is consistent with an origin of the scattering in the warm ionized medium or in _ h ii _ regions , but not in the hot coronal medium , which has much larger minimum lengths . shishov et al . ( 2003 ) studied pulsar psr b0329 + 54 over a wide range of frequencies and found a power spectrum for electron density fluctuations with a slope of @xmath272 for lengths between @xmath2 cm and @xmath273 cm . this spectrum is expected for turbulence in the iroshnikov - kraichnan model ( see _ interstellar turbulence i _ ) . shishov et al . noted how other lines of sight though the galaxy gave different spectra and suggested that the nature of the turbulence varies from place to place . they also found refraction effects on a scale of @xmath274 cm corresponding to electron density fluctuations of strength @xmath275 @xmath64 and suggested these were neutral clouds with an overall filling factor of @xmath10.1 . scattering image anisotropy suggests anisotropic turbulence ( see _ interstellar turbulence i _ ) . lo et al . ( 1993 ) observed 2:1 anisotropy in images of sgr a@xmath276 , wilkinson , narayan & spencer ( 1994 ) found scale - dependent anisotropic images of cygnus x-3 , frail et al . ( 1994 ) observed 3:1 anisotropy in scattering of light from galactic center oh / ir stars , whereas trotter , moran & rodriguez ( 1998 ) observed axial ratios of 1.21.5 for quasar light that scatters through a local _ h ii _ region . a ratio of 4:1 was found for another quasar by rickett , kedziora - chudczer & jauncey ( 2002 ) . spangler & cordes ( 1998 ) observed anisotropic scattering with axial ratio of 1.8 surrounding the cygnus ob association and desai & fey ( 2001 ) observed axial ratios of @xmath11.3 toward the same region . the actual anisotropy of local fluctuations can not be determined from these observations because many different orientations blend on the line of sight ( chandran & backer 2002 ) extreme scattering events are observed in some extragalactic radio sources and a few pulsars . their modulation is strong , @xmath277% , they can last for several months , and they have light profiles that are flat - bottom with spikes at the end , or smooth bottom with no spikes ( fiedler et al . 1994 ) . they may result from supernova shocks viewed edge - on ( romani , blandford & cordes 1987 ) or ionized cloud edges in the galaxy halo ( walker & wardle 1998 ) . some appear correlated with the edges of local radio loops ( fiedler et al . 1994 , lazio et al . 2000 ) , in support of the shock model . the actual scattering process could be a combination of refractive defocusing , during which an intervening electron cloud produces a lens that diverges the light and makes the source dimmer ( romani et al . 1987 , clegg et al . 1998 ) , or stochastic broadening by an excess of turbulence in a small cloud ( fiedler et al . lazio et al . ( 2000 ) found an excess @xmath278 kpc m@xmath200 associated with an event , corresponding to @xmath279 m@xmath200 for ratio @xmath280 of the line - of - sight extent ( @xmath184 ) to the transverse extent . fiedler et al . ( 1994 ) suggested @xmath281 @xmath64 . this high level of scattering along with an observed increase in angular size during the brightness minimum ( lazio et al . 2000 ) suggests the scattering cloud is not part of a power spectrum of turbulence but is an additional au - size feature ( lazio et al . 2000 ) . the amplitude , slope , and anisotropy of the power spectrum of interstellar electron density fluctuations have been observed by scintillation experiments . estimates for the inner scale of these fluctuations range from @xmath54 cm to @xmath262 cm or more . the shorter of these lengths is about the ion gyroradius in the warm ionized medium ( spangler & gwinn 1990 ) . this is much smaller than the mean free path for electron collisions so the fluctuations are collisionless ( see _ interstellar turbulence i _ ) . a lower limit to the largest scale of @xmath110@xmath282 cm was inferred from rotation measures assuming a continuous power law for fluctuations , but this assumption is uncertain . the slope of the power spectrum is usually close to the kolmogorov value , @xmath268 , and distinct from the slope for a field of discontinuities , which is @xmath283 . deviations from the kolmogorov slope appear in some studies , and may be from a large inner scale , @xmath284 cm , an excess of scattering sites having size @xmath1110 au , or a transition in the scaling properties of the turbulence . scintillation arises from both the low - density , diffuse , ionized ism and the higher - density _ h ii _ regions , ionized cloud edges , and hot shells , where the amplitude of the power spectrum increases . on average , the relative fluctuations are extremely small , @xmath285 on scales that dominate the diffraction at ghz radio frequencies , @xmath286 cm ( cordes et al . 1985 ) . for a kolmogorov spectrum , they would be larger on larger scales in proportion to @xmath287 ; if the spectrum is continuous up to pc scales , the absolute fluctuation amplitude there would be @xmath11 ( lee & jokipii 1976 ) . anisotropy of scattered images is at the level of 50% , which is consistent with extremely large intrinsic anisotropies ( @xmath288 ) from mhd turbulence ( sections 4.11 and 4.13 in _ interstellar turbulence i _ ) if many orientations on the line of sight are blended together . the turbulence that is observed directly on resolvable scales in the ism also has important effects on very small scales , down to the level of atomic diffusion , mixing , and viscosity , and continuing far below to the thermal ion gyroradius . the resolvable turbulence in the neutral medium was reviewed in the first part of this series ( _ interstellar turbulence i _ ) , along with general theory and simulations . the smaller scale implications were reviewed here . turbulence helps to disperse the elements made in supernovae and other stellar sources by stretching and folding the contaminated gas until the gradient length approaches the collision mean free path . then atomic diffusion does the final step of interatomic mixing . this mixing and homogenization may occur partly at the source , in the shock fronts and contact discontinuities around the expanding flow following dynamical instabilities , and it may occur partly in the ambient ism after the expansion subsides . observations suggest that the dispersion in elemental abundances among stars inside clusters and field stars of the same age , and the abundance differences between hii regions and diffuse clouds , are only a few percent up to perhaps 30% . observations of clusters also suggest that most of the mixing occurs within several times @xmath2 years . this homogenization has to occur between the injection scale of supernovae and superbubble explosions and the star formation scale containing on the order of a solar mass . the corresponding range of spatial scales is @xmath289 for gas at the ambient density . homogenization can not be significant on scales much larger than a superbubble , because then it would diminish the galactic radial abundance gradient over a hubble time . these gradients enter the problem in another way too because turbulence mixes the gas over galactic radius to increase the elemental dispersion locally at the same time as it mixes this gas locally and leads to homogenization . while the distribution of elemental abundances appears to be fairly narrow , with a dispersion on the order of 10% , it may also have a fat tail caused by occasional odd stars and diffuse clouds with very different abundances . this tail is reminiscent of other fat tails in the distribution functions for turbulent media , and these other tails seem to originate with intermittency . in the case of elemental dispersion , this means that the homogenization process is spotty so some regions survive for long periods of time with very little mixing . several methods for studying turbulent mixing have been employed . because of the wide range of scales involved , these studies often employ simplifying assumptions that appear to capture the essential physics . they include artificial stochastic velocity fields and closure methods using moment equations . direct numerical simulations of ism elemental dispersion have only just begun . the theory suggests that a turbulent medium mixes passive scalars like elemental abundances faster than a gaussian velocity field because of the long - range correlations that are associated with turbulence . turbulence affects interstellar chemistry by mixing regions with different properties , heating the gas intermittently on the viscous scale , and enhancing ion - neutral collisions in regions with strong magnetic field gradients . turbulent mixing is more far - reaching than diffusive mixing because of long - range correlated motions in a turbulent flow . such mixing spreads out each chemical species over a large radial range inside a cloud . turbulent heating promotes temperature - sensitive reactions inside otherwise cold clouds . applications to the formation of ch@xmath59 and oh in diffuse clouds look promising . chemical reaction networks that include these turbulent processes are only beginning to understand some of the implications , and direct simulations of chemistry in a turbulent medium are limited to only a few studies so far . the scattering and acceleration of cosmic rays depends strongly on the presence of turbulence in the ism . the scale for this turbulence is the gyroradius , which is less than the collisional mean free path for most cosmic rays , which have energies less than 1 gev . thus , scattering relies entirely on magnetic irregularities in collisionless plasma turbulence . cosmic ray acceleration is by two processes , both of which involve turbulence : the first - order fermi mechanism accelerates cosmic rays by cycling them through shock fronts where they pick up a relative velocity kick comparable to the shock speed at each passage . this cycling occurs because magnetic irregularities on each side of the front scatter the out - streaming cosmic rays back into the front . the second - order fermi mechanism accelerates cosmic rays through turbulent diffusion . each collision with a randomly moving magnetic irregularity turns the cosmic ray around with a reflection speed in the moving frame that is comparable to the incident speed . over time , the result is a transfer of energy from the turbulence to the cosmic rays . cosmic ray scattering occurs in several ways . fast moving particles can interact weakly but resonantly with numerous magnetic irregularities , and successive interactions of this type can randomly change the pitch angles of their helical motions along the field . this eventually leads to a complete reversal in the direction of motion . cosmic rays can also interact strongly with large magnetic irregularities , as would occur in shock fronts and at the edges of dense cloud complexes . these interactions change the particle directions significantly each time . the nature of cosmic ray diffusion in both momentum and space is not understood well because the structure and strength of the important magnetic irregularities are not observed directly . if mhd turbulence is highly anisotropic on the scale of cosmic ray scattering , with transverse irregularities much stronger than parallel as suggested by theory , then resonant scattering processes during motions along the mean field can be very weak . the structure of magnetic waves below the mean free path is also unclear , as the usual fast and slow magnetosonic modes do not exist . cosmic rays can also generate turbulence by streaming instabilities following particle - wave resonances or by fire - hose and mirror instabilities that operate even without resonances . the expected anisotropy of alfvn wave turbulence diminishes the first of these mechanisms significantly , however . the others are problematic because they require strong cosmic ray pressure anisotropies . as a result , the impact of cosmic rays on turbulence is currently not understood . radio wave scintillation is indirect evidence for interstellar plasma turbulence . these radio observations span a very wide range of spatial scales through a combination of diffraction and refraction effects . the scales are mostly below the collision mean free path , and they are far below the limits of angular resolution . the first of these limits makes it difficult to understand the origin of the density irregularities . unlike the turbulence that scatters cosmic rays , which requires only magnetic irregularities and no density structure , the turbulence that causes scintillation requires small - scale density irregularities in the ionized medium . the associated magnetic irregularities are not observed , and the connection to cosmic ray scattering is unclear , even though the length scales are about the same . the origin of density structures below the collision mean free path is unknown . atomic diffusion should smooth them out on a sound crossing time unless magnetic field irregularities hold them in place . in that case they could be the result of slight temperature variations , with the cooler regions having higher electron densities , all divided up and mixed together by transverse magnetic motions . the nature of these motions below the mean free path is unclear , because , as mentioned above , the usual fast and slow compressional modes do not exist , nor do the usual thermal and pressure - regulated processes . the second of the two limits on spatial scale imply that the geometrical properties of scintillation turbulence are difficult to observe . the scintillation is clearly anisotropic , but whether it is in sheets or filaments , for example , is unknown . the importance of scintillation observations for studies of ism turbulence is that they give the power spectrum of electron density fluctuations fairly accurately . this is usually close to the kolmogorov spectrum of incompressible turbulence . rarely are the spectra so steep that the medium can be interpreted as a superposition of sharp edges , like shock fronts . one recent observation with fairly high precision obtained the relatively shallow iroshnikov - kraichnan power spectrum , leading to the conjecture that the slope varies from region to region . as discussed in _ interstellar turbulence i _ , this variation may arise from a variation in the relative strength of the magnetic field compared to the turbulent motions , with the iroshnikov - kraichnan spectrum present in regions of relatively strong fields . observations suggest that scintillation arises in a distributed fashion from the ambient ionized medium and also from discrete high - density places like the edges of local bubbles , ionized molecular clouds , and hii regions . these discrete regions are likely to be highly turbulent and they also have a juxtaposition of hot and cool gas , which is necessary for isentropic mixing and electron density structure . there are evidently many uncertainties in the nature of ism turbulence on small scales , even though the evidence for this turbulence is pervasive . part of the problem is that none of the features of this turbulence have been observed directly : not the densities , magnetic fields , temperatures , or motions . still , the density irregularities are revealed indirectly through scintillation , the magnetic field irregularities through cosmic ray scattering , the temperature fluctuations through chemistry , and the motions through elemental and chemical mixing . an additional problem is that many small scale effects of ism turbulence rely on details of the theory that are independent of the usual scaling relations , such as viscous heating and elemental diffusion , which arise at the bottom of the cascade in the neutral medium . turbulence in the ionized medium is also below the collisional mean free path , where pressure and thermal effects are relatively unimportant . moreover , the small scale ism processes are often strongly dependent on the large scale processes , such as turbulent shock formation , energy and metal injection , and galactic - scale gradients . this means that direct simulations of small scale turbulence are impossible without simplifying assumptions about the large - scale medium assumptions that require more knowledge about ism turbulence on the large scale than is presently available . while the observations and theory of ism turbulence have come a long way from the first efforts in the 1950s , the details of this new information have led to a growing awareness that the complete problem is far too large to solve any time soon . we rely on future generations of astronomers and physicists to continue this work , and hope that they find this field as intriguing and challenging as we do today . we are grateful to a. brandenburg , d. lambert , f. matteucci , s. oey , and g. tenorio - tagle for helpful comments on section 2 ; e. falgarone , d. hollenbach and j. le bourlot for helpful comments on section 3 ; b. chandran , r. jokipii , a. lazarian , v.s . ptuskin , and r. schlickeiser for helpful comments on section 4 ; b. rickett , s. spangler , and d. stinebring for helpful comments on section 5 and to a. hill for figure [ fig : dynamspect ] . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: wireless networks are ubiquitous and are on their way to become even more prevalent , e.g. , with the advent of internet - of - things . wireless communication is , however , particularly challenging to model algorithmically . in two crucial interrelated aspects , wireless networks on the ground differ from models typically assumed in algorithmic studies . one is the _ communication modeling _ : when is a transmission successfully decoded , as a function of the environment and the _ interference _ from other transmissions . the other is _ variability with time _ : wireless networks are particularly susceptible to changes . both of these are hard to capture accurately with well - defined , clear - cut rules . we aim in this paper to address core information dissemination problems _ local broadcast _ and ( global ) _ broadcast _ in dynamic distributed networks , under very weak assumptions on the communication . to this end , we propose a communication model with significant flexibility that allows for adversarial control , generalizing essentially all known analytic wireless models . it allows for more general interference relationships than treated before . the network can experience adversarial dynamic behavior , both edge changes ( change of signal strengths ) and node insertions / deletions . wireless communication has traditionally been modeled theoretically by graphs , either geometric or general . interference is then also transmitted on graph edges ( precluding a node from receiving a message from a neighbor if another neighbor is transmitting ) , but is sometimes represented by a supergraph . _ fading channel _ or _ physical _ models common in communication engineering , on the other hand , consider interference as cumulative , decreasing with distance but adding up . they have been popular in recent algorithmic studies , adding more realism to the formulation . the standard assumption of _ geometric signal decay _ ( that signal decreases inverse polynomially with distance ) in the sinr model , the prototypical fading channel , is though equally at odds with experimental evidence . ultimately , it may prove futile to hope for a clean deterministic model , or even a purely stochastic one , without a significant dose of unpredictability and non - determinism . wireless communication is commonly closely linked to mobility , as the transceivers are more often than not on the move . dynamic changes to reception conditions have many causes other than node mobility , since almost any changes in the environment affect transmissions due to reflections of signals over multiple paths , antenna characteristics , scattering , and diffraction . these changes are by nature hard to predict , even when assuming a `` mobility model '' . the most robust approach would then be to assume a non - trivial adversarial component . while the study of algorithms in dynamic networks has a long history , little has been done in cases where interference plays a role . * setting and model*. nodes are distributed and autonomous . there is no built - in structure and the nodes have no information besides bounds on model parameters and an upper bound on the number of nodes . communication is locally synchronous , but there is no global clock . we will assume a very general model for when nodes successfully communicate . nodes are located in space , with separation between points given by the relative decrease in strength of signal ( or interference ) sent between the points . this induces a metric space , when these decays are raised to the appropriate power actually , it is a _ quasi - metric _ , since symmetry need not hold . for distributed computation to be possible , the quasi - metric must have _ bounded independence _ ( to be defined precisely ) . _ edge changes _ can occur , with some restrictions , which are changes in the signal strength between the pair of points . the rule for when communication is successful is only partially pre - specified : transmission succeeds on a `` clear channel '' . that is , if a sender is within a _ communication radius _ from the receiver , if no other node transmits within a ( larger ) radius , and if the combined interference from all other transmitting nodes is ( quite ) small , then the transmission succeeds . otherwise , success is up to the adversary , or it can be further specified by the particular model assumptions desired . this captures essentially all known algorithmic wireless models ( including quasi - unit disc graphs , unit - ball graphs , bounded - independence graphs , @xmath0-hop extensions , and sinr ) . the only exception is the radio network model with general graphs , which can not be extended to involve comprehensive interference without a major hit in time complexity . the generality of our model is a key feature . given the vagaries of actual wireless environments , it is preferable for robustness reasons to make minimal assumptions about the communication model . a conservative approach is then to seek algorithms that work in most established models rather than depending on model - specific factors that simplify the life of the algorithm designer . [ [ dynamic - network - model ] ] dynamic network model + + + + + + + + + + + + + + + + + + + + + the network is dynamic , both in terms of nodes and links . nodes can come and go , potentially adversarially . this includes the situation where nodes are wakened by other nodes . we only expect nodes that have been awake for long enough ( logarithmic number of rounds ) to participate in the computation , i.e. , the _ stable _ nodes . in the _ local broadcast _ problem , all stable nodes should inform all of their ( stable ) neighbors ( i.e. , within the communication radius ) . in the _ broadcast _ problem , the nodes in the network are to be informed of a message that initially resides at a source node . the network need not be connected at any instant in time , but we need to make assumptions about what transient paths are usable by an algorithm operating under interference . this leads to a definition of a _ dynamic diameter _ of the network , that becomes the competitive objective for the time complexity . * our approach and results*. the key algorithmic technique is a natural randomized _ contention balancing _ procedure , where a node continuously adjusts its transmission probability based on the interference that it senses . it allows nodes to stabilize quickly from any initial conditions , or after waking up . this routine is a variation on an old story , a simple backoff procedure to manage local contention : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ if a node @xmath1 senses contention in a given round beyond a fixed threshold , then @xmath1 halves its transmission probability in the next round and otherwise doubles it . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ our main technical contribution is to show logarithmic - round convergence of this method to a steady state of nearly balanced contention , from an arbitrary starting configuration and in the presence of network changes . it proves also to be surprisingly tolerant of different communication models . the higher level algorithms are then built on top of this primitive . a crucial component is the use of _ carrier sense _ to detect the cumulative amount of signals in the air . since it is supplied by the cheapest available hardware today as rss ( received signal strength ) readings , we posit that carrier - sense capability should be the default assumption in wireless algorithmics ( while exploring the necessity of different assumptions is interesting theoretically ) . as carrier - sense indicators can provide fine - grained information , we are interested in restraining its use and identifying which aspects are necessary to achieve the results obtained . to this end , we identify several _ primitives _ that carrier - sense can supply , restrict the algorithm to use only a subset of the primitives , and examine which of these are truly necessary . the local broadcast algorithm simply runs the contention balancing procedure , with nodes bowing out when they are sure to have completed their transmission . the broadcast algorithms are based on sparsifying the instance , so that only nodes of constant density actually participate in the global broadcast action . the former is achieved in @xmath2 time , where @xmath3 is the maximum number of neighbors that a node can have , while the latter takes @xmath4 rounds , where @xmath5 is a dynamic diameter . these dissemination algorithms are efficient enough to improve on some of the results known for static versions of the problems . the local broadcast algorithm is strongly optimal , or within constant factors on every instance . in the standard setting ( static , spontaneous case ) , the algorithm is _ uniform _ , in that it need not know the network size . the broadcast algorithm is also optimal and uniform in the same setting , while in the non - spontaneous setting it is faster by a logarithmic factor than the previous algorithm of @xcite that however does not require carrier sense . * closely related work*. there are two largely disjoint bodies of work of wireless algorithmic results , with work on fading models like sinr slowly catching up with the better studied graph - based models . one approach for capturing more realism in sinr model is to move beyond euclidean metrics @xcite , even to general ones @xcite . one can view relative signal decrease as implicitly defining a quasi - distance metric @xcite . link scheduling problems can be formulated on edge - weighted interference graphs @xcite that properly generalize both graph - based and sinr models , linked by a graph - theoretic parameter . distributed dissemination problems , however , necessarily require metric restrictions , such as doubling or _ fading _ metrics @xcite , and limits on communication abilities in order to capture both types of models . the local broadcast and global broadcast problems have been extensively studied in both graph - based radio network models @xcite and the sinr model @xcite . for local broadcast , the best results known in the radio network model are both @xmath6 with and without knowing ( an upper bound on ) @xmath3 @xcite . in the sinr model , with knowledge of @xmath3 , the local broadcast can be accomplished in the same time bound as in the radio network model @xcite . if @xmath3 is not known , the best result is @xmath7 @xcite , which is improved to @xmath2 with free acknowledgments @xcite . this can be further improved to @xmath8 in the spontaneous case , when @xmath3 is known @xcite . the time complexity of non - spontaneous broadcasting in the radio network model is @xmath9 @xcite without collision detection . with collision detection , this lower bound was recently broken in @xcite , where a solution of @xmath10 was given . broadcasting has also been treated in the sinr model under a variety of assumptions . some are stronger than ours ( location information @xcite , power control @xcite ) , while others relax the assumption about the connectivity property , incurring necessarily much higher complexity @xcite . results in our setting , but without carrier sensing , include time complexity of @xmath11 @xcite ( see also @xcite ) , where @xmath12 denotes the maximum ratio between distances of stations connected in the communication graph ; and @xmath13 @xcite . in the spontaneous setting , where the nodes can build an overlay structure along which the message is then propagated , scheideler et al . @xcite used carrier sense to give a dominator algorithms , which can be applied to solve broadcast in @xmath14 rounds . yu et al . @xcite solved the problem in the same time bound using power control , while the algorithm in @xcite that requires neither power control nor carrier sense runs in time @xmath15 . these problems have also been treated in dynamic networks . in the _ unstructured _ model @xcite , where nodes may wake up asynchronously ( modeling the node insertion ) , the local broadcast problem is well studied , even in the sinr setting @xcite , but this model does not consider node deletion . in the _ dual graph _ model @xcite ( originally due to @xcite ) , both the local broadcast @xcite and global broadcast @xcite problems are studied . but this model involves only edge behavior and not node changes ( churn ) . hence , the impact of dynamicity on wireless information dissemination is still largely unexplored . considering wireless interference , there are two classes of wireless network models : graph - based and physical models . basically , the graph - based models define a local and binary type of interference , while the physical models consider fading effect of signal on wireless channels . there are many widely used graph - based models , such as the unit disc graph ( udg ) model @xcite ( unit ball graph ( ubg ) model @xcite ) , the quasi unit disc graph ( qudg ) model @xcite , the protocol model @xcite , the bounded - independence graph ( big ) model @xcite and the @xmath0-hop variants of the above models @xcite . physical models , also known as sinr models @xcite , capture the missing practical issues in wireless interference , such as fading and cumulativeness of wireless interference . most of current works in the sinr model focus on networks embedded in euclidean space , while many of the results hold also for doubling or `` bounded growth '' metrics @xcite . those doubling metrics constrain growth at every ( or arbitrarily small ) granularity , while ours only bounds regions proportional to the transmission range , in particular capturing bounded independence graphs ( big ) . all above models are included by our proposed communication model as shown in sec . [ sec : primitiveproofs ] . for more on wireless models , please refer to @xcite . more detailed related work is introduced in sec . [ drw ] . * our contributions*. we have obtained generalized and improved algorithms for two of the most fundamental dissemination problems , in some cases improving the best results known in static settings . beyond these specific results , we identify the following technical contributions : _ unified model of wireless networks_. the model proposed appears to be the first that allows for the development of pan - model distributed dissemination algorithms . this hopefully prompts further studies crossing the artificial boundary between graph- and fading - based models . _ dynamic networks under interference_. this appears to be the first work to address dynamic networks in the presence of comprehensive interference . _ uniform algorithms_. our algorithms in the static spontaneous setting appear to be the first in fading models that work independent of instance parameters ( number of nodes , max . degree ) . _ primitives for carrier - sense . _ we introduce several primitives or _ capabilities _ that can be implemented using environmental sensing , and propose to study the power of such primitives . _ stabilization mechanisms . _ we identify contention adaptation as a fundamental ability in wireless networks , that appears to be of crucial value to implement other distributed tasks . * roadmap*. the formal model and basic definitions , including the definitions of communication model and carrier sensing primitives , are given in sec . [ sec : model ] . [ sec : balancingcontention ] contains the core technical part of the paper that includes the basic contention balance routine and its analysis . the main results concerning local and global broadcast problems are presented in sections [ sec : localbroadcast ] and [ sec : broadcast ] , respectively . due to space constraints , most proofs are relegated to appendices . we consider a dynamic network of point - size wireless devices ( nodes ) . nodes can transmit messages in time slots / rounds that are sufficiently long to allow a transmission of a single message . no global clock or synchronization of rounds is required , but the clocks of different nodes run at a similar rate , i.e. , the length of a round differs between nodes at most by a factor of 2 . nodes may arrive and leave the network at any time . unless specified otherwise , the nodes are assumed to work _ non - spontaneously _ : they can initially be in sleep state and join the execution of an algorithm only after receiving a message . we say a node is _ alive _ at some point in time if it is present in the network . we assume the total number of nodes in the network is polynomially bounded by a number @xmath16 in each round . we use @xmath17 to denote the set of alive nodes at any fixed point in time and also use @xmath16 to denote the current number of nodes , i.e. @xmath18 . we assume all nodes use the same transmission power @xmath19 for communication in all rounds . _ metrics_. the _ signal strength _ or _ interference _ , depending on context of transmitting node @xmath20 on a node @xmath1 is @xmath21 , where @xmath22 is the _ path loss _ from @xmath20 to @xmath1 . the _ metricity _ of a space @xmath23 is the smallest number @xmath24 such that for every triplet @xmath25 , @xmath26 @xcite . we define @xmath27 if @xmath20 and @xmath1 are different nodes and @xmath28 when @xmath29 . note that @xmath30 is a _ quasi - metric _ , as all metric axioms except symmetry hold . in the rest of the paper , we assume that in each round the metricity of the network is bounded by a fixed constant @xmath24 and will work with values @xmath31 instead of @xmath32 . we assume the quasi - metric @xmath30 has _ bounded independence _ , defined below , roughly stating that there can not be many nodes each causing high interference to a _ fixed _ node , while having low mutual interferences . first , some notations . the _ ball _ with radius @xmath33 centered at @xmath20 is defined as @xmath34 . the _ in - ball _ with radius @xmath33 centered at @xmath20 is defined as @xmath35 ; clearly , @xmath36 . a set @xmath37 is a @xmath33-_packing _ for set @xmath38 if balls of radius @xmath33 centered at nodes in @xmath39 are contained in @xmath38 and are disjoint . @xmath39 is a @xmath40-_cover _ for @xmath38 if the union of balls of radius @xmath40 centered at nodes in @xmath39 contains @xmath38 . note that any maximal @xmath33-packing is a @xmath41-cover , and thus one can bound sizes of covers by packings . we say that @xmath30 has @xmath42-_bounded independence _ , for given @xmath43 and @xmath44 , if for every @xmath45 and every in - ball @xmath46 of radius @xmath47 , the size of a maximum cardinality @xmath48-packing of @xmath46 is at most @xmath49 , where @xmath50 is a constant , possibly depending on @xmath51 . for instance , the euclidean plane is @xmath52-bounded independent , for every @xmath33 . _ neighborhoods , communication graph and dissemination problems_. let @xmath53 denote the maximum transmission distance possible when no other node transmits . as the latter event is arguably very rare , we define the _ communication radius _ @xmath54 as a slightly smaller distance , where @xmath55 is a _ precision _ parameter . we will drop the parameter @xmath55 whenever it is fixed and clear from the context . fix a round @xmath56 . the _ neighborhood _ of a node is @xmath57 , describing who @xmath20 can communicate with directly . the basic operation of interest is when @xmath20 broadcasts a message to its neighbors @xmath58 . the _ communication graph _ is a directed graph @xmath59 , where @xmath60 if and only if @xmath61 . thus , the sequence @xmath62 defines a _ dynamic graph_. the _ vicinity _ of @xmath20 refers to a larger region , @xmath63 , for a parameter @xmath64 . the data dissemination problems that we consider are defined below . we say that a node @xmath20 _ mass - delivers _ in round @xmath56 if it transmits and all its neighbors ( @xmath58 ) receive the message . in the _ local broadcast _ problem , given a node @xmath20 , it is required to minimize the time from the beginning of the algorithm until node @xmath20 mass - delivers at least once , assuming it stays alive during that time . in the _ ( global ) broadcast _ problem , given a distinguished _ source _ node that initially holds a message , the goal is to minimize the time needed to deliver the message to every node in the network through multihop transmissions . _ one hop communication_. when is a transmission successfully received ? suppose a node @xmath20 transmits in round @xmath56 , and let @xmath39 be the set of concurrently transmitting nodes . let @xmath65 and @xmath66 be parameters that depend on the precision @xmath55 . ( , success on a clear channel ) if no other node in @xmath67 transmits _ and _ the total interference at node @xmath20 is at most @xmath68 ( i.e. @xmath69 and @xmath70 ) , then the transmission of node @xmath20 is successfully received by _ all _ its neighbors ( @xmath58 ) . otherwise , the reception is under adversarial control . _ randomized algorithms_. we mainly consider randomized algorithms of the following form : in each round @xmath56 , node @xmath1 makes a transmission with probability @xmath71 , independent of other nodes transmissions in that round . an important notion for the analysis of such algorithms is _ local contention _ , the sum of the transmission probabilities in a close region . the contention in the _ close neighborhood _ of a node @xmath1 in round @xmath56 is @xmath72 , where the radius @xmath73 allows all pair of nodes in @xmath74 to potentially communicate . also , let @xmath75 denote the contention in the larger _ vicinity _ of @xmath1 in round @xmath56 ( @xmath76 will be fixed later ) . we will also use the notation @xmath77 to denote the interference at @xmath1 from nodes outside its vicinity ( in @xmath78 ) in round @xmath56 . the expected value of @xmath77 is @xmath79 . _ sensing primitives_. we assume the nodes have abilities to sense activity on the channel . namely , we assume the nodes are able to detect high and low contention in their vicinity , detect ( under some conditions ) whether their transmission in a given round succeeded and detect a single very near transmission . in the following , we formalize these notions in three primitives : , and . we show in sec . [ sec : primitiveproofs ] how _ all _ these primitives can be implemented with _ basic _ physical carrier sensing and possibly also with other means . ( ) . contention can be probabilistically deduced from measured level of radio activity . we want to relax this ability and will use the following variant , where the outcome of is one of the two values : busy or idle channel . formally , for each node @xmath1 and round @xmath56 : if contention among close neighbors is high ( @xmath80 ) then they _ all _ detect busy channel in round @xmath56 with probability at least @xmath81 , for given @xmath82 , where @xmath83 is a constant , if the contention in the vicinity of @xmath1 is low ( @xmath84 ) _ and _ the interference on @xmath1 from outside its vicinity is above a threshold ( @xmath85 ) then @xmath1 detects idle channel in round @xmath56 with probability at least @xmath86 , for given @xmath87 , where @xmath88 and @xmath89 are constants . if a node @xmath20 has the = ( @xmath55 ) primitive ( depending on the precision parameter @xmath55 ) then : if @xmath20 transmits in round @xmath56 , the interference at @xmath20 is bounded by @xmath90 and the transmission is received by all nodes in @xmath91 , then the outcome of is 1 , where @xmath90 is a parameter . if the transmission is not received by a node @xmath92 then the outcome is @xmath93 . otherwise , the outcome is @xmath93 or @xmath94 , adversarially . ( ) . with = ( @xmath55 ) primitive , a node is able to detect if a transmitter is very close , assuming that it receives the transmitted message . the outcome of is @xmath94 for node @xmath1 in round @xmath56 if @xmath1 _ receives _ a transmission from a node @xmath20 , @xmath95 . otherwise , the outcome of is @xmath93 . this can also be made approximate . _ dynamicity_. we consider a dynamic network where the topology may change adversarially in each round due to node churn ( node arrivals / departures ) and edge changes . we assume that arriving nodes start running the algorithms from an initial configuration , so we do not limit the rate of churn . with edge changes , existing nodes that were not neighbors before , may become neighbors ( e.g. due to mobility ) . the new neighbors may cause too much interference in a too short time , so the edge changes should be limited . we assume the amount of edge changes is bounded for each node @xmath1 , as follows . consider a time interval @xmath96 of length @xmath97 . we require that the number of new neighbors of @xmath1 during @xmath96 ( not counting churn ) is bounded by @xmath98 , where @xmath99 denotes the number of rounds in @xmath96 and @xmath100 is a constant , to be fixed later . we further assume the fraction of rounds in @xmath96 when there are more than @xmath101 new neighbors of @xmath1 is bounded by @xmath102 for every @xmath103 , where @xmath104 . note that there is no restriction on distance changes inside the neighborhood of @xmath1 ( e.g. it is fine for node @xmath1 if its neighbors move , as far as they remain neighbors ) . note that the edge changes may affect the underlying metric , but we require that the upper bounds on metricity and independence are maintained . _ requirements and assumptions_. for the convenience of the reader , we gather together all of our assumptions and requirements in a single place . communication is assumed to succeed in a clear channel ( ) . we assume constant metricity @xmath105 and that @xmath30 has @xmath42-bounded independence with @xmath106 . for the local broadcast problem , we assume that @xmath107 , and for the broadcast problem that @xmath108 . as is standard in fading models , the communication radius @xmath109 is necessarily an @xmath110-fraction of the maximum transmission distance in a clear channel . besides the knowledge required by the primitives that are needed for a particular algorithm , the nodes are assumed to know the precision parameter @xmath55 . a polynomial estimate on the number of nodes , @xmath16 , is needed in dynamic and non - spontaneous algorithms , but not in the static spontaneous problems . knowledge of approximations of model parameters are needed to implement primitives , including @xmath111 , @xmath105 , @xmath53 , @xmath112 , and @xmath68 . knowledge of the maximum degree @xmath3 is not needed . synchronous operation is only assumed in the broadcast algorithm . aspects not defined or constrained are assumed to be under ( adaptive ) adversarial control , including when transmissions that fail are successful , or when nodes appear or disappear from the network . the extent of increases in edge strengths over a period is restricted , as detailed above , while decreases are not and neither are node changes . the dissemination problems are only expected to function with the set of nodes that are sufficiently stable , as detailed in the respective section . in order to keep the contention in the network balanced , we propose a basic procedure called try&adjust , which will be the main building block in our algorithms . the idea is to let each node adapt its transmission probability to the contention detected using the assumed primitive . the parameter @xmath113 describes the passiveness of the newly arriving nodes . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ try&adjust(@xmath114 ) : each node @xmath1 maintains transmission probability @xmath115 in each round @xmath56 , initialized as @xmath116 when @xmath1 enters the network . in round @xmath117 , @xmath1 does : + 1 . transmit with probability @xmath71 , and + 2 . set @xmath118 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the aim for controlling contention is , of course , to ensure that transmissions made have a fair chance of being successful , which means they sufficiently overpower the interference experienced at intended receiver from all other transmissions made in that round . we account for this interference in two ways : the _ contention _ captures the expected interference from the nodes neighbors , while the _ interference _ integrates also the interference from nodes further away . we will measure the contention in the _ vicinity _ of each node @xmath1 , i.e. in @xmath119 , where @xmath76 is a large enough constant . we specify a threshold @xmath120 ( recall @xmath121 from definition ) for measuring contention : if @xmath122 then round @xmath56 is a _ high contention round _ for node @xmath1 and is a _ low contention round _ , otherwise . we further specify a threshold @xmath123 for interference : if @xmath124 then round @xmath56 is a _ high interference round _ for @xmath1 and is _ low interference round _ , otherwise . these thresholds are chosen so as to ensure that in a low - contention / interference round , node @xmath1 will be likely to succeed if it transmits . however , requiring all or most rounds to be low contention for all nodes will lead to high delays instead , it turns out that most rounds will have _ bounded contention _ and low interference , which allows for good progress ; we say node @xmath1 experiences _ bounded contention _ in round @xmath56 if @xmath125 , where @xmath126 is a large enough constant , to be specified later . we say that a round @xmath56 is _ good _ for node @xmath1 if @xmath56 is both bounded contention and low interference round ( in which case , some node in @xmath1 s vicinity has a good chance of successfully transmitting ) . we analyze the properties of try&adjust using the notion of _ a phase _ , the shortest time in which at least @xmath127 rounds occur for all nodes , where @xmath128 is sufficiently large ( given in prop . [ pr : pgoodroundl ] ) . we use @xmath129 to denote a general phase and also the set of rounds in that phase . @xmath130 denotes the number of rounds in a phase @xmath129 ( i.e. @xmath131 ) . the fundamental property of try&adjust ( prop . [ pr : pgoodroundl ] ) is that , for each node and each phase , most of the rounds in the phase are good for that node . this property is then used to show that : if most of the good rounds in a phase have low contention , then node @xmath1 detects idle channel in most of those rounds . otherwise , during a constant fraction of the rounds , a node in the vicinity of @xmath1 mass - delivers . by choosing the parameters carefully , i.e. requiring _ low enough _ contention , we can make sure that during a phase with mostly low contention , the node detects idle channel in _ most of the rounds _ ( more than half ) in a phase , thus leading to an increase of transmission probability by the end of the phase , which , after sufficiently many phases ensures message delivery , w.h.p . on the other hand , during a phase with mostly high contention , there will be many nodes in the vicinity of @xmath1 that successfully transmit , leading to lower contention . these ideas are applied in thms . [ thm : localbroadcast ] and [ thm : broadcast ] . the core idea behind this analysis is based on @xcite . [ pr : pgoodroundl ] let @xmath132 . if constants @xmath133 and @xmath134 are large enough then for each node @xmath1 and phase @xmath129 , with probability @xmath135 , a @xmath136-fraction of the rounds in @xmath129 are good . the proof is rather technical and is deferred to sec . [ sec : contentiondetails ] but the intuition is as follows . the contention in each neighborhood must be bounded most of the time , because when it becomes large , it has a high chance of being decreased due to busy channel . moreover , we show that in expectation , the contention in _ all _ local neighborhoods is bounded , which is then combined with a geometric argument to show that the expected interference at each node is low most of the time . we derive from the fundamental property two useful propositions . the first says that if contention is high , then nodes in the vicinity deliver the message . [ pr : pdecrease ] assume that constants @xmath137 are large enough . for each node @xmath1 and phase @xmath129 , if at least @xmath138-fraction of the rounds of @xmath129 are of high contention , then @xmath139 nodes in @xmath140 mass - deliver , with probability @xmath135 . to this end , we first show that if a round is good for node @xmath20 and a node in its vicinity transmits , then it mass - delivers with constant probability , utilizing both metric assumptions and the properties of good rounds . we then argue that since most rounds are good ( by prop . [ pr : pgoodroundl ] ) and most rounds have by assumption sufficient contention , many rounds will be both good and with sufficient contention , and in each of those , a node in the vicinity of @xmath1 is likely to transmit and succeed . when contention is low in a lot of rounds of a phase , the node will detect idle channel by the primitive in many rounds . this will actually happen during many good rounds , which have the low local contention and low external interference to allow for this detection . [ pr : idle ] assume that @xmath137 are large enough . for each node @xmath1 and phase @xmath129 , if at least @xmath141-fraction of the rounds of @xmath129 are low contention rounds , then with probability @xmath135 , in at least @xmath142-fraction of the rounds of @xmath129 , @xmath1 will detect idle channel and have low contention and low interference . we propose an algorithm for asynchronous local broadcast in dynamic networks . the algorithm is an extension of the try&adjust procedure , where the nodes try to balance the contention in the network and stop transmitting as soon as they deliver their message . we assume the nodes are powered with and primitives . note that the passiveness parameter is set to @xmath143 , which means that the transmission probability of nodes does not get below @xmath144 . localbcast : each node @xmath1 executes try&adjust(1 ) with the following additional step : if @xmath1 transmits and detects , it stops ( i.e. @xmath145 for @xmath146 ) . we will estimate the performance of the algorithm using the notion of _ dynamic degree _ , defined as follows . given a parameter @xmath147 , we denote @xmath148 for node @xmath1 and rounds @xmath149 with @xmath150 , where @xmath151 denotes the in - ball @xmath119 in round @xmath33 . below we prove that if there are not too many node insertions in the neighborhood of a node @xmath1 , then @xmath1 mass - delivers ( delivers to all its neighbors ) in time comparable to its dynamic degree with @xmath76 a constant . the main tools for proving the bound are props . [ pr : pdecrease ] and [ pr : idle ] . first we argue that if there is a phase of mostly low contention , then node @xmath1 will deliver its message , w.h.p . then we show that if the insertions are not too intensive then the contention around @xmath1 will decrease and a phase with mostly low contention will happen . [ thm : localbroadcast ] there is a constant @xmath147 , such that a node @xmath1 performing localbcast asynchronously in a time interval @xmath152 $ ] with @xmath153 mass - delivers , w.h.p . , provided that @xmath154 . note that the assumption @xmath154 is needed only for making the claim w.h.p . : it can be relaxed to higher degree polynomials by only increasing constant factors . let us fix constants @xmath155 so that props . [ pr : pdecrease ] and [ pr : idle ] with @xmath156 . we partition @xmath96 into phases ( for node @xmath1 ) and classify them into two types : ( type a ) phases @xmath129 where at least @xmath138-fraction of rounds are high contention rounds ( i.e. @xmath157 ) , and ( type b ) phases @xmath129 where at least @xmath141-fraction of rounds are low contention rounds ( i.e. @xmath158 ) . node @xmath1 mass - delivers in a type b phase , w.h.p . let @xmath159 be the low contention and interference rounds during phase @xmath129 where @xmath1 detects idle channel . by prop . [ pr : idle ] @xmath160 > 1-o(n^{-3})$ ] . assume for now that the latter happens . for each @xmath161 we have @xmath162 . let us call this operation doubling . the value of @xmath71 at the beginning of the phase is at least @xmath144 , so @xmath163 doubling operations are sufficient to raise it to @xmath164 . the probability can be further halved during the phase at most @xmath165 times . thus , we may assume we have at most @xmath166 halving and at least @xmath167 doubling operations applied to an initial value @xmath164 . if @xmath168 , then the total number of halving operations is less than @xmath169 . it follows that @xmath1 has @xmath170 in at least @xmath171 low contention / interference rounds . by lemma [ le : broadgood ] , in each such round , @xmath1 mass - delivers with probability at least @xmath172 ; hence , if @xmath128 is large enough , @xmath1 mass - delivers in @xmath129 , w.h.p . it remains to argue that there will be a type b phase during time interval @xmath96 . consider a type a phase @xmath129 . [ pr : pdecrease ] implies that with probability @xmath135 , @xmath173 nodes in @xmath119 deliver their message and stop during phase @xmath129 . thus , with probability @xmath174 , there are at most @xmath175 type a phases with @xmath176 a constant , as there are at most @xmath177 nodes in @xmath140 during the time interval @xmath96 ( also recall that @xmath178 ) . we conclude that if @xmath96 consists of at least @xmath179 phases , it will contain a type b phase and @xmath1 will deliver its message w.h.p . * implications for static networks*. in static networks , the parameter @xmath177 is at most @xmath180 if @xmath76 is constant , where @xmath181 is the maximum size of a neighborhood in the network . thus , we obtain the following optimal result ( up to constant factors ) for static networks , as @xmath3 and @xmath163 are lower bounds even when running in the spontaneous mode @xcite . when running localbcast in a static asynchronous network , each node @xmath1 completes local broadcast in @xmath182 rounds , w.h.p . * remark . * in the special case when the nodes can start executing the algorithm simultaneously , i.e. in the _ spontaneous mode _ , the nodes need not know an upper bound on the size of the network . indeed , each node may start running try&adjust with initial probability set to an arbitrary value and with no lower limit . the first phase will be spent for stabilization and can be ignored , while the argument for the rest of the phases is nearly identical to the one in thm . [ thm : localbroadcast ] . for the broadcast problem , we assume nodes communicate in synchronized rounds of equal length . each round consists of two slots . the idea is to use the first slot of each round for disseminating the message with try&adjust and the second slot for notifying nodes which have no uninformed neighbors . the latter is accomplished by using higher precision primitives , namely ( @xmath183 ) and ( @xmath183 ) , when executing try&adjust . this helps to detect a transmission that is successfully received by all nodes in @xmath184 of a node @xmath1 . upon detecting such a transmission , node @xmath1 resends the message in the second slot , in order to inform nodes @xmath20 with @xmath185 that their neighborhood ( @xmath186 ) has been covered . a node @xmath20 can detect that @xmath185 using . the algorithm is presented below . we assume the passiveness parameter @xmath114 of try&adjust is large enough , to be defined later . note that the algorithm works for the non - spontaneous mode , as nodes act only after receiving the message . bcast(@xmath114 ) : initially , only a source node @xmath187 has the message . a node @xmath1 , upon receiving a message , starts executing try&adjust(@xmath114 ) in the first slot of rounds . in addition , in each round @xmath56 , + 1 . if @xmath1 detects in the first slot , it retransmits in the second slot and restarts try&adjust(@xmath114 ) , + 2 . if @xmath1 receives a message in the first slot and detects in the second slot , it restarts try&adjust(@xmath114 ) . in order to evaluate the progress of the algorithm , we use a notion of a _ dynamic distance _ , as defined below . let @xmath188 be a parameter . a sequence @xmath189 is called a _ stable @xmath187-@xmath1 path _ if there is a sequence @xmath190 of time intervals with @xmath191 $ ] , such that @xmath192 , @xmath193 and nodes @xmath194 and @xmath195 are both alive and @xmath196 during @xmath197 . the _ time - length _ of a stable @xmath187-@xmath1 path is @xmath198 . the _ stable @xmath187-@xmath1 distance _ @xmath199 is defined as the minimum time - length of a stable @xmath187-@xmath1 path . note that a stable path need not be connected at any fixed point in time . moreover , most of the nodes might be missing at any given point in time . the core idea behind the analysis of the following theorem is similar to the case of local broadcast : we show that as soon as a neighbor @xmath20 of a node @xmath1 has the message and @xmath20 and @xmath1 keep being neighbors for @xmath200 rounds , @xmath1 will receive a transmission of @xmath20 during those rounds . [ thm : broadcast ] assume the edge change rate @xmath100 is sufficiently small . there are constants @xmath201 , such that when running bcast(@xmath114 ) in the synchronous mode , each node @xmath1 receives the message in @xmath202 rounds w.h.p . let us fix constants @xmath155 so that props . [ pr : pdecrease ] and [ pr : idle ] hold with interference threshold @xmath203 , where @xmath204 and @xmath205 are the parameters corresponding to precision @xmath183 . we set the passiveness parameter of try&adjust to @xmath206 and the stable distance parameter @xmath207 . consider a node @xmath20 that receives the message in round @xmath56 . let @xmath1 be such that @xmath208 in the time interval @xmath209 $ ] ( and both stay alive during @xmath96 ) . the theorem follows by an induction and union bound from the claim below . node @xmath1 gets the message of @xmath20 during @xmath96 , with probability @xmath210 . we prove the claim by contradiction , and assume that @xmath1 can not get the message during @xmath96 . let us split @xmath96 into phases ( for @xmath20 ) and , similar to the proof of thm . [ thm : localbroadcast ] , classify the phases @xmath129 into types : ( type a ) at least @xmath138-th of rounds in @xmath129 are high contention rounds for @xmath20 @xmath211 , and ( type b ) at least @xmath141-th of rounds in @xmath129 are low contention rounds for @xmath20 @xmath212 . we show that with probability @xmath135 , all phases in @xmath96 are of type b. consider a type a phase @xmath129 . we know from prop . [ pr : pdecrease ] that , with probability @xmath135 , there is a set @xmath39 of @xmath139 nodes in @xmath140 that deliver their messages and restart try&adjust during @xmath129 . we split @xmath39 into three subsets , @xmath213 those inserted by churn during @xmath129 , @xmath214 those inserted by edge change during @xmath129 and @xmath215 the rest . consider @xmath214 first . we know that in each local neighborhood , there are at most @xmath216 nodes arriving due to edge changes . note that @xmath217 can be covered with @xmath218 local neighborhoods ; hence , the total number of nodes arriving in @xmath217 due to edge changes is @xmath219 . since @xmath220 , setting @xmath221 a small enough constant gives @xmath222 . now consider @xmath215 . by the primitive , for each node @xmath223 , all nodes in @xmath224 also restart try&adjust , setting their transmission probability to @xmath225 . by the setting of @xmath226 , the probability that a node restarting try&adjust in phase @xmath129 transmits again in @xmath129 is at most @xmath227 . thus , with probability at least @xmath135 , the nodes in @xmath214 constitute an @xmath228-packing of @xmath140 and are at most @xmath229 . as for @xmath213 , the probability that a newly arriving node transmits during phase @xmath129 is at most @xmath227 , so @xmath230 w.h.p . these bounds lead to a contradiction @xmath231 , if constant @xmath128 is large enough . thus , each phase @xmath129 of @xmath96 is of type b with probability @xmath135 . it remains to recall that @xmath96 contains constant number of phases . now assume that @xmath129 is of type b. in a similar way as in theorem [ thm : localbroadcast ] , we can apply prop . [ pr : idle ] and show that if @xmath232 at the beginning of @xmath129 , then by the end of @xmath129 , the effect on @xmath71 is equivalent to applying at least @xmath233 of @xmath234 operations on @xmath235 . since @xmath236 at the beginning of @xmath96 is at least @xmath225 , it will take at most 11 phases to increase the probability to the value @xmath164 . then , a single phase will suffice for @xmath20 to deliver its message ( including to node @xmath1 ) , with probability at least @xmath210 . * implications for static networks*. when the network is static , theorem [ thm : broadcast ] can be reformulated in terms of hop - distance @xmath237 in the communication graph , which is defined as the length of the shortest directed @xmath187-@xmath1 path in @xmath238 : we have that @xmath239 for any node @xmath1 . note also that in this setting nodes that succeeded transmitting or detected need not continue the algorithm , so they stop transmitting . in this case , setting the passiveness parameter to @xmath143 suffices . we call this variant of the algorithm bcast@xmath240 . when running bcast@xmath240 in synchronous non - spontaneous mode in a static network with source @xmath187 , each node @xmath1 receives the message in @xmath241 rounds w.h.p . when the communication graph is strongly connected , the broadcast from any source node is completed in @xmath242 rounds , where @xmath243 is the diameter of the communication graph . in the spontaneous mode , the bound above can be further improved to @xmath244 ; see appendix [ sec : spontaneous ] . this is based on finding a constant - density dominating set in @xmath200 time @xcite and simultaneously propagating along the dominators in @xmath244 time . we can therefore extend the approach based on @xcite to uniform algorithms in bounded - independence metrics . these results are close to best possible . we show below that in order to obtain bounds of that magnitude , it is necessary to have the primitive . to this end , we extend the lower bound construction of ( * ? ? ? * thm . 7 ) for `` compact sinr '' to our setting . this construction leverages the property of our model that there can be arbitrarily many nodes that are mutually close to each other . namely , the bounded - independence metric is strictly more relaxed than the standard euclidean metrics . indeed , there is a @xmath245-round broadcast algorithm for the sinr model that does not need or other carrier sensing primitives @xcite . what the lower bound then illustrates is that to obtain such results , one must depend on opportune traits of the sinr model that we have tried to avoid and are not necessary for problems like local broadcast . thus we can observe concrete tradeoffs depending on model assumptions . [ thm : brlowerbound ] for every ( possibly randomized ) broadcast algorithm @xmath246 that uses neither node coordinates nor primitive , there is a @xmath247-bounded - independence metric space where @xmath246 needs @xmath248 rounds to do broadcast in a @xmath249-broadcastable network , even if the nodes have and primitives and operate spontaneously . 10 n. alon , a. bar - noy , n. linial , and d. peleg . . in _ stoc _ , pages 274285 , 1989 . n. alon , a. bar - noy , n. linial , and d. peleg . a lower bound for radio broadcast . , 43(2):290298 , 1991 . r. bar - yehuda , o. goldreich , and a. itai . on the time - complexity of broadcast in radio networks : an exponential gap between determinism and randomization . in _ podc87 _ , 1987 . l. barenboim and d. peleg . nearly optimal local broadcasting in the sinr model with feedback . in _ sirocco _ , pages 164178 , 2015 . m. h. bodlaender and m. m. halldrsson . beyond geometry : towards fully realistic wireless models . in _ podc _ , pages 347356 , 2014 . m. h. bodlaender , m. m. halldrsson , and p. mitra . . in _ podc _ , pages 355364 . acm , 2013 . i. chlamtac and s. kutten . on broadcasting in radio networks problem analysis and protocol design . , 33(12):12401246 , 1985 . b. s. chlebus , d. r. kowalski , and s. vaya . distributed communication in bare - bones wireless networks . , 2015 . b. n. clark , c. j. colbourn , and d. s. johnson . unit disk graphs . , 86(1):165177 , 1990 . a. e. clementi , a. monti , and r. silvestri . round robin is optimal for fault - tolerant broadcasting on wireless networks . , 64(1):8996 , 2004 . a. czumaj and w. rytter . broadcasting algorithms in radio networks with unknown topology . in _ focs03 _ , 2003 . s. daum , s. gilbert , f. kuhn , and c. newport . . in _ disc _ , pages 358372 , 2013 . b. derbel and e. talbi . . in _ icdcn10 _ , pages 155166 , 2010 . s. dolev , s. gilbert , m. khabbazian , and c. newport . . in _ disc _ , pages 252267 , 2011 . a. fanghnel , t. kesselheim , and b. vcking . . , 412(24):26572667 , 2011 . m. ghaffari , s. gilbert , c. newport , and h. tan . optimal broadcast in shared spectrum radio networks . in _ opodis12 _ , 2012 . m. ghaffari , b. haeupler , and m. khabbazian . randomized broadcast in radio networks with collision detection . in _ podc13 _ , 2013 . m. ghaffari , b. haeupler , n. lynch , and c. newport . bounds on contention management in radio networks . in _ disc _ , 2012 . m. ghaffari , e. kantor , n. lynch , and c. newport . multi - message broadcast with abstract mac layers and unreliable links . in _ podc14 _ , 2014 . m. ghaffari , n. lynch , and c. newport . the cost of radio network broadcast for different models of unreliable links . in podc13 _ , 2013 . o. goussevskaia , t. moscibroda , and r. wattenhofer . . in _ dialm - pomc 08 _ , pages 3544 , 2008 . p. gupta and p. kumar . , 46(2):388404 , 2000 . m. m. halldrsson . . , 9(1):7:17:20 , dec . m. m. halldrsson , s. holzer , and n. a. lynch . a local broadcast layer for the sinr network model . in _ podc _ , pages 129138 , 2015 . m. m. halldrsson , s. holzer , p. mitra , and r. wattenhofer . . in _ soda13 _ , pages 15951606 , 2013 . m. m. halldrsson and p. mitra . . in _ icalp _ , pages 625636 , 2011 . m. m. halldrsson and p. mitra . . in _ podc _ , 2012 . m. m. halldrsson and p. mitra . . fomc _ , 2012 . m. m. halldrsson , y. wang , and d. yu . leveraging multiple channels in ad hoc networks . in _ podc15 _ , 2015 . m. hoefer , t. kesselheim , and b. vcking . approximation algorithms for secondary spectrum auctions . in _ proceedings of the twenty - third annual acm symposium on parallelism in algorithms and architectures _ , spaa 11 , pages 177186 , new york , ny , usa , 2011 . r. impagliazzo and v. kabanets . constructive proofs of concentration bounds . in _ approx - random _ , pages 617631 , 2010 . t. jurdzinski , d. kowalski , m. rozanski , and g. stachowiak . distributed randomized broadcasting in wireless networks under the sinr model . in _ disc _ , pages 373387 , 2013 . t. jurdzinski , d. kowalski , m. rozanski , and g. stachowiak . on the impact of geometry on ad hoc communication in wireless networks . in _ podc 14 _ , 2014 . t. jurdzinski , d. r. kowalski , and g. stachowiak . . in _ fct13 _ , pages 195209 , 2013 . t. jurdzinski , d. r. kowalski , and g. stachowiak . distributed deterministic broadcasting in wireless networks of weak devices . in _ icalp13 _ , 2013 . t. kesselheim and b. vcking . . in _ disc _ , pages 163178 , 2010 . d. kowalski and a. pelc . broadcasting in undirected ad hoc radio networks . in _ podc03 _ , 2003 . f. kuhn , n. lynch , and c. newport . brief announcement : hardness of broadcasting in wireless networks with unreliable communication . in _ disc _ , 2009 . f. kuhn , n. lynch , c. newport , r. oshman , and a. richa . . in _ podc _ , pages 336345 . acm , 2010 . f. kuhn , t. moscibroda , and r. wattenhofer . . in _ mobicom _ , pages 260274 . acm , 2004 . f. kuhn and r. oshman . dynamic networks : models and algorithms . , 42(1):8296 , 2011 . f. kuhn , r. wattenhofer , and a. zollinger . ad hoc networks beyond unit disk graphs . , 14:715729 , 2008 . e. kushilevitz and y. mansour . . , 27(3):702712 , 1998 . c. lenzen and r. wattenhofer . distributed algorithms for sensor networks . , 370(1958 ) , 2012 . t. moscibroda and r. wattenhofer . . , 21(4):271284 , 2008 . c. newport radio network lower bounds made easy . in _ disc14 _ , 2014 . c. scheideler , a. richa , and p. santi . . in _ mobihoc 08 _ , pages 91100 , 2008 . s. schmid and r. wattenhofer . algorithmic models for sensor networks . in _ wpdrts06 _ , 2006 . j. schneider and r. wattenhofer . . in _ podc _ , pages 210219 , 2009 . j. schneider and r. wattenhofer . in _ disc _ , pages 133147 . springer - verlag , 2010 . d. yu , q .- s . hua , y. wang , and f. lau . . dcoss 12 _ , pages 132139 , 2012 . d. yu , q .- s . hua , y. wang , h. tan , and f. lau . . in _ sirocco12 _ , pages 111122 , 2012 . d. yu , q .- s . hua , y. wang , j. yu , and f. lau . . in _ infocom13 _ , pages 24272435 , 2013 . d. yu , y. wang , q .- s . hua , and f. c. m. lau . . in _ distributed computing in sensor systems and workshops ( dcoss ) , 2011 international conference on _ , pages 18 , 2011 . d. yu , y. wang , y. yan , j. yu , and f. lau . speedup of information exchange using multiple channels in wireless ad hoc networks . in _ infocom15 _ , 2015 . _ wireless models : _ considering wireless interference , there are two classes of wireless network models : graph - based and physical models . basically , the graph - based models define a local and binary type of interference , while the physical models consider fading effect of signal on wireless channels . the most classical graph - based model is the radio network model @xcite . in this model , the network is modeled using a communication graph , where each pair of nodes that can communicate with each other is connected by an edge . it defines the interference just from direct neighbors , and a transmission can succeed if and only if there is only one neighbor of the receiver transmitting . there are many widely used variants of the classical radio network models , including : 1 ) the @xmath0-hop model where the interference comes from @xmath0-hop neighbors @xcite ; 2 ) unit disc graph ( udg ) model @xcite which defines the neighborhood using a unit disc ; 3 ) quasi unit disc graph ( qudg ) model @xcite which just defines all pairs of nodes with distance at most @xmath76 for some given @xmath250 $ ] are adjacent , and leave the ` grey ' area in @xmath251 $ ] being determined by an adversary ; 4 ) protocol model @xcite , where each node has a transmission range and an interference range , and a successful transmission occurs if a node falls into the transmission range of a transmitter and outside the interference ranges of all other transmitter ; 5 ) bounded - independence graph ( big ) model @xcite , which defines abstractly and requires that the size of the maximal independent set in the @xmath33-hop neighborhood of each node is bounded by a polynomial function with @xmath33 . though the graph - based interference models miss certain crucial aspects of actual wireless networks , the simple definition of these models can help derive novel insights into distributed solutions to wireless problems . physical models , also known as sinr models @xcite , capture the fading and cumulative features of receptions in actual wireless environments . the default assumption is that interference fades with a polynomial of the distance , and transmission succeeds only if the received signal strength is sufficiently larger than the total interference plus noise . recently , the sinr model has attracted great attentions in the distributed community @xcite . most of these works focus on networks embedded in euclidean space , while many of the results hold also for doubling or `` bounded growth '' metrics @xcite . those doubling metrics constrain growth at every ( or arbitrarily small ) granularity , while ours only bounds regions proportional to the transmission range , in particular capturing bounded independence graphs ( big ) . for more on wireless models , please refer to @xcite . _ local broadcast _ : in the radio network model , probably the first local broadcast result was a randomized algorithm of alon et al . @xcite in a synchronous model , running in @xmath6 rounds . derbel and talbi @xcite later generalized their algorithm to work without knowledge of @xmath3 and their proposed algorithm can accomplish local broadcast in @xmath7 rounds . the decay strategy also yields an @xmath6 time algorithm for local broadcast without knowledge of @xmath3 @xcite . goussevskaia et al . @xcite gave the first results for local broadcast in the sinr model , running in time @xmath6 and @xmath252 with and without knowledge of @xmath3 , respectively . the latter was improved in @xcite and further improved , independently , to @xmath7 time @xcite . with free acknowledgements , this was improved to @xmath253 @xcite . when additionally @xmath3 is known , this was further improved recently to @xmath8 in the spontaneous setting @xcite . the speedup of multiple channels on local broadcast was considered in @xcite _ broadcast _ : the complexity of broadcasting is well understood in graph - based models . in the radio network model , bar - yehuda et al . @xcite presented the decay protocol which can accomplish non - spontaneous broadcast in @xmath254 rounds , where @xmath46 is the diameter . this result was improved to @xmath9 independently by czumaj and rytter @xcite , and kowalski and pelc @xcite . these algorithms can be viewed as clever optimizations of the decay protocol and match the lower bound @xcite . with collision detection , this lower bound was recently broken in @xcite , where a solution of @xmath10 was given . broadcast in multi - channel radio networks was considered in @xcite . for the udg model , an @xmath255 time algorithm was given in @xcite in the spontaneous setting . broadcasting has been treated in the sinr model under a variety of assumptions . some are stronger than ours ( location information @xcite , power control @xcite ) . some relax the assumption about the connectivity property , but incur necessarily super - linear complexity @xcite . results in our setting , but without carrier sensing , include time complexity of @xmath256 @xcite ( see also @xcite ) , where @xmath12 denotes the maximum ratio between distances of stations connected in the communication graph ; and @xmath13 @xcite . in the spontaneous setting , where the nodes can build an overlay structure along which the message is then propagated , scheideler et al . @xcite solved the problem in @xmath257 rounds using physical carrier sensing . yu et al . @xcite solved the problem in @xmath258 rounds using power control , allowing stations to decide the strength of a transmitted signal in each step . and the algorithm in @xcite then runs in time @xmath254 . _ distributed models of temporal variability _ : dynamic networks have been studied extensively in recent years ( see @xcite for a survey ) , but generally not in the presence of interference . the _ dual graph _ model @xcite ( originally due to @xcite ) was designed to capture inherent _ unreliability _ in wireless networks , much of which can be due to dynamicity . the main focus of that work is on extending the radio network model in general graphs . importantly , the dual graph model does not distinguish between interference and communication edges ; only that the unreliable edges can transmit both interference and the usual communication , but their availability is under adversarial control . thus , there is no way to capture interference from further away nodes . most problems become extremely difficult against a powerful adversary , and to get good result , one must assume a much weaker one @xcite . this model only involves edge behavior and not node changes ( churn ) . both the local broadcast @xcite and global broadcast @xcite problems are studied in the dual graph model . a dynamic model that considers node insertion is the _ unstructured _ model @xcite , which admits arbitrary wake - up mode and asynchronous communication . this model was first proposed in the unit - disc setting , and then extended to bounded independence graphs ( big ) @xcite and sinr @xcite . it has been widely used in the solution of a variety of distributed wireless problems @xcite , including local broadcast @xcite , but there are no known global broadcast results . the model neither consider node deletion nor edge changes . hence , the impact of dynamicity on global communication is largely unexplored . [ [ modeling - communication ] ] modeling communication + + + + + + + + + + + + + + + + + + + + + + our communication model captures most known algorithmic wireless models , as it is demonstrated below on the example of sinr , disk - graph based and protocol models . note that we assume below the distance @xmath259 to be symmetric , but the results hold also for `` almost symmetric '' functions , i.e. when there is a constant @xmath260 such that @xmath261 for all @xmath262 . _ sinr model . _ consider a network in a metric space . in the sinr model of communication , if @xmath39 is the set of simultaneously transmitting nodes in the network , a node @xmath1 receives the transmission of node @xmath20 if and only if @xmath263 where constants @xmath113 and @xmath264 denote the minimum sinr threshold and the ambient noise , respectively . note that @xmath265 in this setting . we can implement with parameters @xmath266 and @xmath267 , as shown in the proposition below . [ pr : scc ] if the interference at a node @xmath1 is less than @xmath68 in round @xmath56 , then @xmath1 will deliver its message if it transmits . note that if the interference at node @xmath1 is not more than @xmath68 then there is no node @xmath268 transmitting in round @xmath56 , as otherwise the interference at @xmath1 would be at least @xmath269 ; hence , only nodes in @xmath270 can transmit . consider an arbitrary node @xmath271 . for each node @xmath272 , @xmath273 . then , the interference at @xmath274 is at most : @xmath275 which implies that node @xmath274 receives @xmath1 s transmission : @xmath276 . _ the udg and ubg models . _ these models are described by geometric graphs : a node @xmath20 receives a message from another node @xmath1 if and only if @xmath1 is the only transmitting neighbor of @xmath20 . in the unit disk graph ( udg ) and unit ball graph ( ubg ) models the nodes are located in a metric space and two nodes are connected by an edge if and only if their distance is at most @xmath277 . the functionality of can be modeled as follows : the transmission of a node @xmath1 is received by all its neighbors if there is no other node at distance less than @xmath278 from @xmath1 transmitting simultaneously , i.e. we can set the parameters to @xmath279 and @xmath280 . _ the quasi - udg model . _ the quasi - udg model is an extension of the udg model : a ) if @xmath281 then @xmath20 and @xmath1 are connected by an edge , b ) if @xmath282 then they are disconnected , c ) otherwise , @xmath20 and @xmath1 may be connected or not . in this case may be implemented by setting @xmath279 and @xmath283 , with the adversary constrained to follow the specific static situation captured by the qudg . _ the protocol model . _ in the protocol model , the nodes are in a metric space and there are two radii : @xmath277 the communication radius , and @xmath284 the interference radius . a node @xmath1 receives the transmission of a node @xmath20 if and only if : 1 ) @xmath1 is in the communication range of node @xmath20 , i.e. @xmath281 , and 2 ) there is no transmitting node @xmath274 such that @xmath1 is in the interference range of @xmath274 : for each transmitting node @xmath285 , @xmath286 . may be implemented here by setting @xmath279 and @xmath283 . _ the big model . _ in the bounded independence graph ( big ) model , for a parameter @xmath287 , we are given a graph on the nodes with the property that for every node @xmath1 and every @xmath288 , the maximum independent set in the @xmath0-neighborhood of @xmath1 is @xmath289 . the shortest - path distance metric on the graph is now naturally a @xmath290-bounded independence metric . to fit in our model , the growth parameter @xmath287 must be less than @xmath105 . _ @xmath0-hop variants . _ these graph models can be naturally generalized to a model on interference , where nodes of distance at most @xmath0 cause interference , for some @xmath291 . we capture this by extending @xmath112 as needed . [ [ implementing - primitives - with - physical - carrier - sensing ] ] implementing primitives with physical carrier sensing + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + one way of implementing the primitives mentioned in this paper is to use physical carrier sensing , i.e. we assume the nodes have technology to detect if the interference ( plus noise ) is higher than a given threshold . we show below how to implement , and primitives using carrier sensing . * primitive . * the primitive can be implemented using a carrier sensing threshold @xmath292 and setting the parameter @xmath293 ; busy channel is detected if and only if the interference is at least @xmath96 . we will need the following technical fact . [ ieq ] for every @xmath294 $ ] , @xmath295 , it holds that @xmath296 if @xmath297 in round @xmath56 then all nodes in @xmath298 detect busy channel with probability at least @xmath299 . in particular , we can take @xmath300 if @xmath301 . by the setting of @xmath96 and the definition of @xmath74 , if two nodes in @xmath74 transmit in round @xmath56 , then all nodes in @xmath74 will detect busy channel . hence , the probability of all nodes in @xmath74 detecting busy channel is at least the probability of more than one node transmitting in round @xmath56 . the probability of no node transmitting is @xmath302 by lemma [ ieq ] . the probability of exactly one node transmitting is : @xmath303 where we used the assumption that @xmath304 and lemma [ ieq ] . thus , the probability of detecting busy channel is at least @xmath305 . for every @xmath147 , if @xmath306 and @xmath307 then @xmath1 detects idle channel with probability at least @xmath308 . by the setting of the threshold @xmath96 , if there is no node transmitting in @xmath119 then node @xmath1 will detect idle channel . thus , the probability that @xmath1 detects idle channel is at least @xmath309 using lemma [ ieq ] . * primitive . * in order to detect successful transmission , we can use interference threshold @xmath310 and set @xmath311 , where @xmath312 and @xmath313 are the parameters of . if a node @xmath1 senses that the interference is no higher than @xmath96 , it knows that : 1 . there is no node in @xmath314 transmitting , as otherwise the interference would be at least @xmath315 , 2 . the interference is at most @xmath68 ; thus , it knows that its transmission has been received by all neighbors in @xmath316 by . * primitive . * if a node @xmath1 receives a message from a node @xmath20 then @xmath1 can separate the signal from the interference and measure the received signal strength . as the nodes use uniform power assignment , node @xmath1 knows that @xmath95 if the received signal is stronger than @xmath317 . [ [ implementing - primitives - by - other - means ] ] implementing primitives by other means + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the primitives can frequently be implemented in other ways , often with the logarithmic blowup that explains the differences with the best carrier - sense - free results . * * in an asynchronous system , it may be impossible to implement by other means than carrier sense . in a synchronized system , however , we can be achieved with logarithmic or polylogarithmic factor overhead . consider a given round . for each probability @xmath318 , repeat @xmath319 times : the senders in the original round transmit with probability @xmath320 . using concentration bound with @xmath50 sufficiently large , one can infer the contention within an small approximation , with high probability . such a strategy has been applied , e.g. , in @xcite . * primitive . * a simple strategy is to work with only probabilistic guarantees of a transmission being received by all neighbors . then , simply repeat the protocol until this has been achieved @xmath319 times , which gives an guarantee , w.h.p . this approach underlies , e.g. , the local broadcast algorithms without carrier sense @xcite . * primitive . * this primitive , which is essential for dominator - based strategies for broadcast , can be implemented using power control : by lowering the power on all units appropriately , one can ensure that nodes further away ( by a small constant factor ) will not be able to hear the message due to the ambient noise term , see e.g. @xcite . alternatively , one can assume that distances can be determined in other ways , such as by gps @xcite . recall that we need to prove the following . we assume that at the beginning of the _ first _ phase under consideration , the contention in the whole network is bounded by a constant . this holds for all algorithms in this paper , as the initial probability of nodes is always at most @xmath321 . * * proposition [ pr : pgoodroundl ] . * * let @xmath132 . if constants @xmath133 and @xmath134 are large enough then for each node @xmath1 and phase @xmath129 , with probability @xmath135 , a @xmath136-fraction of the rounds in @xmath129 are good . recall that in a good round , there should be both bounded contention and low interference . the proof is split into two parts , each handling one of these properties . [ pr : pgoodroundl ] follows by simply combining those two parts . we will need the following concentration bounds . @xcite[le : chernoff ] consider a collection of binary random variables @xmath322 , and let @xmath323 . if there are probabilities @xmath324 with @xmath325\leq \prod_{i\in s}p_i$ ] for every set @xmath326 , then it holds for @xmath327 and @xmath328 that @xmath329\leq\left(\frac{e^{\delta}}{(1+\delta)^{1+\delta}}\right)^\mu\leq e^{-\frac{\delta^2\mu}{2(1+\delta/3)}}.\ ] ] if , on the other side , there are probabilities @xmath324 with @xmath325\geq \prod_{i\in s}p_i$ ] for every set @xmath326 , then it holds for @xmath327 and @xmath330 that @xmath331\leq\left(\frac{e^{-\delta}}{(1-\delta)^{1-\delta}}\right)^\mu\leq e^{-\delta^2\mu/2}.\ ] ] first , we show that for each fixed node @xmath20 , the contention in the _ local _ neighborhood @xmath332 is bounded in most of the rounds of a phase . to this end , we show that in each round , the contention is either already low or will be halved with significant probability , then apply a concentration bound to show the claim . recall the constant @xmath333 from the definition of . [ le : phi ] let @xmath129 be a time interval and assume that @xmath334 at the beginning of @xmath129 . then for every @xmath335 , @xmath336 happens at most @xmath337 times during @xmath129 , with probability @xmath338 , where @xmath339 and @xmath340 is the edge change parameter . by the assumptions on edge changes , we have that during phase @xmath129 , the fraction of rounds where more than @xmath341 nodes become a neighbor of @xmath20 because of edge changes is @xmath102 . let @xmath159 denote the remaining set of rounds . we have @xmath342 . in each of those rounds , the contribution of edge changes in @xmath343 is clearly at most @xmath344 . next we bound the number of rounds in @xmath159 , where @xmath345 . consider such a round @xmath161 . let @xmath346 . by the definition of , all nodes in @xmath332 detect busy channel in round @xmath56 and halve their transmission probabilities with probability at least @xmath347 . thus , @xmath348\ge 1-\mu'$ ] , where the additive @xmath164 accounts for the sum of probabilities of the nodes that just join the ball @xmath332 due to node churn ( recall that each of them has an initial probability at most @xmath349 ) and @xmath344 is an upper bound on the contention due to edge changes ( by the definition of @xmath159 ) . for each round @xmath161 , let us define a binary random variable @xmath350 as follows : @xmath351 if @xmath352 or @xmath353 and @xmath354 ( recall that @xmath355 ) and @xmath356 otherwise . by the discussion above , we have : @xmath357&=pr[p_{t}(u)\leq\phi]+pr[p_{t}(u)>\phi\wedge p_{t+1}(u)\leq\frac{5}{6}p_t(u)]\\ & = pr[p_{t}(u)\leq\phi]+pr[p_{t}(u)>\phi]\cdot pr[p_{t+1}(u)\leq\frac{5}{6}p_t(u)|p_{t}(u)>\phi]\\ & \geq pr[p_{t}(u)\leq\phi]+pr[p_{t}(u)>\phi]\cdot ( 1-\mu')\\ & \geq 1-\mu ' . \end{aligned}\ ] ] this implies that for each round @xmath358 and each subset @xmath359 of earlier rounds ( @xmath360 for @xmath361 ) , @xmath362\geq 1-\mu'$ ] and @xmath363\leq \mu'$ ] . thus , for each subset @xmath364 we get @xmath365\le \mu'^{|s| } , $ ] and if we denote @xmath366 , then @xmath367\le \mu ' |h'|=\mu |h|$ ] , where we denote @xmath368 . thus , we can apply chernoff bound ( lemma [ le : chernoff ] ) with @xmath369 to bound @xmath370 with high probability : @xmath371\le 2^{-\mu|h|}.\ ] ] using this bound , we obtain a bound on the number of rounds @xmath56 with @xmath372 . let @xmath373 denote this number . consider a maximal interval @xmath374\subseteq h'$ ] such that @xmath345 for all @xmath375 . for each round @xmath375 , @xmath376 if @xmath351 and @xmath377 otherwise . by maximality of @xmath378 , if @xmath379 _ is not _ the first round of phase @xmath129 , then we have @xmath380 . then a simple calculation shows that @xmath356 for at least @xmath381 part of the rounds of @xmath378 , i.e. @xmath382 . on the other hand , if there is a unique maximal interval @xmath378 starting at the first round of phase @xmath129 and such that @xmath383 , then it must hold that @xmath384 , where @xmath385 . combining these observations , we get that @xmath386 . by combining this bound with ( [ eq : xupperl ] ) , we have : @xmath387 \le pr[x\ge \delta\mu|h| ] < 2^{-\mu|h|}.\ ] ] thus , with probability @xmath388 , in at most @xmath389 fraction of rounds in @xmath129 there can be @xmath345 . recall that for node @xmath1 we need to show that @xmath390 for most rounds in a phase . we prove this by taking a _ constant size _ @xmath73-cover of @xmath119 and applying lemma [ le : phi ] to all nodes in the cover _ [ pr : lowc ] for every @xmath132 and @xmath147 , there are constants @xmath391 and @xmath392 , such that for every node @xmath1 and phase @xmath129 , with probability @xmath135 , at least @xmath136-fraction of rounds in @xmath129 are bounded contention rounds , i.e. @xmath393 . it will be sufficient to choose @xmath128 and @xmath394 such that @xmath395 and @xmath396 . by the definition of @xmath287 and bounded independence , there is a @xmath73-cover @xmath39 of @xmath119 of size @xmath397 . set @xmath398 . let us fix a node @xmath399 . if @xmath129 is the first phase or @xmath400 then we set @xmath401 . otherwise , consider the phase @xmath159 of @xmath127 rounds preceding @xmath129 . by lemma [ le : phi ] , if we set @xmath402 then , with probability @xmath403 , there is a round in @xmath159 where @xmath404 . let @xmath405 be such a round and set @xmath406 $ ] , where @xmath407 is the last round of @xmath129 . clearly , @xmath408 . now we can apply lemma [ le : phi ] for node @xmath20 with @xmath101 as above and conclude that with probability at least @xmath403 , there are at most @xmath409 rounds in @xmath378 ( note that @xmath410 ) where @xmath345 , so there are at most @xmath411 rounds in @xmath129 where @xmath345 . now we can apply the same argument for all nodes in @xmath412 simultaneously and conclude that with probability at least @xmath135 , there are at most @xmath413 rounds in @xmath129 where @xmath345 for _ every _ @xmath399 , if we take @xmath395 to be large enough . in the remaining @xmath414 rounds we have @xmath415 . we extract the following result from the proof of lemma [ le : phi ] , to use it later . the proof is similar to the proof of prop . [ pr : lowc ] . [ co : pexpectation ] for every @xmath335 and each node @xmath1 , the expected number of rounds in each phase @xmath129 where @xmath416 , is @xmath411 . let @xmath378 be the shortest time interval containing @xmath129 such that @xmath417 at the beginning of @xmath378 . let @xmath418 denote the event that @xmath408 . as in the proof of prop . [ pr : lowc ] , we have @xmath419 > 1-n^{-3}$ ] . let @xmath370 denote the number of rounds @xmath375 with @xmath80 . we know from the proof of lemma [ le : phi ] that @xmath420 = o(\phi^{-k})\cdot|\hat h| = o(\phi^{-k})\cdot |h|$ ] and @xmath367=e[x|\mathcal{e}]\cdot pr[\mathcal{e } ] + e[x|\bar{\mathcal{e}}]\cdot pr[\bar{\mathcal{e}}]= o(\phi^{-k})\cdot |h|$ ] . this completes the proof as @xmath421 . next we show that if @xmath76 is appropriately chosen then for each node @xmath1 , @xmath422 happens most of the time during each phase w.h.p . in order to show this , first we split the set of nodes in @xmath423 into local neighborhoods . we show that if in a given round the contention in each local neighborhood is bounded by an appropriate threshold , then the expected interference is small . for each local neighborhood , the expected number of rounds when the contention is higher than its threshold is bounded using corollary [ co : pexpectation ] . combining these results into one we get a bound on the expected number of rounds when @xmath424 is small . it then remains to apply a concentration bound to conclude the proof . [ pr : pinterferenceboundl ] for every @xmath132 and @xmath425 , there are constants @xmath426 and @xmath427 , such that for each node @xmath1 and phase @xmath129 , with probability @xmath135 , at least @xmath136-fraction of rounds in @xmath129 are low interference rounds , i.e. @xmath428 . let @xmath429 denote the in - ball of radius @xmath430 centered at node @xmath1 for @xmath431 . let @xmath432 be an @xmath73-cover of @xmath429 of size @xmath433 for a constant @xmath434 , which exists by bounded independence . let @xmath435 for @xmath436 , where @xmath437 is a large enough constant to be defined later . [ le : pexpectal ] if @xmath438 holds for all @xmath439 and @xmath440 and @xmath441 then @xmath442 . we have , by the definition of expected interference ( explanations below ) , @xmath443 where the first inequality follows from the definition of sets @xmath444 and the assumption of the claim , the second one follows by @xmath435 and @xmath445 , the third one is a rearrangement of the sum , the fourth one follows by the fact that @xmath446 for every @xmath447 and @xmath448 ( here , @xmath449 ) , and the last one follows because @xmath450 . thus , if @xmath441 then the claim follows . let us put the nodes in an arbitrary order and let @xmath451 denote the number of rounds @xmath452 s.t . @xmath453 , where @xmath454 is the @xmath455-th node in @xmath456 . by corollary [ co : pexpectation ] , @xmath457= o(\phi_i^{-k})\cdot|h| $ ] holds for each @xmath458 . let @xmath459 denote the number of rounds in @xmath129 when @xmath460 . by claim [ le : pexpectal ] , @xmath461 $ ] can be bounded as follows , @xmath462\leq\sum_{i}\sum_{j}e[z_{i , j } ] \leq \sum_i|s_i\setminus s_{i-1}|\cdot o(\phi_i^{-k})\cdot |h| \leq \frac{\sigma}{3}|h|,\ ] ] where the last inequality holds if @xmath463 for a large enough constant @xmath464 ; we have @xmath465 and @xmath466 with @xmath467 ( by definition ) , so we have @xmath468 . each @xmath451 can be considered as the sum of binary random variables similar to that in the proof of lemma [ le : phi ] . thus , @xmath459 can also be seen as the sum of binary random variables . moreover , the binary random variables satisfy the conditions of the chernoff bound in lemma [ le : chernoff ] . thus , by taking @xmath469 and @xmath470 , we have : @xmath471\leq 2^{-\sigma|h|/3}\le n^{-3}. $ ] [ le : broadgood ] let @xmath472 , @xmath473 and @xmath474 , where @xmath475 are the parameters of . let @xmath56 be a round such that for a node @xmath20 , @xmath476 and @xmath477 . + if a node in @xmath332 transmits in round @xmath56 then it mass - delivers , with probability at least @xmath478 . let @xmath1 be the transmitting node , which could possibly be @xmath20 itself . the probability that no other node in @xmath479 transmits is @xmath480 where we used lemma [ ieq ] with the assumption that @xmath481 and the bounded contention assumption . on the other hand , we have that @xmath482\ge 9/10 $ ] . indeed , @xmath483\ge 9/10 $ ] follows by the bounded interference assumption and markov inequality , so it suffices to show that @xmath484 , which we have because for each node @xmath485 , @xmath486 , implying that @xmath487 thus , we have that with probability @xmath478 , @xmath1 transmits , no other node in @xmath217 transmits and the interference from @xmath488 at @xmath1 is at most @xmath68 , so @xmath1 delivers its message , by . * proposition [ pr : pdecrease ] * _ assume that constants @xmath137 are large enough . for each node @xmath1 and phase @xmath129 , if at least @xmath138-fraction of the rounds of @xmath129 are of high contention , then @xmath139 nodes in @xmath140 mass - deliver , with probability @xmath135 . _ let @xmath489 . let us choose constants @xmath490 such that prop . [ pr : pgoodroundl ] holds for @xmath491 and @xmath492 and such that @xmath493 is large enough , as stated below , where @xmath475 are the parameters of . we will use the fact that in prop . [ pr : pgoodroundl ] , @xmath394 has the following dependency on @xmath494 : @xmath495 , where @xmath0 is the edge change parameter ( see the proof of prop . [ pr : lowc ] ) . we set @xmath496 . namely , if we choose @xmath137 large enough ( satisfying the relation above ) , then we have that with probability @xmath135 , there are at least @xmath497 rounds @xmath358 such that @xmath498 and @xmath499 . let @xmath56 be such a round . for each node @xmath500 , it holds that @xmath501 and @xmath502 . let us fix a node @xmath500 . first note that @xmath503 , implying that @xmath504 . it remains to bound the expected interference at @xmath20 by nodes in @xmath488 . this interference can be split into two parts : 1 . the interference by nodes in @xmath505 and 2 . the interference by nodes in @xmath506 . the first part can be bounded by @xmath507 using a computation as in the proof of lemma [ le : broadgood ] . the second part can be bounded as follows : the contention in the area @xmath506 is at most @xmath394 and the distance of those nodes to node @xmath20 is at least @xmath76 , so the expected interference by nodes in @xmath506 is at most @xmath508 , and is less than @xmath509 if @xmath76 is large enough , as @xmath510 . here we used the assumption that @xmath511 . recall that there are at least @xmath512 high contention rounds for node @xmath1 , i.e. rounds where @xmath122 . let @xmath56 be such a round . by the bounded independence property , there is a @xmath513-cover @xmath39 of @xmath119 of constant size . by the pigeonhole principle , there is a node @xmath399 such that @xmath514 . by summarizing the above said , we get that there are at least @xmath497 rounds @xmath56 s.t . for all nodes @xmath500 , @xmath515 and @xmath516 . on the other hand , there are at least @xmath512 rounds where there is a node in @xmath119 that has @xmath517 . thus , there are at least @xmath518 rounds @xmath56 with the following property : there is a node @xmath500 such that @xmath519 and @xmath516 . fix such round @xmath56 and node @xmath20 . by lemma [ le : broadgood ] , there is a node @xmath520 that transmits ( the probability of this is at least @xmath521 ) and delivers its message with probability @xmath522 . thus , the expected number of nodes in @xmath119 that deliver their message during @xmath129 is @xmath523 . the proof now follows by applying a chernoff bound with the assumption that the constant @xmath128 is large enough : @xmath524 . * proposition [ pr : idle ] * _ assume that @xmath137 are large enough . for each node @xmath1 and phase @xmath129 , if at least @xmath141-fraction of the rounds of @xmath129 are low contention rounds , then with probability @xmath135 , in at least @xmath142-fraction of the rounds of @xmath129 , @xmath1 will detect idle channel and have low contention and low interference . _ let @xmath525 . assume that @xmath137 large enough , such that prop . [ pr : pgoodroundl ] holds for @xmath526 and @xmath527 , where @xmath528 is the parameter of . let @xmath418 denote the event that there are at least @xmath529 good rounds for @xmath1 in phase @xmath129 . by prop . [ pr : pgoodroundl ] , we have @xmath419 = 1-o(n^{-3})\ge 19/20 $ ] , if @xmath128 is large enough . given @xmath418 , there are at least @xmath530 rounds @xmath56 that are both good and low contention for @xmath1 . let @xmath159 denote the set of such rounds and let @xmath350 be a binary random variable with value 1 if and only if @xmath1 detects idle channel in round @xmath161 . let @xmath531 . for each round @xmath161 , since @xmath56 is a good round and @xmath532 , it holds with probability at least @xmath141 that the interference at node @xmath1 from nodes in @xmath119 is at most @xmath528 . given this and the fact that in low contention rounds we have @xmath533 , @xmath1 will detect idle channel with probability at least @xmath534 in round @xmath56 . thus , in each round @xmath161 , irrespective of prior rounds , we have @xmath535\ge 81/100 $ ] . this implies : @xmath420\ge ( 9 ^ 2/10 ^ 2)|h'|\ge \frac{9 ^ 2}{10 ^ 2}\cdot \frac{9 ^ 2}{10 ^ 2}|h|$ ] and @xmath536 \ge ( 19/20)\cdot\frac{9 ^ 4}{10 ^ 4}|h| > 0.62|h|.\ ] ] just as in the proof of lemma [ le : phi ] , we can now apply chernoff bound to obtain that @xmath537 < n^{-3}$ ] if @xmath128 is large enough . this completes the proof by recalling that all rounds of @xmath370 are low contention and low interference rounds for node @xmath1 . * theorem [ thm : broadcast ] * _ assume the edge change rate @xmath100 is sufficiently small . there are constants @xmath201 , such that when running bcast(@xmath114 ) in the synchronous mode , each node @xmath1 receives the message in @xmath202 rounds w.h.p . _ let us fix constants @xmath155 so that props . [ pr : pdecrease ] and [ pr : idle ] hold with interference threshold @xmath203 , where @xmath204 and @xmath205 are the parameters corresponding to precision @xmath183 . we set the passiveness parameter of try&adjust to @xmath206 and the stable distance parameter @xmath207 . consider a node @xmath20 that receives the message in round @xmath56 . let @xmath1 be such that @xmath208 in the time interval @xmath209 $ ] ( and both stay alive during @xmath96 ) . the theorem follows by an induction and union bound from the claim below . node @xmath1 gets the message of @xmath20 during @xmath96 , with probability @xmath210 . we prove the claim by contradiction , and assume that @xmath1 can not get the message during @xmath96 . let us split @xmath96 into phases ( for @xmath20 ) and , similar to the proof of thm . [ thm : localbroadcast ] , classify the phases @xmath129 into types : ( type a ) at least @xmath138-th of rounds in @xmath129 are high contention rounds for @xmath20 @xmath211 , and ( type b ) at least @xmath141-th of rounds in @xmath129 are low contention rounds for @xmath20 @xmath212 . we show that with probability @xmath135 , all phases in @xmath96 are of type b. consider a type a phase @xmath129 . we know from prop . [ pr : pdecrease ] that , with probability @xmath135 , there is a set @xmath39 of @xmath139 nodes in @xmath140 that deliver their messages and restart try&adjust during @xmath129 . we split @xmath39 into three subsets , @xmath213 those inserted by churn during @xmath129 , @xmath214 those inserted by edge change during @xmath129 and @xmath215 the rest . consider @xmath214 first . we know that in each local neighborhood , there are at most @xmath216 nodes arriving due to edge changes . note that @xmath217 can be covered with @xmath218 local neighborhoods ; hence , the total number of nodes arriving in @xmath217 due to edge changes is @xmath219 . since @xmath220 , setting @xmath221 a small enough constant gives @xmath222 . now consider @xmath215 . by the primitive , for each node @xmath223 , all nodes in @xmath224 also restart try&adjust , setting their transmission probability to @xmath225 . by the setting of @xmath226 , the probability that a node restarting try&adjust in phase @xmath129 transmits again in @xmath129 is at most @xmath227 . thus , with probability at least @xmath135 , the nodes in @xmath214 constitute an @xmath228-packing of @xmath140 and are at most @xmath229 . as for @xmath213 , the probability that a newly arriving node transmits during phase @xmath129 is at most @xmath227 , so @xmath230 w.h.p . these bounds lead to a contradiction @xmath231 , if constant @xmath128 is large enough . thus , each phase @xmath129 of @xmath96 is of type b with probability @xmath135 . it remains to recall that @xmath96 contains constant number of phases . now assume that @xmath129 is of type b. in a similar way as in theorem [ thm : localbroadcast ] , we can apply prop . [ pr : idle ] and show that if @xmath232 at the beginning of @xmath129 , then by the end of @xmath129 , the effect on @xmath71 is equivalent to applying at least @xmath233 of @xmath234 operations on @xmath235 . since @xmath236 at the beginning of @xmath96 is at least @xmath225 , it will take at most 11 phases to increase the probability to the value @xmath164 . then , a single phase will suffice for @xmath20 to deliver its message ( including to node @xmath1 ) , with probability at least @xmath210 . the basic observation is that if all the nodes start running bcast@xmath240 simultaneously , a _ constant density dominating set _ can be computed in @xmath200 rounds ( cf . @xcite ) , where an @xmath33-dominating set of density @xmath538 is a set @xmath39 of nodes such that for each node @xmath20 , @xmath539 . the dominator algorithm is as follows : all nodes run bcast@xmath240 simultaneously , and 1 . if a node @xmath1 stops by then it is a _ dominator _ , if a node @xmath20 stops by detecting of node @xmath1 then it is dominated by @xmath1 . having formed a constant density dominating set , it remains to disseminate the message using only dominators . in order for the set of dominators to derive the connectivity properties of the original graph , we need that @xmath30 forms a @xmath540-bounded independence _ metric _ space . in the static spontaneous setting , there is a uniform algorithm that performs broadcast in @xmath541 rounds , w.h.p . [ thm : static - spont - bcast ] the algorithm for spontaneous broadcast consists of two stages : 1 . compute a constant density @xmath542-dominating set @xmath543 , 2 . transmit the message using only the nodes in @xmath543 . [ [ dominating - set . ] ] dominating set . + + + + + + + + + + + + + + + the dominating set is constructed by running bcast@xmath240 in the spontaneous mode , i.e. all nodes start running the algorithm simultaneously . let @xmath543 be the set of nodes that stop the algorithm by . since the dominated nodes use @xmath544 as a stopping condition , @xmath543 is a @xmath542-dominating set . moreover , @xmath543 is a @xmath545-packing , which implies that each in - ball @xmath46 of radius @xmath546 contains at most @xmath547 dominators . in particular each node @xmath20 is dominated by at most constant number of dominators . the proof the algorithm terminates after at most @xmath200 rounds is almost identical to the proof of theorem [ thm : broadcast ] . [ [ broadcast . ] ] broadcast . + + + + + + + + + + recall that we assume that the communication graph is connected and has diameter @xmath243 . also , since @xmath548 is a metric , the graph is undirected . the broadcast part is as follows : in the first round , the source node transmits the message to its neighbors ; each dominator @xmath20 , upon receiving the message , transmits it in each round with probability @xmath232 until detecting ( @xmath183 ) , where @xmath235 is a small enough constant . note that as soon as all dominators successfully transmit at least once , all nodes will get the message . moreover , if the constant @xmath235 is small enough , then the two algorithms can be run simultaneously : the key point is that the number of dominators in each @xmath549-neighborhood is bounded by a constant ( not depending on @xmath235 ) . this ensures that the nodes _ need not know @xmath16 _ , in order to coordinate the two algorithms . it remains to show that the message will get to all dominators in @xmath244 rounds using the broadcast algorithm . consider a graph @xmath129 defined over the dominating set @xmath543 , where for every pair of nodes @xmath550 , @xmath551 form an edge if @xmath552 . the diameter of @xmath129 is at most @xmath243 . note that for every pair of nodes @xmath553 with @xmath554 , the corresponding dominators @xmath555 and @xmath556 are adjacent in @xmath129 . indeed , by the definition of the dominating set , we have @xmath557 let @xmath558 be two arbitrary nodes in @xmath543 and let @xmath19 be the path of length at most @xmath243 in @xmath238 connecting nodes @xmath555 and @xmath556 . by the observation above , if we replace each node @xmath559 with its dominator @xmath560 , we obtain a path @xmath561 of length @xmath46 in @xmath129 , connecting @xmath555 and @xmath556 . the completes the proof of the claim . for every transmitting node @xmath562 , the probability that @xmath274 delivers its message to all its neighbors in @xmath129 is @xmath563 in each round @xmath56 , if constant @xmath235 is sufficiently small . let @xmath39 denote the set of neighbors of node @xmath274 in @xmath129 , i.e. the nodes of @xmath543 that are at distance at most @xmath564 from @xmath274 . it suffices to show that all nodes in @xmath39 receive the message in a fixed round @xmath56 where @xmath274 transmits with @xmath563 probability . this event holds if the interference at @xmath274 is no more than @xmath68 event @xmath565 , and no other node transmits in @xmath566 event @xmath567 , in round @xmath56 . by the properties of the dominating set , the contention in a ball of radius @xmath568 is at most @xmath569 . similarly to the proof of prop . [ pr : pinterferenceboundl ] , it can be shown that the expected interference by nodes in @xmath570 is @xmath571 . thus , by markov inequality , @xmath572 if @xmath235 is small enough . on the other hand , since the contention in @xmath566 is @xmath571 , @xmath573 follows by lemma [ ieq ] . thus , @xmath574 . given that each informed node in @xmath543 has probability at least @xmath575 of successfully broadcasting the message , the rest of the proof essentially follows along the lines of the proof of ( * ? ? ? * lemma 6 ) . we present a sketch of the proof for completeness of the argument . let @xmath576 be the set of nodes in @xmath543 that have been informed by round @xmath56 , where @xmath577 contains only the source node . let us fix a node @xmath578 and let @xmath579 be the distance ( in graph @xmath129 ) from @xmath1 to the nearest node in @xmath576 . note that @xmath580 . the difference @xmath581 is the progress made in round @xmath56 and is a bernoulli random variable with @xmath582\ge \eta$ ] for all @xmath56 when @xmath583 and @xmath584 otherwise . let @xmath585 be a random variable that has the same distribution as @xmath586 when @xmath583 and is an i.i.d . bernoulli random variable with mean @xmath587 otherwise . let @xmath588 . note that node @xmath1 has been informed by round @xmath56 iff @xmath589 . thus , we need to bound the probability @xmath590 . let @xmath591 . it is easy to show that the sequence @xmath592 is a submartingale . moreover , for a round @xmath593 and constant @xmath594 , @xmath595 implies that @xmath596 . since the sequence @xmath592 is a submartingale , we can apply azuma - hoeffding bound to show that @xmath597 . now the theorem follows by union bound over all nodes , by choosing the constant @xmath260 large enough . * theorem [ thm : brlowerbound ] * _ for every ( possibly randomized ) broadcast algorithm @xmath246 that uses neither node coordinates nor primitive , there is a @xmath247-bounded - independence metric space where @xmath246 needs @xmath248 rounds to do broadcast in a @xmath249-broadcastable network , even if the nodes have and primitives and operate spontaneously . _ assume that @xmath598 . we present the construction using a distance function @xmath548 : it is then straightforward to construct the corresponding path - loss matrix . denote @xmath599 . first , assume the nodes operate non - spontaneously , i.e. a non - source node may need to receive a message in order to start participating in a protocol . recall that @xmath600 . consider @xmath16 points @xmath601 , such that for every @xmath602 , it holds that @xmath603 , @xmath604 , @xmath605 and @xmath606 , where @xmath607 ( see the diagram in fig . [ fig : nonspon ] ) . clearly , this set of points forms a @xmath608-bounded independence metric space . we place @xmath16 wireless nodes at distinct points @xmath601 uniformly at random . let @xmath195 be the node at @xmath609 . we assume , further , that communication only happens according to ; in particular , if the interference at a node is more than @xmath68 then _ none _ of its neighbors receives its transmission . note that broadcast in this network can be completed in 2 steps , starting at any point . note also that @xmath610 can not be directly reached from nodes @xmath195 with @xmath611 . moreover , if at least @xmath612 nodes @xmath195 with @xmath611 transmit simultaneously , _ no node _ receives a message , including the potential communication between @xmath613 and @xmath610 . this follows by observing that @xmath614 : the signal power at the neighboring nodes must be more than the interference threshold , otherwise a node could receive two signals of the same power . assume @xmath195 with some @xmath611 is the source . at the first round of the algorithm , nodes @xmath615 receive the message . let @xmath616 denote the event that @xmath613 transmits and no more than @xmath617 nodes @xmath195 with @xmath611 transmit in round @xmath56 . by the observations above , @xmath610 will not receive the message until @xmath616 happens . moreover , during the subsequent steps of the algorithm , until @xmath616 happens for the first time , all nodes at @xmath615 will have the same history and will be symmetric with respect to and primitives . it remains to bound the expected time until @xmath616 happens for the first time . we can assume w.l.o.g . that in each round there are no more than @xmath187 nodes transmitting . as discussed above , if @xmath610 is not informed in such a round , then the nodes that transmitted in the given round can learn at best that none of them is at @xmath613 and stop transmitting thereafter . thus , we can assume that after each unsuccessful round there are at most @xmath618 nodes that stop transmitting and the probability @xmath619 of success at round @xmath620 ( if no success occurred before round @xmath33 ) is at most @xmath621 where the numerator counts the number subsets of an @xmath622-element set that contain a fixed element ( the node @xmath613 ) and no more than @xmath187 elements in total , and the denominator counts all subsets containing no more than @xmath187 elements . a further straightforward calculation shows that the expected number of rounds until the first success is @xmath248 ( see e.g. ( * ? ? ? Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: an _ orbitope _ is the convex hull of an orbit of a compact algebraic group @xmath2 acting linearly on a real vector space . the orbit has the structure of a real algebraic variety , and the orbitope is a convex , semi - algebraic set . thus , the study of algebraic orbitopes lies at the heart of _ convex algebraic geometry _ the fusion of convex geometry and ( real ) algebraic geometry . orbitopes have appeared in many contexts in mathematics and its applications . orbitopes of finite groups are highly symmetric convex polytopes which include the platonic solids , permutahedra , birkhoff polytopes , and other favorites from ziegler s text book @xcite , as well as the coxeter orbihedra studied by mccarthy , ogilvie , zobin , and zobin @xcite . farran and robertson s regular convex bodies @xcite are orbitopal generalzations of regular polytopes , which were classified by madden and robertson @xcite . orbitopes for compact lie groups , such as @xmath0 , have appeared in investigations ranging from protein structure prediction @xcite and quantum information @xcite to calibrated geometries @xcite . barvinok and blekherman studied the volumes of convex bodies dual to certain @xmath0-orbitopes , and they concluded that there are many more non - negative polynomials than sums of squares @xcite . this paper initiates the study of orbitopes as geometric objects in their own right . the questions we ask about orbitopes originate from three different perspectives : _ convexity _ , _ algebraic geometry _ , and _ optimization_. in convexity , one would seek to characterize all faces of an orbitope . in algebraic geometry , one would examine the zariski closure of its boundary and identify the components and singularities of that hypersurface . in optimization , one would ask whether the orbitope is a spectrahedron or the projection of a spectrahedron . spectrahedra are to semidefinite programming what polyhedra are to linear programming . more precisely , a _ spectrahedron _ is the intersection of the cone of positive semidefinite matrices with an affine space . it can be represented as the set of points @xmath3 such that @xmath4 where @xmath5 are symmetric matrices and @xmath6 denotes positive semidefiniteness . from a spectrahedral description many geometric properties , both convex and algebraic , are within reach . furthermore , if an orbitope admits a representation ( [ eq : lmi ] ) then it is easy to maximize or minimize a linear function over that orbitope . here is a simple illustration . [ ex : hankelsmall ] consider the action of the group @xmath7 on the space @xmath8 of binary quartics and take the convex hull of the orbit of @xmath9 . the four - dimensional convex body @xmath10 is a _ carathodory orbitope_. this orbitope is a spectrahedron : it coincides with the set of all binary quartics @xmath11 such that @xmath12 this representation ( [ 3by3hankel ] ) will be derived in section 5 , where we will also see that it is equivalent to classical results from the theory of positive polynomials @xcite . the hankel matrix shows that the boundary of @xmath10 is an irreducible cubic hypersurface in @xmath13 , defined by the vanishing of the hankel determinant . it also reveals that this four - dimensional carathodory orbitope is @xmath14-neighborly : the extreme points are the rank one matrices , and any two of them are connected by an edge . the typical intersection of @xmath10 with a three - dimensional affine plane looks like an inflated tetrahedron . this three - dimensional convex body is bounded by _ cayley s cubic surface _ , shown in figure [ f : fat_tetrahedron ] . alternative pictures of this convex body can be found in ( * ? ? ? * fig . 3 ) and ( * ? ? ? * fig . 4 ) . the four vertices of the tetrahedron lie on the curve @xmath15 , and its six edges are inclusion - maximal faces of @xmath10 . @xmath16{figures / free - spec}\ ] ] -0.3 cm [ f : fat_tetrahedron ] this article is organized as follows . we begin by deriving the basic definitions and a few general results about orbitopes , and we formulate * ten key questions * which will guide our subsequent investigations . these are organized along the themes of convex geometry ( subsection 2.1 ) , algebraic geometry ( subsection 2.2 ) and optimization ( subsection 2.3 ) . these questions are difficult . they are meant to motivate the reader and to guide our investigations . we are not yet able to offer a complete solution to any of these ten problems . section 3 is concerned with the action of @xmath1 and @xmath0 by conjugation on @xmath17-matrices . these decompose into the actions by conjugation on symmetric and skew - symmetric matrices . the resulting _ schur - horn orbitopes _ are shown be spectrahedra , their algebraic boundary is computed , and their face lattices are derived from certain polytopes known as permutahedra . the spectrahedral representations of the schur - horn orbitopes , stated in theorems [ thm : symspec ] and [ thm : skewsymspec ] , rely on certain schur functors also known as the _ additive compound matrices_. given a compact real algebraic group of @xmath17-matrices , we can form its convex hull in @xmath18 . the resulting convex bodies are the _ tautological orbitopes_. in subsection 4.2 we study the tautological orbitope and its dual coorbitope for the group @xmath1 . both are spectrahedra , characterized by constraints on singular values , and they are unit balls for the operator and nuclear matrix norm considered in @xcite . subsections 4.1 and 4.4 are devoted to the tautological orbitope for @xmath0 . a characterization of its faces is given in theorem [ thm : sontauto ] . the orbitopes of the group @xmath19 are the convex hulls of trigonometric curves , a classical topic initiated by carathodory in @xcite , further developed in @xcite , and our primary focus in section 5 . these @xmath19-orbitopes can be represented as projections of spectrahedra in two distinct ways : in terms of hermitian toeplitz matrices or in terms of hankel matrices . a natural generalization of the rational normal curves in section 5 are the veronese varieties . their convex hulls , the _ veronese orbitopes _ , are dual to the cones of polynomials that are non - negative on @xmath20 , as seen in @xcite . in section 6 we undertake a detailed study of the @xmath21-dimensional veronese orbitope and its coorbitope which arise from ternary quartics . in section 7 we mesh our investigations with a line of research in differential geometry . the _ grassmann orbitope _ is the convex hull of the oriented grassmann variety in its plcker embedding in the unit sphere in @xmath22 . this vector space is the @xmath23-th exterior power of @xmath24 and it is isomorphic to @xmath25 . the facial structure of grassmann orbitopes has been studied in the theory of calibrated manifolds @xcite . we take a fresh look at these orbitopes from the point of view of convex algebraic geometry . theorem [ thm : g2n ] furnishes a spectrahedral representation and the algebraic boundary in the special case @xmath26 , while theorem [ thm : g36bad ] shows that the grassmann orbitopes fail to be spectrahedra in general . let @xmath2 be a real , compact , linear algebraic group , that is , a compact subgroup of @xmath27 for some @xmath28 given as a subvariety . prototypic examples are the _ special orthogonal group _ @xmath29 and the _ unitary group _ @xmath30 . a real representation of @xmath2 is a group homomorphism @xmath31 for some finite dimensional real vector space @xmath32 . we will write @xmath33 for @xmath34 and @xmath35 . the representation is _ rational _ if @xmath36 is a rational map of algebraic varieties . by choosing an inner product @xmath37 on @xmath32 , we may define a @xmath2-invariant inner product as @xmath38 here @xmath39 denotes the _ haar measure _ , which is the unique @xmath2-invariant probability measure on @xmath2 . choosing coordinates on @xmath32 so that @xmath40 becomes the standard inner product on @xmath41 , we can identify the matrix group @xmath2 with a subgroup of the _ orthogonal group _ @xmath42 . the _ orbit _ of a vector @xmath35 under the compact group @xmath2 is the set @xmath43 . this is a bounded subset of @xmath32 . the orbit @xmath44 is a smooth , compact real algebraic variety of dimension @xmath45 here @xmath46 is the _ stabilizer _ of the vector @xmath47 . in particular , the orbit is isomorphic as a @xmath2-variety to the compact homogeneous space @xmath48 . the _ orbitope _ of @xmath2 with respect to the vector @xmath35 is the semialgebraic convex body @xmath49 we tacitly assume that the group @xmath2 and its representation @xmath36 are clear from the context and sometimes we write @xmath50 or @xmath51 for @xmath52 . the dimension of an orbitope @xmath53 is the dimension of the affine hull of the orbit @xmath54 . for small @xmath55 , the connected subgroups of the orthogonal group @xmath1 are known . this leads to the following census of low - dimensional orbitopes of connected groups . @xmath56 we identify @xmath2 with a subgroup of @xmath0 . we assume @xmath57 is an @xmath55-dimensional orbitope in @xmath24 . this implies that @xmath2 fixes no non - zero vector . here is our census : _ @xmath58 : _ there are no one - dimensional orbitopes because @xmath59 is a point . it is known that every proper connected subgroup of @xmath19 or @xmath60 fixes a non - zero vector , so for @xmath61 , the subgroup @xmath2 must be equal to @xmath0 . this establishes the next two cases : _ @xmath62 : _ the only orbitopes in @xmath63 are the discs @xmath64 . _ @xmath65 : _ the only orbitopes in @xmath66 are the balls @xmath67 . the four - dimensional case is where things begin to get interesting : _ @xmath68 : _ the group @xmath69 has connected subgroups @xmath2 of dimension @xmath70 , @xmath14 , @xmath71 and @xmath72 . * if @xmath73 then @xmath74 and the orbitopes are balls in @xmath13 . * if @xmath75 then @xmath76 , acting as the unit quaternions on all quaternions , @xmath77 , by either left or right multiplication . here the orbitopes are also balls in @xmath13 . * if @xmath78 then @xmath79 . these tori act on @xmath13 through an orthogonal direct sum decomposition @xmath80 and their orbitopes are products of two discs . * if @xmath81 then @xmath82 and we obtain four - dimensional orbitopes that are isomorphic , for some positive integers @xmath83 , to the carathodory orbitopes @xmath84\bigr\}\,.\ ] ] these orbitopes were introduced one century ago by carathodory @xcite . their study was picked up in the 1980s by smilansky @xcite and recently by barvinok and novik @xcite . we note that @xmath85 is affinely isomorphic to the hankel orbitope in example [ ex : hankelsmall ] . to compute the dimensions of orbitopes in general we shall need a pinch of representation theory @xcite . a representation @xmath32 of the group @xmath2 is _ irreducible _ if its only subrepresentations are @xmath86 and @xmath32 . if @xmath32 and @xmath87 are irreducible representations , then the space @xmath88 of equivariant linear maps between them is zero unless @xmath89 . schur s lemma states that @xmath90 is a division algebra over @xmath91 , that is , either @xmath91 , @xmath92 , or @xmath93 . if @xmath94 is a complete list of distinct irreducible representations of @xmath2 , and @xmath32 is any representation of @xmath2 , then we have a canonical decomposition into isotypical representations : @xmath95 this is an isomorphism of @xmath2-modules . the map ( [ eq : isotypical ] ) on each summand is @xmath96 . the image of the @xmath97th summand in @xmath32 is called the _ @xmath98-isotypical component of @xmath32 _ , and when it is non - zero , we say that the irreducible representation @xmath98 _ appears _ in the @xmath2-module @xmath32 . we say that @xmath32 is _ multiplicity - free _ if each irreducible representation @xmath98 appears in @xmath32 at most once , so that @xmath99 has rank 1 or 0 over @xmath100 . suppose that @xmath32 contains the trivial representation and write @xmath101 , where @xmath102 is the trivial isotypical component of @xmath32 and @xmath103 does not contain the trivial representation . any vector @xmath104 can be written as @xmath105 , where @xmath106 and @xmath107 . thus we lose no geometric information in assuming that @xmath32 does not contain the trivial representation . this property ensures that the linear span of an orbit coincides with its affine span . hence the affine span of an orbitope decomposes along its isotypical components : @xmath109 to determine the dimension of @xmath110 for @xmath47 in a single isotypical component @xmath32 we proceed as follows . let @xmath111 with @xmath87 irreducible . then @xmath112 and the affine span of @xmath54 is isomorphic to @xmath113 , where @xmath114 is the rank of the @xmath115-module spanned by @xmath116 . hence , the dimension of @xmath117 over @xmath91 equals @xmath118 . in particular , @xmath119 . if @xmath32 is multiplicity free and @xmath47 has a nonzero projection into each isotypical component of @xmath32 , then @xmath120 . we see this in the carathodory orbitopes for the group @xmath19 of @xmath121 rotation matrices . its nontrivial representations are @xmath122 , where a rotation matric acts through its @xmath123th power , for @xmath124 . let @xmath125 be the orbitope of @xmath19 with respect to a general vector @xmath126 . if @xmath127 , then @xmath32 has two isotypical components and @xmath125 has dimension four . if @xmath128 , then @xmath129 consists of a single isotypical component and @xmath130 is two - dimensional , as @xmath131 , the span of @xmath132 is complex one - dimensional . orbitopes are convex bodies , and it is natural to begin their study from the perspective of classical convexity . a point @xmath133 in a convex body @xmath134 is an _ extreme point _ if @xmath135 . thus , the set @xmath136 of extreme points of @xmath137 is the minimal subset satisfying @xmath138 . an extrinsic description of @xmath137 is given by its _ support function _ @xmath139 in terms of the support function , the convex body @xmath137 is the set of points @xmath140 such that @xmath141 for every @xmath142 . each linear functional @xmath142 defines an _ exposed face _ of @xmath137 : @xmath143 an exposed face @xmath144 is itself a convex body of dimension @xmath145 . an exposed face of dimension @xmath146 is called an _ exposed point _ of @xmath137 . it follows that every exposed point is extreme , but the inclusion is typically strict . however , for orbitopes , these two notions coincide . every point in the orbit @xmath147 is exposed in its convex hull . in particular , every extreme point of the orbitope @xmath148 is an exposed point . since @xmath2 acts orthogonally on @xmath32 , the orbit @xmath147 lies entirely in the sphere in @xmath32 that is centered at @xmath146 and contains the point @xmath47 . as every point of the sphere is exposed , the entire orbit consists of exposed points and hence extreme points . a closed subset @xmath149 is a _ face _ if @xmath150 contains the two endpoints to any open segment in @xmath137 it intersects . this includes @xmath151 and @xmath137 itself . an inclusion - maximal proper face of @xmath137 is called a _ facet_. by separation , every face is contained in an exposed face and thus facets are automatically exposed . in general , every exposed face is a face but not conversely . when does an orbitope have only exposed faces ? the exposed faces of a convex body form a partially ordered set with respect to inclusion , called the _ face lattice_. the face lattice is atomic but in general not coatomic as was pointed out to us by stephan weis . a necessary condition is that the polar body ( see below ) has all faces exposed ( cf . @xcite ) . also , it is generally not graded because `` being a face of '' is not a transitive relation . for example , the four - dimensional barvinok - novik orbitope in section [ s : caratheodory ] has triangular exposed faces for which the three vertices are exposed but the edges are not . similar behavior is seen in the convex body on the right of figure [ f : trig_curves ] , which has two triangular exposed facets whose edges and two of three vertices are not exposed . [ question2 ] describe the face lattices of orbitopes . for an orbitope @xmath152 , the faces come in @xmath2-orbits and these @xmath2-orbits come in algebraic families . in particular , the zero - dimensional faces come in a family parametrized by @xmath2 . the description of these families is the point of question [ question2 ] . for instance , the orbitope in example [ ex : hankelsmall ] is a four - dimensional , two - neighborly convex body . its exposed points are parametrized by the circle @xmath153 and the edges come in a two - dimensional family . the _ polar body _ @xmath154 is called the _ coorbitope _ of @xmath2 with respect to @xmath104 . our assumption that @xmath32 does not contain the trivial representation ensures that @xmath146 is the centroid of @xmath51 , and this implies @xmath155 . we shall also make use of the cone over the coorbitope @xmath156 . this is the _ coorbitope cone _ @xmath157 for a convex body @xmath137 the assignment @xmath158 defines an _ ( asymmetric ) norm _ on @xmath32 with unit ball @xmath137 . if @xmath137 is centrally - symmetric with respect to the origin , then @xmath159 is an actual norm . in that case the polar body @xmath160 is also centrally - symmetric and @xmath161 is the _ dual norm_. norms and support functions are related by @xmath162 in particular , if @xmath137 is an orbitope , then @xmath159 and @xmath161 are @xmath2-equivariant norms . every point @xmath133 in a convex body @xmath137 is in the convex hull of finitely many extreme points . we denote by @xmath163 the least cardinality of a set @xmath136 of extreme points with @xmath164 . we call @xmath165 the _ carathodory number _ of @xmath137 . carathodory s theorem ( see e.g. @xcite ) asserts that @xmath166 is bounded from above by @xmath167 . fenchel showed that @xmath168 when the set of extreme points of @xmath137 is connected @xcite . note that the carathodory number of an orbitope @xmath169 in general depends on @xmath47 ( cf . * theorem 4.9 ) ) whereas , for multiplicity - free representations , the dimension of @xmath50 does not . what are the carathodory numbers of orbitopes ? we refer to the recent work of barvinok and blekherman @xcite for more information about the convex geometry of orbitopes and coorbitopes , especially with regard to their volumes . here we look at orbitopes as objects in real algebraic geometry . fix a rational representation @xmath170 of a compact connected algebraic group @xmath2 . every orbit @xmath44 is an irreducible real algebraic variety in @xmath171 , and we may ask for its prime ideal . by the tarski - seidenberg theorem @xcite , the orbitope is a semi - algebraic set . which orbitopes are _ basic _ semi - algebraic sets , i.e. for which triples @xmath172 can @xmath10 be described by a finite conjunction of polynomial equations and inequalities ? the boundary @xmath173 of an orbitope @xmath51 in @xmath171 is a compact semi - algebraic set of codimension one in its affine span @xmath174 . the zariski closure of @xmath173 is denoted by @xmath175 . we call it the _ algebraic boundary _ of @xmath51 . if @xmath176 then the algebraic boundary @xmath175 is the zero set of a unique ( up to scaling ) reduced polynomial @xmath177 whose coefficients lie in the field of definition of @xmath172 . that field of definition will often be the rational numbers @xmath178 . since scalars in @xmath179 have an exact representation in computer algebra , but scalars in @xmath91 require numerical floating point approximations , we seek to use @xmath179 instead of @xmath91 wherever possible . [ ques : algbound ] how to calculate the algebraic boundary @xmath180 of an orbitope @xmath181 ? the irreducible factors of the polynomial @xmath177 that cuts out @xmath182 arise from various singularities in the boundary @xmath183 of the coorbitope @xmath184 . we believe that a complete answer to question [ ques : algbound ] will involve a whitney stratification of the real algebraic hypersurface @xmath185 . recall that a _ whitney stratification _ is a decomposition into locally closed submanifolds ( strata ) in which the singularity type of each stratum is locally constant along the stratum . the faces of a polytope form a whitney stratification of its boundary , which is dual to the stratification of the polar polytope . we expect a similar duality between the whitney stratification of the boundary of an orbitope and of the boundary of its coorbitope . how to compute and study the algebraic boundary @xmath186 of the coorbitope @xmath187 ? is every component of @xmath188 the dual variety to a stratum in a whitney stratification of @xmath185 ? recall that the _ dual variety _ @xmath189 of a subvariety @xmath190 in @xmath171 is the zariski closure of the set of all affine hyperplanes that are tangent to @xmath190 at some regular point . when studying this duality , algebraic geometers usually work in complex projective space @xmath191 rather than real affine space @xmath192 . in some of the examples for @xmath193 seen in this paper , the algebraic boundary @xmath185 of the coorbitope @xmath184 coincides with the dual variety @xmath189 of the orbit @xmath194 . a good example for this is the discriminantal hypersurface in corollary [ cor : discrorbi ] . more generally , we have the impression that the hypersurface @xmath185 is often irreducible while @xmath182 tends to be reducible . for further appearances of dual varieties in convex algebraic geometry see @xcite . @xmath114-th secant variety _ of @xmath15 is the zariski closure of all @xmath195-flats spanned by points of @xmath54 . the study of secant varieties leads to lower bounds for the carathodory number : [ prop : secant ] if @xmath196 then the @xmath114-th secant variety of @xmath15 is the ambient space @xmath171 . let @xmath197 . the set of points that lie in the convex hull of @xmath198 points of @xmath15 is dense in @xmath50 and hence is zariski dense in @xmath171 . the @xmath114-th secant variety contains that set . the lower bound for @xmath199 from proposition [ prop : secant ] usually does not match fenchel s upper bound @xmath200 . for instance , consider the carathodory orbitope @xmath50 in example [ ex : hankelsmall ] . its algebraic boundary @xmath201 equals the second secant variety of the orbit @xmath15 , so proposition [ prop : secant ] implies @xmath202 . this orbitope satisfies @xmath203 but @xmath204 . for which orbitopes @xmath205 is the boundary @xmath206 one of the secant varieties of @xmath15 ? when is the lower bound on the carathodory number @xmath207 in proposition [ prop : secant ] tight ? a fundamental object in convex optimization is the set @xmath208 of positive semidefinite symmetric real @xmath17-matrices . this is the closed basic semi - algebraic cone defined by the non - negativity of the @xmath209 principal minors . it can also be described by only @xmath55 polynomial inequalities , namely , by requiring that the elementary symmetric polynomials in the eigenvalues , i.e. the suitably normalized coefficients of the characteristic polynomial , be non - negative . the algebraic boundary @xmath210 of the cone @xmath208 is the symmetric @xmath17-determinant . all faces of @xmath208 are exposed , isomorphic to @xmath211 for @xmath212 , and indexed by the lattice of linear subspaces ordered by reverse inclusion . spectrahedra inherit these favorable properties . recall that a _ spectrahedron _ is the intersection of the cone @xmath208 with an affine - linear subspace in @xmath213 . if we know that an orbitope is a spectrahedron then this either answers or simplifies many of our questions . [ ques : isspec1 ] characterize all @xmath214-orbitopes that are spectrahedra . polytopes are special cases of spectrahedra : they arise when the affine - linear space consists of diagonal matrices . one major distinction between polytopes and spectrahedra is that the class of spectrahedra is not closed under projection . that is , the image of a spectrahedron under a linear map is typically not a spectrahedron . see section 5 for orbitopal examples . characterizing projections of spectrahedra among all convex bodies is a major open problem in optimization theory ; see e.g. @xcite . here is a special case of this general problem : [ ques : isspec2 ] is every orbitope the linear projection of a spectrahedron ? in questions [ ques : isspec1 ] and [ ques : isspec2 ] , it is important to keep track of the subfield of @xmath91 over which the data @xmath172 are defined . frequently , this subfield is the rational numbers @xmath178 , and in this case we seek to write the orbitope as a ( projected ) spectrahedron over @xmath178 and not just over @xmath91 . semidefinite programming is the problem of maximizing a linear function over a ( projected ) spectrahedron , and there are efficient numerical algorithms for solving this problem in practice . in our context of orbitopes , the optimization problem can be phrased as follows : what method can be used for maximizing a linear function @xmath215 over an orbitope @xmath51 ? equivalently , how to evaluate the norm @xmath216 associated with the coorbitope @xmath156 ? this is equivalent to a non - linear optimization problem over the compact group @xmath2 . we seek to find @xmath34 which maximizes @xmath217 . this maximum is an algebraic function of @xmath47 . in this section we study two families of orbitopes for the orthogonal group @xmath218 . this group acts on the lie algebra @xmath219 by restricting the adjoint representation of @xmath27 . the @xmath1-module @xmath220 decomposes into two distinguished invariant subspaces , namely @xmath221 and @xmath222 . these correspond to the normal and tangent space of @xmath223 at the identity . in matrix terms , the spaces of symmetric and skew - symmetric matrices form two natural representations of @xmath1 for the action @xmath224 with @xmath225 and @xmath226 . for a symmetric matrix @xmath227 we define the _ symmetric schur - horn orbitope _ @xmath228 we shall see that these orbitopes are intimately related to certain polytopes which govern their boundary structure and spectrahedral representation . this connection arises via the classical schur - horn theorem @xcite . the material in the sections below could also be presented in symplectic language , using the moment maps of atiyah - guillemin - sternberg @xcite . the @xmath229-dimensional space @xmath221 decomposes into the trivial @xmath1-representation , given by multiples of the identity matrix , and the irreducible representation of symmetric @xmath17-matrices with trace zero . every symmetric matrix @xmath227 is orthogonally diagonalizable over @xmath91 . the ordered list of eigenvalues of @xmath230 is denoted @xmath231 . the orbit @xmath232 equals the set of matrices @xmath233 that satisfy @xmath234 . we shall see that its convex hull @xmath235 is the set of matrices @xmath233 for which @xmath236 is majorized by @xmath237 . for @xmath238 we say that @xmath239 is _ majorized _ by @xmath133 , written @xmath240 , if @xmath241 , and , after reordering , @xmath242 and @xmath243 , we have @xmath244 for a fixed point @xmath245 , the set of all points @xmath239 majorized by @xmath133 is the convex polytope @xmath246 here @xmath247 denotes the symmetric group , and @xmath248 is the _ permutahedron _ with respect to @xmath133 . this is a well - studied polytope @xcite and is itself an orbitope for @xmath247 . the permutahedron @xmath248 for @xmath249 consists of all points @xmath250 such that @xmath251 and @xmath252$.}\ ] ] let @xmath253 be the linear projection onto the diagonal . let @xmath227 and @xmath254 its symmetric schur - horn orbitope . then the diagonal @xmath255 is majorized by the vector of eigenvalues @xmath237 . in fact , the orbitope @xmath254 maps linearly onto the permutahedron : @xmath256 [ cor : sh1 ] the schur - horn orbitope equals @xmath257 . we have shown that the right hand side equals @xmath258 . we claim that a matrix @xmath259 is in this set if and only if @xmath259 lies in @xmath260 . this is clear if @xmath259 is a diagonal matrix . it follows for all matrices since both sets are invariant under the @xmath1-action . our next goal is to derive a spectrahedral characterization of @xmath254 . consider the natural action of the lie group @xmath27 on the @xmath114-th exterior power @xmath261 . if @xmath262 is any basis of @xmath24 , then the induced basis vectors of @xmath261 are @xmath263 for @xmath264 . a matrix @xmath265 acts on a basis element by sending it to @xmath266 . we denote by @xmath267 the induced map of lie algebras . the linear map @xmath268 is defined by the rule @xmath269 the @xmath270-matrix that represents @xmath271 in the standard basis of @xmath261 is known as the _ @xmath114-th additive compound matrix _ of the @xmath272-matrix @xmath273 . it has the following main property : [ lem : compoundeigen ] let @xmath274 with eigenvalues @xmath275 . then @xmath271 is symmetric and has eigenvalues @xmath276 for @xmath264 . let @xmath277 be an eigenbasis for @xmath273 . then the formula ( [ eq : addcomprule ] ) says @xmath278 hence @xmath279 is an eigenvector of @xmath271 with eigenvalue @xmath280 . this leads to the result that each symmetric schur - horn orbitope @xmath254 is a spectrahedron . [ thm : symspec ] let @xmath227 with ordered eigenvalues @xmath281 . then @xmath282 a matrix @xmath233 is in @xmath254 if and only if @xmath236 is in the permutahedron @xmath283 . from the inequality representation in ( [ eq : permineq ] ) , in conjunction with lemma [ lem : compoundeigen ] , we see that this is the case if and only if the largest eigenvalue of @xmath284 is at most @xmath285 . we shall now derive the description of all faces of the schur - horn orbitope @xmath254 . since @xmath254 is a spectrahedron , by theorem [ thm : symspec ] , we know that all of its faces are exposed faces . hence a face of @xmath254 is the set of points maximizing a linear function @xmath286 . the canonical @xmath1-invariant inner product on @xmath221 is given by @xmath287 and , by identifying spaces , a linear function may be written as @xmath288 . note that the dual space @xmath289 is equipped with the contragredient action , that is , @xmath290 . [ thm : facelatticesh ] every @xmath1-orbit of faces of @xmath254 intersects the pullback of a unique @xmath247-orbit of faces of the permutahedron @xmath283 . in particular , the faces of @xmath254 are products of symmetric schur - horn orbitopes of smaller dimensions corresponding to flags in @xmath24 . let @xmath150 be a face of @xmath254 and let @xmath291 be a linear function maximized at @xmath150 . then the face @xmath292 is given by @xmath293 and we may identify @xmath294 with the orbit @xmath295 . since @xmath273 is orthogonally diagonalizable , the orbit @xmath295 contains the diagonal matrices @xmath296 for @xmath297 . the corresponding faces are the pullbacks of the orbit of faces of @xmath283 given by the linear function @xmath298 . faces of the permutahedron correspond to flags of coordinate subspaces , and @xmath1 translates these to arbitrary flags of subspaces . facets of the permutahedron @xmath248 correspond to coordinate subspaces . replacing these with arbitrary subspaces yields supporting hyperplanes for the schur - horn orbitope @xmath254 . [ cor : sh2 ] the schur - horn orbitope @xmath299 is the set of matrices @xmath233 such that @xmath300 the following three examples show how theorem [ thm : facelatticesh ] translates into explicit face lattices . [ ex : freespec ] let @xmath301 be the diagonal matrix with diagonal @xmath302 . the orbitope @xmath260 is the convex hull of all symmetric rank @xmath70 matrices with trace @xmath70 , and hence @xmath303 . this orbitope plays the role of a `` simplex among spectrahedra '' because every compact spectrahedron is an affine section . the face @xmath304 of the orbitope @xmath254 in direction @xmath274 is isomorphic to @xmath305 where @xmath306 is the unit sphere and @xmath307 is the eigenspace of @xmath273 with maximal eigenvalue . thus , @xmath308 is isomorphic to a lower dimensional schur - horn orbitope for a rank one matrix . we conclude that the face lattice of @xmath254 consists of the linear subspaces of @xmath24 ordered by inclusion . this fact is well known ; see @xcite . the dimension of a face corresponding to a @xmath114-subspace is @xmath309 . the projection @xmath310 is the standard @xmath311-dimensional simplex @xmath312 whose faces correspond to the coordinate subspaces . we now describe continuous analogs to hypersimplices , extending the simplices in example [ ex : freespec ] . fix @xmath313 and let @xmath314 be the diagonal matrix with @xmath114 ones and @xmath315 zeros . the orbitope @xmath316 of the _ @xmath317-spectrahedral hypersimplex _ , as its diagonal projection @xmath318 is the classical @xmath317-hypersimplex ; cf . * example 0.11 ) . for instance , if @xmath68 and @xmath319 then @xmath299 is nine - dimensional and @xmath320 is an octahedron . up to @xmath321-symmetry , the octahedron has one orbit of vertices and edges but two orbits of triangles . the pullback of any edge is a circle , and the pullbacks of the triangles are five - dimensional symmetric schur - horn orbitopes @xmath322 for @xmath323 and @xmath324 . both facets are isomorphic to free spectrahedra . let @xmath227 be a symmetric matrix with distinct eigenvalues , e.g. @xmath325 . the image of @xmath254 under the diagonal map is the classical permutahedron @xmath326 . its face lattice may be described as the collection of all flags of coordinate subspaces in @xmath24 ordered by refinement . we may associate to every @xmath274 the complete flag whose @xmath114-th subspace is the direct sum of the eigenspaces of the first @xmath114 largest eigenvalues . thus , @xmath327 may be identified with the complete flag variety over @xmath91 . as for the facial structure , the face @xmath308 is isomorphic to the convex hull of the orbit @xmath328 . here , the stabilizer decomposes into a product of groups isomorphic to @xmath329 where @xmath330 is the dimension of the @xmath97-th eigenspace of @xmath273 . hence , the face @xmath308 is isomorphic to a cartesian product of generic schur - horn orbitopes and is of dimension @xmath331 . the face only depends on the flag associated to @xmath273 . this implies that the face lattice of @xmath299 is isomorphic to the set of partial flags ordered by refinement . again , in every orbit of flags there is a flag consisting only of coordinate subspaces . these special flags form the face lattice of the standard permutahedron @xmath332 . we regard @xmath299 as a continuous analog of the permutahedron . we conclude this subsection with a discussion of the algebraic boundary @xmath333 of the schur - horn orbitope . let @xmath334 be the smallest subfield of @xmath91 that contains the eigenvalues @xmath335 , and suppose that the @xmath336 are sufficiently general . then the hypersurface @xmath333 is defined in the affine space @xmath337 by the following polynomial of degree @xmath338 in @xmath229 unknowns over the field @xmath334 : @xmath339 however , from a computer algebra perspective , this is not what we want . assuming that @xmath340 has entries in @xmath178 , we prefer not to pass to the field extension @xmath334 , but we want the algebraic boundary @xmath333 to be the @xmath178-zariski closure of the above hypersurface @xmath341 . for instance , suppose that the characteristic polynomial of @xmath340 is irreducible over @xmath178 . then we must take the product of @xmath342 over all permutations of the eigenvalues @xmath335 , and the polynomial @xmath343 that defines @xmath333 over @xmath178 is the reduced part of that product . it equals @xmath344 where @xmath345 denotes the _ tensor sum _ of two square matrices of the same size ( see e.g. @xcite ) . here the product goes only up to @xmath346 because the matrices @xmath259 and @xmath340 have the same trace . for special matrices @xmath340 , the characteristic polynomial may factor over @xmath178 , and in this case the algebraic boundary @xmath333 is cut out by a factor of the polynomial @xmath342 or @xmath343 . the space @xmath222 consists of skew - symmetric @xmath17-matrices @xmath347 . the eigenvalues of @xmath347 are purely imaginary , say @xmath348 , where @xmath349 , with @xmath350 and an additional @xmath146 eigenvalue if @xmath55 is odd . thus @xmath347 is not diagonalizable over @xmath91 , but the adjoint @xmath1-action brings the matrix @xmath347 into the normal form @xmath351 here @xmath352 is a suitable matrix in @xmath353 , @xmath354 is the diagonal matrix with diagonal @xmath355 , and we denote @xmath356 . let @xmath357 be the linear map such that @xmath358 if @xmath347 is in normal form as above , then @xmath359 . we call @xmath360 the _ skew - diagonal _ of @xmath347 . in analogy to the symmetric case , the set @xmath361 of all skew - diagonals arising from @xmath299 is nicely behaved ; in fact , the necessary changes are rather modest . for a point @xmath362 we denote by @xmath363 the vector of absolute values . for @xmath364 let @xmath365 be the set of points @xmath362 such that @xmath366 is _ weakly majorized _ by @xmath367 . this means that @xmath367 and @xmath366 satisfy the majorization conditions except that @xmath368 is allowed . the polytope @xmath369 is the _ @xmath370-permutahedron_. it is the convex hull of the orbit of @xmath133 under the action of the coxeter group @xmath370 , the group of all @xmath371 signed permutations . the @xmath370-permutahedron @xmath365 for @xmath372 consists of all points @xmath362 with @xmath373 $ with $ i \cap j = \emptyset$.}\ ] ] as expected , we have the following analog of the symmetric schur - horn theorem . let @xmath374 with @xmath375 and @xmath376 the skew - symmetric schur - horn orbitope of @xmath347 . then @xmath377 is weakly majorized by @xmath378 . in particular , we have @xmath379 . the same arguments as in the symmetric case yield the following results . [ thm : facelatticeskewsh ] every @xmath1 orbit of faces of the skew - symmetric schur - horn orbitope @xmath376 contains the pullback of a unique @xmath370-orbit of faces of the @xmath370-permutahedron @xmath380 . the skew - symmetric schur - horn orbitope @xmath376 coincides with the set of skew - symmetric matrices @xmath259 such that @xmath381 is weakly majorized by @xmath377 . fix @xmath382 and @xmath383 , and @xmath384 . then the system ( [ eq : bpermineq ] ) consists of @xmath385 linear inequalities , namely , six inequalities @xmath386 , twelve inequalities @xmath387 , and eight inequalities @xmath388 . their solution set is the @xmath389-permutahedron @xmath390 , commonly known as the _ truncated cuboctahedron _ , and it has @xmath391 vertices , @xmath392 edges and @xmath385 facets ( six octagons , twelve squares and eight hexagons ) . a picture is shown in figure [ f : b3permuta ] . @xmath393{figures / bpermut}\ ] ] -0.3 cm let @xmath347 denote a general skew - symmetric @xmath394-matrix . theorem [ thm : facelatticeskewsh ] implies that the facets of the @xmath21-dimensional orbitope @xmath376 come in three families , corresponding to @xmath395-orbits of the octagons , squares , and hexagons in figure [ f : b3permuta ] . the facets arising from the octagons are skew - symmetric schur - horn orbitopes for @xmath69 with skew - diagonal @xmath396 and therefore have dimension six . the facets arising from the squares are the product of a line segment and a disc , coming from @xmath397 with @xmath398 acting by the determinant . the facets arising from the hexagons are @xmath399-orbitopes isomorphic to symmetric schur - horn orbitopes with eigenvalues @xmath400 and therefore have dimension five . we next present a spectrahedral description of an arbitrary skew - symmetric schur - horn orbitope @xmath376 . to derive this , we return to symmetric matrices and their real eigenvalues . let @xmath374 be a matrix with eigenvalues @xmath401 and let @xmath402 then @xmath403 has eigenvalues @xmath404 . for any @xmath405 , the additive compound matrix @xmath406 has largest eigenvalue @xmath407 . we conclude that each skew - symmetric skew - horn orbitope @xmath380 is a spectrahedron : [ thm : skewsymspec ] let @xmath374 with @xmath408 . then @xmath409 from this theorem we can derive a description of the algebraic boundary as before , and again the issue arises that the @xmath410 lie in an extension @xmath334 over the field of definition of @xmath347 , which will usually be @xmath178 . at present we do not know whether theorems [ thm : symspec ] and [ thm : skewsymspec ] can be extended to obtain spectrahedral representations of the respective orbitopes over @xmath178 . we close this section with one more example of a skew - symmetric schur - horn orbitope . consider the skew - symmetric schur - horn orbitope @xmath376 for some @xmath411 with @xmath412 . according to theorem [ thm : skewsymspec ] , a spectrahedral representation is @xmath413 the projection @xmath361 to the skew diagonal is the _ crosspolytope _ @xmath414 . this a regular polytope with symmetry group @xmath370 , and it has only one orbit of faces in each dimension . the orbitope @xmath376 is the @xmath26 instance of the _ grassmann orbitope _ @xmath415 . these are important in the theory of calibrated manifolds , and we shall study them in section 7 . we argued in section [ sec : setup ] that , given a compact group @xmath2 acting algebraically on @xmath416 , we can identify @xmath2 with a subgroup of @xmath1 , or even of @xmath0 when @xmath2 is connected . the ambient space @xmath417 is itself an @xmath418-dimensional real representation of the group @xmath2 . the action of @xmath2 on @xmath419 is by left multiplication . the orbit of the identity matrix @xmath420 under this action is the group @xmath2 itself . we call the corresponding orbitope @xmath421 the _ tautological orbitope _ for the pair @xmath422 . this orbitope lives in @xmath423 , and it serves as an initial object because it maps linearly to every orbitope @xmath52 in @xmath32 . tautological orbitopes of finite permutation groups have been studied under the name of _ permutation polytopes _ ( see @xcite ) . the most famous of them all is the _ birkhoff polytope _ for @xmath424 , which was studied for other coxeter groups by mccarthy , ogilvie , zobin , and zobin @xcite . in this section we investigate the tautological orbitopes for the full groups @xmath1 and @xmath0 . similar to the schur - horn orbitopes in section [ sec : schurhorn ] , the facial structure is governed by polytopes arising from the projection onto the diagonal . we begin with the example @xmath425 . the group @xmath60 of @xmath426 rotation matrices has dimension three . its tautological orbitope is a convex body of dimension nine . the following spectrahedral representation was suggested to us by pablo parrilo . [ prop : pabloso3 ] the tautological orbitope @xmath427 is a spectrahedron whose boundary is a quartic hypersurface . in fact , a @xmath426-matrix @xmath428 lies in @xmath427 if and only if @xmath429 we first claim that @xmath427 coincides with the set of all @xmath426-matrices @xmath430 where @xmath431 runs over all positive semidefinite @xmath432-matrices having trace @xmath70 . positive semidefinite @xmath432-matrices with both trace @xmath70 and rank @xmath70 are of the form @xmath433 their convex hull is the free spectrahedron of example [ ex : freespec ] . the image of the above rank @xmath70 matrices @xmath434 under the linear map ( [ eq : pablosu2 ] ) is precisely the group @xmath60 . this parametrization is known as the _ cayley transform_. geometrically , it corresponds to the double cover @xmath435 . the claim follows because the linear map commutes with taking the convex hull . the symmetric @xmath432-matrices @xmath431 with @xmath436 form a nine - dimensional affine space , and this space is isomorphic to the nine - dimensional space of all @xmath426-matrices @xmath428 under the linear map given in . we can express each @xmath437 in terms of the @xmath438 by inverting the linear relations @xmath439 , @xmath440 , etc . the resulting symmetric @xmath432-matrix @xmath441 is precisely the matrix in the statement of proposition [ prop : pabloso3 ] . the ideal of the group @xmath399 is generated by the entries of the @xmath426-matrix @xmath442 , while the prime ideal of @xmath60 is that same ideal plus @xmath443 . we can check that the prime ideal of @xmath60 coincides with the ideal generated by the @xmath444-minors of the matrix . thus the group @xmath60 is recovered as the set of matrices of rank one . proposition [ prop : pabloso3 ] implies that @xmath427 is affinely isomorphic to the free spectrahedron for @xmath68 , that is , to the set of positive semidefinite @xmath432-matrices with trace equal to @xmath70 . this implies a characterization of all faces of the tautological orbitope for @xmath60 . first , all faces are exposed because @xmath427 is a spectrahedron . all of its proper faces are free spectrahedra , for @xmath445 , so they have dimensions @xmath146 , @xmath14 and @xmath446 , as seen in example [ ex : freespec ] . we now examine @xmath447 . as before , let @xmath448 denote the projection of the @xmath17-matrices onto their diagonals . [ lem : ontodiagonal ] the projection @xmath449 of the tautological orbitope for the orthogonal group @xmath1 to its diagonal is precisely the @xmath55-dimensional cube @xmath450^n$ ] . the columns of a matrix @xmath451 are unit vectors . thus every coordinate @xmath452 in bounded by @xmath70 in absolute value and @xmath449 is a subset of the cube . for the reverse inclusion , note that all @xmath453 diagonal matrices with entries @xmath454 are orthogonal matrices . the cube @xmath455^n$ ] is the special @xmath456-permutahedron @xmath457 . as with schur - horn orbitopes , the projection onto this polytope reveals the facial structure . as general endomorphisms are not normal , the key concept of diagonalizability is replaced by that of _ singular value decomposition_. recall that for any linear map @xmath226 there are orthogonal transformations @xmath458 such that @xmath459 is diagonal with entries @xmath460 . these entries are the _ singular values _ of @xmath259 . we shall see in ( [ eq : onspec ] ) that @xmath461 is a spectrahedron , hence all of its faces are exposed faces . the following result recursively characterizes all faces of this tautological orbitope . [ thm : tautoon ] let @xmath462 be a linear function on @xmath18 with @xmath463 . then the face of @xmath461 in direction @xmath215 is isomorphic to @xmath464 where @xmath465 . let @xmath466 be a linear function with @xmath467 , so that @xmath468 . we fix a singular value decomposition @xmath469 of the matrix @xmath273 . here @xmath470 is a diagonal matrix @xmath17 with its first @xmath471 entries positive and remaining @xmath472 entries zero . this matrix also defines a linear function @xmath473 on @xmath18 . cyclic invariance of the trace ensures that the faces @xmath474 and @xmath475 are isomorphic . the subset of @xmath1 at which @xmath476 is maximized is the subgroup @xmath477 . the convex hull of this subgroup equals @xmath475 . it coincides with the tautological orbitope for @xmath478 . we interpret theorem [ thm : tautoon ] geometrically as saying that the tautological orbitope for @xmath1 is a continuous analog of the @xmath55-dimensional cube . every face of the cube is a smaller dimensional cube and the dimension of a face maximizing a linear functional @xmath215 is determined by the support of @xmath215 . the role of the support is now played by the rank of the matrix @xmath273 . this behavior yields information about the carathodory number of the tautological orbitope . the carathodory number of the orbitope @xmath479 is at most @xmath480 . by ( * ? * lemma 3.2 ) , the carathodory number of a convex body @xmath137 is bounded via @xmath481 since every proper face is isomorphic to @xmath482 for some @xmath483 the result follows by induction on @xmath55 . the base case is @xmath484 for which @xmath485 is a @xmath70-simplex . note that the orbit @xmath486 coincides with the orbit of the identity matrix @xmath420 under the action of the product group @xmath487 by both right and left multiplication . hence the tautological orbitope @xmath461 is also an @xmath487-orbitope for that action . we shall now digress and study these orbitopes in general . after we have seen ( in theorem [ thm : ononspec ] ) that these are spectrahedra , we shall resume our discussion of @xmath461 . the group @xmath488 acts on @xmath18 by simultaneous left and right translation . the action is given , for @xmath489 and @xmath226 , by @xmath490 ky fan proved in @xcite that the schur - horn theorem for symmetric matrices under conjugation by @xmath1 generalizes to arbitrary square matrices under this @xmath491 action . now , singular values play the role of the eigenvalues . the following is a convex geometric reformulation : [ lem : fan ] for a square matrix @xmath226 let @xmath492 denote its orbitope under the action ( [ eq : fanact ] ) of the group @xmath493 . then the image @xmath494 of the projection to the diagonal is the @xmath456-permutahedron with respect to the singular values @xmath495 . we shall refer to @xmath496 as the _ fan orbitope _ of the matrix @xmath259 . from lemma [ lem : fan ] , one easily deduces the analogous results to theorems [ thm : facelatticesh ] and [ thm : facelatticeskewsh ] . the facial structure of the fan orbitope @xmath492 is determined by the facial structure of the @xmath456-permutahedron @xmath497 specified by the singular values of the matrix @xmath259 . the description of the @xmath456-permutahedron in terms of weak majorization was stated in . rephrasing these same linear inequalities for the singular values , and using lemma [ lem : fan ] , now leads to a spectrahedral description of the fan orbitopes . for that we make use of the alternative characterization of singular values as the square roots of the ( non - negative ) eigenvalues of @xmath498 . using schur complements , it can be seen that the @xmath499-matrix @xmath500 has the eigenvalues @xmath501 . we form its additive compound matrices as before . [ thm : ononspec ] let @xmath259 be a real @xmath17-matrix with singular values @xmath502 . then its fan orbitope @xmath492 equals the spectrahedron @xmath503 fix an integer @xmath504 . the _ ky fan @xmath133-norm _ is defined by @xmath505 this function is indeed a norm on @xmath18 , and its unit ball is the fan orbitope @xmath492 where @xmath259 is the diagonal matrix with @xmath133 diagonal entries @xmath506 and @xmath507 diagonal entries @xmath146 . theorem [ thm : ononspec ] shows that the unit ball in the ky fan @xmath133-norm is a spectrahedron . two norms of special interest in applied linear algebra are the _ operator norm _ @xmath508 and the _ nuclear norm _ @xmath509 . indeed , these two norms played the key role in work of fazel , recht and parrilo @xcite on compressed sensing in the matrix setting . both the operator norm and the nuclear norm are closely tied to their vector counterparts . the unit balls in the operator and nuclear norm are both fan orbitopes . the projection to the diagonal yields the unit balls for the @xmath510 and the @xmath511-norm on @xmath24 . these two unit balls are polytopes in @xmath24 , namely , the @xmath55-cube and the @xmath55-crosspolytope respectively . fazel _ et al . _ showed in ( * ? ? ? 2.1 ) that the nuclear norm ball has the structure of a spectrahedron , and semidefinite programming duality allows for linear optimization over the operator norm ball . the spectrahedral descriptions of both unit balls in @xmath18 are special instances of theorem [ thm : ononspec ] , to be stated explicitly once more . the operator norm ball consists of all matrices @xmath190 whose largest singular value is at most @xmath70 . this is equivalent to @xmath512 the nuclear norm ball consists of all matrices @xmath190 for which the sum of the singular values is at most @xmath70 . this is equivalent to saying that the sum of the largest @xmath55 eigenvalues of the symmetric @xmath513 matrix @xmath514 is at most @xmath70 . hence the nuclear norm ball in @xmath18 is the spectrahedron defined by the linear matrix inequality @xmath515 we are now finally prepared to return to the main aim of this section , which is the study of tautological orbitopes . the proof of theorem [ thm : tautoon ] implies that the convex hull of @xmath1 equals the set of @xmath516-matrices whose largest singular value is at most @xmath70 . in other words : the operator norm ball in @xmath18 is equal to the tautological orbitope of the orthogonal group @xmath1 . the coorbitope @xmath517 is the nuclear norm ball . both of these convex bodies are spectrahedra . they are characterized in ( [ eq : onspec ] ) and ( [ eq : onspec2 ] ) respectively . it would be worthwhile to explore implications of our geometric explorations of orbitopes for algorithmic applications in the sciences and engineering , such as those proposed in @xcite . we next discuss the faces of the tautological orbitope of the group @xmath0 . the relevant convex polytope is now the convex hull @xmath518 of all vertices @xmath519 with an even number of @xmath520-entries . this polytope is known as the _ demicube _ or _ halfcube_. it is a permutahedron for the coxeter group of type @xmath521 , i.e. signed permutations with an even number of sign changes . the facet hyperplanes of @xmath518 are derived by separating infeasible vertices of @xmath455^n$ ] by hyperplanes through the @xmath55 neighboring vertices : @xmath522^n \,:\ , \sum_{i \not\in j } x_i - \sum_{i \in j } x_i \le n-2 \,\,\,\ , \hbox{for all $ j \subseteq [ n]$ of odd cardinality } \bigr\}.\ ] ] while @xmath523 is only a line segment , we have @xmath524 for @xmath525 . for example , the halfcube @xmath526 is the tetrahedron with vertices @xmath527 , @xmath528 , @xmath529 and @xmath530 . note that its facet inequalities appear as the diagonal entries in the symmetric @xmath432-matrix ( [ eq : magic ] ) : @xmath531 this observation is explained by results of horn ( cf . @xcite ) on the diagonals of special orthogonal matrices . these imply the following lemma about the tautological orbitope of @xmath0 : the projection of @xmath532 onto the diagonal equals the halfcube @xmath518 . just like in lemma [ lem : ontodiagonal ] , it is clear that @xmath518 is a subset of @xmath533 . the converse is derived from the linear algebra fact that the trace of any matrix in @xmath534 is at most @xmath535 . for @xmath536 $ ] let @xmath537 be the diagonal matrix with @xmath538 if @xmath539 and @xmath540 if @xmath541 . let @xmath542 . then @xmath543 for all @xmath544 of odd cardinality . this means that @xmath545 satisfies the linear inequalities in ( [ eq : sonfacets ] ) and hence lies in @xmath518 . there is a variant of singular value decomposition with respect to the restricted class of orientation preserving transformations . for every matrix @xmath226 there exist rotations @xmath546 such that @xmath547 is diagonal . the diagonal entries are called the _ special singular values _ and denoted by @xmath548 . the main difference to the usual singular values is that @xmath549 may be negative ; only the first @xmath550 entries of @xmath551 are non - negative . we need to make this distinction in order to understand the faces of @xmath552 . [ thm : sontauto ] the tautological orbitope of @xmath0 has precisely two orbits of facets . these are the tautological orbitopes for @xmath553 and the free spectrahedra of dimension @xmath554 . up to @xmath521-symmetry , the halfcube @xmath518 has only two distinct facets , namely an @xmath311-dimensional halfcube and an @xmath311-dimensional simplex . a typical halfcube facet @xmath555 arises by maximizing the linear function @xmath556 over @xmath518 , and a typical simplex facet @xmath557 arises by maximizing the linear function @xmath558 . pulling back @xmath215 along the diagonal projection @xmath559 , we see that the facet of @xmath552 corresponding to the halfcube facet is the convex hull of all rotations @xmath542 that fix the first standard basis vector @xmath560 . pulling back @xmath476 along @xmath559 , we see that the facet of @xmath552 corresponding to the simplex facet is the convex hull of @xmath561 . this facet is isomorphic to the convex hull of all @xmath562 such that @xmath563 . since @xmath564 is orientation reversing , one eigenvalue of @xmath564 is @xmath565 , and @xmath566 forces all other eigenvalues to be equal to @xmath70 . hence the facet in question is the symmetric schur - horn orbitope for the diagonal matrix @xmath567 . example [ ex : freespec ] implies that this is a free spectrahedron . in subsection 4.1 we exhibited a spectrahedral representation for the tautological orbitope @xmath427 , and in that case , the two facet types of theorem [ thm : sontauto ] collapse into one type . at present , we do not know how to generalize the representation ( [ eq : magic ] ) to @xmath0 for @xmath568 . orbitopes for @xmath19 were first studied by carathodory @xcite . the coorbitope cone ( [ eq : coorbitopecone ] ) dual to such a _ carathodory orbitope _ consists of non - negative trigonometric polynomials . this leads to the toeplitz spectrahedral representation of the universal carathodory orbitope in theorem [ th : toeplitz ] , which implies that the convex hulls of all trigonometric curves are projections of spectrahedra . the universal carathodory orbitope is also affinely isomorphic to the convex hull of the compact even moment curve , whose coorbitope cone consists of non - negative univariate polynomials . this leads to the representation by hankel matrices in theorem [ th : hankel ] . the irreducible representations @xmath569 of @xmath19 are indexed by non - negative integers @xmath123 . here , @xmath570 is the trivial representation . when @xmath571 is positive , the representation @xmath569 of @xmath19 acts on @xmath63 , and it sends a rotation matrix to its @xmath123th power : @xmath572 for a vector @xmath573 we consider the direct sum of these representations @xmath574 the _ carathodory orbitope _ @xmath575 is the convex hull of the orbit @xmath576 under the action @xmath577 on the vector space @xmath578 . this orbit is the _ trigonometric moment curve _ @xmath579 this curve is also identified with the matrix group @xmath580 lying in the space @xmath581 of block - diagonal @xmath582-matrices with @xmath23 blocks of size @xmath583 . thus @xmath575 is isomorphic to the convex hull of @xmath580 , and can therefore also be thought of as a tautological orbitope . we distinguish between isomorphisms of orbitopes that preserve the @xmath19-action , and the weaker notion of affine isomorphisms that preserve their structure as convex bodies . [ l : all_cara ] any orbitope of the circle group @xmath19 is isomorphic to a carathodory orbitope @xmath575 , where @xmath584 has distinct coordinates , and it is affinely isomorphic to a carathodory orbitope where the coordinates of @xmath259 are relatively prime integers . let @xmath51 be an orbitope for @xmath19 . we may assume that its ambient @xmath19-module @xmath32 has no trivial components and is the linear span of @xmath51 . then @xmath32 has the form @xmath577 for @xmath585 with distinct non - zero components , that is , @xmath32 is multiplicity free . this is because @xmath586 , where we identify @xmath63 with @xmath92 and @xmath19 with the unit circle in @xmath92 . the orbitope @xmath51 is generated by a vector @xmath587 with non - zero coordinates . by complex rescaling , we may assume that @xmath588 , showing that @xmath51 is isomorphic to @xmath575 . lastly , if the coordinates of @xmath589 have greatest common divisor @xmath123 , then @xmath577 is the composition @xmath590 , where @xmath591 and @xmath51 is affinely isomorphic to an orbitope for the module @xmath592 . we henceforth assume that @xmath593 where the @xmath594 are relatively prime . when @xmath595 , carathodory @xcite studied the facial structure of @xmath575 . an even - dimensional cyclic polytope is the convex hull of finitely many points on carathodory s curve ( see @xcite ) . smilansky @xcite studied the four - dimensional carathodory orbitopes ( @xmath26 ) , and this was recently extended by barvinok and novik @xcite who studied @xmath596 . the corresponding curve ( [ caracurve ] ) is the _ symmetric moment curve _ which gives rise to a remarkable family of centrally symmetric polytopes with extremal face numbers . many questions remain about the facial structure of the barvinok - novik orbitopes @xmath596 . see @xcite for details . we now focus on the _ universal carathodory orbitope _ @xmath597 in @xmath598 . this convex body has the following spectrahedral representation in terms of hermitian toeplitz matrices . [ th : toeplitz ] the universal carathodory orbitope @xmath599 is isomorphic to the spectrahedron consisting of positive semidefinite hermitian toeplitz matrices with ones along the diagonal : @xmath600 we note that the complex spectrahedron ( [ toeplitz ] ) can be translated into a spectrahedron over @xmath91 as follows . consider a hermitian matrix @xmath601 where @xmath150 is real symmetric and @xmath2 is real skew - symmetric . the hermitian matrix @xmath602 is positive definite if and only if @xmath603 a _ trigonometric curve _ is any curve in @xmath24 that is parametrized by polynomials in the trigonometric functions sine and cosine , or equivalently , any curve that is the image under a linear map of the universal trigonometric moment curve where @xmath595 . the convex hull of any trigonometric curve is a projected spectrahedron . in particular , all carathodory orbitopes are projected spectrahedra . @xmath604{figures / trig_curve_1.eps } } \put(135,0){\includegraphics[height=120pt]{figures / trig_curve_2.eps } } \put(280,0){\includegraphics[height=120pt]{figures / trig_curve_3.eps } } \put ( 0,-17){$(\cos(\theta),\sin(\theta),\cos(2\theta))$ } \put(135,-17){$(\cos(2\theta),\sin(2\theta),\cos(\theta))$ } \put(280,-17){$(\cos(\theta),\sin(2\theta),\cos(3\theta))$ } \end{picture}\ ] ] -0.3 cm figure [ f : trig_curves ] shows the convex hulls of three trigonometric curves in @xmath66 . the left and middle convex bodies are each the intersection of two convex quadratic cylinders ( @xmath605 and @xmath606 for the former ; @xmath607 and @xmath608 for the latter ) and hence are spectrahedra . the rightmost convex body is visibly not a spectrahedron . its exposed points are @xmath609 for @xmath610 , and it has two algebraic families of one - dimensional facets . in addition , there are two two - dimensional facets , namely the equilateral triangles for @xmath611 and @xmath612 . the six edges of these two triangles are one - dimensional non - exposed faces , as there is no linear function which achieves its minimum on this body along these edges . also , exactly one vertex of each triangle at @xmath613 and @xmath614 is exposed . we conclude that this convex body is a projected spectrahedron but it is not a spectrahedron . the coorbitope cone dual to @xmath615 consists of all affine - linear functions that are non - negative on @xmath615 . these correspond to non - negative trigonometric polynomials : @xmath616 we identify each point @xmath617 in @xmath618 with the laurent polynomial @xmath619\ ] ] with @xmath620 , @xmath621 , @xmath349 , and @xmath622 . then @xmath623 and @xmath624 if and only if @xmath625 is non - negative on the unit circle @xmath153 of @xmath92 . roots of @xmath625 occur in pairs @xmath626 and those on @xmath153 have even multiplicity . choosing one root from each pair gives the factorization @xmath627 where @xmath628 $ ] has degree @xmath23 and the coefficient vectors of @xmath602 and @xmath629 are complex conjugate . this factorization is the classical fejr - riesz theorem . utilizing the monomial map @xmath630 with @xmath631 , this is equivalent to the following : a trigonometric polynomial @xmath632 is non - negative on the unit circle if and only if there is a non - zero vector @xmath633 such that @xmath634 a point @xmath635 belongs to the carathodory orbitope @xmath636 if and only if @xmath637 the sum on the left equals the hermitian inner product in @xmath638 of the coefficient vector @xmath639 of the polynomial @xmath632 and the vector @xmath640 with @xmath641 as in ( [ toeplitz ] ) . the formula ( [ circlesos2 ] ) expresses @xmath639 as the image of the hermitian matrix @xmath642 under some linear projection @xmath643 . if @xmath644 denotes the linear map dual to @xmath643 then @xmath645 is precisely the hermitian toeplitz matrix in ( [ toeplitz ] ) . we conclude that the sum in ( [ circlepairing ] ) equals @xmath646 thus the point @xmath647 represented by a hermitian toeplitz matrix @xmath190 lies in @xmath615 if and only if @xmath648 for all @xmath649 if and only if @xmath190 is positive semidefinite . the proof of theorem [ th : toeplitz ] elucidates the known results about the facial structure of @xmath615 . [ neighborly ] the universal carathodory orbitope @xmath650 is a neighborly simplicial convex body . its faces are in inclusion - preserving correspondence with sets of at most @xmath23 points on the circle . a laurent polynomial @xmath625 as in lies in the boundary of the coorbitope cone @xmath651 if and only if it is non - negative on the unit circle @xmath153 but not strictly positive . it supports the face of @xmath615 spanned by the points of the trigonometric moment curve corresponding to its zeros in @xmath153 . each zero has multiplicity at least @xmath14 , so there are at most @xmath23 such points , and conversely any subset of @xmath652 points supports a face . since any fewer than @xmath653 points on the curve are affinely independent and since all faces are exposed , these faces are simplices . corollary [ neighborly ] implies that the carathodory number @xmath654 equals @xmath655 , as no point in the interior of @xmath615 lies in the convex hull of @xmath23 points of the orbit , but @xmath654 is at most one more than the maximal carathodory number of a facet , by lemma 3.2 of @xcite . carathodory orbitopes are generally not spectrahedra because they can possess non - exposed faces . smilansky @xcite showed that if we write @xmath656 for a point on the trigonometric moment curve with weights @xmath70 and @xmath71 then the faces of @xmath657 are exactly the points @xmath658 of the orbit , the line segments @xmath659 , where @xmath660 , and the triangles @xmath661 in particular , each line segment @xmath662 is a non - exposed edge of @xmath657 . we conclude that the barvinok - novik orbitope @xmath657 is not a spectrahedron . the toeplitz representation ( [ toeplitz ] ) of the universal carathodory orbitope @xmath615 reveals complete algebraic information . for example , the algebraic boundary @xmath663 is the irreducible hypersurface of degree @xmath655 defined by the determinant of that @xmath664-matrix . the curve itself is the set of all positive definite hermitian toeplitz matrices of rank one . the @xmath583-minors of the matrix ( [ toeplitz ] ) generate the prime ideal @xmath665 of this rational curve . the union of the @xmath666-dimensional faces of @xmath615 is the set of positive definite hermitian toeplitz matrices of rank @xmath667 , as a point lies on a @xmath666-dimensional face if and only if it is the convex combination of @xmath667 points of the curve . the zariski closure of this stratum is the set of all rank @xmath667 hermitian toeplitz matrices which is defined by the vanishing of the @xmath668 minors of that matrix . this is also the @xmath667th secant variety of the carathodory curve . lastly , this rank stratification is a whitney stratification of the algebraic boundary . the derivation of the algebraic description of @xmath575 for arbitrary @xmath669 requires the process of elimination . for instance , the ideal @xmath670 of the trigonometric moment curve ( [ caracurve ] ) can be computed from the ideal of @xmath444-minors for @xmath671 by eliminating all unknowns @xmath672 with @xmath673 . the equation of the algebraic boundary @xmath674 is obtained by the same elimination applied to a certain ideal of larger minors of the toeplitz matrix ( [ toeplitz ] ) . we refer to recent work of vinzant @xcite for a detailed study of the edges of barvinok - novik orbitopes . an analysis of the algebraic boundary the orbitope @xmath675 can be found in @xcite . the cone over the degree @xmath23 moment curve is the image of @xmath63 in its @xmath23th symmetric power @xmath676 under the map @xmath677 this map is naturally @xmath19-equivariant . we define the _ compact moment curve _ to be the image @xmath678 of the unit circle under the map @xmath679 . this restricted map equals @xmath680 the convex hull of the curve @xmath678 is an orbitope . by lemma [ l : all_cara ] , it is isomorphic to some carathodory orbitope @xmath575 . the following lemma makes this identification explicit . [ l : identification ] if @xmath681 is odd , then @xmath682 is isomorphic to the barvinok - novik orbitope @xmath683 . if @xmath681 is even , then @xmath682 is isomorphic to @xmath684 , which is affinely isomorphic to the universal carathodory orbitope @xmath685 . complexifying the @xmath19-module @xmath577 where @xmath589 gives the @xmath686-module with symmetric weights @xmath687 . thus the underlying real @xmath19-module of this @xmath686-module is @xmath688 . the lemma follows because the complexified representation @xmath689 of @xmath690 has weights @xmath691 , and this representation is spanned by the pure powers of linear forms , so every weight appears in the linear span of the orbit . suppose now that @xmath692 is even . we describe the moment curve and the carathodory orbitope in coordinates @xmath693 for @xmath694 . fix the @xmath695-hankel matrix @xmath696 [ th : hankel ] the even moment curve consists of all vectors @xmath697 such that the hankel matrix @xmath698 has rank one , is positive semidefinite , and satisfies the linear equation @xmath699 its convex hull consists of all @xmath700 such that @xmath698 is positive semidefinite and satisfies . thus the carathodory orbitope @xmath701 has a second _ hankel representation _ as a spectrahedron . the proof of this theorem follows from the well - known fact that every non - negative polynomial in one variable is a sum of squares of polynomials . it uses the duality in @xcite . observe that the points of the even compact moment curve satisfy , which comes from the polynomial identity @xmath702 . as @xmath703 has just one copy of the trivial representation , this is the only affine equation that holds on @xmath704 . the dual to @xmath703 is the space @xmath705_{2n}$ ] of real homogeneous polynomials of degree @xmath706 in @xmath707 and @xmath708 . the coefficients of such a polynomial gives coordinates for @xmath705_{2n}$ ] . the coorbitope cone @xmath709 dual to the orbitope @xmath710 is the cone of homogeneous polynomials of degree @xmath706 that are non - negative on @xmath63 . thus a point @xmath711 lies in the orbitope @xmath710 if and only if it satisfies and @xmath712 for every non - negative polynomial @xmath713 . since non - negative homogeneous polynomials in @xmath707 and @xmath708 are sums of squares ( cf . @xcite ) , we only need these inequalities to hold when @xmath714 is a square . writing @xmath715 for the coefficient vector of the polynomial @xmath716 , the sum becomes @xmath717 where @xmath698 is the hankel matrix . this proves the theorem . the hankel representation of the universal carathodory orbitope arose by considering the image of the circle @xmath718 in @xmath703 and its relation to non - negative binary forms . generalizing from @xmath63 to @xmath719 gives the veronese orbitopes whose coorbitope cones ( [ eq : coorbitopecone ] ) consist of non - negative @xmath23-ary forms . when @xmath720 , non - negative forms are not necessarily sums of squares , except for quadratic forms and the exceptional case of ternary quartics . the set of decomposable symmetric tensors is the image of the veronese map @xmath721 the @xmath722-orbits through any two non - zero decomposable tensors are scalar multiples of each other and are thus isomorphic . we define the _ veronese orbitope _ @xmath723 to be the convex hull of the orbit through the specific decomposable tensor @xmath724 . that orbit is also the image @xmath725 of the unit @xmath726-sphere under the @xmath472-th veronese embedding of @xmath719 . suppose that @xmath727 is even . then the orbit @xmath725 can be identified with @xmath728 since @xmath729 is two - to - one with @xmath730 . the dual vector space to @xmath731 is the space of homogeneous forms of degree @xmath706 on @xmath719 . the only invariant forms are those proportional to the form @xmath732 , so both @xmath731 and its dual space contain one copy of the trivial representation , and @xmath733 lies in the hyperplane of @xmath731 defined by @xmath734 . the dual cone to the veronese orbitope @xmath735 is the cone of non - negative forms of degree @xmath706 in @xmath736 . see also ( * ? ? ? * example ( 1.2 ) ) . we write @xmath737 for the veronese coorbitope cone consisting of non - negative forms . the cone @xmath738 of non - negative forms contains the cone @xmath739 of sums of squares , but when @xmath720 , @xmath740 , and @xmath741 , hilbert @xcite showed that the inclusion is strict . we refer to @xcite for a recent study which compares the dimension of the faces of these cones . the cone @xmath739 is naturally a projection of the positive semidefinite cone . hence its dual cone @xmath742 is a spectrahedron : it can be realized as the intersection of the positive semidefinite cone with a certain linear space of generalized hankel matrices , discussed in detail in reznick s book @xcite . this spectrahedron @xmath743 is strictly larger than the veronese orbitope @xmath744 when @xmath720 , @xmath740 , and @xmath741 . in fact , @xmath744 is precisely the convex hull of the subset of extreme points in @xmath743 that have rank one . we present a detailed case study of the exceptional case of ternary quartics , when @xmath745 . the veronese orbitope @xmath746 is a @xmath747-dimensional convex body . let @xmath748 be the @xmath21-dimensional cone over the veronese orbitope @xmath749 . as all non - negative ternary quartics are sums of squares , we have the following identities of cones : @xmath750 we next present reznick s spectrahedral representation of @xmath749 . for this , we identify @xmath751 with its dual , and we introduce coordinates @xmath752 where the indices @xmath753 are the exponents of monomials in variables @xmath754 of degree @xmath755 . the ternary quartic corresponding to @xmath700 is @xmath756 where @xmath757 is the multinomial coefficient . the inner product @xmath758 is @xmath60-invariant . given a ternary quartic @xmath759 in the notation , we associate to it the following symmetric @xmath760-matrix with hankel structure as in ( * ? ? ? ( 5.25 ) ) : @xmath761 the veronese orbitope @xmath762 is a spectrahedron . it consists of all positive semidefinite hankel matrices @xmath763 as in @xmath764 that satisfy the equation @xmath765 it is shown in ( * ? ? ? 5 ) that the quartic @xmath759 is non - negative if and only if the hankel matrix @xmath763 is positive semidefinite . furthermore , the equation is the affine equation @xmath766 which defines the hyperplane containing the orbit @xmath767 . by contrast , the veronese coorbitope is not a spectrahedron . [ th : sos ] the convex cone of non - negative ternary quartics @xmath768 is not a spectrahedron . its facets have dimension twelve and the intersection of any two facets is an exposed face of dimension nine . it also has maximal non - exposed faces of dimension nine . we thank greg blekherman who explained this to us . the cone @xmath769 is full - dimensional in the @xmath21-dimensional space @xmath736 . its facets come from its defining linear inequalities @xmath770 for @xmath771 , let @xmath772 be the facet exposed by the inequality @xmath773 , which consists of those non - negative ternary quartics @xmath239 that vanish at @xmath133 . since the boundary of @xmath769 is @xmath747-dimensional and we have a two - dimensional family of isomorphic facets , we see that each facet @xmath772 is @xmath774-dimensional . a non - negative form that vanishes at @xmath771 must also have its two partial derivatives ( in local coordinates at @xmath133 ) vanish , which gives three linear conditions on the facet @xmath772 . concretely , if we take @xmath775 $ ] with local coordinates @xmath776 , then the constant and linear terms of the inhomogeneous quartic @xmath777 must vanish . consequently , @xmath778 where @xmath602 , @xmath779 , and @xmath780 are , respectively , the terms of degrees @xmath14 , @xmath71 , and @xmath755 in @xmath239 . these are binary forms . their @xmath781 coefficients parametrize the linear span of the facet @xmath772 , showing again that @xmath772 is 12-dimensional . the quadratic form @xmath782 is the hessian of @xmath777 at @xmath775 $ ] in these coordinates . a form @xmath239 lies in the relative interior of the facet @xmath772 if and only if , given any @xmath783 which vanishes at @xmath133 along with its partial derivatives , so that it lies in the linear span of @xmath772 , there is an @xmath784 such that @xmath785 also lies in @xmath772 . these conditions are equivalent to 1 . @xmath239 has no other zeros in @xmath786 , and 2 . the hessian of @xmath239 at @xmath133 is a positive definite quadratic form . a form @xmath787 lies in the boundary of @xmath772 when one of these conditions fails , that is , either 1 . @xmath239 vanishes at a second point @xmath788 , or 2 . the hessian of @xmath239 has a double root at some point @xmath789 . faces of type ( 1 ) have the form @xmath790 . these are nine - dimensional and occur in a four - dimensional family parametrized by pairs of distinct points in @xmath786 . the union of all such faces is a semialgebraic subset of dimension @xmath791 in the boundary of @xmath769 . faces of type ( 2 ) also have dimension nine . the condition that the hessian has a double root at a point @xmath792 gives two linear conditions on the coefficients of @xmath239 . there is an additional condition that the cubic part @xmath779 of @xmath239 in also vanishes at @xmath793 , for otherwise @xmath239 takes negative values along the line through @xmath133 corresponding to @xmath793 . a face of type ( 2 ) is the limit of faces @xmath790 of type ( 1 ) as @xmath794 approaches @xmath133 along the line corresponding to @xmath793 . these faces form a three - dimensional family on which @xmath60 acts faithfully and transitively . let us now examine the exposed faces of @xmath769 . let @xmath795 be a symmetric tensor in the dual cone to @xmath769 . then @xmath215 is the sum of decomposable symmetric tensors , @xmath796 and so it supports the face @xmath797 . thus the exposed faces of @xmath769 are intersections of facets , and they consist of non - negative ternary quartics that vanish at a given set of points . since the faces of type ( 2 ) are not of this form , they are not exposed . our next agenda item is a discussion of the algebraic boundaries of the convex bodies and cones discussed above . the algebraic boundary of the cone @xmath748 is characterized by the vanishing of the determinant of the hankel matrix @xmath763 in . from this we conclude : the algebraic boundary of the veronese orbitope @xmath749 is the variety of dimension @xmath798 and degree six which is defined by the linear equation and the hankel determinant @xmath799 . the extreme points of @xmath749 are precisely the hankel matrices @xmath763 of rank @xmath70 . we observed in the proof of theorem [ th : sos ] that the boundary of @xmath769 consists of non - negative quartics @xmath239 that vanish at some point @xmath133 of @xmath786 , and that the partial derivatives of @xmath239 necessarily also vanish at @xmath133 . that is , the plane quartic curve defined by @xmath800 is singular at @xmath133 . thus the algebraic boundary of @xmath769 consists of singular ternary quartics . working in @xmath801 , and its dual space of ternary quartics , this algebraic boundary is seen to be the dual variety to the veronese surface which consists of rank @xmath70 hankel matrices @xmath763 . [ cor : discrorbi ] the algebraic boundary of the coorbitope cone @xmath769 is an irreducible hypersurface of degree @xmath802 . its defining polynomial is the discriminant @xmath803 of the ternary quartic @xmath804 the discriminant @xmath803 is a homogeneous polynomial of degree @xmath802 in the @xmath21 indeterminates @xmath805 . in what follows we shall present an explicit expression for @xmath803 . that expression will be derived from a beautiful classical formula due to sylvester which can be found in section 3.4.d , starting on page 118 , of the book by gel@xmath806fand , kapranov , and zelevinsky @xcite . according to ( 1.7 , page 434 ) , the discriminant @xmath803 is proportional to the resultant @xmath807 of the three partial derivatives of the quartic @xmath239 . here @xmath808 denotes the resultant of three ternary cubics , and the precise relation is @xmath809 . we write @xmath810 for the space of linear forms on @xmath66 , and we introduce the linear map @xmath811 next , for an exponent vector @xmath812 of degree @xmath813 and any variable @xmath814 , we choose a decomposition of the cubic partial derivative @xmath815 where @xmath816 , @xmath817 , and @xmath818 are forms of degree @xmath819 , @xmath820 , and @xmath821 , respectively . then @xmath822 is a quartic polynomial . finally , we define a linear map @xmath823 by sending @xmath824 , where @xmath825 is the basis dual to the monomial basis of @xmath826 . ( sylvester @xcite ) the discriminant @xmath803 is proportional to the resultant of the ternary cubics @xmath827 which is equal to the determinant of the linear map @xmath828 this is an irreducible homogeneous polynomial of degree @xmath802 in the @xmath21 coefficients @xmath805 . we write this map explicitly as a @xmath829 matrix @xmath830 whose rows are indexed by the @xmath21 monomials @xmath831 of degree @xmath832 and whose columns are indexed by @xmath21 auxiliary quartics , and whose entry in a given row and column is the coefficient of that monomial in that auxiliary quartic . nine of the quartics come from the map @xmath833 in . they are @xmath834 the other six are the polynomials @xmath835 , @xmath836 , @xmath837 , @xmath838 , @xmath839 , and @xmath840 from @xmath841 . we only describe @xmath835 and @xmath838 as the others may be recovered from these by symmetry . for @xmath835 , note that each partial derivative of @xmath239 has six terms divisible by @xmath707 , three divisible by @xmath708 and not @xmath707 , and a unique term involving @xmath842 . this leads to a decomposition , and @xmath835 is the determinant of the @xmath843 matrix : @xmath844 by a similar reasoning , we find that @xmath838 is the determinant of the @xmath843 matrix : @xmath845 this concludes our discussion of the algebraic boundary of the coorbitope cone @xmath769 . we close with the remark that the notations @xmath846 and @xmath847 are consistent with those used in the paper @xcite where these cones consist of concentration matrices and sufficient statistics of a certain gaussian model . the _ grassmann orbitope _ @xmath415 is the convex hull of the grassmann variety of oriented @xmath23-dimensional linear subspaces of @xmath24 in its plcker embedding in the unit sphere in @xmath22 . equivalently , this is the highest weight orbitope for the group @xmath0 acting on @xmath22 : @xmath848 faces of the grassmann orbitope are of considerable interest in differential geometry since , according to the fundamental theorem of calibrations , they correspond to area - minimizing @xmath23-dimensional submanifolds of @xmath24 . references to this subject include the seminal article on calibrated geometries by harvey and lawson @xcite and the beautiful expositions by morgan @xcite . in this section we review basic known facts about @xmath415 and we initiate its study from the perspectives of combinatorics , semidefinite programming , and algebraic geometry . vectors in @xmath22 are written in terms of plcker coordinates relative to the standard basis : @xmath849 the plcker vector @xmath133 lies in the _ oriented grassmann variety _ if and only if it is decomposable , i.e. @xmath850 for some pairwise orthogonal subset @xmath851 of @xmath852 . this happens if and only if @xmath133 lies in the unit sphere in @xmath22 and satisfies all quadratic plcker relations , the relations among the @xmath853-minors of a @xmath854-matrix . these relations generate the prime ideal called the _ plcker ideal _ @xmath855 . thus the oriented grassmann variety is the algebraic subvariety of @xmath22 defined by the ideal @xmath856 the convex hull of that real algebraic variety is the @xmath857-dimensional grassmann orbitope @xmath415 . [ ex : g24 ] the grassmann orbitope @xmath858 is the convex hull of the variety of @xmath859 as suggested by ( * ? ? ? * proposition 3.2 ) , we perform the orthogonal change of coordinates @xmath860 this is simultaneous rotation by @xmath861 in each of the coordinate planes spanned by the pairs @xmath862 , @xmath863 , and @xmath864 . in these new coordinates , the prime ideal ( [ eq : pluckerideal24 ] ) equals @xmath865 this reveals that @xmath858 is the direct product of two three - dimensional balls of radius @xmath866 . we next examine the case @xmath867 and arbitrary @xmath55 . the vectors @xmath133 in @xmath868 can be identified with skew - symmetric @xmath17-matrices , and this brings us back to the orbitopes in section 3.2 . if @xmath133 is a real skew - symmetric matrix whose eigenvalues are @xmath872 , where @xmath873 , then the matrix @xmath874 is hermitian and its eigenvalues are the real numbers @xmath875 . recall that the operator @xmath876 computes the @xmath114-th additive compound matrix of a given matrix . [ thm : g2n ] let @xmath877 and @xmath878 . the grassmann orbitope equals the spectrahedron @xmath879 its algebraic boundary @xmath880 is an irreducible hypersurface of degree @xmath881 , defined by a factor of the determinant of the matrix @xmath882 . the proper faces of @xmath869 are @xmath883-orbitopes for @xmath884 . every face @xmath150 is associated with an even - dimensional subspace @xmath885 equipped with an orthogonal complex structure and the extreme points of @xmath150 correspond to complex lines in @xmath885 . everything in this theorem is also true for the small cases @xmath886 , with the exception that the quartic hypersurface @xmath887 is not irreducible , as was seen in example [ ex : g24 ] . in light of theorem [ thm : facelatticeskewsh ] , the reducibility of @xmath858 arises because the two - dimensional crosspolytope equals the square , which decomposes as a minkowski sum of two line segments . this may also be seen as the stabilizer of a decomposable tensor in @xmath69 is @xmath888 , where @xmath889 is the identity matrix , and @xmath890 , so @xmath858 is also an orbitope for @xmath891 . we begin with the last statement about the face lattice of @xmath869 . this result is well - known in the theory of calibrations , where it is usually phrased as follows : _ every face of the grassmannian of two - planes in @xmath24 consists of the complex lines in some @xmath892-dimensional subspace of @xmath24 under some orthogonal complex structure_. see @xcite . to derive the spectrahedral representation ( [ eq : spectrag2n ] ) , we note that the eigenvalues of @xmath893 are the sums of any @xmath114 distinct numbers of the eigenvalues of the skew - symmetric matrix @xmath133 . these are @xmath894 if @xmath55 is even and @xmath895 if @xmath55 is odd . in light of corollary [ rmk : grassinsec3 ] , we can apply the results in section 3.2 to conclude that @xmath133 lies in @xmath869 if and only if @xmath896 for all choices of signs . in terms of polyhedral geometry , this condition means that the vector @xmath897 lies in the crosspolytope @xmath898 . since all @xmath899 eigenvalues of @xmath893 are bounded above by the maximum of the @xmath881 special eigenvalues @xmath900 , we conclude that @xmath901 if and only if @xmath902 . to compute the algebraic boundary @xmath903 we consider the expression @xmath904 this is a symmetric polynomial of degree @xmath905 in the squared eigenvalues @xmath906 , and hence it can be written as a polynomial in the coefficients of the characteristic polynomial @xmath907 the resulting polynomial has degree @xmath881 in the entries @xmath908 of the matrix @xmath909 , and it vanishes on the boundary of the orbitope @xmath869 . it can be checked that it is irreducible for @xmath877 . let @xmath910 and consider the characteristic polynomial of our hermitian matrix : @xmath911 the algebraic boundary of the grassmann orbitope @xmath912 is derived from the polynomial @xmath913 we rewrite this expression in terms of the @xmath21 unknowns @xmath914 to get an irreducible polynomial of degree @xmath915 with @xmath916 terms . this is the defining polynomial of the hypersurface @xmath917 . for @xmath918 we use the same polynomial in @xmath919 but now there is one more eigenvalue @xmath146 . the defining polynomial of @xmath920 has @xmath921 terms of degree @xmath915 in the @xmath922 matrix entries @xmath914 . we now come to the harder case @xmath923 . the grassmann orbitope @xmath924 is a @xmath925-dimensional convex body . its facial structure was determined by dadok and harvey in @xcite , and independently by morgan in @xcite . we present their well - known results in our language . 1 . one - dimensional faces from pairs of subspaces that satisfy the angle condition ( [ eq : threeangles ] ) ; 2 . three - dimensional faces that arise as @xmath60-orbitopes and these are round @xmath71-balls ; 3 . @xmath774-dimensional faces that are @xmath926-orbitopes , from the lagrangian grassmannian . harvey and morgan @xcite extended these results to @xmath927 . we will here focus on @xmath923 . our first goal is to explain the faces described in theorem [ thm : morgan36 ] , starting with the edges . let @xmath928 and @xmath929 be three - dimensional linear subspaces in @xmath930 with corresponding unit length plcker vectors @xmath133 and @xmath794 . we define @xmath931 to be the minimum angle between any unit vector @xmath932 and any unit vector @xmath933 . next , @xmath934 is the minimum angle between two unit vectors @xmath935 and @xmath936 such that @xmath937 and @xmath938 . finally , we define @xmath939 to be the angle between two unit vectors @xmath940 and @xmath941 such that @xmath942 and @xmath943 . we refer to ( * ? ? ? * lemma 2.3 ) for the fact that @xmath944 and @xmath945 are well defined by this rule . the angle condition referred to in part ( 1 ) of theorem [ thm : morgan36 ] is the inequality @xmath946 the convex hull of @xmath133 and @xmath794 is an exposed edge of @xmath924 if and only if the condition ( [ eq : threeangles ] ) holds . the maximal facets in part ( 3 ) of theorem [ thm : morgan36 ] are known to geometers as _ special lagrangian facets _ , and we represent them as orbitopes as follows . the _ special unitary group _ @xmath926 consists of all complex @xmath426-matrices @xmath434 with @xmath947 and @xmath948 . if @xmath949 and @xmath950 , so that @xmath951 , then we can identify @xmath434 with the real @xmath394-matrix @xmath952 note that this matrix lies in @xmath953 if and only if @xmath954 , @xmath955 , and @xmath956 . hence the transformation @xmath957 realizes @xmath926 as a subgroup of @xmath953 . the orbit @xmath958 is known as the _ special langrangian grassmannian_. this is a five - dimensional real algebraic variety in the unit sphere in @xmath959 . we call its convex hull @xmath960 the _ special lagrangian orbitope_. it is obtained from the @xmath925-dimensional grassmann orbitope @xmath924 by maximizing the linear function @xmath961 , which takes value @xmath70 on that face . the face @xmath962 is @xmath774-dimensional because it satisfies the two affine equations @xmath963 plus the six independent linear equations that cut out the langrangian grassmannian : @xmath964 the exposed faces of type ( 2 ) in theorem [ thm : morgan36 ] are not facets . each of them is the intersection of two special lagrangian facets . for example , consider the two linear functionals @xmath965 which were discussed in @xcite . each of them supports a special lagrangian facet . the intersection of these two @xmath747-dimensional facets is a three - dimensional ball . indeed , the linear functional @xmath966 is bounded above by @xmath70 on @xmath924 , and the subset at which the value equals @xmath70 is contained in the linear span of the six basis vectors @xmath967 where @xmath968 . in the intersection of this linear space with @xmath924 we find the grassmann orbitope @xmath858 from example [ ex : g24 ] , with our linear functional @xmath969 being represented by the scaled coordinate @xmath970 . the face of @xmath858 where @xmath970 attains its maximal value @xmath70 is a @xmath71-ball . this concludes our discussion of the census of faces given in theorem [ thm : morgan36 ] . the first assertion was proved by dadok and harvey in ( * ? ? ? * theorem 7 ) . for the second assertion we proceed as follows . we apply the technique in ( * ? ? ? * lemma 2.2 ) and restrict to the linear subspace @xmath973 of @xmath959 . the intersection of @xmath974 is the @xmath975-orbitope of @xmath976 , where the action is by unitary diagonal @xmath426-matrices , while the intersection @xmath977 is the @xmath978-orbitope of @xmath979 , where the action is by special unitary diagonal @xmath426-matrices , as described in @xcite . we claim that the @xmath978-orbitope @xmath977 is @xmath14-neighborly . this can be seen by examining the bivariate trigonometric polynomials of the form @xmath980 indeed , for any choice of @xmath981 and @xmath982 , we can find eight real coefficients @xmath983 such that @xmath984 and its derivatives vanish at @xmath981 and @xmath982 but @xmath984 is strictly positive elsewhere . the results in @xcite imply that every exposed edge of @xmath974 is also an exposed edge of @xmath924 . we believe that morgan s technique can be adapted to show that _ every _ exposed edge of @xmath977 is an exposed edge of @xmath962 . for the purpose of proving theorem [ thm : g36bad ] , however , we only need to identify _ one _ exposed edge of @xmath977 that is an exposed edge of @xmath962 . the claim that such exposed edges exist can be derived from ( * ? ? ? * theorem 12 ( i ) ) . with the help of philipp rostalski , we also obtained a computational proof of that claim . this was done as follows . we selected various random choices of points @xmath981 and @xmath985 in @xmath63 . each choice specifies two three - dimensional subspaces @xmath928 and @xmath929 of @xmath930 for which equality holds in the angle condition ( [ eq : threeangles ] ) . the corresponding line segment is not an exposed edge of @xmath924 . to show that the line segment between @xmath928 and @xmath929 is exposed in @xmath962 , we use a technique from semidefinite programming . first we compute the eight coefficients @xmath986 of the supporting function @xmath984 as above . this gives a linear function @xmath987 on @xmath959 . we then run a first - order lasserre relaxation ( cf . @xcite ) to minimize this linear function subject to the linear and quadratic constraints that cut out the special lagrangian grassmannian . the optimal value is zero , the optimal lasserre moment matrix has rank two , and its image in @xmath959 lies in the relative interior of the line segment between @xmath928 and @xmath929 . we then re - optimize for various perturbations of the linear function @xmath987 . the output of each run is a rank one moment matrix which certifies either @xmath928 or @xmath929 as optimal solution of the optimization problem . this proves that the face of @xmath962 exposed by @xmath987 is the line segment between @xmath928 and @xmath929 . nicholas mccarthy , david ogilvie , nahum zobin , and veronica zobin , _ convex geometry of coxeter - invariant polyhedra _ , trends in banach spaces and operator theory ( memphis , tn , 2001 ) , contemp . math . 321 , amer . soc . , providence , ri , 2003 , pp . 153179 . kristian ranestad and bernd sturmfels , _ the convex hull of a variety _ , to appear in the julius borcea memorial volume , ( eds . petter brndn , mikael passare and mihai putinar ) , trends in mathematics , birkhuser verlag , arxiv:1004.30 . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: quantum chromodynamics , the theory describing the interactions of quarks and gluons , has been shown to provide excellent descriptions of a wide variety of phenomena ranging from hadron spectroscopy via lattice field theory to hard scattering phenomena via perturbation theory . yet qcd is a theory whose degrees of freedom are the quarks and gluons quanta that can not be observed directly due to their confinement in hadrons . on the other hand , experiments deal with hadrons and leptons . in order to calculate observables measured in high energy scattering processes a technique is needed that allows conversion from a description in terms of hadrons to one in terms of quarks and gluons ( collectively referred to as partons ) . for hadrons in the initial state this conversion is provided by parton distribution functions ( pdfs ) , while for final state hadrons a similar role is played by parton fragmentation functions . the pdfs , denoted by @xmath0 , for each type of quark and antiquark and for the gluon ( @xmath1 ) , allow one to essentially describe a beam of hadrons as an effective beam of quarks and gluons . in the infinite momentum frame , the pdfs can be interpreted as probability densities describing how the parent hadron s momentum is shared amongst the different types of partons , as a function of the hadron s momentum fraction @xmath2 . with knowledge of the pdfs and the feynman rules for qcd , one can calculate hard scattering cross sections and differential distributions that can then be compared to data . pdfs are traditionally determined by simultaneously fitting a wide variety of data for large momentum transfer processes . the parameters of the fits describe the pdfs at some initial momentum transfer scale , while evolution equations are then used to calculate the pdfs at all other scales needed for the calculations . the data sets used for such fits often include deep inelastic scattering ( dis ) of charged leptons on proton and deuterium targets , or neutrinos on heavy nuclear targets , lepton pair production on proton and deuterium targets , and the production of photons , vector bosons , or jets at large values of transverse momentum . while the pdfs provide us with a detailed description of the partonic substructure of hadrons , they contain only partial information . partons ( both quarks and gluons ) have nonzero spin , so that the fundamental distributions in nature are the pdfs for a specific helicity ( spin projection along the direction of motion ) , @xmath3 and @xmath4 , corresponding to parton spins aligned and antialigned with that of the hadron , respectively . experiments with unpolarized beams and targets are therefore sensitive only to the sums of the helicity pdfs , @xmath5 , while information on the differences , @xmath6 , can be obtained from measurements involving polarized beams and/or targets . it is the purpose of this topical review to give a current `` snapshot '' of the status of our knowledge of pdfs , both unpolarized and polarized . to this end , brief reviews of both the theory and the types of data used in the fits will be presented , and theoretical issues that need to be addressed in the perturbative calculations will be outlined . throughout this review we shall attempt to answer the question `` what do we know and how do we know it ? '' recently there have been several excellent reviews of pdfs @xcite , updating the early treatises of pdfs in refs . @xcite , which have focused on various aspects of unpolarized pdfs , such as the impact on the phenomenology of the large hadron collider ( lhc ) . the spin structure of the nucleon , including spin - dependent pdfs , was also reviewed in refs . the present pedagogical review should be viewed as complementary to these efforts , emphasizing the different aspects of hadron structure that can be revealed through the study of unpolarized and polarized pdfs , at both small and large parton momentum fractions @xmath2 . there is an old saying in experimental particle physics that `` today s discovery is tomorrow s calibration . '' a similar effect occurs for theoretical calculations . what was once touted as a test of qcd , now often plays the role of providing a description of backgrounds for searches for new types of particles . of course , one wants these calculations to be as accurate as possible and this , in turn , requires precise knowledge of the pdfs that enter into the calculation . the error estimates on pdfs will be discussed at some length for this reason . on the other hand , there are many aspects of hadron structure that are currently not understood well , or at all , either theoretically or experimentally , and clearly identifying and delineating the limits of our knowledge in the context of global pdf fits serves a valuable purpose . the plan of the review is as follows . sec . 2 contains an overview of the necessary qcd theory required to understand the results of global fits . sec . 3 contains the review of unpolarized pdfs , while sec . 4 summarizes the status of polarized pdfs . finally , sec . 5 provides some speculations as to how our knowledge of pdfs may be improved through anticipated results of ongoing and future experiments . as noted in the introduction , pdfs are necessary ingredients for obtaining predictions for hard scattering hadron hadron and lepton hadron processes . the cross section for a typical hadron hadron process involving collisions of hadrons @xmath7 and @xmath8 , with momenta @xmath9 and @xmath10 , respectively , producing a state @xmath11 in addition to other hadrons ( collectively denoted by @xmath12 ) , can be written in the form @xmath13 where @xmath14 denotes an @xmath15-th order parton parton `` cross section '' that produces the desired final state , and the functions @xmath16 and @xmath17 are the pdfs of flavor @xmath18 in hadron @xmath7 and flavor @xmath19 in hadron @xmath8 , respectively . the state @xmath11 might denote a high-@xmath20 jet , a lepton pair , a photon or weak vector boson , for example , and @xmath21 is a scale that characterizes the hard scattering . for high-@xmath20 jet or photon production the scale @xmath22 . the form ( [ eq : xsec ] ) can also be applied to the case when the hadrons @xmath7 and @xmath8 are polarized ; in the remainder of this section , however , we shall for illustration focus primarily on the unpolarized case . the parton - level cross sections will generally possess ultraviolet singularities that must be renormalized . doing so leads to the introduction of the running coupling @xmath23 which depends on a renormalization scale @xmath24 . there can also be infrared singularities associated with loop graphs ; these will cancel corresponding singularities from the emission of real soft gluons , provided that the observable is suitably defined , _ i.e. , _ that it is `` infrared safe . '' finally , there will be collinear singularities associated with , in this case , the initial partons emitting additional partons at zero angle . these collinear configurations correspond to internal propagators going on - shell and , as such , correspond to long - distance physics . such singular terms can be _ factorized _ and absorbed into the pdfs . this process introduces a factorization scale , @xmath25 , that separates the long - distance and short - distance hard scattering physics . as shown in eq . ( [ eq : xsec ] ) , the partonic hard scattering cross section can be expanded in powers of the running coupling . the dependence on @xmath24 that enters via the running coupling cancels that which appears in the partonic cross sections . similarly , the dependence on @xmath25 in the pdfs cancels against the @xmath25 dependence in the partonic cross sections . these cancellations represent the fact that the physical cross section , if calculated to all orders in perturbation theory , should not depend on the two scales introduced to control the ultraviolet and collinear singularities . however , at any fixed order in perturbation theory these cancellations will only be approximate . indeed , at order @xmath26 one has the relations @xmath27 these results suggest that the dependence on the renormalization and factorization scale choices will decrease as one carries out the perturbative calculations to higher order . this is indeed the case for the processes typically used in global fits for pdfs . ( if a new kinematic configuration opens up at a higher order , then the corrections at that order can be large . this happens when going from lo to nlo for heavy quark production , for example , in which case the scale dependence can actually increase at that order . the reduction then occurs at the next order . ) these processes are all known to at least nlo and some to nnlo , as will be discussed below . in principle , the two scales @xmath24 and @xmath25 can be chosen independently . however , in processes where there is one large scale characterizing the hard scattering it is often the case that the choice @xmath28 is used . this simplifying choice will be utilized in the results discussed in the sections to follow . the dependence on the renormalization scale of the running coupling is governed by the qcd @xmath29 function via the renormalization group equation , @xmath30 the qcd @xmath29 function is calculable in perturbation theory and can be written as @xmath31 where the coefficients for @xmath32 flavors are @xmath33 the strong coupling @xmath34 calculated using the expression for @xmath29 in eq . ( [ eq : alphas ] ) with terms up to @xmath35 is referred to as the @xmath36loop running coupling . lo calculations retain only @xmath37 , while nlo calculations retain @xmath38 , and nnlo calculations retain also @xmath39 . note that the @xmath29 function is negative , resulting in a decrease of @xmath34 with increasing scale , a phenomenon known as asymptotic freedom . this behavior is essential for the applicability of perturbation theory for hard scattering processes . in eq . ( [ eq : xsec ] ) both the pdfs and the partonic cross sections depend on the factorization scale @xmath25 . as seen in eq . ( [ eq : scale ] ) , this scale dependence cancels up to the order of perturbation theory used . this fact allows one to calculate the @xmath41 dependence of the pdfs . indeed , the singularities associated with collinear parton emission are universal and factorize from the hard scattering subprocesses . the factorization scale dependence of the parton distributions is then also universal , reflecting the independence of physical quantities on the scale @xmath25 . this leads to the @xmath40 evolution , or renormalization group equations ( rge ) , @xmath42 where the bjorken @xmath2 variable is ( related to ) the longitudinal momentum fraction of the partons . the evolution kernels @xmath43 or `` splitting functions '' stem from the collinear divergences absorbed in the distributions and represent the ( collinear ) resolution of a parton @xmath44 in a parton @xmath45 . they are calculated perturbatively as a series expansion in @xmath46 , @xmath47 at present these have been computed up to 3 loops ( @xmath48 ) , which is the necessary accuracy for a nnlo analysis . the evolution of spin - dependent pdfs @xmath49 follows in a similar manner , with a corresponding set of spin - dependent splitting functions @xmath50 . in this case the splitting functions have been computed to 2 loops , enabling analyses to be performed at nlo accuracy . traditionally , information on the pdfs of the nucleon has come from the process of deep - inelastic scattering ( dis ) of leptons from protons or nuclei , beginning with the pioneering experiments at slac in the late 1960s . in the one - boson exchange approximation , the differential dis cross section can be written as a product of lepton and hadron tensors , @xmath51 where @xmath52 is the fine structure constant , @xmath53 is the laboratory solid angle of the scattered lepton , and @xmath54 is the incoming ( outgoing ) lepton energy . for neutral - currents , the summation is over @xmath55 and the interference @xmath56 , while for charged - currents only @xmath57 exchange contributes . the leptonic tensor @xmath58 depends on the charge @xmath59 and helicity @xmath60 of the lepton , @xmath61 where @xmath62 and @xmath63 are the initial and final electron momenta , and @xmath64 and @xmath65 are the electron vector and axial - vector charges , respectively . the factors @xmath66 in eq . ( [ eq : sigtens ] ) denote the ratios of the corresponding propagators and couplings to the photon propagator and coupling squared @xcite @xmath67 where @xmath68 is the fermi weak interaction coupling constant , and @xmath69 is the @xmath70 boson mass . the hadronic tensor @xmath71 contains all of the information about the structure of the hadron target . using constraints from lorentz and gauge invariance , together with parity conservation , it can be decomposed into spin - independent and spin - dependent contributions , @xmath72 , \label{eq : wmunu}\end{aligned}\ ] ] where @xmath73 and @xmath74 are the nucleon and exchanged boson four - momenta , @xmath75 , and @xmath76 . the nucleon polarization four - vector @xmath77 satisfies @xmath78 and @xmath79 . for spin - averaged scattering , the nucleon structure is parametrized in terms of the vector @xmath80 and @xmath81 structure functions , and the vector - axial vector interference @xmath82 structure function , which requires weak currents . these are generally functions of two variables ( such as @xmath2 and @xmath40 ) , but become functions of @xmath2 only in the bjorken limit , in which both @xmath40 and @xmath83 , but @xmath2 is fixed . in this limit the @xmath80 and @xmath81 structure functions become proportional , according to the callan - gross relation , @xmath84 , and in the parton model are given in terms of quark @xmath85 and antiquark @xmath86 distribution functions , @xmath87,\ ] ] where @xmath88 is interpreted as the probability to find a quark of flavor @xmath85 and charge @xmath89 in the nucleon with light - cone momentum fraction @xmath2 . at finite energies , the logarithmic @xmath40 dependence from the evolution equations described in sec . [ ssec : q2evol ] , as well as residual @xmath40 dependence associated with power corrections ( see sec . [ ssec : power ] below ) , give corrections to the simple parton model expectations . the spin - dependent structure functions @xmath90 and @xmath91 can be extracted from measurements where longitudinally polarized leptons are scattered from a target that is polarized either longitudinally or transversely relative to the electron beam . for longitudinal beam and target polarization , the difference between the cross sections for spins aligned and antialigned is dominated at high energy by the @xmath90 structure function . the @xmath91 structure function can be determined with additional measurement of cross sections for a nucleon polarized in a direction transverse to the beam polarization . in practice one often measures the polarization asymmetry @xmath92 , which is given as a ratio of spin - dependent and spin - averaged structure functions , @xmath93.\end{aligned}\ ] ] at small values of @xmath94 , one then has @xmath95 . if the @xmath40 dependence of the polarized and unpolarized structure functions is similar , the polarization asymmetry @xmath92 will be weakly dependent on @xmath40 . in analogy with the unpolarized @xmath80 structure function , in the parton modelthe structure function @xmath90 can be expressed at lo in terms of differences between quark distributions with spins aligned ( @xmath96 ) and antialigned ( @xmath97 ) relative to that of the nucleon , @xmath98 , @xmath99 . \label{eq : g1quark}\ ] ] the @xmath91 structure function , on the other hand , does not have a simple parton model interpretation . however , its measurement provides important information on the subleading , higher twist contributions which parametrize long - range nonperturbative parton - parton correlations in the nucleon ( see sec . [ ssec : power ] ) . as discussed in sec . [ sec : qcd ] , the pdfs are universal , the collinear singularities associated with initial parton collinear emission having been absorbed into the pdfs . therefore , the parton distributions appearing , for example , in the expressions for dis structure functions , are the same that describe the structure of the incoming hadrons in hadronic production . repeating the general form given in eq . ( [ eq : xsec ] ) ( with @xmath100 ) , one has , for spin - unpolarized scattering , @xmath101 and an analogous expression for the spin - dependent cross section , involving the difference of cross sections with hadrons @xmath7 and @xmath8 polarized in the same and opposite directions , @xmath102 here @xmath103 and @xmath104 are the spin - dependent pdfs for flavor @xmath18 in a hadron @xmath7 and flavor @xmath19 in a hadron @xmath8 , respectively , and @xmath105 is the corresponding spin - dependent partonic cross section . as indicated in eqs . ( [ eq : xsec2 ] ) and ( [ eq : xsec_pol ] ) , the partonic cross sections are calculable in fixed order perturbation theory as series expansions in @xmath106 , which starts with different powers depending on the process . it will be useful for subsequent sections to describe the parton kinematics for several examples of hadron hadron processes . although all processes used in global fits are known at least to nlo , the lo parton kinematics nevertheless serves as a useful guide to what region of parton momentum fraction and which combinations of parton flavors will be constrained by a given set of data . consider high-@xmath20 dijet production in lowest order where two jets are produced with approximately balancing transverse momenta @xmath20 and rapidities denoted by @xmath107 . two - body phase constrains the momentum fractions to be @xmath108 where the dimensionless ratio @xmath109 . for centrally produced dijets one sees that the parton momentum fractions are both approximately @xmath110 . however , if one or both of the jets is produced at far forward or backward rapidity values , then the @xmath2 range can be considerably expanded towards large or small values thereby allowing a `` tuning '' of the probed @xmath2 range . for the case of inclusive production of a jet with transverse momentum @xmath20 and rapidity @xmath111 there is , at lo , an integration over one of the parton momentum fractions , say @xmath112 , given by @xmath113 again , one sees that for @xmath111 near 0 the ranges of the momentum fractions are centered near @xmath110 . another important class of experiments involves the production of a system of mass @xmath114 , rapidity @xmath111 , and transverse momentum @xmath20 . examples of such systems include @xmath115 , @xmath116 , and charged lepton pairs @xmath117 . the four momentum of the produced system is conveniently given as @xmath118 where @xmath119 , and @xmath120 is the azimuthal angle . the rapidity @xmath111 and , equivalently , the feynman@xmath2 variable @xmath121 are defined by @xmath122 which lead to the relations @xcite @xmath123 where @xmath124 . depending on the detected final state it is possible to define different cross sections for hadron hadron collisions . in the drell - yan mechanism @xcite a quark from one hadron and an antiquark from the other one annihilate into an intermediate vector boson ( @xmath125 , @xmath116 or @xmath57 ) which subsequently decays into a lepton pair . for lepton pairs of invariant mass @xmath126 , the process is dominated by virtual photon exchange . the experiments usually consist of proton(beam)nucleon ( proton or some other nucleus ) collisions and the cross sections are extracted from the detection of muon pairs ( dimuon production ) produced in the decay of the virtual photons . the double differential cross section for lepton pair production at lo is given by @xmath127 , \label{eq : dy}\end{aligned}\ ] ] where the two parton momentum fractions are given by @xmath128 from eq . ( [ eq : dy ] ) one can see that for @xmath129 collisions this process is sensitive separately to sea and valence distributions , in contrast to the neutral current dis cross sections , where they enter as @xmath130 . inclusion of the drell yan data in global qcd analyses of pdfs is in fact instrumental in fixing ratio @xmath131 for @xmath132 at lo ( see eq . ( [ eq : sig_pd_pp ] ) below ) . another alternative is the use of charged current neutrino dis structure functions , although this process entails the use of model dependent corrections for the heavy nuclear targets typically employed in such measurements . nuclear corrections can be avoided if one instead uses the time reversed processes such as @xmath133 . a process closely related to drell yan dilepton production is the hadronic production of electroweak bosons @xcite . for @xmath134 scattering , the inclusive production cross section for @xmath57 bosons , for example , is given at lo by @xmath135 where @xmath136 are the ckm matrix elements , and the sum runs over all light quark and antiquark flavors in both hadrons . the production of @xmath137 bosons is then sensitive primarily to the products @xmath138 , while @xmath139 production is sensitive to @xmath140 . at large @xmath70 boson rapidity , or equivalently @xmath132 , @xmath57 production is sensitive to the @xmath141 and @xmath142 quark pdfs in the proton , respectively . provided the antiquark distributions at @xmath143 are known to sufficiently accuracy , the @xmath144 cross section ratio would then be a clean probe of the @xmath145 pdf ratio , which is presently not well determined at large @xmath112 . conversely , if the quark pdfs are known in particular regions of @xmath146 , then the @xmath147 ratios can also be used to probe the antiquark distributions . similar considerations apply when one or both of the protons are polarized , as for the case of the spin program at rhic . for collisions of longitudinally polarized protons from unpolarized protons , @xmath148 , the generalization of eq . ( [ eq : wprod ] ) simply involves replacing one of the quark or antiquark distributions by the corresponding polarized quark ( antiquark ) pdf , @xmath149 . single - spin asymmetries can then constrain ratios of polarized to unpolarized pdfs , @xmath150 , for various flavors , over specific regions of @xmath146 ( see sec . [ ssec : pol_sea ] ) . double - spin asymmetries , formed by combinations of inclusive hadron or jet production cross sections in the scattering of polarized protons from polarized protons , @xmath151 , have also been studied in order to constrain the polarized gluon distribution @xmath152 ( see sec . [ ssec : pol_gluon ] ) . the treatment of heavy quarks in high - energy processes generally requires more care than for light quarks . the production mechanism of heavy quarks in dis is predominantly the photon gluon fusion ( pgf ) process , which makes it particularly sensitive to the gluon distribution in the nucleon . although in principle there could exist an `` intrinsic '' ( initial state ) heavy quark content of the nucleon , existing measurements are well described through `` extrinsic '' ( generated in the hard scattering ) heavy quark production only ( _ cf . _ @xcite and references therein ) , so that any intrinsic heavy quark content of the nucleon is small and restricted to large values of bjorken @xmath2 ( see sec . [ ssec : hqresults ] below ) . the natural framework for describing the pgf mechanism is known as the `` fixed flavor number scheme '' ( ffns ) . in this approach the light - quark flavors ( @xmath153 ) and the gluon are considered as massless partons within the nucleon . this scheme allows one to compute the contributions of heavy quarks ( @xmath154 ) perturbatively as final - state quantum fluctuations , taking into account the full dependence of the production cross section on the mass @xmath155 of the heavy quark . as is common in the @xmath156 renormalization scheme , the evaluation of the strong running coupling @xmath106 is nevertheless based on the usual variable flavor number scheme for the @xmath29function governing its scale dependence . this procedure automatically resums the contributions to the strong coupling from heavy quarks , and consequently improves the stability of the perturbative expansion @xcite . complete calculations of ( heavy quark ) dis structure functions are presently known at lo @xcite and at nlo @xcite . beyond this , approximations based on threshold resummation have been made @xcite ( nnlo * in sec . [ ssec : hqresults ] ) , but thus far the nnlo coefficient functions are only partially known @xcite , which constitutes a major drawback of any precision qcd analysis at nnlo accuracy . another aspect of these calculations is the mass definition used ; while the original calculations of the amplitudes employ the pole mass definition , it has been reported @xcite that the use of the running mass definition improves the stability of the perturbative series . in many situations calculations within the ( fully massive ) ffns scheme become unduly complicated , and in many cases even impossible due to the unknown massive ( @xmath157 ) matrix elements at nnlo or even nlo . for this reason it is common to generate parton distributions in the so - called `` variable flavor number scheme '' ( vfns ) , where the heavy quarks are also considered to be massless constituents of the nucleon . here , the required nlo and nnlo cross sections have been computed for a variety of important production processes , in particular at hadron colliders . in the vfns scheme the effective heavy quark distributions are generated perturbatively from the nonperturbative distributions of light quarks and gluons using the boundary conditions of ref . @xcite , typically at the unphysical `` thresholds '' @xmath158 . from there one proceeds to their renormalization group evolution with an increased number of flavors @xmath159 and ( eventually ) 6 . for situations where the ( threshold ) invariant mass of the produced system far exceeds the mass of the interacting heavy flavor in the ffns , the vfns predictions have been shown @xcite to deviate from the ffns typically by about 10% , although the exact amount depends on the particular process and energy scale . this is usually within the margins of the renormalization and factorization scale uncertainties and other theoretical ambiguities related to pdfs . eventually , one nevertheless has to _ assume _ that these massless `` heavy '' quark distributions are relevant asymptotically , in that they can correctly describe the relevant cross sections at scales much larger than the heavy quark masses involved . unfortunately this is not the case for dis structure functions , for which the strictly massless approach , known in this context as the zero mass vfns , or zm - vfns , is well known to be experimentally inadequate , particularly near the heavy quark production thresholds . to remedy this the heavy - quark mass effects are re inserted in what are known as general - mass vfnss ( gm - vfns ) , for which several prescriptions are available in the literature @xcite ( see also ref . these schemes interpolate ( in a model - dependent way ) between the ffns results near production threshold and the asymptotic results of the zm - vfns . however , since the interpolating schemes are based on the same massive matrix elements as the ffns , they do not have complete information on the @xmath160 heavy flavor wilson coefficients ( even though they are sometimes referred to as `` nnlo gm - vfnss '' ) . the importance for global pdf analysis of finite heavy quark mass effects in the calculation of dis structure functions had been previously emphasized in refs . @xcite , but was only universally recognized after the analysis of @xcite . currently the abm @xcite and jr @xcite collaborations use the ffns for the calculation of dis structure functions in their global analyses , while the mstw @xcite , ct @xcite , herapdf @xcite and nnpdf @xcite groups use variants of gm - vfns . the calculations of hadron collider cross sections are carried out in the vfns in all cases . the elegant machinery that has been developed within the framework of perturbative qcd to analyze leading twist pdfs is , strictly speaking , valid only at high values of @xmath40 and @xmath70 where all hadron mass scales are suppressed , @xmath161 . in real experiments performed at a finite beam energy @xmath162 , however , the maximum values of @xmath40 and @xmath70 are limited , which inevitably restricts the available coverage in bjorken @xmath2 . this is especially relevant at large @xmath2 , where in dis the invariant mass squared of the produced hadronic system is given by @xmath163 with @xmath164 the mass of the target nucleon . for fixed @xmath40 , as @xmath165 the final state hadron mass @xmath70 decreases as one descends into the region dominated by nucleon resonances at @xmath166 gev . the resonance region may be treated using the concept of quark - hadron duality @xcite , although this goes beyond the scope of the usual pqcd analysis . in the region of low @xmath40 , power corrections to the bjorken limit results that scale as powers of @xmath167 become increasingly important . in the operator product expansion , these are associated with higher twist corrections , which arise from multi - parton correlations and characterize the long - range nonperturbative interactions between quarks and gluons . of tremendous interest in their own right as providing glimpses into the dynamics of quark confinement , the power corrections are viewed as troublesome backgrounds to efforts aimed solely at extracting leading twist pdfs . to avoid the complications from the higher twist corrections , the usual strategy in global pdf analyses is to apply cuts specifying minimum values of @xmath40 and @xmath168 . in many analyses of unpolarized scattering data , the cuts are of the order @xmath169 gev@xmath170 and @xmath171 gev@xmath170 @xcite , which in practice restrict the range of @xmath2 that can be accessed to @xmath172 . for spin - dependent pdfs , the scarcity of high - energy data forces most global analyses to use less restrictive cuts , of the order gev@xmath170 and @xmath174 gev@xmath170 . on the other hand , there are a number of important reasons for needing to know the large-@xmath2 behavior of pdfs , and several recent unpolarized pdf analyses @xcite have been performed with relaxed cuts of @xmath175 and @xmath176 gev@xmath170 which have allowed an expanded reach into large @xmath2 . the improvement in the large-@xmath2 kinematic coverage amounts to about 1300 more data points for proton and deuteron targets , representing some @xmath177 increase in the total number of dis data points compared with the more restrictive cuts . most importantly , the resulting leading twist pdfs fits have proved to be very stable even when such low cuts have been applied @xcite . with the inclusion of the kinematic regions in which subleading @xmath178 effects play a non - negligible role , it is obviously crucial to account for the power corrections that might otherwise obfuscate the leading twist pdfs . among the different categories of @xmath178 effects , the simplest are the target mass corrections ( tmcs ) , which are formally associated with matrix elements of leading twist operators @xcite and are of kinematical origin . others include genuine higher twist corrections , which arise from dynamical , multi - parton correlations , as well as higher order perturbative qcd corrections , which can also resemble power suppressed contributions at low @xmath40 . the standard method to compute tmcs in dis is based on the operator product expansion , and was first formulated by georgi and politzer @xcite , and expressions for all unpolarized and polarized structure functions now exist in both @xmath2 and mellin space @xcite . an alternative method based on collinear factorization ( cf ) in momentum space was developed by ellis , furmanski and petronzio @xcite , and extended by various authors @xcite ( see also the recent reviews of tmcs in refs . the advantage of the ope method is that tmcs can be calculated to all orders in @xmath178 in dis , whereas tmcs in the cf approach have only been computed to @xmath179 . on the other hand , the ope is limited to inclusive dis , while the cf framework can be applied to computing tmcs also in other processes @xcite . since typically the non - dis data are taken at very high @xmath40 where tmc effects are very small , the ope method is usually adopted . all of the tmc methods also suffer to some extent from the threshold problem , whereby the target mass corrected structure function remains nonzero as @xmath165 @xcite . for the purposes of global fits , however , the region where the threshold effects become problematic is @xmath180 gev @xcite , which is mostly outside of where even the most liberal cuts in @xmath70 and @xmath40 are made @xcite . from the seminal work of de rjula _ et al . _ @xcite , the appearance of the threshold problem in the analysis of tmcs is attributed to the neglect of dynamical higher twist corrections , both of which scale as powers in @xmath178 . ( in fact , higher order perturbative qcd corrections can also resemble power suppressed contributions at low @xmath40 . ) regardless of their origin , the various power suppressed corrections that are not included in a leading twist calculation can be absorbed into phenomenological functions ; for example , for an unpolarized structure function @xmath181 , @xmath182 where @xmath183 denotes the leading twist contribution including tmcs . the higher twist corrections are sometimes assumed to be multiplicative , with the functions @xmath184 proportional to the leading twist contribution , @xmath185 . possible additional @xmath40 dependence of the higher twist contributions , from radiative @xmath186 corrections , is usually neglected . the leading twist pdfs were found in ref . @xcite to be essentially independent of the tmc prescription adopted , with the ht parameters able to compensate for the variations due to the different tmc formulations . the isospin dependence of the ht corrections was also studied in refs . existing data do not allow for an accurate determination of the @xmath187 corrections @xmath188 , and attempts to include them in global fits produce anomalously small values of @xmath189 and @xmath190 in conjuction with large compensating twist4 and twist6 contributions @xcite , possibly due to overfitting and blurring of the scaling violations . in the polarized case , the greater scarcity of data means that typically one can not afford the luxury of @xmath40 and @xmath70 cuts as stringent as those applied in some of the unpolarized pdf analyses . most global analyses of spin - dependent pdfs therefore include structure function measurements down to @xmath191 gev@xmath170 , where higher twist corrections are believed to be important . the higher twist contributions to @xmath90 and @xmath91 may be treated in an analogous way to the unpolarized @xmath181 structure functions , with the important difference that the twist3 contributions to @xmath91 are not @xmath40-supressed . phenomenological pdf analyses exist which include ht contributions to @xmath90 @xcite , @xmath91 @xcite , and to both functions simultaneously @xcite . while the focus in the present work is on the leading twist pdfs , the higher twist contributions to @xmath90 and @xmath91 are also of intrinsic interest in themselves , containing information , for example , about the correlations of color electric and magnetic fields in the nucleon with the nucleon s spin @xcite . since nucleons bound in a nucleus are not free , the parton distributions @xmath192 in a nucleus @xmath7 deviate from a simple sum of pdfs in the free proton and neutron , @xmath193 , where @xmath194 is the number of protons . this phenomenon is especially relevant at small values of @xmath2 , where nuclear shadowing ( or screening ) effects suppress the nuclear to free isoscalar nucleon ( @xmath195 ) ratio , @xmath196 , and at large @xmath2 , where the effects of fermi motion , nuclear binding , and nucleon off - shellness give rise to the `` nuclear emc effect '' @xcite . in addition , for spin - dependent pdfs , the different polarizations of the bound nucleons and nuclei need to be taken into account . in the nuclear impulse approximation , where scattering is assumed to take place incoherently from partons inside individual nucleons , the pdf in a nucleus can be expressed as a convolution of the pdf in a bound nucleon and a momentum distribution function @xmath197 of nucleons in the nucleus @xcite . coherent rescattering effects involving partons in two or more nucleons give rise to nuclear shadowing corrections to the impulse approximation . in general , at large @xmath40 the pdfs in a nucleus and in a nucleon are related by @xmath198 where the additive term @xmath199 represents nucleon off - shell or relativistic corrections , and @xmath200 parametrizes the shadowing corrections . a similar expression can be written for spin - dependent pdfs . the momentum distribution , or `` smearing function '' , @xmath197 , can be computed from nuclear wave functions , incorporating nuclear binding and fermi motion effects . at @xmath201 the smearing function has a simple probabilistic interpretation in terms of the light - cone momentum fraction of the nucleus carried by the struck nucleon , where @xmath202 and @xmath203 are the four - momenta of the nucleon and nucleus , respectively , and @xmath204 is the nuclear mass . in this case the smearing function is normalized to unity , @xmath205 . at finite @xmath40 , however , the smearing function depends in addition on the parameter @xmath206 , where @xmath207 and @xmath208 are the energy and three - momentum transfer , respectively , which characterizes the deviation from the bjorken limit @xcite . typically , the function @xmath197 is steeply peaked around @xmath209 , becoming broader with increasing mass number @xmath7 as the effects of binding and fermi motion become more important . in the limit of zero binding , @xmath210 , and one recovers the free - nucleon case . this is the usual assumption made in most global pdf analyses . recently , several analyses @xcite have accounted for the nuclear effects by explicitly calculating the corrections from microscopic nuclear models , or attempted to constrain them phenomenologically @xcite . the most straightforward calculation is for the simplest nucleus the deuteron , for which both nonrelativistic and relativistic wave functions are available @xcite , constrained by high - precision nucleon - nucleon scattering data . experiments with deuterium targets play a vital role , in fact , in determining the flavor decomposition of the proton pdfs . traditionally , the standard method for disentangling the @xmath141 and @xmath142 pdfs has been through charged lepton dis , which for a proton target is sensitive at large @xmath2 to the combination @xmath211 . dis from a neutron would constrain @xmath212 ; however , the absence of free neutron targets has necessitated the use of deuterium as effective neutron targets . while the nuclear corrections in the deuteron are typically a few percent , in some regions of kinematics , most notably at large @xmath2 , they can give rise to large uncertainties in the extracted @xmath142 quark pdf in particular . an alternative , purely phenomenological approach @xcite has attempted to constrain the nuclear corrections in deuterium directly from the data . however , without an independent , high - precision measurement of the @xmath142 quark pdf from processes other than inclusive dis , it is difficult to unambiguously separate the effects of the nuclear corrections from uncertainties in the @xmath142 quark pdf , especially at high values @xmath2 . until future experiments ( see ref . @xcite for a discussion ) are able to determine the @xmath142 quark pdf independent of uncertainties in nuclear models , a more practical approach adopted by the cj collaboration @xcite has been to produce a set of global pdfs for a range of nuclear models , corresponding to mild ( cj12min ) , medium ( cj12mid ) , and strong ( cj12max ) nuclear corrections in the deuteron . the current uncertainties in pdfs from nuclear corrections can have important consequences far beyond the fixed - target experiments where they are encountered most directly . in fact , the effects of the nuclear smearing corrections are not suppressed at large @xmath40 , and must be considered at all scales wherever data at @xmath213 are used @xcite . through @xmath40 evolution , pdfs at large @xmath2 and small @xmath40 evolve to lower @xmath2 and higher @xmath40 , so that large-@xmath2 uncertainties in fixed - target experiments can have significant consequences for collider measurements @xcite , examples of which were discussed for selected observables at the tevatron and lhc in ref . @xcite . for neutrino scattering , in order to increase the relatively low rates and obtain sufficient statistics , experiments have often resorted to using heavier nuclear targets , such as iron or lead . this is particularly relevant for determinations of the strange quark pdf , which is typically extracted from opposite - sign dimuon events in @xmath207 and @xmath214 charm production , @xmath215 or @xmath216 @xcite . such extractions are complicated by the presence of nuclear corrections in neutrino structure functions @xcite , as well as effects of the nuclear medium on the charm quark propagation in the final state @xcite . uncertainties in the strange quark pdf can have significant impact on @xmath70 and @xmath194 boson measurements at the lhc , for example , so that understanding of the nuclear effects will have impact far beyond lepton nucleus dis @xcite . in order to systematically study the nuclear dependence of pdfs , without assuming specific relations between pdfs in nuclei and nucleons , several groups have parametrized nuclear pdfs directly @xcite . one approach has been to parametrize the @xmath7 dependence of the initial proton pdf parameters , and then perform a global fit of the available hard scattering data on nuclear targets . this implicitly assumes a smooth @xmath7 dependence for the pdfs , which is reasonable for large @xmath7 , but may break down for light nuclei such as deuterium . an alternative method is to fit to one type of data set and examine how the results differ from those on proton targets . this can be useful for testing whether the nuclear corrections are consistent with those obtained from various data sets . such comparisons have revealed , for instance , a controversy about whether or not the @xmath7 dependence of the neutrino - nucleus data from the nutev collaboration is compatible with that observed in charged lepton dis @xcite . differences between neutrino and electromagnetic nuclear interactions can arise , for instance , from the presence of the parity - odd @xmath82 structure function in @xmath217 scattering , which does not contribute to charged - lepton scattering @xcite . for spin - dependent scattering , the scarcity of data and their larger uncertainties at very small @xmath2 and at high @xmath2 , where nuclear corrections are most prominent , has meant that almost all global analyses have thus far relied exclusively on the effective polarization _ ansatz _ , in which the polarized pdf in the nucleus @xmath218 is related to the polarized pdfs in the proton and neutron as @xmath219 , with @xmath220 the average polarizations of the proton ( neutron ) in the nucleus . in practice , only polarized deuterium and @xmath221he nuclei have been used in dis experiments , in addition to protons . the new global analysis of helicity pdfs by the jam collaboration @xcite is the first systematic attempt to incorporate the effects of nuclear smearing in dis structure functions of the deuteron and @xmath221he . this will be important as future spin - dependent scattering experiments probe the nucleon spin structure at increasingly large values of @xmath2 @xcite . the concept of global fitting for pdfs relies on the formalism outlined in the previous sections . partonic cross sections for various hard scattering processes are convoluted with scale - dependent pdfs to generate results for physical observables which can then be compared to data . the pdfs are parametrized at some convenient input scale @xmath222 and then evolved using the appropriate evolution equations to the scales needed for each calculation . the values of the input parameters are estimated using a @xmath189 minimization technique . all of the global pdf fitting groups use some variation of this technique . a typical parametrization at the input scale @xmath222 for a generic polarized or unpolarized parton ( quark , antiquark or gluon ) distribution function @xmath223 is @xmath224 where @xmath225 represents some smoothly varying function usually chosen as a polynomial in @xmath2 or @xmath226 , although exponential functions are used as well . in addition , there are parametrizations inspired by a statistical model of the nucleon @xcite , as well as those based on neural networks @xcite and self - organizing maps @xcite . some of the parameters in the input distributions can be determined from physical constraints . for example , in the unpolarized case valence quark number is conserved , @xmath227 while the momentum sum rule requires @xmath228 = 1 , \label{eq : mom_sum_rule}\ ] ] where we use the notation @xmath229 , and the number of flavors at the input scale @xmath230 is usually taken to be @xmath231 ( see sec . [ ssec : hq ] ) . in the polarized case the first moments of the charge - conjugation even ( or @xmath11-even ) distributions can be related to octet baryon weak decay constants . for the isovector combination , corresponding to the bjorken sum rule , @xmath232 where @xmath233 is the nucleon axial charge , while for the su(3 ) octet one has @xmath234 where the octet axial charge @xmath235 is extracted from hyperon @xmath29-decays assuming su(3 ) flavor symmetry @xcite . note that the sum rules ( [ eq : sum_rules])([eq : hyperon ] ) are preserved under @xmath40 evolution . while most of the pdf groups use similar procedures and data , one can nevertheless obtain pdf results that can be rather different . some of the reasons for this include the following : * differences in the specific parametrizations and input scale @xmath230 . * differences in data selection , choices of data sets , kinematic cuts , and the specific treatment of the correlated errors of the data . this typically limits the amount of data available at large values of @xmath2 . * differences in the theoretical framework used , including the particular form of the solutions of the rge employed , the treatment of heavy quarks , inclusion of higher twist contributions , target mass effects , and nuclear corrections . these differences should be born in mind when comparing the results from different pdf groups . pdfs are an essential ingredient for producing predictions for processes in high energy experiments . the uncertainties in these predictions depend , in part , on how well determined the pdfs are themselves . it is important to bear in mind that the predictions that are compared to data are convolutions of pdfs with partonic hard scattering cross sections . there are thus three main sources of pdf uncertainties : the fitted data , the partonic cross sections , and the parametrizations used to describe the pdfs . the following list describes the main sources of uncertainty in the determination of pdfs . * the experimental errors on the fitted data can be directly propagated to the fitted pdfs . standard techniques include the hessian @xcite , lagrange @xcite , and monte carlo @xcite methods . * uncertainties due to the use of perturbation theory : these can be estimated to some extent by doing lo , nlo , and nnlo fits , although not all processes are known to nnlo accuracy . * scale dependence : the perturbative predictions depend to some extent on the choices made for the renormalization and factorization scales for each process . these choices will change the results for each process and the fitted pdfs must compensate these changes ( see sec . iii of ref . @xcite for an example of the effects of choosing different scales ) . * choice of the value of the running coupling @xmath236 : some pdf determinations fit @xmath236 while others use the global average value . this is mostly a question of philosophy . on the one hand , since the strong coupling is a parameter of qcd , then there is no freedom to choose a different value for each process . thus , information from data types not included in the global fits ( such as data from @xmath237 processes ) provides valid constraints on the value of the coupling and are included in determining the global average . on the other hand , fitting the value of the strong coupling and comparing it to the global average provides an interesting consistency check on the description of the data provided by qcd . * choice of data sets and kinematic cuts : these choices can affect the fitted pdfs and the user should be aware of these differences . * treatment of heavy quarks : the various schemes that are currently used differ at higher orders and such differences can affect the fitted pdfs . * parametrization dependence : fitted pdfs can differ simply through the choice of the initial parametrizations . extensive efforts are made to choose flexible parametrizations which are well - constrained by data , but there is no control over the behavior outside the kinematic region covered by the data . this can lead to different extrapolations at very large or very small values of @xmath2 . a method for estimating this remaining uncertainty has been suggested in ref . @xcite . of all these sources of error , the easiest to treat is the first the propagation of the experimental errors . the hessian method is described in detail in ref . the elements of the hessian matrix are given by @xmath238 where @xmath239 denotes the @xmath240 pdf parameter . the hessian matrix is generated during the actual minimization procedure and its inverse is the error matrix . the eigenvectors of the error matrix can then be used to define eigenvector parameter sets which , in turn , can be used to calculate error bands for the pdfs or for specific processes . one particular subtlety is that the error bands generally depend on a @xmath189 tolerance . mathematically , one expects the @xmath241 parameter errors to correspond to an increase of @xmath189 by one unit from the minimum value . however , it has been suggested @xcite that inconsistencies between different data sets may require a larger value to be used . this `` @xmath189 tolerance '' varies between groups and allowance must be made for this when comparing the resulting error bands . the lagrange multiplier @xcite method is useful when one wants to determine the pdf error on a specific observable @xmath12 such as the cross section for @xmath70 or @xmath194 production . let @xmath242 denote the @xmath189 for the global data set . then one minimizes the function @xmath243 where @xmath18 denotes the set of pdf parameters and minimization is done for a range of values of @xmath244 . the end result is a relation between @xmath242 and the value @xmath12 of the chosen observable . once one specifies the @xmath189 tolerance , _ i.e. _ , the range of allowable @xmath189 values , this maps out a range of values @xmath245 where @xmath246 is the value of @xmath12 at the global minimum . an alternative to the usual linear propagation of errors ( hessian method ) which can be useful for minima that are not well behaved or defined , is the so called monte carlo method . in order to propagate the experimental errors a number of ` replica ' data sets are generated by using random numbers and the original errors to generate new data points @xcite . these replica data sets are then fitted and the resulting replica pdf sets are treated using standard statistics ; the central values are given by the average over replicas , and the uncertainties by the envelope of predictions . the main motivation for pursuing global fitting as a technique for determining pdfs is that the use of a wide range of data types with different kinematic coverage places many constraints simultaneously on the pdfs . each type of observable depends on a particular linear combination of pdfs or products of pdfs . obtaining the best fit to all of the observables simultaneously has proven to be an efficient method for extracting pdfs . nevertheless , it remains true that specific observables may be sensitive to a specific pdf or combination of pdfs . in this section some examples using simplified lo kinematics will be presented since these can be useful in understanding how pdf errors can be reduced by future measurements . * deep - inelastic scattering * experiments provide direct information on pdfs since both the structure functions and the cross sections depend linearly on the pdfs . the following list summarizes the main dependences . 1 . charged lepton neutral current measurements on a proton target constrain the combination @xmath247 where we have assumed that only single photon exchange contributes ( @xmath194 boson exchange involves a different linear combination of pdfs ) . at large values of @xmath2 the antiquark pdfs become negligible leaving the combination @xmath248 . if one had a neutron target , then the linear combination would be @xmath249 which becomes @xmath250 at large values of @xmath2 . of course , neutron targets are not available , so deuterium is often used for this purpose requiring that nuclear corrections be made in order to extract the neutron target information as discussed in sec . [ ssec : nuclear ] . 2 . charged current neutrino interactions constrains the combinations @xmath251 where the plus ( minus ) sign corresponds to @xmath252 ( @xmath253 ) . the actual linear combinations depend on the type of target used ; high statistics neutrino experiments employ different types of heavy targets for which model - dependent nuclear corrections must be made . note that one can , in principle , isolate specific combinations of @xmath85 or @xmath86 pdfs , depending on the type of target used . a special case is provided by charm production in neutrino and antineutrino scattering as these are proportional to the @xmath254 and @xmath255 pdfs , respectively . the experimental signal is the production of opposite sign muon pairs requiring one to take into account the charm fragmentation and decay as well as nuclear corrections , depending on the type of target employed . the gluon pdf enters in dis at @xmath256 . thus , it is mainly constrained through the @xmath40 dependence of the structure functions and rosenbluth ( or longitudinal / transverse ) separated @xmath257 data ( see sec . [ ssec : gluon ] ) . * vector boson production * ( lepton pairs , @xmath115 , and @xmath116 ) in lo proceed through @xmath258 annihilation . a few key examples are : 1 . lepton pair production in proton proton and proton neutron collisions depend on the combinations @xmath259 where @xmath260 , and the @xmath261indicate contributions from the @xmath254 , @xmath262 and @xmath19 quarks . of course , the @xmath263 cross section is obtained from data from a deuterium target . these cross sections can be used to constrain the @xmath264 ratio , for example . 2 . @xmath115 production constrains products of the form @xmath265 with specific weights given by the appropriate ckm matrix elements , whereas @xmath116 production constrains @xmath258 . for @xmath266 collisions at the tevatron at large values of rapidity , one has approximately , @xmath267 where @xmath268 denotes the proton ( antiproton ) , and the @xmath261 represents contributions from heavier quarks . note that these results are written in terms of proton pdfs . if the rapidity is large and positive , then @xmath269 and one can neglect the antiquark terms so that these cross sections directly constrain the @xmath270 pdfs . due to the missing neutrino from the @xmath70 decays , one can not directly reconstruct the rapidity distributions . what is usually presented is the charged lepton rapidity asymmetry for @xmath115 production . in this case the decay process means that the constraints on the pdfs are less direct , but such measurements still provide useful constraints on the @xmath145 ratio at moderate values of @xmath2 . of particular interest is the model dependent determination of the @xmath70 rapidity asymmetry by the cdf collaboration @xcite , which directly constrains the @xmath145 ratio at large values of @xmath2 since the asymmetry is for the @xmath70 and not the charged lepton from the decay . recently the atlas collaboration has presented an analysis @xcite using @xmath70 decay lepton and @xmath116 rapidity distribution data which shows potential for constraining the strange quark pdf . this analysis makes use of the fact that the @xmath116 cross section receives a significant contribution from @xmath271 annihilation while the @xmath70 cross sections help constrain the @xmath272 and @xmath273 pdfs . the other major class of observables includes inclusive * jet or photon , dijet * , and * photon + jet production*. each of these constrain products of pdfs summed over all flavors . nevertheless , there are certain pdfs that can be constrained by these observables . this follows since the @xmath141 and @xmath142 pdfs are well constrained by the dis data and the vector boson and lepton pair production data provide strong constraints on the @xmath272 and @xmath273 pdfs . hence , these data have the greatest impact on the gluon distribution . 1 . direct photon production in lo has two subprocesses : @xmath274 and @xmath275 . hence , this process was though to be a good candidate for constraining the gluon , especially at fixed target energies where the available @xmath110 range extended out to about 0.6 . however , photons can also be created by bremsstrahlung from the charged quarks often referred to as the fragmentation process . indeed , one can define photon fragmentation functions to help describe this production component . it turns out that at fixed target energies this component receives very large soft gluon corrections , requiring threshold resummation techniques @xcite . this , combined with some apparent disagreements between experimental data sets @xcite meant that photon production has not fulfilled the original promise of constraining the gluon pdf . recently , however , an analysis @xcite suggests that the use of _ isolated _ photon collider data may well help constrain the gluon pdf . the use of isolation cuts reduces the fragmentation contribution and the use of higher energy collider reduces the need for threshold resummation . high-@xmath20 jet production has played a significant role in constraining the gluon pdf . even though the quark and antiquark pdfs are well constrained by other types of data , there are still significant contributions from @xmath276 and @xmath277 subprocesses @xcite . dijet production triple differential cross sections yield more information than single jet cross sections because the rapidity of the second jet is also constrained , thereby helping to constrain the momentum fractions of the pdfs . by tuning the rapidity ranges for the two jets one can explore different regions for these momentum fractions . an example is given in ref . photon + jet production offers similar constraints , but now the subprocesses are weighted by the squared charge of the parton to which the photon couples . as with unpolarized measurements , historically most constraints on spin - dependent pdfs have come from * polarized charged - lepton dis * experiments . 1 . in the one - photon exchange approximation , the difference of inclusive dis cross sections for leptons polarized longitudinally with spin parallel and antiparallel to the target hadron polarization measures @xmath11-even quark combinations @xmath278 . for proton targets , one has the combination @xmath279 while for the neutron the combination would be @xmath280 with contributions from heavy quark polarization expected to be negligible . in practice , polarized @xmath221he targets are usually used as effective sources of polarized neutron , since the neutron carries almost 90% of the spin of @xmath221he , while polarized deuterons , to which the proton and neutron spins contribute equally , are used to provide the isoscalar combination @xmath281 because of the greater sensitivity of the proton @xmath90 structure function measurements to the polarized @xmath141 quark pdf , and uncertainties associated with nuclear corrections when extracting the neutron from @xmath221he or deuterium data , the uncertainty on the @xmath282 pdf is significantly larger at high @xmath2 than on @xmath283 . 2 . at nlo , the polarized gluon distribution @xmath152 also enters in the @xmath90 structure function . the @xmath40 evolution of the flavor singlet contribution to @xmath90 can then be used to constrain @xmath152 , as for the unpolarized gluon pdf discussed above , although the constraints are weaker in the polarized because @xmath284 at small @xmath2 . * semi - inclusive dis * provides additional independent combinations of spin - dependent pdfs that can be used to reconstruct individual quark and antiquark flavors . 1 . semi - inclusive production of hadrons @xmath285 in the current fragmentation region , primarily pions or kaons , is proportional to products of spin - dependent pdfs and quark @xmath286 hadron fragmentation functions , @xmath287 where @xmath288 is the fraction of the quark s energy carried by the hadron @xmath285 . the fragmentation functions @xmath289 can be determined from other reactions , such as inclusive hadron production in @xmath290 annihilation . one can weight particular quark or antiquark flavors by selecting favored ( such as @xmath291 or @xmath292 ) or unfavored ( @xmath293 or @xmath294 ) fragmentation functions for specific hadrons ( in this case @xmath295 ) . information on the polarized strange quark pdf @xmath296 in particular can be obtained from data on @xmath297 production . 2 . the polarized gluon distribution @xmath152 can be constrained by semi - inclusive dis data on charmed or high-@xmath20 hadron production through the photon gluon fusion process . recently the analysis of this process has been performed at nlo , extending earlier lo extractions of @xmath152 . inclusive particle production in * polarized proton proton collisions * provides an additional method of determining spin - dependent sea quark and gluon distributions . 1 . @xmath57 production cross sections for scattering longitudinally polarized protons from unpolarized protons , @xmath148 , depend on products of spin - dependent and spin - averaged pdfs , for example , @xmath298 at large positive or negative rapidities , @xmath132 or @xmath299 , the cross sections ( or asymmetries ) are dominated by a single flavor , while at mid - rapidities both @xmath141 and @xmath142 flavors contribute . inclusive jet or @xmath300 production in double - polarized proton proton scattering , @xmath301 , is sensitive to the polarized gluon pdf . the first evidence for a small , but nonzero @xmath152 was recently observed by the star collaboration at rhic in jet data at @xmath302 gev . direct photon production in polarized @xmath134 scattering has also been suggested @xcite as a means of probing @xmath152 . using the technology outlined in sec . [ sec : qcd ] , a number of global qcd analyses of the world s high - energy scattering data have produced sets of proton pdfs , up to next - to - leading order ( nlo ) or next - to - next - to - leading order ( nnlo ) accuracy . the efforts have indeed been global , with groups in europe and the us in the forefront of the data analyses . the pdfs sets include parametrizations from the mstw @xcite and abm @xcite groups , both of which use standard global fitting methodology ; the nnpdf @xcite collaboration , which uses a newer approach based on neural networks ; the jr @xcite pdfs , which are dynamically generated through @xmath40 evolution from a low @xmath40 input scale ; and the herapdf @xcite group , which includes only data from the h1 and zeus experiments at hera . the us - based efforts have been centered around the cteq collaboration @xcite , which at present involves two derivative analyses of nucleon pdfs , by the ct @xcite and cj ( cteq - jefferson lab ) @xcite groups , as well as the ncteq @xcite analysis of nuclear pdfs . the cj collaboration in particular has focused on developing the methodologies needed for describing data over a broad energy range including the low-@xmath40 and @xmath70 domain @xcite . in this section we summarize the results for unpolarized pdfs from the various pdf fitting groups , discussing their similarities and contrasting their differences . we will focus mostly on the physics issues , rather than on technical aspects of pdf fitting . of course with new data arriving or soon anticipated at the lhc , jefferson lab , and other facilities , the global fitting efforts are constantly evolving , so that the information presented here can only be viewed as a snapshot of the field at the present time . ( red ) , @xmath303 ( blue ) , @xmath304 ( green ) , @xmath305 ( violet ) , @xmath306 ( orange ) and @xmath307 ( black ) pdfs for the cj12mid fit @xcite at @xmath308 gev@xmath170 , shown on logarithmic _ ( left ) _ and linear _ ( right ) _ scales in @xmath2 . note that in the left panel the gluon is scaled by a factor 1/10.,title="fig:",width=302 ] ( red ) , @xmath303 ( blue ) , @xmath304 ( green ) , @xmath305 ( violet ) , @xmath306 ( orange ) and @xmath307 ( black ) pdfs for the cj12mid fit @xcite at @xmath308 gev@xmath170 , shown on logarithmic _ ( left ) _ and linear _ ( right ) _ scales in @xmath2 . note that in the left panel the gluon is scaled by a factor 1/10.,title="fig:",width=272 ] as a typical example of modern pdfs and their uncertainties , fig . [ fig : allpdfs ] shows the @xmath309 , @xmath303 , @xmath304 , @xmath305 , @xmath306 and @xmath307 pdfs ( with the gluon scaled by a factor 1/10 ) for the cj12 fit @xcite ( for the case of moderate nuclear corrections , cj12mid ) at a scale of @xmath308 gev@xmath170 . the general behavior of the pdfs is similar for all the parametrizations @xcite , particularly where sufficient data exist to constrain the distributions in specific regions of @xmath2 and @xmath40 . in regions where data are scarce , such as at very low @xmath2 ( @xmath310 ) or high @xmath2 ( @xmath311 ) for certain observables , or where data on specific pdfs ( such as the strange quark ) are subject to large experimental or theoretical uncertainties , the details of the pdfs can depend strongly on the assumptions adopted in the extrapolations . and @xmath312 , and the @xmath305 and @xmath313 sea distributions , for the available 3-flavor nnlo pdf sets from jr09 @xcite ( red solid line ) , mstw08 @xcite ( blue dashed ) and abm11 @xcite ( green dotted ) , at @xmath308 gev@xmath170.,width=453 ] , but for the singlet - sector distributions @xmath314 , @xmath315 and @xmath316 and the gluon @xmath307 pdf.,width=453 ] some of these differences are evident in figs . [ fig : nspdf ] and [ fig : spdf ] , which illustrate the flavor nonsinglet combinations @xmath317 , and the singlet - sector distributions @xmath318 , respectively , for the nnlo mstw08 @xcite , abm11 @xcite , nnpdf @xcite and jr09 @xcite pdf sets . while the valence @xmath319 and @xmath320 pdfs are fairly well constrained in the intermediate-@xmath2 region , at large @xmath2 there are significant uncertainties on the @xmath320 distribution in particular . the su(2 ) nonsinglet distribution @xmath321 is well determined by data from the drell - yan reaction over the range @xmath322 , but is not constrained at higher @xmath2 . the strange nonsinglet distribution is the most difficult to determine , requiring a combination of neutrino and antineutrino scattering data , which are subject to large experimental and nuclear uncertainties . for the @xmath11-even distributions in fig . [ fig : spdf ] , the @xmath323 and @xmath324 are the dominant quark distributions at large @xmath2 , but the strange quark pdf becomes increasing more important at small @xmath2 . numerically , the gluon distribution dominates all other pdfs at small @xmath2 , but has sizable uncertainties at large @xmath2 values . in the following sections we discuss features of each of the pdfs in more detail . valence quarks give the global properties of the nucleon , such as its charge and baryon number . knowledge of their momentum distributions is important for many reasons , especially at high values of @xmath2 where a single quark carries most of the nucleon s momentum . the large-@xmath2 region is in fact a unique laboratory for studying the nonperturbative flavor and spin dynamics of quarks @xcite , as well as testing predictions from perturbative qcd for the behavior of pdfs in the limit @xmath165 @xcite . reliable determination of pdfs at large @xmath2 is also important for searches for new physics beyond the standard model in collider experiments at the lhc . through perturbative qcd evolution , uncertainties in pdfs from fixed - target experiments at high @xmath2 and low @xmath40 can propagate to larger @xmath40 to affect cross sections at smaller @xmath2 values @xcite . this is especially true for the forward production of particles of mass @xmath155 at large rapidities @xmath111 , whose cross sections are given by products of pdfs with one evaluated at small @xmath325 and the other at large @xmath326 . the production of heavy @xmath327 and @xmath328 bosons , for example , is sensitive to @xmath142 quark pdf uncertainties at high rapidities , exceeding 100% in the @xmath329 channel , which places limits on the accuracy of cross section measurements for masses near the kinematic thresholds @xcite . furthermore , understanding pdfs at large @xmath2 is vital for the analysis of neutrino oscillation experiments @xcite , where one of the most significant uncertainties comes from neutrino nucleus cross sections at the interface of the dis and resonance regions . the growing need to better understand large-@xmath2 pdfs and their uncertainties has been reflected in the greater attention being paid recently to the physics of the large-@xmath2 region , with dedicated experiments planned at jefferson lab following its 12 gev upgrade @xcite , as well as at proposed new facilities such as the lhec @xcite and the electron - ion collider @xcite . it has also been a catalyst for the recent concerted theoretical efforts by the cj collaboration @xcite in their global qcd analyses of pdfs extending into the lower @xmath40 and @xmath168 regions , with the aim of providing better constraints on pdfs at large values of @xmath2 @xcite . and @xmath142 quark distributions from the cj12mid pdfs @xcite , with pdf errors only ( solid , red shaded ) with combined pdf and nuclear uncertainties ( dashed , blue shaded ) , and with a fit excluding all deuterium data ( dot - dashed , green shaded ) , relative to the reference cj12mid set at an arbitrary scale @xmath330 gev@xmath170 . note the different vertical scales for the @xmath141 and @xmath142 quark pdfs.,title="fig:",width=283 ] and @xmath142 quark distributions from the cj12mid pdfs @xcite , with pdf errors only ( solid , red shaded ) with combined pdf and nuclear uncertainties ( dashed , blue shaded ) , and with a fit excluding all deuterium data ( dot - dashed , green shaded ) , relative to the reference cj12mid set at an arbitrary scale @xmath330 gev@xmath170 . note the different vertical scales for the @xmath141 and @xmath142 quark pdfs.,title="fig:",width=283 ] the current constraints on the @xmath141 and @xmath142 quark distributions are illustrated in fig . [ fig : dunod ] for the cj12mid pdf set @xcite , where the uncertainties arising from nuclear corrections and pdf ( experimental ) errors are indicated separately . since the @xmath141 quark pdf is relatively well constrained by the proton dis data at large @xmath2 , the effect of the nuclear uncertainties is minor , increasing the total uncertainty at intermediate @xmath2 by a few percent . for the @xmath142 quark distribution , on the other hand , whose determination at present requires both proton and deuterium dis data , the nuclear correction uncertainties in the deuteron significantly increase the overall error for @xmath331 . however , even though the use of deuterium data introduces the need to confront the problem of nuclear effects and their uncertainties , without these data the error bands on the @xmath142 quark pdf would be even larger over most of the @xmath2 range , as fig . [ fig : dunod ] demonstrates . in addition to accounting for the finite-@xmath40 ( tmc and higher twist ) and nuclear effects which become increasingly important as @xmath2 tends to 1 at fixed @xmath40 , the behavior of pdfs in the @xmath165 limit also depends critically on the functional form used for the @xmath2 dependence . the conventional parametrizations utilized in most global pdf analyses assume a @xmath332 dependence for both the @xmath141 and @xmath142 quark pdfs , as in eq . ( [ eq : param ] ) above , so that in the limit as @xmath165 , the @xmath145 ratio tends either to zero or infinity . @xcite allowed for a more flexible parametrization of the valence @xmath142 quark pdf , in which the @xmath320 distribution receives a small admixture from the @xmath319 pdf , @xmath333 with @xmath19 and @xmath262 as free parameters . provided the @xmath142 quark has a softer momentum dependence ( larger exponent @xmath334 ) than the @xmath141 quark , which agrees with phenomenology , the ratio of the distributions will approach a constant value , @xmath335 , as @xmath165 . a nonzero , finite value in this limit is in fact expected in several different nonperturbative models of nucleon structure @xcite . ratio for the _ cj12min ( dot - dashed ) , cj12mid ( solid ) and cj12max ( dashed ) pdfs @xcite , and _ ( right ) _ cj12mid ( solid ) compared with the ct10 @xcite ( dot - dashed ) , mstw08 @xcite ( dashed ) and abkm09 @xcite ( dot - dot - dashed ) parametrizations at @xmath330 gev@xmath170.,title="fig:",width=283 ] ratio for the _ ( left ) _ cj12min ( dot - dashed ) , cj12mid ( solid ) and cj12max ( dashed ) pdfs @xcite , and _ ( right ) _ cj12mid ( solid ) compared with the ct10 @xcite ( dot - dashed ) , mstw08 @xcite ( dashed ) and abkm09 @xcite ( dot - dot - dashed ) parametrizations at @xmath330 gev@xmath170.,title="fig:",width=283 ] the ratio of the @xmath142 to @xmath141 quark distributions is illustrated in fig . [ fig : dupdf ] for several modern nlo pdf sets , including the cj12 @xcite , ct10 @xcite , mstw08 @xcite ( dashed ) and abkm09 @xcite distributions . as expected , the total uncertainty on the @xmath145 ratio at large @xmath2 increases when different nuclear correction models are considered , with the central values extrapolated to @xmath336 increasing from @xmath337 for the cj12min pdfs , with the minimum nuclear correction model , to @xmath338 for the cj12max set with the maximum nuclear correction model from ref . @xcite . combining all uncertainties , the cj12 analysis found @xmath339 } \pm 0.10 \,\rm{[nucl]}$ ] as @xmath165 , where the first error is from the pdf fits and the second is from the nuclear correction models . in contrast , using the more restrictive , conventional parametrizations , the @xmath145 ratios for the mstw08 and abkm09 distributions tend to zero in the @xmath165 limit , while that for ct10 tends to infinity . this behavior distorts somewhat the error bands , making the mstw08 ratio appear to have an anomalously small uncertainty , while the ct10 result has a much larger one . the relative errors , however , are similar for the mstw08 and ct10 pdfs , while the cj12 and abkm09 uncertainties are reduced at larger @xmath2 because of the additional high-@xmath2 data used in those analyses . unfortunately , the extrapolated values of @xmath145 at @xmath336 span most of the range of available predictions @xcite , so that discrimination between the physical mechanisms that lead to the different @xmath165 behaviors will only be possible with constraints from new experiments that do not involve deuterium data and the nuclear model uncertainties @xcite . inclusive electromagnetic dis probes @xmath11-even combinations of pdfs , @xmath340 , weighted by their squared charges , or in terms of valence and sea quark ( or antiquark ) distributions , @xmath341 . the shapes of the quark valence and sea distributions differ markedly as a function of @xmath2 . as we have seen in the preceding section , the valence quark pdfs dominate at intermediate and large values of @xmath2 , but vanish as @xmath342 for @xmath343 . the sea , in contrast , dominates at small @xmath2 , but is strongly suppressed as @xmath165 . for the light antiquarks @xmath272 and @xmath273 , structure function data with proton targets are sensitive at small @xmath2 to the combination @xmath344 , while deuterium targets ( nuclear corrections aside ) probe the isosinglet combination @xmath345 . under @xmath40 evolution , the singlet quark distribution mixes with the gluon , which leads to the rapid growth in the antiquark pdfs with decreasing @xmath2 . with increasing @xmath40 , the uncertainties on @xmath345 also decrease . inclusive dis measurements can therefore provide important constraints on the behavior of both the @xmath346 and @xmath86 distributions in the regions where their contributions are dominant . however , in the intermediate-@xmath2 region , @xmath347 , where the magnitude of valence and sea quark pdfs is comparable , additional information is required to uniquely determine the distributions individually . here data from neutrino scattering can allow access to the parity - violating @xmath82 structure function , which measures @xmath11-odd combinations of pdfs , @xmath348 . combined with the @xmath11-even pdfs from the parity - conserving @xmath80 and @xmath81 structure functions , one can then uniquely determine the @xmath85 and @xmath86 distributions individually . unfortunately , neutrino scattering cross sections are usually statistics limited compared with electromagnetic data , and with few exceptions , have necessitated the use of nuclear targets such as iron or lead as a means of increasing the rates . this subsequently brings with it the inherent complications associated with accounting for nuclear corrections in relating the nuclear structure functions with those in a free nucleon @xcite . more direct constraints on @xmath86 distributions can be obtained from lepton pair production in inclusive hadron hadron scattering , which involves products of pdfs evaluated at different values of beam ( @xmath112 ) and target ( @xmath143 ) momentum fractions . as discussed in sec . [ sssec : h - h ] , the drell - yan process or @xmath57-boson production in @xmath134 scattering at nonzero rapidity , for example , allows the antiquark distribution at small @xmath112 to be extracted with knowledge of the corresponding quark distribution at @xmath143 . in particular , the drell - yan reaction @xcite has been used by the na51 collaboration at cern @xcite and the e866/nusea collaboration at fermilab @xcite to determine the ratio of @xmath273 to @xmath272 distributions in the proton . naive expectations from perturbative qcd based on @xmath349 splitting suggest that since the masses of the @xmath141 and @xmath142 quarks are very similar , the production of @xmath350 and @xmath351 pairs should be nearly identical , even with the inclusion of higher - order radiative corrections @xcite . the first indication of a significant asymmetry in the proton sea was obtained by the new muon collaboration ( nmc ) @xcite at cern , who performed an accurate measurement of the ratio of dis cross sections for hydrogen and deuterium , from which the gottfried sum @xmath352 was determined @xcite , @xmath353 where charge symmetry is assumed ( see sec . [ ssec : csv ] below ) . for a flavor symmetric su(2 ) proton sea , the prediction for the gottfried sum would be @xmath354 . in contrast , the nmc measurement found @xmath355 @xcite , indicating a strong violation of flavor symmetry in the light antiquark sea , @xmath356 following the suggestion by ellis and stirling @xcite that the light antiquark sea could be more directly probed in the drell - yan process , the na51 experiment used the ratio of dimuon cross sections for @xmath357 and @xmath134 scattering to extract the ratio @xmath358 at an average @xmath359 at a feynman-@xmath2 of @xmath360 . subsequently , the e866/nusea experiment at fermilab measured the @xmath361 ratio at large @xmath121 , where at leading order @xmath362,\ \ \ \ \ x_a \gg x_b . \label{eq : sig_pd_pp}\end{aligned}\ ] ] the results , illustrated in fig . [ fig : dbar_ubar ] , confirmed a significant deviation of the cross section ratio from the naive perturbative qcd expectation , indicating a large asymmetry between @xmath273 and @xmath272 . and @xmath134 scattering as a function of the fractional momentum @xmath143 of the target parton . the data points from the fermilab e866/nusea collaboration @xcite are compared with the cross section ratio calculated using the jr09 @xcite ( red solid line ) , mstw08 @xcite ( blue dashed ) , and abm11 @xcite ( green dotted ) , and nnpdf23 @xcite ( yellow dot - dashed ) pdf parametrizations.,width=453 ] the earliest efforts to explain an enhancement of @xmath273 quarks over @xmath272 in the proton invoked the effects of pauli blocking , which arise from the presence of different numbers of valence @xmath141 and @xmath142 quarks in the proton @xcite . estimates of this effect are very difficult to quantify reliably @xcite , with some calculations @xcite even suggesting that it produces a small excess of @xmath272 over @xmath273 . alternative explanations have focused on the role of chiral symmetry breaking and the pion cloud of the nucleon @xcite . because of the heisenberg uncertainty principle , a component of the proton s wave function involves its dissociation into a nucleon and a virtual pion state . the preferential coupling of a proton to a @xmath363 , with the @xmath364 containing a valence @xmath365 pair , implies a natural mechanism for generating an excess of @xmath273 over @xmath272 . while many models have been proposed to compute this effect quantitatively ( see refs . @xcite for reviews ) , the underlying physics principles are model independent and simply the chiral symmetry properties of qcd @xcite . while most of the modern pdf parametrizations are able to accommodate an enhanced @xmath273 sea over the range @xmath366 , there is some uncertainty in the trend of the @xmath264 ratio at larger @xmath2 . the new drell - yan experiment e906/seaquest at fermilab @xcite will extend the @xmath2 coverage up to @xmath367 , and a proposal at the j - parc facility in japan @xcite would extend the range even further . most pdf analyses have been performed under the assumption of charge symmetry , or independence of interactions under rotations in isospin space . charge symmetry implies that the @xmath141 quark distribution in the proton is identical to the @xmath142 quark distribution in the neutron , and _ vice versa _ , and is believed to be accurate in nature at the @xmath368 level . this symmetry is broken explicitly by light quark mass differences , @xmath369 , as well as by electromagnetic corrections . such effects lead to nonzero values of the `` majority '' and `` minority '' charge symmetry violating asymmetries in the nucleon , defined as @xmath370 respectively , and similarly for the antiquarks @xmath371 . although there is currently no direct evidence for charge symmetry violation ( csv ) in pdfs , it has been predicted in several nonperturbative models for both valence and sea quark pdfs @xcite . in addition , electromagnetic radiative effects can be included in the @xmath40 evolution equations ( sec . [ ssec : q2evol ] ) , in analogy with usual gluon radiation , by adding a term that accounts for the emission of photons by quarks @xcite , @xmath372 here @xmath373 are the new qed splitting functions ( expanded in terms of the electromagnetic coupling @xmath52 ) , and the sum over partons includes also the photon pdf @xmath374 . even with charge symmetric pdfs at the input scale , the different electric couplings of the photon to @xmath141 and @xmath142 quarks in the modified evolution equations will induce nonzero csv distributions ( [ eq : csv_u ] ) and ( [ eq : csv_d ] ) at a higher scale . the valence quark asymmetries resulting from the qed - modified @xmath40 evolution ( [ eq : qeddglap ] ) are illustrated in fig . [ fig : csv_ssbar ] at a scale @xmath375 gev@xmath170 . the perturbatively generated csv distributions from glck _ et al . _ @xcite are calculated with an average isoscalar quark mass @xmath376 mev , with the qed radiative corrections computed after fixing the qcd evolution effects . the majority valence @xmath377 distribution is negative at intermediate @xmath2 values , and is of opposite sign to the minority valence distribution @xmath378 , which is smaller in magnitude . qualitatively , these features are similar to the nonperturbative csv valence asymmetries computed in the bag model @xcite , which arise from differences between the spectator @xmath379 and @xmath380 diquark masses , as well as the proton neutron mass difference . the mrstqed @xcite analysis of csv was similar to that in ref . @xcite , but involved unequal photon pdfs in the proton and neutron resulting from qed radiation with quark masses @xmath381 mev . since the photon now also contributes to the momentum sum rule , eq . ( [ eq : mom_sum_rule ] ) , the csv photon distributions induce charge symmetry violation in the valence quark pdfs , assuming the sea and gluon are charge symmetric . despite the different assumptions , the resulting distributions turn out to be similar to those in fig . [ fig : csv_ssbar ] . and `` minority '' @xmath378 valence quark distributions for the radiative qed fit of ref . @xcite at @xmath375 gev@xmath170 , compared with the bag model estimates have been taken from ref . _ ( right ) _ strange antistrange asymmetry at @xmath382 gev@xmath170 ( appropriate for the nutev experiment ) for the asymmetric gjr08 @xcite ( red solid ) and the mstw2008 @xcite ( blue dashed ) pdfs.,title="fig:",width=253 ] and `` minority '' @xmath378 valence quark distributions for the radiative qed fit of ref . @xcite at @xmath375 gev@xmath170 , compared with the bag model estimates have been taken from ref . _ ( right ) _ strange antistrange asymmetry at @xmath382 gev@xmath170 ( appropriate for the nutev experiment ) for the asymmetric gjr08 @xcite ( red solid ) and the mstw2008 @xcite ( blue dashed ) pdfs.,title="fig:",width=364 ] while the overall magnitude of the possible csv effects is rather small , they can nevertheless play an important role in understanding the discrepancy between the nutev measurement of @xmath383 @xcite and the standard model value @xcite . the nutev collaboration extracted the paschos - wolfenstein ( pw ) ratio @xcite of total neutral to charged current cross sections , involving the difference between @xmath207 and @xmath384 cross sections . in addition to assuming small uncertainties on the nuclear corrections necessary for translating the measured iron target data to an isoscalar nucleon target , the analysis also neglected any csv effects . including csv corrections , the experimental pw ratio would receive a contribution proportional to @xmath385 @xcite . depending on the sign of the csv distributions , the asymmetry can therefore either remove some of the discrepancy with the nutev result , or exacerbate it . the signs and magnitude of the predicted perturbative qed and nonperturbative quark model asymmetries in fig . [ fig : csv_ssbar ] would in fact remove approximately 2/3 of the discrepancy . an additional correction to the pw ratio can arise from differences between strange and antistrange distributions in the nucleon , @xmath386 , which we discuss in the next section . traditionally the strange quark distribution has been more difficult to determine experimentally than the non - strange sea . in practice one has often assumed that the @xmath254 and @xmath255 pdfs ( which are often assumed to be identical ) are proportional to the @xmath345 distributions , @xmath387 with @xmath388 ranging between @xmath389 and 0.5 . the @xmath254 and @xmath255 distributions can be directly determined , however , through the measurement of dimuon pairs in neutrino dis , which are produced through the charged current by scattering from strange quarks with subsequent generation of charm . in particular , while neutrino scattering produces charmed hadrons ( @xmath390 ) , antineutrino scattering results in anticharmed hadrons ( @xmath391 ) . the semileptonic decay of the charmed hadrons ( such as @xmath392 and @xmath393 mesons ) then yields @xmath394 pairs along with the associated hadronic debris . because of the low rates involved in neutrino scattering experiments , in order to increase the statistics these have typically made use of heavy nuclear targets , such as iron . since the nuclear effects for neutrino and charged lepton scattering are in general different @xcite , the nuclear corrections introduce an additional source of uncertainty in extractions of the strange quark pdf . a number of nonperturbative models have in fact predicted an asymmetry between the @xmath254 and @xmath255 distributions in the nucleon , which can be probed by taking the difference between the @xmath207 and @xmath384 cross sections . for example , in analogy with the pion cloud models invoked to explain the @xmath273 excess over @xmath272 in the proton in sec . [ ssec : lightsea ] , the dissociation of a nucleon to a hyperon and a virtual kaon naturally produces unequal @xmath254 and @xmath255 distributions , although the magnitude and even the sign is difficult to determine unambiguously @xcite . in addition , an asymmetry can also be generated through perturbative qcd evolution at nnlo because of unequal valence @xmath141 and @xmath142 distributions @xcite . the ccfr @xcite and nutev @xcite collaborations at fermilab have collected the highest statistics data on inclusive charm production in @xmath207 and @xmath384 scattering on an iron target , enabling the most direct constraints to be placed on the strange quark pdfs as a function of @xmath2 . the latter in particular was the first analysis using a complete nlo description of charm production . the cross section for ( anti)neutrino charm production was computed to nlo in refs . @xcite within the ffns , in which , apart from the gluon , only the light quark flavors ( @xmath395 ) are included as massless partons ( see secs . [ ssec : hq ] and [ ssec : hqresults ] ) . the fully differential nlo calculation , including acceptance corrections , was performed in ref . ( note , however , that the use of the expressions in @xcite together with vfns distributions , as implemented in the nutev analysis @xcite , is strictly not consistent . ) comparison of the @xmath207 and @xmath384 events found a preferred fit for the @xmath396 distribution that peaked at @xmath397 , with the compensating negative contribution required by the sum rule @xmath398 to be restricted to the unmeasured region at @xmath399 @xcite . for the first moment , the value obtained at @xmath382 gev@xmath170 was @xmath400 , where the errors are statistical , systematic , and external . the latter includes a dominant contribution from the charm semileptonic branching ratio uncertainty , as well as from nuclear corrections @xcite . the strangeness asymmetry in the nucleon is shown in fig . [ fig : csv_ssbar ] at @xmath382 gev@xmath170 , as relevant for the nutev experiment , for several different pdf parametrizations . the mstw08 @xcite strange asymmetry is fitted to the ccfr and nutev data assuming the form @xmath401 where the factor @xmath402 ensures zero net strangeness in the nucleon . the strange asymmetry in ref . @xcite is generated radiatively from the boundary condition @xmath403 at @xmath404 gev@xmath170 , but allowing @xmath405 to have the form in eq . ( [ eq : sminus ] ) . the resulting @xmath405 asymmetry at the experimental scale is similar to the mstw08 fit , and to the phenomenological extraction in ref . @xcite , within the currently sizable uncertainties . the values for the momentum weighted asymmetry are the order @xmath406 for the mswt08 fit @xcite and @xmath407 for the dynamically generated pdfs @xcite at scales @xmath408 gev@xmath170 . the cteq collaboration found a strange asymmetry in the range @xmath409 @xcite . these values have the correct sign and similar magnitude to that required to reconcile the nutev @xmath383 measurement with the standard model value @xcite , once the charge symmetry violating effects in sec . [ ssec : csv ] are also included @xcite . finally , a recent measurement by the atlas collaboration @xcite of inclusive @xmath57 and @xmath194 boson production in @xmath134 collisions at the lhc found a significantly larger value of the strange to non - strange sea ratio in eq . ( [ eq : kappa ] ) , @xmath410 at @xmath411 for @xmath412 gev@xmath170 , than in previous analyses . a reanalysis of the atlas and hera data by the nnpdf group @xcite found , however , that the uncertainty on the extracted @xmath388 was significantly underestimated , and that a more complete global analysis gives @xmath388 values consistent with the traditional values . heavy quark production contributes considerably to inclusive dis ; for example , the charm ( bottom ) contributions to the dominant @xmath81 structure function constitute up to 30% ( 3% ) of the total in the small@xmath2 region . besides these contributions , heavy quark electroproduction is directly accessed experimentally in semi inclusive dis , typically by detecting charm ( bottom ) mesons in the final state . because the main production mechanism is photon gluon fusion , data on this process provide valuable constraints on the gluon pdf in the nucleon . furthermore , they provide an appropriate context for testing the different approaches to these calculations that have been adopted in various pdf analyses ( see sec . [ ssec : hq ] ) . since it is by far the most relevant , we focus the discussion on the charm flavor , although similar considerations often also apply to bottom ( top contributions to this process , on the other hand , are negligible at accessible kinematics ) . the most relevant data on heavy quark electroproduction are the ( h1 + zeus ) combination of hera measurements @xcite . a comparison of these data with ffns calculations at lo , nlo and nnlo * ( see sec . [ ssec : hq ] ) in fig . [ fig : charm ] ( left ) shows excellent agreement between the ffns calculations . in fact , the experimental analysis of the hera data is based primarily on the ffns fully differential ( exclusive ) calculations of ref . @xcite . despite this fact , the constructs in the gm - vfnss are often motivated by arguing that pseudo - mass - divergences ( non - collinear logarithms of the form @xmath413 ) which appear in the wilson coefficients of the massive calculations need to be resummed . however , the ffns results are notably stable even for very large values of @xmath414 @xcite , in particular at the largest @xmath40 values accessible at hera , suggesting little need to resum these supposedly `` large logarithms '' in the context of global pdf analyses . finally , in addition to the perturbative generation of charm , it is also possible for charm to be produced through nonperturbative mechanisms , such as those discussed in refs . @xcite ( and references therein ) . first postulated to account for some early charm production data in hadronic reactions , these `` intrinsic '' charm contributions share several characteristic features , even though their details depend somewhat on the models . in particular , they are typically valence - like , with significantly harder @xmath2 distributions than those generated perturbatively , and can produce large asymmetries between the @xmath262 and @xmath415 distributions . measurements of the charm structure function @xmath416 by emc @xcite , especially the two data points at the highest @xmath40 in fig . [ fig : charm ] ( right ) , have been interpreted as indicating a nonzero intrinsic charm contribution at large @xmath2 . the evidence is not conclusive , however , in view of the disagreement evident in fig . [ fig : charm ] ( right ) between the emc data and theoretical predictions at smaller @xmath2 , where the predictions are in agreement with the more recent hera results in fig . [ fig : charm ] ( left ) . the current limits on the normalization of the nonperturbative charm quark distribution range from @xmath417 to 2% , depending on which data sets are fitted and which models of intrinsic charm considered @xcite . although current global pdf analyses do not require an intrinsic charm component , the existence of nonperturbative charm remains an intriguing issue which can be addressed by high - precision measurements of charm production at future facilities such as j - parc @xcite , gsi - fair @xcite , jefferson lab at 12 gev @xcite , after@cern @xcite , or an electron - ion collider @xcite . in addition to the heavy quark production discussed above , there are three primary sources of information on the gluon distribution : the @xmath40 dependence of the dis structure function @xmath81 at low values of @xmath2 ; the longitudinal structure function @xmath257 at all @xmath2 values ; and jet ( both at hera and at hadron colliders ) and photon production cross sections at moderate to high values of @xmath2 . furthermore , the momentum sum rule ( [ eq : mom_sum_rule ] ) provides an additional constraint on the gluon pdf . since the normalizations of the valence pdfs are fixed by the flavor sum rules in eq . ( [ eq : sum_rules ] ) and the contributions of the light antiquarks are relatively well determined by dis and lepton pair production data , the momentum sum rule effectively constrains the gluon contribution . the effect is an anticorrelation between the low- and high-@xmath2 behavior of the gluon pdf : a decrease ( increase ) of the gluon distribution at low @xmath2 leads to an increase ( decrease ) at high values of @xmath2 . interestingly , this can result in more constrained gluon distributions in the medium-@xmath2 region ( see fig . [ fig : gluon ] ) . gev@xmath170.,width=377 ] to understand the relationship between the scale dependence of @xmath81 and the gluon pdf , consider eq . ( [ eq : rge ] ) for the scale dependence of the quark pdf . multiply both sides by @xmath2 and by the squared parton charge and sum over the active flavors . in this way one derives ( using the lo expression for @xmath81 ) , @xmath418 \label{eq : f2q2}\ ] ] where @xmath419 and @xmath420 are the quark quark and quark gluon splitting functions . at small values of @xmath2 the gluon term dominates the integrand so that the scale dependence of @xmath81 places a constraint on the gluon pdf . a second constraint on the gluon from dis comes from the longitudinal structure function @xmath257 , which is defined as @xmath421 in lowest order , where only quarks contribute to @xmath80 and @xmath81 , one has @xmath422 so that @xmath423 . the first nonzero contribution to @xmath257 starts at @xmath256 where , at small @xmath2 , there is a significant contribution from the subprocess @xmath424 , which again imposes strong constraints on the gluon pdf at small @xmath2 . at larger @xmath2 values the valence quark contributions increase , the gluon pdf decreases , and the subprocess @xmath425 dominates . however , since the same combinations of quark distributions are well determined from @xmath81 , where they enter with one power of @xmath34 less and the gluon does not contribute , precision measurements of the longitudinal structure function provide valuable information on the gluon pdf at large @xmath2 @xcite . note also that the gluon pdf is accompanied by the same powers of @xmath34 in all dis structure functions , starting with @xmath426 . this results in a correlation between the value of @xmath34 obtained in global pdf analysis and the size and shape of the gluon distribution , with larger @xmath34 leading to a smaller gluon pdf in the small-@xmath2 region and ( via the momentum sum rule correlation ) a larger gluon pdf at large @xmath2 . this can be seen in fig . [ fig : gluon ] , where the jr09 @xcite and abm11 @xcite analyses obtain @xmath34 values and small-@xmath2 gluon pdfs of similar size , while the mstw08 fit @xcite finds a larger @xmath34 value and a smaller gluon pdf in this region ( which even turns negative ) . both possibilities describe well the inclusive cross sections measurements at hera . this ambiguity can be reduced by using processes with a different leading @xmath34 power , typically jet production at the tevatron , which , as noted in sec . [ sssec : unpol ] , can provide constraints on the gluon pdf . to understand this , recall that the calculation of the jet cross section involves a sum over @xmath427 , @xmath276 and @xmath277 induced subprocesses where `` @xmath85 '' stands for any flavor of quark or antiquark . at low values of jet @xmath110 , the @xmath277 subprocesses dominate , at intermediate values all three contribute , and eventually as @xmath110 approaches one only the @xmath427 terms survive see fig . 3 in ref . @xcite , for example . jet data at the tevatron extend to @xmath110 values of approximately 0.5 , so these data provide some constraints on the gluon pdf , at least in the mid-@xmath2 range where the @xmath276 subprocesses make substantial contributions . as discussed in ref . @xcite , isolated photon production at collider energies has the potential to tighten the error bands in the @xmath2 range below about 0.2 . it remains a challenge to find processes that will constrain the gluon beyond @xmath428 . the fundamental problem is that the valence pdfs here are larger than the gluon pdf . most hard processes that receive contributions from @xmath276 subprocesses also receive contributions from @xmath427 processes , which reduce the sensitivity to the gluon pdf . one counterexample is provided by direct photon production in @xmath134 collisions . the two dominant subprocesses are initiated by @xmath258 and @xmath276 initial states . at large values of @xmath110 the fragmentation process is suppressed ; it can be further reduced by photon isolation cuts . the single photon cross section involves an integration over the awayside jet rapidity which tends to smear out the @xmath2 values of the contributing pdfs . there are regions where one @xmath2 is small and one is large one might think that this could constrain the large-@xmath2 gluon . however , the favored configuration will be where the gluon or @xmath86 is at small values of @xmath2 while the valence quark is at the larger values . instead , consider the photon + jet cross section with equal and opposite photon and jet rapidities , in which case both momentum fractions will be equal to @xmath429 using lo kinematics . going to large values of @xmath110 and @xmath430 would then yield information on the gluon at large @xmath2 , since the @xmath86 contribution is much smaller there than that of the gluon . this would also be the case for lepton pair @xcite and vector boson production in @xmath134 collisions , provided that one considers @xmath431 in order for the parton momentum fraction to be large . the analysis in ref . @xcite suggests that photon + jet data with increased statistics from the lhc may help constrain the gluon and light quark pdfs . another source of information on the gluon pdf is top quark pair production , for which the lowest order subprocess is @xmath432 . in ref . @xcite the constraints on the gluon pdf provided by @xmath433 data from the tevatron and lhc were examined , with the results indicating that these data place strong constraints on the large-@xmath2 gluon pdf . also of note is the analysis of ratios and double ratios of cross sections measured at different lhc energies @xcite . such ratios have reduced systematic errors and have the potential to provide strong constraints on large-@xmath2 pdfs . in addition to the issues discussed above , an obstacle for the determination of the gluon distribution , and the highly correlated @xmath34 values from pdf analysis , is that the calculations for the relevant processes are not available beyond nlo , with the important exceptions of the ( light quark contributions to the ) inclusive dis structure functions and drell - yan cross sections . for example , the relative uncertainties of the jr09 gluon pdf in fig . [ fig : gluon ] are below about 10% for @xmath434 , and increase as one proceeds to higher values of @xmath2 , where the constraints of the @xmath81 structure function from fixed target experiments used in that fit are weaker . the uncertainty at large @xmath2 could in principle be reduced by including jet data . however , this would also affect the central values obtained for the gluon pdf , changing them in the relevant region by an amount proportional to the missing nnlo corrections , which in the case of jet production are expected to be large @xcite . global nnlo fits which include jet data , such as the mstw08 @xcite parametrization in fig . [ fig : gluon ] , generally find larger gluon pdfs in the high-@xmath2 region and correspondingly also larger values of @xmath34 . yet another factor which can alter the pdfs and the value of @xmath34 obtained in a particular analysis are the kinematic cuts applied ; of particular relevance are the cuts on the dis structure function data and the related treatment of contributions beyond leading twist @xcite . although data selection ( data sets and cuts ) bring about certain arbitrariness in global pdf analysis , it can not account for all the existing differences between the various pdf sets , which are also affected by theoretical issues such as heavy quark schemes ( sec . [ ssec : hq ] ) and the particular solutions of the rge equations used . even within a given framework there are uncertainties due to the inability of the estimation procedure to find the optimal solution , for example , shortcomings of the parametrization to reproduce the optimal shape of the distributions , and the statistical estimation procedure . these effects , referred in general as ` procedural bias ' , induce a dependence of the results on the choice of the scale at which the input distributions are parametrized , which in principle should not depend on these choices . for particularly sensitive quantities such as the gluon distribution and @xmath34 , this uncertainty can be comparable to the experimental uncertainty @xcite , as can be seen in fig . [ fig : alphas ] . values obtained for different parametrizations in ref . the band indicates the uncertainties due to propagation of the experimental errors for a particular parametrization ( nnlo20).,width=377 ] another example of this is the differences between `` standard '' distributions , generated from input pdfs at values for the input scale @xmath435 gev@xmath170 , and the so called dynamical distributions @xcite , for which input scale is chosen to be @xmath436 gev@xmath170 . at these low scales the input distributions naturally tend to valence like ( positive definite ) shapes @xcite that vanish in the small@xmath2 limit , not only for valence but also for the sea quark and gluon input densities . it was shown in refs . @xcite , where also `` standard '' distributions were generated , that this implies that the dynamical distributions at small @xmath2 are more restricted and have smaller uncertainties than their standard counterparts . furthermore , this procedure typically results in smaller @xmath34 values and steeper gluons in the small@xmath2 region , although both approaches provide comparable descriptions of data , in particular , of the inclusive dis cross sections from hera . an alternative approach to fitting @xmath34 as part of the global fitting is to use the world average value for @xmath236 , thereby avoiding the correlations during the fitting process . since @xmath34 is a parameter of the qcd lagrangian , it should be the same for all processes , not just those in the global fit . of course , interesting information could be obtained by determining the pdfs both with a fixed value of @xmath236 and with a fitted value . comparing the results could indicate which processes in the global fit are chiefly responsible for any differences . the moments of leading twist pdfs can be evaluated from first principles using lattice qcd . within the operator product expansion , the moments of pdfs are related to matrix elements of local operators between hadron states , which can be computed numerically on the lattice . the lattice pdf moments can be related to the quark and antiquark distributions in the nucleon as @xmath437 , \label{eq : mom_unp } \\ \langle x^n \rangle_{\delta q } & = & \int_0 ^ 1 dx\ , x^n\ [ \delta q(x)+(-1)^n \delta \bar q(x ) ] , \label{eq : mom:_pol}\end{aligned}\ ] ] for the unpolarized and helicity distributions ( see sec . [ sec : spin ] ) , respectively . the lattice moments alternate between the total ( or @xmath11-even ) @xmath438 and valence ( @xmath11-odd ) @xmath348 distributions , depending on whether @xmath15 is even or odd . while reconstructing the @xmath2 dependence of the pdfs from a finite number of calculated moments is challenging @xcite , comparison of the lattice moments with the phenomenological pdf moments can provide important tests of lattice techniques and provide constraints on specific pdfs which may be difficult to measure experimentally . because of the discretization of space - time on the lattice , the rotational symmetry of the continuum is broken to the hyper - cubic subgroup , which introduces mixing with lower dimensional operators under renormalization . as a result , only the lowest few moments , corresponding to @xmath439 , can at present be reliably computed . as well as the usual approximations of working at a finite lattice spacing @xmath18 , in a finite lattice volume @xmath440 , and at unphysically large values of the quark mass , simulations usually include contributions only from `` connected '' diagrams , representing operator insertions on quark lines connected to the source . meaningful comparisons of lattice data can therefore only be made with nonsinglet combinations , such as @xmath441 , where the `` disconnected '' contributions cancel . significant progress has been made in recent years in the computation of matrix elements of twist - two operators on the lattice , using dynamical ( unquenched ) configurations and quark ( or pion , by the gell - mann oaks renner relation ) masses approaching the physical limit . unfortunately , the results for the lowest nontrivial ( @xmath442 ) nonsinglet moment of the unpolarized distributions , @xmath443 , lie systematically in the range @xmath444 at a scale @xmath445 gev@xmath170 , which is some 40% above the phenomenological pdf moments of @xmath446 @xcite . a smaller , but persistent 10% overestimate of the lowest ( @xmath447 ) moment of the nonsinglet spin - dependent distribution , @xmath448 , has posed a serious challenge to lattice simulations @xcite . it has been suggested @xcite that the discrepancy between the calculated and experimental pdf moments could be resolved by the nontrivial chiral behavior of the moments , when extrapolating the results from the lowest @xmath449 values at which lattice data exist to the physical mass . verification of this behavior will require lattice simulations at very small quark masses , as well as at large enough volumes to fully include the physics of the pion cloud @xcite ( see also sec . [ ssec : lightsea ] ) . while it may take some time before the lattice parameters can become sufficiently close to those in the physical realm , such comparisons will continue to reflect the important synergy between lattice qcd and pdf phenomenology . considerable progress has been made in understanding the spin structure of the nucleon since the first precision polarized dis experiments at cern in the late 1980s @xcite indicated an anomalously small fraction of the proton spin carried by quarks . a rich program of spin - dependent inclusive and semi - inclusive dis , as well as polarized proton proton scattering experiments has followed , vastly improving our knowledge of spin - dependent pdfs of the nucleon over the last two decades ( for a recent review , see ref . @xcite and references therein ) . while the spin - dependent data have not been as abundant as those available for constraining spin - averaged pdfs , the existing empirical information from polarized lepton and hadron facilities at cern @xcite , slac @xcite , desy @xcite , jefferson lab @xcite and rhic @xcite has enabled several dedicated global qcd analyses of spin - dependent parton distributions to be performed . these include the nlo analyses from the more established dssv @xcite group , the european - based lss @xcite and bb @xcite groups , as well as the japanese asymmetry analysis collaboration ( aac ) @xcite , using the standard global fitting methodology outlined in sec . [ sec : qcd ] . more recent efforts have been made by the nnpdf collaboration @xcite , extending the neural network approach to the polarized sector , and the jefferson lab angular momentum ( jam ) collaboration @xcite , which has focused on utilizing data over a large kinematic range that includes the large-@xmath2 and low-@xmath40 regions , with a simultaneous fit of unpolarized pdfs . and @xmath450 , the antiquark @xmath451 , @xmath452 and @xmath453 , and polarized gluon @xmath454 distributions at @xmath191 gev@xmath170 , for the dssv09 @xcite , lss10 @xcite , bb10 @xcite , aac09 @xcite , and jam13 @xcite pdf sets.,width=548 ] representative examples of the spin - dependent pdfs are shown in fig . [ fig : spinpdfs ] , illustrating the total ( or @xmath11-even ) @xmath455 and @xmath456 pdfs , as well as the sea quark @xmath457 , @xmath458 and @xmath459 and polarized gluon @xmath152 distributions , for the dssv09 @xcite , lss10 @xcite , bb10 @xcite , aac09 @xcite , and jam13 @xcite pdf sets at @xmath191 gev@xmath170 . the bb10 @xcite and jam13 @xcite analyses are based on inclusive dis data only , while the lss10 @xcite fit includes also semi - inclusive dis data . the dssv09 pdfs are constrained in addition by polarized @xmath134 scattering data . the latest aac analysis @xcite is based on inclusive dis data and a @xmath297-factor approximation for the nlo corrections for the @xmath134 data . as for the unpolarized case in fig . [ fig : nspdf ] , the @xmath455 distribution is the best constrained polarized pdf , mostly by measurements of the proton @xmath90 structure function over a relatively broad range of @xmath2 . the corresponding @xmath456 distribution , which has the opposite sign , is smaller in magnitude compared with the @xmath455 with somewhat larger uncertainties , especially at larger @xmath2 values . the uncertainty band for the dssv09 parametrization of @xmath455 and @xmath456 is smaller than the variation between the different pdf sets , which reflects the fact that the systematic uncertainties associated with the choice of data sets and parametrization assumptions , as well as other theoretical inputs , are currently larger than the experimental errors . the polarization of the sea is considerably smaller , and more strongly dependent upon assumptions about the flavor dependence in the analysis of semi - inclusive dis data . at present there is no conclusive experimental evidence of a nonzero light quark sea , @xmath457 and @xmath458 , although there is a slight trend towards a more negatively polarized @xmath273 distribution than @xmath272 , with a positive @xmath460 . the polarization of the strange sea is also very small , in contrast to suggestions from the early spin - dependent dis data analyses of a large negative @xmath459 . the polarized gluon distribution is essentially unconstrained by existing inclusive and semi - inclusive dis data , and almost all information on @xmath152 comes from measurements of @xmath461 production in semi - inclusive dis @xcite , and inclusive pion and jet production in polarized @xmath134 scattering @xcite . the data are consistent with a small value of @xmath462 , consistent with zero , although new measurements from rhic have the promise of resolving a small nonzero distribution . at large values of @xmath2 , the spin - dependent pdfs are even more sensitive to the quark - gluon dynamics responsible for spin - flavor symmetry breaking than the spin - averaged valence quark pdfs discussed in sec . [ ssec : valence ] . however , while considerable progress has been made in determining spin - dependent pdfs over the last two decades , especially in the small-@xmath2 region , relatively little attention has been paid to structure function measurements at large @xmath2 . the dearth of data in the valence region is especially striking for the neutron , where , with the exception of several data points from jefferson lab hall a @xcite for the polarization asymmetry @xmath463 extending to @xmath464 , there are essentially no constraints for @xmath465 . on the other hand , there are a number of predictions for the behavior of polarized pdfs and ratios of polarized to unpolarized pdfs in the limit as @xmath165 that differ even in sign . at large @xmath2 there are dramatically different predictions for the valence @xmath283 and @xmath282 distributions from nonperturbative and perturbative calculations . in the simple su(6 ) spin - flavor symmetric quark model , the equal probabilities of the spin-0 and spin-1 spectator diquark configurations lead to the simple relation @xmath466 , and corresponding polarized to unpolarized ratios @xmath467 and @xmath468 . in this limit the polarization asymmetry of the proton , which at lo can be written @xmath469 is predicted to be @xmath470 , while that of the neutron , @xmath471 would vanish , @xmath472 . while spin - flavor symmetry is of course broken in nature ( in some cases badly ) , with the unpolarized @xmath142 quark distribution significantly softer than the @xmath141 , for example , the details of this breaking can affect the spin - dependent pdfs in strikingly different ways . the mechanism of spin-1 diquark suppression , which leads to the vanishing of the @xmath145 ratio as @xmath165 , also predicts that @xmath473 in this limit , while @xmath474 remains unchanged from the su(6 ) expectation . arguments based on helicity conservation in perturbative qcd , on the other hand , predict that for valence quarks in the ground state of the nucleon the scattering at large @xmath2 is predominantly from configurations in which the quark is coupled to a diquark of zero helicity @xcite . in this case the pdfs associated with a quark polarized in the same direction as the nucleon are favored over the helicity - antialigned configurations , @xmath475 as @xmath165 . this model leads to the expectation that @xmath473 and @xmath476 in the @xmath165 limit , and consequently a rapid approach to unity of both the proton and neutron @xmath92 asymmetries . an additional feature of the helicity conservation model is that it correlates the @xmath165 behavior of the pdfs with the large-@xmath40 behavior of elastic form factors @xcite , which can be measured in elastic lepton nucleon scattering experiments at jefferson lab and elsewhere @xcite . this is also closely related to the notion of quark - hadron duality , which connects averages over structure functions in the resonance region at low @xmath70 with those extrapolated from the deep - inelastic continuum at higher @xmath70 ( lower @xmath2 ) @xcite . since the existing dis data are for the most part consistent with a vanishingly small neutron asymmetry at @xmath477 , a very dramatic change in @xmath463 would be expected at larger @xmath2 . an even more rapid transition at high @xmath2 may occur if quark orbital angular momentum plays an important role in nucleon structure @xcite . and @xmath478 distributions and their uncertainties for the jam13 pdf set @xcite at @xmath191 gev@xmath170 , illustrating the effects of the nuclear smearing ( green dotted ) , target mass ( blue dashed ) and higher twist ( red solid ) corrections , relative to the reference fit ( yellow dot - dashed ) . note that the yellow and green bands overlap for @xmath479 and the green and blue bands overlap for @xmath478 . _ ( right ) _ corresponding ratios of polarized to unpolarized @xmath480 and @xmath481 distributions.,title="fig:",width=302 ] and @xmath478 distributions and their uncertainties for the jam13 pdf set @xcite at @xmath191 gev@xmath170 , illustrating the effects of the nuclear smearing ( green dotted ) , target mass ( blue dashed ) and higher twist ( red solid ) corrections , relative to the reference fit ( yellow dot - dashed ) . note that the yellow and green bands overlap for @xmath479 and the green and blue bands overlap for @xmath478 . _ ( right ) _ corresponding ratios of polarized to unpolarized @xmath480 and @xmath481 distributions.,title="fig:",width=302 ] experimentally , the extraction of leading twist spin - dependent pdfs at high @xmath2 faces similar challenges to those discussed in sec . [ ssec : valence ] for the unpolarized distributions . in particular , information on the structure of the neutron is typically obtained from experiments involving polarized deuterium or @xmath221he nuclei , requiring the nuclear corrections to be understood . analysis of high-@xmath2 data also necessitates careful treatment of finite-@xmath40 corrections such as target mass and higher twist contributions . the effects of these corrections are illustrated in fig . [ fig : pol_val ] , which shows the total @xmath455 and @xmath456 pdfs from the jam13 analysis @xcite for various fits with or without the finite-@xmath40 and nuclear corrections . at small values of @xmath2 their effects are negligible ; however , at @xmath482 they can give up @xmath483 corrections for @xmath455 and more @xmath177 corrections for @xmath456 ( the relative correction can be even larger at @xmath484 , although the pdfs are not constrained in this region ) . the same effects are more clearly illustrated through the polarization ratios @xmath480 and @xmath481 in fig . [ fig : pol_val ] , where the unpolarized distributions are fitted simultaneously within the same analysis as the polarized @xcite . this in principle eliminates any bias arising from the use of spin - averaged pdfs taken from analyses performed under different sets of assumptions . in the intermediate-@xmath2 region the ratios for both @xmath141 and @xmath142 quarks are generally consistent with the symmetric quark model expectations , with the @xmath480 increasing towards unity at larger @xmath2 . the @xmath481 ratio , on the other hand , remains negative for all @xmath2 where it is constrained , and shows no indication of the upturn predicted by the helicity conservation models . the nuclear and finite-@xmath40 effects can significantly impact the limiting behavior as @xmath165 , and clearly additional data are needed in order to constrain their @xmath2 dependence at high @xmath2 . a dedicated program of large-@xmath2 structure function measurements is planned at the 12 gev energy upgraded jefferson lab facility @xcite . more complete information on spin - dependent pdfs will also be available once the @xmath91 structure function of the nucleon is measured more accurately @xcite . this will provide additional information on possible higher twist corrections that enter in the extraction of leading twist pdfs from dis polarization asymmetries . as for the unpolarized inclusive charged lepton dis measurements , inclusive @xmath90 structure function experiments measure @xmath11-even combinations of pdfs , @xmath278 . the individual quark and antiquark distributions can be separated with the help of semi - inclusive charged lepton scattering data , with coincident measurement of fast pions or kaons in the final state . typically such experiments measure the semi - inclusive polarization asymmetry , which at lo can be written @xmath485 where @xmath486 is the fragmentation function for the scattered quark or antiquark to produce a hadron @xmath285 ( @xmath487 ) in the current fragmentation region with a fraction @xmath488 . for large @xmath288 , the produced hadron has a high probability of containing the scattered parton , hence providing a tag on the initial state distribution of quarks and antiquarks . the fragmentation functions @xmath289 are extracted from independent fits to single hadron production cross sections in @xmath290 , @xmath134 and other reactions . , from the compass semi - inclusive dis measurement at @xmath489 gev@xmath170 @xcite , compared with several parametrizations and model calculations , and the unpolarized @xmath305 asymmetry ( solid curve ) . ( figure from ref . @xcite . ) _ ( right ) _ single polarization asymmetry @xmath490 for inclusive @xmath57 production in @xmath491 scattering from the star experiment at rhic , at central electron pseudorapidity @xmath492 . ( figure from ref . copyright ( 2011 ) by the american physical society.),title="fig:",width=340 ] , from the compass semi - inclusive dis measurement at @xmath489 gev@xmath170 @xcite , compared with several parametrizations and model calculations , and the unpolarized @xmath305 asymmetry ( solid curve ) . ( figure from ref . _ ( right ) _ single polarization asymmetry @xmath490 for inclusive @xmath57 production in @xmath491 scattering from the star experiment at rhic , at central electron pseudorapidity @xmath492 . ( figure from ref . copyright ( 2011 ) by the american physical society.),title="fig:",width=226 ] the program of semi - inclusive dis measurements from proton and deuteron targets at hermes @xcite , compass @xcite , and earlier smc @xcite , have produced intriguing glimpses into the flavor content of the polarized sea . the extracted sea quark and antiquark asymmetries are generally rather small , and consistent with zero within the current uncertainties . for the light quark polarized sea , there is a slight trend towards a positive @xmath457 ( with an @xmath2-integrated value of @xmath493 from the latest compass data @xcite ) and a negative @xmath458 ( with an @xmath2-integrated value of @xmath494 ) , so that the difference @xmath460 is slightly positive , at the @xmath495 level @xcite , as fig . [ fig : pol_sea ] indicates . this is in contrast to the large negative values of the unpolarized @xmath496 asymmetry ( see sec . [ ssec : lightsea ] ) , which would disfavor models such as the chiral quark soliton model in the large-@xmath497 limit @xcite that predict large @xmath498 . very small antiquark spin asymmetries would be expected in models based on the pion cloud of the nucleon , such as those discussed in sec . [ ssec : lightsea ] . one should caution , however , that the experimental distributions in fig . [ fig : pol_sea ] were extracted assuming lo expressions for the semi - inclusive asymmetries , which , strictly speaking , can not be directly compared with nlo pdfs @xcite . the comparison should therefore be viewed as a qualitative one , although the general indication of a small polarized non - strange sea is unlikely to be qualitatively modified in an nlo treatment . the semi - inclusive dis measurements are also consistent with a very small strange quark polarization , @xmath296 . the hermes data @xcite give a slightly positive value when integrated between @xmath499 and 0.6 , while the compass data @xcite are consistent with zero . this is in contrast with the analysis of inclusive dis data on protons , deuteron and @xmath221he , which suggests a small negative value . the extractions of @xmath296 from the inclusive and semi - inlcusive data are somewhat dependent on the assumptions of su(3 ) flavor symmetry for the axial charges of hyperons , and on the specific fragmentation functions used in the semi - inclusive analysis . future measurements of kaon fragmentation functions @xmath500 should improve the accuracy of the flavor decomposition of the proton s helicity distributions . a complementary method , discussed by bourrely and soffer @xcite , of constraining the @xmath457 and @xmath458 distributions , which does not depend on knowledge of fragmentation functions , is through longitudinal polarization asymmetries in inclusive @xmath57 boson production from @xmath134 scattering with one proton polarized and the other unpolarized , @xmath501 . the asymmetry for @xmath139 production , for example , can then be written at lo as @xcite @xmath502 where @xmath503 is the scattering angle of the electron in the partonic center - of - mass frame , while the corresponding @xmath137 asymmetry at lo is @xmath504 at large negative rapidity , where @xmath299 , the @xmath139 asymmetry then directly probes the @xmath457 pdf , @xmath505 , while at large positive rapidity one has @xmath132 , and the asymmetry is sensitive to the @xmath282 pdf at large @xmath2 , @xmath506 . for @xmath137 production , the term proportional to @xmath507 will be similarly enhanced at small @xmath299 ; however , the angular factor will produce a suppression of this contribution at backward angles relative to the @xmath508 term , so that both terms will give competing contributions . the first data from the star experiment ( run 9 ) at rhic @xcite are shown in fig . [ fig : pol_sea ] for zero electron rapidity , and are generally in agreement with nlo calculations using existing pdf parametrizations . data from run 12 at rhic have subsequently been taken at both forward and backward rapidities @xcite , and will provide important constraints on the antiquark polarization . information on the polarized gluon distribution @xmath152 can be obtained in several complementary ways . firstly , from the @xmath40 evolution equations , the evolution of the quark singlet contribution to the @xmath90 structure function mixes with the gluon contribution , so that in principle the study of scaling violations in @xmath90 can constrain @xmath152 . in practice , however , the @xmath40 range and precision of the available inclusive dis data are not sufficient to provide meaningful constraints , and dis - only fits of spin - dependent pdfs effectively leave @xmath152 undetermined . values extracted at lo from charm and high-@xmath20 hadron production data from compass , smc and hermes . ( figure from ref . copyright ( 2013 ) by the american physical society . ) _ ( right ) _ longitudinal double - spin asymmetry @xmath509 for inclusive jet production , as a function of the jet transverse momentum , in polarized @xmath510 scattering from the star experiment at rhic @xcite , compared with calculations based on several pdf parametrizations . ( figure from ref . @xcite.),title="fig:",width=321 ] values extracted at lo from charm and high-@xmath20 hadron production data from compass , smc and hermes . ( figure from ref . copyright ( 2013 ) by the american physical society . ) _ ( right ) _ longitudinal double - spin asymmetry @xmath509 for inclusive jet production , as a function of the jet transverse momentum , in polarized @xmath510 scattering from the star experiment at rhic @xcite , compared with calculations based on several pdf parametrizations . ( figure from ref . @xcite.),title="fig:",width=321 ] more direct information on gluon polarization has come from the production of charm mesons and high-@xmath20 hadrons in semi - inclusive dis , through the process of photon gluon fusion ( see sec . [ ssec : hqresults ] ) . data on hadron production from hermes , smc and compass , and open charm data from compass have been analyzed over a range of @xmath2 values , averaging from @xmath511 to @xmath512 , and @xmath40 between @xmath513 gev@xmath170 and 13 gev@xmath170 . most of the analyses have been performed at lo , finding values of @xmath462 consistent with zero , albeit with large uncertainties , as illustrated in fig . [ fig : pol_g ] . the recent compass experiment @xcite was the first to be analyzed at nlo , yielding @xmath514 over the range @xmath515 with an average @xmath516 . an alternative reaction that can be used to constrain @xmath152 is the inclusive production of jets or neutral pions in polarized proton proton scattering . the longitudinal double - spin asymmetry for this process , @xmath517 where @xmath518 is the differential cross section for longitudinally polarized protons with equal ( opposite ) helicities , is then sensitive to the polarized quark gluon scattering cross section at the parton level . following an early measurement of @xmath509 for multi-@xmath519 pairs from two - jet type events by the fermilab e581/704 collaboration @xcite , which found a small @xmath462 at @xmath520 , most of the experimental effort at measuring @xmath152 in @xmath134 scattering has been by the phenix and star collaborations at rhic . the star experiment measures inclusive jet production in @xmath521 at an invariant center of mass energy @xmath302 gev . the @xmath509 data from the 2006 and 2009 runs @xcite shown in fig . [ fig : pol_g ] provide the clearest evidence to date for a small but nonzero @xmath152 . a nonzero asymmetry is also observed by star in preliminary data on dijet events at mid - rapidity @xcite , which may provide even better constraints on the shape of @xmath152 . the phenix data on inclusive neutral pion production , @xmath522 , from 2005 to 2009 suggest a very small asymmetry up to @xmath523 gev @xcite . taken together , the inclusive rhic data from several different channels suggest a small but non - vanishing gluon polarization in the proton . though not as large as had been proposed in some early explanations of the small contribution of quarks to the proton spin , the precise role played by @xmath152 in the proton spin budget will be intriguing to understand from future measurements . in this topical review we have presented a summary of the current knowledge of spin - averaged and ( longitudinal ) spin - dependent pdfs . while great progress has been made in recent years , there are still regions of bjorken-@xmath2 where the pdf uncertainty bands are uncomfortably large . this is primarily due to either lack of data in kinematic regions that would otherwise provide constraints on the pdfs , or to the particular pdf being smaller in that region than those of other flavors , or both . one might naively conclude that if existing data do not constrain particular pdfs in certain domains , then it may not be too important to seek ways to constrain them there . however , this viewpoint overlooks two important points : ( i ) there is intrinsic interest in the behavior of the different pdfs and relations to the properties of the parent hadrons , and ( ii ) it is possible that new phenomena may occur in extended kinematic regions which would require precise knowledge of the pdfs there . as an example , suppose one were searching for a massive state produced at large rapidity in hadron hadron collisions . this would typically require one parton to have a large value of @xmath2 and the other a small value . the errors at large @xmath2 may well dominate the uncertainty in the resulting prediction . the key here is to measure standard model processes in expanded kinematic regions and include these processes in the global fits . the resulting reduction in the errors can then be used to improve searches for new phenomena in these regions . moreover , if one reduces the errors at large values of @xmath2 , this will necessarily lead to reductions in the errors at intermediate @xmath2 . since deuterium targets in dis are a prime source of information on the @xmath142 quark pdf at large values of @xmath2 , nuclear corrections are a significant source of uncertainty in its determination . several planned experiments at the 12 gev energy upgraded jefferson lab @xcite will use techniques that reduce the effects of nuclear corrections on the extraction of @xmath145 at large @xmath2 . constraints on the @xmath145 ratio in the proton can also be obtained , free of nuclear effects , through the measurement of @xmath57 charge asymmetries in @xmath134 collisions at very high rapidity @xcite , although this would require reconstruction of the @xmath57 distributions themselves from the decay lepton distributions and , unlike in @xmath524 collisions at the tevatron , would require precise knowledge of the @xmath272 and @xmath273 distributions @xcite . another important area of investigation is the flavor separation of the pdfs . for example , currently our knowledge of the @xmath254 and @xmath255 pdfs comes primarily from one neutrino experiment which used an iron target @xcite , thus requiring the use of model - dependent nuclear corrections . other sources , for example weak vector boson production at the lhc @xcite , can provide additional constraints , especially as the statistics of the measurements improve . in a complementary effort , the hermes collaboration has measured @xcite @xmath525 production in semi - inclusive dis and used these data to place constraints on the strange quark pdf . while an interesting analysis , it does depend on knowledge of the @xmath525 fragmentation functions , and a recent , more complete analysis @xcite of the hermes data suggests a somewhat smaller @xmath254 pdf than that extracted in ref . better knowledge of the fragmentation functions would clearly help in the interpretation of the semi - inclusive dis measurements . ideally , a high statistics neutrino experiment on _ hydrogen _ could further provide knowledge of both the @xmath254 and @xmath255 pdfs through the measurement of charm production ( using the @xmath526 and @xmath527 subprocesses ) without having to make model - dependent nuclear corrections . note that such an experiment could also yield information on the @xmath145 ratio at high values of @xmath2 . information on the su(2 ) flavor asymmetry of the proton sea will be provided by experiment e906/seaquest @xcite at fermilab , which is currently taking data on muon pair production using a 120 gev proton beam on a variety of targets . it is designed to improve our knowledge of @xmath528 over the @xmath2 range out to @xmath529 , whereas the currently available data run out of statistics at @xmath530 . the experiment will further enable a study of the effects of nuclear binding on the sea quark pdfs . there is also the possibility of follow - on measurements of lepton pair production at the nuclear and particle physics facility at j - parc @xcite in japan . the question of charge symmetry violation in pdfs can be addressed in future through study of drell - yan lepton pair production in @xmath531 scattering from the deuteron , which could be measured at the fermilab main injector or with the compass experiment at cern @xcite , or via measurement of @xmath532 electroproduction ratios in semi - inclusive dis from isoscalar targets @xcite . another avenue to explore will be parity - violating dis on a deuterium target @xcite , for which the parity - violating asymmetry arising from @xmath533 interference is expected to be independent of hadronic structure at large @xmath2 in the absence of csv effects @xcite . beyond searches for csv effects , a program of parity - violating dis studies at the 12 gev jefferson lab @xcite could open up a new window on the vector @xmath534 interference structure functions , as well as provide glimpses into the parity - odd @xmath535 structure function . for heavier quarks , the issue of intrinsic charm will require measuring the charm structure function at large values of @xmath2 , which may be possible at a future electron - ion collider facility @xcite . the semi - inclusive production of prompt photons and charmed jets in @xmath134 collisions at the lhc , @xmath536 , may produce an enhancement at large transverse momenta from intrinsic charm @xcite . measurements of jet , dijet , isolated @xmath519 , and @xmath519 + jet production at the lhc will all play a role in helping to further constrain the gluon pdf . of particular interest would be @xmath519 + jet production where both the @xmath519 and the jet have equal and opposite large rapidities . the dominant @xmath274 subprocess would thus require both a high-@xmath2 gluon and a high-@xmath2 quark . furthermore , continued progress on resummation , especially for the fragmentation contribution at fixed target energies may provide additional information on the gluon at high values of @xmath2 . in the polarized sector , several experiments planned at jefferson lab will measure the longitudinal polarization asymmetries for hydrogen , deuterium and @xmath221he to unprecedented large values of @xmath2 @xcite . these will constrain the spin - dependent valence quark pdfs , especially the @xmath282 distribution , to values of @xmath537 . this should help answer long - standing questions about the behavior of the @xmath538 ratios as @xmath165 , and reveal insights into the role of quark orbital angular momentum in the nucleon . the flavor asymmetry of the polarized light quark sea , @xmath460 , will be further probed via @xmath70 boson production in polarized @xmath134 scattering at rhic at @xmath539 gev . both the star and phenix experiments anticipate significantly improved statistics and kinematic coverage in their 2012 and 2013 runs @xcite . the strange quark polarization , despite appearing to be smaller than initially surmised from early polarized dis experiments , is neverthess important to pin down , as it allows a portal to nonperturbative qcd effects in the nucleon . improved constraints on kaon fragmentation functions should lead to more reliable determinations of @xmath296 in semi - inclusive dis . finally , the polarized gluon distribution is being actively investigated in semi - inclusive deep - inelastic hadron production at compass , and through double - spin asymmetries for pion and jet production in polarized @xmath134 collisions at rhic . preliminary results from star on jet and dijet cross sections hint at nonzero values for @xmath152 , which will need to be confirmed by further measurements at small @xmath2 . access to @xmath152 below @xmath540 will be possible with future running at @xmath541 gev and at large forward rapidity @xcite , although reaching values down to @xmath542 will only be possible with an electron - ion collider @xcite . on the theory front , continued progress on nnlo calculations will eventually result in global fits where all the processes are treated consistently at nnlo , for both unpolarized and polarized observables . the increased precision and decreased scale dependence will help to reduce the uncertainty bands on the pdfs . the parametrization dependence of global fits will continue to be explored , allowing for greater flexibility and reducing bias introduced with the use of specific forms , such as for the @xmath145 ratio at large @xmath2 . improvements in electroweak radiative corrections are being sought , incorporating , for example , pdfs of a photon with @xmath543 @xmath40 evolution . progress on computing moments of pdfs in lattice qcd is also anticipated , with more reliable simulations performed on larger lattice volumes and at quark masses near the physical limit . a new approach @xcite to calculating the @xmath2 dependence of the pdfs directly in the infinite momentum frame may provide complementary constraints on pdfs that are difficult to access experimentally . we thank a. accardi and k. griffioen for helpful comments and a careful reading of the manuscript . the work of p.j .- d . and was supported by doe contract no . de - ac05 - 06or23177 , under which jefferson science associates , llc operates jefferson lab . the work of j.f.o . was supported by doe contract no . de - fg02 - 13er41942 . 10 a. accardi , _ large - x connections of nuclear and high - energy physics _ , to appear in mod . phys . a ; pos ( confinement x ) 227 , + http://pos.sissa.it/archive/conferences/171/227/confinement%20x_227.pdf . jefferson lab experiment e12 - 06 - 109 , _ the longitudinal spin structure of the nucleon _ , d. crabb _ et al . _ , spokespersons . jefferson lab experiment e12 - 06 - 110 , _ measurement of the neutron spin asymmetry @xmath463 in the valence quark region using an 11 gev beam in hall c _ , j .- p . chen _ et al . _ , spokespersons . jefferson lab experiment e12 - 06 - 122 , _ measurement of the neutron asymmetry @xmath463 in the valence quark region using 8.8 and 6.6 gev beam energies and the bigbite spectrometer in hall a _ , b. wojtsekhowski _ et al . _ , spokespersons . a. accardi _ et al . _ , _ electron ion collider : the next qcd frontier - understanding the glue that binds us all _ , bnl-98815 - 2012-ja , jlab - phy-12 - 1652 , arxiv:1212.1701 [ nucl - ex ] . jefferson lab experiment e12 - 10 - 102 [ bonus12 ] , _ the structure of the free neutron at large @xmath2-bjorken _ , s. bltmann , m. e. christy , h. fenker , k. griffioen , c. e. keppel , s. kuhn and w. melnitchouk , spokespersons . jefferson lab experiment e12 - 10 - 103 [ marathon ] , _ measurement of the @xmath544 , @xmath145 ratios and @xmath545 emc effect in dis off the tritium and helium mirror nuclei _ , g. g. petratos , j. gomez , r. j. holt and r. d. ransome , spokespersons . fermilab e906 experiment , _ drell - yan measurements of nucleon and nuclear structure with the fermilab main injector _ , d. f. geesaman and p. e. reimer , spokespersons ; http://www.phy.anl.gov/mep/seaquest/index.html . a. airapetian _ et al . _ [ hermes collaboration ] , in print ; h. jackson , talk given at the _ xxi international workshop on deep inelastic scattering and related subjects ( dis2013 ) _ , marseilles , france , april 22 - 26 , 2013 . jefferson lab experiment e12 - 09 - 002 , _ precise measurement of @xmath532 ratios semi - inclusive dis : charge symmetry violating quark distributions _ , k. hafidi _ et al . _ , spokespersons . c. y. prescott _ et al . _ , b * 77 * , 347 ( 1978 ) . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: quantum information theory is an area of scientific investigation that has witnessed an enormous progress in the last decade @xcite . in the framework of quantum information science , paradigmatic systems of condensed matter physics such as , e.g. , spin chains and harmonic lattices , are analyzed from the point of view of their information content and ability to process and transfer information . fundamental concepts of statistical mechanics and probability , such as the shannon - von neumann entropy @xcite , play a central role in the quantification of bipartite quantum entanglement @xcite . on the other hand , exciting advances in quantum information research are in a sense paying back the debt : the mathematical and theoretical toolkit developed for the characterization and quantification of quantum entanglement has proven useful to tackle questions and improve our understanding in the investigation of strongly correlated systems and quantum phase transitions @xcite . perhaps the most interesting development stemming from the application of the tools of entanglement theory concerns the study of correlation scaling in diversely connected systems , with its relations to conformal field theory and the proof of longstanding conjectures on entropic area laws in spin and harmonic lattice systems @xcite . these investigations and the associated results are of fundamental interest because they allow to enhance the control and establish firmer bounds and limits of applicability on many important tools of simulation and numerical analysis of complex many - body systems , a fundamental task in condensed matter , such as the density matrix renormalization group ( dmrg ) algorithms @xcite , matrix product representations @xcite , the entanglement renormalization ansatz @xcite , and weighted graph states methods @xcite . the pioneering application of quantum information concepts to condensed matter was the observation that two - body entanglement , as quantified by the concurrence , in the ground state of a cooperative system , exhibits peculiar scaling features approaching a quantum critical point @xcite . these seminal studies helped to clarify that at quantum phase transitions @xcite , the dramatic change in the ground state of a many - body system is associated to , or reflected by , a change in the way that quantum fluctuations are correlated , that is , in the way entanglement is distributed among the elementary constituents . many recent efforts have thus been aimed at understanding the behavior of different measures of entanglement at quantum phase transitions , and assessing the corresponding enhancement of properties useful for applications in quantum technology and quantum engineering @xcite . however , since most entanglement measures can be often rewritten in terms of the conventional n - point correlation functions , the presence of a divergence in the former at quantum criticality can be directly traced back to a divergence in the latter @xcite , although in some particular instances entanglement - based studies allowed to discover novel types of phase transitions @xcite . finally , a more technologically oriented product of the interaction between these two areas of research is the study of quantum spin chains as natural information carriers and distributors @xcite , able to realize tasks such as quantum state transfer @xcite and storage @xcite , possibly with maximal fidelity in chains with suitably engineered couplings and dynamics . quantum engineering and quantum technology are being indeed developed at a fast pace . quantum devices are vigorously pursued for applications ranging from nano - sciences to quantum computation and entanglement - enhanced metrology @xcite . despite a large variety of possible implementations involving different physical systems , many relevant properties of such devices can be in fact investigated in a unified setting by appropriate mappings to quantum spin models @xcite . control of separability and entanglement in ground states of quantum spin models plays an important role in entanglement - enabled quantum technology applications , and needs to be characterized and quantified in different physical regimes @xcite . from a more fundamental perspective , the determination of exact solutions endowed with precisely known properties of separability or entanglement , can be of great relevance in the study of advanced models of condensed matter and cooperative systems that are in general not exactly solvable . in this paper we exploit quantum information tools to tackle another important problem in condensed matter physics : the determination and characterization of exact ground states of cooperative systems characterized by the property of being in the form of a fully factorized , tensor product of single - particle pure states , with no quantum correlations between the individual constituents . historically , the occurrence of totally factorized ( unentangled ) ground states of quantum many - body systems was first discovered in the one - dimensional anisotropic heisenberg model with nearest - neighbor interactions by kurmann , thomas and mller @xcite , by adopting a direct method in terms of product state ansatz . this result was later re - derived and extended to two dimensions using quantum monte carlo numerical methods @xcite . the direct product - state ansatz method of kurmann , thomas and mller has been later extended to anisotropic rings with interactions of arbitrary range , by exploiting the observation that factorized ground states break the parity symmetry @xcite . complex quantum systems exhibiting cooperative behaviors , whose ground states are typically entangled @xcite , may thus admit , for some non trivial values of the hamiltonian parameters , a ground state which is completely separable . the occurrence of factorization at finite or even strong values of the couplings is thus an effect of a delicate balancing between interactions and external fields . the phenomenon of ground state factorization is particularly intriguing as it appears to be associated with the presence of an `` entanglement phase transition '' with no classical counterpart @xcite . furthermore , for the purposes of quantum engineering applications , that employ distributed entanglement in order to manipulate and transfer information @xcite , factorization points need to be exactly identified and either avoided , in order to guarantee the reliable implementation of quantum devices , or properly exploited for the dynamical creation of strongly entangled multipartite states of large assemblies of microsystems ( graph and cluster states ) . finally , and also quite importantly , for models not admitting exact general solutions , achieving knowledge of the exact ground state , even if only for the restricted nontrivial set of parameters associated to factorization , would allow ( i ) to prove the existence of an ordered phase and characterize it ; ( ii ) to build variational or perturbative approximation schemes around the exact factorized solution , that may then be used as test benchmarks for the validity and the precision of numerical algorithms and simulations . unfortunately , the direct method based on the ansatz of a factorized solution is neither sufficiently general nor mathematically tractable , apart from the notable exceptions of one - dimensional spin chains with short - range @xcite , infinite - range @xcite , and some types of long - range exchange interactions @xcite . in a recent work @xcite we introduced an analytic method that allows to determine exactly the existence of factorized ground states and to characterize their properties in quantum spin models defined on regular lattices of any size ( finite as well as at the thermodynamic limit ) , in any spatial dimension , and with spin - spin interactions of arbitrary range . in correspondence to rigorously established ground state factorizability , the method also allows to determine novel sets of exact solutions in generally non exactly solvable models . all the previous particular findings can be rigorously re - derived and extended within a unified framework inspired by concepts of quantum information science . in the present work we extend the original method introduced in ref . @xcite and we develop a general self - contained theory of ground state factorization in ( frustration - free ) quantum spin systems : with no _ a priori _ assumption on the nature of the magnetic ordering , we derive a complete and closed set of conditions that have to be satisfied by the hamiltonian interaction parameters in order for the ground state to be factorized , at precisely determined values of the external magnetic field , with well defined values of the magnetization and of the energy . besides the completely rigorous re - derivation of the few previously known results on quantum factorization points @xcite , our general analytic method allows to determine novel , exact factorization points and factorized ground states in , generally non exactly solvable , spin-@xmath0 models as well as in the corresponding hamiltonian models of any higher spin . these results hold true regardless of spatial dimensionality and interaction range , and can be easily extended even to non - translationally invariant systems . we illustrate the versatility of the method with a series of different applications , including models with nearest - neighbor interactions ( for which we provide the full factorization diagram , generalizing the results of refs . @xcite and of our recent work @xcite ) , models with long - range interactions ( notably including , among others , the fully connected infinite - range lipkin - meshkov - glick ( lmg ) model @xcite and systems with unbalanced next - to - nearest - neighbor interactions on cubic lattices ) , and models with spatial anisotropies . the paper is organized as follows . an excursus on the problem of ground state factorization is presented in sec . [ sec.history ] . in sec . [ sec.squoexe ] we recall the quantum information toolkit upon which our method is based , in particular the formalism of single - spin , or single - qubit , unitary operations and associated entanglement excitation energies , previously introduced for the characterization and quantification of bipartite entanglement from an abstract geometrical perspective @xcite . the general method for the rigorous determination of factorized ground states , that generalizes the scheme introduced in ref . @xcite , is discussed in full detail in sec . [ sec.method ] . relevant applications and examples , including models defined on lattices of different spatial dimensions and with different interaction ranges as well as extensions to models with spatial anisotropies and with dimerized interactions are discussed in sec . [ sec.short ] , sec . [ sec.long ] , sec . [ sec.lmg ] , and sec . [ sec.anisotropy ] . concluding remarks and outlooks on future perspectives are presented in sec . [ sec.concl ] . quantum fluctuations in the ground states of cooperative many - body systems are typically highly correlated , that is , those states contain in the most general instance a strong degree of entanglement distributed among the individual components . this is one of the reasons why their exact characterization is often hard to be accomplished by analytical or even numerical methods . however , there may exist particular solutions corresponding to special values of the hamiltonian parameters such that the ground state becomes `` classical - like '' , in the sense of being a full product state of single - site factors ( completely separable , fully factorized state ) . it is quite surprising that despite the inherent appeal of such exact , special solutions , and the renewed interest stemming from possible applications to quantum technology , very few analytical advances were obtained in the last quarter of a century . the first systematic work on ground state factorization in quantum spin models was completed by kurmann , thomas , and mller in 1982 @xcite . they focused on the one - dimensional nearest - neighbor anisotropic heisenberg - like antiferromagnetic spin-@xmath0 model in transverse field , with the assumption that all the spin - spin couplings take nonnegative values . et al . _ identified a precise value of the external magnetic field , the `` factorizing field '' , corresponding to which the system admits a fully factorized ground state . they employed a very direct , brute force " method , i.e. formulating a product state ansatz and verifying if and for what values of the hamiltonian parameters it satisfies the stationary schrdinger problem with lowest energy . even though this approach appears obvious and extremely easy to pursue , the positive result obtained by kurmann _ _ in the case of the 1-d heisenberg model was enabled by some special properties of the case they analyzed . in fact , in order to verify the eigenvalue equation for the product state ansatz , one needs to decompose the total hamiltonian of the system into a series of pairwise terms , which could be done easily for the model studied by kurmann _ _ since all the `` pairwise components '' of the factorizing magnetic field had the same weight in this particular case . however , this direct method of determining factorization is in fact significantly less obvious , useful , and comprehensive , when one tries to apply it to more general cases . for instance , considering models with interaction terms of growing spatial range , and/or increasing the spatial dimensionality of the lattice , and/or treating extensions to spatially anisotropic or non - translationally invariant models , results in increasingly hard `` guesses '' concerning the partitioning of the external field into pairwise components and in increasingly nontrivial verification steps . depending on the structure of the model and on the form of the product state ansatz , the direct method may either fail to detect factorization points at all , or it may detect only a subset of all physically realizable factorized solutions , or finally it may lead to intractable sets of conditions . these limitations of the direct method become evident , as we will see in the following , even in the simplest 1-d @xmath1 model with nearest - neighbor interactions analyzed by kurmann and coworkers . in their original work , due to the insufficient generality of their proposed product ansatz , they failed to identify a wide range of situations that allow the existence of factorized ground states . as we will see when discussing models with competing interactions of different spatial ranges , a further serious fault of the direct method , that becomes incurable in the presence of frustration , is its incorrect assessment of factorized ground states that are in fact excited energy eigenstates of the system . following the seminal analysis by kurmann _ _ , the direct method has been applied to the fully connected infinite - range lipkin - meshkov - glick model by dusuel and vidal @xcite , and to the anisotropic heisenberg chain with the same interactions of arbitrary range along the three spatial directions by hoeger et al . @xcite and by rossignoli _ we would like to remark that these works , that include the results of refs . @xcite as special cases , provide an interesting characterization of factorization points in finite ( one - dimensional ) translationally invariant lattice spin models with periodic boundary conditions as those points where the parity symmetry is broken . a general and rigorous approach to the problem of ground state separability , completely different from the direct method based on the product state ansatz , has been introduced by us in ref . @xcite and is fully developed in the present work . this analytic method allows _ inter alia _ to complete the factorization diagram of the anisotropic spin-@xmath0 models without restrictions on the sign of the couplings , and to go significantly beyond the limitations of the previously mentioned studies . our approach is based on an additional characterization of product states ( including product ground states ) that employs concepts and tools of quantum information and entanglement theory . the strong point of the quantum informatic approach to ground state factorization lies in its generality and its ensuing independence from the types and ranges of interactions present in the hamiltonian , the lattice size , and the spatial dimension , which makes it applicable in the most general cases and therefore especially useful in all the situations that are either intractable or incomplete if one resorts to the direct method or to numerical techniques . in fact , in the original formulation of the quantum informatic method @xcite there was still trace of a residual `` ansatz '' in the initial stage , since the magnetic ordering of the candidate factorized ground state had to be in some sense guessed , and imposed _ a priori_. moreover , in the final stage , we were not yet able to derive a completely general set of conditions in order to determine in each situation whether a candidate factorized state that turned out to belong to the spectrum of the hamiltonian was actually a ground state or belonged to the set of excited states . this had then to be decided case by case in each specific instance . in the present work we proceed further and develop the complete method to determine and characterize factorized ground states in quantum cooperative systems . we provide the systematics that allows to determine the occurrence of full ground state separability given a generic quantum spin hamiltonian . the magnetic ordering is derived as a consequence of the theory and not imposed _ a priori_. a complete set of general conditions is established for the verification of ground state factorization , and the elegance of the method does not hinder practical usefulness in the application to specific models and situations . we would like to recall that numerical advances on the problem of ground state factorization have been realized by the monte carlo study of ground state factorization in two - dimensional anisotropic nearest - neighbor spin models @xcite . unfortunately , beyond this important result the numerical approach appears to have a very restricted potential for extensions , since the required computational power increases rapidly with the spatial dimension and/or with the range of interactions . we will show that the difficulties inherent to the direct analytic approach and to the numerical method are intrinsically absent in our formalism . in the present section we provide a brief overview of some recent results on the geometrical interpretation of the entanglement measure known as `` tangle '' ( linear entropy ) @xcite and the consequences it entails for the relationship between energy and pairwise entanglement @xcite . such results embody the premises for our all - analytic characterization of ground state separability in quantum spin systems . here and in the following we consider generic systems made of @xmath2 spin-@xmath0 elementary constituents , or , in the language of quantum information , @xmath2 qubits . for such a system , we introduce the set of single - qubit unitary operations ( _ squos _ ) @xmath3 @xcite , defined as the unitary transformations that act separately as the identity on any spin of the system except the @xmath4-th one , on which they act instead as hermitian , unitary , and traceless operators : @xmath5 in eq . ( [ e.uk ] ) @xmath6 stands for the identity operator on the @xmath7-th spin of the system while the generic hermitian , unitary , and traceless operator @xmath8 can be expressed as a linear combination of the standard spin operators ( @xmath9 ) defined on the @xmath4-th spin : @xmath10 in the previous definition the parameters @xmath11 and @xmath12 take values respectively in the ranges @xmath13 $ ] and @xmath14 . physically , @xmath15 corresponds to a rotation in the spin space , and the traceless condition imposes the rotation operation to be orthogonal to the identity , so basically a combination of spin flip and phase flip . a squo transforms every pure state @xmath16 into a new state @xmath17 , that in general differs from @xmath18 . we may quantify the action of the squo in terms of the trace distance between the original and the transformed state , defined as @xmath19 . the distance @xmath20 varies in the interval @xmath21 $ ] and vanishes if and only if the two states coincide , meaning that the considered squo leaves the original state unchanged . for a given initial state and for any site @xmath4 , one can determine the _ extremal - squo _ ( e - squo ) that minimizes such a distance . one can prove that the e - squo is uniquely determined by the following conditions on the angular variables @xcite : @xmath22 where @xmath23 denotes the expectation value of the spin operator @xmath24 on the unperturbed state @xmath18 ( @xmath25 ) . the e - squo plays a crucial role in our analysis : for any pure state defined on a general system of interacting qubits one can prove @xcite that the square of the distance associated to the action of the e - squo coincides with the linear entropy @xmath26 $ ] of the @xmath4-th qubit ( where @xmath27 denotes the reduced density matrix of qubit @xmath4 ) . this quantity , also known as the _ tangle _ @xmath28 in the literature , is an entanglement monotone for qubit systems and satisfies important monogamy constraints @xcite . being a monotonic function of the von neumann entropy of entanglement @xcite , it quantifies the bipartite block entanglement present in the state @xmath18 between spin @xmath4 and the remaining @xmath29 constituents , and is thus a proper measure of single - site entanglement @xcite . one has the following relation @xcite : @xmath30 \ ; .\ ] ] this geometric - operational characterization of the tangle allows to develop a quantum informatic approach to the problem of characterizing the correlation properties of the ground state in many - body quantum systems @xcite . let us consider a collection of spin-@xmath0 systems defined on a regular lattice , with hamiltonian @xmath31 , and let @xmath18 denote from now on the ground state of @xmath31 . if the system is in the ground state , the application of a squo on a generic spin of the lattice is necessarily associated to an energy transfer that one must provide to the system in order to realize the selected squo . in other words , a squo perturbs the system , and this perturbation results in an increase of the average energy ( obviously , in general the transformed state needs not to be an eigenstate of the hamiltonian ) . such energy deviation from the ground - state value can be quantitatively defined as @xmath32 because @xmath18 is the ground state of @xmath31 , @xmath33 is a non negative defined quantity . considering the special relation between the e - squo and the tangle we may expect that the associated amount of energy difference @xmath34 , appropriately named ` entanglement excitation energy ' ( exe ) , will have as well a strong , direct connection with the tangle and in general with the single - site entanglement in the ground state . in fact , starting from the above definition of eq . ( [ e.ees ] ) and taking into account the property of the e - squo of leaving unchanged a fully disentangled state , it has been established in full generality @xcite that if the system is invariant under spatial translation and the hamiltonian of the system does not commute with any possible squo , then the vanishing of the exe is a _ necessary and sufficient _ condition to admit a fully factorized ground state @xmath18 @xcite . hence , if the system that we want to analyze satisfies all the former hypotheses , the vanishing of the exe provides an additional condition , besides the vanishing of the tangle @xmath28 , that one needs to include in establishing a general analytic approach to the problem of ground state factorization . the general method that was first partially developed in ref . @xcite for the determination of existence , location , and exact form of factorized ground states rests on the following two main observations : the ground state @xmath18 of a spin-@xmath0 hamiltonian @xmath31 is factorized if and only if the single - site entanglement ( tangle ) @xmath28 vanishes for all spins @xmath4 in the lattice ; provided that @xmath35\neq 0 $ ] for every possible squo @xmath36 of the form eq . ( [ e.uk ] ) , the ground state @xmath18 of a spin-@xmath0 hamiltonian @xmath31 is factorized if and only if the exe @xmath37 vanishes @xmath38 . in ref . @xcite , the strategy to the determination and understanding of factorization for a given hamiltonian bore on the above two facts and on the requirement to assume as working point a phase endowed with some kind of magnetic order . by imposing the vanishing of the exe and of the tangle , we showed how to determine uniquely the form of the completely disentangled state compatible with the assumed magnetic order , candidate to be the ground state of the system at a particular value of the external magnetic field ( labeled as `` factorizing field '' ) . the final steps of the procedure concern the determination of exact analytic conditions for the candidate factorized state to be an eigenstate of the hamiltonian with lowest energy . in ref . @xcite , we derived a closed set of eigenstate conditions , but establishing whether a factorized eigenstate is actually a ground state had to be left to a case by case analysis . in the present paper we generalize the method , extend its scope , and apply it to different classes of quantum spin models . firstly , we will show that there is no need for _ a priori _ assumptions on the magnetic order . in fact , we will prove that the existence of a factorized ground state is a necessary and sufficient condition for the existence of quantum phase transitions and ordered phases in translationally - invariant quantum spin models with exchange interactions and in the presence of external fields . moreover , the actual structure of the factorized ground state determines automatically the kind of ordered phase . further , we derive a complete set of rigorous conditions for the candidate disentangled state to be an eigenstate and a ground state . in particular , for systems with no frustration we prove that if the factorized state is an eigenstate , then it is always a ground state of the system . we illustrate the method in detail by considering its application to the general , translationally invariant , exchange hamiltonian @xmath31 for spin-@xmath0 systems on a @xmath20-dimensional regular lattices , with spin - spin interactions of arbitrary range and arbitrary anisotropic couplings : @xmath39 here @xmath40 ( and similarly @xmath41 ) is a @xmath20-dimensional index vector identifying a site in the lattice , @xmath42 @xmath43 stands for the spin-@xmath0 operator on site @xmath40 , @xmath44 is external field directed along the @xmath45 direction , @xmath46 is the distance between two lattice sites , and @xmath47 is the spin - spin coupling along the @xmath48 direction . translational invariance is ensured by the fact that all the couplings depend only on the distance @xmath49 between the spins . this type of hamiltonian encompasses a large variety of models and spans over several universality classes including , among others , the ising , xy , heisenberg , and xyz symmetry classes . besides their importance in quantum statistical mechanics and in the theory of quantum critical phenomena , these models play an important role in the study of various schemes of quantum information and quantum communication tasks with quantum many - body systems @xcite . it is a rather straightforward exercise to verify that the hamiltonian eq . ( [ e.hamiltonian ] ) never commutes with the squos @xmath50 . translational invariance and non - commutativity of the hamiltonian with the squos guarantee that the vanishing of the exe is a necessary and sufficient condition for the full separability of the ground state [ statement ( ii ) above ] . since the external field @xmath44 is uniform on the lattice , it forces the existence of a site - independent , nonvanishing magnetization @xmath51 along the @xmath45 axis : @xmath52 . on the other hand , at a factorization point , statement ( i ) above imposes the vanishing of the single - site entanglement on all sites of the lattice . therefore , if a factorization point exists , the magnetizations along the directions orthogonal to the external field must assume the following form : @xmath53 where @xmath54 is the modulus of the projection of the magnetization on the @xmath55 plane and the local orientations @xmath56 , and hence the type of magnetic ordering in the @xmath55 plane , remain undetermined . since @xmath51 is site - independent , then @xmath57 is site - independent as well . taking into account eq . ( [ e.minimumpoint ] ) of sec . [ sec.squoexe ] one sees that , at least at the factorization point , @xmath58 , i.e. one of the two angles that fix the orientation of the e - squo , does not depend on the site index . therefore @xmath59 @xmath60 . collecting the former results and considering that in the presence of a factorized ground state , if it exists , all the correlation functions are products of single - site expectations values , we obtain that the condition on the vanishing of the exe associated to the generic site @xmath61 reads : @xmath62 in the former equation @xmath63 is the _ factorizing field _ , i.e. the value , if it exists , of the external field for which ground state factorization occurs . since we are looking for a fully factorized ground state , eq . ( [ e.exe ] ) must be satisfied simultaneously for all @xmath61 and , henceforth , eq . ( [ e.exe ] ) must be site - independent . this fact in turn implies that the following relation must be satisfied by every angle @xmath56 : @xmath64 eq . ( [ e.exe ] ) must hold independently of the site on which the e - squo acts and the value of the constant @xmath65 must be identified by determining the expression of the energy density , i.e. the energy per site , associated to the factorized ground state . given the total ground - state energy @xmath66 and the total number of sites @xmath2 , the latter quantity can be written as @xmath67 where @xmath68 is the coordination number , that is the number of sites at distance @xmath49 from a given site . in order to determine the magnetic order in the candidate factorized ground state , one can now exploit the fact that , by definition , it must minimize the energy of the system . therefore such state will be simply characterized by the magnetic order that minimizes the per - site contribution @xmath65 , depending on the hamiltonian parameters and the geometric structure of the lattice . thus the general conditions for full separability of the ground state lead to the uncovering of the magnetic order in the @xmath55 plane . in the original formulation of the quantum - informatic method @xcite one had to assume from the beginning the type of ordered phase , and this resulted in constraints on the magnetization and limitations in the range of the coupling coefficients . the outcome of the minimization of the energy density depends on the presence or the absence of frustration in the system , either due to the lattice geometry , or to the structure of the hamiltonian , or to both causes . for hamiltonians belonging to the class of eq . ( [ e.hamiltonian ] ) , frustration can be due both to the geometry of the lattice , e.g. for systems with nearest - neighbor antiferromagnetic interactions defined on a lattice that admits closed loops with an odd number of spins , and to the competition of interactions of different spatial orders , e.g. in systems defined on linear chains with nearest - neighbor ferromagnetic and next - nearest - neighbor antiferromagnetic interactions . frustration effects in systems with interactions of different spatial ranges arise whenever it happens that the minimization of the energy associated to interactions over a certain spatial scale precludes the possibility of minimizing the energy associated to interactions over a different spatial scale . in the absence of frustration , i.e. when the lattice geometry and the interactions between the spins are such to allow all energy minimizations simultaneously , it is straightforward to prove that every possible factorized ground state must be characterized by one of only four magnetic orders out of the many possible ones . the complete proof of this statement is reported in appendix [ appendixfourcases ] . this restricted set includes ferromagnetic ordering along the @xmath69 direction ( @xmath70 ) or along the @xmath71 direction ( @xmath72 ) , and antiferromagnetic ordering along the @xmath69 direction ( @xmath73 ) or along the @xmath71 direction ( @xmath74 ) , the type of order effectively present in the ground state can be determined by comparing the values of the different `` net interactions '' along the @xmath69 and @xmath71 directions , where the net interactions are defined as follows : @xmath75 with @xmath76 . the net interaction along the @xmath45 direction is defined equivalently as @xmath77 as a function of the net interactions , the magnetic orders in the system are determined by the values of @xmath78 : @xmath79 it is interesting to observe that , in terms of the net interactions , one can immediately establish the presence or absence of frustration in the system : any frustration arising from the presence of competing interactions would imply that @xmath80 , with @xmath76 depending on which of the two axes is characterized by a non vanishing value of the magnetization . we remark that here and in the following we are not considering the particular situation in which a _ saturation _ occurs rather than a true ground state factorization : this instance will be discussed separately in subsec . [ s.sec.saturation ] . from now on , without loss of generality , we will specialize our analysis to the the case of an antiferromagnetic ordering along the @xmath69 direction . with trivial modifications , all the steps and results that will be obtained in the following hold as well in the remaining three cases of eq . ( [ fourcases ] ) . at the end of the procedure we will present a summary table collecting the main results for all the four types of magnetic orders compatible with factorization . given an antiferromagnetic ordering along the @xmath69 direction , and keeping in mind that in such case @xmath73 , eq . ( [ e.magnxy ] ) implies that @xmath81 while @xmath82 with @xmath83 . the @xmath84 sign reflects the fact that the antiferromagnetic order discriminates between two sublattices , each characterized by the opposite sign of the staggered magnetization along @xmath69 . consequently , the general condition eq . ( [ e.exe ] ) takes the form @xmath85 \ , , \end{aligned}\ ] ] that admits two solutions for @xmath86 , @xmath87 however , since the e - squo must be unique @xcite , the two solutions must coincide . factorization thus requires that @xmath88 moreover , the requirement of a vanishing tangle ( single - site entanglement ) imposes @xmath89 combining eqs.([e.mz][e.mx ] ) via eq . ( [ e.minimumpoint ] ) yields a closed expression for the phase @xmath86 as a function of the hamiltonian parameters and of the factorizing field @xmath63 : @xmath90 therefore , eq . ( [ e.theta ] ) , together with the knowledge of @xmath91 , once inserted into eq . ( [ e.ok ] ) , determines , independently of the value of the magnetizations , the form of the e - squo at factorization : @xmath92 the form of the candidate factorized state @xmath93 is then readily determined by imposing that the action of the e - squo on all lattice sites leaves the ground state invariant : @xmath94 > from eqs.([e.inalterate][e.oz ] ) we find then that the candidate factorized ground state @xmath93 has to be a tensor product of single - site states that are the exact eigenstates of the operators @xmath95 with eigenvalue @xmath0 : @xmath96 equation ( [ e.candidate ] ) expresses the general form that a factorized ground state must assume given an antiferromagnetic order along @xmath69 . this result is independent of the actual values assumed by the staggered magnetizations @xmath97 s . we remark that in order to establish the general form of the candidate ground state it is essential to impose the vanishing of the exe . in fact , without requiring it , the phase @xmath86 and the candidate factorized ground state become explicitly dependent on the magnetizations . then , imposing the conditions for eq . ( [ e.candidate ] ) to be an eigenstate of the hamiltonian , one merely obtains a relation between the factorizing field @xmath63 and the magnetizations @xmath98 s . if the analytic expression of at least one of the @xmath98 s is known , it is still possible to obtain an expression of the field @xmath63 in terms of the hamiltonian parameters . however , this actually occurs in few special , exactly solvable , cases ( for instance , the xy model @xcite ) . otherwise , in general one needs to resort to numerical evaluations of the magnetizations or introduce other approximations in the analysis . this problem is completely eliminated in the general analytic framework based on the invariance of the factorized ground states under the action of e - squos and on the vanishing of the exe at factorization . before proceeding further let us clarify a relevant point . because of the even - odd symmetry of the energy spectra of the system , as elucidated also in refs . @xcite , the assignment of one of the two different staggered magnetizations to each sublattice is arbitrary . according to this observation , one sees that , inverting the sign of the magnetization along the @xmath69 direction on each site , all the conditions for factorization are still satisfied but the candidate ground state takes the form @xmath99 obviously the two states in eq . ( [ e.candidate ] ) and in eq . ( [ e.candidate-1 ] ) are distinguishable and both @xmath93 and @xmath100 are legitimate candidate ground states of the system , as well as general linear combinations of the two ( the latter will be in general highly entangled ) . here and in the following we assume to be working in a situation of broken symmetry , i.e. after that a small perturbing external field along the @xmath69 direction has lifted the degeneracy between @xmath93 and @xmath100 , e.g. lowering the energy of the former and rising that of the latter . after a time long enough to ensure convergence to equilibrium , i.e. relaxation to the state @xmath93 , the perturbation is switched off : in the thermodynamic limit , this ensures that the system will remain indefinitely in the state @xmath93 . having determined the exact form of the candidate factorized ground state , the subsequent step concerns the determination of the conditions for its occurrence , i.e. the conditions under which a state of the form eq . ( [ e.candidate ] ) is indeed the eigenstate of @xmath31 , eq . ( [ e.hamiltonian ] ) , with the lowest energy eigenvalue . to this aim , taking the value of the external field at the factorization point , @xmath101 , it is useful to decompose the total hamiltonian as a sum of pairwise hamiltonian terms @xmath102 , so that @xmath103 , where each term @xmath102 involves only the degrees of freedom of a single pair of spins : @xmath104 here the quantity @xmath105 plays the role of the `` component '' of the factorizing field that acts on the selected pair of spins and obeys the following relation : @xmath106 the above ensures that @xmath107 satisfies eq . ( [ e.theta ] ) . proving that the candidate factorized ground state @xmath93 is an eigenstate of the total hamiltonian @xmath31 is equivalent to prove that it is a simultaneous eigenstate of all pair hamiltonians @xmath108 , or , more precisely , that the projection of @xmath93 onto the subspace of two given spins @xmath40 and @xmath109 which is still a pure state since @xmath93 is a tensor product of single - spin pure states is an eigenstate of @xmath108 for every pair @xmath110 . to proceed , we associate to each lattice site a set of orthogonal spin operators defined as follows : @xmath111 it is immediate to observe that @xmath112 . by inverting eqs . ( [ e.axayaz ] ) , we can conveniently re - express the standard spin operators as functions of the new set of operators @xmath113 : @xmath114 inserting eq . ( [ e.sxsysz ] ) in eq . ( [ e.couple ] ) we obtain the expression of the pair hamiltonian @xmath115 as a function of the sets of operators @xmath113 and @xmath116 : @xmath117 -2 h_f^r \sin \theta \left [ { a_{\underline{i}}^x \left ( a_{\underline{j}}^z -\frac{1}{2}\right ) + \left ( a_{\underline{i}}^z-\frac{1}{2}\right ) a_{\underline{j}}^x } \right ] \nonumber \\ & & + \ a_{\underline{i}}^x a_{\underline{j}}^x \left ( \sin^2 \theta j_z^r+\cos^2\theta ( -1)^r j_x^r \right)+ a_{\underline{i}}^y a_{\underline{j}}^y j_y^r ( -1)^r - h_f^r \cos \theta \left ( { a_{\underline{i}}^z + a_{\underline{j}}^z } \right ) \ , , \end{aligned}\ ] ] where we recall that @xmath49 is the distance between sites @xmath40 and @xmath109 and that , in the presence of an antiferromagnetic order along the @xmath69 direction , @xmath118 . the spin - pair projection @xmath119 of @xmath93 is an eigenstate of the pair hamiltonian @xmath120 either if all terms in eq . ( [ e.pairhij ] ) admit @xmath119 as an eigenstate or if they annihilate it . by construction , recalling that @xmath121 and that @xmath122 is the eigenstate of @xmath123 with eigenvalue equal to @xmath0 , the spin - pair projection of @xmath93 is an eigenstate of the pair hamiltonian if the following condition holds : @xmath124 for general models of the type eq . ( [ e.hamiltonian ] ) with a given , arbitrary , maximum spatial range of interaction @xmath125 , eq . ( [ e.condition ] ) admits the following solution : @xmath126 for all distances @xmath127 , i.e. for all distances associated with nonvanishing couplings ( we recall that @xmath128 and @xmath129 ) . the set of equations ( [ e.solcondition ] ) determines the conditions that must be satisfied simultaneously by the phase @xmath86 , which is unique , in order for @xmath93 to be an eigenstate of the total hamiltonian eq . ( [ e.hamiltonian ] ) . ( [ e.solcondition ] ) discriminate quite clearly between short - range models , i.e. models with only one nonvanishing coupling at @xmath130 ( models with nearest - neighbor interactions ) and all other possible models containing finite- or infinite - range interaction terms . specifically , models with only nearest - neighbor interactions are characterized by the fact that eqs . ( [ e.solcondition ] ) reduce to a single condition and hence if the value of the r.h.s . of eq . ( [ e.solcondition ] ) falls in the interval @xmath21 $ ] ( associated to the permitted values of @xmath131 in terms of the interaction parameters ) , one can immediately conclude that the models under investigation admit a factorized eigenstate . on the other hand , for models containing interaction terms of longer range so that eqs . ( [ e.solcondition ] ) include two or more conditions factorized energy eigenstates are allowed if and only if the r.h.s . of all conditions in the set ( [ e.solcondition ] ) take values in the interval @xmath21 $ ] and , moreover , they all coincide . summing term by term over the index @xmath49 all the relations in eq . ( [ e.condition ] ) , taking into account eq . ( [ e.theta ] ) , and solving for @xmath63 , one eventually obtains the exact expression of the factorizing field as a function of the net interactions : @xmath132 we remark that eq . ( [ e.factorizingfield ] ) , which was derived in ref . @xcite using some unnecessary auxiliary assumptions , is completely general and holds for lattices of arbitrary spatial dimension and for spin - spin interactions of arbitrary range . accordingly , the angle @xmath86 that determines the direction of the e - squo can be also expressed as a function of the net interactions as follows @xmath133 in the case of a maximum range of interaction @xmath134 , relation eq . ( [ e.theta2 ] ) for the net interactions is a necessary but not sufficient condition for factorization , and further use of eqs . ( [ e.condition ] ) and eqs . ( [ e.solcondition ] ) is needed , as we will show with explicit examples in sec . [ sec.long ] and sec . [ sec.anisotropy ] . we are finally left with the problem of establishing conditions for a factorized energy eigenstate @xmath93 to be indeed a ground state of the system . a very simple sufficient but not necessary condition is that the spin - pair projection of @xmath93 onto the subspace of a pair of spins be the simultaneous ground state of every pair hamiltonian @xmath108 . in fact , given the relation existing between all the pair hamiltonians @xmath108 in eq . ( [ e.couple ] ) and the total hamiltonian @xmath31 eq . ( [ e.hamiltonian ] ) , it follows that the simultaneous ground state of all pair hamiltonians @xmath108 is also ground state of @xmath31 . to proceed , let us re - express the pair hamiltonians eq . ( [ e.pairhij ] ) in matrix form in the basis spanned by the eigenstates of the operators @xmath135 . one has @xmath136 where @xmath137 and @xmath138 . making use of eq . ( [ e.theta2 ] ) and eq . ( [ e.hfij ] ) the eigenvalues of the matrix eq . ( [ e.matrix ] ) can be put in the form @xmath139 \nonumber \\ \varepsilon_1 & = & \frac{1}{8 } \left [ 3 ( -1)^r j_x^r - j_z^r - \left ( { j_z^r -(-1)^r j_x^r } \right)\cos \left ( { 2\theta } \right ) \right ] \nonumber \\ \varepsilon_2 & = & \frac{1}{8 } \left [ - 3\left ( { ( -1)^rj_x^r + j_z^r } \right ) + \left ( { j_z^r-(-1)^r j_x^r } \right)\cos \left ( { 2\theta } \right ) \right]\nonumber \\ \varepsilon_3 & = & \frac{1}{8 } \left [ { - 3(-1)^rj_x^r + 5j_z^r + \left ( { j_z^r- ( -1)^r j_x^r } \right)\cos \left ( { 2\theta } \right ) } \right ] \nonumber \\ & & \end{aligned}\ ] ] the eigenvalue @xmath140 is the one associated to the projection @xmath119 of the factorized eigenstate @xmath93 onto the four - dimensional hilbert space associated to the pair of spins @xmath141 . therefore , the spin - pair projection is the ground state of the pair hamiltonian if @xmath142 . imposing this condition yields the following inequalities : @xmath143 & \ge & 0 \ ; ; \label{e.suffcondition.1 } \\ ( -1)^r j_x^r - j_z^r & \le & 0 \ ; . \label{e.suffcondition.2}\end{aligned}\ ] ] notably , these inequalities are automatically satisfied for any interacting spin system that is not frustrated and that verifies all of the eqs . ( [ e.solcondition ] ) . namely , for every even value of the distance @xmath49 in the interaction range ( @xmath127 ) eq . ( [ e.suffcondition.2 ] ) implies that the denominator in the corresponding eq . ( [ e.solcondition ] ) is non - positive . therefore , an acceptable value of @xmath131 can be obtained only if also the numerator is non - positive and exceeds or equals the denominator . these conditions in turn imply the order relation , @xmath144 . on the other hand , comparing eq . ( [ e.suffcondition.2 ] ) and eq . ( [ e.suffcondition.1 ] ) , one has that @xmath145 . therefore , comparing the two order relations yields finally @xmath146 ( @xmath49 even ) . this condition is always verified in systems with an antiferromagnetic order along the @xmath69 direction and in absence of frustration and therefore the projection of the factorized energy eigenstate on the hilbert space of every pair of spins is indeed the ground state of every pair hamiltonian . on the other hand , in the case of odd @xmath49 , eq . ( [ e.suffcondition.2 ] ) implies that the denominator of eq . ( [ e.solcondition ] ) is non - negative and , similarly to the former case , the numerator must be non - negative and not exceeding the denominator . hence we obtain @xmath147 while from eq . ( [ e.suffcondition.1 ] ) we recover @xmath148 that , in turn , implies @xmath149 . again , such inequalities are always verified in the presence of an antiferromagnetic order and in the absence of frustration . [ cols="^,^,^,^ " , ] collecting all these results , we have proved that the following holds : * theorem*. _ for any cooperative system of spin-@xmath0 particles described by hamiltonians @xmath31 of the type [ eq . ( [ e.hamiltonian ] ) ] , characterized by an anti - ferromagnetic order along the @xmath69 axis as emerging from eq . ( [ fourcases ] ) and in the absence of frustration , the simultaneous verification of all eqs . ( [ e.condition ] ) is necessary and sufficient for the fully factorized state eq . ( [ e.candidate ] ) to be the exact ground state of @xmath31 when the external magnetic field takes the value @xmath101 determined by eq . ( [ e.factorizingfield])_. this central result holds as well , with obvious modifications , when one considers the other different ordered phases that form the set identified in eq . ( [ fourcases ] ) . in tab . [ t.summary ] we summarize and compare results for each of the four different possibilities . before moving to apply the general method to specific models and examples of conceptual and physical relevance , we discuss briefly the meaning of factorization and its marked differences with the phenomenon of saturation " ( see below ) . as we have seen , and as it will appear even clearer in the discussion of the examples in the following sections , the physical mechanism of ground state factorization in quantum spin systems is due to a delicate kind of balancing " between the coupling strengths that regulate the intensity ( and range ) of the interactions and the aligning effect of the external field , that tends to orient all spins along a given direction and to destroy all quantum correlations . the remarkable aspect of factorization is then that it occurs , according to well defined conditions and constraints , at precisely defined , finite values of the couplings and of the fields in an ordered ( or symmetry - broken ) phase . this means that factorization occurs when the system is relatively strongly interacting and the external aligning field is relatively weak . moreover , we will see that the factorization point is always a precursor from below ( from the ordered phase ) of the quantum critical point , as first observed for the @xmath1 model with nearest - neighbor interactions @xcite . saturation is the phenomenon of ground state factorization that occurs trivially when the value of the external field grows unboundedly compared to all other hamiltonian parameters . therefore its aligning effect on the spins prevails against all other effects and quantum correlations are suppressed . the main physical difference between proper factorization and saturation is then clear and is reflected at the level of the hilbert space of states . namely , suppose that the system under study admits a fully factorized state that is the true ground state of the system at @xmath101 . then , when one moves away from the factorizing field , i.e. when @xmath150 , the factorized state ceases to be an eigenstate of the hamiltonian . on the contrary , in the presence of saturation , the trivially ensuing factorized state is always an eigenstate of the hamiltonian and remains the ground state of the system when perturbing the value of the external field @xmath44 . let us formalize the above discussion . consider a system that admits a fully factorized ground state . in this case the pair hamiltonian in eq . ( [ e.pairhij ] ) admits as eigenstate a factorized spin - pair state . however , eq . ( [ e.pairhij ] ) is the expression of the pair hamiltonian only at @xmath101 . if we consider a value of the external field different from the factorizing field we must include in eq . ( [ e.pairhij ] ) a term proportional to @xmath151 . obviously , for nonvanishing values of @xmath152 the presence of such term prevents the factorized state from being an eigenstate of the hamiltonian of the system . on the contrary , if @xmath153 and hence if @xmath154 the extra hamiltonian term vanishes and the factorized state @xmath155 remains an energy eigenstate of the system for all values of the external field . this result can be understood from a different point of view . when @xmath156 , from eqs.([e.solcondition ] ) we immediately obtain that @xmath157 , which implies that the total magnetization along the @xmath45 axis is preserved , @xmath158=0 $ ] . therefore the hamiltonian of the system and the total magnetization operator admit a complete set of common eigenstates , discriminated by the eigenvalue of the total magnetization : the factorized state is the one ( and only one ) characterized by the maximum eigenvalue . we now move on to show that the factorized eigenstate , in the case of saturation , is the system s ground state in an infinite range of values of the external field . here and in the following we recall that since the external field is directed along the conventional @xmath45 direction , only saturation with spin alignment along the @xmath45-axis is allowed . at very low values of @xmath44 the ground state of the system will possess a small total magnetization along the field direction . however , as @xmath44 increases , also the total ground - state magnetization along the @xmath45 direction will increase until all spins align with the field and saturation occurs . by increasing @xmath44 further , it is not possible for the system to evolve in a state with higher values of the magnetization along the field direction . hence , the factorized state remains the ground state of the system for all values of the external field @xmath159 , where by @xmath160 we denote the value of the field for which the system reaches saturation ( `` saturating field '' ) . let us consider the matrix representation , similarly to eq . ( [ e.matrix ] ) , of the pair hamiltonian at arbitrary values of @xmath44 . taking into account the condition @xmath153 we have @xmath161 it is immediate to verify that if @xmath162 , the projection of the factorized state onto the considered pair of spins is associated to the lowest energy . re - summing over all values of the spatial range index @xmath49 and taking into account the coordination number @xmath68 of the lattice , we have that the factorized state is the ground state of the total system if @xmath163 eq . ( [ e.hfsaturation ] ) is the rigorous condition for saturation . we notice that it coincides with eq . ( [ e.factorizingfield ] ) in the limit @xmath164 , meaning that in this simple instance no ground - state factorization occurs other than saturation . in the following sections we apply the general theory of factorization to different quantum spin models . we will derive a series of exact results concerning the occurrence of factorized ground states in systems with various types of two - body interactions of different spatial ranges defined on regular lattices of arbitrary spatial dimensionality . the general method enables to determine many novel exact factorized ground state solutions as well as the existence and the nature of ordered phases in models that are , in general , non exactly solvable . moreover , it allows to recover quite straightforwardly the existing analytical and numerical results . in the process , we will come to appreciate that ground state factorization is a phenomenon more common than previously believed . determining the existence of exact factorized solutions in non exactly solvable models should in turn allow to envisage controlled approximation schemes , e.g. perturbative or variational , to investigate the physics of non exactly solvable quantum spin models in the vicinity of quantum factorization points . in this and in the following sections we will consider quantum spin models of the @xmath1 type eq . ( [ e.hamiltonian ] ) defined on regular lattices of arbitrary dimension . let us first consider models with short - range interactions including only nearest - neighbor couplings : @xmath165 as in sec . [ sec.method ] we restrict ourselves to the case in which such models are defined on regular lattices of a generic dimension , whose geometry does not induce any frustration , i.e. in which any closed loop is characterized by an even number of spins . the absence of geometrical frustration together with the fact that all the couplings beyond the nearest neighbor spins vanish , ensure that it is always possible for the system to satisfy all the conditions on the net magnetic interactions . therefore , as shown in sec . [ sec.method ] , if the r.h.s . of eq . ( [ e.solcondition ] ) or , better , if the r.h.s . of the equation relative to each possible magnetic order ( see tab . [ t.summary ] ) takes an acceptable value ( @xmath166 ) then the system admits a fully factorized ground state . in the case of antiferromagnetic ordering along the @xmath69 direction , which is obtained if @xmath167 , from eq . ( [ e.solcondition ] ) it follows that a factorized ground state exists if @xmath168 . analogous conditions are obtained for the other types of magnetic orders compatible with full ground state separability . the results are summarized in fig . [ f.factor ] , where @xmath169 is set assume positive values for reference . a very similar diagram of factorization holds in the case @xmath170 . and the coupling @xmath169 is set to be positive for reference . analogous results hold in the case @xmath170 . blank regions : no factorization allowed . rightmost ( yellow online ) shaded region : presence of a fully factorized ground state supporting an antiferromagnetic order along the @xmath69 direction . leftmost ( light blue online ) shaded region : fully factorized ground state with a ferromagnetic order along @xmath69 . bottommost ( light red online ) shaded region : fully factorized ground state with a ferromagnetic order along the @xmath71 direction . topmost ( light green online ) shaded region : fully factorized ground state with an antiferromagnetic order along @xmath71 . the solid line @xmath172 corresponds to models that exhibit saturation rather than true factorization ; the dashed line @xmath173 accommodates for models with unresolved superpositions of two fully factorized ground states each one supporting a different magnetic order . either one of the two possible ground states is then `` chosen '' by the system depending on the way the symmetry is broken , see text for further details . all plotted quantities are dimensionless.,width=321 ] as one can see from fig . [ f.factor ] , the type of magnetic order corresponding to the factorized ground state determines four different regions in the space of the hamiltonian parameters ( one for each different ordering ) associated to the existence of a factorized ground state . the four different domains of the interaction parameters are separated by two bisectrices that individuate special cases . the line @xmath172 is constituted by the set of points in which saturation occurs in place of genuine factorization ( see subsec . [ s.sec.saturation ] ) . on the other hand , the line @xmath174 , corresponds to models that , for a given value of the factorizing field @xmath63 and of the e - squo orientation @xmath86 , admit two degenerate , but distinct factorized states . the choice of the ground state is then determined by the usual mechanisms of symmetry breaking : if an infinitesimal magnetic field is applied in the @xmath55 plane , the axis of magnetic alignment in the ground state is determined by the order in which the two limits @xmath175 and @xmath176 are realized . it is important to observe that all the shaded regions of fig . [ f.factor ] correspond to exact analytical ground - state solutions of the investigated short - range models , which are , in general , non - exactly solvable . the exact , fully factorized ground states are tensor products of the form reported in the fourth column of tab . [ t.summary ] , in which the value of the orientation parameter @xmath86 of the e - squo is retrieved by solving the equation in the second column of tab . [ t.summary ] . obviously , such an exact solution holds only at the factorizing field @xmath101 . in the case of antiferromagnetic order either along the @xmath69 or the @xmath71 direction , the latter is given by ( see tab . [ t.summary ] ) @xmath177 while in the case of a ferromagnetic phase , again unrespectfully if along @xmath69 or along @xmath71 , it reads @xmath178 given that the ground state is known exactly at the factorization point , all the relevant physical observables can be evaluated straightforwardly . for instance , the ground state energy density at factorization reads @xmath179 in the presence of an antiferromagnetic ordering , while in the ferromagnetic case it is @xmath180 both the factorizing field and the ground state energy , unlike the e - squo orientation @xmath86 and the expression of the factorized ground states , are functions of the number of nearest neighbors per site ( the coordination number @xmath181 , where the index @xmath182 denotes the nearest - neighbor range of the interactions ) . hence , they depend on the geometry and the spatial dimension of the lattice . to begin with , let us consider one - dimensional lattices ( spin chains , @xmath183 ) , for which the coordination number @xmath184 . substituting this value in the previous equations , we immediately recover , among others , the value of the factorizing field and the ground state energy obtained analytically by kurmann _ et al . _ for the @xmath1 spin chain using the direct method @xcite . et al . _ had to restrict their investigation to coupling constants @xmath185 taking values in the interval @xmath21 $ ] , thus establishing ground state factorization only in a restricted set of conditions . instead , the general method allows to assess the occurrence of factorized ground states associated to different types of magnetic orderings in the entire region of the hamiltonian parameters . one of the most important advantages of the general method is that it is easily applicable irrespective of the spatial dimensionality , while the direct method becomes increasingly difficult to apply with growing spatial dimensions of the lattice . numerical techniques can overcome some of the limits of the direct method , although they fall short of the power and extension of the general analytic method . the most important numerical result has been obtained by applying quantum monte carlo techniques to the study of the @xmath1 model in @xmath186 space dimensions ( square lattice ) @xcite . applying the present method to the @xmath1 model on the square lattice allows not only to immediately reobtain analytically the previous numerical findings , but also , again , to extend the investigation to the entire space of the hamiltonian parameters by simply taking into account that for the square lattice @xmath187 . moreover , the flexibility of the general method allows to include in the analysis different planar models . for instance , considering models defined on planar hexagonal lattices , the previous analysis needs to be modified only in that the value of the coordination number , in this case is @xmath188 . let us now move to @xmath189 and investigate ground state factorization in nearest - neighbor @xmath1 models on three - dimensional regular lattices , for which there were so far no results ( apart the preliminary ones obtained in our previous work @xcite ) either analytic or numerical ( such a task would be extremely demanding if approached either via the direct method or via numerical techniques ) . resorting to the general method , we can instead fully characterize the factorized ground state and the associated physical quantities just by entering the correct value of the coordination number , e.g. @xmath190 for a simple cubic lattice . obviously , the results extend straightforwardly also for models defined on lattices of arbitrary higher dimension @xmath191 . the analytic method can be applied with minor complications to models with interactions of longer spatial range . as already mentioned in sec . [ sec.method ] , concerning ground state factorization , the main difference between models with short- and finite - range interactions is that for the latter there exist more than one equation in the set ( [ e.solcondition ] ) . each of these equations is associated to a different value of the range @xmath127 , where @xmath125 is the maximum interaction range , that is the maximum distance between two directly interacting spins : all the couplings vanish for pairs of spins with inter - spin distance greater or equal to @xmath192 . clearly , the fact that all the equations have to be solved simultaneously yields tighter conditions on factorizability compared to the case of models with nearest - neighbor interactions . namely , it is not sufficient that the r.h.s . of each eq . ( [ e.solcondition ] ) takes values in the interval @xmath21 $ ] . it is also necessary to require that all these values coincide . this constraint yields a set of further conditions that must be satisfied by all the nonvanishing couplings . for instance , let us suppose that the system is in an antiferromagnetic ordered phase along the @xmath69 direction . labeling by @xmath193 the value of @xmath131 and imposing that for every @xmath49 the r.h.s . of eqs.([e.solcondition ] ) must equate @xmath193 , we find that @xmath194 where , to ensure the absence of frustration , @xmath195 must obey the following inequality : @xmath196 . analogously , for an antiferromagnetic order along @xmath71 : @xmath197 and to exclude frustration we must also have that @xmath198 . in the presence of a ferromagnetic phase , respectively for an ordering along @xmath69 and along @xmath71 , the constraints and the no - frustration condition become @xmath199 eqs . ( [ e.exe.cantix][e.exe.cferroy ] ) generalize the instance analyzed in ref . @xcite , in which we limited the analysis to the case of vanishing @xmath169 that , according to eqs . ( [ e.exe.cantix][e.exe.cferroy ] ) , implies @xmath200 for all @xmath127 . especially in the presence of a ferromagnetic order in the lattice , eqs.([e.exe.cferrox][e.exe.cferroy ] ) allow us to simplify the expressions of both the factorizing field and the energy per site . namely , taking into account the definition of the ferromagnetic net interactions eq . ( [ e.netinteractions ] ) and the expression of the factorizing field in the presence of a ferromagnetic order ( see tab . [ t.summary ] ) , we have @xmath201 respectively for an order along the @xmath69 and the @xmath71 directions . similarly , for the energy density we obtain @xmath202 we now discuss a specific example ( out of many possible ones ) which demonstrates the power of the analytic method in conditions where the more conventional direct approach based on the factorized ansatz of refs . @xcite is completely ineffective without the supplement of a remarkable intuition in guessing the correct forms of the possible factorized ground states . let us consider a @xmath1-type spin model defined on a spatially isotropic cubic lattice of arbitrary size , possessing nearest - neighbor and next - nearest - neighbor interactions , and admitting arbitrarily different couplings along the @xmath69 , @xmath71 , and @xmath45 directions . the coordination numbers associated to the two types of interaction can be straightforwardly computed from the geometry of the problem and are @xmath190 and @xmath203 . suppose the model is deduced from an explicit experimental realization of a spin lattice and we are provided with explicit values of the various interaction strengths . for example , @xmath204 . the application of our `` on - demand '' ground - state factorization search engine is now immediate . from eq . ( [ fourcases ] ) we have that the candidate factorized state must possess antiferromagnetic order along the @xmath69 direction . therefore the general conditions for the existence of a factorization point , as determined by eq . ( [ e.solcondition ] ) , can be summarized as follows @xmath205 this condition must hold together with the frustration - free constraint . it is readily verified that @xmath206 provides a solution to the combined set of necessary and sufficient conditions for ground state factorization . therefore if the experimentalist tunes the external field @xmath44 at the factorizing value @xmath207 , obtained from eq . ( [ e.factorizingfield ] ) , the engineered cubic spin lattice system will relax in a completely factorized ground state of the form given in table [ t.summary ] ( second row , last column ) with @xmath208 . many other applications can be considered to models with arbitrary spin - spin interactions of longer range , and they can be treated with the same simplicity within the framework of the general analytic method . for instance , a very interesting example of a model with infinite - range interactions will be treated in the next section . however , before moving to the study of this case , we should comment on the important issue of the interplay between factorization and frustration in quantum spin systems with multiple spatial scales of interaction . the investigation of general models including many interaction terms of different spatial ranges needs to be carried out with particular care , because the interplay between the different interactions can lead in general to important effects of frustration that tend to suppress the occurrence of ground state factorization . the very delicate and tricky nature of the problem is at the origin of a recent incorrect prediction about the existence of factorized ground states in one - dimensional @xmath1 models on finite rings with long - range interactions , obtained exploiting the direct method @xcite . in ref . @xcite the author , resorting to the direct method , concludes that all the factorized energy eigenstates in one - dimensional long - range @xmath1 models , with interaction terms of spatial ranges that are in integer ratio with the total number of sites , are also the states of lowest energy . however , the proof of this statement obtained using only the direct method requires that there must be no frustration in the system @xcite . unfortunately , the interplay between the interactions of different spatial range can in general introduce frustration effects ( e.g. in the case of antiferromagnetic nearest - neighbor and next - to - nearest neighbor interactions , and so on ) . the existence of these effects nullifies all proofs , based on the direct method , of the existence of fully separable ground states . in fact , immediate counterexamples can be given of fully factorized energy eigenstates that are not ground states . let us consider for instance an @xmath209 model on a finite ring of six spins with nearest - neighbor and next - to - nearest - neighbor antiferromagnetic interactions , and @xmath210 . following ref . @xcite , this model should admit a factorized ground state . in fact , a factorized energy eigenstate with antiferromagnetic order does indeed exist , with an energy @xmath211 . however , solving the model by exact diagonalization yields that the lowest energy is @xmath212 and that the associated ground state is entangled . essentially the same type of counterarguments nullifies a recent claim that the factorized energy eigenstates are the factorized ground states of mixed - spin models of ferrimagnetism of the @xmath1 type @xcite . a general analysis of ground state factorization in ferrimagnetic models with mixed spin-@xmath0-spin-@xmath182 interactions will be presented elsewhere @xcite , based on a proper extension of the formalism of squos to include spin-@xmath182 systems , as sketched in the final part of ref . @xcite . in conclusion , the study of factorization in quantum spin models with many different finite- and long - range interaction terms requires a very careful analysis and the use of the general analytic method whenever frustration effects are present . a rigorous and systematic study of the crucial ( and subtle ) interplay between frustration and factorization in quantum spin systems will be the subject of a forthcoming paper @xcite . a very interesting limit of the cases analyzed in subsec . [ sec.long ] is given by models with infinite range interactions , such as the fully connected or lipkin - meshkov - glick ( lmg ) model @xcite , extensively studied in condensed matter physics @xcite , which is described by the following hamiltonian @xmath213 with @xmath214 . the pre - factor @xmath215 ( with @xmath2 denoting the total number of spins ) ensures the linear divergence of the ground state energy as @xmath2 increases . let us remark that normalizing the interaction part of the lipkin - meshkov - glick hamiltonian requires some care : being a connected interaction , it scales as @xmath216 , i.e. as the product of the total number of spins and the number of pairs of spins . therefore , to ensure that the interaction energy is extensive , i.e. scales with @xmath2 , the interaction part of the hamiltonian must be normalized by @xmath29 . usually , for calculational ease , the interaction part is normalized by @xmath2 , so that it scales as @xmath29 . using this convention is strictly speaking incorrect , but is clearly harmless for large @xmath2 and in the thermodynamic limit . for instance , concerning the separability of the ground state , it introduces a slight , spurious @xmath2-dependence of the factorization point through the multiplicative factor @xmath217 , and the error vanishes in the thermodynamic limit . however , on fundamental physical grounds , the factorization points of translationally invariant systems , if they exist , must be independent of the size of the system , and must take the same value both at finite @xmath2 and in the thermodynamic limit . applying the kurmann - thomas - mller factorized ansatz @xcite and the @xmath218 normalization , dusuel and vidal obtained an analytic expression of the factorizing field for the lmg model that is @xmath2-dependent through the multiplicative factor @xmath217 @xcite . although defined according to an anomalous convention , this result is essentially exact in the limit of large @xmath2 . we first observe that the lmg model in eq . ( [ e.exe.hlmg ] ) can be obtained as the limiting case of a generic model described by eq . ( [ e.hamiltonian ] ) assuming that the range of interactions diverges ( i.e @xmath219 ) while the individual strengths become independent of the distance ( @xmath220 ) , scale with @xmath2 ( @xmath221 ) , and vanish along the @xmath45-direction ( @xmath222 ) . therefore , assuming @xmath223 , @xmath224 and @xmath222 in eq . ( [ e.hamiltonian ] ) we recover eq . ( [ e.exe.hlmg ] ) . because of the sign of the interactions and of the range of values that can be assumed by @xmath225 , we immediately obtain that the energy per site ( [ e.energy ] ) is minimized by a state possessing ferromagnetic order along the @xmath69 direction . moreover , taking into account that the coupling strengths are proportional to the ratio @xmath218 , we have that the net interactions converge to finite values in the thermodynamic limit : @xmath226 , @xmath227 . it is then rather straightforward to apply the analytic method to the lmg model and hence prove , exactly , that the ground state is fully factorized at the factorizing field @xmath228 , which is , correctly , independent of @xmath2 . the form of the fully separable ground state is the tensor product of local states of the form @xmath229 with @xmath230 and the energy density reads @xmath231 . these results are in agreement with those obtained , using the direct method , by dusuel and vidal @xcite , modulo a proper normalization factor . the versatility and generality of the analytic method can be demonstrated further by considering extensions of the translationally - invariant models discussed in the previous sections . in fact , till now , we have always considered systems in which the coupling amplitudes depend only on the distance between the spins involved . in the present subsection we shall focus instead on models which contemplate the possibility of spatial anisotropies , i.e. given the spin @xmath232 associated to site @xmath40 , the interaction coupling with spin @xmath233 associated to site @xmath109 depends not only on the distance @xmath234 but also on the location of site @xmath109 relative to site @xmath40 . to simplify the notations , the physical analysis , and the number of possible cases that need to be considered , we limit ourselves to models with short - range interactions , i.e. with all the couplings vanishing for @xmath235 . we will comment briefly on general models with spatial anisotropies and interactions of arbitrary range at the end of this section . let us assume that for each site @xmath40 the total number of nearest neighbors is @xmath236 , where @xmath237 is the number of nearest neighbors whose coupling strength with @xmath232 takes a certain value @xmath238 , and @xmath239 is the number of nearest neighbors whose coupling with @xmath232 is given by a different value @xmath240 . obviously , we could consider situations more complicated at will , with an arbitrary number of @xmath241 different types of nearest neighbors . for the given example , the system hamiltonian reads @xmath242 , \nonumber \\\end{aligned}\ ] ] where the index @xmath243 runs over the first @xmath237 nearest neighbors , and the index @xmath244 runs over the remaining @xmath239 nearest neighbors for each spin @xmath232 . according to the different values of @xmath237 and @xmath239 this hamiltonian describes various possible models . for instance , if @xmath245 and @xmath246 eq . ( [ e.exe.hamiltonian.anis ] ) describes @xmath1 models on ladders constituted by two coupled linear chains ; the case @xmath247 and @xmath246 corresponds to models defined on two coupled square lattices . in both cases , every spin in each chain ( plane ) , besides the usual interactions within a single chain ( plane ) involving @xmath237 nearest neighbors , is also coupled to @xmath239 companion spins located on the second chain ( plane ) . the instance @xmath248 corresponds to a @xmath1-type model defined on a linear chain with nearest - neighbor interactions of alternating strengths . the factorization properties of this latter model have been studied by g. l. giorgi in the limit @xmath249 that , belonging to the @xmath209 symmetry class , can be solved exactly @xcite . as we have seen in sec . [ sec.method ] , the first step of the method is to single out the magnetic order that minimizes the energy associated to the candidate factorized ground state , at fixed magnetization along the @xmath45 direction ( the direction of the external field ) . provided there are no effects of frustration , in the present case we obtain that every possible hypothetical factorized ground state can assume one of eight different magnetic orders . these eight possibilities stem from the four possible orders of eq . ( [ fourcases ] ) , that the system can support when the effects of @xmath240 can be neglected , duplicated according to the two possible arrangements , parallel or antiparallel , induced by the interactions associated to @xmath240 . the four possible orderings along a single direction are represented in fig . [ f.disegno ] for a model of two transversally coupled one - dimensional spin chains . or @xmath71 ) characterizing the candidate factorized ground states of models with spatially anisotropic interactions . the case considered corresponds to a ladder of two coupled one - dimensional spin chains with coordination numbers @xmath245 and @xmath246 . topmost row : parallel ( a ) and antiparallel ( b ) ferromagnetic order . bottommost row : parallel ( c ) and antiparallel ( d ) antiferromagnetic order.,width=321 ] by imposing the minimization of the energy per site , eq . ( [ e.energy ] ) , the kind of order actually present in the ground state can be determined comparing the antiferromagnetic and ferromagnetic net interactions along the @xmath69 and @xmath71 axes , defined as @xmath250 where the @xmath84 sign discriminates between parallel / antiparallel ordering with respect to @xmath251 . as a function of the net interactions , the type of magnetic order present in the system is determined by the value of the minimum , i.e. of @xmath252 . accordingly , we have the following correspondences : @xmath253 obviously , also in the presence of spatial anisotropies we may observe that , in terms of the net interactions , the fact that in the system there is no frustration ( and especially no frustration arising from the spatial anisotropies ) implies that latexmath:[$\mu = - z_1^{(1 ) } according to which axis is characterized by the given magnetic order . without loss of generality , let us consider the situation in which the system realizes a parallel antiferromagnetic order along the @xmath69 direction , see fig . [ f.disegno](c ) . all other cases can be treated analogously . following once again the steps described in sec . [ sec.method ] , and in complete analogy with eq . ( [ e.condition ] ) , it is possible to derive the set of conditions that the orientation @xmath86 of the e - squo must satisfy simultaneously in order for the system to admit a factorized energy eigenstate : @xmath255 for the two conditions to be satisfied simultaneously the two r.h.s . must coincide and must assume values in the interval @xmath21 $ ] . hence , by equating them it is possible to derive an equation that plays the same role of eqs . ( [ e.exe.cantix][e.exe.cferroy ] ) . in the hypothesis that the model under investigation satisfies both eqs . ( [ e.exe.theta.anis.1][e.exe.theta.anis.2 ] ) , what is left to prove is that the associated factorized state be indeed the ground state . following the route illustrated in sec . [ sec.method ] it is not difficult to verify that the following inequalities must hold : @xmath256 it is straightforward to establish that they are both satisfied , provided there are no frustration effects . namely , for spins interacting via couplings of the first kind @xmath238 ( referring to fig . [ f.disegno ] , these are the intra - chain interactions ) , the second of eqs . ( [ e.exe.suffcondition.anis.1 ] ) yields that the denominator in the r.h.s . of eq . ( [ e.exe.theta.anis.1 ] ) is non - negative and , therefore , the corresponding numerator must be non - negative and not exceeding the denominator . hence , we obtain @xmath257 . on the other hand , from the first of eqs . ( [ e.exe.suffcondition.anis.1 ] ) it follows that @xmath258 which , in turn , implies @xmath259 . the latter relation is in agreement with the hypothesis of absence of frustration . furthermore , for pairs of neighboring spins interacting with a coupling of the second kind @xmath240 ( referring to fig . [ f.disegno ] , these are the inter - chain interactions ) , the second of eqs . ( [ e.exe.suffcondition.anis.2 ] ) yields that the denominator in the r.h.s . of eq . ( [ e.exe.theta.anis.2 ] ) must be non - positive . hence , also the corresponding numerator must be non - positive and therefore @xmath260 . simultaneously , the first of eqs . ( [ e.exe.suffcondition.anis.2 ] ) implies @xmath261 and therefore @xmath262 , again in complete agreement with the hypothesis of absence of frustration . collecting all these results we can conclude that in the absence of frustration , if a system described by the model hamiltonian eq . ( [ e.exe.hamiltonian.anis ] ) satisfies simultaneously all conditions in eqs.([e.exe.theta.anis.1][e.exe.theta.anis.2 ] ) , there exist a factorizing field @xmath63 at which the ground state is fully factorized and characterized by a parallel antiferromagnetic order along @xmath69 . the factorizing field is determined according to the procedure described in sec . [ sec.method ] , and it is not difficult to verify that it reads : @xmath263 the exact form of the factorized ground state reads @xmath264 in eq . ( [ e.antiferroparrallelolungox ] ) , the superscript `` @xmath265 '' ( `` @xmath266 '' ) denotes the upper ( lower ) chain of the ladder model considered in fig . [ f.disegno](c ) . we observe that @xmath267 does not depend on the choice of the chain , implying that the two generic @xmath4-th spins in both legs of the ladder are in the same state equipped with an antiferromagnetic order with respect to the nearest neighbor sites along the corresponding chain . analogous results hold , with the appropriate trivial modifications , when considering the other possible magnetic orders compatible with factorization ( see eq . ( [ e.exe.eightcases ] ) ) . the present work has been motivated both by the need for exact solutions to complex many - body quantum spin models , in particular by the question `` when is a mean - field solution exact ? '' , and by the necessity to acquire improved knowledge and control on the structure of quantum correlations for potential technological applications of spin systems . we have introduced a simple and rather powerful all - analytic general method to determine the conditions for the existence and the properties of fully factorized ( fully separable ) ground states in translationally invariant quantum spin models with general two - body exchange interactions and subject to external fields . the theory developed in the present paper builds on and extends the scheme originally introduced in ref . it provides a readily useful procedure to establish , in terms of the hamiltonian parameters , whether the ground state is entangled or completely factorized and , in the latter case , what is the exact form of the state , the expression of the ground - state energy and of the magnetic observables , and the type of ordered phase ( magnetic order ) compatible with factorization . the task is achieved by exploiting tools imported from quantum information theory such as the formalism of single - qubit unitary operations introduced in refs . @xcite and the corresponding entanglement excitation energies , whose property of vanishing if and only if a quantum ground state is totally factorized plays a central role in the analysis of ground - state separability . using these tools we have completed the program initiated in ref . @xcite by determining a general set of exact relations and inequalities that , for frustration - free systems , allow to establish rigorously the occurrence of all types of possible factorized ground states . we have applied the analytic method to spin-@xmath0 models with general heisenberg - like exchange interactions of arbitrary range ( short , finite , long ) , either isotropic or anisotropic , defined on regular lattices of arbitrary dimensions . the method is insensitive to the size of the system and applies rigorously in the thermodynamic limit as well as for finite lattices . the method allows to establish novel exact , fully factorized ground state solutions of generally non exactly solvable models , corresponding to nontrivial sets of values of the hamiltonian parameters . key results include , for instance , the rather straightforward determination of the existence and form of factorized ground states in @xmath1-type models with different types of short - range and long - range interactions on cubic lattices , a task that would be of formidable complexity if tackled with numerical techniques or resorting to the simple - minded direct method based on the product ansatz . the complete factorization diagram for nearest - neighbor heisenberg - like anisotropic models might be particularly useful for those technological implementations which employ spin systems as information processors exploiting their ground state entanglement @xcite . we have provided in fig . [ f.factor ] a `` minefield '' map of the unentangled working points that need to be avoided , when engineering the couplings and tuning the external fields , in order to achieve satisfactory quantum transmission performances . on the other hand , there are also alternative tasks in quantum information which instead require as initial working points factorized states , e.g. for engineering , via suitable dynamics , specific classes of long - distance entangled states useful for quantum state transfer and quantum teleportation @xcite or the instantaneous creation of strongly entangled graphs or cluster states involving a mesoscopic or macroscopic number of spins for one - way quantum computation @xcite : in these instances the minefield may turn into a treasure map . the analysis of ground - state factorization would be particularly useful in the study of quantum spin models on open - end lattices . such systems are not translationally invariant , and therefore the analytic method requires to be suitably extended to this type of instances . this generalization is under way and should be in reach in the near future . notably , according to a general theorem by kurmann , thomas , and mller on factorization @xcite , given any hamiltonian of the form eq . ( [ e.hamiltonian ] ) involving higher spins @xmath268 , the ground state of that spin-@xmath269 system is fully factorized at the same value of the external field @xmath101 eq . ( [ e.factorizingfield ] ) , at which factorization occurs in the corresponding spin-@xmath0 model . therefore , the method and the results derived in the present paper have a much wider scope of application and can be straightforwardly generalized to interacting systems with arbitrarily high values of the spin , provided they are of the same hamiltonian structure as in the spin-@xmath0 case . absence of frustration effects , either of geometric or dynamical nature , has been essential to the theoretical scheme derived in the present work . for instance , we have determined the factorization diagram for specific examples of frustration - free systems endowed with isotropic interaction terms of different spatial ranges , and for systems with nearest - neighbor anisotropic interactions . however , it is clear that there are many extremely interesting instances of more complex systems that would in general be subject to frustration effects . continuing a step - by - step strategy of applying the general method to models of increasing complexity , natural further stages of investigation would include models subject to frustration , e.g. , models on linear chains with competing isotropic interactions of different range such as a 1-d @xmath1-type model with nearest - neighbor ferromagnetic interactions and next - to - nearest neighbor antiferromagnetic interactions . increasing further the degree of complexity , one could then consider models with competing anisotropic interactions of different ranges , that according to the type of anisotropy could be subject to both geometrically- and dynamically - induced frustrations . when considering frustrated models , except for very special cases , even when the considered system admits a fully factorized state as an energy eigenstate , the set of conditions derived in the present work are not sufficient to establish that such an eigenstate is indeed the ground state of the system . in fact , a competition arises between factorization , which requires a magnetic ordering , and frustration , which tends to destroy magnetic orders , ultimately leading to a chaotic correlating behavior . the latter unavoidably tends to enhance ground - state entanglement , thus suppressing the occurrence of situations where the most favorable state , in terms of energy content , is completely uncorrelated . in such cases , a naive application of the direct method based on the factorized ansatz , for instance to models including anisotropic interaction terms of different magnetic types on many different spatial scales ( or even on all scales ) , leads unavoidably to overestimate the factorizing effect of the balancing between interactions and external fields and thus to badly incorrect predictions on the occurrence of fully factorized ground states in these classes of models , as done , unfortunately , in the second part of ref . @xcite . in a forthcoming work @xcite we will present a systematic and thorough study of some important classes of frustrated quantum spin models in order to establish the conditions for the occurrence of true ground state factorization and to characterize it in terms of compatibility thresholds with frustration . an intriguing issue that arises in connection with frustration is the relationship between factorization and the existence of ordered phases . for frustration - free hamiltonian systems that conserve the even - odd parity , spontaneous symmetry breaking and the existence of an ordered phase are necessary conditions for ground state factorization , because factorized states have no definite parity . however , in the presence of frustration there may exist factorized energy eigenstates without ground state factorization . it would then be very interesting to understand whether in these situations there is still a causal relation between factorization and the existence of phase transitions in the system . factorization might then be used as an heuristic tool to gain insight in the phase diagram of frustrated quantum systems . it would also be important to understand whether chiral interactions can enhance factorization beyond the limits imposed by the presence of frustration effects . in fact , since factorization can be seen as `` mean field becoming exact '' , it seems to require quite naturally a balancing between interactions and external fields as the only possibility for its occurrence . however , one can not exclude , in principle , that other mechanisms might lead to the same effect without requiring the presence of external driving fields . future investigations will be concerned with the application of the analytic method to the determination of the spectrum of factorization , that is the study of the occurrence of factorized excited states in different classes of quantum spin hamiltonians , and the implications of factorization diagrams on the structural and informational properties of quantum spin models for potential applications in quantum technology . we acknowledge cnr - infm research and development center `` coherentia '' , infn , and isi foundation for financial support . one of us ( f.i . ) , would like to thank ylenia dautilia for invaluable exchanges . in sec . [ sec.method ] we have determined the general conditions for ground state factorization by selecting as candidate factorized ground states only those characterized by a ferromagnetic or antiferromagnetic order along the @xmath69 or @xmath71 directions depending on the value of the minimum value @xmath270 in the set of the net interactions [ see eq . ( [ fourcases ] ) ] , and neglecting all other possible magnetic arrangements . in the following , we will prove that indeed , given any spin hamiltonian of the type eq . ( [ e.hamiltonian ] ) , if a factorized ground state exists , then it is characterized by one of the four above - mentioned magnetic orders . let us consider the most general factorized state compatible with the constraint of a site - independent magnetization @xmath51 : @xmath271 where the local spin state @xmath272 is eigenstate of @xmath273 with eigenvalue equal to @xmath0 . let us suppose that there exists an hamiltonian of the type eq . ( [ e.hamiltonian ] ) that admits , for some particular value of the external field @xmath101 , the state in eq . ( [ e.candidate.gen ] ) as an eigenstate . under these hypotheses we must have that the projection of the factorized eigenstate onto the four - dimensional hilbert space of a pair of spins in the lattice must be an eigenstate of the pair - hamiltonian term @xmath102 defined as in eq . ( [ e.couple ] ) . following the same strategy as in sec . [ sec.method ] let us introduce for each site @xmath274 a set of three auxiliary mutually orthogonal spin operators among which one , that we name @xmath123 , coincides with @xmath275 : @xmath276 by inverting eqs . ( [ e.axayaz.gen ] ) , we can conveniently re - express the conventional single - spin operators as functions of the new set of operators @xmath113 : @xmath277 substituting eq . ( [ e.sxsysz.gen ] ) into the pair hamiltonian we obtain the expression for @xmath115 at factorization as a function of the sets of operators @xmath113 and @xmath116 , which reads @xmath278 -2 h_f^r \sin \theta \left [ { a_{\underline{i}}^x \left ( a_{\underline{j}}^z -\frac{1}{2}\right ) + \left ( a_{\underline{i}}^z-\frac{1}{2}\right ) a_{\underline{j}}^x } \right ] \nonumber \\ & & + a_{\underline{i}}^x a_{\underline{j}}^x \left ( j_z^r-2 h_f^r \cos \theta\right)+ a_{\underline{i}}^y a_{\underline{j}}^y \left ( j_x^r \sin\varphi_{\underline{i } } \sin \varphi_{\underline{j } } + j_y^r \cos \varphi_{\underline{i } } \cos \varphi_{\underline{j } } \right ) - h_f^r \cos \theta \left ( { a_{\underline{i}}^z + a_{\underline{j}}^z } \right ) \nonumber \\ & & + \left ( a_{\underline{i}}^z a_{\underline{j}}^y \cos \theta + a_{\underline{i}}^x a_{\underline{j}}^y \sin \theta \right)\left(-j_x^r \cos \varphi_{\underline{i } } \sin \varphi_{\underline{j}}+ j_y^r \sin \varphi_{\underline{i } } \cos \varphi_{\underline{j } } \right ) \\ & & + \left ( a_{\underline{i}}^y a_{\underline{j}}^z \cos \theta + a_{\underline{i}}^y a_{\underline{j}}^x \sin \theta \right)\left ( -j_x^r \sin \varphi_{\underline{i } } \cos \varphi_{\underline{j}}+ j_y^r \cos \varphi_{\underline{i } } \sin \varphi_{\underline{j } } \right ) \ , , \nonumber\end{aligned}\ ] ] > from eq . ( [ e.pairhij.gen ] ) , one has that to ensure that the projection of the factorized state onto the four - dimensional hilbert space of spins @xmath232 and @xmath233 is an eigenstate of the pair hamiltonian , it is necessary to require that the terms associated to the operators @xmath281 , @xmath282 , @xmath283 and @xmath284 vanish . this condition yields that @xmath285 solving eq . ( [ e.condition2.1.gen ] ) and eq . ( [ e.condition2.2.gen ] ) we obtain that if @xmath286 , then the phases @xmath287 obey to the following relations : @xmath288 with @xmath241 integer . it is immediate to verify that these relations are consistent only with a ferromagnetic or with an antiferromagnetic order along the @xmath69 or the @xmath71 direction . this concludes the proof . g. vidal , j. i. latorre , e. rico , and a. kitaev , phys . * 90 * , 227902 ( 2003 ) ; v. e. korepin , phys . lett . * 92 * , 096402 ( 2004 ) ; g. c. levine , phys . rev . lett . * 93 * , 266402 ( 2004 ) ; g. refael and j. e. moore , phys . 93 * , 260602 ( 2004 ) ; p. calabrese and j. cardy , j. stat . mech . * p06002 * ( 2004 ) ; m. b. plenio , j. eisert , j. dreiig , and m. cramer , phys . rev . lett . * 94 * , 060503 ( 2005 ) ; m. cramer , j. eisert , and m. b. plenio , phys . rev . lett . * 98 * , 220603 ( 2007 ) ; j. eisert , m. cramer , and m. b. plenio , arxiv:0808.3773 . g. vidal , phys . * 99 * , 220405 ( 2007 ) ; m. rizzi , s. montangero , and g. vidal , phys . rev . a * 77 * , 052328 ( 2008 ) ; v. giovannetti , s. montangero , and r. fazio , phys . rev . lett . * 101 * , 180503 ( 2008 ) . s. bose , phys . lett . * 91 * , 207901 ( 2003 ) ; m. christandl , n. datta , a. ekert , and a. j. landahl , phys . lett . * 92 * , 187902 ( 2004 ) ; d. burgarth and s. bose , phys . rev . a * 71 * , 052315 ( 2005 ) ; a. kay , phys . rev . a 73 , 032306 ( 2006 ) . y. makhlin , g. schn , and a. shnirman , rev . * 73 * , 357 ( 2001 ) ; v. giovannetti , s. lloyd , and l. maccone , science * 306 * , 1330 ( 2004 ) ; f. dellanno , s. de siena , and f. illuminati , phys . rep . * 428 * , 53 ( 2006 ) . m. j. hartmann , f. g. s. l. brando , and m. b. plenio , phys . . lett . * 99 * , 160501 ( 2007 ) ; d. g. angelakis , m. f. santos , and s. bose , phys . a 76 , 031805(r ) ( 2007 ) ; a. kay and d. g. angelakis , europhys . lett . * 84 * , 20001 ( 2008 ) . a. srensen and k. mlmer , phys . 83 * , 2274 ( 1999 ) ; a. b. kuklov and b. v. svistunov , phys . lett . * 90 * , 100401 ( 2003 ) ; l .- m . duan , e. demler , and m. d. lukin , phys . * 91 * , 090402 ( 2003 ) . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: recently there has been much progress in the study of synchronization of nonlinear maps @xcite and flows @xcite . this may lead to the clarification of the intelligence supposed to come from the synchronization among the neurons in the neural network , e.g.@xcite . especially the globally coupled map lattice ( gcml ) , a system of identical maps coupled via their mean field , has surprisingly rich clustering phases in the space of the nonlinearity parameter and the coupling . in the ordered two - cluster phase , gcml exhibits a posi - nega switch between clusters of synchronizing maps controlled by external inputs@xcite . furthermore even in the turbulent regime there emerges some hidden correlation which was first observed as the violation of the law of large numbers ( lln ) in the fluctuation of the mean field in the time series@xcite . gcml is a natural extension of the spin - glass models@xcite and these interesting features may be relevant to the intelligence activity . all the maps in gcml contribute to the mean field with an equal weight and hence there is no notion of distance between maps . it is conjectured that the intriguing properties of gclm , both the posi - nega switch and the hidden coherence , come from this bona fide scaling invariance of gcml@xcite . however , neither the reason why there occurs complete swapping of the maps between clusters in the posi - nega switch in the two - cluster phase nor the global feature of the turbulent phase has been elucidated so far . the purpose of this article is twofold . firstly , we revisit gcml . we show that there is a natural mechanism which guarantees the complete swapping of maps in the two - cluster posi - nega switch . the key to this mechanism is the motion of the mean field during the transient process . we show that even though whole maps are released from tightly synchronizing clusters to chaotic violent movements during the transient process , still they never mix across their mean field . we clarify the mechanism for this phenomenon . we also perform an extensive survey of the turbulent regime of gcml . we show that this regime has a prominent period three window and is full of periodicity remnants . a correlator analysis detects quasi - periodic clusters with high mixing rate among them almost everywhere in the turbulent regime of the large - size gcml . the so - called hidden coherence occurs in - between the regions of the quasi - clusters and the valleys with no violation of lln as the most modest appearance of the periodicity remnants . secondly , we construct a new extended globally coupled map lattice ( egcml ) . our egcml is a two - dimensional map lattice and the maps are under all - to - all interaction as in the gcml . but instead of the uniform interaction in gcml , the interaction in egcml obeys an inverse power law with respect to the distance between the maps on the lattice . egcml locates at an intermediate position between the nearest neighbor coupling map model and the gcml with uniform interaction . thus egcml facilitates a test whether the salient feature of gcml is due to the scale invariant interaction in it . we find that egcml indeed inherits the intriguing properties of gcml though they turn out diffuse to some extent . for instance , in the two - cluster phase of egcml , the two clusters follow the same bifurcation tree with that of gcml when the population unbalance between the clusters are increased but the synchronization of maps in each cluster is looser in egcml than in gcml . we also observe in the turbulent regime of the egcml the same periodic window as well as the quasi - periodic clusters with high mixing rate . gcml exhibits the basic posi - nega switch as well as more intricate controlled switches among coded clusters at various hierarchical levels in its simplest construction@xcite and certainly serves as a model for intelligence activity . but the very fact that it is a system of globally coupled maps with an equal coupling makes it lose the notion of the distance between maps . in a way it is a zero dimensional chaotic field theory . apart from the intelligence one is interested to know what is the ( classical ) field theory of chaotic maps on the lattice at criticality . in this aspect the synchronizing maps on the lattice may be regarded as the analog of higgs condensation in the field theory@xcite . we can use egcml to see the general trend how the spatial clustering of maps occurs under the scale invariant interaction on the lattice . we find that egcml does exhibit a self - organized formation of the spatial clusters ( condensation ) due to the synchronization between the maps during the chaotic transient process in the switch and after it . in the egcml switch the positive ( negative ) cluster becomes negative ( positive ) evolving in the same two - cluster attractor as before with only a few percent change of the population unbalance between the two clusters in most of the runs . however , there occurs the mixing of maps between the clusters . the organization of this article is as follows . in sec . ii we pay attention to two important rules in the gcml . these are the reciprocity of the couplings among the maps and an invariance rule that the maps do not change their mean field under the interaction . we then construct egcml based on these rules . in sec . iii we compare the phase diagrams of both models and show that they are almost the same each other . in sec . iv we analyze the two - cluster regime of both models . for the gcml we clarify how the basic posi - nega switch between two clusters in the two - cluster regime can be realized . we then turn to egcml and investigate the switch process in it . we show that amazing spatial clusters are formed during the chaotic transient process in the switch . in sec . v we present the results of an extensive statistical analysis over the whole turbulent regime of both gcml and egcml . we show that even in the turbulent regime the maps are under coherence with various manifestations of periodicity effect . we conclude in sec . vi . the coupled map lattice model in general is a system of one - dimensional field on a lattice @xmath1 with number of sites @xmath2 which evolves simultaneously by an iteration of the following two steps : @xmath3 in the first step all @xmath4 on the lattice ( called as map conventionally ) are simultaneously mapped by a nonlinear function @xmath5 . the function @xmath5 could be different site by site but we consider here the simplest case that @xmath5 is a simple logistic map @xmath6 common to all sites . in the second step the maps undergo interaction between themselves . the @xmath7 constant matrix @xmath8 represents the coupling between the maps at @xmath9 and @xmath10 on the lattice and the parameter @xmath11 is an overall coupling constant . thus the model in the simplest form has only two parameters , the nonlinear parameter @xmath12 and the coupling constant @xmath11 . if we rewrite ( [ generic ] ) in terms of @xmath13 , we obtain @xmath14 which is a familiar form in the neural network analysis @xcite . in gcml @xcite we consider n maps @xmath15 which evolve as @xmath16 where @xmath17 is the mean field on the lattice @xmath18 thus the coupling matrix of gcml is simply @xmath19 and clearly satisfies the reciprocity @xmath20 from ( [ global1 ] ) and ( [ global2 ] ) we can also derive a relation @xmath21 this relation guarantees that the mean field value is kept invariant when the maps undergo the interaction at the second step of the interaction . these two relations serve as a guide to construct the map lattice models in general . let us examine the implication of the invariance rule ( ii ) . the nonlinearity of @xmath5 generally magnifies the variance among the maps . therefore , the first step may be regarded as a _ defocusing lens _ with some _ random prisms _ installed in it . the larger the parameter @xmath12 is , the more strongly the defocusing lens acts on the system of maps . in the second step every @xmath22 is pulled to the mean field @xmath17 at a fixed rate @xmath23 . thus the second step acts as a _ focusing lens_. the larger the coupling @xmath11 is the more strongly the focusing lens acts . since ( [ global1 ] ) is a conformal contraction transformation the position of the _ center of mass _ @xmath17 of the system is kept invariant during the interaction . this is what the invariance condition ( ii ) implies . thanks to the invariance condition the fluctuation of the mean field of the maps is caused only at the first defocusing process . the maps in gcml evolve in time repeating this two - step process of defocusing and focusing . under the battle of the two conflicting tendencies the maps eventually form an attractor in which fluctuations among maps is maximally suppressed by the averaging effect of the interaction via the mean field . what is the condition on the coupling matrix @xmath24 for a general map model of the form ( [ generic ] ) satisfies the invariance condition ( ii ) ? by summing up ( [ generic ] ) over @xmath25 and using ( ii ) we obtain a relation @xmath26 at any step of the iteration this must hold . thus the necessary and sufficient condition on the coupling @xmath24 is @xmath27 from reciprocity ( i ) this is equivalent to @xmath28 this type of a model is known to show an interesting spatial cluster formation similar to the spin glass@xcite . on the other hand it is also known that the interesting two features of the gcml , i.e. the posi - nega switch and hidden coherence are lost . nonetheless it is instructive to check that this model satisfies ( i ) and ( ii ) like gcml . let us denote the sub - lattice of @xmath1 as @xmath29 which is a set of the nearest neighbor sites to @xmath25 including @xmath25 itself . the model is @xmath30 where @xmath17 is the mean field of the maps in @xmath29 at @xmath25 , i.e. @xmath31 the coupling matrix is then @xmath32 where @xmath33 is the number of maps in the sub - lattice @xmath34 in dimension @xmath35 . but if @xmath36 , then @xmath37 . hence @xmath38 and the condition ( i ) follows . also we have @xmath39 and ( ii ) follows . note that the interaction represents a diffusion process . for instance in @xmath40 we can rewrite ( [ ising1 ] ) and ( [ ising2 ] ) as @xmath41 and the second term is nothing but the laplacian on the one - dimensional lattice . we now wish to construct the following model . firstly it must have all to all interaction and satisfy the relations ( i ) and ( ii ) in order to have the same defocusing and focusing dynamics as in the gcml . secondly we wish to investigate the consequence of distance dependent interaction ; in particular , we wish to know to what extent the salient features of the gcml such as the posi - nega switch or the hidden coherence depend on the fact that the model is scale invariant trivially by the absence of distance scale . do these features disappear if the interaction between the maps decreases with distance ? certainly at the nearest interaction limit , we know neither the posi - nega switch nor the violation of the law of large numbers occurs . most crucial case is the interaction which decreases with distance but in a power law . in contrast to the exponential law this does not introduce the length scale and yet the interaction is all to all . thus the model we would like to construct is an extended gcml ( egcml ) in which the interaction decays with a power law and satisfies the conditions ( i ) and ( ii ) . below we show there is such a natural extension of gcml . in this article we choose @xmath42 for simplicity though our construction goes through for all dimensions . let us denote the set of sites at distance @xmath43 from @xmath25 as @xmath44 . the symbol @xmath45 is the memento of @xmath46 since in @xmath42 such a set is a loop ( in fact a square on the lattice space ) of radius @xmath43 around @xmath25 . we denote by @xmath47 the number of sites in @xmath48 . for @xmath42 , @xmath49 and @xmath50 . we denote the maximal radius of the loop realized in the lattice @xmath1 as @xmath51 . this is the perimeter of the lattice and for @xmath42 , @xmath52 $ ] . to be precise we present our result for @xmath53 odd but we have checked no change of results occurs for @xmath53 even . in order to treat all of the sites in @xmath1 democratically and to minimize the finite size effect we impose the periodic boundary conditions . for any point @xmath25 on the lattice we calculate the effect of all maps within @xmath51 taking into account the periodic boundary condition . the egcml is then compactly described as @xmath54 where @xmath55 at @xmath25 is given by @xmath56 in words the effect of all the maps on the lattice to the map at @xmath25 is determined by a double average . firstly any one of the maps at an equal distance @xmath43 from @xmath25 must affect the map at @xmath25 with the same weight . thus they are averaged with an equal weight and this gives @xmath57 . then , the effects of maps with various distances from @xmath25 must be accounted for . in the egcml this is done again by taking an equal weight average of @xmath57 over @xmath43 from zero to @xmath51 which gives @xmath55 as a natural non - local extension of the diffusion process . the first average must be an equal weight average to respect the rotational symmetry . the second average must be again an equal weight average . it is necessary for the invariance condition ( ii ) as we show below . the coupling matrix element @xmath24 is the weight factor for the @xmath58 to contribute to @xmath59 via the double average and is given by @xmath60 thus in egcml any map @xmath58 at equal distance @xmath43 from @xmath25 contributes with an equal weight to the mean field @xmath55 and the interaction decreases with an inverse power with respect to @xmath43 for @xmath42 . the reciprocity ( i ) follows just as in the nearest neighbor case since @xmath61 is equivalent to @xmath62 . also the invariance condition ( ii ) can be proved as : @xmath63 the gcml phase diagram in the @xmath64 plane given in @xcite is reproduced in fig . [ fig1]a to facilitate the comparison with the egcml phase diagram . for gcml we analyzed statistically the cluster composition of @xmath65 maps in the final attractor from 1000 random initial configurations and we verified the original phase diagram . the various phases are the outcome of the battle between chaotic motions of maps ( the defocusing effect ) and the interaction via the mean field forcing the maps to move coherently ( the focusing lens effect ) . since the gcml phase structure is elucidated in ref.@xcite we recapitulate here only the essential features necessarily in the following discussion . let us consider a line at @xmath66 for definiteness which is well - beyond the critical value @xmath67 of a single logistic map to the chaos@xcite . for a sufficiently large @xmath11 ( @xmath68 the maps in the final attractor are strongly bunched together in a cluster and evolve as a unity in the same chaotic motion as that of one single logistic map . this is the coherent phase . for @xmath11 in the range @xmath69 $ ] the interaction via the mean field can no longer exerts strong bunching and the final maps divide into two clusters . the maps in each cluster are still tightly synchronizing each other , while the two clusters mutually oscillate opposite in phase . this is as a solution of a minimum fluctuation in the mean field . we discuss this ordered two - cluster phase extensively in sec . iv . for smaller @xmath11 , the number of final clusters increases but the number of clusters at a given @xmath11 does not depend on the number of maps @xmath2 . the typical number of clusters at various @xmath11 ranges is indicated in the phase diagram . for further smaller @xmath11 ( @xmath70 ) the binding of the maps is almost resolved and the number of clusters is proportional to @xmath2 . this region is the turbulent phase and we extensively analyze this regime in sec . v. final remark is that the boundary values of @xmath11 between phases simultaneously increase ( decrease ) in order to maintain the balance between the conflicting tendencies if the coupling @xmath12 is increased ( decreased ) . we show in fig . [ fig1]b the corresponding phase diagram of the egcml . we used @xmath71 map lattice and analyzed statistically the final attractor configuration using typically 200 runs for each @xmath72 . for each map at the lattice site @xmath25 must be individually calculated using the inverse power law . for @xmath73 egcml , the necessary cpu time is about a hundred times of that for gcml of the same size . ] there is a remarkable agreement between two phase diagrams except for slight shifts in the phase boundaries . for instance , for the egcml at @xmath74 , the coherent phase , the ordered two - cluster phase , and the turbulent regime are respectively realized at @xmath75 , @xmath76 and @xmath77 to be compared with the boundaries quoted above . this agreement between two phase diagrams indicates strongly that egcml is in the same universality class@xcite with gcml . the ordered two - cluster phase of gcml is realized in a band in the @xmath79 parameter space with @xmath80 $ ] at @xmath81 and @xmath82 $ ] at @xmath83 ( see fig . [ fig1]a ) . in this phase , the maps divide into two clusters after some transient steps for most of the initial values . the clusters oscillate around the unstable fixed point @xmath84 mutually opposite in phase and the maps in each of the two clusters are in tight synchronization . in order to distinguish the two clusters we follow the convention in @xcite and name the cluster which takes a value above ( below ) @xmath84 at an even step @xmath85 from the first of the iteration as the positive ( negative ) cluster . we denote the number of maps in the positive ( negative ) cluster as @xmath86(@xmath87 ) and define the population ratio parameter @xmath88 . the values of maps in the positive ( negative ) cluster are uniformly @xmath89 if the synchronization of maps in each cluster is really tight . the maps are evolving in two clusters repeating the two step process of mapping and interaction and the dynamics of maps is reduced to that of two clusters : @xmath90 and the mean field of maps at time @xmath85 is given by @xmath91 where the second equality comes from the invariance condition ( ii ) for the two - cluster configuration . thus when a set of three numbers , @xmath12 , @xmath11 and @xmath78 is given the attractor of the two clusters is uniquely determined . for the discussion of the posi - nega switch below let us explain these by an explicit sample in fig . [ fig2 ] . in the left box we illustrate the mechanism of eqs . ( [ twocluster1 ] ) and ( [ twocluster2 ] ) and in the right box we show a sample attractor of two clusters realized at @xmath92 , @xmath93 and the corresponding motion of the mean field . the two clusters , represented by white and black circles , evolve in time under the iteration of mapping ( solid arrows ) and focusing by interaction ( dashed arrows ) . the mean field ( a short bar ) is kept invariant under the interaction as dictated by the invariance condition ( ii ) and ( [ twocluster2 ] ) . the first time @xmath94 in this run happened to be even . hence the positive cluster is the white circle moving as @xmath95 and the negative one is the black circle moving as @xmath96 . we see that they move in period four and mutually opposite in phase . the mean field is closer to the negative cluster , which means that the positive cluster is the minority cluster and @xmath78 must be less than half . numerically @xmath97 for this attractor . note that it is probabilistic whether the first time @xmath94 is even or odd . from another initial configuration the same attractor _ modulo translation in time _ may be equally realized . if the translation is odd steps in time we will have the same figure but with @xmath94 is now odd . the dominant cluster ( black circle ) must be then named as the positive cluster and @xmath98 by definition . in fig . [ fig3]a and b we show the attractors of gcml and egcml at @xmath92 and @xmath93 respectively as a function of @xmath78 . we used @xmath99 runs for gcml and @xmath100 runs for egcml . in one run for gcml , @xmath65 maps are iterated from a random initial configuration and to account for the first transient process the values of maps during the @xmath101 steps after the first @xmath102 iteration are plotted at the fraction of maps with @xmath103 . for instance the run shown in fig . [ fig2 ] with @xmath94 even contributes four points @xmath104 at @xmath97 and other four points @xmath105 at @xmath98 . then the run is repeated and the results of all runs are superposed in order to see the dependence of the attractor on the population unbalance between the clusters . note that this is equivalent to plot the maps only at the last even iteration steps at @xmath78 when the results from many initial configurations are superposed . the gcml attractor in fig . [ fig3]a agrees with that found in ref.@xcite . when @xmath78 approaches an upper threshold @xmath106 from below ( or a lower threshold @xmath107 from above ) the attractor undergoes successive period doubling and at the thresholds the maps fall into chaos ( @xmath106=@xmath108 and @xmath109 for @xmath92 ) . for egcml in fig . [ fig3]b the final attractor of @xmath110 maps is measured in a similar way but in order to shorten the large computing time necessary for egcml a routine is included in the iteration program which judges whether the maps have fallen into a periodic attractor of a certain period @xmath111 less than or equal to @xmath112 or not is calculated at every even step @xmath85 and when it stops to vary during ten successive steps the routine is activated at @xmath94 . it calculates the judgment condition @xmath113 and if this is satisfied for some @xmath114 the maps are judged to have fallen into an attractor of a period @xmath111 . if not , the routine is inactivated and the next stabilization of @xmath86 is waited for . ] . if the judgment condition is satisfied the maps are plotted at the fraction of maps with @xmath115 only during the next one cycle of the attractor and the next run is started . this not only shortens the calculation time but also enables the statistical analysis of the frequency of the occurrence of the attractors and of the fluctuation of maps in a given attractor since indefinitely duplicated plotting of the maps is avoided . if on the other hand the maps do not fall into an attractor all through the @xmath102 steps the maps are plotted for the next @xmath101 steps just as in the gcml run . in egcml we find that the maps in a cluster synchronize more loosely than those in gcml . the dispersion of maps in each cluster is approximately 0.1 for all @xmath78 to be compared with the gap size @xmath116 between the two bands of the attractor at @xmath117 . even though the maps do not mix between the clusters in the final stable attractor this fluctuation hides the details of bifurcation structure with respect to @xmath78 in a direct superposition of all runs in terms of the bitmap graphics . thus we have done a statistical analysis of the plot density and removed those points with density less than one percent of the maximum density from the diagram . this procedure is justified because we truncated each run just after one cycle of the periodic attractor so that each run contributes to the points on the attractor diagram with an equal weight . it was a remarkable instance in this work when we observed the same bifurcation diagram with gcml turned out in the central region of the egcml attractor ( @xmath118 $ ] ) as is shown in fig . [ fig3]b . in egcml the synchronization of maps in a cluster becomes looser than that of gcml but still the dynamics of final clusters is common for both gcml and egcml and controlled by @xmath78 . we have already seen above that egcml is endowed with a phase diagram remarkably similar to that of gcml and now we find that the attractors in both models in the ordered two - cluster regime are also similar . the posi - nega switch in gcml uses the ability of @xmath78 to control the attractor . but before delving into the detail discussion of the posi - nega switch let us examine to what extent the reduction of dynamics of maps is realized . if the reduction is really effective down to two degrees of freedom , all @xmath86(@xmath87 ) maps in the positive ( negative ) cluster should have a single value @xmath119 ( @xmath120 ) . then eqs . ( [ global1 ] ) and ( [ global2 ] ) should reduce to ( [ twocluster1 ] ) and ( [ twocluster2 ] ) or equivalently to a matrix coupled two - map model @xmath121 with @xmath122 and @xmath123 . from the remarkable agreement between gcml attractor ( fig . [ fig3]a ) and two - map model attractor ( fig . [ fig3]c ) we see that in the two - cluster phase of gcml the dynamics of the @xmath2 maps reduces to that of two maps for all @xmath78 between the thresholds . in egcml the coupling @xmath24 is not a single constant but it depends on the distance between @xmath25and @xmath124 . the interaction acts on a map at a site @xmath25 via the position dependent mean field @xmath55 which is a double sum given by ( [ tsm2 ] ) and ( [ tsm3 ] ) contrary to a single sum @xmath17 given by ( [ global2 ] ) in gcml . therefore the reduction of egcml to two maps we find in fig . 3b requires an explanation . what is important here is not the uniformness of the couplings but the scale invariance of them . in both gcml and egcml the interaction has no cutoff scale and hence it prevails over the lattice and the maps are forced to synchronize among themselves due to the averaging effect of the interaction via the mean field . thus the possible difference of egcml from gcml comes solely from the variation of the site dependent @xmath125 around @xmath17 . if the variation is small both models will show similar attractors . egcml is constructed in such a way to assure this small variance when the interaction forces the maps to synchronize . first let us consider a simple case where the nonlinear parameter @xmath12 is so small that the chaotic randomness in each map can not overcome the averaging effect . then after iterations all the maps on the lattice will synchronize and take approximately the same values with only a small variance . if we ignore this variance , the value @xmath22 common to all @xmath25 becomes a multiplicative factor to the sum for the mean field and @xmath126 follows from the invariance conditions ( [ unitaritygcml1 ] ) for gcml and ( [ unitarityegcml ] ) for egcml . in gcml the variance is negligible but in egcml the weaker synchronization causes a variance typically to the order @xmath127 in @xmath125 and the egcml attractor turns out reflecting this variance . but otherwise egcml behaves just like gcml . now in the two - cluster phase at higher @xmath12 the maps can not be bunched together in a single cluster and divide into two clusters , a and b. the maps in each of the clusters take approximately the same value and synchronize among themselves . hence we may write @xmath128 here @xmath129 ( @xmath130 ) is the collective coordinate for a ( b ) ( the average of maps taken over a ( b ) only ) and the @xmath131 ( @xmath132 ) represents small fluctuation around the @xmath133 ( @xmath130 ) . now let us analyze @xmath125 for egcml . an important observation is that the double sum for @xmath125 can be written as a sum of two double sums , one only over cluster a and the other only over cluster b. this decomposition allows one to apply the above argument to each of the clusters and the @xmath134 ( @xmath135 ) becomes a multiplicative factor to the double sum for cluster @xmath136 ( @xmath137 ) . as for the remaining double sum of couplings , some more consideration is necessary . in egcml a ` polarized ' distribution of maps is in principle allowed in which , for instance , the maps in the cluster @xmath136 ( @xmath137 ) accumulate around a given point @xmath25 and maps in cluster @xmath137 ( @xmath136 ) accumulate far from @xmath25 . but egcml has no interaction to induce such a polarization and the randomness in each map also acts to break the polarization . we have numerically checked the dominance of unpolarized distributions . therefore the remaining double sum for egcml again counts the number of maps in each cluster just in the same way the single sum for gcml does in ( [ twocluster2 ] ) . thus we have also in egcml @xmath138 here the approximation takes into account both the polarization in fluctuation and the small variance of maps in each cluster due to approximate synchronization . thus the collective coordinates @xmath133 and @xmath130 of the clusters is described by the matrix coupled two - map model and the maps in each of the clusters fluctuate around the collective coordinates . this explains the approximate reduction of egcml to two maps . the posi - nega switch found in gcml@xcite is based on the ability of @xmath78 as the control parameter of the gclm attractor . if maps are successively transported from one cluster to the other by input ( @xmath139 ) the population unbalance can be enhanced and the attractors undergo successive bifurcation . when @xmath78 reaches the threshold the whole maps go into a grand chaotic motion almost bunched together in a unit . see fig . [ fig4 ] for a typical sample of this transition . this state induced by an excess of the last one map is unstable as we see in fig . [ fig3]a that there is no self - organized stable attractor above @xmath106 ( or below @xmath140 ) . thus after some transient time the maps again form the allowed stable configuration , that is , tightly bound two clusters mutually oscillating opposite in phase . just as before let us again label the new cluster as positive if it takes a value above @xmath84 at even step @xmath85 ( counting from the start of iteration ) and the other as negative . an amazing finding in @xcite is that the composition of the new two clusters is limited only in two ways . ( 1 ) all the maps involved in the positive cluster before the chaotic transient are now involved in the negative cluster . ( 2 ) all the maps in the old positive cluster are again in the new positive cluster . the case ( 1 ) is the posi - nega switch . in both cases the maps behave as if they keep a _ memory _ of their former clusters . but how is it possible ? the maps pass the chaotic transient process where many two - cluster states as well as many other channels are open . what is the mechanism that protects the maps from mixing ? before answering this old puzzle let us point out some trivial aspects of the posi - nega switch . first there is no essential difference between case ( 1 ) and ( 2 ) . how long the chaotic transient process persists and in particular whether the duration time is even or odd is ( almost ) probabilistic . it is of course deterministic but highly dependent on the map values at the onset of the transient process . whether a cluster after the transition is positive or negative depend on the parity ( even or odd ) of the duration time by definition and there is no sense to distinguish ( 1 ) and ( 2 ) when the parity is indefinite . second we should note that in the two - cluster phase of gcml with fixed @xmath12 and @xmath11 the self - organized attractor is determined solely by the population unbalance between the two clusters . in case ( 1 ) @xmath141 and in case ( 2 ) @xmath142 . in either case , the population _ unbalance _ is the same before and after the transition and hence the attractor before and after must be the same . to be precise , to put the attractor before the transition upon that after the transition we need a temporal translation of odd steps in case ( 1 ) and even steps in case ( 2 ) . modulo this difference the attractor before and after the transition are the same in either case . there is no chance of appearance of a new attractor _ provided that _ the population _ unbalance _ before and after the transition is the same . therefore the real puzzle boils down to a single question . what is the mechanism which keeps the population unbalance between two clusters unchanged before and after the chaotic transient process ? we solve this puzzle below in two steps . first we discuss how the chaotic transient process starts and ends . this provides the temporal boundary conditions of the transient process . then we show there is a very simple mechanism which transfers the population unbalance rate through the transient process . the logistic map @xmath143 copies the region @xmath144 $ ] onto itself twice - fold as follows the limits are @xmath145 and the unstable fix point is @xmath146 but the argument here is independent of the particular parametrization of the map . all we need is the universal folding nature of the map . ] . i+ii+iii \nonumber \\ i & \equiv & [ -x_l , 0 ] \stackrel{f}{\longrightarrow } i+ii+iii \nonumber \\ ii&\equiv & [ 0 , x^ * ] \stackrel{f}{\longrightarrow } iii \\ iii&\equiv&[x^ * , x_l ] \stackrel{f}{\longrightarrow } ii+i ~. \nonumber \\ \end{aligned}\ ] ] the first two mappings are monotonously increasing while the last is monotonously decreasing . see fig . [ fig5 ] . below the threshold the two clusters are moving mutually opposite in phase around @xmath84 so that one must be in region iii and the other in region i or ii . when the unbalance is increased by sending the maps of the minority cluster to the majority cluster the mean field comes closer to the majority cluster and oscillates in phase with it . the interaction conformally contracts the maps to the mean field at every iteration step , and the minority cluster is pulled to the mean field more strongly at this highly unbalanced situation . in a few iterations after the unbalance exceeds the threshold by the last input , the non - dominant cluster crosses the boundary @xmath84 . this is the very time of the start of grand chaotic transient motion of maps . at the next step of iteration all maps must move into the same direction . the synchronization opposite in phase ceases and the maps start almost coherent evolution . the mean field evolves together in the middle of them . the focusing effect of the interaction now works only to bunch the maps close together around the mean field . the high nonlinearity of each map then becomes manifest and all maps move around the three regions bunched together in a pseudo - cluster with a slight variance around their mean field . in the typical sample run shown in fig . [ fig4 ] the mean field ( depicted by a short sold bar ) oscillates closely to the dominant cluster ( connected by the solid line ) for several steps . at time @xmath148 the minority cluster ( the dashed line ) is pulled down across the @xmath84 and at the next step both clusters are mapped upwards together . then the grand chaotic motion starts at @xmath149 . the pseudo - cluster is formed by the external input and is unstable . to bunch it stably the coupling must be larger ; it must have the value for the coherent region ( e.g. , @xmath150 for @xmath92 ) . the tiny variance between maps in the pseudo - cluster is soon magnified and the transient pseudo - cluster is broken to the self - organized stable two clusters again . in the following we use the fact that the maps starts and exits the transient process with small variance among them . let us define the relative coordinates @xmath151 with respect to the mean field @xmath152 since all of the maps are now close to the mean field we may expand @xmath153 as @xmath154 averaging this over @xmath155 using @xmath156 we obtain a relation @xmath157 but the interaction does not affect the mean field ( the condition ( ii ) ) . so we obtain @xmath158 therefore the evolution equation of the relative coordinates is given by @xmath159 we used ( [ averagerelation ] ) once more in order to derive the third line . the last line is crucial . the factor @xmath160 is common to all maps and in particular it acts with a common sign . thus at every step in the chaotic transient process the maps separated by the mean field never mix each other . this is the mechanism which keeps the population unbalance unchanged before and after the chaotic transient process . the sign of the factor at @xmath85 depends on the mean field value @xmath161 . hence case ( 1 ) and ( 2 ) may occur equally . we note that the invariance condition ( ii ) is used in proving the mechanism . this condition is really important to explore the model which shares the property of gcml . in the above resolution we examined the case in which the maps are confined in a pseudo - cluster and their variance around the mean field is kept small due to synchronization . numerically we can trace the motion of the pseudo - cluster . almost always the maps moves in a pseudo - cluster around the mean field with squared variance less than @xmath162 . but sometimes it intermittently splits into two or more tight clusters around the mean field . this does not invalidate above resolution . the danger of the mixing of members between clusters occurs only when they come close each other . our mechanism gives a guarantee that no mixing occurs even in such emergence . in fig . [ fig6]a we show a posi - nega switch in @xmath163 gcml ( @xmath164 ) . in the upper ( lower ) diagram the maps at the even ( odd ) iteration steps are plotted and connected by lines . a similar figure was given in the first report of the gcml switch@xcite but the evolution of maps only at the even iteration steps was presented . by such a diagram only it looks as though the attractor undergoes a transition to a new periodic attractor after the posi - nega switch . but as we noted above , the attractor before and after the switch is the same modulo temporal transport of odd steps . we can see clearly in fig . [ fig6]a the attractor on the whole , the even and odd iteration steps together , is precisely the same before and after the chaotic transition . as indicated by the arrows the switch from positive to negative is trivially a change in the count ( even or odd ) of iteration steps . what is not trivial is that the memory of the composition of two clusters is kept during the posi - nega switch . in fig . [ fig6]b we present a posi - nega switch event of @xmath165 gcml at even iteration steps only as a three - dimensional landscape plot . the peaks ( valleys ) before the chaotic transient process become the valleys ( peaks ) which is the posi - nega switch . in the transient process the maps form a surface which is chaotically oscillating but at any instance the segment of the surface looks remarkably horizontal . this shows that in the transient process the maps are in quasi - coherent motion with only a tiny variance . in fig . [ fig6]c we show a _ print circuit pattern _ of the same switch event only at even iteration steps distinguishing the maps with respect to the mean field value . those maps with ( @xmath166 ) are shown with white squares and the others by black squares . just as we proved above , the pattern clearly shows that there is no mixing of maps across the mean field all through the switch event . it can be also seen in fig . [ fig6]c that the transient process ceases when the last transferred map ( @xmath167 ) comes back to its home cluster . this is a typical way of the end of gcml switch . this completes our resolution of the posi - nega switch mechanism . the egcml has almost the same phase diagram with gcml as we showed in sec . iii and the attractor of egcml in the two - cluster phase has the same bifurcation tree structure with that of gcml . but as we showed in fig . [ fig3]b the attractor of egcml turns out with some variance . that is , the dynamics of clusters in egcml is reduced to that of matrix - coupled of two maps just as in gcml but the maps in each egcml cluster are in phase - synchronization rather than in tight synchronization in contrast to gcml . the all to all interaction with no cutoff scale in egcml is effective enough to maintain the reduction to two - cluster dynamics for any @xmath78 between @xmath168 $ ] . in the numerical simulation of egcml we find that the two clusters of egcml undergo a switch process by inputs which is similar to that in gcml . a typical sample of the egcml switch is shown in fig . [ fig7 ] . but let us first list the general differences of egcml switch from gcml switch below and add description particular to fig . [ fig7 ] afterwards . \(1 ) _ slower reaction _ : by an injection of a pulse we transfer a map in the minority cluster to the majority cluster . in gcml the reaction of the attractor to this input is quite rapid and the attractor shifts to the new movement only after a few steps of the iteration . on the other hand in egcml the attractor reacts to the input gradually in several iteration steps this has a simple reason . the synchronization of maps in an egcml cluster is looser . when a map is transferred there occurs a negotiation among maps in each cluster to reform the clusters and then the two clusters mutually move to the new attractor . this is similar to the two step phase - synchronization found in the globally coupled flows@xcite . \(2 ) _ self - organized transition _ : in egcml the effect of one pulse depends on the unbalance between clusters at the time of injection . if the unbalance is not high just a slow reaction of the attractor described above occurs . if the unbalance is high the map transportation by an input is followed by a self - movement of another map between clusters in the same direction . this occurs after a few tens of steps for the egcml of the size @xmath169 . if the unbalance is close to the threshold determined by the above statistical analysis ( fig . [ fig3]b ) the transportation of a single map triggers a cascade . the last injected pulse induces follower maps to move from a cluster to the other cluster in the same direction and in some 100 iteration steps the move grows up to a series of avalanches at various scales . the corresponding change of the spatial distribution of positive and negative clusters is similar to a percolation transition . under this process no more external input is given and the maps are themselves forming a metastable state . both the existence of a threshold and the formation of clusters at various scales suggest a remarkable similarity to the self - organized criticality , e.g. @xcite , but further investigation on a larger size lattice , in particular the measurement of the critical exponent , is required to verify the relation . \(3 ) _ the mixing _ : in egcml the two - cluster attractor after the chaotic transient process is for most cases approximately the same with that before . the change of the population unbalance is only a few percent in most of the runs . but there occurs swapping of maps between clusters and the composition of the clusters in terms of maps is often different before and after the egcml switch . the posi - nega switch in gcml is realized by the fact that no mixing of maps across the mean field occur during the chaotic transient process . in egcml the mixing does occur since each map is under the influence of position - dependent @xmath55 . thus the egcml switch is a posi - nega switch as a cluster process but not as the map process . in view of the brain dynamics this difference may be immaterial since what is important is the activity of whole neurons rather than the identity of individual neurons . \(4 ) _ self - organized spatial clusters in egcml _ : before the switch the maps in the positive cluster and those in the negative cluster are almost randomly distributed on the lattice . the transition to chaotic transient process goes like a percolation transition . when the self - organized mass movement starts , the positive maps form several large - size spatial clusters and so do the negative maps . since these spatial clusters are formed by the long - range interaction the formation of them is more remarkable in the lattice of larger size . the spatial clusters can be best traced in the chaotic process by distinguishing the maps by the mean field of whole maps . at short time scale often we observe a posi - nega switch of spatial patterns . but due to mixing the spatial clustering pattern changes at longer time scale . now we explain fig . [ fig7]a which is a typical egcml switch event sampled in @xmath170 lattice ( @xmath171 ) . _ preparation : _ the maps are iterated from random start and a few tens of pulses are injected during the first 1500 iteration steps until the two synchronizing clusters exhibit period 32 oscillation mutually opposite in phase . _ start of chaotic process : _ at @xmath172 we injected ten pulses simultaneously to randomly chosen maps on the lattice . this was necessary to show the whole process in one figure . had we injected the pulses one by one with say a hundred steps in between we had been able to see the accelerated cascade transition as described in ( 2 ) above . but still we can see in fig . [ fig7]a that the process after @xmath173 is a self - organized cascading process . _ the posi - nega switch between clusters : _ at @xmath174 the transient process ceased and a periodic attractor of two clusters appeared . as indicated by arrows we see clearly that it is the same attractor as before . the change of the unbalance of population between the two clusters is less than one percent in this switch . _ the spatial clusters : _ the snap - shot patterns of the maps at the even iteration steps are shown in the boxes below the evolution plot . in the upper boxes we distinguish the maps with respect to the mean field while in the lower boxes we used the unstable fixed point @xmath84 . in either way we can see that spatial clusters are formed by the self - interaction of maps during the chaotic transient process . we see clearly that the maps undergo a posi - nega switch among themselves between snap - shots 4 - 7 in the upper boxes . on the other hand in the lower boxes the pattern becomes sometimes totally black and sometimes totally white . hence , we see that the distinction by the mean field is a better way to sense the spatial cluster formation on the lattice . a distinction by the mean field is sensitive to the fluctuation around the collective coordinate . the clustering pattern extracted by this method is the long - range fluctuation dynamics of maps on the lattice . this long - range interaction dynamics can be seen more remarkably on a larger lattice which allows the formation of the spatial pattern of longer wavelength . in fig . [ fig7]b we show the samples of amazing spatial patterns formed during the chaotic transient process in the run of @xmath175 lattice ( @xmath176 ) . the snap - shots were sequentially taken at every four iteration steps . we observe that a small seed of fluctuation tends to grow into a circular dense pattern by the chaotic diffusion process on the lattice in accord with the isotropic interaction in egcml . it is interesting to note the similarity of the patterns observed on the lattice to the galaxies in the universe and in particular the appearance of patterns like globular nebula . the egcml may be used as a toy model for the formation of density fluctuations in a nonlinear diffusion process , though further study is required to test this speculation . the turbulent regime is generically the region of small @xmath11 and known by the so - called hidden coherence phenomenon . generally it is understood that in this region maps evolve almost randomly at small lattice size @xmath2 and that there occurs the hidden coherence at large @xmath2 . but actually the turbulent regime has a subtler feature . we are going to show that it is a regime where the grand periodic or quasi - periodic motion of the whole maps is in action almost everywhere . but first let us quote briefly previous observations in the literature in three items . \(i ) when gcml is iterated from a random configuration with a small coupling ( @xmath177 ) and a large nonlinear parameter ( @xmath178 ) it reaches a final state which is an ensemble of maps and tiny clusters , each moving chaotically due to high nonlinearity of the map . ( we show below that they are under certain coherence almost everywhere in the turbulent regime even though each element looks in an independent random motion . ) the number of elements increases proportionally to the number of whole maps in sharp contrast to the ordered regime where the dynamics of maps reduces to that of a few clusters@xcite . \(ii ) it was found in the statistical analysis of the mean field fluctuations that there emerges certain coherence between elements for large lattice size @xmath2@xcite . ( we show below some reservation is necessary to this statement . ) let us denote the mean field of the @xmath2 maps on the lattice @xmath1 at time @xmath85 as @xmath179 if @xmath180 are really @xmath2 independent random variables with a common distribution , the mean square deviation ( msd ) of the fluctuation ( @xmath181 ) should decrease proportionally to @xmath182 by the law of large numbers ( lln ) and the distribution of the @xmath183 must be a gaussian for sufficiently large @xmath2 by the central limit theorem ( clt ) . however the time series analysis of the mean field fluctuation shows that there is a threshold @xmath184 in @xmath2 ( depending on both @xmath12 and @xmath11 ) above which msd ceases to shrink even though the distributions remains in a gaussian shape . even though the elements ( maps and tiny clusters ) look evolving independently each other the violation of lln must reflect some hidden coherence between the elements . in fact an introduction of a very small independent noise term in each map restores lln@xcite . \(iii ) this violation of lln observed in the time averaged distribution of @xmath183 reflects that the probability distribution @xmath185 that a map takes a value @xmath186 explicitly depends on time . a noise intensity analysis of ensembles successfully proves lln@xcite . if lln should hold in the time average , @xmath185 would have to be the fixed point distribution of the frobenius - perron(fp ) evolution equation@xcite . it has been shown that the fixed point distribution is unstable due to the periodic windows inherent to the chaotic logistic map@xcite . the coherence manifests itself in the mutual information@xcite . on the other hand the temporal correlation function similar to the edwards - anderson order parameter for the spin glass@xcite decays to zero exponentially . thus it may not be due to freezing between two elements@xcite . all these are important observations for the turbulent regime of gcml . but this is not the whole story . especially one should not simply conclude by the above list that the turbulent regime of gcml ( and also egcml ) is a regime with complete randomness at small @xmath2 and with the hidden coherence for larger @xmath2 . the msd and distributions previously reported @xcite are only for the limited regions of the turbulent regime where the mean field distributions looks semi - gaussian and the violation of lln for large @xmath2 is remarkable . but there do exist regions in the turbulent regime where the coherence is suppressed strongly and the msd obeys lln in a good approximation . oppositely there are also large regions where the final attractors are two or three clusters which evolve stably ( until the end of @xmath187 iterations ) in period three ; a grand periodicity in the turbulent regime . furthermore even in a large part of the hidden coherence region the distribution is not really gaussian ; it is only semi - gaussian ( trapezoidal ) or has visible peaks on top of semi - gaussian distribution . we can detect the remains of the periodic coherent motion by an analysis of the decay of a temporal correlation function for these regions . the turbulent regime is actually the region which is full of periodicity remnants and the hidden coherence is realized in between the genuinely turbulent regions and manifestly periodic regions . the egcml shares the same features with gcml . in the following we show these facts by an extensive analysis over the whole turbulent regime . we measure the distribution of @xmath183 and its msd at every point on the grid on the @xmath188 plane in the turbulent regime at a fixed value of the nonlinear parameter @xmath12 . here we choose @xmath189 as a canonical value but our observations below generally hold for other values of @xmath12 larger than @xmath190 . at @xmath191 the turbulent regime is from @xmath192 to @xmath193 for gcml and from @xmath192 to @xmath194 for egcml ( see fig . [ fig1 ] a and b ) . thus we set a @xmath195 @xmath188 grid where the lattice size @xmath196 varies from @xmath197 to @xmath198 with an increment of @xmath199 two is necessary for two - dimensional egcml with the periodic boundary condition . for gcml we keep the increment the same for the sake of comparison . ] and the coupling @xmath11 varies from @xmath200 to @xmath194 with an increment @xmath201 . at every point on the @xmath188 grid with @xmath12 fixed at @xmath202 , the maps are iterated from a random start and the mean field @xmath183 is measured during the time interval between @xmath203 and @xmath204 . thus the result for one run at @xmath191 is a @xmath195 table of distributions and each distribution is a histogram of @xmath205 @xmath183 . we have repeated the run several times for the same @xmath12 and verified that the general result is independent from the change of the initial configuration of randomly produced maps . we have also checked in two ways that our @xmath188 grid is sufficiently fine to see the general feature of the turbulent regime . firstly we randomly chose more than a hundred squares of the grid and subdivided them . using this finer grid we have verified that either the variation of @xmath11 or @xmath2 within each square does not cause any significant change of the distribution ( a local check ) . secondly the compiled @xmath206 distributions are sufficient to make an animated show of the distribution . we observed by this animation that the distribution changes systematically the shape with variation of either @xmath11 or @xmath2 over the whole @xmath188 grid though the onset of the apparent periodic distribution is quite rapid ( a global check ) . we find that a special care must be taken over for the transient time when we are on the edge of the periodic window region . there the maps may fall sometimes into the final periodic attractor after , for instance , @xmath207 iterations but sometimes only after @xmath208 iterations depending on the random initial values . for this edge region we have done a separate analysis using a sequence of runs , each consists of @xmath204 to @xmath187 iterations . we comment on this region in sec . v d. this simple but tedious analysis distributions for both models . ] was the start of the work to unveil the real feature of the turbulent regime . in fig . [ fig8]a and b we show the msd of the mean field ( @xmath183 ) distributions at @xmath189 as a surface plot over the @xmath209 grid for gcml and egcml respectively . both msd and @xmath2 are plotted in the logarithmic scale so that the linear edge of the surface in the front panel at @xmath192 is of course due to lln . for a non - zero but very small @xmath11 @xmath210 , the lln still holds to a good approximation but for a larger @xmath11 we can clearly see that the surfaces have many peaks . the global features of the msd surfaces may be summarized as follows . * the valleys * : at the valleys of the surface the msd still approximately respects lln and we may call these valleys as genuinely turbulent regions . apart from the valleys lln does not hold . this does not imply the real violation of the lln in the ensemble average@xcite . what is meant by the violation of lln here is that there is a larger fluctuation in the time series of @xmath183 than expected by lln . we are interested in detecting the coherence among the evolving elements by the enhanced msd . for both gcml and egcml the peaks divide into two classes except that the peaks are more parallel each other for gcml . below we quote figures for gcml . * the first - class peaks * : we can see two very high and broad peaks at @xmath211 and @xmath212 and lln is violated extremely on them . here the excess msd is observed also for small @xmath2 . let us call them as the first - class peaks of the msd surface . * the second - class peaks * : separated by the large first - class peaks there are also several minor peaks . their locations for @xmath191 are @xmath213 let us call them as the second - class peaks . provided that one keeps eyes only on the second - class peaks the msd first decreases with increasing @xmath2 following lln and then above certain threshold @xmath214 the msd ceases to decrease and remains approximately in a plateau . this is a characteristic behavior of msd at the hidden coherence . however we show below that the @xmath183 distribution is not a gaussian but is a distorted gaussian in the bulk of the second - class peak regions . now we analyze directly the table of @xmath195 distributions of the mean field @xmath183 at @xmath191 for both gcml and egcml to understand how the distribution looks like for the enhanced msd regions . before this analysis we had thought from the literature@xcite that the distributions are all gaussian with possible msd enhancement at least for the central region of the @xmath183 histogram . distribution is limited in the range @xmath215 $ ] , it can not be a finite width gaussian distribution over @xmath216 . when we discuss whether the @xmath183 distribution is gaussian or not we concern whether the essence of the clt , that the convolution of independent random distributions peaks like gaussian in the central region , is in action or not . the ratio of the fourth to the second moment must be three for a perfect gaussian and for the @xmath183 distribution a small deviation from three is reported @xcite . but it is sensitive to the tail region and the dynamics in the central region is obscured in a ratio . for instance , the rank three distribution below is often two overlapping gaussian peaks due to the periodic clusters with high mixing but a ratio of moments can not tell it . ] but to our first surprise we find that there are a variety of distributions involved in the table of gcml and egcml . we assign each distribution a pattern classification number or a rank from zero to four by its shape . 1 . the distribution is gaussian . compared with the gaussian distribution at @xmath192 with common @xmath12 and @xmath2 , the msd agrees within 20 percent error . 2 . still gaussian but with more than 20 percent increase of the msd . at large @xmath2 we see the msd can become even more than twice . a singly peaked distribution but the shape is distorted from gaussian . typically a gaussian - like peak polarized to the right or left , or trapezoidal . either it has a few sharp peaks on top of a broad band , or it is an apparent overlapping gaussian distribution . it manifestly shows the periodic motion of elements . the distribution consists of a few sharp peaks only . at rank zero the mean field distribution obeys both lln and clt and the maps may be thought as independent random numbers with a common probability distribution . oppositely at rank four the maps are in periodic motion and so is the mean field . the ranks are organized in a way that the periodicity of the elements becomes more manifest with an increase of the rank . in fig . [ fig9]a and b we show the rank distribution on the @xmath188 grid as a density plot for gcml and egcml respectively . we find the followings . \(1 ) there is a remarkable match between the msd surface plots and the distribution density plots . the msd surface is high ( low ) wherever the rank of the distribution is high ( low ) . the rank distribution plot is almost a contour plot of the msd surface . this is highly non - trivial . that the rank plot is a contour plot of a surface means that there occurs no jump of ranks when we vary either @xmath11 or @xmath2 . for instance to go to rank four from rank zero one has to pass the intermediate rank regions . this indicates that the dynamics of maps changes continuously ( not necessarily smoothly ) when either @xmath11 or @xmath2 is varied over the turbulent regime . \(2 ) the fist class peaks of the msd surface occur just at the rank three and four distributions . the first - class peaks are due to the periodic coherent motions of elements . \(3 ) in the second - class peaks the mean field distribution is rank zero ( both clt and lln hold . ) for small @xmath2 . when @xmath2 is increased with @xmath11 fixed , it becomes rank one ( clt holds but lln violated ) above certain threshold . with a further increase of @xmath2 it almost always become rank two ( even clt is not fully operative ) . no higher rank distribution ( rank three or four ) appears in the second - class peak regions . only in small band(s ) of @xmath11 ( for instance , only in a narrow band around @xmath217 for @xmath191 ) the rank remains one for large @xmath2 . \(4 ) the threshold between rank zero and rank one distribution precisely agrees with the break point @xmath214 of the drop of msd following lln . for reference we superpose the approximate curve for the @xmath214 ( @xmath218 ) given by @xcite in fig . [ fig9 ] and indeed it runs roughly between the rank zero and rank one regions . we consider the fact in ( 3 ) that the rank one distribution changes smoothly to the rank two distribution with increasing @xmath2 is a key to the hidden coherence ; it indicates that the coherence between maps hidden in the rank one distribution region becomes manifest in the adjacent rank two distribution region . hereafter we focus our attention to gcml . in fig . [ fig10 ] we show a sample of ( 3 ) in the left three boxes . we fix @xmath12 and @xmath11 ( @xmath219 and give the @xmath183 distributions at various @xmath2 ( @xmath220 ) which are sampled in @xmath204 iterations from random starts discarding the first @xmath102 steps . we also display in each box the @xmath183 distribution with @xmath11 set to zero as a reference distribution . in the box for @xmath221 the two distributions are indistinguishable . the @xmath183 distribution is hence rank zero and both lln and clt hold . at @xmath222 the distribution with @xmath223 becomes a gaussian with enhanced msd . it is rank one and indicates the hidden coherence ; clt holds but lln does not and @xmath224 ) . but the continuous change of dynamics to higher ranks , such as from the second box to the third box was not noted . ] . now in the third box at @xmath225 the distribution turns out to be the rank two distorted gaussian distribution . even clt no longer holds . in the fourth box we also exhibit the @xmath183 distribution at @xmath226 , @xmath223 and @xmath65 . this is in our classification the rank three distribution with two overlapping gaussian peaks . in the model with @xmath191 , we can move to this dynamics region by varying @xmath11 , i.e. by traversing the turbulent regime . this point is further substantiated in the following subsections . at this point we need some explanation on the terminology of the _ hidden _ coherence . the interesting phenomenon that above some @xmath214 the lln in the time series is violated certainly occurs but now we have found that it is restricted to the second - class peak regions only . there are valleys where lln approximately holds even for large @xmath2 and also the first - class peak regions where lln is violated even for small @xmath2 . and now even in the second - class peak regions the place where the clt holds ( with lln violated ) is again restricted . for @xmath191 it is only a narrow band around @xmath217 and it moves to the larger @xmath11 for larger @xmath12 . thus if a distinction between the rank one and two distribution is made and the hidden coherence is defined strictly as the phenomenon that clt holds with no approximation ( i.e. excluding the rank two distribution ) but with broken lln , it can live only in a very restricted region among the second - class peak regions . we will further show below that a careful analysis of the decay exponent of the temporal correlator can also unveil the coherence among maps . but still we consider the first discovery@xcite of the onset of the violation of lln _ even in the region where the clt holds ( rank one ) _ is remarkable . our point in emphasizing the distinction between rank one and two distributions is that the recognition that the rank one dynamics region is adjacent in the @xmath188 space to the rank two dynamics region which in turn is adjacent to the rank three and so on enables us to track down the change of dynamics and thus leads us to a working hypothesis that the hidden coherence is the most modest manifestation of the coherent quasi - periodic motion of maps ubiquitous in the turbulent regime . now that we have identified the various regions in the turbulent regime with distinct manifestation of periodicity let us to proceed to clarify how the dynamics of maps in these regions are related each other . to collect necessary information it is best to traverse the turbulence regime at fixed @xmath227 and varying @xmath11 as a tuning parameter . in fig . [ fig11 ] we present a sample record of such an expedition at @xmath228 . the first column shows the mean field distributions as well as the single map distributions . the former is the average during the steps from @xmath203 to @xmath229 , and the latter is the average during the last @xmath230 steps of @xmath204 iterations . in the second column the rank of the @xmath183 distribution is denoted at the top in each box . the evolution of a hundred selected maps is displayed for the last seven steps . in order to chase the periodic motion of elements we devise a simple temporal correlator of the _ relative coordinate vector _ @xmath231 of maps around the mean field : @xmath232 the average over @xmath85 is taken for the last @xmath99 steps . this correlator is displayed in the third column . we show it just as defined by ( [ correlator ] ) but for the measurement of the decay exponent we quote values obtained by the signed logarithm plot ( @xmath233log@xmath234 ) of @xmath235 . for smaller lattice analysis we also averaged over data with different initial configurations to improve the statistics but this did not cause any significant change for the exponent estimation . the last column shows the return maps of @xmath183 ( the set of points ( @xmath236 ) from the last @xmath99 steps . let us start at pure randomness @xmath192 . the map average distribution has many sharp peaks with fractal hierarchy structure reflecting the unstable fixed points of a single map in the chaotic region . the @xmath183 distribution obeys of course both lln and clt and naturally sharp gaussian ( rank zero ) for @xmath65 . the evolution plot shows a simple logistic pattern and the return map is a simple tiny cloud of points . the correlator is one at @xmath237 by definition and decays almost instantly . we can foresee that periodicity three window in the first - class peaks is waiting us ( @xmath238)$ ] but let us choose @xmath239 to see if any precursor of it emerges . we find that the @xmath183 distribution has become rank one . the map average distribution has lost some sub - peaks since any one of the maps is pulled by the mean field in fluctuation and it has become more difficult for it to stay at the unstable sub - fixed points for long time . the coherence is still invisible neither in the map evolution nor in the return map . but we find that remarkably the correlator now oscillates in period three and decays exponentially ( @xmath240 steps for @xmath241 decay ) . this period three damping oscillation of the correlator starts at @xmath242 except for the valleys and the decay exponent becomes smaller for larger @xmath11 . at @xmath243 the @xmath183 distribution is rank two . we are now very close to the first - class peaks . the correlator survives longer period ( @xmath244 steps for @xmath241 decay ) in period three motion . but the return map does not sense this . from the evolution diagram we can immediately tell that this failure of the return map is due to mixing of the maps between the quasi - clusters at a high rate . this inadequacy of a return map should be noted . at @xmath245 we are on the edge of the first - class peak regions . now very careful analysis is necessary to account for the first transient behavior of the maps . the truncation of the first @xmath102 steps is not sufficient and we have to watch the evolution carefully until at least during @xmath204 steps . we show two typical samples at @xmath245 . in the first sample at @xmath245 the maps remain in a few quasi - clusters with mixing until the end of the run . the unstable quasi - clusters oscillate in period three and the correlator decays to @xmath241 only after @xmath246 due to the smaller mixing than before . now we can see in the return map three clouds of points turn out . the map average distribution has only three prominent peaks and all other sub peaks are lost . the corresponding @xmath183 distribution is rank three with both a visible peak and a shoulder on top of a broad band . the high msd of the @xmath183 distribution comes from the large separations between the dominant unstable fixed points of the maps in interaction . this distribution is an example of overlapping gaussian distributions . for the perfect periodic motion of the tight clusters , sharp delta peaks would be observed in both map and @xmath183 distributions . due to the decay of the quasi - clusters caused by the mixing the sharp delta peaks become broad and turn into overlapping gaussian distributions . we frequently observed more prominent overlapping two or three gaussian peaks in the whole analysis of the turbulent regime using both @xmath12 and @xmath11 as tuning parameters . ( see , for instance , the fourth box in fig . [ fig10 ] . ) in the second sample at @xmath245 the maps literally dropped into period three attractor of three clusters at @xmath247 . as the map evolution plot shows the map positions in each cluster exhibit some variance in evolution so that we would better describe the clusters as quasi - clusters . the broad band in this rank three @xmath183 distribution is an artifact of the transient motion before the maps fall into the final attractor . there is no mixing at all between the clusters . the @xmath183 distribution is a sharp single peak and the @xmath183 return map is confined in a very small range . this is due to the fact that the three clusters happen to be almost equally populated in this run so that the mean field fluctuation is minimized . at @xmath248 we are at the summit of the first - class peak . here the maps almost always drop into period three attractor of two clusters . by only two clusters in period three motion the mean field must also fluctuates in period three . thus the @xmath183 distribution has sharp three peaks and msd turns out high . the value of msd is solely determined by the dynamics of clusters and @xmath2 independent . also the return map is widely spread . while we traverse the summit of the first - class peaks till @xmath249 the attractor remains in period three . sometimes it bifurcates to period six . in further expedition we depart from the first - class peaks and go down into the second class peak regions . everything then occurs in reverse way in this descent till @xmath250 . from @xmath251 we traverse the second - class peaks as well as the valleys . at the second - class peaks the correlator survives above @xmath241 until @xmath252 but with no apparent low periodicity . then above @xmath253 the correlator starts again catching the precursor of the period two opposite phase motion in the two - cluster regime of the gcml . thus the indeterminacy of the period of quasi - clusters beyond second peak regions comes from the fact that the regions are just in between the period three window and two - cluster opposite - phase motion regions . after @xmath254 the correlator shows perfect period two motion and the decay is smoothly prolonged with increasing @xmath11 . at @xmath255 it only decreases by a few percent even at @xmath256 . correspondingly , above @xmath254 the middle peak in the map average distribution starts approaching to the peak at higher @xmath186 and above @xmath257 the map average distribution splits into two ( no events in the middle gap region ) . still in the evolution plot we can see the maps are in loose clusters with mixing between each other . the high msd of the rank two and three distributions above @xmath254 is clearly induced by the widely spread dominant unstable fixed points of low periodicity . we have done a similar analysis at various @xmath12 up to @xmath226 and verified that all above features at @xmath191 are unchanged . the various manifestation of the periodicity is summarized in table [ table1 ] for @xmath191 . only a minor difference for larger @xmath12 is that the regions of rank one distribution above the first rank msd peaks become larger shifted to larger @xmath11 and also the rank three distribution starts to appear there . since the rate of the mixing of maps between clusters is the key to the turbulent regime we have also measured the temporal correlator of the clustering pattern matrix similar to the anderson - edwards order parameter@xcite . it exhibits the same variation of the decay exponents with our correlator . we examine below how the dynamics in various regions are related each other . ( 1 ) let us first compare the dynamics of the periodic regions with that at @xmath258 . at @xmath258 maps evolve almost independently with long visit at all of the unstable fixed points of the logistic map . thus the map average distribution has many peaks with fractal structure peaking . since the maps behave as almost independent random variables with the same probability distribution , naturally both lln and clt hold . at the periodic window in the first - class msd peak regions , on the other hand , the @xmath23 focusing effect tunes the maps to form tightly bound two or three clusters mutually oscillating in period three attractor and this leads to the rank four distribution with sharp delta peaks . the dynamics of clusters in the periodic attractor is solely determined by @xmath12 , @xmath11 and the effective couplings between the clusters determined by the population _ ratios _ among the clusters ) does exhibit the same period three attractor at the same value of @xmath11 . ] . the msd of the @xmath183 distribution is again determined by the @xmath12 , @xmath11 and the population _ ratios _ between the clusters , and there is no room for the explicit @xmath2 dependence . similar coherent dynamics is realized also in the ordered two - cluster regime beyond the turbulent regime ( @xmath259 for @xmath191 ) . the turbulent regime has a period three window in the center of it and the two - cluster regime beyond it . we discuss below how these affect other regions . \(2 ) when @xmath11 deviates from the range necessary for the period three cluster dynamics ( or for the period two - cluster dynamics ) , the perfect synchronization of maps among each cluster is no longer realized . but what we observed in the last subsection is that even for some large deviation of @xmath11 the maps still move under certain coherence if @xmath2 is beyond the threshold @xmath214 . such coherent motion of maps can be best described by _ unstable quasi - clusters _ the decay life of which is given by the inverse of the exponent of the correlator . the maps now divide into a few masses _ at any instance _ of the evolution . in a short time scale these masses oscillate mutually in the same periodicity as before as we observed in both the map evolution plot and the correlator . but the variance of map positions among each mass is not so small and the synchronization of maps in each mass is also not perfect . hence we call these masses as quasi - clusters . due to the fail of perfect synchronization , the quasi - clusters mix each other by the exchange of maps and soon they loose their identities . so the quasi - clusters are unstable and have short life time . but still at any instance of evolution there do exist quasi - clusters . the quasi - clusters defined at a certain time decay but in the evolution of maps new quasi - clusters are kept created . we may regard these quasi - clusters as _ remnants _ of the perfect periodic motion of maps . in fact when we approach the periodic regions the decay exponent of the correlator decreases smoothly , that is , the life time of the quasi - clusters increases smoothly , signaling the phase transition from turbulence to periodicity . in pure turbulence the system is invariant under the exchange of maps , while in the system of maps divided into clusters this permutation symmetry is broken . simultaneously the system in pure turbulence is invariant under the temporal translation , while the system in periodic motion is only invariant under the temporal translation just for the period . the self - organized state of maps in periodic quasi - clusters emerges as the state of broken symmetry from pure turbulence by a weak self - coupling and it emerges only for large @xmath2 . this is interestingly reminiscent of the onset of ordered parameter at the spontaneous break down of the global symmetry in the field theory where an infinite number of dynamical degree of freedom is necessary in order to support the ordered vacuum . see , for instance , weinberg@xcite where the broken symmetry state of a chair is elucidated . it is important that the maps in the quasi - clusters approximately follow the previous cluster attractor in the life time of the quasi - clusters . this means that the @xmath183 distribution is controlled by the scale of the previous attractor approximately and again determined by the @xmath12 , @xmath11 and the population _ ratios _ and not by @xmath2 . thus there emerges a @xmath183 distribution whose msd does not decrease with @xmath2 and lln is violated . depending on the rate of mixing and the allowed population ratios at the @xmath11 , distributions of various ranks emerge . \(3 ) if the deviation of @xmath11 from the necessary range for periodic cluster attractor is not large , the failure in the map synchronization is not also large . the quasi - clusters decay by low rate mixing of members so the correlator survives for long time exhibiting clearly the periodicity of the previous periodic cluster attractor . but the would - be delta - peaks of both the average map distribution and the @xmath183 distribution at no mixing now change into overlapping gaussian distributions with visible separation between the peaks . these changes of the correlator and the @xmath183 distribution at small deviation of @xmath11 are precisely what we observe around the first - class msd peaks and near the upper boundary of the turbulent regime adjacent to the ordered two - cluster phase . with further deviation of @xmath11 the mixing rate becomes higher . the correlator decays more quickly and the @xmath183 distribution becomes from rank three to rank two - the distorted gaussian distribution . we observe this when we go down from the first - class peaks to the region of @xmath258 or when we approach the first - class peak regions from above . in the former case the correlator keeps exhibiting the period three damped oscillation even if we go down into the rank one @xmath183 distribution region . in the latter case the periodicity of the correlator becomes indefinite below @xmath254 for @xmath191 since naturally we are coming down from two - cluster regime with period two to the period three window , and we finally loose the track of a periodicity remnant ( rank two distribution ) when we go further into the narrow band of rank one @xmath183 distribution region . this is the genuine hidden coherence region . \(4 ) the place the hidden coherence can live has now become very limited since we have improved the sensitivity to the coherence effect . in the old analysis@xcite , the distinction between rank one and two distributions was ignored and neither the change of the decay exponent of the correlator with @xmath11 nor the periodicity of it was analyzed . the periodic window and the valleys were not discussed also . then the coherence seemed to live hidden everywhere in the turbulent regime if @xmath260 . now , both the periodic window with manifest coherence and the valleys with valid lln should be excluded first of all . furthermore , in the regions where the correlator has clearly caught the periodicity remnant , it is no longer legitimate to describe the coherence as hidden . thus , regarding the correlator the coherence is hidden only in a limited region in between the period - three and period - two remnant regions among the second - class msd peak regions . this is the region with @xmath251 for @xmath191 . see table [ table1 ] and eq . ( [ second ] ) . here the correlator does show longer decay time ( @xmath241 decay within @xmath261 steps ) than that at @xmath258 but it decays with large fluctuation showing no low periodicity . this is natural for the region in between two different periodicity regions . but not only the correlator but also the shape of the @xmath183 distribution is a vital quantity to keep track of the change the dynamics . almost everywhere in the possible residence of the hidden coherence regarding the correlator , the @xmath183 distribution is rank two ( rather than gaussian ) and naturally understood as the smooth continuation of the rank three overlapping gaussian distribution . the clt is apparently violated here and the distribution is still controlled by the scale of the attractor . therefore , the place where the coherence is literally hidden from both the correlator and distribution analyses is now very limited . it is the narrow band at @xmath262 for @xmath191 and around @xmath263 for @xmath226 where the @xmath183 distribution is really gaussian with enhanced msd ( rank one ) and the correlator does not show the low periodicity . here the coherence seen in the enhanced msd can be only substantiated either by the finite mutual information or by the instability of the fp fixed point distribution . but in ( 3 ) above we successfully tracked down the smooth change of distribution ( and the corresponding change of the decay exponent of the correlator ) with the deviation of @xmath11 from rank four down to rank two distribution regions which surround the literally hidden coherence region . this gives a new way to look at the hidden coherence . higher rate mixing with further smooth deviation of @xmath11 should necessarily weaken the coherence among the maps and change the distorted gaussian ( rank two ) to the gaussian distribution . if the scale of the attractor still remains to govern the map dynamics the @xmath183 distribution with @xmath2-independent msd would emerges . we propose this last statement as a working hypothesis for the origin of the hidden coherence . up to this point we have heavily depended on the numerical simulation . our concern is the statistical properties of the large - size gcml ( egcml ) and the monte - carlo simulation is naturally the basic approach to the phenomenology of such system . we have shown that a faithful extensive analysis of the data unveils amazing new features of the system . but , of course , if by some analytical method , one could relate the systematic manifestation of periodicity in the turbulent regime to the periodic behavior of the element of the system , namely the period three window of a single logistic map , it would be certainly a step forward to the real understanding of the turbulent regime . in this last subsection we discuss such an analytic approach . let us consider a typical case where the n maps are divided into equally populated three clusters @xmath264 ( @xmath265 ) and the clusters move cyclically round the three fixed positions @xmath266 . in such a maximally symmetric and period three motion of gcml maps , @xmath267 , the meanfield of all maps at time @xmath85 becomes a time - independent constant which is nothing but an equal weight average @xmath268 of @xmath269 . to see this let us first rewrite the meanfield using the invariance rule ( [ sumrule ] ) as @xmath270 then for the configuration in case the last average can be calculated as @xmath271 where @xmath272 denotes the coordinate of the cluster @xmath273 at time n. therefore , if such a configuration is produced by the interaction between the maps via the meanfield , any one of gcml maps must be evolving by a common evolution equation @xmath274 we can cast the evolution equation ( [ consthevolution ] ) with a constant meanfield @xmath268 and @xmath275 to a simple logistic map @xmath276 with a reduced coupling constant @xmath277 by a linear scale transformation @xcite @xmath278 the reduction rate @xmath279 is given by @xmath280 the period three window of the logistic map ( [ yevolution ] ) starts at @xmath281 by the tangent bifurcation and after the sequential bifurcation it closes at @xmath282 by the crisis . therefore the reduction rate @xmath279 must be for @xmath191 in the range @xmath283 and at a given rate @xmath279 within this range , ( [ rconstarint ] ) gives a curve of constraint on the @xmath284 plane . there is another constraint from self - consistency . each of the maps @xmath285 is related to the gcml maps @xmath286 by the linear transformation ( [ lineartransformation ] ) and so is the average value @xmath287 of maps @xmath285 to @xmath268 : @xmath288 since @xmath287 is simply an equal weight average of the period three stable orbits of ( @xmath289 ) , it can be estimated solely by the property of the logistic map without any recourse to the gcml evolution equation . for instance at the tangent bifurcation point ( @xmath290 ) , we have @xmath291 and from matching of the coefficients we obtain @xmath292 . numerically we find that @xmath293 slightly below the threshold @xmath294 , drops sharply ( @xmath295 ) to @xmath296 at the threshold @xmath297 , and @xmath287 gradually changes around @xmath298 until the end of the window ( @xmath299 ) . by eliminating @xmath268 from ( [ rconstarint ] ) and ( [ yconstaraint ] ) we obtain @xmath300 the estimated @xmath11 for @xmath191 from ( [ epsilonestimates ] ) is @xmath301 the estimates a , b , c and d are evaluated respectively below the threshold , at the threshold , at the first bifurcation point at the middle , and at the upper edge of the period three window in the order of increasing @xmath277 . before comparing with the simulation result we should note that a maximally symmetric configuration of equally populated three clusters in period three motion is assumed in deriving ( [ epsilonestimates ] ) . thus , the value of @xmath11 necessary for such a configuration should be deduced from the genuinely period three region , namely from the interval from b to c. we can predict that @xmath11 must be in the range ( 0.0365 , 0.0422 ) for the formation of such a maximally symmetric gcml configuration . for such a configuration the meanfield should not fluctuate in time and it will show up as a deep valley in the gcml msd surface . we have added two exterior estimates in ( [ estimates ] ) expecting that these extensions will cover the asymmetrically populated clusters in period three motion . now let us compare ( [ estimates ] ) with the range of the period three first - class peaks found in section v b , namely , @xmath302 and @xmath212 . firstly there is a remarkable agreement between the range of peaks and the range of estimates a to d in ( [ estimates ] ) . the whole period three window of a single logistic map shows up as the whole first - class peaks . a closer comparison with the observations in section v d gives further support for the link of the structure observed in the numerical simulation and the period three window of the logistic map . * we noted that in the deep valley of the msd surface between the two first - class peaks the formation of three clusters in period three motion . the valley region @xmath303 remarkably matches the above estimated range ( 0.0365,0.0422 ) for the maximally symmetric configuration . * we noted that in the upper first - class peak ( @xmath212 ) the maps almost always form only two clusters in period three motion . it is natural that the larger @xmath11 reduces the number of clusters from three to two . as we noted the extremely high msd in the first - class peak is induced by the asymmetric population configuration ( @xmath304 ) . furthermore we have verified by simulation that the first - class peak regions shift to higher @xmath11 when @xmath12 is increased precisely in accord with the prediction by ( [ epsilonestimates ] ) . it is remarkable that the maps with high nonlinearity are driven into coherent synchronizing clusters with very small interaction parameter @xmath11 once it takes the tune up value . a lot of further effort has to be devoted to fully appreciate the implication . the analytic approach here is only a step forward to this direction . a survey of the overall relation of peak - valley structure of the msd surface to the hierarchical windows of a single logistic map as well as the dynamical variation of the population configuration is now underway . in this article we have revisited the gcml and first resolved the puzzle of the posi - nega switch realized the two - cluster phase . the puzzle was that the maps posses the memory of the cluster to which they subject even though they pass through a transient chaotic process where many channels are open . we have proved that maps never mix across the mean field of them in the transient process . we have then analyzed the turbulent regime of gcml . to date the so - called hidden coherence has been the main target of interest in this region . we have presented our new finding that there exists a remarkable period three attractor regions in this regime . the turbulent regime locates in - between the genuinely turbulent region around @xmath192 and two - cluster regime of period two attractor and has this period three attractor regions in the center of it . we have shown that the remnant of the periodic coherence can be found almost everywhere in the turbulent regime by the measurement of the temporal correlator of the fluctuation of maps around the mean field and by the extensive analysis of the @xmath183 distributions over the whole turbulent regime . the dynamics of the turbulent regime can be well described by the quasi - clusters with mixing of elements between each other . the hidden coherence may be regarded the most modest manifestation of this remnant periodicity . by an analytic approach we have shown that the prominent periodicity manifestation in the turbulent regime is caused by the period three window of a single logistic map . a new extended globally coupled map lattice(egcml ) is constructed in which the interaction between maps decreases proportionally to the distance . egcml at @xmath42 has the same phase diagram with that of gcml . the bifurcation tree structure of the attractor of egcml in the two - cluster regime is also the same with that of gcml . the maps in each egcml cluster are phase synchronizing rather than synchronizing . the posi - nega switch between the two clusters is realized also in egcml . in the chaotic transient process in egcml - switch , formation of amazing spatial clusters of maps is observed . egcml may be viable for the simulation of chaotic diffusion process . we have also shown that the turbulence regime of egcml is also not immune from periodicity just like gcml turbulent regime . as a whole this work is an exploration of order in the chaos . the posi - nega switch in gcml is realized because there occurs no mixing in the chaotic transient process . in the chaotic transient process in a egcml - switch , interesting spatial clusters are formed . and the turbulent regime of both of gcml and egcml is dominated by the quasi - clusters of maps in periodic motion . it is a pleasure to note the author s debt of gratitude to k. kaneko . this work was started several years ago stimulated by the pioneering articles on gcml by him@xcite and the compact model gcml has been an unfailing source of interest since then . the author is thankful to fumio masuda who worked on this subject together with the author until the last one year of completion and the nice collaboration time with him is memorized . thanks also go to hidehiko shimada for many crucial suggestions , hayato fujigaki for his collaboration on the study of coupled flows done in parallel with this work , and kengo kikuchi for his unfailing effort on the analytic approach to gcml now underway . last but not at least the author wishes to thank wolfgang ochs at the max - planck institute for sharing his insight in this research and reading the manuscript . this work was supported by the faculty collaborative research grant from meiji university , grant - in - aids for scientific research from ministry of education , science and culture of japan , and grant for high techniques research from both organizations . k. kaneko , phys . lett . * 65 * , 1391 , ( 1990 ) ; + k. kaneko , physica ( amsterdam ) * 55d * , 368 , ( 1992 ) ; ibid . , * d86 * , 158 , ( 1995 ) . k. kaneko , physica ( amsterdam ) * 23d * , 436 ( 1986 ) ; + j. p. crutchfield and k. kaneko , _ in directions in chaos _ , edited by b. -l . hao ( world scientific , singapore , 1987 ) . see , for instance , + j.j . hopfield , proc . usa ( biophysics)*79 * , 2554 ( 1982 ) ; + w.a . little , math . * 19*,101 ( 1974 ) ; + d. j. amit , h. gutfreund and h. sompolinski , phys . rev . * a32 * , 1007 ( 1985 ) . .the various periodicity manifestation in the turbulent regime of gcml ( @xmath191 and @xmath65 ) . note that in each range of @xmath11 there are also valleys with an approximate lln along with listed peaks with the broken lln . [ table1 ] [ cols="^,^,^,^ " , ] Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: as their name suggests , compact groups are small systems of several galaxies in a compact configuration on the sky . the first example was found over one hundred years ago by stephan ( 1877 ) who observed it visually using the 40-cm refractor of the observatoire de marseille . stephan s quintet is a small group of five galaxies , three of which show strong tidal distortions due to gravitational interaction . a second example was found 71 years later by seyfert ( 1948 ) from a study of harvard schmidt plates . seyfert s sextet ( figure 1 ) is one of the densest groups known , having a median projected galaxy separation of only @xmath1 kpc ( the hubble constant @xmath2 km s@xmath3 mpc@xmath3 ) . the palomar observatory sky survey ( poss ) provided a new and extensive resource for the systematic investigation of small groups of galaxies . two catalogs , the _ atlas of interacting galaxies _ ( vorontsov - velyaminov 1959 , 1975 ) and the _ atlas of peculiar galaxies _ ( arp 1966 ) , contain galaxies or galaxy groups selected on the basis of visible signs of interaction or peculiar appearance . in addition to stephan s quintet and seyfert s sextet , these include many new compact groups , including a striking chain of five galaxies , vv 172 . prior to these , shakhbazian ( 1957 ) had discovered a small dense cluster of 12 faint red galaxies that appeared so compact that they were initially mistaken for stars . over the next two decades , shakhbazian and collaborators examined over 200 poss prints covering 18% of the sky and cataloged 376 additional `` compact groups of compact galaxies '' ( shakhbazian 1973 , shakhbazian & petrosian 1974 , baier et al 1974 , petrosian 1974 , 1978 , baier & tiersch 1975 - 79 ) . apart from occasional photographic or spectroscopic observations ( eg . mirzoian et al 1975 , tiersch 1976 , massey 1977 , shakhbazian & amirkhanian 1979 , vorontsov - velyaminov et al 1980 , vorontsov - velyaminov & metlov 1980 ) these systems initially received little attention . however , interest in them is growing . although the majority seem to be small clusters , they share some of the properties , and pose some of the same questions , as do compact groups . when redshifts were measured for galaxies in the first compact groups ( burbidge & burbidge 1959 , 1961a ) , surprises were found . both stephan s quintet and seyfert s sextet contain a galaxy with a discordant redshift . it seemed unlikely that a foreground or background galaxy would appear so often projected within such compact systems ( burbidge & burbidge 1961a ) . this impression was further reinforced with the discovery of yet another discordant redshift in vv 172 ( sargent 1968 ) . are these examples of physical association between objects of widely different redshifts , as has been advocated for many years by arp ( 1987 ) ? even if the discordant galaxies are ignored , the velocity dispersions of these systems are generally higher than would be expected given their visible mass ( burbidge & burbidge 1959 , 1961b , 1961c , burbidge & sargent 1971 ) . it was argued that such groups must be unbound and disrupting ( eg . ambartsumian 1961 ) , although limber and mathews ( 1960 ) showed that the virial theorem could be satisfied for stephan s quintet , given the uncertainties in the projection factors , if the individual galaxy masses were considerably larger than those of isolated galaxies . the observations can of course also be explained if the bulk of the mass is in a non - visible form . in hindsight , this was one of the earliest indications of the possible existence of dark matter in galactic systems . a new problem emerged with the realization that bound groups would be unstable to orbital decay resulting from gravitational relaxation processes ( peebles 1971 ) . a simple calculation indicated that the dynamical - friction timescale was much shorter than the hubble time ( hickson et al 1977 ) , as was soon confirmed by numerical simulations ( carnevali et al 1981 ) . at the same time , it became increasingly clear that mergers played an important role in the evolution of many , if not all , galaxies ( press & schechter 1974 , ostriker & tremaine 1975 ) . compact groups emerged as prime locations for investigations of the dynamical evolution of galaxies . motivated by the desire for a homogeneous sample that could be subject to statistical analysis , rose ( 1977 ) , and later , hickson ( 1982 ) produced the first catalogs of compact groups having specific , quantitative , selection criteria . subsequent detailed investigation , at many wavelengths , has produced a large body of observational data for the hickson catalog . as a result , it has now become possible to address some of the outstanding questions concerning the nature of compact groups and their role in galaxy evolution . not surprisingly , new questions have been raised and new controversies have appeared . however , much progress has been made in resolving both old and new issues . this review is organized as follows : in section 2 , the definition of a compact group is discussed , along with methods of identification and surveys that have been made . in sections 3 - 5 , observed properties of these systems are summarized and discussed . sections 6 and 7 focus mainly on interpretation of the observations , and on implications of these results . all work on compact groups of galaxies can not possibly be discussed in this short paper , although an attempt is made to touch upon most current topics . other recent reviews of compact groups and closely related subjects include those by white ( 1990 ) , hickson ( 1990 , 1997 ) , whitmore ( 1992 ) , kiseleva & orlov ( 1993 ) , sulentic ( 1993 ) , and mamon ( 1995 ) . by `` compact group '' , we mean a small , relatively isolated , system of typically four or five galaxies in close proximity to one another . such groups do not necessarily form a distinct class , but may instead be extreme examples of systems having a range of galaxy density and population . because of this , the properties of the groups in any particular sample may be strongly influenced by the criteria used to define the sample . the early surveys used qualitative criteria that , while successful in finding many interesting individual objects , do not easily allow one to draw broad conclusions about the groups as a whole . thus , the focus in recent years has been on samples selected using specific , quantitative , criteria . these criteria define the minimum number and magnitude range of the galaxies , and also consider the galaxy spatial distribution . the use of quantitative selection criteria was pioneered by rose ( 1977 ) who searched for groups that have three or more galaxies that are brighter than a limiting magnitude of 17.5 , and that have a projected surface density enhancement of a factor of 1000 compared to the surrounding background galaxy density . searching an area of 7.5% of the sky , he found 170 triplets , 33 quartets , and 2 quintets . unfortunately , the sample received little follow - up study . sulentic ( 1983 ) re - examined the 35 rose groups which contain four or more galaxies and found that only a third actually satisfied the selection criteria . this is testimony to the difficulty of visual searches . a more fundamental problem is that the fixed magnitude limit in the selection criteria makes the sample susceptible to strong distance - dependent biases . in an attempt to reduce such effects , hickson ( 1982 , 1993 , 1994 ) adopted a relative magnitude criterion , selecting systems of four or more galaxies whose magnitudes differ by less than 3.0 . a distance independent ( to first order ) compactness criterion was employed : @xmath4 , where @xmath5 is the mean surface brightness of the group calculated by distributing the flux of the member galaxies over the smallest circular area containing their geometric centers . to avoid including the cores of rich clusters , an isolation criterion was necessary so as to reject the group if a non - member galaxy , not more than 3 mag fainter than the brightest member , occurred within three radii of the center of the circle . ( a non - member galaxy is a galaxy which if included in the group would cause the group to fail one or more of the selection criteria . ) from a search of 67% of the sky ( all the poss prints ) , and using magnitudes estimated from the poss red prints , exactly 100 groups were found satisfying these criteria ( hereafter hcg s ) . as the hcg sample is now the most widely studied , it is important to examine the biases introduced by the criteria , and by the visual search procedure . any sample selected on the basis of surface density will suffer from geometric and kinematic biases . the former occurs because non - spherical systems will be preferentially selected if they are oriented to present a smaller cross - sectional area ( eg . prolate systems pointed towards us ) ; the latter because we will preferentially select systems that , owing to galaxy orbital motion , are momentarily in a more compact state ( transient compact configurations ) . thus , a compact group might result from a chance alignment or transient configuration within a loose group ( mamon 1986 ) . this question will be considered in more detail in section 6.2 . other biases arise from the subjective nature of the search procedure . the original catalog contains a few mis - identifications , such as compact galaxies mistaken for stars , and marginal violations of the isolation criteria . in addition , when photometry was obtained for the galaxies in the catalog ( hickson et al 1989 ) , it was found that some groups would not satisfy the selection criteria if photometric magnitudes are used . attempts to rectify these problems have been made by hickson et al ( 1989 ) and sulentic ( 1997 ) . however , it should be emphasized that changes , such as using photometric magnitudes in the selection criteria , are not corrections to the catalog , but are actually the imposition of _ additional _ a postiori selection criteria . the resulting subsample is by no means complete because the new criteria are applied only to the visually - selected hcg catalog and not to the entire sky . because of the difficulty of identifying faint groups , the hcg catalog starts to become significantly incomplete at an integrated magnitude of about 13 ( hickson et al 1989 , sulentic & rabaa 1994 ) . a more subtle effect results from the difficulty of recognizing low - surface - brightness groups . because of this , the catalog also becomes incomplete at surface brightnesses fainter than 24 ( hickson 1982 ) . yet another effect is that groups may be more noticeable if the magnitude spread of their members is small . thus , the catalog may also be incomplete for magnitude intervals greater than about 1.5 ( prandoni et al 1994 ) . these effects are of critical importance in statistical analyses of the sample . one immediate conclusion is that the actual number of groups which satisfy the selection criteria may be considerably larger than the number found by a subjective search . it has recently become feasible to find compact groups by automated techniques . mamon ( 1989 ) used a computer to search tully s ( 1987 ) catalog of nearby galaxies and identified one new compact group , satisfying hickson s criteria , in the virgo cluster . prandoni et al ( 1994 ) applied similar criteria to digital scans of @xmath6 deg@xmath7 around the southern galactic pole and detected 59 new southern compact groups ( scgs ) . observations are presently underway to obtain accurate photometry and redshifts for this sample ( iovino private communication ) . the digitized palomar sky survey ii also offers new opportunities for the identification of compact groups ( de carvalho & djorgovski 1995 ) . an alternative approach is to identify groups of galaxies from redshift information , as was first done by humason et al ( 1956 ) . with the advent of large - scale redshift surveys , it has become possible to identify a reasonably - large sample of compact groups in this way . barton et al ( 1996 ) have compiled a catalog of 89 redshift - selected compact groups ( rscgs ) found by means of a friends - of - friends algorithm applied to a complete magnitude - limited redshift survey . galaxies having projected separations of @xmath8 kpc or less and line - of - sight velocity differences of 1000 km s@xmath3 or less are connected and the sets of connected galaxies constitute the groups . the numerical values were chosen to best match the characteristics of the hcg sample , and indeed , many of those rscgs that have at least four members are also hcgs . there are some significant differences , however : because foreground and background galaxies are automatically eliminated by the velocity selection criteria , this technique is more effective at finding groups in regions of higher galaxy density , which would fail the hcg isolation criterion . this criterion requires that the distance to the nearest neighbor be at least as large as the diameter of the group . the rscg criteria , on the other hand , require only that the nearest - neighbor distance be larger than the threshold distance ( @xmath8 kpc for the rscgs ) , which may be considerably smaller . it will therefore allow the inclusion of groups that are physically less isolated ( and therefore less physically distinct ) than would the hcg criterion . one would also expect that the numbers of groups found in a given volume by the less - restrictive rscg criteria to be larger than by the hcg criteria , as seems to be the case . while the redshift - selection method compliments the hcg angular - selection technique , one should keep in mind that it also selects groups according to apparent ( projected ) density the velocity information serves only to reject interlopers . thus it will be subject to some of the biases discussed above . also , because the galaxy sample used is magnitude - limited , rather than volume - limited , there will be redshift - dependent biases in the rscgs . however , the well - defined selection criteria and the completeness of the sample , should allow a quantitative determination of the effects of this bias . the space distribution and environment of compact groups provides important clues to their nature . the median redshift of the hcgs is @xmath9 , placing most of them at distances well beyond the virgo cluster ( hickson et al 1992 ) . a cursory inspection reveals that they are fairly uniformly distributed and show no preference for rich clusters . this is at least partly due to the isolation criterion . however , galaxies in rich clusters have rather different kinematical and morphological properties than do those in compact groups , so one might justifiably argue that small clumps of galaxies within clusters are not compact groups . a natural question is whether or not compact groups are associated with loose groups . rood & struble ( 1994 ) observed that 70% of the hcgs are located within the bounds of cataloged loose groups and clusters . studies of the distribution of galaxies in redshift space ( vennik et al 1993 , ramella et al 1994 , sakai et al 1994 , garcia 1995 , barton et al 1996 ) indicate that compact sub - condensations do occur within loose groups and filaments . vennik et al ( 1993 ) and ramella et al ( 1994 ) find that most hcgs are indeed associated with loose groups . while the above studies show that compact groups trace large - scale structure , it is also clear that they prefer low - density environments . sulentic ( 1987 ) , rood & williams ( 1989 ) , kindl ( 1990 ) , and palumbo et al ( 1995 ) have examined the surface density of galaxies surrounding the groups . they generally agree that about two - thirds groups show no statistically significant excess of nearby neighbors . this is not inconsistent with the redshift - space results because most of the hcg associations identified by ramella et al ( 1996 ) contain fewer than 5 excess galaxies . thus while compact groups are associated with loose groups and filaments , these tend to be low - density and sparsely - populated systems . are the galaxies in compact groups in any way distinct from those in their immediate environments ? rood & williams ( 1989 ) and kindl ( 1990 ) both found that compact groups , including those in rich environments , contain a significantly smaller fraction of late - type ( spiral and irregular ) galaxies than do their neighborhoods . this result is of particular importance to the question of the physical nature of compact groups , discussed below in section 6 . in addition , many independent studies have found significant differences between galaxies in compact groups and those in other environments . these are examined in section 5 . the first studies of individual compact groups ( burbidge & burbidge 1959 , 1961b , 1961c , burbidge & sargent 1971 , rose & graham 1979 , kirshner & malumuth 1980 ) indicated short dynamical times and mass - to - light ratios intermediate between those of galaxies and rich clusters . however , because of the small number of galaxies , estimates of the space velocities and physical separations of galaxies in individual groups are highly uncertain . meaningful dynamical conclusions about systems containing only four or five galaxies requires statistical analysis of large homogeneous samples . by 1992 , velocities had been measured for almost all 462 galaxies in the hcg catalog ( hickson et al 1992 ) . the distribution of galaxy velocities , relative to the median of each group , is shown in figure 2 . it can be seen that the majority of velocities fall within a roughly gaussian distribution of standard deviation @xmath10 km s@xmath3 ( the sharp peak at zero velocity results from the use of the median ) . this characteristic velocity is quite similar to velocity dispersions found in loose groups , and much smaller than typical velocity dispersions in rich clusters . in addition to the gaussian core , a flat component is seen in the velocity distribution of figure 2 . this is expected as some galaxies which are not physically related to the group will appear projected on the group by chance . this component contains about 25% of the total number of galaxies . whether of not chance projection can account for such a large number of `` discordant '' galaxies is still a matter of some debate , and is discussed further in section 6.1 . for a system of characteristic linear size @xmath11 and internal velocity @xmath12 , a characteristic dynamical time is @xmath13 . a characteristic mass density is @xmath14 , from which one can estimate the total mass within the region occupied by the galaxies . after removing galaxies whose velocities differ from the group median by more that 1000 km s@xmath3 , hickson et al ( 1992 ) found that @xmath15 and obtained a mean mass - to - light ratio of @xmath16 ( solar units ) for the hcgs . similar values were found for several shakhbazian groups ( tikhonov 1986 , amirkhanian & egikian 1987 , amirkhanian 1989 , lynds et al 1990 , amirkhanian et al 1991 ) . since the mass - to - light ratios of individual hcg galaxies is on the order of @xmath17 ( rubin et al 1991 ) , the galaxies appear to contain only @xmath18% of the total mass . the spatial distribution and luminosities of the member galaxies provides further clues to the nature of compact groups . if they are primarily projections or transient configurations , the luminosity function should be the same as that of the parent systems , and the spatial distribution of the galaxies should be consistent with a random distribution . if they are bound physical systems , the luminosity function and spatial distribution might show features that reflect the origin or subsequent evolution of compact groups . the shapes of compact groups were first investigated by arp ( 1973 ) who concluded that galaxy `` chains '' were unusually predominant . however , rose ( 1977 ) determined that the ellipticities of his groups were consistent with a random distribution of galaxies . using the larger hcg sample , hickson et al ( 1984 ) and malykh & orlov ( 1986 ) reached the same conclusion as arp the groups are typically more elongated than would be a random distribution of galaxies . an immediate consequence of this result is that compact groups can not easily be explained as random projections or chance crossings , as this would largely erase any inherent ellipticity of a parent loose group . from static simulations , hickson et al ( 1984 ) concluded that the observed ellipticities are best matched by three - dimensional shapes that are intrinsically prolate . the same result was found by oleak et al ( 1995 ) in a recent study of the shapes of 95 shakhbazian compact groups . these conclusions , however , are not unique . hickson et al ( 1984 ) also found the shapes to be consistent with those seen in dynamical simulations of compact groups seen in projection as subgroups within loose groups . in addition , one must always be concerned about possible selection biases . it may be that highly elongated groups ( such as vv 172 ) are more easily noticed in visual searches . it will be interesting to see if these results are confirmed by studies of groups found by automated searches . if the intrinsic shapes of compact groups are related to their formation process , one might expect to see a relationship between the orientation angle of a group and the environment . palumbo et al ( 1993 ) examined the environments of the hcgs , and found that the orientations of the major axes of the groups were consistent with an isotropic distribution . if compact groups are not simply projection effects , they might be expected to show a centrally - concentrated surface density profile , as is seen in clusters of galaxies . although the number of galaxies in individual compact groups is small , with a large sample it is possible to estimate a mean profile . by scaling and superimposing the hcgs , hickson et al ( 1984 ) found evidence for central concentration . mendes de oliveira & girard ( 1994 ) , using a similar analysis , concluded that the mean surface density profile is consistent with a king ( 1962 ) model with typical core radius of 15@xmath19 kpc . most recently , montoya et al ( 1996 ) have analyzed the profiles of the 42 hcg quartets which have accordant redshifts . their technique uses the distribution of projected pair separations and thus avoids assumptions about the location of the group center . they find a smaller core radius ( @xmath20 kpc for a king model ) . the fact that montoya et al ( 1996 ) find a consistent density profile for all groups , without any scaling , is particularly interesting . this would not be expected if most groups are chance alignments within loose groups . it also implies that compact groups have a unique scale , which seems counter to the concept of hierarchical clustering . montoya et al ( 1996 ) suggest that this arises as a result of a minimum mass density and velocity dispersion that is required for the groups to be virialized ( mamon 1994 ) . there have been several studies of the morphological types of galaxies in compact groups ( hickson 1982 , williams & rood 1987 , sulentic 1987 , hickson et al 1988b ) . most studies agree that the fraction @xmath21 of late type galaxies is significantly less in compact groups than in the field . hickson et al ( 1988b ) find @xmath22 for the hcgs ; prandoni et al ( 1994 ) obtain @xmath23 for the scgs . both these values are substantially lower than those found for field galaxy samples ( @xmath24 , gisler 1980 , nilson 1973 ) . also well established is morphological type concordance , observed in both the hcgs and scgs ( sulentic 1987 , hickson at al 1988b , prandoni et al 1994 ) . a given compact group is more likely to contain galaxies of a similar type ( early or late ) than would be expected for a random distribution . white ( 1990 ) has pointed out that such concordance could result from a correlation of morphological type with some other property of the group . the strongest such correlation found to date is between morphological type and velocity dispersion ( hickson et al 1988b ) . as figure 3 shows , groups with higher velocity dispersions contain fewer late - type ( gas - rich ) galaxies . they also tend to be more luminous . the importance of velocity dispersion , in addition to local density , on the galaxy morphology had previously been emphasized by de souza et al ( 1982 ) . a crucial clue is that the morphology - density relation seen in clusters and loose groups ( dressler 1980 , postman & geller 1984 , whitmore & gilmore 1992 ) is not the dominant correlation in compact groups ( hickson et al 1988b ) , although some effect is present ( mamon 1986 ) . this suggests that the velocity dispersion is more fundamental , ie of greater physical relevance to the formation and evolution of galaxies in compact groups , than is apparent physical density . there is much evidence that interaction is occurring in a large fraction of galaxies in compact groups . the strongest direct support comes from kinematical studies . rubin et al ( 1991 ) found that two thirds of the 32 hcg spiral galaxies that they observed have peculiar rotation curves . these show asymmetry , irregularity and in some cases extreme distortion , characteristic of strong gravitational interaction . this result has recently been challenged by mendes de oliveira et al ( preprint ) who obtained h@xmath25 velocity maps for 26 hcg spiral galaxies and found that only 1/3 showed abnormal rotation curves . they suggest that the difference is due to the more - complete spatial sampling of their data . in their study , rubin et al ( 1991 ) observed 12 hcg elliptical galaxies and detected nuclear emission in 11 of them . this high fraction suggests that interactions and mergers may be supplying gas to these galaxies . this idea received independent support from radio observations in which neutral hydrogen emission was detected in three compact groups which contain only elliptical galaxies ( huchtmeier 1994 ) . zepf & whitmore ( 1993 ) found that elliptical galaxies in compact groups tend to have lower internal velocity dispersions than do ellipticals in other environments having the same effective radii , absolute magnitudes and colors . they therefore do not lie on the fundamental plane defined by other elliptical galaxies . this discrepancy correlates with isophote shape in that those galaxies that have `` disky '' or irregular isophotes tend to have lower velocity dispersion . both zepf & whitmore ( 1993 ) and bettoni & fasano ( 1993 , 1995 , 1996 , fasano & bettoni 1994 ) report that hcg elliptical galaxies are less likely to have `` boxy '' isophotes , and more likely to have irregular isophotes . such effects are consistent with results of simulations of tidal encounters ( balcells & quinn 1990 ) . the luminosity function of compact groups was first estimated by heiligman & turner ( 1980 ) . they examined a sample consisting of stephan s quintet , seyfert s sextet and eight more compact groups from the arp and vorontsov - velyaminov catalogs , and concluded that compact groups contain relatively fewer faint galaxies than does a comparable field galaxy sample . analysis of the relative luminosities within individual hcgs ( hickson 1982 ) , and studies of several shakhbazian groups ( kodaira et al 1991 ) , showed a similar effect , although tikhonov ( 1987 ) found a luminosity function similar to that of field and cluster galaxies . the larger hcg sample allows the question of the galaxy content of compact groups to be addressed with greater certainty . the standard technique for determination of the luminosity function weights each galaxy by @xmath26 , where @xmath12 is the volume of the smallest geocentric sphere containing the group , and @xmath27 is the volume of the largest such sphere within which the group could have been detected . using this approach sulentic and rabaa ( 1994 ) obtained a luminosity function for hcg galaxies similar to that of field galaxies . however , mendes de oliveira and hickson ( 1991 ) argued that the standard calculation does not address the selection effects of the hcg sample . for example , the luminosity range within an individual group is limited by the 3-mag range of the selection criteria . because of this , fainter galaxies within compact groups are not included in the catalog . in order to account for such biases they used a modeling technique in which galaxies were drawn from a trial luminosity function and assigned to groups . redshifts were given to each group according to the observed distribution , and groups that failed to meet the hcg selection criteria were rejected . the luminosity distribution of the resulting galaxy sample was then compared to the observations and the process repeated with different trial luminosity functions . their best - fit luminosity function is deficient in faint galaxies , although a normal field - galaxy luminosity function is not excluded . to avoid the selection problem , ribeiro et al ( 1994 ) obtained deeper photometry for a subsample of the hcgs in order to include the fainter galaxies explicitly . since redshifts are not known for these galaxies , a correction for background contamination was made statistically . the luminosity function that they obtained is similar to that of field galaxies . most recently , the luminosity function for the rscgs has been computed by barton et al ( 1996 ) . they find it to be mildly inconsistent with that of field - galaxies , in the same sense as that of mendes de oliveira & hickson ( 1991 ) for the hcgs . figure 4 summarizes these estimates of the luminosity function of compact galaxies in terms of the schechter ( 1976 ) parameters @xmath28 and @xmath29 . how do we interpret these apparently conflicting results ? prandoni et al ( 1994 ) have argued that the hcg catalog is biased toward groups with a small magnitude range @xmath30 , because the scgs have a larger fraction of high @xmath30 groups . however , it is not known what fraction of such groups are physically real , as few redshifts have yet been obtained . such a bias could affect the luminosity function of mendes de oliveira & hickson ( 1991 ) , particularly at the faint end , but it is not evident that the bias is sufficient to account for the apparent faint - galaxy deficit . on the other hand , the small sample used by ribeiro et al ( 1994 ) may not be representative of compact groups in general . hickson ( 1997 ) points out that the ribeiro et al ( 1994 ) sample has a spiral fraction of 0.60 , substantially higher than that of the whole hcg catalogue , and contains 7 of the 16 hcgs found to be in a high density environment by palumbo et al ( 1995 ) . this suggests that the their sample has more than the usual amount of field galaxy contamination . de carvalho et al ( 1994 ) , note that the faint galaxies form a more extended distribution than do the brighter galaxies . thus they may be a dynamically distinct component , or simply unrelated field galaxies . finally , the redshift - selected rscg sample also shows mild evidence for a faint galaxy deficiency . while the faint end of the lf in compact groups appears to be depleted , there is evidence that the bright end may be enhanced . limber and matthews ( 1960 ) were the first to remark that `` the members of stephan s quintet are to be classed among the brightest of galaxies '' . it is possible that this may be in part due to interaction - induced star formation , at least for the spiral galaxies . on the other hand mendes de oliveira & hickson ( 1991 ) compared their luminosity function of elliptical galaxies in compact groups with those in the virgo and coma clusters ( as reported by sandage et al 1985 and thompson & gregory 1980 ) and found that that the compact - group ellipticals have a luminosity enhancement of more than 1 mag compared to cluster ellipticals . sulentic and rabaa ( 1994 ) find a similar enhancement in their morphological - type - specific luminosity function . this suggests that compact group elliptical galaxies may have a unique formation mechanism . from the luminosity function , one can estimate the contribution of compact groups of galaxies @xmath31 to the total galaxian luminosity density @xmath32 . mendes de oliveira & hickson ( 1991 ) obtained a ratio of @xmath33 . applying the same analysis to the luminosity function of ribeiro et al ( 1994 ) gives a ratio of 3.3% . for the rscgs , the figure is comparable : for groups of four or more galaxies , barton et al ( 1996 ) obtain a compact group abundance of @xmath34 mpc@xmath3 which leads to a luminosity density ratio of approximately 3% . these are surprisingly high figures considering the short dynamical times of most compact groups . evidence is accumulating that tidal interactions play an important role in triggering starburst activity in galaxies ( eg . maccagni et al 1990 , campos - aguilar & moles 1991 , kormendy & sanders 1992 , sanders & mirabel 1996 ) . compact groups , with their high galaxy density and evident signs of galaxy interaction should be ideal systems in which to study such effects . many hcgs do in fact contain galaxies showing starbursts or harboring active galactic nuclei ( agn ) . for example , hcg 16 is found to contain a seyfert 2 galaxy , two liners , and three starburst galaxies ( ribeiro et al 1996 ) . hcg 31 contains five galaxies showing signs of recent starburst activity ( rubin et al 1990 , iglesias - pramo & vlchez preprint ) . seyfert galaxies are also found in hcg 77 , 92 , 93 and 96 . the general degree of star formation activity in compact group galaxies can be determined from infrared observations . to date , studies have been based primarily on data from the iras satellite . hickson et al ( 1989 ) found sources in 40 hcg from a search of the point source catalog . they concluded that the ratio of far - infrared - to - optical luminosity is greater by about a factor of two in compact group galaxies , compared to that of isolated galaxies . this result was disputed by sulentic and de mello rabaa ( 1993 ) who argued that the low spatial resolution of the data made the assignment of infrared flux to individual galaxies ambiguous . they concluded that redistribution of the flux could result in little or no infrared enhancement , a conclusion echoed by venugopal ( 1995 ) . however , in cases of doubt , hickson et al ( 1989 ) identified the infrared galaxy on the basis of radio emission . the well - known correlation between infrared and radio continuum emission makes it unlikely that the results are much in error . analysis of improved data ( eg . allam et al 1996 ) should soon resolve questions about the identifications and infrared fluxes . the resolution problem can be avoided by considering the infrared colors of the sources instead of the infrared / optical ratio . zepf ( 1993 ) compared the ratio of 60 @xmath35 m to 100 @xmath35 m fluxes of compact group galaxies with those of isolated galaxies and also with those of galaxies believed to be currently merging . he found that the compact group sample was significantly different from both other samples , and estimated that about 1/3 of the compact group galaxies had warm colors ( larger 60/100 @xmath35 m ratios ) similar to those of merging galaxies . another approach to interpreting the infrared results was taken by menon ( 1991 ) who emphasized that the strong correlation between radio and infrared radiation indicates that these likely originate from a common region . in compact group spirals the radio emission is primarily nuclear whereas in isolated spirals it originates in the disk . if this is also true for the infrared flux , there must be an enhancement of the infrared / optical ratio , in the nuclear region , of more than an order of magnitude . this idea is supported by recent millimeter - wavelength observations ( menon et al 1996 ) in which co emission is detected in 55 of 70 iras - selected hcg galaxies . the inferred ratio of infrared luminosity - to - h@xmath36 mass showed an enhancement which correlates with the projected nearest - neighbor distance . further clues are provided by radio continuum studies . nonthermal emission from spiral galaxies can arise from both disk and nuclear sources . disk emission is predominantly due to supernova remnants and is thus related to the star formation rate . nuclear emission can arise both from star formation and from an active nucleus . menon ( 1995 ) observed 133 spiral galaxies in 68 hcg , and found that overall they typically show less continuum emission than those in isolated environments , which is consistent with the neutral hydrogen observations . however , when considering the nuclear regions alone , the radio emission is found to be an order of magnitude higher compared to isolated spirals . the implication is that star formation and/or agn activity is substantially enhanced in the nuclear regions of many compact group spiral galaxies . this is generally consistent with a picture in which galaxy interactions remove gas from the outer regions of galaxies , while simultaneously allowing gas to flow inwards toward the nucleus , resulting in enhanced star formation in the nuclear region , and possibly fueling an active nucleus . although there is a clear example of tidal interaction stimulating disk radio emission in at least one compact group ( menon 1995a ) , statistical evidence for a link between interactions and radio emission in compact groups is only now accumulating . if interactions are stimulating nuclear radio emission , one would expect the radio luminosity to be correlated with some index describing the degree of interaction such as the projected distance to the nearest neighbor . evidence in support of this was found by vettolani & gregorini ( 1988 ) who observed that early - type galaxies which have a high ratio of radio - to - optical emission show an excess of nearby neighbors . a similar effect was observed by malumian ( 1996 ) for spiral galaxies in groups . examining compact group galaxies , menon ( 1992 ) found that elliptical and s0 galaxies detected at a wavelength of 20 cm had closer neighbors than the undetected galaxies . the effect was not found for spiral galaxies , but if one considers only the detected galaxies , there is a significant correlation between radio - to - optical luminosity and nearest neighbor distance for both early and late type galaxies ( tk menon , private communication ) . continuum radio emission has also been detected in a number of hcg elliptical galaxies . unlike those found in cluster ellipticals , the radio sources are low - luminosity and compact . where spectral indices are available , they indicate that the radio emission arises from an agn rather than from starburst activity ( tk menon , private communication ) . in the hcg sample , there is a significant preference for radio - loud elliptical galaxies to be first - ranked optically ( menon & hickson 1985 , menon 1992 ) . the probability of radio emission does not correlate with absolute luminosity , but instead correlates with _ relative _ luminosity within a group . spiral hcg galaxies do not show this effect . although the tendency of radio galaxies in rich clusters to be first - ranked has been known for many years , it is surprising to find a similar effect in small groups , where the number of galaxies and luminosity range is small , the gravitational potential well is much less clearly defined , and it is unlikely that any individual galaxy holds a central location . the effect of optical rank on radio emission had been previously noted in other small groups by tovmasian et al ( 1980 ) although these authors made no distinction between elliptical and spiral galaxies . it is difficult to imagine any explanation for this result in which the compact group is not a true physical system . it would appear that , regardless of absolute luminosity , only the first - ranked ( presumably the most massive in the group ) elliptical galaxy can develop a radio source . stars stripped from galaxies by tidal forces should accumulate in the potential well of the group . and may be detectable as diffuse light . in an early photographic study , rose ( 1979 ) found no evidence for diffuse light in his groups , and was lead to the conclusion that most of his groups must be transient configurations . however , bergvall et al ( 1981 ) were successful in detecting ionized gas and a common halo around a compact quartet of interacting early - type galaxies , and evidence for a common halo in vv 172 was reported by sulentic & lorre ( 1983 ) . diffuse light can clearly be seen in hcg 94 , and has been found in hcg 55 ( sulentic 1987 ) , but pildis et al ( 1995b ) did not detect any in seven other compact groups . analysis by mamon ( 1986 ) indicated that while the expected diffuse light should be detectable with modern techniques , it would generally be very faint . estimates of the total amount of diffuse light in the detected groups are rather uncertain as they depend sensitively on subtraction of the galactic light , and the sky background . deeper photometry and improved image processing techniques may yet reveal diffuse light in other compact groups ( sulentic 1997 ) . the mass and distribution of cool galactic and intergalactic gas , can be obtained from observations of the 21-cm line of neutral hydrogen . the first such study of a large sample of compact groups is that of williams & rood ( 1987 ) who found a median hi mass of @xmath37 . they concluded that compact groups are typically deficient in neutral hydrogen by about a factor of two compared to loose groups . this effect is consistent with similar deficit in continuum radio emission seen in the disks of compact group spiral galaxies ( menon 1995 ) , and suggests that interactions in compact groups has removed much of the gas from the galaxies . simulations suggest that in addition to an outflow of gas , inflow also occurs which may fuel nuclear star - formation , as suggested by the strongly - enhanced radio emission seen in the nuclear regions of compact group spiral galaxies ( menon 1995 ) . high - resolution studies of individual groups ( williams & van gorkom 1988 , williams et al 1991 ) showed clearly that the gas is not confined to the galaxies . in two of three groups studied , the radio emission originates from a common envelope surrounding the group and in the third group there are signs of tidal distortion . these results strongly indicate that at least these compact groups are physically dense systems and not chance alignments or transient configurations in loose groups . they also show that many groups have evolved to the point that gas contained within individual galaxies has been distributed throughout the group . in contrast to the hi results , initial co - line observations of 15 compact - group galaxies ( boselli et al 1996 ) indicated a normalized molecular gas content similar to that of isolated spiral galaxies . however , this result is based on normalizing the flux by the optical area of the galaxy , rather than by the infrared luminosity , and may be biased by the relatively small sizes of compact - group galaxies . further co studies , currently in progress , should soon settle this question . x - ray observations of hot gas in clusters of galaxies can reveal the amount , distribution , temperature and metallicity of the gas , as well as the relative amount and distribution of the total gravitating mass . temperature , metallicity ( fraction of solar abundance ) , and bolometric luminosities are estimated by fitting a spectral model , such as that of raymond & smith ( 1977 ) to the data . x - ray emission from compact `` poor clusters '' was first reported by schwartz et al ( 1980 ) , who concluded that their x - ray properties were similar to those of rich clusters . using the einstein observatory , bahcall et al ( 1984 ) first detected x - ray emission from stephan s quintet . the x - ray map revealed that the emission is diffuse and not centered on individual galaxies . however , we now know that most of this emission is associated with a shock front rather than gas trapped in the group potential well ( sulentic et al 1995 ) . although several other groups ( bahcall et al 1984 , biermann & kronberg 1984 ) were detected by the einstein observatory , further progress required the improved sensitivity of the rosat x - ray observatory . pointed rosat observations revealed massive hydrogen envelopes surrounding the ngc 2300 group , a bright elliptical - spiral pair with two fainter members ( mulchaey et al 1993 ) , and hcg 62 , a compact quartet of early - type galaxies ( ponman & bertram 1993 ) , and showed that these systems are dominated by dark matter . subsequent investigations detected x - rays from 18 additional compact groups , either from individual galaxies , or from diffuse gas ( ebeling et al 1994 , pildis et al 1995a , sarraco & ciliegi 1995 , sulentic et al 1995 ) . these studies showed that the physical properties of individual systems span a wide range , but that the ratio of gas - to - stellar mass is significantly lower than in rich clusters . moreover , the detected compact groups all contained a majority of early - type galaxies . no spiral - rich groups were detected ( although mulchaey et al 1996b pointed out that they might be found from qso absorption spectra ) . this result is consistent with the fact that x - ray - selected groups ( henry et al 1995 ) and loose groups ( mulchaey et al 1996a ) tend to be spiral poor , and led to the suggestion that spiral - rich compact groups might not be physically dense systems at all . the most extensive x - ray study of compact groups to date is that of ponman et al ( 1996 ) . these authors combined pointed and survey - mode observations of a complete sample of 85 hcgs and detected diffuse emission in 22 groups . they conclude that , when the detection limits are considered , diffuse emission is present in at least 75% of the systems . significantly , they detected diffuse emission in several spiral - rich groups . in these the surface brightness is lower and the x - ray emission has a lower characteristic temperature , as would be expected given the lower velocity dispersions of spiral - rich compact groups . the diffuse x - ray luminosity was found to correlate with temperature , velocity dispersion , and spiral fraction , but not with optical luminosity . the last result suggests that the gas is mostly primordial and not derived from the galaxies . the correlations with temperature and velocity dispersion appear to be consistent with a single relation for clusters and groups ( figure 5 ) . the total mass in compact groups typically exceeds the stellar and gas mass by an order of magnitude . pildis et al ( 1995 ) derived baryon fractions of 12 - 19% . davis et al ( 1996 ) obtains 10 - 16% for the ngc 2300 group . these are comparable to the fractions found for poor clusters ( dellantonio et al ( 1995 ) and are about half the typical values found for rich clusters . however , the derived baryon fraction depends sensitively on the radius within which it is measured and on the assumed background level ( henriksen & mamon 1994 ) . both the total mass , which is dominated by dark matter , and the gas mass continue to increase with radius . consequently , both the baryon fraction and the gas fraction are poorly determined . the contribution of compact groups to the x - ray luminosity function has been estimated by pbeb , who find that on the order of 4% of the total luminosity in the range @xmath38 erg s@xmath3 comes from hcgs . this is higher than the contribution of hcg galaxies to the local optical luminosity density estimated at 0.8% by mendes de oliveira & hickson ( 1991 ) , but it is comparable to the value found by ribeiro et al ( 1994 ) . the metallicity inferred for the x - ray - emitting gas in compact groups is relatively low . pbeb obtain a mean metallicity of 0.18 solar , compared to the value , 0.3 - 0.4 solar , found in rich clusters . this is comparable to the low value ( @xmath39 solar ) found for the ngc 2300 group ( davis et al 1996 ) . these figures suggest that the gas is largely primordial , a result supported by the absence of a correlation between x - ray and optical luminosity . however , given the limited spectral resolution of rosat , these low metalicities can not yet be considered secure . the higher spectral resolution and sensitivity of the asca satellite should provide more definitive results . recent observations of hcg 51 and the ngc 5044 group found metal abundances comparable to those of clusters ( fukazawa et al 1996 ) . the nature of the discordant redshift members of compact groups has been a subject of debate for many years ( eg . burbidge & burbidge 1961a , burbidge & sargent 1971 , nottale & moles 1978 , sulentic 1983 ) . if the frequency of discordant galaxies is inconsistent with the statistics of chance projection , it might signify the need for new physical theories ( arp 1987 ) , or for gravitational amplification of background galaxies ( hammer & nottale 1986 ) . initial estimates of the chance probability of finding discordant galaxies in groups like stephan s quintet , seyfert s sextet and vv 172 , were very small ( burbidge & sargent 1971 ) . however , such probabilities were recognized to be difficult to determine reliably , because the a priori probability of _ any _ particular configuration of galaxies is also very small ( burbidge & sargent 1971 ) . only with a well - defined sample of groups and a complete characterization of selection effects , can meaningful estimates of the probabilities be made . the explicit selection criteria of the hcg catalog in principal makes this sample suitable for a quantitative statistical investigation of the discordant - redshift question . sulentic ( 1987 ) first concluded that the number of discordant redshifts in the catalog is too large to explain by chance . on the other hand , hickson et al 1988a and mendes de oliveira ( 1995 ) , applying the selection criteria more rigorously , found no strong statistical evidence for this . their result , however , may be biased by incompleteness in the hcg catalog : there seem to be too - few low surface brightness groups in the catalog , and the `` missing '' groups may have a higher fraction of discordant redshifts ( sulentic 1997 ) . in order to address the incompleteness issue , iovino & hickson ( 1996 ) combined observational results from both the hcg and scg catalogs with monte - carlo simulations . their technique exploits the unbiased nature of the scg catalog and the complete redshift coverage of the hcg sample . they conclude that for all except the two highest - surface - brightness quintets ( stephan s quintet and seyfert s sextet ) , the number of discordant redshifts is consistent with chance projections . for these two , the chance probabilities are low . however , for both of these systems there is independent physical evidence that the discordant galaxies are at the cosmological distances that correspond to their redshifts and are therefore not group members ( kent 1981 , wu et al 1994 ) . one should not assume that the situation is now completely settled . further studies will be possible when redshifts have been obtained for the scg galaxies . there are still other questions that have not been adequately addressed , such as reported redshift quantization ( cocke & tifft 1983 ) . however , at this point it appears that the frequency of discordant galaxies does not require a new interpretation of galaxy redshifts . in fact , physical evidence suggests the opposite . the discordant galaxies all have physical properties consistent with a cosmological distance . for example those with higher redshift tend to be smaller and fainter than other members of the group , and vice versa ( mendes de oliveira 1995 ) . because we can measure only three phase - space dimensions for galaxies in compact groups ( two components of position and one of velocity ) , the groups are subject to projection effects . because of this , they may not be physically dense , or even physically related systems . the following interpretations have so far been suggested for compact groups : 1 . transient dense configurations ( rose 1977 ) 2 . isolated bound dense configurations ( sulentic 1987 , hickson & rood 1988 ) 3 . chance alignments in loose groups ( mamon 1986 , walke & mamon 1989 , mamon 1995 ) 4 . filaments seen end - on ( hernquist et al 1995 ) 5 . bound dense configurations within loose groups ( diaferio et al 1994 , governato et al 1996 ) evidence for and against physical association and high density in the hcg sample , to 1988 , was summarized by hickson & rood ( 1988 ) , and by walke and mamon ( 1989 ) respectively . since that time , several new results have emerged . from an analysis of optical images , mendes de oliveira & hickson ( 1994 ) concluded that 43% of all hcg galaxies show morphological features indicative of interaction and/or merging , and that 32% of all hcgs contain three or more interacting galaxies . these percentages are likely to rise with more - detailed studies and sophisticated image analysis ( longo et al 1994 ) . this high frequency of interactions observed in compact groups is difficult to reconcile with the chance alignment and filament hypotheses , even if the alignments contain physical binaries ( mamon 1995 ) . the high fraction of hcgs showing diffuse x - ray emission is very strong evidence that a large fraction of these systems are physically dense , and are not transient configurations or projection effects . although the exact numbers are not final , due to the faintness of the sources and the problems of contamination by sources associated with the individual galaxies , it seems evident that many groups are dense bound systems . the correlations seen between x - ray and optical properties , and the fact that the x - ray properties of compact groups are not inconsistent with those of clusters reinforces this conclusion . ostriker et al ( 1995 ) have argued that the relatively low x - ray luminosities of compact groups might not be due to a low gas fraction , but instead could be understood if the groups are filaments seen in projection ( hernquist et al 1995 ) . however , ponman et al ( 1996 ) point out that in order to explain even the fainter compact groups , gas temperatures @xmath40 kev and densities @xmath41 @xmath42 would be required . these appear to be ruled out by both observations ( briel & henry 1995 ) and simulations ( diaferio et al 1995 , pildis et al 1996 ) . even if compact groups are physically dense , they may not be as dense as they appear . as mentioned in section 2 , a sample of groups selected on the basis of high apparent density will be biased by the inclusion of looser systems which appear more compact due to geometrical or kinematic effects . is this bias large ? its magnitude can be estimated as follows : consider @xmath43 galaxies randomly located within a circle of radius @xmath11 on the sky . what is the probability @xmath44 that they will fall within some circular subarea of radius @xmath45 ? the answer can be obtained using analytic expressions derived by walke & mamon ( 1989 ) . from their equations 1 and 6 ( setting @xmath46 and @xmath47 ) we obtain f(x , n ) = n!^n _ 0^x dr _ 0 ^ 1-r d where dn = 2^n-1 n r^2n-3(n-2 ) ! 2ddr is the number of possible configurations with radius between @xmath48 and @xmath49 and distance from the center between @xmath50 and @xmath51 . here we have neglected a small edge contribution that is unimportant for small values of @xmath52 ( walke and mamon s case 3 ) . this gives f(x , n ) = n^2 x^2n-2 now , an observer would infer a galaxy space density that is higher by a factor @xmath53 , so the average apparent space density enhancement is < > & = & _ 0 ^ 1 dx + & = & _ 0 ^ 1 n^2 x^2n-5 dx + & = & 2n^2(n-1)(2n-5)(2n-3)(n-2 ) thus we expect to typically overestimate the space density by about a factor of 12.0 for triplets ( or quartets containing a physical binary ) , 3.2 for true quartets and 2.0 for quintets . a key question that remains is the position of compact groups in the clustering hierarchy . are compact groups distinct entities ( sulentic 1987 ) or an intermediate stage between loose groups and triplets , pairs and individual galaxies ( barnes 1989 , white 1990 , cavaliere et al 1991 , rampazzo & sulentic 1992 , diaferio et al 1994 ) . some compact groups are purely projection effects , others may be small clusters ( ebeling et al 1995 ) , but most appear to be real . it seems that they can arise naturally from subcondensations in looser groups , but further studies are needed to better determine both the observed space density of groups as a function of population and the timescales involved in the evolutionary process . this question is related to that of the formation mechanism of compact groups . two mechanisms have been discussed in the literature . diaferio et al ( 1994 ) conclude that compact groups form continually from bound subsystems within loose groups . this gains some support from the observation ( see section 3 ) that most hcgs are embedded in loose groups , although it is not obvious that these loose groups are sufficiently rich ( sulentic 1997 ) . governato et al ( 1996 ) , proposed a model in which merging activity in compact groups is accompanied by infall of galaxies from the environment . this naturally explains the observed mix of morphological types , and it allows compact groups to persist for longer times . where do the shakhbazian groups fit in this picture ? recent studies ( tikhonov 1986 , amirkhanian & egikian 1987 , amirkhanian et al 1988 , 1991 , amirkhanian 1989 , kodaira et al 1988 - 91 , stoll et al 1993 - 1996 ) show that these objects are typically compact clusters or groups of early - type galaxies . although the systems were selected on the basis of red colors and compact appearance of their galaxies , both of these factors result from their large distances because k - corrections and contrast effects become significant . the galaxies are in fact relatively normal , although luminous ( del olmo et al 1995 ) . however , the number of blue ( gas rich ) galaxies in these systems does seem to be very small . thus it appears the shakhbazian groups are mostly small clusters , possibly intermediate in physical properties between classical compact groups and clusters . if the groups are dynamically bound , galaxy mergers should commence within a few dynamical times ( carnevali et al 1981 , ishizawa et al 1983 , barnes 1985 , ishizawa 1986 , mamon 1987 , 1990 , zheng et al 1993 ) . both n - body and hydrodynamic simulations indicate that the dark matter halos of individual galaxies merge first , creating a massive envelope within which the visible galaxies move ( barnes 1984 , bode et al 1993 ) . kinematic studies of loose groups ( eg . puche & carignan 1991 ) indicate that the dark matter is concentrated around the individual optical galaxies . in contrast , the x - ray observations indicate that in most compact groups , the gas and dark matter is more extended and is decoupled from the galaxies . this may explain the observation that galaxies in compact groups typically have mass - to - light ratios 30% to 50% lower than more isolated galaxies ( rubin et al 1991 ) . is there any observational evidence that galaxies in compact groups are merging ? by 1982 it was evident that first - ranked galaxies in compact groups did not appear to be merger products , because the fraction of first - ranked galaxies that are type e or s0 is the same as for the general population of hcg galaxies ( hickson 1982 ) . if mergers were a dominant effect , the first - ranked galaxies would be expected to be more often elliptical . the same conclusion was reached by geller & postman ( 1983 ) who found that the luminosities of first - ranked galaxies were consistent with a single luminosity distribution for all group galaxies . of course this may just mean that in small groups the first - ranked galaxy is not necessarily the most evolved . rather , one should ask if _ any _ galaxies in compact groups show indications of merging . the relative paucity of merging galaxies in compact groups was first noted by tikhonov ( 1987 ) , from a visual inspection of optical images . zepf & whitmore ( 1991 ) realized that elliptical galaxies formed by recent mergers of gas - rich systems should have bluer colors than normal . examining the hcgs , they found only a small enhancement in the fraction of early - type galaxies having blue colors , a conclusion reinforced by an independent study by moles et al ( 1994 ) . on the other hand , caon et al ( 1994 ) argued that the large effective radii of compact group elliptical galaxies is indicative of an origin by merging or accretion of companions . zepf ( 1993 ) estimated that roughly 7% of the galaxies in compact groups are in the process of merging . this conclusion was based on roughly consistent frequencies of _ ( a ) _ optical signatures of merging , _ ( b ) _ warm far - infrared colors , and _ ( c ) _ sinusoidal rotation curves . however , few galaxies show all of these effects simultaneously . the merging fraction may thus be as high as 25% if one allows that any one of these criteria would be considered to be sufficient to indicate a merger ( hickson 1997 ) . given the small numbers of objects in these studies , it is fair to say that the fraction of merging galaxies is highly uncertain at present . it seems safe to conclude that current observations do not rule out a significant amount of merging in compact groups . detailed studies of individual compact groups can be quite revealing . many galaxies that at first appear normal are revealed to have peculiar morphology or spectra when examined more closely . many , perhaps most , compact groups clearly contain galaxies that are dynamically interacting . however , the groups likely span a range of evolutionary states . at the extreme end are high - density groups like seyfert s sextet , hcg 31 , hcg 62 , hcg 94 ( pildis 1995 ) and hcg 95 ( rodrigue et al 1995 ) in which we find strong gravitational interactions . at the other end are lower density compact groups , such as hcg 44 , which most likely are in a less - advanced stage of evolution . this picture is supported by radio observations : seyfert s sextet and hcg 31 are both embedded in extended hi clouds whereas in hcg 44 the hi is associated with individual galaxies ( williams et al 1991 ) . it seems clear that the groups as we now see them can persist for only a fraction of a hubble time . simulations indicate that merging should destroy the group on a time scale @xmath54 that is typically an order of magnitude larger than @xmath55 , depending on the distribution of dark matter ( cavaliere et al 1983 , barnes 1984 , navarro et al 1987 , kodaira et al 1990 ) and initial conditions ( governato et al 1991 ) . assuming that the groups are in fact bound dynamical systems , we can draw two conclusions : _ ( a ) _ there must be an ongoing mechanism for forming or replacing compact groups , and _ ( b ) _ there must be a significant population of relics of merged groups . what are the end - products of compact groups ? it is tempting to identify them with field elliptical galaxies , following a suggestion first made by toomre ( 1977 ) . simulations ( weil & hernquist 1994 ) indicate that multiple mergers in small groups of galaxies best reproduce the observed kinematical properties of elliptical galaxies . the resulting galaxies are predicted to possess small kinematic misalignments , which can be detected by detailed spectroscopic and photometric studies . neverthess , it remains to be demonstrated that these merger remnants can reproduce the tight correlation between size , luminosity and velocity dispersion found in present - day elliptical galaxies . if compact groups have lifetimes on the order of @xmath54 , and form continuously , then the number of relics , per observed group , is expected to be on the order of @xmath56 . thus , the number of relics could exceed that of present day groups by as much as an order of magnitude . mamon ( 1986 ) estimated that , if all hcgs are real , then the relics would account for about 25% of luminous field elliptical galaxies . as we have seen , the true space density of compact groups is uncertain by at least a factor of two , and may be underestimated because of selection biases . there is then the potential problem of producing too many relics . a second problem is the fact that the integrated luminosities of compact groups are typically a factor of three to four times greater than luminosities of isolated elliptical galaxies ( sulentic and rabaa 1994 ) . it is possible that interaction - induced star formation has boosted the luminosities of some compact - group galaxies , and that some degree of fading of the merger product is expected . however , at this point it is not clear whether or not the relics can be identified with isolated elliptical galaxies . despite these problems , a fossil compact group may have actually been found . ponman et al ( 1995 ) have detected a luminous isolated elliptical galaxy surrounded by diffuse x - ray emission which is consistent with the expected end - product of a compact group . if more objects like this are found , it may be possible to compare their space density with that expected for compact group relics . interactions are often implicated in the development of active nuclei in galaxies ( eg . freudling & almudena prieto 1996 ) . the hcg catalog includes several examples of compact groups containing both starburst galaxies and agn . several recent examples of associations between starburst galaxies or agn and what appear to be compact groups have been reported : del olmo & moles ( 1991 ) have found a broad - line agn in shakhbazian 278 ; zou et al ( 1995 ) find that the luminous infrared source iras 23532 coincides with a compact group that includes a seyfert 1 as well as a starburst galaxy . if this association extends to qsos , one would expect to find numerous compact groups at redshifts @xmath57 , where the comoving number density of qsos peaks ( eg . hartwick & schade 1984 ) . the tendency for qsos to have close companions has been known for some time ( eg . stockton 1982 , bahcall et al 1997 ) . recently , several examples of compact groups associated with luminous infrared galaxies , agn and qsos at @xmath58 have been found using hst ( pascarelle et al 1996 , francis et al 1996 , matthews et al 1994 , tsuboi & nakai 1994 , hutchings 1995 , hutchings et al 1995 ) . these observations provide support to the idea that tidally - triggered star formation is a predominant factor in the galaxy formation process ( lacey & silk 1991 , lacey et al 1993 ) . in this model disk star formation occurs relatively late , after the compact group has formed and tidal interactions are strong . this seems at least qualitatively consistent with the fragmentary nature of high - redshift galaxies observed with the hubble space telescope ( schade et al 1995 ) , although these fragments appear to be much less luminous and more irregular than most present - day compact group galaxies . the model also offers a possible explanation for the excess numbers of faint blue galaxies found in field galaxy count as dwarf galaxies undergoing star formation at a redshift of @xmath59 . compact groups may possibly play a role in the formation of other systems . we have seen that giant galaxies may be formed as the end product of compact - group evolution . at the other end of the scale , dwarf galaxies have physical properties distinct from normal galaxies , which suggests a unique formation mechanism . one possibility is that they form during gravitational interactions from tidal debris ( duc & mirabel 1994 ) . if this is the case , one would expect to find evidence for this in compact groups of galaxies . from an examination of condensations in tidal tails , hunsberger et al ( 1996 ) concluded that the fraction of dwarf galaxies produced within tidal debris in compact groups is not negligible . there is also evidence that star clusters form from tidal debris . longo et al ( 1995 ) have found an excess population of unresolved blue objects around hcg 90 which appear to be recently formed star clusters . these may be similar to the population of new star clusters recently reported in the merger remnant ngc 7252 ( whitmore et al 1993 ) . because compact groups have a high galaxy surface density , they may form effective gravitational lenses . gravitational amplification of background field galaxies was proposed by hammer and nottale ( 1986 ) as a possible explanation for the presence of the high - redshift discordant member of this group . mendes de oliveira and giraud ( 1994 ) and montoya et al ( 1996 ) find that most hcgs are too nearby to produce strong lensing effects . however , because the critical mass density required for strong lensing depends reciprocally on distance , analogous systems 5 - 10 times more distant should produce a non - negligible fraction of giant arcs . studies of small groups may provide clues to the overall structure of the universe . the baryon fractions found in clusters of galaxies appear to be inconsistent with a density parameter @xmath60 , unless the dark matter is more prevalent outside clusters ( white 1992 , babul & katz 1993 ) . compact groups provide a means to study dark matter in such regions . the baryon fractions found for compact groups do appear to be lower than those for clusters . david et al ( 1995 ) argued that the gas is the most extended component ; galaxies being the most compact and the dark matter being intermediate . they concluded that the baryon fraction approaches 30% on large enough scales , which is comparable to the values found for clusters . given the constraints of standard big - bang nucleosynthesis this would imply that the density parameter @xmath61 is at most 0.2 . on the other hand , the infall picture of compact - group evolution ( governato et al 1996 ) requires a high - density @xmath62 universe . in a low - density universe the infall rate is insufficient . as there is at present no other clear mechanism for avoiding the overproduction of relics by merging compact groups , this may be a strong argument for a high - density universe . during the last two decades we have seen a resurgence of interest in compact groups . while initially little more than a curiosity , these systems are now viewed as potentially important sites of dynamical evolution , shaping the structure of many galaxies . it now seems clear that while many compact groups are contaminated by projections , a large fraction of at least the high - surface - brightness hcgs are physically dense . they form by gravitational relaxation processes within looser associations of galaxies . the densest are generally in an advanced stage of evolution characterized by strong interactions , starburst and agn activity , stripping of stellar and dark matter halos , and merging . they contain large amounts of dark matter and primordial x - ray - emitting gas trapped within the gravitational potential well . despite this progress , many questions remain unanswered . what are the end products of compact group evolution , and do they have properties consistent with any know population of objects ? what is the space density of such relics ? where do compact groups fit in the overall clustering hierarchy ? what is their role in the evolution of galaxies both past and present ? given the current interest and research activity in this area , it is likely that many of these questions may soon be addressed . it is a pleasure to thank the observatories of brera and capodimonte for hospitality during the initial work on this review . i have benefitted from discussions with many individuals , but i would like to acknowledge particularly the contributions of a iovino , e kindl , g longo , g mamon , c mendes de oliveira , tk menon , g palumbo , h rood , and j sulentic . i thank g mamon , a sandage and j sulentic for providing helpful comments on an earlier version of the manuscript . financial support was provided by the natural sciences and engineering research council of canada and nato . [ a]llam s , assendorp r , longo g , braun m , richter g. 1996 . _ _ 117:39 - 82 [ a]mbartsumian ba 66:536 - 40 [ a]mirkhanian as . 1989 . ssr bjurak . 61:25 - 28 [ a]mirkhanian as , egikian ag . 1987 . _ astrofizika _ 27:395 - 97 [ a]mirkhanian as , egikian ag , silchenko ok . _ pisma astron . zh . _ 14:404 - 08 [ a]mirkhanian as , eghikian ag , tikhonov na , shakhbazian rk . _ astrofizika _ 35:67 - 75 [ a]rp h. 1966 . _ ap . j. suppl . _ 14:1 - 20 [ a]rp h. 1973 . _ 185:797 - 808 [ a]rp h. 1987 . _ quasars , redshifts and controversies _ , berkeley : interstellar media [ b]abul a , katz n. 1993 . _ ap . j. lett . _ 406:l51-l54 [ b]ahcall jn , kirharos s , saxe dh , schneider dp . 1997 . _ ap . j. _ 479 , in press [ b]ahcall na , harris de , rood hj . _ 284:l29-l33 [ b]aier fw , petrosian , mb , tiersch h , shakhbazian rk . _ astrofizika _ 10:327 - 35 [ b]aier fw , tiersch h. 1975 . _ astrofizika _ 11:221 - 28 [ b]aier fw , tiersch h. 1976 . _ astrofizika _ 12:7 - 12 [ b]aier fw , tiersch h. 1976 . _ astrofizika _ 12:409 - 15 [ b]aier fw , tiersch h. 1978 . _ astrofizika _ 14:279 - 82 [ b]aier fw , tiersch h. 1979 . _ astrofizika _ 15:33 - 35 [ b]alcells n , quinn pj . j. _ 361:381 - 93 [ b]arnes j. 1984 . _ mnras _ 208:873 - 85 [ b]arnes je . _ mnras _ 215:517 - 36 [ b]arnes je . 1989 . _ nature _ 338:123 - 26 [ b]arton e , geller mj , ramella m , marzke ro , da costa ln . 112:871 - 86 [ b]ergvall n , ekman a , lauberts a. 1981 . _ astron . _ 95:266 - 77 [ b]ettoni d , fasano g. 1993 . 105:1291 - 307 [ b]ettoni d , fasano g. 1995 . _ astron . j. _ 109:32 - 55 [ b]ettoni d , fasano g. 1996 . _ _ 118:429 - 39 [ b]iermann p , kronberg pp . 1984 . _ physica scr . _ t7:169 [ b]ode pw , cohn hn , lugger pm . 1993 . _ ap . j. _ 416:17 - 25 [ b]oselli a , mendes de oliveira c , balkowski c , cayatte v , casoli f. 1996 . _ 314:738 - 44 [ b]riel ug , henry jp . 1995 . astrophys . _ 302:l9-l12 [ b]urbidge em . , burbidge gr . 130:23 - 25 [ b]urbidge em . , burbidge gr 66:541 - 50 [ b]urbidge em . , burbidge gr . 1961b . _ ap . j. _ 134:244 - 47 [ b]urbidge em . , burbidge gr . 1961c . _ ap . j. _ 134:248 - 56 [ b]urbidge em , sargent wlw . 1971 . 35:351 - 78 [ c]ampos - aguilar a , moles m. 1991 . _ 241:358 - 64 [ c]aon n , capaccioli m , donofrio m , longo g. 1994 . astrophys . _ 286:l39-l42 [ c]arnevali p , cavaliere a , santangelo p. 1981 . _ ap . j. _ 249:449 - 61 [ c]avaliere a , colafrancesco s , scaramella , r. 1991 . _ ap . j. _ 380:15 - 23 [ c]avaliere a , santangelo p , tarquini g , vittorio n. 1983 . in _ clustering in the universe , d gerbal , a mazure , pp . gif - sur - yvette : ed . front . [ c]ocke wj , tifft wg . _ ap . j. _ 268:56 - 59 [ d]a costa ln , geller mj , pellegrini ps , latham dw , fairall ap , marzke ro , willmer cna , huchra jp , calderon jh , ramella m , kurtz mj , _ ap . j. lett . _ 424:l1-l4 [ d]avid lp , jones c , forman w. 1995 . _ ap . j. _ 445:578 - 90 [ d]avis ds , mulchaey js , mushotzky rf , burstein d. 1996 . _ ap . j. _ 460:601 - 11 [ d]e carvalho rr , djorgovski sg 1995 . soc . _ 187:53.02 [ d]e carvalho rr , ribeiro alb , zepf se . _ 93:47 - 63 [ d]ellantonio ip , geller mj , fabricant dg . 1995 . 110:502 - 12 [ d]el olmo a , moles m. 1991 . _ 245:27 - 30 [ d]el olmo a , moles m , perera j. 1995 . in _ groups of galaxies _ , asp conference series , 70:117 - 126 [ d]e souza re , capelato hv , arakaki l , logullo c. 1982 . _ ap . j. _ 263:557 - 63 [ d]iaferio a , geller mj , ramella m. 1994 . _ astron . j. _ 107:868 - 79 [ d]iaferio a , geller mj , ramella m. 1995 . 109:2293 - 304 [ d]ressler a. 1980 . _ ap . j. _ 236:351 - 65 [ d]uc p - a , mirabel if . 1994 . _ , 289:83 - 93 [ e]beling h , mendes de oliveira c , white da . 1995 . _ mnras _ 277:1006 - 32 [ e]beling h , voges w , boehringer h. 1994 . _ ap . j. _ 436:44 - 55 [ f]asano g , bettoni d. 1994 . _ astron . 107:1649 - 67 [ _ ap . j. _ ] 457:490 - 99 [ f]reudling w , almudena prieto m. 1996 . _ _ 306:39 - 48 [ f]ukazawa y , makishima k , matsushita k , yamasaki n , ohashi t , et al _ 48:395 - 407 [ g]arcia a. 1995 . _ 297:56 - 60 [ g]eller mj , postman m. 1983 . _ ap . j. _ 274:31 - 38 [ g]isler g. 1980 . 85:623 - 25 [ g]overnato f , chincarini g , bhatia r. 1991 . _ ap . j. _ 371:l15-l18 [ g]overnato f , tozzi , p , cavaliere a. 1996 . _ ap . j. _ 458:18 - 26 [ h]ammer f , nottale l. 1986 . _ 155:420 - 22 [ h]eiligman gm , turner el . 1980 . _ ap . j. _ 236:745 - 49 [ h]enriksen mj , mamon ga . 1994 . _ ap . j. lett . _ 421:l63-l66 [ h]enry jp , gioia i m , huchra jp , burg r , mclean b , et al . 1995 . _ ap . j. _ 449:422 - 30 [ h]ernquist l , katz n , weinberg dh . 1995 . _ ap . j. _ 442:57 - 60 [ h]ickson p. 1982 . _ ap . j. _ 255:382 - 91 [ h]ickson p. 1990 . in _ paired and interacting galaxies , _ int . astron . union colloq . huntsville , al : nasa marshall space flight center [ h]ickson p. 1993 . _ ap . lett . & commun . _ 29:1 - 207 [ h]ickson p. 1994 . _ atlas of compact groups of galaxies _ , basel : gordon & breach [ h]ickson p. 1997 . _ ap . lett . & commun . _ in press [ h]ickson p , kindl e , huchra jp . 1988a . _ ap . j. _ 329:l65-l67 [ h]ickson p , kindl e , huchra jp . 1988b . _ ap . 331:64 - 70 [ h]ickson p , kindl e , auman jr . _ 70:687 - 98 [ h]ickson p , mendes de oliveira c , huchra jp , palumbo gg 399:353 - 67 [ h]ickson p , menon tk , palumbo ggc , persic m. 1989 . _ ap . 341:679 - 84 [ h]ickson p , ninkov z , huchra jp , mamon ga . 1984 . in _ clusters and groups of galaxies _ , ed . f mardirossian , g giuricin , m mezzetti , pp . dordrecht : reidel [ h]ickson p , richstone do , turner el . 213:323 - 26 [ h]ickson p , rood hj . 1988 . _ ap . j. _ 331:l69-l72 [ h]uchtmeier w. 1994 . _ 286:389 - 94 [ h]unsberger sd , charlton jc , zaritsky d. 1996 . _ ap . j. _ 462:50 - 56 [ h]umason ml , mayall nu , sandage ar . 1956 . 61:97 [ h]utchings jb 1995 . 109:928 - 34 [ h]utchings jb , crampton d , johnson a. 1995 . 109:73 - 80 [ i]ovino a , hickson p. 1996 . _ mnras _ in press [ i]shizawa t. 1986 . _ astrophys . space sci . _ 119:221 - 25 [ i]shizawa t , matsumoto r , tajima t , kageyama h , sakai h. 1983 . _ 35:61 - 76 [ k]ent sm _ 93:554 - 57 [ k]indl e. 1990 . a photometric and morphological study of compact groups of galaxies and their environments , ph.d . thesis , university of british columbia , vancouver , canada , 168 pp . [ k]irshner rp , malumuth em . 236:366 - 72 [ k]iseleva lg , orlov vv . 1993 . _ vistas astron . _ 36:1 - 30 [ k]odaira k , doi m , ichikawa s - i , okamura s. 1990 . _ 1:283 - 95 [ k]odaira k , iye m , okamura s , stockton a. 1988 . _ 40:533 - 45 [ k]odaira k , sekiguchi m , sugai h , doi m. 1991 . _ 43:169 - 76 , [ k]ormendy j , sanders db . 390:l53 - 56 [ l]acey c , guiderdoni b , rocca - volmerange b , silk j. 1993 . _ 402:15 - 41 [ l]acey c , silk j. 1991 . _ 381:14 - 32 [ l]imber dn , mathews wg . 132:286 - 305 [ l]ongo g , busarello g , lorenz h , richter g , zaggia s. 1994 . _ astron . astrophys . _ 282:418 - 24 [ l]ongo g , grimaldi a , richter g. 1995 . astrophys . _ 299:l45-l48 [ l]ynds cr , khachikian ee , amirkhanian as . _ pisma astron . zh . _ 16:195 - 98 [ m]accagni d , gioia i m , henry jp , maccacaro t , vettolani gp 100:1461 - 67 [ m]alumian vh . 1996 . 317:101 [ m]alykh sa , orlov vv . 1986 . _ astrofizika _ 24:445 - 51 [ m]amon ga . 307:426 - 30 [ m]amon ga j. _ 321:622 - 44 , [ m]amon ga . _ 219:98 - 100 [ m]amon ga . 1990 . in _ paired and interacting galaxies , _ int . union colloq . huntsville , al : nasa marshall space flight center [ m]amon ga . 1995 . in _ groups of galaxies , _ asp conf . 70:83 - 94 [ m]arzke ro , huchra jp , geller mj . 1944 . _ ap . j. _ 428:43 - 50 [ m]assey p. 1977 . _ 89:13 - 18 [ m]atthews k , soifer bt , nelson j , boesgaard h , graham jr . , et al . _ 420:l13-l16 [ m]endes de oliveira c. 1995 . _ mnras _ 273:139 - 45 [ m]endes de oliveira c , giraud e. 1994 . _ ap . j. _ 437:l103-l106 [ m]endes de oliveira c , hickson p. 1991 . _ ap . 380:30 - 38 [ m]endes de oliveira c , hickson p. 1994 . _ ap . j. _ 427:684 - 95 [ m]enon tk . 372:419 - 23 [ m]enon tk _ mnras _ 255:41 - 47 [ m]enon tk 110:2605 - 9 [ m]enon tk _ mnras _ 274:845 - 852 [ m]enon tk , hickson , p. 1985 . _ j. _ 296 , 60 - 64 , [ m]enon tk , leon s , combes f. 1996 . _ bull . am . 188:4502 [ m]irzoian lv , miller js , osterbrock de . 1975 , _ ap . 196:687 - 8 [ m]oles m , del olmo a , perea j , masegosa j , marquez i , costa , v. 1994 . _ _ 285:404 - 14 [ m]ulchaey js , davis ds , mushotzky rf , burstein d. 1993 . _ ap . j. lett . _ 404:l9-l12 [ m]ulchaey js , davis ds , mushotzky rf , burstein d. 1996a . _ ap . 456:80 - 97 [ m]ulchaey js , mushotzky rf , burstein d , davis ds . 1996b . _ ap . j. lett . _ 456:l5-l8 [ n]avarro jf , mosconi m , garcia lambas d. 1987 . _ mnras _ 228:501 - 11 [ n]ilson pn . _ uppsala general catalogue of galaxies , _ uppsala obs . [ n]ottale l , moles m. 1978 . _ 66 , 355 - 58 [ o]leak h , stoll d , tiersch h , macgillivray h. 1995 . 109:1485 - 89 [ o]striker jp , lubin lm , hernquist l. 1995 . _ j. _ 444:l61-l64 [ o]striker jp , tremaine sd . _ 202:l113-l117 [ p]alumbo ggc , saracco p , hickson p , mendes de oliveira c , 1995 . 109:1476 - 84 [ p]alumbo ggc , saracco p , mendes de oliveira c , hickson p , tornatore v , baiesi - pillastrini gc . 1993 . _ ap . j. _ 405:413 - 18 [ p]alumbo g , saracco p , hickson p , mendes de oliveira c. 1995 . _ astron . 109:1476 - 84 [ p]ascarelle sm , windhorst ra , driver sp , ostrander ej , keel wc . 1996 . _ ap . j. lett . _ 456:l21-l24 [ p]eebles pje . _ physical cosmology . _ princeton : princeton univ . press [ p]etrosian mb . 1974 . _ astrofizika _ 10:471 - 76 [ p]etrosian mb . _ astrofizika _ 14:631 - 35 [ p]ildis ra . 455:492 - 96 [ p]ildis ra , bregman jn , evrard ae . 1995a . _ ap . j. _ 443:514 - 26 [ p]ildis ra , bregman jn , schombert jm 110:1498 - 506 [ p]ildis ra , evrard ae , bregman jn . 1996 . 112:378 - 87 [ p]onman tj , bertram d. 1993 . _ nature _ 363:51 - 54 [ p]onman tj , allan dj , jones lr , merrifield m , mchardy i m , lehto hj , luppino ga . _ nature _ 369:462 - 64 [ p]onman tj , bourner pdj , ebeling h , bohringer h. 1996 . _ mnras _ 283:690 - 708 [ p]ostman m , geller mj . 1984 . _ 281:95 - 99 [ p]randoni i , iovino a , macgillivray ht . 107:1235 - 44 [ p]ress wh , schechter p. 1974 . _ ap . j. _ 187:425 - 38 [ p]uche d , carignan c. 1991 . _ ap . j. _ 378:487 - 95 [ r]amella m , diaferio a , geller mj , huchra jp . 1994 . _ astron . 107:1623 - 28 [ r]ampazzo r , sulentic jw . 1992 . _ astron . astrophys . _ 259:43 - 60 [ r]aymond jc , smith bw _ 35:419 - 39 [ r]ibeiro alb , de carvalho rr , zepf se . 1994 . _ mnras _ 267:l13-l16 [ r]ibeiro alb , de carvalho rr , coziol r , capelato hv , zepf se . _ 463:l5-l8 [ r]odrigue m , schultz a , thompson j , colegrove t , spight ld , et al . 109:2362 - 67 [ r]ood hj , struble mf . 1994 . _ 106:413 - 16 [ r]ood hj , williams ba 339:772 - 82 [ r]ose ja . 211:311 - 18 [ r]ose ja 231:10 - 22 [ r]ose ja , graham ja . 1979 . _ ap . j. _ 231:320 - 26 [ r]ubin vc , ford wkj , hunter da . 1990 . _ ap . j. _ 365:86 - 92 [ r]ubin vc , hunter da , ford wkj . 1991 . _ ap . j. suppl . _ 76:153 - 83 [ s]akai s , giovanelli r , wegner g. 1994 . 108:33 - 43 [ s]andage a , binggeli b & tammann ga . 90:1759 - 71 [ s]anders db , mirabel if . 1996 . _ 34:749 - 92 [ s]aracco p , ciliegi p. 1995 . _ astron . _ 301:348 - 58 [ s]argent wlw _ 153:l135 - 37 [ s]chade d , lilly sj , crampton d , hammer f , le fevre o , tresse l. 1995 . _ j. _ 451:l1-l4 [ s]chechter p. 1976 . _ ap . 203:297 - 306 [ s]chwartz da , schwarz j , tucker w. 1980 . _ ap . j. _ 238:l59-l62 [ s]eyfert ck . j. _ 53:203 - 4 [ s]hakhbazian rk 1957 . _ 177:11 [ s]hakhbazian rk 1973 . _ astrofizika _ 9:495 - 501 [ s]hakhbazian rk , amirkhanian as . 1979 . _ astrofizika _ 14:455 - 66 [ s]hakhbazian rk , petrosian mb . _ astrofizika _ 10:13 - 20 [ s]tephan me . mnras _ 37:334 - 39 [ s]tockton a. 1982 . _ 257:33 - 39 [ s]toll d , tiersch h , oleak h , baier f , macgillivray ht . 1993a . _ 314:225 - 67 [ s]toll d , tiersch h , oleak h , baier f , macgillivray ht . 1993b . _ 314:317 - 60 [ s]toll d , tiersch h , oleak h , macgillivray ht _ 315:11 - 61 [ s]toll d , tiersch h , oleak h , macgillivray ht . 1994b . _ 315:97 - 150 [ s]toll d , tiersch h , braun m. 1996a . _ 317:239 - 57 [ s]toll d , tiersch h , braun m. 1996b . _ 317:315 - 32 [ s]ulentic jw . 1983 . _ 270:417 - 21 [ s]ulentic jw . 1987 . _ ap . j. _ 322:605 - 17 [ s]ulentic jw . 1993 . in _ progress in new cosmologies : beyond the big bang hc arp et al , pp . new york : plenum press [ s]ulentic jw . j. _ in press [ s]ulentic jw , de mello rabaa df . 1993 . _ ap . j. _ 410:520 - 25 [ s]ulentic jw , lorre jj . _ 120:36 - 52 [ s]ulentic jw , pietsch w , arp h. 1995 . _ 298:420 - 26 [ s]ulentic jw , rabaa cr 429:531 - 39 [ t]hompson la , & gregory sa . 242:1 - 7 [ t]iersch h. 1976 . _ 297:301 - 3 [ t]ikhonov na . _ soobsc . spets . _ 49:69 - 84 [ t]ikhonov na _ astrofizika _ 27:253 - 64 [ t]ovmasian gm , shakhbazian ets , shirbakian ms . 1980 . _ astrofizika _ 15:513 - 14 [ t]oomre a. 1977 . in _ the evolution of galaxies and stellar populations , _ ed . bm tinsley , rb larson , pp . new haven : yale univ . [ t]suboi m , nakai n. 1994 . jpn . _ 46:l179-l182 [ t]ully rb . _ nearby galaxies catalog _ , cambridge : cambridge univ . press [ v]ennik j , richter gm , longo g. 1993 . 314:393- [ v]enugopal vr . _ mnras _ 277:455 - 57 [ v]ettolani g , gregorini l. 1988 . _ astron . _ 189:39 - 41 [ v]orontsov - velyaminov ba _ atlas and catalog of interacting galaxies , _ vol 1 . moscow : sternberg inst . [ v]orontsov - velyaminov ba . _ atlas of interacting galaxies , part ii _ , _ astron . suppl . _ 28:1 [ v]orontsov - velyaminov ba , dostal va , metlov vg . _ soviet astron . _ 6:394 - 97 [ v]orontsov - velyaminov ba , metlov vg . 1980 . _ soviet astr . _ 6 , 199 - 201 [ w]alke dg , mamon ga . 1989 . _ _ 225:291 - 302 [ w]eil ml , hernquist l. 1994 . _ ap . j. lett . _ 431:l79-l82 [ w]hite sdm . 1990 . in _ dynamics and interactions of galaxies , _ ed . r weilen , pp . 380 - 88 , heidelberg : springer - verlag [ w]hite sdm . 1992 . in _ clusters and superclusters of galaxies _ , nato advanced study institute , pp . 1 - 15 . [ w]hitmore bc . 1992 . in _ physics of nearby galaxies : _ nature _ or nurture ? , _ tx thuan , c balkowski , j tran thanh van , _ 12th moriond astrophys . meet . giv - sur - yvette : ed . front . [ w]hitmore bc , gilmore dm 367:64 - 68 [ w]hitmore bc , schweizer f , leitherer c , borne k , robert c. 1993 . _ astron . j. _ 106:1354 - 70 [ w]illiams ba , mcmahon pm , van gorkom jh . 101:1957 - 68 [ w]illiams ba , rood hj . _ 63:265 - 94 [ w]illiams ba , van gorkom jh 95:352 - 55 [ w]u w , rabaa cr , sulentic jw 1994 . soc . _ 26:1494 [ z]epf se . 407:448 - 55 [ z]epf se , whitmore bc . 1991 . _ ap . j. _ 383:542 - 49 [ z]epf se , whitmore bc . 1993 . _ ap . j. _ 418:72 - 81 [ z]epf se , whitmore bc , levison hf . 1991 . _ 383:524 - 41 [ z]heng j , valtonen mj , chernin ad . 1993 . 105:2047 - 53 [ z]ou z - l , xia x - y , deng z - g , wu h. 1995 . _ astron . astrophys . _ 304:369 - 73 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: long before the development of general relativity , john michell wondered about the gravitational influence of objects on the light they emit , and how one might go about finding objects so dense that light could not escape their surfaces . `` [ i]f any other luminous bodies should happen to revolve about them we might still perhaps from the motions of these revolving bodies infer the existence of the central ones with some degree of probability , as this might afford a clue to some of the apparent irregularities of the revolving bodies , which would not be easily explicable on any other hypothesis '' @xcite . more than 200 years later , precisely this method has been employed to determine the mass of the black hole ( bh ) at the center of our milky way . sgr a@xmath10 is one of the most tightly constrained bhs in the universe , with a total mass uncertainty ( statistical and systematic ) of less than 10% ( see * ? ? ? * and references therein ) . using stellar motions to measure the bh masses in more distant galaxies is complicated by our current inability to spatially resolve the orbits of individual stars . instead , one must rely on the luminosity - weighted line - of - sight velocity distribution ( losvd ) within different spatial resolution elements , and model the most likely contribution to the gravitational potential from the unseen bh ( beyond that provided by the stars themselves and any dark matter ) . the most common numerical approach to constraining the bh mass ( @xmath0 ) and mass - to - light ratio of the stars ( @xmath11 ) is the orbit superposition method of @xcite , which simultaneously optimizes the fit to the surface brightness distribution and the line - of - sight kinematics of stars within the nuclear region . application of that technique has yielded more than 50 bh mass estimates ( e.g. , * ? ? ? * ; * ? ? ? a complimentary mass scale has been developed for the bhs in active galactic nuclei ( agns ) , based on the technique of reverberation mapping ( rm ; * ? ? ? * ; * ? ? ? measuring the size - scale and velocity of gas near the bh has provided mass estimates for @xmath1250 bhs @xcite , but systematic uncertainties remain due to the unknown geometry and dynamics of the gas . this uncertainty is traditionally encapsulated in a scaling ( or projection ) factor , @xmath13 an empirical calibration for the rm mass scale @xmath14 has been established by assuming that those agns also lie on the tight correlation between bh mass and bulge stellar velocity dispersion ( the @xmath15 relation ) found in quiescent galaxies . this was first done by @xcite , who obtained an average rm mass scale of @xmath16 . subsequent determinations have ranged from @xmath17 @xcite to @xmath18 @xcite , with the most recent value being @xmath19 @xcite . for ngc 4151 , this gives a current rm - derived bh mass of @xmath20 ( based on the revised estimate of * ? ? ? it is important to note that the error generally quoted on a rm - derived bh mass in an individual galaxy is the formal uncertainty in the virial product and does not include the uncertainty in the nominal mean scale factor @xmath21 . what has been missing is an independent bh mass measurement for a sample of galaxies with rm estimates . early results have suggested that the empirically calibrated rm masses are consistent with the values produced by other techniques @xcite , but the few agns near enough to be probed by stellar dynamics have not yielded particularly tight constraints ( in part , owing to the complicating factor of the bright agn overwhelming the stellar absorption features at the smallest galactic radii ) . while only a few galaxies lend themselves to multiple methods for measuring the masses of their central bhs , such independent bh mass determinations are crucial to isolating the systematic errors inherent in each method . it is only via independent measurements that it is possible to derive robust bh masses . ngc 4151 is a particularly interesting galaxy since it is one of the rare objects for which there already exist two independent dynamical measurements for the mass of the bh . @xcite use optical spectroscopy of stellar absorption lines and orbit superposition modeling to derive an upper limit for the bh mass of @xmath22 when the galactic bulge is assumed to be edge - on . when the bulge is assumed to have the same inclination as the large - scale disk , they derive a bh mass of 4 - 5@xmath23 , which they caution is likely to be a biased solution since it is associated with a very noisy @xmath1 surface that also gives a poor fit to the data ( in terms of absolute @xmath1 value ) . @xcite use adaptive optics ( ao ) to obtain high spatial resolution , near - infrared spectroscopy of the molecular , ionized , and highly ionized gas in ngc 4151 . from the kinematics of gas in the vicinity of the bh , they derive a dynamical mass of @xmath24 for the bh . being somewhat smaller than the value obtained by @xcite , this suggests a need to revisit the stellar dynamical bh mass measurement . two serious concerns with the result of @xcite motivate the need for the high spatial resolution observations that are presented in this paper . first , based on the rm mass and the bh mass derived by @xcite , the sphere of influence of the bh is estimated to be @xmath25 , significantly smaller than the spatial resolution of the spectra used by @xcite . second , low spatial resolution integral field spectroscopy of a larger field of view @xcite shows that the true kinematic major axis of the bulge ( derived from the 2d velocity field ) is oriented @xmath26 from the axis assumed by @xcite . the observations and modeling presented in this paper are designed to overcome both of these drawbacks and to provide a more robust comparison with the gas dynamical measurement of @xcite . to enhance the comparison between the rm and stellar dynamical mass scales for bhs , we have obtained near - ir integral field spectroscopy with ao in order to study the stellar dynamics in the inner regions of the reverberation - mapped agn , ngc 4151 . in conjunction with our previous optical spectroscopy and multi - wavelength imaging data , we are able to place the tightest dynamical constraints to date on the bh mass in ngc 4151 . in section 2 , we describe the observational data . in section 3 , we discuss our stellar dynamical modeling . section 4 describes our results . section 5 contains further discussion and our conclusions . ngc 4151 is a nearby ( @xmath27 ) galaxy with a prominent bulge , faint spiral arms and a large scale bar ( see sections 3.4 and 3.5 for more details on the morphology and distance of the galaxy ) . the galaxy s nuclear activity has been studied extensively since the original paper by seyfert ( 1943 ) ; in particular , the analysis of uv and optical emission lines have been used for determining the bh mass via rm ( e.g. , * ? ? ? the analysis presented here uses new near - ir spectroscopy performed with ao , as well as previously published optical long - slit spectroscopy , and space- and ground - based imaging ( primarily from * ? ? ? * hereafter paper i ) to provide a mass estimate independent of the rm value . observations of ngc 4151 and three giant stars ( used as velocity templates ) were obtained with the near - infrared integral field spectrograph ( nifs ; * ? ? ? * ) on gemini north between 16 feb 2008 and 24 feb 2008 in queue mode ( program i d gn-2008a - q-41 ) . the nifs image - slicing integral field unit ( ifu ) provides @xmath28 near - ir spectra across a @xmath29 field , with a spatial sampling of @xmath30 , or @xmath31 `` spaxels '' ( spatial pixels ) , where the smaller pixel scale is along each of the 29 image slices . because previous observations of the k - band stellar absorption lines in ngc 4151 show the features to be weak in the central regions @xcite , we focus on the co ( 6 - 3 ) bandhead at a rest - frame wavelength of 1.62 @xmath32 m , which has been found to be quite prominent in previous nifs spectra @xcite . the @xmath2-band spectra cover a wavelength range of @xmath33m@xmath34 m ( centered at 1.6490 @xmath32 m ) , with a dispersion of 1.60 pixel@xmath35 across the 2040 spectral pixels , and a velocity resolution of 56.8 . to take advantage of the spatial sampling afforded by nifs , the observations were taken with the altair ao system @xcite , using the agn in ngc 4151 as the ao guide star . the observations employed the altair field lens @xcite to provide a more uniform point spread function ( psf ) across the nifs field . the level of ao correction depends upon the prevailing seeing conditions at the time . the median uncorrected seeing was @xmath36 , and more than 95% of the data were obtained in seeing better than @xmath37 , which provided an ao - corrected nuclear fwhm between @xmath38 and @xmath39 . to estimate the psf , we model the contribution from the agn to the spectrum in each spaxel across the field of view , taking advantage of the agn s strong [ ] @xmath40 m emission line , at an observed wavelength of @xmath41 m . although the [ ] emission is spatially extended in ngc 4151 , the off - nuclear emission line is narrower than the nuclear line profile , and tight constraints can be placed on the amount of agn spectrum that ends up in each spaxel ( see 4.2 of * ? ? ? the analysis was restricted to a wavelength region around the [ ] line ( @xmath42m@xmath43 m ) , and to control the effects of noise , the spectra were box - car smoothed over 5 pixels . a slice was taken through the center of the resulting 2d map , cutting along the smaller pixel scale ( @xmath44 ) . the psf profile shows the standard core+halo structure delivered by ao systems , which we model as a 2-part gaussian , where the core has @xmath45 , the halo has @xmath46 , and each component has equal amplitude . the data for ngc 4151 were acquired at a position angle of @xmath47 , which aligns the nifs slitlets perpendicular to the radio jet ( matching the orientation used by * ? ? ? * ; * ? ? ? * ; * ? ? ? the typical observing sequence included a 9-point dither pattern with separations of 0206 , and offsets to a sky field ( 210 away ) approximately every third frame . individual exposures were 120 s long , and the observations were split into sequences of 80 minutes in duration ( in some instances , shortened by changing weather conditions ) . arc lamps were taken in the middle of each observing sequence , and telluric standards were obtained before and after each sequence . the telluric stars were chosen to be 2mass sources to simultaneously provide photometric calibration : hd 98152 was observed before ngc 4151 , and hd 116405 afterwards . darks , flats , and ronchi mask calibrations were obtained during the daytime after each night s science observations . in total , 253 on - source frames were obtained , for a total exposure time of 30360 s. a total of 112 sky frames were also acquired . the giant stars we observed were included so as to provide a measure of the spectral broadening of the instrument , allowing the intrinsically narrow - featured stars to serve as templates that could be broadened to match the actual stellar velocity distribution of ngc 4151 . these velocity template stars were observed without ao , and were selected to span a range of spectral type : hd 35833 ( g0 ) , hd 40280 ( ) , hip 60145 ( m0 ) . the observations of each star consist of four on - source and two sky exposures , each with an integration time of 5.3 s. the nifs data were reduced with a combination of version 1.9.1 of the gemini iraf package and our own iraf scripts . calibrations were processed independently for each night . the observations of ngc 4151 were grouped by proximity to a telluric standard , typically dividing the observing sequence in half . the data were flat - fielded , dark - subtracted , wavelength - calibrated , and spatially rectified , then combined , corrected for telluric features , and flux - calibrated . the flux calibration provided by the two telluric / flux standards differs systematically by @xmath48% , implying a 0.3 mag discrepancy in the 2mass photometry for the two stars . because the absolute flux levels in the nifs data play no role in our subsequent analysis , we do not attempt to correct for this offset . since the image quality of some frames is significantly worse than others , we select a subsample of individual frames with fwhm values below a certain upper limit . with a threshold of 016 in the fwhm ( measured at a uniform wavelength of 1.6598 @xmath32 m to avoid emission features in the spectra ) , 236 frames are included in the final datacube . we spatially shift the data to account for the dither pattern and then median - combine the images , yielding a datacube containing 29@xmath4969 spatial pixels and 2040 spectral pixels . the near - ir spectra at each spatial position in the final nifs datacube are combined onto a bin of size @xmath1202-square ( 2@xmath495 raw spaxels ) , resulting in a cube of 15@xmath4915 spaxels . this choice of binning the spaxels arose from two main considerations . first , prior to carrying out the axisymmetric dynamical modeling , the kinematic data need to be symmetrized about the kinematic major axis and reflected about the rotation axis , to avoid modeling inconsistencies . this process is best accomplished with square bins , especially when the symmetry axes of the galaxy are not aligned with the axes of the spectrograph ( as is generally the case ) . second , since the halo of the psf has @xmath46 and this component comprises 50% of the overall psf amplitude , the spatial resolution of the data is not significantly altered by using larger bins . we reduce the observations of the telluric / flux standard stars and the velocity template stars with procedures similar to those we apply to the ngc 4151 data . we median - combine the four frames for each star , extract spectra with a fixed aperture of 1@xmath505 , and then we normalize each extracted spectrum for the velocity template stars by a low - order fit to the continuum . in addition to the new data from nifs , we utilize the results from the optical spectra of ngc 4151 that were analyzed in paper i. the data consist of long - slit spectra of the calcium triplet region ( @xmath51 ) taken with the mayall 4 m telescope at kitt peak national observatory in 2001 , and with the 6.5 m mmt observatory at mt . hopkins in 2004 . for the data from kitt peak , the spectroscopic slit was 2wide and was placed at a position angle of 135@xmath52 . to enhance the signal - to - noise , spectra were extracted with an aperture that increased in size away from the galaxy center , extending to radii of @xmath5314 . the spectral resolution was @xmath54 . for the observations with the mmt , the slit was 1 wide , at a position angle of 69@xmath52 . spectra were extracted in @xmath55 bins to a distance of 12on either side of the agn . the spectral resolution was @xmath56 . the seeing conditions for the two datasets were @xmath121.8 and @xmath123 , respectively . we refer the reader to paper i for full details . to model the surface brightness profile of ngc 4151 in the near - ir , we make use of the @xmath2-band imaging from the early data release of the ohio state university bright spiral galaxy survey ( osubsgs ; * ? ? ? the data were obtained with the 1.8 m perkins telescope at lowell observatory . the image has a plate scale of 15 pixel@xmath35 , and we estimate the seeing to be 2 . we photometrically calibrate the osubsgs imaging using 2mass point - sources in the field , deriving a photometric zero - point of @xmath57 mag . to bridge the gap in the surface brightness information between the small field - of - view of the ao - assisted nifs data and the large - scale low resolution imaging data of the osubsgs , we use the _ hubble space telescope _ images from paper i. these data were taken with the f550 m filter on the high resolution channel of the advanced camera for surveys ( acs ) , and cover a field - of - view of @xmath1225@xmath4929 . we refer the reader to paper i for additional details . we also make use of @xmath58- and @xmath59-band optical imaging from the sloan digital sky survey ( sdss ) to assist in constraining the mass - to - light ratio ( @xmath11 ) . the images have an exposure time of 53.9 s in each band , with 0396 pixels and @xmath1210-seeing in both filters . we use the photometrically calibrated and sky - subtracted images that sdss makes available . as in paper i , our modeling relies on simultaneously fitting both the luminosity distribution of the bulge and the observed integrated line - of - sight velocity distributions ( losvd ) of stars obtained with various spectrographs over a range of spatial scales . we detail below the methods of deriving those quantities from the data described above . as discussed in [ sec : nifs_details ] , the nifs instrument produces a grid of spatial pixels ( `` spaxels '' ) of size @xmath30 which were combined onto a bin of size 02-square , resulting in a cube of 15@xmath4915 spaxels . the spectrum from each binned spaxel is then analyzed using the `` penalized pixel fitting '' ( ppxf ) package for idl by @xcite . the ppxf routine computes the velocity shift , the velocity dispersion , and the higher - order moments of the stellar losvds ( parameterized as gauss - hermite ( gh ) polynomials ) that , when convolved with the velocity template star spectrum , provide the best match to the observed data . the software allows emission features to be masked in the fit , and also provides for multiplicative and additive polynomials to be included ( which we use to account for the contribution of the agn continuum ) . our ppxf analysis fits the losvd up to 4th - order gh terms , including 2nd - order multiplicative and additive polynomials , and fits the spectra over the wavelength range @xmath60m@xmath61 m , with four small wavelength gaps around agn emission lines and residual sky features . figure [ fig.spec ] shows four example spectra and ppxf fits , from different positions within the datacube . we obtain the smallest uncertainties on the kinematic parameters when we fit the agn spectra with the m0 velocity template , and allowing simultaneous use of all three templates does not measurably improve the results . we also compared the velocity measurements to those obtained when using nifs velocity templates ( spectral types , , , and a ) acquired by watson et al . ( 2008 ; gemini program gn-2006b - sv-110 ) . the m - type stars provided better matches to the relative strengths of the absorption lines in each ngc 4151 spectrum , and the kinematic uncertainties were smaller than for the k - type stars . compared to the kinematics derived from our m0 velocity template , the differences were typically less than 1@xmath4 , suggesting our fits are robust against the effects of template - mismatch . our results are consistent with those of @xcite , who found better matches to @xmath2-band galaxy spectra with m - type stars than with k - type stars . figure [ fig : vel_fields_orig ] ( top row ) shows maps of the line - of - sight velocity ( @xmath62 ) , velocity dispersion ( @xmath63 ) , and the 3rd and 4th gh moments ( @xmath64 ) in 02 spaxels . the @xmath65 coordinates are aligned with the major and minor kinematic axes , respectively . the color scale is in km / s for @xmath62 and @xmath63 , and in dimensionless units for @xmath66 and @xmath67 . we then bi - symmetrize the 2d losvd map ( fig . [ fig : vel_fields_orig ] bottom row ) around the kinematic minor axis at a position angle of @xmath68 ( @xmath69 relative to the nifs y - axis ) , using an idl routine developed by @xcite . this procedure enhances the s / n by combining the measurements on either side of the kinematic minor axis in a symmetric ( @xmath63 and @xmath67 ) , or anti - symmetric ( @xmath62 and @xmath66 ) way . data point is multiplied by -1 and averaged with the @xmath66 value at a position mirrored on the opposite side of the bh . ] such a procedure is necessary in order to obtain sensible results from an axisymmetric dynamical modeling code , which assumes such an underlying symmetry . the kinematic data were bi - symmetrized in a weighted mean , using the individual uncertainties reported by ppxf to derive the weights . the final kinematic uncertainties are computed as the errors in the weighted mean values . figure [ fig : nifs_sau ] shows the line - of - sight stellar velocity fields from nifs ( left ) and sauron ( right ) . the latter data are from @xcite and show a similar large scale rotational velocity to that seen in the inner 3@xmath493 region observed with nifs . although the large scale line - of - sight velocity fields in both panels show the same global rotational velocity , the central region of the sauron velocity field within 5@xmath49 5 ( right ) is dominated by agn continuum . therefore , the innermost parts allow no velocity measurement , and appear as a black oval region in the image . thus , the sauron data do not provide constraints on the bh mass , although they could provide constraints on the mass - to - light ratio ( @xmath11 ) . however , we do not include the sauron data in our modeling , relying instead on the long - slit kinematics obtained from mmt and kpno to provide the constraints on @xmath11 at large radii . this is partly a pragmatic choice since the use of ifu data on both small and large scales would significantly increase the total number of kinematic constraints that would need to be fitted and , therefore , would require a significantly larger orbit library than we use here , making the problem significantly more computationally expensive than we are able to handle . we calculate the @xmath2-band surface brightness profile using our nested series of images to probe different spatial scales . for large radii ( @xmath70 ) , we use the photometrically calibrated osubsgs images . at intermediate radii ( @xmath71 ) , we use the acs optical image , allowing a significant region of overlap with the osubsgs data . the acs flux level is then rescaled to provide the best match to the @xmath2-band data in the overlap region . our use of the optical data could potentially bias the surface brightness profile we derive if there are significant variations in extinction or stellar population at radii of @xmath72 . however , there are no independent indications of stellar population changes in that part of ngc 4151 , and the radial profile of the acs image is relatively smooth at those distances . those radii are also well isolated from the psf of the agn in the acs image . in the central regions , we derive a stellar flux profile directly from the nifs datacube . one of the outputs of the ppxf modeling is the derived amplitude of the stellar template that best matches the galaxy data . as our template spectra are continuum - normalized , these amplitudes constitute relative measurements of the stellar flux in each spaxel , calculated in a way that naturally omits the agn contribution to the light profile . the unbinned data are used to probe radii of 0@xmath73 ( excluding the agn - dominated central spaxel ) . the overall nifs stellar profile is then scaled in amplitude to match the overlapping acs data from larger radii . we then fit the full @xmath2-band surface brightness profile using the multi - gaussian expansion ( mge ) package @xcite . the mge routine fits the smallest necessary number of nested 2d gaussians with a common center and radially varies their amplitudes and axis ratios to obtain the best possible smooth fit to the observed 2d surface brightness distribution , taking account of the psf . the deprojection of a 2d surface brightness distribution to derive the 3d luminosity distribution suffers from the well known degeneracy arising from `` konus densities '' that are invisible in projection @xcite . a number of different methods have been used to derive the 3d luminosity density , none of which guarantee a unique deprojection . the mge method has the advantage that , once an angle of inclination is assumed , each 2d gaussian has a unique 3d luminosity density whose potential is easily computed and which is positive everywhere @xcite . the code outputs the amplitude ( total counts ) , width in pixels ( @xmath74 ) , and observed axis ratio ( @xmath75 ) for each gaussian . we convert those values to a central surface brightness , @xmath76 , for each gaussian as : @xmath77 where @xmath78 is the @xmath2-band surface brightness of the sun at a distance where 1 pc@xmath79 subtends 1 arcsec@xmath79 . for an absolute @xmath2-band magnitude of 3.32@xmath530.03 mag @xcite , this gives @xmath80 mag arcsec@xmath81 . the surface brightness is computed from the mge fits as : @xmath82 where @xmath83 is the osubsgs @xmath2-band image zero - point of 22.2 mag , @xmath84 is the total counts , @xmath85 is the plate scale of @xmath86 pixel@xmath35 , and the @xmath2-band extinction is @xmath87=0.016 mag ( * ? ? ? * retrieved from the nasa / ipac extragalactic database ) . for the mge modelling , a 1-component psf was used , having an effective width of @xmath88 . the output of our mge fit to the surface brightness distribution is given in table [ tab : mge_sb ] . for each gaussian in the mge fit , the first column gives an identifying number , the second column gives the @xmath2-band luminosity density , the third column gives the standard deviation @xmath4 ( in arcsec ) , and the fourth column gives the flattening @xmath75 ( ellipticity ) . it is clear from the values of @xmath75 in the third column that the bulge surface brightness distribution is close to circular since most of the gaussians have @xmath89 ( except for the innermost and outermost gaussians ) suggesting that the bulge is probably close to spherical . in figure [ fig : sb_profile ] , we show the composite radial surface brightness profile obtained by patching together the three imaging datasets . lcccgaussian & surface density , @xmath90 & gaussian @xmath4 & axis ratio @xmath75 + number & ( @xmath91/pc@xmath92 ) & ( ) & + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) + + 1 & 9.3963e+04 & 0.126 & 0.7906 + 2 & 1.3200e+05 & 0.232 & 1.0000 + 3 & 7.8100e+04 & 0.549 & 1.0000 + 4 & 2.0951e+04 & 1.421 & 1.0000 + 5 & 6.7191e+03 & 2.799 & 1.0000 + 6 & 3.7009e+03 & 6.703 & 1.0000 + 7 & 6.8485e+02 & 14.513 & 0.9550 + 8 & 3.1906e+02 & 64.987 & 0.7052 + [ tab : mge_sb ] to provide a rough constraint on the relationship between the luminosity distribution and the stellar mass distribution , we combine optical imaging from sdss with the osubsgs @xmath2-band image . we construct ( @xmath93 ) and ( @xmath94 ) color maps of ngc 4151 , finding the bulge colors of ( @xmath93)=1.1@xmath530.1 mag , and ( @xmath94)=2.4@xmath530.2 mag to be quite uniform across the bulge ( aside from the central few arcseconds , where the agn dominates the emission ) . @xcite calculated @xmath2-band mass - to - light ratios ( @xmath95 ) in ( @xmath93)-(@xmath94 ) color - space , based on a monte carlo analysis of @xcite stellar population synthesis models . examining our measured colors in the @xcite color - space indicates that @xmath8 . our initial exploration with dynamical models spanning the entire range @xmath96 showed that the best fit values were always in the range @xmath97 , hence this range of values is used in the models below . ngc 4151 was classified by @xcite as ( r)sab(rs)ab : a disk galaxy with a pseudo - outer ring ( r ) , a `` mixed '' bar ( ab ) ( i.e. , possibly barred but is not strongly so ) , with mixed spiral / inner ring classification ( rs ) , and an intermediate to large bulge ( ab ) . figure [ fig : slit_orient ] ( left ) shows the palomar observatory sky survey ( poss ) image of ngc 4151 in the red optical filter . in part because of the very circular bulge , the classification of this galaxy as barred has been questioned by authors as far back as @xcite , who speculated that the elongation was not a bar , but was in fact made up of material that was ejected from the nucleus in a former explosive event . however , we believe that ngc 4151 may be an example of a galaxy containing a `` barlens '' , an oval - shaped structure that is likely to be the vertically thick aspect of a bar seen from a particular viewing angle @xcite . in paper i , we assumed that the kinematic major axis is aligned with the bar , since spectral information was only available along two long - slits . the kpno slit ( represented by the red line in fig . [ fig : slit_orient ] left ) was oriented along the bar at a position angle of 135@xmath52 @xcite . the mmt slit ( green line in fig . [ fig : slit_orient ] left ) was placed at a position angle of 69@xmath52 . the middle panel of figure [ fig : slit_orient ] shows the region within the white box from the left panel . the grey box in the middle panel represents the orientation and shape of the nifs field of view . it is quite clear from a comparison of the nifs velocity field and the sauron velocity field in fig . [ fig : nifs_sau ] , that the rotation axis of the disk and bulge are not perpendicular to the bar as was assumed in paper i. this is an important difference between the models presented here and the models presented in paper i. to first order , the bh mass derived by the stellar dynamical modeling depends linearly on the assumed distance to the galaxy where r is the linear radius within which the velocity ( or velocity dispersion ) is measured , and the conversion between angular and linear radii is inversely proportional to distance in the nearby universe . ] . for consistency with the analysis in paper i , we adopt a distance for ngc 4151 of 13.9 mpc ( based on the recession velocity of 998 measured by * ? ? ? however , as noted in paper i , the actual distance to the galaxy is rather uncertain , with published estimates spanning a range of 10 mpc to 30 mpc . recently , @xcite compiled a group - averaged tully - fisher distance of 11.2@xmath531.1 mpc in the extragalactic distance database . however , that group distance is based on just four galaxies , which have a standard deviation in their distance moduli of more than four times the individual distance modulus errors . furthermore , it is unclear whether the contribution to the group average distance from ngc 4151 ( of @xmath124 mpc ) accounts for the agn contribution to the galaxy brightness . thus , we can not regard the @xcite value as being any more reliable than other distances , and until a more accurate distance can be obtained , we continue to use the value of 13.9 mpc . we apply what is now the standard stellar dynamical modeling approach of orbit superposition ( e.g * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) to fit the observed losvds and luminosity distribution of ngc 4151 . the orbit superposition ( or `` schwarzschild '' ) method of dynamical modeling uses line - of - sight kinematical and imaging data to constrain the mass distributions of external galaxies . the implementation of the code used here differs from the method detailed in ( * ? ? ? * hereafter vme04 ) in two small aspects . it is ( a ) designed to fit generalized axisymmetric models that are obtained from direct deprojection of the surface brightness profiles of the individual galaxies using a mge code , following the method outlined in @xcite ; ( b ) modified to fit integral field kinematics in addition to long - slit kinematics . we restrict our models to the surface brightness distribution within 50 i.e. , the bulge region using the mge decomposition from table [ tab : mge_sb ] . for each dynamical model , we assume a constant @xmath2-band mass - to - light ratio ( @xmath95 ) for the stars value for the stars and the best - fit model values of @xmath95 described below are completely consistent with the expectations derived from the stellar photometry in sec . [ sec : mtol ] . even if we were to include such a halo , its contribution to the inner 15 region , where the nifs kinematics exist , would likely be small . as we will show , the nifs data entirely determine our final bh mass , justifying our decision to not include a dark matter halo . ] , with values varying from @xmath98 . the resulting density profile is then converted to a gravitational potential using an axisymmetric multipole expansion scheme @xcite . the gravitational potential of a point mass ( @xmath0 ) , representing the central bh , is then added . models are constructed for every pair of parameters ( @xmath0 , @xmath95 ) , building an orbit library for each combination . the large - scale disk of ngc 4151 is inclined to the line of sight at @xmath99 @xcite . however , since the bulge appears very close to circular in projection ( see fig . [ fig : slit_orient ] and tab . [ tab : mge_sb ] ) , it is probably nearly spherical or being viewed close to face on . thus , we run models on a 2d grid of ( @xmath100 ) values for three different inclination angles @xmath101 ( edge - on ) , @xmath102 and @xmath103 ( close to face - on ) . we also run a limited set of models for which the value of @xmath0 is held fixed while the inclination angle of the model is varied with values @xmath104 , and @xmath105 , in an effort to set constraints on the inclination of the bulge to the line - of - sight . for inclination angles of @xmath101 and @xmath106 , we run orbit libraries for 18 bh masses between @xmath107 , using a baseline @xmath108 . these models are then scaled to create libraries for 18 different @xmath95 values , producing a total of 324 models for each inclination angle . we also run orbit libraries for @xmath102 , using 10 bh masses across the same mass range and 18 @xmath95 values for each @xmath0 ( 180 models ) . the range of bh mass values was set by initial experimentation with a coarse grid on a wide range of values . the lowest two values of @xmath0 are set to 10@xmath109 and 10@xmath110 , both of which consistently give poor fits to the data . the upper limit of @xmath0 was raised until it was clear that models once again gave very poor fits to the data . the combined gravitational potential of the bulge and the bh is used to integrate a large library of @xmath111 stellar orbits selected on a grid with @xmath112 energy values , @xmath113 angular momentum values ( @xmath114 ) at each energy , and @xmath115 pseudo third integral values ( @xmath116 ) at each angular momentum value ( see vme04 for details ) . for instance , models with @xmath117 used @xmath118 , @xmath119 , @xmath120 . each orbit is integrated for 100 orbital periods and its average `` observed '' contribution to each of the `` apertures '' in which kinematical data are available is stored . the orbital kinematics are subjected to the same type of instrumental effects as the real data ( i.e. psf convolution , pixel binning ) . self - consistent orbit - superposition models are constructed by linearly co - adding orbits in each library to simultaneously fit several elements within the observed apertures ( imaging and spectroscopic ) : the 3d mass distribution ( for each value of @xmath95 ) , the projected surface density distribution , and the losvds . the fits use a non - negative least squares optimization algorithm ( nnls ; * ? ? ? * ) , which gives the weighted superposition of the orbits that best reproduce both the self - consistency constraints ( i.e. , the 3d mass distribution and surface brightness distribution ) and the observed kinematical constraints ( @xmath121 ) . vme04 showed that as the ratio of orbits to constraints decreases , the error bars decrease artificially . in the majority of the models presented here we have 1240 constraints that are fitted by a library of 9936 orbits . this size of orbit library gives an orbits - to - constraints ratio of 8 which is only slightly above the ratio ( 5 ) at which vme04 find that solutions begin to be biased . in section [ sec : robust_size ] we carry out a limited exploration with a 50% larger orbit library and find that our results are unchanged . smoothing ( or `` regularization '' ) of the orbital solutions is typically performed in this type of analysis , and a variety of methods have been employed to that end . regularization mainly helps to reduce the sensitivity of the solutions to noise in the data by requiring a smoother sampling of the orbit libraries used in the solution . following @xcite , we run a trial series of models with a regularization scheme that requires local smoothing in phase space . we find that for small smoothing parameters ( @xmath122 ) the resultant best - fit @xmath0 values do not differ from the models without regularization . however , the @xmath1 value for the best - fit model is larger and the error bars on @xmath0 are smaller . vme04 showed that it is impossible ( without computationally expensive approaches such as generalized cross - validation ) to determine the ideal regularization parameter . not regularizing the solutions can yield large error bars , but choosing too large a value of the smoothing parameter can introduce a bias in the best - fit value of @xmath0 . in this paper , we present models without regularization , since imposing regularization constraints significantly increases the total number of constraints and therefore decreases even further the orbits - to - constraints ratio . for the libraries used here , models without regularization provide the most conservative error bars on the estimated value of @xmath0 . since the mass and surface mass density constraints are _ self - consistency constraints _ rather than observational constraints , they do not have observational errors associated with them . there are two ways in which the absence of observational errors are handled in the literature . @xcite and vme04 include self - consistency constraints in the nnls minimization problem with errors corresponding to predefined level of relative accuracy in the fit ( e.g. 5% -10% ) , while others @xcite solve the orbit superposition problem by requiring the self - consistency constraints ( mass and surface brightness ) to be fitted as separate as linear ( in)equality constraints with predetermined absolute accuracy . @xcite state that they require the self - consistency constraints to be fitted to an accuracy of 0.02 . we also attempted to fit self - consistency constraints as linear equalities but find that that since the values of the mass constraints vary by 5 orders of magnitude ( from @xmath123 to 10 ) , using a fixed numerical accuracy fails to fit the mass constraints at small radii ( where the logarithmic radial bins have the smallest mass values per bin and where the fit to mass constraints most strongly affects the estimated @xmath0 ) . for the models presented here we include the self - consistency constraints in the optimization problem and require the optimization problem to fit each of the 472 self - consistency constraints ( 280 mass constraints and 192 aperture surface brightness constraints ) with an error corresponding to a relative accuracy - per - constraint of 5% . for our preferred model with @xmath101 , the fractional accuracy per self - consistency constraint varies from 0.27% ( for the best fit model ) to 2.2% for the worst fit model , i.e. , always better than our required accuracy . figure [ fig : mass_err ] shows contours of constant fractional error in the fit to the self - consistency constraints ( mass and surface brightness in apertures ) over the 2d parameter space ( @xmath0 , @xmath95 ) for models with @xmath101 . the star is the location of the model with 0.27% accuracy per self - consistency constraint , and the contours are spaced at intervals of 0.1% relative accuracy per constraint . in previous applications of our schwarzschild modeling code ( vme04 , * ? ? ? * paper i ) , the error in the fit to the self - consistency constraints was essentially random over the entire 2d parameter space ( @xmath124 ) . although the self - consistency constraints were always included in the @xmath1 , they contributed so little to the total @xmath1 that the best fit solution was always driven by the fit to the kinematical constraints . we will show that this is not true for our current ngc 4151 dataset . despite the fact that the fits to the 3d mass distribution and surface brightness distribution are very good , because the fit vary systematically rather than randomly across the parameter space , the best fit solution is altered by the decision of whether or not to include the fit to the self - consistency constraints in the @xmath1 . based on the arguments above , we compute the @xmath1 in two different ways : ( 1 ) the total @xmath125 includes all the constraints in the nnls optimization problem : 280 mass constraints , 192 surface brightness constraints , and 768 kinematical constraints ( giving a total of @xmath126 constraints ) ; ( 2 ) @xmath127 which only considers the quality of the fit to @xmath128 kinematical constraints ( @xmath129 and @xmath67 in 192 apertures : 136 nifs spaxels , 15 kpno apertures , and 41 mmt apertures ) . for each of the three assumed angles of inclination for the bulge , we construct models by fitting these constraints using an orbit library consisting of @xmath117 orbits . for one inclination angle ( @xmath101 ) we do a limited exploration with a library consisting of @xmath130 orbits . after generating orbit libraries for a grid of parameters ( @xmath131 ) the best fit parameters are determined by the minimum in the 2d @xmath1-contour plot . in this section , we present the results of our modeling . in section [ sec : define_chi2 ] we describe results for inclination angles @xmath101 , @xmath102 and @xmath103 , showing 2d contour plots of @xmath125 and @xmath127 using a full 2d grid of models for each of these inclination angles . in section [ sec : inclination ] we examine the dependence on inclination angle for 4 additional models using a single @xmath0 . in section [ sec : robust ] we explore the robustness of our solutions by examining their sensitivity to the size of the orbit library and to the effect of restricting the set of kinematic constraints to those from the nuclear region only . we also present some of the best - fit solutions and show how well different pairs of @xmath100 fit both the 1d and 2d kinematical data . in section 5.4 , we discuss our results in the context of the relation . figure [ fig:2dchi2 ] shows 2d contour plots of @xmath125 in the plane of model parameters @xmath0 and @xmath95 for @xmath101 ( top ) , @xmath102 ( middle ) , and @xmath103 ( bottom ) . the grid of points shows the model parameters ( @xmath100 ) for which the orbit libraries is constructed and the optimization problem is solved . the first 6 contour levels in all the @xmath1 contour plots that follow correspond to @xmath132 , 68.3% confidence ) , @xmath133 , 90% confidence ) , @xmath134 , 95.4% confidence ) , @xmath135 , 99% confidence ) , @xmath136 , 99.73% ) , and @xmath137 , 99.99% ) , with subsequent contours equally spaced in @xmath138 . for @xmath101 , the minimum value of @xmath125 is obtained for @xmath139 and @xmath140 , with @xmath141 . for inclination angle of @xmath102 , the best fit model has @xmath142 and @xmath143 , with @xmath144 . for inclination angle of @xmath103 , the best fit model has @xmath145 and @xmath140 , with @xmath146 . table [ tab : bestfit ] summarizes the results and includes minimum @xmath1 values and the 1@xmath4 and 3@xmath4 error bars . the errors on a given parameter are obtained by marginalizing the 2d @xmath147 values over the other parameter . figure [ fig:1dchi2 ] ( top ) shows the marginalized 1d @xmath138 versus @xmath0 for the three different inclination angles , as indicated by the legends . the horizontal dotted lines are at @xmath148 ( 1@xmath4 , 68.5% confidence interval for 1 degree - of - freedom ) , @xmath149 ( 2@xmath4 , 90% confidence interval ) , and @xmath150 ( 3@xmath4 , 99% confidence interval ) . the 1@xmath4 and 3@xmath4 errors on @xmath0 and @xmath95 are given in table [ tab : bestfit ] . likewise , figure [ fig:1dchi2 ] ( bottom ) shows the marginalized 1d @xmath138 versus @xmath95 for each of the three inclinations of models to the line - of - sight . it is important to note that with the standard methods used to derive the velocity profiles ( vps ) from the spectra , the errors in the gauss - hermite ( gh ) coefficients are not independent . the errors associated with the even moments ( @xmath4 and @xmath67 ) are correlated , and errors associated with the odd moments ( @xmath151 and @xmath66 ) are also correlated ( see , e.g. , * ? ? ? * ; * ? ? ? @xcite point out that an important consequence of correlated errors in the gh coefficients is that the errors obtained by comparing data to dynamical models using the @xmath1 estimator are significantly underestimated . in this paper we quote both 1-@xmath4 and 3-@xmath4 errors , and often consider the 3-@xmath4 errors as better representations of the true modeling errors . lcccccccc & & & & + & & & & & & & & + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) + + + 90@xmath52 & 9936 & 709.1 & 4.68 & @xmath152 & @xmath153 & 0.304 & @xmath154 & @xmath155 + & & & & & & & + 60@xmath52 & 9936 & 767.5 & 5.42 & @xmath156 & @xmath157 & 0.305&@xmath158 & @xmath159 + & & & & & & & + 23@xmath52 & 9936 & 1050.8 & 3.69 & @xmath160 & @xmath161 & 0.304 & @xmath158 & @xmath155 + & & & & & & & + + + 90@xmath52 & 9936 & 186.7 & 7.76 & @xmath162 & @xmath163 & 0.332 & @xmath164 & @xmath165 + & & & & & & & + 60@xmath52 & 9936 & 193.1 & 8.62 & @xmath166 & @xmath167 & 0.326&@xmath159 & @xmath168 + & & & & & & & + 23@xmath52 & 9936 & 348.2 & 8.50 & @xmath169 & @xmath170 & 0.329 & @xmath171 & @xmath172 + & & & & & & & + [ tab : bestfit ] the best fit solutions for both @xmath0 and @xmath95 for the three different inclination angles are consistent with each other within 3@xmath4 . however @xmath173 is the smallest for the edge - on model ( @xmath101 ) , increases for @xmath102 and is significantly larger for @xmath103 implying that it may be possible to constrain the inclination of the bulge ( see section [ sec : inclination ] ) . marginalizing @xmath125 over the three values of inclination gives a best fit @xmath174 and @xmath175 . figure [ fig:2dchi2_kin ] shows 2d contour plots of @xmath176 for the same sets of models as in figure [ fig:2dchi2 ] , with inclination angles @xmath177 ( from top to bottom ) . we emphasize that the model solutions that are used to generate figure [ fig:2dchi2_kin ] are identical to those used to generate figure [ fig:2dchi2 ] only the quantities used in computing the @xmath1 of the fit to the data differ . figure [ fig:1dchi2_kin ] shows the same data marginalized over @xmath95 ( top ) and marginalized over @xmath0 ( bottom ) . it is clear from these two figures that the best - fit solutions obtained using @xmath178 give both a larger @xmath179 and a larger @xmath180 . we will discuss the possible causes for this in section [ sec : robust ] and section [ sec : discuss ] . in contrast to the results of paper i , where @xmath101 models gave an upper limit of @xmath181 and @xmath103 models gave a best fit @xmath0 between @xmath182 , figures [ fig:1dchi2 ] and [ fig:1dchi2_kin ] show that best - fit @xmath0 is relatively insensitive to inclination ( although the actual best - fit value depends on how @xmath1 was computed ) . table [ tab : bestfit ] shows that the minimum values of @xmath183 and @xmath184 for @xmath101 are lower than the corresponding @xmath1 values for the other two inclination angles . we now try to constrain the inclination angle of the bulge of ngc 4151 by assuming a constant bh mass of @xmath185 . with this value of @xmath0 we compute orbit libraries for four additional inclination angles : @xmath186 , @xmath187 , @xmath188 , and @xmath105 . figure [ fig : chi_inc ] shows the values of @xmath183 for models with this value of @xmath0 and @xmath189 as a function of the inclination of the bulge to the line - of - sight . it is clear that @xmath183 decreases steadily with increasing inclination angle with a minimum at @xmath101 . ( a similar dependence on inclination angle is obtained for @xmath190 , although all the models give much worse fits as assessed by the @xmath125 values . ) the inclination angle of @xmath103 that is inferred from the velocity field is strongly disfavored . since the disk of this galaxy is clearly seen to be close to face - on , the edge - on orientation of the bulge preferred by our models is rather puzzling : the surface brightness isophotes of the bulge are nearly circular , hence the best - fit edge - on orientation implies that the bulge is nearly spherical in shape ; but in this case the rotation axis of the bulge must be misaligned with the rotation axis of the disk by @xmath191 . previous determinations of the orientations of the various components of ngc 4151 have shown a remarkable diversity . while the circular isophotes of the large - scale disk suggest the nearly face - on orientation with @xmath103 @xcite which motivated our testing of that value of inclination the components associated with the agn have been inferred to lie further from our line of sight . models of the bi - conical narrow - line region ( nlr ) in the uv and optical @xcite , and in the near - ir @xcite have typically found inclinations of @xmath192 . closer to the agn , @xcite fit the absorption using a pc - scale nuclear disk inclined at @xmath193 . on even smaller scales , an inclination of @xmath194 for the agn accretion disk was obtained from modeling the x - ray spectrum @xcite , while @xcite suggest that the radio jet in ngc 4151 lies very close to the plane of the sky and is interacting directly with circumnuclear gas in the plane of the galaxy ( i.e. , @xmath195 ) . this lack of alignment between the different spatial scales could be the consequence of a merger in the galaxy s recent past . it has been claimed that the periodic variability in the nuclear activity of ngc 4151 is indicative of a bh binary , with a separation on the order of @xmath196 au @xcite . while distinguishing between a single bh and a close binary is far beyond the capability of our dataset , the remnants of a merger could also contribute to the complicated dynamics we observe on the size scales of the bulge . in section [ sec : robust ] , we will show that the kinematics in the inner 5 show evidence for a bar in this region , implying that the internal dynamics of the bulge are likely not that of a spherical or axisymmetric system . a full understanding of why the kinematics of our model prefers an edge - on configuration to the more inclined configuration requires the system to be modeled by a bar dynamical modeling code , which does not exist at the present time . since the result of this section show that models with edge - on inclination ( @xmath101 ) are greatly preferred over more face - on models , in the rest of this paper we confine our analysis to models with @xmath101 . lllllcccccc & & & & & & + & & & & & & & & & & + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) + + @xmath183 & 15092 & 1240 & 641.9 & 0.52 & 4.68 & @xmath197 & @xmath198 & 0.31 & @xmath158 & @xmath199 + & & & & & & & & & & + @xmath178&15092 & 768 & 170.2 & 0.22 & 7.32 & @xmath200 & @xmath201 & 0.31 & @xmath199 & @xmath202 + & & & & & & & & & & + + + @xmath183 & 9936 & 960 & 583.1 & 0.61 & 3.82 & @xmath203 & @xmath204 & 0.32 & @xmath199 & @xmath205 + & & & & & & & & & & + @xmath178 & 9936 & 680 & 101.7 & 0.14 & 3.70 & @xmath206 & @xmath207 & 0.35 & @xmath199 & @xmath202 + [ tab : robustbestfit ] vme04 showed that the solution derived from the schwarzschild method was sensitive to two important factors : ( a ) the size of the orbit libraries used in obtaining the best - fit bh mass ; and ( b ) the spatial resolution of the kinematic data and whether or not they resolved the sphere - of - influence of the bh . they showed that biased solutions could result when the size of the orbit libraries was too small for the available data . they also showed that when the kinematic data did not resolve the sphere - of - influence of the bh , spurious results could arise . in the next two sections we perform two tests to assess the robustness of our solutions . in particular , our goal is to assess which of the two values of @xmath0 obtained previously ( 5@xmath208 or 8@xmath209 ) is more robust . in this section , we construct a set of 288 models with @xmath101 but with orbit libraries containing @xmath210 orbits ( as compared to @xmath211 used thus far ) to fit the same set of 1240 constraints that were fitted in section [ sec : define_chi2 ] . large orbit libraries are constructed for 16 bh masses between 10@xmath109 and 15@xmath212 and for 18 values of @xmath213 . figure [ fig:1dchi_i9015k ] shows the 1d @xmath183 ( solid curves ) and @xmath178 ( dashed curves ) obtained by marginalizing over the mass - to - light ratio @xmath95 ( top panel ) and obtained by marginalizing over @xmath0 ( bottom panel ) . the resulting best - fit values for @xmath0 and @xmath95 , and their 1@xmath4 and @xmath214 errors , are shown in the top two rows of table [ tab : robustbestfit ] . the reduced @xmath1 values are given in column five . it is clear from a comparison with results in table [ tab : bestfit ] that the best - fit @xmath0 values are relatively insensitive to an increase in the size of the orbit library by @xmath1250% . the best - fit value of @xmath95 obtained by using @xmath178 ( dashed curves in the bottom panel of fig . [ fig:1dchi_i9015k ] ) is somewhat lower than the value of @xmath215 obtained with the smaller library , but is still within 2@xmath4 of this value . the test with the larger orbit libraries confirms the result we obtained with @xmath117 orbits per library , implying that our best - fit parameters are not biased by having too small an orbit library . furthermore , the discrepancy between the best - fit values obtained using @xmath147 and @xmath178 persists even with larger orbit libraries . figure [ fig : kinenifs_90 ] shows examples of model fits to the 2 dimensional nifs velocity fields ( from left to right , , and the gh moments @xmath66 and @xmath67 ) for the edge - on ( @xmath101 ) case with @xmath210 . note that while the best - fit solutions listed in tables [ tab : bestfit ] & [ tab : robustbestfit ] are determined by marginalizing over one parameter in the 2d contour surface ( which itself is smoothed with a kernel ) , the 2d grid of models is discrete . the derived best - fit value ( shown as stars on the 2d @xmath1 maps of figs . [ fig:2dchi2 ] and [ fig:2dchi2_kin ] ) does not overlap with any model on the grid , but lies between 4 grid points . we therefore show velocity fields and 1d kinematics for the model whose parameters are closest to the best - fit solution . in figure [ fig : kinenifs_90 ] , the top row shows velocity fields for @xmath216 ( the model closest to the best - fit obtained with for @xmath210 ) and the middle row shows velocity fields for @xmath217 , @xmath218 ( the model closest to the best - fit obtained with for @xmath210 ) . ( the bottom row of this figure is for a model with @xmath219 , @xmath218 , obtained when only nifs kinematical data are fitted , and will be discussed in section [ sec : robust_nuclear ] . ) the fitted velocity in figure [ fig : kinenifs_90 ] should be compared with the bi - symmetrized velocity fields in the bottom row of figure [ fig : vel_fields_orig ] . the four white pixels in each panel of the figure were not fitted because it was determined that their extremely high velocity dispersion values ( @xmath220 ) were spurious ( see fig . [ fig : vel_fields_orig ] ) since their values were larger than the central velocity dispersion despite being far from the center . in figure [ fig : kinenifs_90 ] the odd - moments of the losvd ( , @xmath66 ) are fairly well fitted by both the low @xmath0 model ( top row ) and the high @xmath0 model ( middle row ) . however , the nifs velocity dispersion for the model with @xmath221 , @xmath108 ( top row ) is much lower than in the data ( bottom of fig . [ fig : vel_fields_orig ] ) . the model with the larger value of @xmath217 does a slightly better job of reproducing the higher values within @xmath222 , but fails to fit the at the edges of the nifs field , where model values are significantly lower ( i.e. bluer pixel colors ) than the observed values seen in figure [ fig : vel_fields_orig ] ( bottom ) . this is because from nifs in the inner 15 region is overall larger than that obtained with kpno ( @xmath223 ) and mmt ( @xmath224 ) . inconsistent velocity dispersion values in the _ same physical region _ drives the solution to fit the data with the smallest error bars ( i.e. , mmt data ) thereby under - estimating from nifs . also note that neither model is able to fit both the low central and high outer @xmath67 values seen in the bi - symmetrized nifs velocity fields ( fig . [ fig : vel_fields_orig ] bottom row , right - most panel ) . the difference between the model fits and the data is more clearly seen in figure [ fig : kinematics_90_1d ] which shows the one - dimensional line - of - sight velocity ( ) , velocity dispersion ( ) and the gh moments of the losvd ( @xmath66 and @xmath67 ) that we obtain for the edge - on ( @xmath101 ) model . the red and blue curves show the best fits for same models as in the top and middle panels of figure [ fig : kinenifs_90 ] with values and @xmath95 indicated by the line legends in the top left panel of the first row ( these two models used libraries with @xmath210 orbits ) . the green curves in the top four nifs panels are obtained when only the nifs data are fitted ( with @xmath211 ) , this model is discussed in the next section . the open circles in the panels show the three different kinematic datasets used in the modeling , along with their 1@xmath4 uncertainties . ( the top four panels show kinematics in the nifs spaxels lying closest to the major axis of the model ; the middle four panels show data from the kpno long - slit ; the bottom four panels show data from the mmt long - slit . ) the most striking feature of the fits is that the red curve for @xmath225 ( red curve ) the model closest to the best - fit solution derived from the minimum in @xmath183 ( from figure . [ fig:1dchi2 ] ) clearly _ underestimates _ the nifs velocity dispersion ( top right panel ) for the central - most points . the uniformly large central @xmath226 values in the inner @xmath227 region are much better fitted by larger values of @xmath228 and @xmath218 ( blue curve ) . both the models do an equally good job of fitting , , and @xmath66 from the mmt data , but overestimate from the kpno data . another important point to note is that the values of the gh parameters @xmath67 are _ negative _ in the innermost regions of each of the three datasets . neither model is able to fit the negative @xmath67 values measured with the mmt spectrograph nor the central dips to negative values in @xmath67 seen in the nifs data and kpno data . in spherical and axisymmetric galaxies , @xmath67 is negative when the velocity distribution is tangentially biased ; it is positive when orbits are predominantly radial ; and it is zero when the velocity distribution in isotropic @xcite . however , negative @xmath67 values have also been associated specifically with the presence of a bar . in recent @xmath229-body simulations of barred galaxies by @xcite , the losvds have negative @xmath67 values in the region where the bar dominates . the negative @xmath67 parameters arise due to the kinematic properties of the special orbits that constitute the bar , and due to the fact that from an external observer s point of view , the bar has a large tangential velocity component due to its pattern speed . even after a bar buckles and forms a boxy bulge , negative @xmath67 values can be seen in face - on systems @xcite . another signature of the kinematics associated with a nuclear bar is undulations in the rising part of the inner rotation curve @xcite , which we also see in the mmt data for ngc 4151 . finally , while @xmath66 is always anti - correlated with the line - of - sight velocity in axisymmetric systems , @xcite show that @xmath66 is _ correlated _ with along the major axis of the bar . for the data obtained with kpno ( where the slit lies along the major axis of the large scale bar ) @xmath66 is correlated with over the radial range @xmath230 ( figure [ fig : kinematics_90_1d ] ) , providing additional kinematical evidence for a bar . therefore it appears that despite the very circular photometric contours of the bulge in ngc 4151 , and the visual appearance of a rather weak bar at radii outside the bulge , there are several clear kinematical signatures of a small scale bar in the vicinity of the nuclear bh . this is consistent with the properties of `` barlens '' features that have recently been identified in real and simulated galaxies @xcite . as pointed out by @xcite , barlenses frequently masquerade as classical bulges because of their nearly circular central isophotes , especially when viewed nearly face - on , but have bar - like kinematics . the combination of bars and bhs can have significant implications for the central dynamics of galaxies . @xcite analyze a suite of @xmath229-body simulations of barred galaxies in which point masses representing bhs were grown adiabatically . they find that the growth of a bh of a given mass causes a 5 - 10% larger increase in @xmath63 in a barred galaxy than in an otherwise identical axisymmetric galaxy . @xcite find that if a disk galaxy with a bulge has a pre - existing black hole , the formation and evolution of a bar can result in a 15 - 40% increase in the central velocity dispersion . both studies find that the increase in the observed line - of - sight velocity dispersion in the barred systems is a consequence of three separate factors : ( a ) mass inflow due to angular momentum transport by the bar , ( b ) velocity anisotropy due to the presence of bar orbits , and ( c ) weak dependence on orientation of position angle of the bar . if ngc 4151 does have a bar and is thus non - axisymmetric , that fact also has implications for our modeling results . in stellar dynamical modeling of axisymmetric galaxies , fitting a negative @xmath67 requires a larger fraction of tangential orbits , which contribute little to the line - of - sight velocity dispersion . fitting a given velocity dispersion with such an orbit population will imply a larger enclosed mass than would be required if @xmath67 was zero or positive ( which is typically the case in axisymmetric systems with bhs ) . @xcite argue that if an axisymmetric stellar dynamical modeling code is used to derive the value of @xmath0 in a barred galaxy with a high central @xmath231 and a negative @xmath67 , the central bh mass will be systematically overestimated . if the large values of @xmath231 in the nuclear regions and the negative @xmath67 values in the mmt data are a consequence a nuclear bar , then the discrepancy between @xmath127 and @xmath125 could reflect this predicted bias in @xmath0 . in this section we examine the consequences of fitting only the nifs kinematical data ( ignoring the kinematical constraints at larger radii obtained from the lower spatial resolution long - slit spectra from kpno and mmt ) . there are three reasons why the nifs data alone may provide better constraints on the mass of the bh . first , the nifs data have higher spatial resolution ( 02 ) than the long slit data ( 1 - 2 ) . for a @xmath232 , the sphere - of - influence of the bh ( assuming @xmath233 ; see section [ sec : msigma ] below ) is @xmath1216 pc , which corresponds to @xmath12 024 for our assumed distance of 13.9 mpc for this galaxy . therefore the nifs data used in our modeling are barely able to resolve the sphere - of - influence of the bh , but are quite close to doing so . if @xmath234 , the sphere - of - influence of the bh would be 26 pc ( 038 ) and the bh sphere - of - influence would be resolved by the nifs data . the long slit data do not provide constraints on @xmath0 and are included mainly to provide constraints on the mass - to - light ratio of the stars , which is assumed to be independent of radius . second , as discussed previously , the nifs velocity dispersion within @xmath235 is significantly larger than the velocity dispersion on same physical scale obtained from the long - slit data . however , since the error bars on the long - slit mmt data are about half those from the nifs data , the optimization code tries to fit them better than the nifs data . in figure [ fig : kinenifs_90 ] ( top two panels ) , the need to fit the lower velocity dispersion values from mmt manifests as low velocity dispersion values ( blue ) at the edges of the nifs field . these low dispersion values outside the sphere - of - influence of the bh also force the model to adopt a lower mass - to - light ratio , which is then compensated for in the inner region by requiring a much larger to fit the nifs velocity dispersion . neglecting the long - slit data would allow the optimization code to raise the mass - to - light ratio in the inner @xmath532 region ( where the velocity dispersion is quite flat ) . finally , in figure [ fig : kinematics_90_1d ] , we see kinematic evidence for the bar in mmt and kpno long - slit data ( negative @xmath67 values and undulations in the rising part of the rotation curve seen in mmt data ; correlation between @xmath66 and @xmath236 in kpno data ) occurring on scales of 3 - 5 , well outside the nuclear region . except for the central value of @xmath67 in the nifs dataset ( which , as we see from fig . [ fig.spec ] , is contaminated by the agn ) , all the other @xmath67 values are positive , which is what is expected for orbits around a central bh . neglecting the long - slit kinematical data could help overcome biases introduced by the bar kinematics . assuming that the bh mass should primarily be constrained by kinematic constraints at the smallest radii we now test the effect of not including the long - slit data in the nnls optimization problem . the expectation is that neglecting the long - slit data should bring the best - fit solutions obtained from @xmath183 and @xmath178 closer to each other . we construct a full sets of models ( 18 values of @xmath0 and 18 values of @xmath95 ) in which we only fitted the nuclear kinematical data obtained with the nifs spectrograph , with an orbit library of @xmath211 . figure [ fig : nifs_only ] shows @xmath138 values obtained by marginalizing over @xmath95 ( top ) and by marginalizing over @xmath0 ( middle ) . the solid curves show @xmath237 while the dashed curves show @xmath238 . the dotted horizontal lines show the 1@xmath4 , 2@xmath4 and 3@xmath4 confidence levels , respectively ( for 1 degree - of - freedom ) . the top plot shows that the minimum @xmath178 ( dashed curve ) has decreased quite significantly and is now in agreement with the minimum @xmath183 ( solid curve ) . the best - fit values of @xmath95 ( middle panel ) are still inconsistent within 1@xmath4 but are consistent within 3@xmath4 . the bottom - most panel shows the error on the fit to the mass and surface brightness distribution . contours are spaced at intervals of 0.1% error per constraint . it is clear that the fit to the self - consistency constraints is uniform over a much larger range of @xmath0 and @xmath95 values implying that the fit to the mass is no longer a significant factor in determining the location of the minimum . the best fit black hole mass and m / l ratio obtained using @xmath237 and @xmath238 , along with the reduced @xmath138 and 1@xmath4 and 3@xmath4 error bars , are shown in the bottom two rows of table [ tab : robustbestfit ] . since the best fit @xmath0 values and their 1@xmath4 error bars from both methods of computing @xmath1 are nearly the same we use the average of the two best - fit values as the solution and sum their 1@xmath4 errors in quadrature to obtain @xmath239 . figure [ fig : nifs_only ] shows that the @xmath95 values obtained from @xmath178 are consistent with the best fit value obtained with @xmath183 at the 3@xmath4 level , but both are well within the expected range @xmath240 . therefore we use the average of the two best - fit m / l values and sum their 3@xmath4 errors in quadrature to get @xmath241 . the bottom panel of figure [ fig : kinenifs_90 ] shows a 2d kinematic map for a model with @xmath219 and @xmath242 , with parameters closest to the best - fit solution obtained with @xmath125 ( see 3rd row of table [ tab : robustbestfit ] ) . this model achieves adequately large values of @xmath231 ( see red / orange regions ) over the central nifs field by raising the m / l ratio very slightly from @xmath243 to @xmath244 . the 1d kinematic fit of this model to the nifs data along the kinematic major axis of the model is shown by the green curves in figure [ fig : kinematics_90_1d ] , from which it is also clear that the lower value of @xmath0 is able to fit the high velocity dispersion values even better than the high bh mass when the model attempted to fit both the nifs and long - slit kinematic data . this test largely confirms our hypothesis that the two main causes of the large value of @xmath0 obtained in the previous section were ( a ) inconsistencies between @xmath231 values from nifs and the long - slit data in the same spatial region , ( b ) the need to fit the low @xmath67 values and high central @xmath231 values with a low m / l ratio . thus , we see that by not including the kinematics data with poorer spatial resolution ( which are clearly unaffected by the bh as can be seen in figure [ fig : kinenifs_90 ] ) we are able to simultaneously fit the nifs kinematics and the mass and surface brightness distributions with a lower value of @xmath0 . in this section , we determine how the new nifs kinematic measurements in the nuclear region of ngc 4151 affect the galaxy s location in the @xmath15 relation . the aperture used to define the appropriate @xmath4 value for the relation is a historically contentious issue ( e.g. * ? ? ? * ; * ? ? ? * ) , but the nifs data naturally lend themselves to one of those standards : @xmath245 , which is the velocity dispersion within an aperture of radius @xmath246/8 , where @xmath246 is the bulge effective radius @xcite . ( despite having to adopt one particular standard , we note that @xcite found no systematic differences between estimates of @xmath245 ( inside of @xmath246/8 ) and estimates of @xmath4 that extended out to @xmath246 , likely because of the steep radial surface brightness gradient in most bulges . ) the @xmath246 value for ngc 4151 has been measured to be @xmath247 @xcite , while the nifs field - of - view extends to a radius of between @xmath86 and @xmath248 . thus , we sum the nifs datacube across all of the spaxels into a single spectrum , and fit for the dispersion using ppxf . from the nifs data , we find @xmath249 . this is somewhat larger than either the @xmath250 measured by @xcite , or the @xmath251 measured by @xcite . both of those earlier measurements used the calcium triplet absorption lines , but whereas the former covered a similar region of the galaxy as our nifs spectra ( @xmath252 ) , the latter used an aperture of @xmath253 . previous studies of bulge kinematics have found near - ir and optical data to give consistent results ( see * ? ? ? * and references therein ) , while the discrepancies seen in some luminous ir galaxies are in the sense of _ smaller _ @xmath4 values from the co - bandheads @xcite . our new measurement of @xmath254 is @xmath1225% above the previous velocity dispersion measurements , while our best - fit stellar dynamical estimate of @xmath255 is 17% below our earlier stellar dynamical estimate @xcite . previously , ngc 4151 had been an outlier from the agn @xmath15 relation in the direction of low-@xmath4 or high-@xmath0 . the new @xmath245 and @xmath0 measurements helps to bring ngc 4151 closer to the best - fit relation of @xcite , though it does remain on the low-@xmath4 side . finally , we note that if the bh masses in barred galaxies have been overestimated due to the use of axisymmetric dynamical modeling codes , such galaxies would lie even further below the standard relation than has been found previously ( * ? ? ? * and references therein ) . this demonstrates the importance of developing robust bar dynamical modeling approaches in the future . we have conducted ao - assisted near - ir integral field spectroscopy of the local seyfert galaxy ngc 4151 , using the nifs instrument on gemini north . we used an axisymmetric orbit - superposition code to fit the observed surface brightness distribution within 50 , nifs kinematics within @xmath5315 , and long slit kinematics along two different position angles in order to estimate the best - fit values of the mass of the bh ( @xmath0 ) and the mass - to - light ratio of the stars ( @xmath95 ) . models were constructed for 10 - 18 values of @xmath0 , 18 values of @xmath95 and 3 different inclination angles . the main results of this paper are summarized below . we use the @xmath1 estimator to determine the model which gives the best - fit to the data but find that the solution depends on whether the @xmath1 includes both self - consistency constraints ( which are not strictly speaking `` data '' ) and kinematic constraints or only kinematic constraints . when we use both types of constraints and all the available kinematical data ( from low - resolution long - slit data and high resolution nifs data ) the best - fit solution is @xmath256 and @xmath175 , for a model with @xmath101 . however , this best fit model gives a central @xmath231 that is too low to fit the data obtained with the nifs instrument . when we use @xmath127 ( determined by only considering the fit to the kinematic data ) , the best fit model has @xmath257 and @xmath258 , and provides a better fit to the nuclear kinematical data from nifs spectrograph , although it gives a slightly worse fit to the mass and surface brightness distributions . an interesting point worth noting is that @xcite found a best - fit value of @xmath259 in ngc 4151 when they fitted the kinematics of the h@xmath260 line - emitting gas within 1 of the center ( see figure 43 in their paper ) . however , if they included the kinematics of gas within 2 they obtained a larger value of @xmath261 , although with a somewhat worse reduced-@xmath1 . it is intriguing that both our stellar dynamical modeling and their gas dynamical modeling , while preferring lower values of @xmath262 , also suggest that _ higher _ values of @xmath0 may be obtained when lower spatial resolution data are included in the fit . models are generated for three different inclination angles of the bulge @xmath263 and show that the best - fit @xmath0 and @xmath95 values are relatively insensitive to inclination angle . we examine an additional 4 inclination angles for fixed bh mass values ( @xmath264 ) and fixed mass - to - light ratio ( @xmath265 ) . we find that edge - on models ( @xmath101 ) give the smallest @xmath1 values , while models with the inclination of the large scale disk ( @xmath103 ) are strongly disfavored . this suggests that the bulge must be nearly spherical but its rotation axis is significantly misaligned from the rotation axis of the disk . however , it is well known from previous work @xcite that inclination of a spheroid is extremely difficult to determine via dynamical modelling , especially in the presence of a bar . this determination of the inclination of the bulge should therefore be treated with caution . a detailed examination of the kinematics in the inner 5 of ngc 4151 shows evidence for bar kinematics , which manifest as high values of @xmath231 , low or negative values of @xmath67 and @xmath66 values which are correlated with @xmath236 . we hypothesize that the discrepancy between the @xmath1 obtained from including all the constraints and that obtained by only including the kinematic constraints is likely a consequence of two factors : ( a ) a stellar bar , which is not possible to model with our axisymmetric code , and ( b ) inconsistencies between the velocity dispersion values obtained from the low - spatial resolution long - slit data and the high spatial resolution nifs data over the same spatial range . this hypothesis was tested by fitting only the nifs data . when we fit only the kinematical data obtained with the nifs spectrograph ( neglecting all kinematical constraints beyond @xmath266 - i.e those data that show kinematical evidence for a bar and data that is discrepant with high - resolution nifs data ) , the non - negative optimization problem was able to find a much better fit to the nifs data and the same best - fit @xmath0 was obtained by both methods of computing @xmath1 . the stellar - dynamical modeling carried out in this paper gives a best - fit value of @xmath267 ( obtained by averaging over the @xmath125 and @xmath127 values in the lower two rows of table [ tab : robustbestfit ] ) which is consistent at the 1@xmath4 level with the reverberation mapping - based mass of @xmath268 ( 1@xmath4 errors ) obtained by using data from @xcite , but relying on a recently updated empirical calibration of the rm mass scale by @xcite . the best fit m / l ratio @xmath269 ( 3@xmath4 error ) is consistent with the photometrically derived mass - to - light ratio of @xmath270 . the new nifs data yields @xmath254 , the velocity dispersion within @xmath271 , which is @xmath1225% larger than previous values , while the new best estimate of @xmath272 17% smaller than but fully consistent with both our previous stellar dynamical upper limit @xcite and the dynamical estimate based on h@xmath260 line - emitting gas @xcite . the larger value of @xmath245 and smaller value of @xmath0 help to bring ngc 4151 closer to the best - fit relation of @xcite . the analysis in this paper demonstrates that biased estimates of bh masses can arise when an axisymmetric orbit superposition code is used to model a galaxy with a weak , but kinematically identifiable barred galaxy , possibly resulting in an over - estimate in @xmath0 if the m / l ratio is constrained primarily by data beyond the sphere - of - influence of the bh . this confirms the prediction made from the recent analysis of @xmath229-body simulations of barred galaxies with bhs by @xcite , and points to the need for new dynamical modeling tools capable of modeling a stellar bar . when such codes are applied to the existing sample of barred galaxies , our results suggest that it will likely enhance the discrepancy between barred and unbarred galaxies in the @xmath15 relation @xcite . our modeling of this complex system suggests that the standard practice of fitting kinematical constraints over a large range of radii with a constant mass - to - light ratio ( for exceptions to this practice , see * ? ? ? * ; * ? ? ? * ) can bias the mass of the central bh and the derived mass - to - light ratio . to some extent , the biases can be overcome by only considering the kinematics very close to the central bh , although this gives larger errors on the estimated solutions of @xmath0 and @xmath95 . our tests of robustness demonstrate that the use of such high spatial resolution nuclear kinematical data in such modeling is very valuable . however , the complications of the stellar dynamical bar imply that additional work is required in order to perform the crucial test of the rm mass scale calibration in this galaxy . only with improved non - axisymmetric modeling methods and/or bh mass measurements in other reverberation - mapped agns will we ultimately be able to assess whether the growth history of the bhs that we see accreting at present is systematically different from those currently in quiescence . we thank remco van den bosch for valuable input and for providing his symmetrization routine . we thank gaelle dumas and eric emsellem for providing us with the sauron integral field kinematics . we thank linda watson for providing copies of her nifs stellar velocity template spectra . we thank the anonymous referee for suggestions that helped to improve our manuscript . we are grateful to michele cappellari for making his analysis routines publicly available and for his continuing efforts in adding software features . m. valluri and jb were supported by national science foundation grant ast-0908346 , and m. valluri acknowledges support from the university of michigan s elizabeth crosby award . bmp and rwp are grateful to the national science foundation for support of this work through grant ast-1008882 to the ohio state university . mb gratefully acknowledges support by the national science foundation under grant ast-1253702 . m. vestergaard acknowledges support from a freja fellowship granted by the dean of the faculty of natural sciences at the university of copenhagen and a marie curie international incoming fellowship . the research leading to these results has received funding from the people programme ( marie curie actions ) of the european union s seventh framework programme fp7/2007 - 2013/ under rea grant agreement no . 300553 ( mv ) . the dark cosmology centre is funded by the danish national research foundation . based on observations obtained at the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the science and technology facilities council ( united kingdom ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , ministrio da cincia e tecnologia ( brazil ) and ministerio de ciencia , tecnologa e innovacin productiva ( argentina ) . this research was supported in part through computational resources and services provided by advanced research computing and the flux cluster at the university of michigan , ann arbor . this work made use of data from the ohio state university bright spiral galaxy survey , which was funded by grants ast-9217716 and ast-9617006 from the united states national science foundation , with additional support from the ohio state university . based on observations made with the nasa / esa hubble space telescope , obtained from the data archive at the space telescope institute . stsci is operated by the association of universities for research in astronomy , inc . under the nasa contract nas 5 - 26555 . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . this research has made use of the simbad database , operated at cds , strasbourg , france . , m. c. , denney , k. d. , cackett , e. m. , dietrich , m. , fogel , j. k. j. , ghosh , h. , horne , k. , kuehn , c. , minezaki , t. , onken , c. a. , peterson , b. m. , pogge , r. w. , pronik , v. i. , richstone , d. o. , sergeev , s. g. , vestergaard , m. , walker , m. g. , & yoshii , y. 2006 , , 651 , 775 , m. c. , walsh , j. l. , barth , a. j. , baliber , n. , bennert , v. n. , canalizo , g. , filippenko , a. v. , ganeshalingam , m. , gates , e. l. , greene , j. e. , hidas , m. g. , hiner , k. d. , lee , n. , li , w. , malkan , m. a. , minezaki , t. , sakata , y. , serduke , f. j. d. , silverman , j. m. , steele , t. n. , stern , d. , street , r. a. , thornton , c. e. , treu , t. , wang , x. , woo , j .- h . , & yoshii , y. 2009 , , 705 , 199 , m. c. , walsh , j. l. , barth , a. j. , yoshii , y. , woo , j .- h . , wang , x. , treu , t. , thornton , c. e. , street , r. a. , steele , t. n. , silverman , j. m. , serduke , f. j. d. , sakata , y. , minezaki , t. , malkan , m. a. , li , w. , lee , n. , hiner , k. d. , hidas , m. g. , greene , j. e. , gates , e. l. , ganeshalingam , m. , filippenko , a. v. , canalizo , g. , bennert , v. n. , & baliber , n. 2010 , , 716 , 993 , g. , de vaucouleurs , a. , corwin , jr . , h. g. , buta , r. j. , paturel , g. , & fouqu , p. 1991 , third reference catalogue of bright galaxies . volume i : explanations and references . volume ii : data for galaxies between 0@xmath273 and 12@xmath273 . volume iii : data for galaxies between 12@xmath273 and 24@xmath273 . ( springer - verlag : new york ) , p. b. , frogel , j. a. , pogge , r. w. , quillen , a. c. , berlind , a. a. , davies , r. l. , depoy , d. l. , gilbert , k. m. , houdashelt , m. l. , kuchinski , l. e. , ramrez , s. v. , sellgren , k. , stutz , a. , terndrup , d. m. , & tiede , g. p. 2002 , , 143 , 73 , k. , richstone , d. , tremaine , s. , lauer , t. r. , bender , r. , bower , g. , dressler , a. , faber , s. m. , filippenko , a. v. , green , r. , grillmair , c. , ho , l. c. , kormendy , j. , magorrian , j. , & pinkney , j. 2003 , , 583 , 92 , c. l. , merritt , d. , olling , r. , valluri , m. , bender , r. , bower , g. , danks , a. , gull , t. , hutchings , j. , kaiser , m. e. , maran , s. , weistrop , d. , woodgate , b. , malumuth , e. , nelson , c. , plait , p. , & lindler , d. 2001 , , 550 , 668 , p .- y . , cappellari , m. , emsellem , e. , bournaud , f. , michel - dansac , l. , alatalo , k. , blitz , l. , bois , m. , bureau , m. , davies , r. l. , davis , t. a. , de zeeuw , p. t. , duc , p .- a . , khochfar , s. , krajnovi , d. , kuntschner , h. , morganti , r. , mcdermid , r. m. , naab , t. , oosterloo , t. , sarzi , m. , scott , n. , serra , p. , weijmans , a .- m . , & young , l. m. 2012 , , 424 , 1495 , p. j. , hart , j. , conroy , p. g. , pfitzner , m. l. , bloxham , g. j. , jones , d. j. , downing , m. d. , dawson , m. , young , p. , jarnyk , m. , & van harmelen , j. 2003 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 4841 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . m. iye & a. f. m. moorwood , 15811591 , c. a. , valluri , m. , peterson , b. m. , pogge , r. w. , bentz , m. c. , ferrarese , l. , vestergaard , m. , crenshaw , d. m. , sergeev , s. g. , mchardy , i. m. , merritt , d. , bower , g. a. , heckman , t. m. , & wandel , a. 2007 , , 670 , 105 ( paper i ) , s. , gebhardt , k. , bender , r. , bower , g. , dressler , a. , faber , s. m. , filippenko , a. v. , green , r. , grillmair , c. , ho , l. c. , kormendy , j. , lauer , t. r. , magorrian , j. , pinkney , j. , & richstone , d. 2002 , , 574 , 740 , j .- h . , treu , t. , barth , a. j. , wright , s. a. , walsh , j. l. , bentz , m. c. , martini , p. , bennert , v. n. , canalizo , g. , filippenko , a. v. , gates , e. , greene , j. , li , w. , malkan , m. a. , stern , d. , & minezaki , t. 2010 , , 716 , 269 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: [ [ stress - fluctuation - formalism . ] ] stress fluctuation formalism . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + among the fundamental properties of any solid material are its elastic constants @xcite . they describe the macroscopic response to weak external deformations and encode information about the potential landscape of the system . in their seminal work @xcite squire , holt and hoover derived an expression for the isothermal elastic constants extending the classical born theory @xcite to canonical ensembles of imposed particle number @xmath11 , volume @xmath12 and _ finite _ ( mean ) temperature @xmath5 . they found a correction term @xmath13 to the born expression for the elastic constants which involves the mean - square fluctuations of the stress @xcite @xmath14 with @xmath15 being the inverse temperature , @xmath16 boltzmann s constant , @xmath17 the instantaneous value of a component of the stress tensor @xmath18 and greek letters denoting the spatial coordinates in @xmath19 dimension . perhaps due to the fact that the demonstration of the stress fluctuation formalism " in ref . @xcite assumed explicitly a well - defined reference position and a displacement field for the particles , it has not been appreciated that this approach is consistent ( at least if the initial pressure is properly taken into account ) with the well - known pressure fluctuation formula for the compression modulus @xmath20 of isotropic fluids @xcite which has been derived several decades earlier , presumably for the first time in the late 1940s by rowlinson @xcite . in any case , the stress fluctuation formalism provides a convenient route to calculating the elastic properties in computer simulations which has been widely used in the past @xcite . it has been generalized to systems with nonzero initial stress @xcite , hard - sphere interactions @xcite and arbitrary continuous potentials @xcite , or to the calculation of local mechanical properties @xcite . [ [ low - temperature - limit - and - non - affine - displacements . ] ] low - temperature limit and non - affine displacements . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in particular from ref . @xcite it has become clear that the stress fluctuation correction to the leading born term does not necessarily vanish in the zero - temperature limit . this is due to the fact that when a system is subjected to a homogeneous deformation , the ensuing particle displacements need not follow the macroscopic strain affinely @xcite . the stress fluctuation term quantifies the extent of these non - affine displacements , whereas the born term reflects the affine part of the particle displacements . how important the non - affine motions are , depends on the system under consideration @xcite . while the elastic properties of crystals with one atom per unit cell are given by the born term only , stress fluctuations are significant for crystals with more complex unit cells @xcite . they become particularly pronounced for polymer like soft materials @xcite and amorphous solids @xcite . this is revealed by recent simulation studies of various glass formers @xcite , like lennard - jones ( lj ) mixtures @xcite , polymer glasses @xcite or silica melts @xcite . many of these simulation studies concentrate on the mechanical properties deep in the glassy state , exploring for instance correlations between the non - affine displacement field and vibrational anomalies of the glass ( boson peak " ) @xcite , the mechanical heterogeneity at the nanoscale @xcite or the onset of molecular plasticity in the regime where the macroscopic deformation is still elastic @xcite . [ [ shear - stress - and - shear - modulus . ] ] shear stress and shear modulus . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + surprisingly , the temperature dependence of the elastic constants of glassy materials does not appear to have been investigated extensively using the stress fluctuation formalism @xcite . following the pioneering work by barrat _ @xcite , we present in this work such a study where we focus on the behavior of the shear modulus @xmath0 of two isotropic colloidal glass - formers in @xmath1 and @xmath2 dimensions sampled , respectively , by means of molecular dynamics ( md ) and monte carlo ( mc ) simulations @xcite . since we are interested in isotropic systems one can avoid the inconvenient tensorial notation of the full stress fluctuation formalism as shown in fig . [ fig_sketch ] . focusing on the shear in the @xmath21-plane we use @xmath22 for the mean shear stress and @xmath23 for its instantaneous value . as an important contribution to the shear modulus we shall monitor the shear stress fluctuations @xmath24 as a function of temperature @xmath5 . quite generally , the shear modulus @xmath0 may then be computed according to the stress fluctuation formula @xcite @xmath25 where @xmath26 stands for the so - called affine shear elasticity " corresponding to an assumed _ affine _ response to an external shear strain @xmath3 . for pairwise interaction potentials it can be shown that @xmath27 i.e. @xmath26 comprises the so - called born - lam coefficient " @xmath28 , a simple moment of the first and the second derivative of the pair potential with respect to the distance of the interacting particles @xcite , and the excess contribution @xmath29 to the total pressure @xmath6 . these notations will be properly defined in sec . [ theo_stressfluctu ] below where a simple derivation will be given . since eq . ( [ eq_gmodtt ] ) is not the only operational definition of the shear modulus discussed in this work , an index has been added reminding that the shear stress @xmath30 is used for the determination of @xmath0 . [ [ physical - motivation - and - systems - of - interest . ] ] physical motivation and systems of interest . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + if an infinitesimal shear stress @xmath4 is applied at time @xmath31 , an isotropic solid exhibits , quite generally , an elastic response quantified in strength by a finite static ( @xmath32-independent ) shear modulus ( @xmath33 ) as shown by the top curve in fig . [ fig_sketch](b ) , whereas a liquid has a vanishing modulus ( @xmath34 ) and flows on long time scales as shown by the bottom curve . the shear modulus @xmath0 can thus serve as an order parameter " distinguishing liquid and solid states @xcite . upon melting , the shear modulus of crystalline solids is known to vanish discontinuously with @xmath5 . a natural question which arises is that of the behavior of @xmath35 for amorphous solids and glasses in the vicinity of the glass transition temperature @xmath9 . interestingly , qualitative different theoretical predictions have been put forward by mode - coupling theory ( mct ) @xcite and some versions of replica theory @xcite predicting a _ discontinuous jump _ at the glass transition whereas other versions of replica theory @xcite suggest a _ continuous _ transition in the static limit of large sampling times @xmath32 . ( see ref . @xcite for a topical recent discussion of the replica approach . ) the numerical characterization of @xmath35 for colloidal glass - former is thus of high current interest . since the stress fluctuation formalism can readily be added to the standard simulation codes of colloidal glass - formers @xcite at imposed particle number @xmath11 , box volume @xmath12 , box shape ( @xmath36 ) and temperature @xmath5 , called @xmath37-ensembles , this suggests the use of eq . ( [ eq_gmodtt ] ) to tackle this issue . note that the latter stress fluctuation relation should also be useful for wide range of related systems , beyond the scope of the present work , including the glass transition of polymeric liquids @xcite , colloidal gels @xcite and self - assembled networks @xcite as observed experimentally for hyperbranched polymer chains with sticky end - groups @xcite or bridged networks of telechelic polymers in water - oil emulsions @xcite . [ [ validity - through - a - solid - liquid - transition . ] ] validity through a solid - liquid transition . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + this begs the delicate question of whether the general stress fluctuation formalism and especially eq . ( [ eq_gmodtt ] ) for the shear modulus only holds for elastic solids with a well - defined displacement field , as suggested in the early literature @xcite , or if the approach may actually be also applied _ through _ a solid - liquid transition up to the liquid state @xcite with @xmath38 and thus @xmath39 as it should for a liquid . by reworking the theoretical arguments leading to eq . ( [ eq_gmodtt ] ) it will be seen that this is in principal the case , at least if standard thermodynamics can be assumed to apply . the latter point is indeed not self - evident for the colloidal glasses we are interested in where _ ( i ) _ some degrees of freedom are frozen on the time scale probed and _ ( ii ) _ the measured response to an imposed infinitesimal shear @xmath4 may depend on the sampling time @xmath32 , especially for temperatures @xmath5 around @xmath9 as shown in fig . [ fig_sketch](b ) . to be on the safe side , it thus appears to be appropriate to test numerically the applicability of the stress fluctuation formula , eq . ( [ eq_gmodtt ] ) , using the corresponding strain fluctuation relations which are more fundamental on experimental grounds . [ [ strain - fluctuation - relations . ] ] strain fluctuation relations . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + panel ( a ) of fig . [ fig_sketch ] shows a schematic setup for probing experimentally the shear modulus @xmath0 of isotropic systems confined between two rigid walls . we assume that the bottom wall is fixed and the particle number @xmath11 , the ( mean ) normal pressure @xmath6 and the ( mean ) temperature @xmath5 are kept constant . in linear response either a shear strain @xmath40 or a ( mean ) shear stress @xmath4 may be imposed . the thin solid line indicates the original unstrained reference system with @xmath36 and @xmath41 , the dashed line the deformed state . from both the mechanical and the thermodynamical point of view , the shear modulus is defined as @xmath42 where the derivative may also be taken at @xmath41 . in practice , one obtains @xmath0 by fitting a linear slope to the @xmath43-data at vanishing strain . obviously , the noise level is normally large for small strains . from the numerical point of view it would thus be more appropriate to work at imposed mean stress @xmath41 , to let the strain freely fluctuate and to sample the _ instantaneous _ values @xmath44 . the shear modulus @xmath0 is then simply obtained by linear regression @xmath45 where the index indicates that both the instantaneous values @xmath46 and @xmath30 are used . ( note that imposing the strain @xmath3 also fixes the instantaneous strain @xmath46 which thus can not fluctuate . ) the associated dimensionless correlation coefficient @xmath47 allows to determine the quality of the fit . we do not know whether eq . ( [ eq_gmodgt ] ) and eq . ( [ eq_cmodgt ] ) have actually been used in a real experiment , however since no difficulty arises to work in a computer simulation at constant @xmath4 and to record @xmath44 , we use this linear regression as our key operational definition of @xmath0 which is used to judge the correctness of the stress fluctuation relation , eq . ( [ eq_gmodtt ] ) , computed at constant @xmath3 . to avoid additional physics at the walls , we use of course periodic boundary conditions @xcite ( the central image being sketched ) as shown by the panels on the right - hand side of fig . [ fig_sketch ] . as discussed in sec . [ theo_legendre ] , it is possible to determine @xmath48 at constant volume ( @xmath49-ensemble ) as shown by panel ( c ) or constant normal pressure ( @xmath50-ensemble ) as shown by panel ( e ) . the latter ensemble has the advantage that one can easily impose in addition the same normal pressure @xmath6 for all temperatures @xmath5 as it is justified for experimental reasons . if thermodynamics holds , it is known ( sec . [ theo_strainfluctu ] ) that @xmath51 should hold @xcite . using eq . ( [ eq_gmodgt ] ) this implies a second operational definition @xmath52 which has the advantage that only the instantaneous shear strain @xmath46 has to be recorded . the main technical point put forward in this paper is that all operational definitions yield similar results @xmath53 even for temperatures where a strong dependence on the sampling time @xmath32 is observed . in other words , eq . ( [ eq_gmods ] ) states that this time dependence applies similarly ( albeit perhaps not exactly ) to the various moments computed for the three operational definitions of @xmath0 in different ensembles . [ [ outline . ] ] outline . + + + + + + + + the paper is organized as follows : several theoretical issues are regrouped in sec . [ sec_theo ] . we begin the discussion in sec . [ theo_strainfluctu ] by reminding a few useful general thermodynamic relations . a simple derivation of the stress fluctuation formula , eq . ( [ eq_gmodtt ] ) , for the shear modulus @xmath54 in constant-@xmath3 ensembles is given in sec . [ theo_stressfluctu ] . ( the analogous stress fluctuation formula for the compression modulus @xmath20 in constant-@xmath12 ensembles @xcite is rederived in appendix [ app_kstressfluctu ] . ) shear stress fluctuations @xmath55 in ensembles with different boundary conditions are discussed in sec . [ theo_legendre ] . for constant-@xmath4 ensembles @xmath56 will be shown to reduce to the affine shear elasticity @xmath26 , i.e. the functional @xmath54 must vanish for all temperatures . as pointed out recently @xcite , the truncation of the interaction potentials used in computational studies implies an impulsive correction for the affine shear elasticity @xmath26 . this is summarized in sec . [ theo_trunc ] . the numerical models used in this study are presented in sec . [ sec_algo ] . we turn then in sec . [ sec_res ] to our main numerical results . starting with the high-@xmath5 liquid limit , sec . [ res_thigh ] shows that all our operational definitions correctly yield a vanishing shear modulus . moving to the opposite low-@xmath5 limit we present in sec . [ res_dynmat ] the elastic response for a topologically fixed network of harmonic springs constructed from the dynamical matrix " of our 2d model glass system @xcite . as shown in sec . [ res_tlow ] , the same behavior is found qualitatively for low-@xmath5 glassy bead systems . the temperature dependence of stress fluctuations and shear moduli in different ensembles is then discussed in sec . [ res_tvari ] . as stated above , one central goal of this work is to determine the behavior of the shear modulus around @xmath9 . we conclude the paper in sec . [ sec_conc ] . thermodynamic relations and numerical findings related to the compression modulus @xmath20 are regrouped in appendix [ app_k ] . we assume in the following that standard thermodynamics and thermostatistics @xcite can be applied and remind a few useful relations . let us denote an extensive variable by @xmath57 and its instantaneous value by @xmath58 , the conjugated intensive variable by @xmath59 and its instantaneous value by @xmath60 @xcite . with @xmath61 being the free energy at temperature @xmath5 and imposed @xmath57 , the intensive variable @xmath59 and the associated modulus @xmath62 are given by @xmath63 alternatively , one may consider the legendre transformation @xmath64 for which @xmath65 in our case , the extensive variable is @xmath66 , the intensive variable @xmath67 @xcite and the modulus @xmath68 @xcite . we emphasize that there is an important difference between extensive and intensive variables : if the extensive variable @xmath57 is fixed , @xmath58 can not fluctuate . if , on the other hand , the intensive variable @xmath69 is imposed , not only @xmath58 but also @xmath60 may fluctuate @xcite . in our case this means , that @xmath70 for the @xmath37-ensemble , while the shear stress fluctuations @xmath55 defined above do clearly not vanish in the @xmath49-ensemble shown in fig . [ fig_sketch](d ) . assuming @xmath59 to be imposed , it is well known that @xcite @xmath71 with @xmath15 being the inverse temperature . for our variables @xmath72 and @xmath73 , eq . ( [ eq_dxdi ] ) implies eq . ( [ eq_dgamdtau ] ) from which it is seen that eq . ( [ eq_gmodgt ] ) is consistent with eq . ( [ eq_gmodgg ] ) . the latter relation is also obtained directly using eq . ( [ eq_dxdx ] ) . as discussed in textbooks @xcite a simple average @xmath74 of some observable @xmath75 does not depend of the chosen ensemble , at least not if the system is large enough ( @xmath76 ) . a correlation function @xmath77 of two observables @xmath75 and @xmath78 may differ , however , depending on whether @xmath57 or @xmath59 are imposed . as demonstrated by lebowitz , percus and verlet in 1967 @xcite one verifies that @xmath79 if the extensive variable and at least one of the observables are identical , the left - hand side of eq . ( [ eq_dadb ] ) vanishes : if @xmath80 and @xmath81 , this yields eq . ( [ eq_dxdi ] ) . if @xmath82 , this implies eq . ( [ eq_dxdx ] ) . more importantly , for @xmath83 one obtains using eq . ( [ eq_ftom ] ) @xmath84 if it is possible to probe the intensive variable fluctuations in both ensembles , this does in principle allow to determine the modulus @xmath62 . ( the latter equation might actually be taken as the definition of @xmath62 in cases where the thermodynamic reasoning becomes dubious . ) we emphasize that the modulus @xmath62 is an intrinsic property of the system and does _ not _ depend on the ensemble , albeit it is determined using ensemble depending correlation functions . for a thermodynamically stable system , @xmath85 , eq . ( [ eq_didi ] ) implies that @xmath86 and that both fluctuations must become similar in the limit where the modulus @xmath62 becomes small . for the shear stress fluctuations @xmath55 , eq . ( [ eq_didi ] ) corresponds to the transformation @xmath87 and one thus expects @xmath55 to be larger for @xmath49-systems than for @xmath37-systems , while in the liquid limit where @xmath88 the imposed boundary conditions should not matter . we shall verify eq . ( [ eq_muftransform ] ) numerically below . a glance at the stress fluctuation formula , eq . ( [ eq_gmodtt ] ) , suggests the identity @xmath89 assuming eq . ( [ eq_dgamdtau ] ) and eq . ( [ eq_muamuftau ] ) , it follows for the correlation coefficient @xmath90 defined above that @xmath91 if the measured @xmath44 were indeed perfectly correlated , i.e. @xmath92 , this implies @xmath93 and thus @xmath94 . in fact , for all situations studied here we always have @xmath95 i.e. the affine shear elasticity sets only an upper bound to the shear modulus and a theory which only contains the affine strain response must _ overpredict _ @xmath0 . we show now directly that eq . ( [ eq_gmodtt ] ) and , hence , eqs . ( [ eq_muamuftau],[eq_cmodgt_mua],[eq_gbound ] ) indeed hold and this under quite general conditions . [ [ introduction . ] ] introduction . + + + + + + + + + + + + + as shown in fig . [ fig_sketch](a ) , we demonstrate eq . ( [ eq_gmodtt ] ) from the free energy change associated with an imposed arbitrarily small pure shear strain @xmath3 in the @xmath21-plane assuming a constant particle number @xmath11 , a constant volume @xmath12 and a constant temperature @xmath5 ( @xmath37-ensemble ) . it is supposed that the total system hamiltonian may be written as the sum of a kinetic energy and a potential energy . since for a plain shear strain at constant volume the ideal free energy contribution does not change , i.e. is irrelevant for @xmath4 and @xmath0 , we may focus on the excess free energy contribution @xmath96 due to the conservative interaction energy of the particles . the ( excess ) partition function @xmath97 of the unperturbed system at @xmath36 is the boltzmann - weighted sum over all states @xmath98 of the system which are _ accessible _ within the measurement time @xmath32 . the argument @xmath99 is a short - hand for the unstrained reference . following the derivation of the compression modulus @xmath20 by rowlinson @xcite the partition function @xmath100 of the sheared system is supposed to be the sum over the _ same _ states @xmath98 , but with a different _ metric _ @xcite corresponding to the macroscopic strain which changes the total interaction energy @xmath101 of state @xmath98 and , hence , the weight of the sheared configuration for the averages computed . this is the central hypothesis made . interestingly , it is not necessary to specify explicitly the states of the unperturbed or perturbed system , e.g. , it is irrelevant whether the particles are distinguishable or not or whether they have a well - defined reference position for defining a displacement field . [ [ shear - stress - and - modulus - for - general - potential . ] ] shear stress and modulus for general potential . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + assuming eq . ( [ eq_fexgamma ] ) to hold we compute now the mean shear stress @xmath4 and the shear modulus @xmath0 for a general interaction potential @xmath101 . we note for later convenience that @xmath102 for the derivatives of the free energy and @xmath103 for the derivatives of the excess partition function where a prime denotes the derivative of a function @xmath104 with respect to its argument @xmath105 . using eq . ( [ eq_dzdgam ] ) and taking finally the limit @xmath106 one verifies for the shear stress that @xmath107 _ defining _ the instantaneous shear stress @xcite . note that the average taken is defined as @xmath108 using the weights of the unperturbed system . the shear stress thus measures the average change of the total interaction energy @xmath101 taken at @xmath36 . the shear modulus @xmath0 is obtained using in addition eq . ( [ eq_ddlogzddgam ] ) and eq . ( [ eq_ddzddgam ] ) and taking finally the @xmath106 limit . confirming thus the stress fluctuation formula eq . ( [ eq_gmodtt ] ) stated in the introduction , this yields @xmath109 with @xmath26 being the affine shear elasticity " @xmath110 already mentioned above and @xmath55 as defined in eq . ( [ eq_muf ] ) . the comparison of eq . ( [ eq_gstressfluctu ] ) and eq . ( [ eq_muftransform ] ) confirms eq . ( [ eq_muamuftau ] ) . [ [ comment . ] ] comment . + + + + + + + + the affine shear elasticity @xmath26 corresponds to the change ( second derivative ) of the total energy which would be obtained if one actually strains _ affinely _ in a computer simulation a given state @xmath98 without allowing the particles to relax their position . as shown for athermal ( @xmath111 ) amorphous bodies @xcite , the positions of the particles of such a strained configuration will of course in general change slightly to minimize the interaction energy relaxing thus the elastic moduli . this is also of relevance for thermalized solids where the non - affine displacements of the particles are driven by the minimization of the free energy . it is for this reason that the shear - stress fluctuation term @xmath112 must occur in eq . ( [ eq_gstressfluctu ] ) correcting the _ overprediction _ of the shear modulus @xmath0 by the affine shear elasticity @xmath26 , eq . ( [ eq_gbound ] ) . this point has been overlooked in the early literature @xcite and only appreciated much later @xcite as discussed in barrat s review @xcite . interestingly , as has been shown by lutsko @xcite , @xmath55 and other similarly defined stress fluctuations become temperature independent in the harmonic ground state approximation for @xmath111 . probing the stress fluctuations in a low - temperature simulation allows thus to determine the elastic moduli of athermal solids . please note that up to now we have not used explicitly the coordinate transformations ( metric change ) associated with an affine shear strain . this is needed to obtain operational definitions for the instantaneous shear stress @xmath30 ( and thus for @xmath4 and @xmath55 ) and the affine shear elasticity . [ [ coordinate - transformation . ] ] coordinate transformation . + + + + + + + + + + + + + + + + + + + + + + + + + + as shown in fig . [ fig_sketch](a ) , we assume a pure shear strain which transforms the @xmath105-coordinate of a particle position or the distance between two particles as @xmath113 leaving all other coordinates unchanged @xcite . the squared distance @xmath114 between two particles thus transforms as @xmath115 where we need to keep the @xmath116-term for calculating the shear modulus below . we note for later reference that this implies @xmath117 for , respectively , the first and the second derivative with respect to @xmath3 . let us now consider an arbitrary function @xmath118 . using eq . ( [ eq_strain_rone ] ) it is readily seen that its first derivative with respect to @xmath3 may be written as @xmath119 in the second step we have introduced the components @xmath120 and @xmath121 of the normalized distance vector between both particles . more generally , we denote by @xmath122 the @xmath123-component of a normalized vector with @xmath124 stating only the small-@xmath3 limit for the second derivative of @xmath118 with respect to @xmath3 we note finally that @xmath125 where we have dropped the argument @xmath99 for the distance @xmath126 of the unperturbed reference system . [ [ pair - interaction - potentials . ] ] pair interaction potentials . + + + + + + + + + + + + + + + + + + + + + + + + + + + + we assume now that the interaction energy @xmath101 of a configuration @xmath98 is given by a pair interaction potential @xmath127 between the particles @xcite with @xmath128 being the distance between two particles , i.e. @xmath129 where the index @xmath130 labels the interaction between the particles @xmath131 and @xmath132 with @xmath133 . using the general result eq . ( [ eq_tauhat_us ] ) for the shear stress it follows using eq . ( [ eq_strain_fronea ] ) that @xmath134 where we have dropped the argument @xmath99 in the @xmath106 limit . we have thus rederived for the shear stress ( @xmath135 ) the well - known kirkwood expression for the general excess stress tensor @xcite @xmath136 the affine shear elasticity @xmath26 is obtained using eq . ( [ eq_muadef ] ) and eq . ( [ eq_strain_frtwo ] ) . this yields @xmath137 with the first contribution @xmath28 being the so - called born - lam coefficient " @xcite @xmath138 it corresponds to the first term in eq . ( [ eq_strain_frtwo ] ) . the second contribution @xmath139 stands for the normal excess stress in the @xmath140-direction , i.e. the @xmath141 component of the excess stress tensor . [ [ isotropic - systems . ] ] isotropic systems . + + + + + + + + + + + + + + + + + + in this work we focus on isotropic materials under isotropic external loads . the normal ( excess ) stresses @xmath142 for all directions @xmath123 are thus identical , i.e. @xmath143 with @xmath144 being the excess pressure , @xmath6 the total pressure , @xmath145 the ideal gas pressure and @xmath19 the spatial dimension . this implies that the affine shear elasticity is given by @xmath146 as already stated in the introduction , eq . ( [ eq_muamubpex ] ) . we note that for isotropic @xmath19-dimensional systems it can be shown @xcite that the born - lam coefficient can be simplified as @xmath147 the @xmath19-dependent prefactor stems from the assumed isotropy of the system and the mathematical formula @xcite @xmath148 ( @xmath149 being the kronecker symbol @xcite ) for the components of a unit vector in @xmath19 dimensions pointing into arbitrary directions . by comparing the excess pressure @xmath29 , eq . ( [ eq_sigaapex ] ) , with the underlined second contribution to the born coefficient in eq . ( [ eq_mub ] ) this implies that @xmath150 it is thus inconsistent to neglect the explicit excess pressure in eq . ( [ eq_muamubpex ] ) while keeping the second term of the born - lam coefficient @xmath28 . this approximation is only justified if the excess pressure @xmath29 is negligible ( and not the total pressure @xmath6 ) . [ [ different - ensembles . ] ] different ensembles . + + + + + + + + + + + + + + + + + + + + being simple averages neither @xmath29 , @xmath26 or @xmath28 does depend for sufficiently large systems on the chosen ensemble , i.e. irrespective of whether @xmath3 or @xmath4 is imposed one expects to obtain the same values . as already reminded in sec . [ theo_strainfluctu ] , this is different for fluctuations in general @xcite . for instance , it is obviously pointless to use the shear - strain fluctuation formulae @xmath48 or @xmath151 in a constant-@xmath3 ensemble . a more interesting result is predicted if @xmath152 is computed in the wrong " constant-@xmath4 ensemble . since @xmath153 , eq . ( [ eq_muftransform ] ) , one actually expects to observe @xmath154 for all temperatures @xmath5 if our thermodynamic reasoning holds through the glass transition up to the liquid state . we will test numerically this non - trivial prediction in sec . [ sec_res ] . up to now we have only considered for simplicity of the presentation @xmath37- and @xmath49-ensembles at constant volume @xmath12 . since on general grounds pure deviatoric and dilatational ( volumetric ) strains are decoupled ( both strains commute ) , one expects the observable @xmath54 also to be applicable in the @xmath155-ensemble shown in fig . [ fig_sketch](d ) and the observables @xmath48 and @xmath151 to be applicable in the @xmath50-ensemble as sketched in fig . [ fig_sketch](e ) . in the latter ensemble one also expects to find @xmath156 and @xmath153 . this will also be tested below . [ [ direct - derivation - of - ensuremathmu_mathrmf - for - imposed - tau . ] ] direct derivation of @xmath55 for imposed @xmath4 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we demonstrate now directly eq . ( [ eq_muamuftau ] ) for the shear stress fluctuations using the general mathematical identity @xmath157 with @xmath105 being an _ unconstrained _ coordinate , @xmath158 some property , @xmath159 a force " with respect to some energy " @xmath160 and the average being boltzmann weighted , i.e. @xmath161 . ( it is assumed in eq . ( [ eq_intbyparts ] ) that @xmath160 decays sufficiently fast at the boundaries . ) following work by zwanzig @xcite we express first the instantaneous shear stress @xmath30 given above by the kirkwood formula , eq . ( [ eq_tauhatkirkwood ] ) , by the alternative virial representation @xcite @xmath162 the second term , sometimes called the inner virial " , stands for a sum over all particles @xmath131 with @xmath163 being the @xmath140-coordinate of the particle and @xmath164 the @xmath105-coordinate of the force acting on the particle . this contribution indeed vanishes on average as can be seen using eq . ( [ eq_intbyparts ] ) , @xmath165 which shows that @xmath166 as it must . the stress fluctuations may thus be expressed as @xmath167 where we have used the integration by part , eq . ( [ eq_tauhatfluctu ] ) . this step requires that all particle positions are unconstrained and independent ( generalized ) coordinates . note that this is possible at imposed @xmath4 , but can not hold at fixed @xmath3 . assuming then pair interactions and using eq . ( [ eq_tauhatkirkwood ] ) one can reformulate eq . ( [ eq_stressfluctuvirial ] ) within a few lines . without further approximation this yields @xmath168 where the sum runs now over all interactions @xmath130 between particles @xmath133 . @xmath169 and @xmath170 refer to components of the distance vector of the interaction @xmath130 and @xmath171 to the @xmath105-component of the central force @xmath172 between a particle pair at a distance @xmath173 . note that in eq . ( [ eq_stressfluctdecoupled ] ) the stress fluctuations stemming from different interactions are decoupled , i.e. @xmath174 characterizes the self- or two - point contributions of directly interacting beads . since @xmath175 and , hence , @xmath176 this is identical to the affine shear elasticity @xmath26 derived at the end of sec . [ theo_stressfluctu ] . we have thus confirmed eq . ( [ eq_muamuftau ] ) . interestingly , the argument can be turned around : 1 . the stress fluctuation contribution term @xmath177 is obtained by the simple derivation given in this paragraph . 2 . using the general legendre transformation , eq . ( [ eq_muftransform ] ) , this implies @xmath178 for the @xmath37- and the @xmath155-ensemble . as already stressed , our thermodynamic derivation of eq . ( [ eq_gmodtt ] ) is rather general , not using in particular a well - defined displacement field for the particle positions . [ [ truncation . ] ] truncation . + + + + + + + + + + + it is common practice in computational condensed matter physics @xcite to truncate a pair interaction potential @xmath127 , with @xmath126 being the distance between two particles @xmath131 and @xmath132 , at a conveniently chosen cutoff @xmath179 . this allows to reduce the number of interactions to be computed and energy or force calculations become @xmath180-processes . however , the truncation also introduces technical difficulties , e.g. , instabilities in the numerical solution of differential equations as investigated especially for the md method @xcite . without restricting much in practice the generality of our results , we assume below that * the pair potential @xmath127 is short - ranged , i.e. that it decays within a few particle diameters , * it scales as @xmath181 with the reduced dimensionless distance @xmath182 where @xmath183 characterizes the length scale of the interaction @xmath130 and * the same reduced cutoff @xmath184 is set for all interactions @xmath130 . for instance , for monodisperse particles with constant diameter @xmath185 , as for the standard lennard - jones ( lj ) potential @xcite , @xmath186 the scaling variable becomes @xmath187 and the reduced cutoff @xmath188 . the effect of introducing @xmath189 is to replace @xmath190 by the truncated potential @xmath191 with @xmath192 being the heaviside function @xcite . even if eq . ( [ eq_utrunc ] ) is taken _ by definition _ as the new system hamiltonian , it is well known that impulsive corrections at the cutoff have to be taken into account in general for the excess pressure @xmath29 and other moments of the first derivatives of the potential @xcite . this is seen by considering the derivative of the truncated potential @xmath193 [ [ shifting - of - truncated - potential . ] ] shifting of truncated potential . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + these corrections with respect to first derivatives can be avoided by considering a properly shifted potential @xcite @xmath194 since @xmath195 . with this choice _ no _ impulsive corrections arise for moments of the instantaneous shear stress @xmath30 and of other components of the excess stress tensor @xmath196 , eq . ( [ eq_sigab ] ) . specifically , if the potential is shifted , all impulsive corrections are avoided for the shear stress fluctuations @xmath55 , eq . ( [ eq_muf ] ) . [ [ correction - to - the - born - coefficient . ] ] correction to the born coefficient . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as shown in ref . @xcite , the standard shifting of a truncated potential is , however , insufficient in general for properties involving second ( and higher ) derivatives of the potential . this is particulary the case for the born coefficient @xmath28 , eq . ( [ eq_mub ] ) , which is required to compute the affine shear elasticity @xmath26 , eq . ( [ eq_muamubpex ] ) . the second derivative of the truncated and shifted potential being @xmath197 this implies that the born coefficient reads @xmath198 with @xmath199 being the bare born coefficient and @xmath200 the impulsive correction which needs to be taken into account . the latter correction may be readily computed numerically from the configuration ensemble using @xcite @xmath201 being a weighted radial pair distribution function which is related to the standard radial pair distribution function @xmath202 @xcite . [ [ mixtures - and - polydisperse - systems . ] ] mixtures and polydisperse systems . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + below we shall consider model systems for mixtures and polydisperse systems where @xmath183 may differ for each interaction @xmath130 . for such mixed potentials @xmath190 , @xmath203 and @xmath204 and their derivatives take in principal an explicit index @xmath130 , i.e. one should write @xmath205 , @xmath206 , @xmath207 and so on . ( this is often not stated explicitly to keep a concise notation . ) for example one might wish to consider the generalization of the monodisperse lj potential , eq . ( [ eq_lj ] ) , to a mixture or polydisperse system with @xmath208 where @xmath209 and @xmath183 are fixed for each interaction @xmath130 . in practice , each particle @xmath131 may be characterized by an energy scale @xmath210 and a diameter " @xmath211 . the interaction parameters @xmath212 and @xmath213 are then given in terms of specified functions of these properties @xcite . the extensively studied kob - andersen ( ka ) model for binary mixtures of beads of type a and b @xcite , is a particular case of eq . ( [ eq_ljgeneralized ] ) with fixed interaction ranges @xmath214 , @xmath215 and @xmath216 and energy parameters @xmath217 , @xmath218 and @xmath219 characterizing , respectively , aa- , bb- and ab - contacts . the expression eq . ( [ eq_histolame ] ) remains valid for such explicitly @xmath130-dependent potentials . [ [ shear - modulus - and - compression - modulus . ] ] shear modulus and compression modulus . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + since @xmath220 , eq . ( [ eq_histolame ] ) implies in turn @xmath221 with @xmath222 being the uncorrected bare shear modulus . as we shall see in sec . [ res_thigh ] , the correction @xmath200 is of importance for the _ precise _ determination of @xmath54 close to the glass transition where the shear modulus must vanish . an impulsive correction has also to be taken into account for the compression modulus @xmath223 computed from the normal pressure fluctuations in a constant volume ensemble as discussed in appendix [ app_k ] . using the symmetry of isotropic systems , eq . ( [ eq_etabmub ] ) , one verifies @xcite @xmath224 with @xmath225 being the uncorrected compression modulus . [ [ introduction.-1 ] ] introduction . + + + + + + + + + + + + + the computational data presented in this work have been obtained using two extremely well studied models of colloidal glass - formers which are described in detail elsewhere @xcite . we present first both models and the molecular dynamics ( md ) and monte carlo ( mc ) @xcite methods used to compute them in the different ensembles ( @xmath37 , @xmath49 , @xmath155 and @xmath50 ) compared in this study . we comment then on the quench protocol and locate the glass transition temperature @xmath9 for both models via dilatometry , as shown in fig . [ fig_quench_rho ] . this is needed as a prerequisite to our study of the elastic behavior in sec . [ sec_res ] . as sketched in fig . [ fig_sketch ] , periodic boundary conditions are applied in all spatial directions with @xmath226 for the linear dimensions of the @xmath19-dimensional simulation box . time scales are measured in units of the lj time @xmath227 @xcite for our md simulations ( @xmath228 being the particle mass , @xmath185 the reference length scale and @xmath229 the lj energy scale ) and in units of monte carlo steps ( mcs ) for our mc simulations . lj units are used throughout this work ( @xmath230 ) and boltzmann s constant @xmath16 is also set to unity . [ [ kob - andersen - model . ] ] kob - andersen model . + + + + + + + + + + + + + + + + + + + the kob - andersen ( ka ) model @xcite for binary mixtures of lj beads in @xmath1 dimensions has been investigated by means of md simulation @xcite taking advantage of the public domain lammps implementation @xcite . we use @xmath231 beads per simulation box and molar fractions @xmath232 and @xmath233 for both types of beads @xmath234 and @xmath235 . following ref . @xcite we set @xmath236 , @xmath237 and @xmath238 for the interaction range and @xmath239 , @xmath240 and @xmath241 for the lj energy scales . only data for the usual ( reduced ) cutoff @xmath242 are presented . for the @xmath155-ensemble we use the nos - hoover barostat provided by the lammps code ( fix npt command " ) @xcite which is used to impose a constant pressure @xmath243 for all temperatures . starting with an equilibrated configuration at @xmath244 well above the glass transition temperature @xmath245 ( as confirmed in the inset of fig . [ fig_quench_rho ] ) , the system is slowly quenched in the @xmath155-ensemble . the imposed mean temperature @xmath246 varies linearly as @xmath247 with @xmath248 being the constant quench rate . after a tempering time @xmath249 we compute over a sampling time @xmath250 various properties for the isobaric system at fixed temperature such as the moduli @xmath54 represented by the filled spheres in fig . [ fig_gt ] . fixing only then the volume ( @xmath37-ensemble ) the system is again tempered ( @xmath249 ) and various properties as the ones recorded in table [ tab_ka ] are computed using a sampling time @xmath250 . note that in the @xmath37-ensemble the equations of motion are integrated using a velocity verlet algorithm @xcite with a time step @xmath251 and systems are kept at the imposed temperature @xmath5 using a langevin thermostat @xcite with friction coefficient @xmath252 . by redoing for several temperatures around and below @xmath9 the tempering and the production runs we have checked that the presented results remain unchanged ( within numerical accuracy ) and that ageing effects are absolutely negligible . averages are taken over four independent configurations . as one may expect from the discussion in sec . [ theo_legendre ] , very similar results have been found for both ensembles for simple averages , for instance for the affine shear elasticity @xmath26 , and for the shear modulus @xmath54 , eq . ( [ eq_gmodtt ] ) . unfortunately , we are yet unable to present data for the ka model data at imposed shear stress @xmath4 ( @xmath49- , @xmath50-ensembles ) . .some properties for the ka model at normal pressure @xmath243 and shear stress @xmath41 : the mean density @xmath253 , the mean energy per volume @xmath254 , the impulsive correction term @xmath200 ( sec . [ theo_trunc ] ) , the ( corrected ) born - lam coefficient @xmath28 , the affine shear elasticity @xmath255 , the shear modulus @xmath54 obtained in the @xmath37-ensemble , the hypervirial @xmath256 and the affine dilatation elasticity @xmath257 discussed in appendix [ app_k ] and the compression modulus @xmath258 obtained for the @xmath37-ensemble . the affine elasticities @xmath26 and @xmath259 are the natural scales for , respectively , the shear modulus @xmath0 and the compression modulus @xmath20 . [ tab_ka ] [ cols="^,^,^,^,^,^,^,^,^,^,^,^",options="header " , ] [ [ local - mc - moves - for - the - plj - model . ] ] local mc moves for the plj model . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for all ensembles considered here local mc moves are used ( albeit not exclusively ) where a particle is chosen randomly and a displacement @xmath260 , uniformly distributed over a disk of radius @xmath261 , from the current position @xmath262 of the particle is attempted . the corresponding energy change @xmath263 is calculated and the move is accepted using the standard metropolis criterion @xcite . the maxium displacement distance @xmath261 is chosen such that the acceptance rate @xmath234 remains reasonable , i.e. @xmath264 @xcite . as may be seen from the inset in fig . [ fig_quench_e_plj ] where the acceptance rate @xmath265 is shown as a function of temperature @xmath5 , we find that temperatures below @xmath266 are best sampled using @xmath267 , whereas @xmath268 is a good choice for the interesting temperature regime between @xmath269 and @xmath270 around the glass transition temperature @xmath271 . ( see the main panel of fig . [ fig_quench_rho ] for the determination of @xmath9 . ) various data sets are presented in sec . [ sec_res ] for a fixed value @xmath272 which allows a reasonable comparison of the sampling time dependence for temperatures around the glass transition . this value is not necessarily the best choice for tempering the system and for computing static equilibrium properties at the given temperature , especially below @xmath273 . [ [ collective - plane - wave - mc - moves . ] ] collective plane - wave mc moves . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + local mc jumps in the @xmath37-ensemble become obviously inefficient for relaxing large scale structural properties @xcite for the lowest temperatures computed for the plj model ( @xmath274 ) as may be seen , e.g. , from the finite , albeit very small shear stresses @xmath4 which happen to occur naturally in some of the quenched configurations ( especially if the quench rate @xmath275 is too large ) and can only with difficulty be relaxed using local jumps _ alone _ @xcite . we have thus crosschecked and improved the values obtained using only local mc moves by adding global mc move attempts for all particles with longitudinal and ( more importantly ) transverse plane waves commensurate with the simulation box . the random amplitudes and phases of the waves are assumed to be uniformly distributed . note that the maximum amplitude @xmath276 of each wave is chosen inversely proportional to the length @xmath277 of the wavevector in agreement with continuum theory . this allows to keep the same acceptance rate for each global mode if @xmath278 . as one expects , the plane - wave displacement attempts becomes inappropriate for large @xmath279 where continuum elasticity breaks down @xcite and , hence , the acceptance rate @xmath280 gets too large to be efficient @xcite . note that the maximum amplitude @xmath276 has to be chosen much smaller for the longitudinal waves as for the transverse ones , i.e. since the systems are essentially incompressible , the transverse modes naturally are more effective for quenching and equilibrating the systems . these global plane - wave moves have been added for the tempering of all presented configurations below @xmath270 with @xmath281 and for production runs over sampling times @xmath282 for static properties for @xmath283 . details concerning the collective plane - wave mc moves must be given elsewhere . [ [ mc - sampling - of - different - ensembles . ] ] mc sampling of different ensembles . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in this paper we shall be concerned more with the consequences of affine displacements associated with collective mc moves changing the overall box volume @xmath12 and shear strain @xmath3 . using again a metropolis criterion for accepting a suggested change of @xmath12 or @xmath3 , these pure dilatational and pure shear strain fluctuations are used to impose a mean normal pressure @xmath284 and a mean shear stress @xmath41 . for a shear strain altering mc move one first chooses randomly a strain fluctuation @xmath285 with @xmath286 . assuming an imposed shear stress @xmath4 the suggested @xmath285 is accepted if @xmath287 with @xmath288 being a uniformly distributed random variable with @xmath289 . the energy change @xmath263 associated with the affine strain may be calculated using eq . ( [ eq_strain_r ] ) . since @xmath290 is supposed in this study , the last term in the exponential drops out . ( as noted in sec . [ theo_stressfluctu ] , the volume remains unchanged in a pure shear strain and , hence , no translational entropy contribution appears . ) the maximum shear strain step @xmath291 is fixed such that at the given temperature the acceptance rate remains reasonable @xcite . as described in more detail in ref . @xcite , one similarly chooses a relative volume change @xmath292 with @xmath12 being the current volume @xcite . the volume altering move is accepted at an imposed pressure @xmath6 if @xmath293 where @xmath263 may be computed using eq . ( [ eq_strain_r_epsilon ] ) . the logarithmic contribution corresponds to the change of the translational entropy . using eq . ( [ eq_timposed ] ) after each mcs for local mc moves ( and global plane - wave mc moves if these are added ) realizes the @xmath49-ensemble shown in fig . [ fig_sketch](c ) , using eq . ( [ eq_pimposed ] ) the @xmath155-ensemble shown in fig . [ fig_sketch](d ) and using in turn both strain fluctuations the @xmath50-ensemble in fig . [ fig_sketch](e ) . we remind that @xmath37- and @xmath155-ensembles are used to determine @xmath54 and @xmath49- and @xmath50-ensembles to obtain @xmath48 and @xmath151 . as shown in fig . [ fig_quench_rho ] , we quench the configurations starting from @xmath294 in the @xmath155- and the @xmath50-ensemble using a constant quench rate @xmath275 . note that the smallest quench rate @xmath295 for one configuration in the @xmath155-shown did require alone a run over six months using one core of an intel xeon e5410 processor . interestingly , similar results are obtained with the @xmath50-ensemble only using a rate @xmath296 . ( due to the additional computation for the shear strain fluctuations this is , however , only a factor five faster . ) the configurations created by the @xmath155-quenches are used ( after tempering ) for the sampling in the @xmath37- and @xmath155-ensembles , the configurations obtained by the @xmath50-quenches for the sampling in the @xmath49- and @xmath50-ensembles . only two independent configurations have been sampled following the full protocol for each temperature and each of the four ensembles . instead of increasing further the number of configurations ( which will be done in the future ) we have focused in the present preliminary study on long tempering times and production runs over @xmath282 mcs which have been redone several times for a few temperatures ( summing up to total runs of @xmath297 mcs for @xmath269 and @xmath298 ) to check for equilibration problems and to verify that ageing effects can be ignored . we firmly believe that the breaking of the time translational invariance is not a relevant issue for the strong sampling time dependence reported below for the shear moduli . [ [ algo_sample ] ] glass transition temperature . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the inset of fig . [ fig_quench_rho ] presents for the ka model how the glass transition temperature @xmath9 may be obtained using standard dilatometry where the rescaled density @xmath299 is traced as a function of temperature @xmath5 as in ref . @xcite for a similar polymer glass problem . this reveals a linear low-@xmath5 ( solid line ) and a linear high-@xmath5 regime . the intercept of both lines confirms the well - known value @xmath300 from the literature @xcite . the main panel presents the unscaled number density @xmath253 for plj systems at constant normal pressure @xmath284 as a function of temperature @xmath5 for different quench rates @xmath275 in mcs . only local mc moves with a fixed maximum particle displacement @xmath301 for all temperatures @xmath5 are used for the examples given . the crosses refer to a quench in the @xmath50-ensemble where the box shape is allowed to fluctuate at @xmath41 . all other data refer to the @xmath155-ensemble with @xmath36 . the solid line and the dashed line are linear fits to the best density estimates indicated by the stars obtained for , respectively , low and high temperatures . by matching both lines one determines @xmath302 . a similar value is given if @xmath299 is plotted as a function of @xmath5 . the main panel of fig . [ fig_quench_e_plj ] shows the interaction energy per particle @xmath303 of the plj model as a function of temperature for the same quench protocols as in fig . [ fig_quench_rho ] . the solid line and the dashed line are linear fits to the energy for , respectively , low and high temperatures . this confirms @xmath302 for the plj model @xcite . we begin our discussion by focusing on the high temperature liquid limit where all reasonable operational definitions of the shear modulus @xmath0 must vanish . the results presented in fig . [ fig_gt_thigh ] have been obtained for the plj model at mean pressure @xmath284 , mean shear stress @xmath304 and mean temperature @xmath305 , i.e. far above the glass transition temperature @xmath302 . only local mc monomer displacements with a maximum jump displacement @xmath272 are reported here . instantaneous properties relevant for the moments are written down every @xmath306 mcs and averaged using standard gliding averages @xcite , i.e. we compute mean values and fluctuations for a given time interval @xmath307 $ ] and average over all possible intervals of length @xmath32 . the horizontal axis indicates the latter interval length @xmath32 in units of mcs . the vertical axis is made dimensionless by rescaling the moduli with the ( corrected ) affine shear elasticity @xmath308 from table [ tab_plj ] . ( being a simple average , @xmath26 is found to be identical for all computed ensembles . ) in agreement with eq . ( [ eq_gbound ] ) , the ratio @xmath309 for all operational definitions drops rapidly below unity . the dashed line indicates the asymptotic power - law slope @xmath310 @xcite . this can be understood by noting that at imposed @xmath4 the shear strain freely diffuses in the liquid limit , i.e. @xmath311 , which implies that @xmath312 . for both @xmath49- and @xmath50-ensembles we compare our key definition @xmath48 , eq . ( [ eq_gmodgt ] ) , to the observable @xmath151 , eq . ( [ eq_gmodgg ] ) . in both cases we confirm @xmath313 for larger times as suggested by the thermodynamic relation eq . ( [ eq_dgamdtau ] ) which has been directly checked @xcite . we have also computed the dimensionless correlation coefficient @xmath314 , eq . ( [ eq_cmodgt ] ) , which characterize the correlation of the measured instantaneous shear strains @xmath46 and shear stresses @xmath30 . as indicated for the @xmath50-ensemble ( stars ) , @xmath90 vanishes rapidly with @xmath32 , i.e. @xmath46 and @xmath30 are decorrelated . the dash - dotted line indicates the power - law exponent @xmath315 expected from eq . ( [ eq_cmodgt_mua ] ) and @xmath316 . note that by plotting @xmath314 _ vs. _ @xmath317 , eq . ( [ eq_cmodgt_mua ] ) has been directly verified for both @xmath49- and @xmath50-ensembles ( not shown ) . the stress fluctuation formula , eq . ( [ eq_gmodtt ] ) , is represented in fig . [ fig_gt_thigh ] by squares and diamonds for , respectively , @xmath37- and @xmath155-ensembles . the bare shear modulus @xmath222 _ without _ the impulsive correction @xmath200 for the born coefficient @xmath28 discussed in sec . [ theo_trunc ] is given by small filled symbols . the impulsive correction @xmath318 ( table [ tab_plj ] ) computed independently from the weighted radial pair correlation function , eq . ( [ eq_histolame ] ) , is shown by a horizontal line . the corrected moduli @xmath319 vanish indeed as expected . see ref . @xcite for a systematic numerical characterization of @xmath200 as a function of the reduced cutoff @xmath189 . as may also be seen there , a truncation correction is also necessary for the ka model . the @xmath200 for different temperatures may be found in table [ tab_ka ] for the ka model and in table [ tab_plj ] for the plj model . we assume from now on that the impulsive corrections are properly taken into account without stating this technical point explicitly . [ [ introduction.-2 ] ] introduction . + + + + + + + + + + + + + we turn now to the opposite low-@xmath5 limit focusing specifically on the plj model at @xmath320 . obviously , such deeply quenched colloidal glasses must behave as amorphous solids with a constant shear modulus for large times ( albeit not infinite times ) as sketched in fig . [ fig_sketch](b ) by the top curve . before presenting in sec . [ res_tlow ] the elastic properties of these glasses , we discuss first conceptually simpler substitute systems formed by _ permanent _ spring networks . we do this for illustration purposes since questions related to ergodicity and ageing are by construction irrelevant and the thermodynamical relations reminded in sec . [ sec_theo ] should hold rigorously . [ [ network - hamiltonian . ] ] network hamiltonian . + + + + + + + + + + + + + + + + + + + + assuming ideal harmonic springs the interaction energy reads @xmath321 with @xmath322 being the spring constant and @xmath323 the reference length of spring @xmath130 . the sum runs over all springs @xmath130 between topologically connected vertices @xmath131 and @xmath132 of the network . we assume here a strongly and homogeneously connected network where every vertex @xmath131 is in contact with many neighbors @xmath132 . such a network may be constructed from a plj configuration at @xmath320 keeping the particle positions for the positions of the vertices and replacing each lj interaction @xmath130 by a spring of spring constant @xmath322 and reference length @xmath323 . the simplest choice we have investigated is to set @xmath324 and @xmath325 which corresponds to a well - defined ground state of energy @xmath326 and pressure @xmath327 at @xmath328 ( not shown ) . we discuss here instead a slightly more realistic choice of @xmath322 and @xmath323 which corresponds to the _ same _ dynamical matrix " as the plj configuration at @xmath329 and @xmath284 , i.e. the same second derivative of the total interaction potential with respect to the particle positions @xcite . this choice imposes the setting @xmath330 with @xmath331 being the interaction potential for reduced distances @xmath332 . ( impulsive corrections at @xmath333 are neglected here . ) [ [ sampling . ] ] sampling . + + + + + + + + + taking advantage of the fact that all interacting vertices are known , these networks can be computed using global mc moves with longitudinal and transverse planar waves as described in sec . [ sec_algo ] . no local mc jumps for individual vertices have been added . note that although some @xmath322 and @xmath323 may even be negative , this does not affect the global or local stability of the network . presumably due to the fact that our spring constants are rather large , we have not observed either any buckling instability if volume fluctuations are allowed . for networks at the same pressure @xmath6 , shear stress @xmath4 and temperature @xmath5 as the original plj configuration , we obtain , not surprisingly , the same affine shear elasticity @xmath334 as for the reference . the same applies for other simple averages . [ [ shear - modulus . ] ] shear modulus . + + + + + + + + + + + + + + figure [ fig_gt_dynmat ] presents various measurements of the shear modulus @xmath0 for different ensembles as a function of sampling time @xmath32 . as in sec . [ res_thigh ] , we use gliding averages and the vertical axis has been rescaled by the affine shear elasticity @xmath26 . as can be seen , we obtain in the long - time limit @xmath335 ( bold line ) irrespective of whether the shear modulus is determined from @xmath48 or @xmath151 in the @xmath49- or @xmath50-ensembles or using @xmath54 in the @xmath37- or @xmath155-ensembles at fixed shear strain @xmath3 . note that even in these cases the shear modulus decreases first with time albeit the network is perfectly at thermodynamic equilibrium and no ageing occurs by construction . as shown by the dash - dotted line , @xmath336 for short times , i.e. in this limit the response is affine . as indicated by the stars , we have also computed the correlation coefficient @xmath314 as a function of time . from @xmath92 for small @xmath32 , it decreases somewhat becoming @xmath337 for large times in agreement with eq . ( [ eq_cmodgt_mua ] ) . if @xmath54 is computed in the wrong " @xmath49- or @xmath50-ensembles , it is seen to vanish rapidly with time as shown by the filled squares . this decay is of course expected from eq . ( [ eq_muamuftau ] ) as discussed in sec . [ theo_stressfluctu ] and sec . [ theo_legendre ] . [ [ shear - stress - fluctuations . ] ] shear - stress fluctuations . + + + + + + + + + + + + + + + + + + + + + + + + + + to emphasize the latter point we show in fig . [ fig_muft_dynmat ] the stress fluctuations @xmath338 for different ensembles . this is also done to have a closer look at the time dependence of various contributions to @xmath339 . as before , the vertical axis has been rescaled with @xmath340 . we also indicate the rescaled values of @xmath341 ( thin top line ) and @xmath342 ( dash - dotted line ) computed by a gliding average over time intervals @xmath307 $ ] . being simple averages both properties do not depend on the ensemble probed ( not shown ) and become virtually immediately @xmath32-independent as can be seen from the indicated horizontal lines . this is quite different for the different shear stress fluctuations @xmath55 presented which increase monotonously from @xmath343 for short times to their plateau value for long times . in agreement with eq . ( [ eq_muamuftau ] ) we find that @xmath344 for @xmath49- and @xmath50-ensembles and as predicted by eq . ( [ eq_muftransform ] ) we confirm @xmath345 for @xmath37- and @xmath155-ensembles . the asymptotic limit is reached after about @xmath346 mcs in the first case for imposed @xmath4 , but only after @xmath347 for imposed @xmath3 . this is due to the fact that @xmath174 corresponds to the fluctuations of the self - contribution of individual springs as discussed in sec . [ theo_legendre ] , while @xmath348 also contains contributions from stress correlations of different springs . it should be rewarding to analyze in the future finite system - size effects and the time dependence of three - dimensional ( 3d ) systems using the dynamical matrix associated to the ka model in the @xmath329 limit . we return now to the colloidal glasses where the topology of the interactions is , of course , _ not _ permanently fixed ( at least not at finite temperatures ) and the shear stresses may thus fluctuate more strongly . as an example , we present in fig . [ fig_gt_tlow ] the shear modulus @xmath349 as a function of sampling time @xmath32 for plj beads at mean pressure @xmath284 , mean shear stress @xmath350 and mean temperature @xmath320 , i.e. the same conditions as in the previous subsection . all simple averages , such as the affine shear elasticity @xmath351 , are the same for all ensembles albeit different quench protocols have been used . for all examples given we use local mc jumps with @xmath301 which allows to keep a small , but still reasonable acceptance rate @xmath352 . we have also performed runs with additional global mc moves using longitudinal and transverse plane waves . this yields similar data which is shifted horizontally to the left ( not shown ) . as one would expect , the shear moduli presented in fig . [ fig_gt_tlow ] are similar to the ones shown in fig . [ fig_gt_dynmat ] for the permanent spring network . the filled triangles show @xmath54 computed in the wrong " @xmath50-ensemble . as expected from eq . ( [ eq_muftransform ] ) these values vanish rapidly with time . if on the other hand @xmath48 , @xmath151 and @xmath54 are computed in their natural ensembles , all these measures of @xmath0 are similar and approach , as one expects for such a low temperature , the same finite asymptotic value @xmath353 indicated by the bold horizontal line . the stars indicate the squared correlation coefficient @xmath354 which is seen to collapse perfectly on the rescaled modulus @xmath317 which shows that eq . ( [ eq_cmodgt_mua ] ) even holds for small times before the thermodynamic plateau has been reached . similar reduced shear moduli @xmath355 have been also obtained for the ka model in the low-@xmath5 limit as may be seen from table [ tab_ka ] . these results confirm that about half of the affine shear elasticity is released by the shear stress fluctuations in agreement with refs . interestingly , this value is slightly smaller as the corresponding value for the permanent dynamical matrix model presented above . since some , albeit very small , rearrangement of the interactions must occur for the plj particles at finite temperature , this is a reasonable finding . the rescaled shear stress fluctuations @xmath356 for the same ensembles are presented in fig . [ fig_muft_tlow ] . while the two - point correlations @xmath357 ( thin horizontal line ) and @xmath358 ( dash - dotted horizontal line ) do not dependent on time , one observes again that @xmath356 increases monotonously from zero to its long - time asymptotic plateau value . in agreement with eq . ( [ eq_muamuftau ] ) we find @xmath344 for the @xmath49 and the @xmath50 ensembles , while @xmath345 ( dashed horizontal line ) for the @xmath37 and the @xmath155 ensembles . ( similar behavior has been obtained for the ka model in the @xmath37-ensemble for temperatures @xmath274 . ) as may be better seen for the data set obtained for @xmath359 using the @xmath37-ensemble ( small filled squares ) where @xmath360 , the stress fluctuations @xmath361 do not rigorously become constant , but increase extremely slowly on the logarithmic time scales presented . there is , hence , even at low temperatures always some sampling time dependence if the interactions are not permanently fixed as for the topologically fixed networks discussed above . however , this effect becomes only sizeable above @xmath269 for the ka model and above @xmath266 for the plj model . we shall now attempt to characterize this temperature dependence . [ [ stress - fluctuations . ] ] stress fluctuations . + + + + + + + + + + + + + + + + + + + + we turn now to the characterization of the temperature dependence of the shear modulus @xmath0 and various related properties . we focus first on the best values obtained for the longest simulation runs available . figure [ fig_muamuf_t ] presents the stress fluctuations obtained for the ka model using the @xmath37-ensemble ( open spheres ) and for the plj model using @xmath37- and the @xmath50-ensembles . the affine shear elasticity @xmath26 obtained for each system is represented by small filled symbols . as shown for both models by a bold solid line for the low-@xmath5 and by a dashed line for the high-@xmath5 regime , @xmath8 decays roughly linearly with @xmath5 . interestingly , the slopes match precisely at @xmath362 for both models . as may be seen from the large triangles for the @xmath50-ensemble , we confirm for all temperatures that @xmath363 as expected from eq . ( [ eq_muamuftau ] ) . for systems without box shape fluctuations , it is seen that @xmath364 becomes _ non - monotonous _ with a clear maximum at @xmath365 . in the liquid regime for @xmath366 where the boundary conditions do not matter , we obtain @xmath367 , i.e. all shear stress fluctuations decay with temperature . below @xmath9 the boundary constraint ( @xmath36 ) becomes more and more relevant with decreasing @xmath5 which reduces the shear stress fluctuations . according to eq . ( [ eq_muftransform ] ) , @xmath364 must thus decrease with increasing @xmath0 as the system is further cooled down . that @xmath364 must necessarily be non - monotonous with a maximum at @xmath9 is one of the central results of the presented work . [ [ shear - modulus - as - a - function - of - temperature . ] ] shear modulus as a function of temperature . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the temperature dependence of the ( unscaled ) shear moduli @xmath35 is presented in fig . [ fig_gt ] . data obtained by md simulation of the ka model ( @xmath368 , @xmath300 ) are indicated for the @xmath37-ensemble ( open spheres ) and the @xmath155-ensemble ( filled spheres ) . all other results refer to the best values obtained by mc simulations of the plj model ( @xmath284 , @xmath302 ) using both local and global moves and different ensembles as indicated . for the plj model the values @xmath54 obtained in the @xmath37 ensemble ( squares ) are seen to be essentially identical for all @xmath5 to the moduli @xmath48 ( crosses ) and @xmath151 ( triangles ) obtained in ensembles with strain fluctuations . as shown by the small filled triangles , @xmath54 vanishes in the @xmath50 ensemble for all @xmath5 in agreement with eq . ( [ eq_muamuftau ] ) . ( we remind that the impulsive correction @xmath200 has to be taken into account if @xmath54 is used . ) decreasing the temperature further below @xmath9 the shear moduli are seen to increase rapidly for both models . as shown by the solid and the dashed lines both models are well fitted by a cusp - singularity @xmath369 with fit constants @xmath370 for the ka model and @xmath371 for the plj model where we have set in both cases @xmath372 . confirming the md simulations by barrat _ @xcite , this suggests that the transition is very sharp , albeit _ continuous _ in qualitative agreement with the predictions from ref . @xcite . if we fit the data close to @xmath9 with an additional off - set @xmath373 , a slightly _ negative _ value is obtained which is not compatible with mct @xcite . however , admittedly both the number of data points close to @xmath9 and their precision ( due to the small number of independent configurations ) are yet not sufficient to rule out completely a positive , albeit ( presumably ) small off - set . a slightly different representation of the data is given in fig . [ fig_gt_rescal ] where the reduced shear modulus @xmath374 is plotted as a function of the reduced temperature @xmath375 for both models . using the independently determined glass transition temperature @xmath9 and affine shear elasticity @xmath8 as scales to make the axes dimensionless , the rescaled data are found to collapse ! this is of course a remarkable and rather unexpected result considering that two different models in two different dimensions have been compared . the bold line indicates the cusp - singularity @xmath376 with a prefactor which is compatible to the ones used in eq . ( [ eq_cuspcont ] ) considering the typical values of @xmath8 in both models . whether this striking collapse is just due to some lucky coincidence or makes manifest a more general universal scaling is impossible to decide at present . ( please note that the corresponding data for the compression modulus shown in fig . [ fig_kt ] does not scale . ) no rescaling of the vertical axis is necessary for the squared correlation coefficient @xmath377 shown in the inset for two different ensembles of the plj model . the same continuous cusp - like decay ( bold line ) is found as for the modulus @xmath0 . this is again consistent with the thermodynamic reasoning put forward at the end of sec . [ theo_strainfluctu ] , eq . ( [ eq_cmodgt_mua ] ) . [ [ sampling - time - dependence . ] ] sampling time dependence . + + + + + + + + + + + + + + + + + + + + + + + + + a word of caution must be placed here . the results presented in figs . [ fig_muamuf_t]-[fig_gt_rescal ] correspond to the longest simulation runs we have at present been able to perform . obviously , this does _ not _ mean that they correspond to the limit for asymptotically long sampling times , if the latter limit exists , or at least an intermediate plateau of respectable period of time . focusing on the plj model in the @xmath50-ensemble we attempt in fig . [ fig_gt_t ] to give a tentative characterization of the sampling time dependence . we plot @xmath378 as a function of @xmath32 for different temperatures as indicated . to make the data comparable all simulations for @xmath379 have been performed with local mc jumps of same maximum distance @xmath272 . as we have already seen in fig . [ fig_gt_thigh ] , the shear modulus decays inversely with time in the liquid regime at high temperatures . this limit is indicated by the thin line . with decreasing @xmath5 the ( unscaled ) shear modulus increases and a shoulder develops . note that even for @xmath298 , i.e. slightly below the glass transition temperature @xmath380 estimated via dilatometry ( fig . [ fig_quench_rho ] ) , @xmath378 decays strongly with time . a more or less @xmath32-independent shoulder ( on the logarithmic scales used ) can only be seen below @xmath273 . we emphasize that we present here a sampling time effect expressing the fact that the configuration space is more and more explored with increasing @xmath32 and _ not _ a time correlation function @xcite . please note also that this time dependence has nothing to do with equilibration problems or ageing effects . time translational invariance is perfectly obeyed in our simulations as we have checked by rerunning the sampling _ after _ additional tempering over at least @xmath347 mcs for all temperatures . similar @xmath32-dependencies for different temperatures @xmath5 have also been observed for @xmath378 and @xmath339 for the plj model and for @xmath339 for the ka model ( not shown ) . a different representation of the data for both models is shown in fig . [ fig_gt_t ] where the reduced shear modulus @xmath381 is plotted against the reduced temperature @xmath382 for different sampling times @xmath32 as indicated . similar behavior is found for both models . the shear modulus decreases with @xmath32 and this the more the larger the temperature @xmath5 . for smaller temperatures the different data sets appear to approach more readily as one expects from the @xmath32-independent shoulder for the plj model in fig . [ fig_gt_t ] . as a consequence the glass transition is seen to become sharper with increasing @xmath32 . the data approach the continuous cusp - singularity indicated by the bold line . however , it is not clear from the current data if the data converge to an at least _ intermediately _ stable behavior . longer simulation runs are clearly necessary for both models to clarify this issue . similar data have also been obtained for the plj model from the independent determination of @xmath378 and @xmath383 in the @xmath49-ensemble and for @xmath339 in the @xmath37- and @xmath155-ensemble . the stress fluctuation formalism is a powerful method for computing elastic moduli for canonical ensembles using simulation boxes of constant volume and shape @xcite . focusing on the shear modulus @xmath0 of two well - known glass - forming colloidal model systems in @xmath1 and @xmath2 dimensions , which we have sampled by means of md and mc simulations , we have addressed the general question of whether the stress fluctuation method may be used through a solid - liquid transition where a well - defined displacement field ceases to be defined . deliberately setting the experimentally motivated linear regression eq . ( [ eq_gmodgt ] ) as the fundamental definition , the shear modulus @xmath0 has been computed from the strain and stress fluctuations ( assuming a fixed finite measurement time window @xmath384 ) in ensembles where either a shear strain @xmath36 ( @xmath37- and @xmath155-ensembles ) or a ( mean ) shear stress @xmath41 ( @xmath49- and @xmath50-ensembles ) have been imposed ( fig . [ fig_sketch ] ) . working at constant mean normal pressure we have computed and compared the temperature dependence for various simple averages and fluctuations contributing to the shear modulus in different ensembles . for the ka model only the currently available results for @xmath37- and @xmath155-ensembles have been presented . as has been stressed in sec . [ theo_stressfluctu ] and sec . [ theo_legendre ] , the stress fluctuation representation @xmath54 at fixed shear strain does not rely _ in principle _ on a solid - like reference state for a displacement field of tagged particles if thermodynamics can be assumed to hold . it thus holds formally through the glass - transition temperature @xmath9 up to the liquid regime ( albeit with a trivial value @xmath34 ) as does the better known relation for the compression modulus @xmath20 ( appendix [ app_k ] ) @xcite . as emphasized in sec . [ theo_trunc ] , impulsive corrections due to the truncation of the pair potentials have to be properly taken into account , especially at high temperatures ( fig . [ fig_gt_thigh ] ) , for the precise determination of the affine shear elasticity @xmath255 @xcite . confirming eq . ( [ eq_gmods ] ) , it has been shown for the plj model that @xmath54 at @xmath36 has the _ same _ long - time behavior as the conceptually more direct observables @xmath48 and @xmath151 obtained using the strain fluctuations at constant mean shear stress @xmath41 . the standard thermodynamic relations comparing different ensembles appear thus to hold also _ in practice _ ( at least within our numerical precision ) through the glass transition for all @xmath5 and this albeit 1 . some degrees of freedom get quenched at low temperatures on the time window @xmath384 computationally accessible , 2 . all operational definitions of @xmath0 are transient in the sense that they vanish for @xmath385 ( figs . [ fig_gt_t ] and [ fig_gt_t ] ) and 3 . it is not self - evident that in the latter limit and for large temperatures @xmath3 and @xmath4 can still be treated as a pair of conjugated thermostatistical variables . the latter point compactly epitomized by the thermodynamic relation eq . ( [ eq_dgamdtau ] ) has been shown to hold , however , with remarkable precision up to very high temperatures for the plj model . as predicted by general legendre transformation , @xmath54 vanishes for all temperatures @xmath5 , if @xmath4 rather than @xmath3 is imposed ( figs . [ fig_gt_dynmat ] , [ fig_gt_tlow ] and [ fig_gt ] ) . the shear - stress fluctuations @xmath174 are thus given by the affine response under an external load @xmath26 ( fig . [ fig_muamuf_t ] ) , i.e. @xmath174 reduces to a simple two - point pair correlation function if pair potentials are considered as discussed in sec . [ theo_legendre ] . the same holds above @xmath9 for constant @xmath3 , since the boundary conditions are irrelevant for the liquid state . this symmetry with respect to the boundary conditions ( @xmath386 ) is broken at the glass transition for a large , but finite @xmath384 : at constant @xmath3 the stress fluctuations reveal a strong non - monotonous behavior with a clear maximum at @xmath9 ( fig . [ fig_muamuf_t ] ) . the shear modulus is the order parameter characterizing this symmetry breaking . alternatively , as we have discussed in sec . [ theo_legendre ] , @xmath387 may be seen as an order parameter comparing the ratio of the non - affine to the affine shear responses . since @xmath388 for @xmath389 , this implies that the stress fluctuations contain higher - order correlations _ reducing _ the free energy of the system . the increase of @xmath0 below @xmath9 is reasonably fitted for both models by a _ continuous _ cusp singularity , eq . ( [ eq_cuspcont ] ) , in qualitative agreement with recent replica calculations @xcite . a jump discontinuity , as suggested by mode - coupling theory @xcite and another replica theory @xcite , is not compatible with the currently available data shown in fig . [ fig_gt ] and fig . [ fig_gt_rescal ] . however , as shown in fig . [ fig_gt_t ] and fig . [ fig_gt_t ] , our data depend strongly on sampling time @xmath32 and the computation of larger @xmath32 could lead to a sharper transition . thus a jump discontinuity can not be ruled out completely . ( we are not aware of any other investigation of the sampling time effect for the shear modulus . ) at present we believe , however , that it would be surprising if our data could be reconciled with a discontinuity at @xmath9 . it seems more likely that below @xmath9 , where the choice of the boundary conditions does matter as shown , the definition of the glassy shear modulus used in refs . @xcite do not correspond to the key operational definition @xmath48 of the shear modulus considered by us . in any case , our work shows that any theory and numerical scheme used to determine the shear modulus using a stress or strain fluctuation relation ( in the low - wavevector limit or at a small finite @xmath279 ) must cleary specify and take into account whether the shear stress or the shear strain are macroscopically imposed . otherwise , such an approach is void . [ [ beyond - the - presented - work . ] ] beyond the presented work . + + + + + + + + + + + + + + + + + + + + + + + + + + future work should clearly focus on the more detailed description of the sampling time dependence . by extrapolating appropriately for the large-@xmath32 asymptotic behavior , this may allow to settle the theoretical debate . as emphasized above , one shortcoming of the present work is that the time - scale used in our mc simulations was slightly arbitrary , especially if additional non - local particle moves are used in the low-@xmath5 limit and if box volume and shape changing strains are included . the comparison of @xmath32-dependent properties for different temperatures @xmath5 becomes thus delicate . our mc study for the 2d soft particles should thus in any case be recomputed with langevin thermostat md dynamics as used for the data of the ka model presented . the @xmath32-dependence of the latter model , which is a better glass former than the 2d plj model @xcite , has still to be worked out for the key operational definition @xmath48 in the @xmath49- or @xmath50-ensembles @xcite . we plan also in the near future to reconsider more carefully our previous investigation of the glass transition of polymer melts @xcite taking properly into account the impulsive truncation corrections and comparing the shear moduli @xmath390 ( @xmath50-ensemble ) and @xmath54 ( @xmath50-ensemble ) around @xmath9 . in addition , it should be rewarding to compare our shear moduli with the values obtained from the displacement of particles following the procedure chosen in ref . this procedure relies on the idea that the particles fluctuate around a well - defined reference position . in the low-@xmath5 limit where the interaction network can be replaced by the dynamical matrix , this should yield the same results . however , for larger temperatures where the harmonic approximation must break down , this approach is questionable ( both for the analysis of experimental and computational data ) and should be compared to the moduli obtained directly from the strain and stress fluctuations of the overall simulation box . [ [ a - simple - scalar - model . ] ] a simple scalar model . + + + + + + + + + + + + + + + + + + + + + + we are currently working on an extremely simplified scalar model for permanent and transient networks in @xmath2 dimensions which may help to clarify the theoretical debate @xcite . as sketched in the panel ( a ) of fig . [ fig_band ] , in this model beads ( filled spheres ) are glued on rigid rods which are only allowed to move horizontally . in its most simple implementation this is done by means of global mc moves using transverse plane waves with a wavevector pointing in the @xmath140-direction . the glued beads are connected by ideal harmonic springs where the spring constants @xmath322 and reference lengths @xmath323 are chosen randomly according to fixed , narrow or broad distributions . the stiffness of the contacts between rods may differ ( even including negative @xmath322 and @xmath323 as for the dynamical matrix studied in sec . [ res_dynmat ] ) . the shear modulus may be computed either from the shear strain fluctuations in the @xmath49- or the shear stress fluctuations in the @xmath37-ensemble . the shear stresses can be relaxed by either allowing the particles to move along the rods using longitudinal plane waves in the @xmath105-direction and/or by breaking and reconnection of springs . the latter step is currently done by applying a chemical potential for the springs . physically , the study of such transient networks is motivated by systems of hyperbranched polymer chains with sticky end - groups @xcite or telechelic polymers in water - oil emulsions @xcite . the associated relaxation frequency @xmath391 for a given spring @xmath130 is assumed to be set by an activation barrier @xmath392 which may differ for each spring . depending on the distribution of this activation barrier the shear modulus @xmath349 thus decays more or less rapidly with sampling time @xmath32 as sketched in panel ( b ) of fig . [ fig_band ] . the shear modulus becomes constant , if the random percolating network is permanent ( @xmath393 , @xmath394 ) or the relaxation negligible ( top solid line ) . the system behaves as a liquid ( bottom line ) for large ( mean ) @xmath395 , while a shoulder is seen in the intermediate frequency range . while this is confirmed by preliminary simulations , we do at present not understand the role of the quenched noise injected in the model which is seen to strongly influence the shear stress fluctuations @xmath55 ( especially if fluffy layers happen to appear ) . also the understanding of finite system - size effects is of course crucial for such a low - dimensional model . ( only systems containing @xmath396 particles have been considered so far . ) as sketched in panel ( c ) the key question is to describe accurately for asymptotically long simulation runs and asymptotically large box sizes the shape of @xmath35 around the glass transition temperature @xmath397 where the shear modulus vanishes and @xmath348 should again have a maximum . thanks the cnrs and the irtg soft matter for supporting her sabbatical stay in strasbourg , p.p . the rgion alsace and the irtg soft matter and f.w . the daad for funding . we are indebted to a. blumen ( freiburg ) and h. meyer , o. benzerara and j. farago ( all ics , strasbourg ) for helpful discussions . we have focused in the main part of this paper on the shear modulus @xmath0 of our glass - forming model systems and the comparison of the three observables @xmath48 , @xmath151 and @xmath54 . being isotropic the complete macroscopic elastic description of our systems only requires the numerical determination of _ one _ additional elastic modulus , the isothermal compression modulus @xcite , @xmath398 characterizing the elastic response with respect to a volumetric ( dilatational ) strain fluctuation as shown in fig . [ fig_sketch](d ) . as discussed in the standard textbooks @xcite , the extensive variable is here the volume @xmath399 and the conjugated intensive variable the negative pressure @xmath400 where @xmath401 and @xmath402 stand for the instantaneous values of the volume and the pressure . with these notations eq . ( [ eq_pkv ] ) is obviously consistent with eq . ( [ eq_ftom ] ) or eq . ( [ eq_htom ] ) . in analogy with @xmath48 for the shear modulus @xmath0 , the compression modulus @xmath20 may be determined in an @xmath155- or an @xmath50-ensemble using the linear regression relation @xmath403 where the index @xmath404 indicates that both the measured instantaneous volume @xmath401 and pressure @xmath402 have been used . the corresponding correlation coefficient is @xmath405 since from eq . ( [ eq_dxdi ] ) we have @xmath406 if @xmath6 is imposed , this implies that @xmath407 should be identical to the better known relation @xcite @xmath408 which corresponds to the operational definition @xmath151 discussed above . the compression modulus @xmath20 may of course also be determined using the pressure fluctuations in systems of imposed volume @xmath12 as sketched in fig . [ fig_sketch](c ) . as ( to our knowledge ) first derived by rowlinson @xcite , the corresponding stress fluctuation formula reads @xmath409 being the fluctuation of the total normal pressure ( corresponding to @xmath55 ) and @xmath259 the affine dilatational elasticity " ( corresponding to @xmath26 ) . we shall give below a proper operational definition for @xmath259 . here we only note that due to the general legendre transformation eq . ( [ eq_didi ] ) for intensive variables one knows immediately that @xmath410 this implies by comparison with eq . ( [ eq_kmodpp ] ) that @xmath411 in analogy to eq . ( [ eq_muamuftau ] ) for the shear , i.e. @xmath259 can be directly be determined from the pressure fluctuations in @xmath155- and @xmath50-ensembles . as before for the correlation coefficient @xmath90 , eq . ( [ eq_cmodgt_mua ] ) , it follows from eq . ( [ eq_etaaetafp ] ) that @xmath412 since @xmath413 , the affine dilatational elasticity sets an upper bound to the compression modulus @xmath20 . we reformulate here briefly the derivation of the stress fluctuation relation @xmath258 for the compression modulus @xmath20 of an isotropic system at imposed volume @xmath12 by rowlinson @xcite . with @xmath414 being the volume of the unperturbed simulation box we assume a small relative volume change @xmath415 with all coordinates equally deformed as , e.g. @xmath416 for the @xmath105-coordinate in @xmath19-dimensions . ( the argument @xmath99 denotes again the reference system . ) using a similar rescaling trick as for the shear modulus @xmath0 in sec . [ theo_stressfluctu ] , the interaction energy @xmath417 of a configuration @xmath98 of the strained system is expressed in terms of the coordinates ( state ) of the unperturbed system and the explicit metric parameter @xmath229 . due to the volume change we have now also to take into account the ideal gas ( kinetic ) contributions . we derive first the excess contribution @xmath29 to the total pressure @xmath418 and the excess contribution @xmath419 to the total compression modulus @xmath420 valid for an arbitrary conservative potential . the general relations eqs . ( [ eq_dlogzdgam]-[eq_ddzddgam ] ) stated above for an imposed shear strain @xmath3 still hold with the relative volume change @xmath229 replacing @xmath3 . using eq . ( [ eq_dzdgam ] ) it follows for the excess pressure that @xmath421 _ defining _ the instantaneous excess pressure @xcite . in the second step we have taken the limit @xmath422 and have dropped the argument @xmath99 . the average is again evaluated using the weights of the unperturbed system , eq . ( [ eq_average ] ) . using eq . ( [ eq_ddlogzddgam ] ) and eq . ( [ eq_ddzddgam ] ) one obtains for the compression modulus @xmath423 which is again understood to be taken at @xmath424 . the first term @xmath425 in eq . ( [ eq_kexstressfluctu ] ) corresponds to the change of the system energy assuming an affine strain transformation for all particle positions . it gives the excess contribution @xmath426 to the total affine dilatational elasticity @xmath427 . the second term corresponds to the excess contribution @xmath428 to the total normal pressure fluctuation @xmath429 mentioned in sec . [ app_kstrainfluctu ] . it corrects the overprediction of the affine strain contribution @xmath426 @xcite . we remind that for an ideal gas the isothermal compression modulus @xmath430 is given by the ideal gas pressure @xmath145 and that the ideal gas pressure fluctuation @xmath431 becomes @xmath432 at constant volume @xcite . interestingly , using again the general legendre transform eq . ( [ eq_didi ] ) one sees that @xmath433 for the ideal gas pressure fluctuations at imposed pressure . due the assumed additivity of the total system hamiltonian , the stress fluctuation formula for the total compression modulus may now be written @xmath434 while this result is useful for constant-@xmath12 ensembles , especially if mc simulations are used , it is yet not in a form exactly equivalent to the general legendre transform eq . ( [ eq_kmodpp ] ) . since the ideal and excess pressure fluctuations are decoupled , @xmath435 , this implies @xmath429 . one may thus rewrite eq . ( [ eq_kconstantv ] ) as @xmath436 with @xmath437 where we have used eq . ( [ eq_etafidp ] ) in the second step . assuming eq . ( [ eq_etaaexid ] ) assures that @xmath259 becomes equivalent to @xmath438 . this has the nice feature that not only the functional @xmath258 , eq . ( [ eq_kmodpp ] ) , should give the correct compression modulus @xmath20 in the @xmath37 and the @xmath49 ensembles , but also that @xmath258 must vanish properly if the wrong " @xmath155 and @xmath155 ensembles are used , just as @xmath54 vanishes for @xmath49- or @xmath50-ensembles . up to now we have not taken advantage of the fact that the conservative interaction potential is assumed to be a pair potential , eq . ( [ eq_upairinteraction ] ) . this will be used now to express @xmath29 and @xmath426 in terms of simple ( ensemble independent ) two - point correlation functions . equation ( [ eq_strain_x_epsilon ] ) implies that the squared distance @xmath114 between two particles transforms as @xmath439 it follows that @xmath440 where we have taken finally for both derivatives the limit @xmath422 . for the first two derivatives of a general function @xmath441 with respect to @xmath229 this implies @xmath442 for @xmath422 and dropping the argument @xmath99 on the right - hand sides . according to the general result eq . ( [ eq_peshat_us ] ) for the normal pressure and using eq . ( [ eq_strain_fronebeps ] ) one confirms for pair interactions the expected virial relation @xcite @xmath443 for the instantaneous excess pressure where the interactions between pairs of particles @xmath133 are again labeled by an index @xmath130 . using eq . ( [ eq_strain_rtwo_epsilon ] ) the excess contribution to the affine dilatational elasticity becomes @xmath444 is sometimes called the hypervirial " contribution to the compression modulus @xcite . it corresponds to the first term indicated in eq . ( [ eq_strain_frtwobeps ] ) . by comparing eq . ( [ eq_etab ] ) with eq . ( [ eq_mub ] ) it is readily seen that for isotropic systems @xmath445 i.e. @xmath256 and @xmath28 and , hence , the affine dilatational elasticity @xmath259 and affine shear elasticity @xmath26 are closely related . ( this justifies to call @xmath256 a born coefficient " . ) please note that eq . ( [ eq_kmodpp ] ) or eq . ( [ eq_kconstantv ] ) together with eq . ( [ eq_etaaex ] ) are perfectly consistent with the relations stated in the literature @xcite . we emphasize finally that , as discussed for the affine shear elasticity @xmath26 in sec . [ theo_stressfluctu ] , it is inconsistent to neglect the explicit excess pressure contribution to @xmath259 in eq . ( [ eq_etaaex ] ) but to keep the underlined term of the born coefficient @xmath256 , eq . ( [ eq_etab ] ) @xcite . the affine dilatational elasticity @xmath446 obtained for both models is presented in the inset of fig . [ fig_kt ] as a function of the reduced temperature @xmath447 . ( the values may also be found in table [ tab_ka ] and table [ tab_plj ] . ) we remind that for determining the born coefficient @xmath256 impulsive corrections need to be considered as discussed at the end of sec . [ theo_trunc ] . as shown by the small filled triangles , @xmath259 compares well with the total pressure fluctuation @xmath448 in an @xmath50-ensemble . the observed data collapse is expected from eq . ( [ eq_etaaetafp ] ) and eq . ( [ eq_etafidp ] ) . as indicated by bold and dashed lines for , respectively , the low- and high-@xmath5 limit @xmath259 decreases essentially linearly with temperature . the coefficients of the lines ( not shown ) are consistent with eq . ( [ eq_etabmub ] ) relating @xmath259 and @xmath26 and the fits given in fig . [ fig_muamuf_t ] for @xmath26 . we emphasize that the affine dilatational elasticity @xmath259 ( and thus @xmath26 ) is not only larger for the ka model but , more importantly that , is also increases more rapidly with decreasing @xmath5 than the plj model . the main panel of fig . [ fig_kt ] shows the best long - time values of the compression modulus @xmath20 for both models . the vertical axis is rescaled by @xmath259 . the rescaled stress fluctuation formula , eq . ( [ eq_kconstantv ] ) , for the @xmath37-ensemble is given for both models . for the plj model we also indicate @xmath407 ( crosses ) and @xmath449 ( large triangles ) for the @xmath50-ensemble . no rescaling with @xmath259 is necessary for the squared correlation coefficient @xmath450 ( stars ) also presented . a perfect collapse of all plj data is found confirming the equivalence of all operational definitions used . for both models we observe a strong increase of @xmath451 around @xmath452 with decreasing temperature . interestingly , the rescaled data of both models is similar but not identical . the mentioned increase around @xmath452 is clearly more pronounced for the ka model . note that if the unscaled @xmath20 is directly plotted against @xmath5 , this effect becomes even stronger ( not shown ) consistently with the scaling of @xmath259 presented in the inset . we note finally that for very low temperatures we observe for both models that the ratio @xmath453 approaches unity , i.e. the compression modulus is essentially dominated by the affine response . we remind that this is different for the reduced shear modulus for both models which is observed to approach @xmath454 in the low-@xmath5 limit ( fig . [ fig_gt_rescal ] ) , i.e. the shear stress fluctuations @xmath364 are more relevant than the pressure fluctuations @xmath455 in this limit . for the plj model we present in fig . [ fig_kt_thigh ] the sampling time dependence of various operation definitions of @xmath20 for the liquid limit at temperature @xmath456 . the horizontal axis indicates the sampling time @xmath32 in units of mcs , the vertical axis is made dimensionless using @xmath259 as natural scale . as shown for @xmath407 and @xmath449 obtained in the @xmath50-ensemble and for @xmath258 obtained using eq . ( [ eq_kconstantv ] ) in the @xmath37-ensemble , all operational definitions applied in their natural ensemble converge _ rapidly _ within @xmath457 mcs to the same plateau . ( this is a general finding for all temperatures . ) the plateau value for @xmath456 is indicated by the bold solid line . note that @xmath451 is smaller unity as expected from the discussion at the end of sec . [ app_kstrainfluctu ] . the small filled symbols refer to @xmath258 computed in the wrong " @xmath50-ensemble . the small diamonds have been computed using directly eq . ( [ eq_kconstantv ] ) . as indicated by the dashed line , this yields @xmath458 for large times . only if we compute @xmath436 using eq . ( [ eq_etaaexid ] ) , it does vanish properly as shown by the small triangles . this shows numerically that eq . ( [ eq_kconstantv ] ) , albeit perfectly fine for constant-@xmath12 ensembles , is just not the correct formulation allowing to make manifest the general legendre transformation eq . ( [ eq_etaftransform ] ) . this is , however , achieved using eq . ( [ eq_etaaexid ] ) . this finding holds rigorously for all temperatures as can be seen from the scaling of the small triangles in the inset of fig . [ fig_kt ] . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: many solids have a crystalline structure . however , ideal crystals are absent in nature , and most of their physical properties , such as plasticity , melting , growth , etc . , are defined by defects of the crystalline structure . therefore , a study of defects is a topical scientific question of importance for applications in the first place . a broad experimental and theoretical investigations of defects in crystals started in the 1930s and continues nowadays . at present , a fundamental theory of defects is absent in spite of the existence of dozens of monographs and thousands of articles . one of the most promising approaches to the theory of defects is based on riemann cartan geometry , which involves nontrivial metric and torsion . in this approach , a crystal is considered as a continuous elastic medium with a spin structure . if the displacement vector field is a smooth function , then there are only elastic stresses corresponding to diffeomorphisms of the euclidean space . if the displacement vector field has discontinuities , then we are saying that there are defects in the elastic structure . defects in the elastic structure are called dislocations and lead to the appearance of nontrivial geometry . precisely , they correspond to a nonzero torsion tensor , equal to the surface density of the burgers vector . the idea to relate torsion to dislocations appeared in the 1950s [ 14 ] . this approach is still being successfully developed ( note reviews [ 511 ] ) , and is often called the gauge theory of dislocations . a similar approach is also being developed in gravity @xcite . it is interesting to note that e cartan introduced torsion in geometry @xcite having the analogy with mechanics of elastic media in mind . the gauge approach to the theory of defects is being developed successfully , and interesting results are being obtained in this way [ 1417 ] . we note in this connection two respects in which the approach proposed below is essentially different . in the gauge models of dislocations based on the translational group or on the semidirect product of the rotational group with translations , one usually chooses the distortion and displacement fields as independent variables . it is always possible to fix the invariance under local translations such that the displacement field becomes zero because it transforms by simple translation under the action of the translational group . in this sense , the displacement field is the gauge parameter of local translations , and physical observables are independent of it in gauge - invariant models . the other disadvantage of the gauge approach is the equations of equilibrium . einstein type equations are usually considered for distortion or vielbein , with the right hand side depending on the stress tensor . this appears unacceptable from the physical point of view because of the following reason . consider , for example , one straight edge dislocation . in this case , the elastic stress field differs from zero everywhere . then the torsion tensor ( or curvature ) is also nontrivial in the whole space due to the equations of equilibrium . this is wrong from our point of view . indeed , we can consider an arbitrary domain of medium outside the cutting surface and look at the creation process for an edge dislocation . the chosen domain was a part of the euclidean space with identically zero torsion and curvature before the defect creation . it is clear that torsion and curvature remain zero because the process of dislocation formation is a diffeomorphism for the considered domain . in addition , the cutting surface may be chosen arbitrary for the defect creation , leaving the dislocation axis unchanged . then it follows that torsion and curvature must be zero everywhere except at the axis of dislocation . in other words , the elasticity stress tensor can not be the source of dislocations . to avoid the apparent contradiction , we propose a radical way out : we do not use the displacement field as an independent variable at all . this does not mean that the displacement field does not exist in real crystals . in the proposed approach , the displacement field exists and can be computed in those regions of medium that do not contain cores of dislocations . in this case , it satisfies the equations of nonlinear elasticity theory . the proposed geometric approach allows considering other defects that do not relate directly to defects of elastic media . the intensive investigations of other defects were conducted in parallel with the study of dislocations . the point is that many solids have not only elastic properties but also a spin structure . for example , there are ferromagnets , liquid crystals , spin glasses . in this case , there are defects in the spin structure which are called disclinations @xcite . they arise when the director field has discontinuities . the presence of disclinations is also connected to nontrivial geometry . namely , the curvature tensor equals the surface density of the frank vector . the gauge approach based on the rotational group @xmath0 was also used for describing disclinations @xcite . @xmath0-gauge models of spin glasses with defects were considered in @xcite . the geometric theory of static distribution of defects which describes both types of defects dislocation and disclinations from a single standpoint was proposed in @xcite . in contrast to other approaches , it involves the vielbein and @xmath0 connection as the only independent variables . the torsion and curvature tensors have direct physical meaning as the surface densities of dislocations and disclinations , respectively . covariant equations of equilibrium for the vielbein and @xmath0 connection similar to those in a gravity model with torsion are postulated . to define the solution uniquely , we must fix the coordinate system ( fix the gauge ) because any solution of the equations of equilibrium is defined up to general coordinate transformations and local @xmath0 rotations . the elastic gauge for the vielbein @xcite and lorentz gauge for the @xmath0 connection @xcite were proposed recently . we stress that the notions of a displacement vector and rotational angle are completely absent in our approach . these notions can be introduced only in those domains where defects are absent . in this case , equations for vielbein and @xmath0 connection are identically satisfied , the elastic gauge reduces to the equations of nonlinear elasticity theory for the displacement vector , and the lorentz gauge leads to the equations for the principal chiral @xmath0 field . in other words , to fix the coordinate system , we choose two fundamental models : the elasticity theory and the principal chiral field model . to show the advantages of the geometric approach and to compare it with the elasticity theory , we consider in detail a wedge dislocation in the frameworks of the elasticity theory and the proposed geometric model . we show that the explicit expression for the metric in the geometric approach is simpler and coincides with the induced metric obtained within the elasticity theory only for small relative deformations . as an application of the geometric theory of defects , we consider two examples in the last sections of the present review . first , we solve the problem of phonon scattering on a wedge dislocation . the problem of phonon scattering is reduced to the integration of equations for extremals for the metric describing a wedge dislocation ( because phonons move along extremals in the eikonal approximation ) . as a second application , we consider the quantum mechanical problem of impurity or vacancy motion inside a cylinder whose axis coincides with a wedge dislocation . the wave functions and energetic spectrum of the impurity are found explicitly . the presence of defects results in nontrivial riemann cartan geometry . this means that for describing the phenomena that relate ingeniously to elastic media , we must make changes in the corresponding equations . for example , if a phonon propagation in an ideal crystal is described by the wave equation , then the presence of defects is easily taken into account . for this , the flat euclidean metric has to be replaced by a nontrivial metric describing the distribution of defects . the same substitution must be made in the schrdinger equation to describe other quantum effects . it is shown nowadays that the presence of defects essentially influences physical phenomena . the schrdinger equation in the presence of dislocations was considered in [ 2546 ] for different problems . problems related to the wave or laplace equations were considered in [ 4754 ] . the influence of the nontrivial metric related to the presence of defects was investigated in electrodynamics @xcite and hydrodynamics @xcite . scattering of phonons on straight parallel dislocations was studied in [ 5759 ] . another approach to the theory of defects based on affine geometry with nonzero nonmetricity tensor was considered recently in @xcite . we consider infinite three dimensional elastic media . we suppose that the undeformed medium in the defect - free case is invariant under translations and rotations in some coordinate system . then , the medium in this coordinate system @xmath1 , @xmath2 , is described by the euclidean metric @xmath3 , and the system of coordinates is called cartesian . thus , in the undeformed state , we have the euclidean space @xmath4 with a given cartesian coordinate system . we also assume that torsion ( see the appendix ) in the medium equals zero . let a point of the medium has coordinates @xmath1 in the ground state . after deformation , this point has the coordinates @xmath5 .45 in the initial coordinate system , see fig . [ feldef ] . the inverse notation is used in the elasticity theory . one usually writes @xmath6 . these are equivalent because both coordinate systems @xmath7 and @xmath1 cover the whole @xmath4 . however , in the theory of defects considered in the next sections , the situation is different . generally , the elastic medium fills the whole euclidean space only in the final state . here and in what follows , we assume that fields depend on coordinates @xmath8 that are coordinates of points of the medium after the deformation and cover the whole euclidean space @xmath4 . in the presence of dislocations , the coordinates @xmath1 do not cover the whole @xmath4 in the general case because part of the media may be removed or , conversely , added . therefore , the system of coordinates related to points of the medium after an elastic deformation and defect creation is more preferable . in the linear elasticity theory , relative deformations are assumed to be small ( @xmath9 ) . the functions @xmath10 are then components of a vector field that is called the displacement vector field and is the basic variable in elasticity theory . in the absence of defects , we assume that the displacement field is a smooth vector field in the euclidean space @xmath4 . the presence of discontinuities and singularities of the displacement field is interpreted as a presence of defects in elastic media . in what follows , we consider only static deformations with the displacement field @xmath11 independent of time . then the basic equations of equilibrium for small deformations are ( see , e.g. , @xcite ) @xmath12 where @xmath13 is the stress tensor , which is assumed to be symmetric . the tensor of small deformations @xmath14 is given by the symmetrized partial derivative of the displacement vector : @xmath15 lowering and raising of the latin indices is performed with the euclidean metric @xmath16 and its inverse @xmath17 . the letters @xmath18 and @xmath19 denote constants characterizing elastic properties of media and are called lam coefficients . functions @xmath20 describe the total density of nonelastic forces inside the medium . we assume in what follows that such forces are absent : @xmath21 . equation ( [ estati ] ) is newton s law , and eqn ( [ ehook ] ) is hook s law relating stresses to deformations . in a cartesian coordinate system and for small deformations , the difference between upper and lower indices disappears because raising and lowering of indices is performed with the help of the euclidean metric . one usually forgets about this difference due to this reason , and this is fully justified . but in the presence of defects , the notion of cartesian coordinate system and euclidean metric is absent , and the indices are raised and lowered with the help of riemannian metric . therefore , we distinguish the upper and lower indices as is accepted in differential geometry , having the transition to elastic media with defects in mind . the main problem in the linear elasticity theory is the solution of the second - order equations for the displacement vector that arise after substitution of ( [ ehook ] ) into ( [ estati ] ) with some boundary conditions . many known solutions are in good agreement with experiment . therefore , one may say that equations ( [ ehook ] ) , ( [ estati ] ) have a solid experimental background . we now look at the elastic deformations from the standpoint of differential geometry . from the mathematical standpoint , map ( [ eeldef ] ) is itself a diffeomorphism of the euclidean space @xmath4 . the euclidean metric @xmath16 is then induced by the map @xmath22 . it means that in the deformed state , the metric in the linear approximation is given by @xmath23 i.e. , is defined by the tensor of small deformations ( [ edefte ] ) . we note that in the linear approximation , @xmath24 and @xmath25 . in riemannian geometry , the metric uniquely defines the levi civita connection @xmath26 ( christoffel s symbols ) , eqn ( [ etigam ] ) . we can compute curvature tensor ( [ ecurva ] ) for these symbols . this tensor equals identically zero , @xmath27 , because the curvature of the euclidean space is zero , and the map @xmath22 is a diffeomorphism . the torsion tensor is equal to zero for the same reason . thus , an elastic deformation of the medium corresponds to the trivial riemann cartan geometry , with zero curvature and torsion tensors . the physical interpretation of metric ( [ emetri ] ) is as follows . the external observer fixes cartesian coordinate system corresponding to the ground undeformed state of the medium . the medium is then deformed , and external observer discovers that the metric becomes nontrivial in this coordinate system . if we assume that elastic perturbations in the medium ( phonons ) propagate along extremals ( lines of minimal length ) , then their trajectories in the deformed medium are defined by eqns ( [ exteqr ] ) . trajectories of phonons are now not straight lines because the christoffel s symbols are nontrivial ( @xmath28 ) . in this sense , metric ( [ emetri ] ) is observable . here , we see the essential role of the cartesian coordinate system @xmath1 defined by the undeformed state , with which the measurement process is connected . we assume that the metric @xmath29 given in the cartesian coordinates corresponds to some state of elastic media without defects . the displacement vector is then defined by the system of equations ( [ emetri ] ) , and its integrability conditions are the equality of the curvature tensor to zero , in accordance with theorem [ tlocev ] in the appendix . in the linear approximation , these conditions are known in elasticity theory as the saint venant integrability conditions . we make a remark that is important for the following consideration . for appropriate boundary conditions , the solution of the elasticity theory equations ( [ estati ] ) , ( [ ehook ] ) is unique . from the geometric standpoint , this means that elasticity theory fixes diffeomorphisms . this fact is used in the geometric theory of defects . equations of nonlinear elasticity theory written in terms of the metric or vielbein are used for fixing the coordinate system . we start with the description of linear dislocations in elastic media ( see , e.g. , @xcite ) . the simplest and most undespread examples of linear dislocations are shown in fig . [ fdislo ] . we cut the medium along the half - plane @xmath30 , @xmath31 , move the upper part of the medium located over the cut @xmath32 , @xmath31 by the vector @xmath33 towards the dislocation axis @xmath34 , and glue the cutting surfaces . the vector @xmath33 is called the burgers vector . in the general case , the burgers vector may not be constant on the cut . for the edge dislocation , it varies from zero to some constant value @xmath33 as it moves from the dislocation axis . after the gluing , the media comes to the equilibrium state called the edge dislocation , see fig . [ fdislo]_a_. if the burgers vector is parallel to the dislocation line , it is called the screw dislocation ( fig . [ fdislo]_b _ ) . the same dislocation can be made in different ways . for example , if the burgers vector is perpendicular to the cutting plane and directed from it in the considered cases , then the produced cavity must be filled with medium before gluing . it is easy to imagine that the edge dislocation is also obtained as a result , but rotated by the angle @xmath35 around the @xmath34 axis . this example shows that a dislocation is characterized not by the cutting surface but by the dislocation line and the burgers vector . straight linear dislocations . ( _ a _ ) the edge dislocation . the burgers vector @xmath33 is perpendicular to the dislocation line . ( _ b _ ) the screw dislocation . the burgers vector @xmath33 is parallel to the dislocation line.,title="fig : " ] straight linear dislocations . ( _ a _ ) the edge dislocation . the burgers vector @xmath33 is perpendicular to the dislocation line . ( _ b _ ) the screw dislocation . the burgers vector @xmath33 is parallel to the dislocation line.,title="fig : " ] + from the topological standpoint , the medium containing several dislocations or even the infinite number of them is the euclidean space @xmath4 . in contrast to the case of elastic deformations , the displacement vector in the presence of dislocations is no longer a smooth function because of the presence of cutting surfaces . at the same time , we assume that partial derivatives of the displacement vector @xmath36 ( the distortion tensor ) are smooth functions on the cutting surface . this assumption is justified physically because these derivatives define deformation tensor ( [ edefte ] ) . in its turn , partial derivatives of the deformation tensor must exist and be smooth functions in the equilibrium state everywhere except , possibly , the dislocation axis , because otherwise equations of equilibrium ( [ estati ] ) have no meaning . we assume that the metric and vielbein are smooth functions everywhere in @xmath4 except , may be , dislocation axes , because the deformation tensor defines the induced metric ( [ emetri ] ) . the main idea of the geometric approach amounts to the following . to describe single dislocations in the framework of elasticity theory we must solve equations for the displacement vector with some boundary conditions on the cuts . this is possible for small number of dislocations . but , with an increasing number of dislocations , the boundary conditions become so complicated that the solution of the problem becomes unrealistic . besides , one and the same dislocation can be created by different cuts which leads to an ambiguity in the displacement vector field . another shortcoming of this approach is that it can not be applied to the description of a continuous distribution of dislocations because the displacement vector field does not exist in this case at all because it must have discontinuities at every point . in the geometric approach , the basic variable is the vielbein which by assumption is a smooth function everywhere except , possibly , dislocation axes . we postulate new equations for the vielbein ( see section [ sequil ] ) . in the geometric approach , the transition from a finite number of dislocations to their continuous distribution is simple and natural . in that way , the smoothing of singularities occurs on dislocation axes in analogy with smoothing of mass distribution for point particles in passing to continuous media . we now develop the formalism of the geometric approach . in a general defect - present case , we do not have a preferred cartesian coordinate frame in the equilibrium because there is no symmetry . therefore , we consider arbitrary coordinates @xmath37 , @xmath38 , in @xmath4 . we use greek letters for coordinates allowing arbitrary coordinate changes . then the burgers vector can be expressed as the integral of the displacement vector @xmath39 where @xmath40 is a closed contour surrounding the dislocation axis , fig . [ fburco ] . .45 this integral is invariant under arbitrary coordinate transformations @xmath41 and covariant under global @xmath0-rotations of @xmath1 . here , components of the displacement vector field @xmath42 are considered with respect to the orthonormal basis in the tangent space , @xmath43 . if components of the displacement vector field are considered with respect to the coordinate basis @xmath44 , the invariance of the integral ( [ eburge ] ) under general coordinate changes is violated . in the geometric approach , we introduce new independent variable the vielbein instead of partial derivatives @xmath45 @xmath46 the vielbein is a smooth function on the cut by construction . we note that if the vielbein was simply defined as the partial derivative @xmath47 , then it would have the @xmath48-function singularity on the cut because functions @xmath49 have a jump . the burgers vector can be expressed through the integral over a surface @xmath50 having contour @xmath40 as the boundary @xmath51 where @xmath52 is the surface element . as a consequence of the definition of the vielbein in ( [ edevid ] ) , the integrand is equal to zero everywhere except at the dislocation axis . for the edge dislocation with constant burgers vector , the integrand has a @xmath48-function singularity at the origin . the criterion for the presence of a dislocation is a violation of the integrability conditions for the system of equations @xmath53 : @xmath54 if dislocations are absent , then the functions @xmath49 exist and define transformation to a cartesian coordinates frame . in the geometric theory of defects , the field @xmath55 is identified with the vielbein . next , we compare the integrand in ( [ eburg2 ] ) with the expression for the torsion in cartan variables ( [ ecurcv ] ) . they differ only by terms containing the @xmath0 connection . this is the ground for the introduction of the following postulate . in the geometric theory of defects , the burgers vector corresponding to a surface @xmath50 is defined by the integral of the torsion tensor : @xmath56 this definition is invariant with respect to general coordinate transformations of @xmath37 and covariant with respect to global rotations . thus , the torsion tensor has straightforward physical interpretation : it is equal to the surface density of the burgers vector . the physical interpretation of the @xmath0 connection is given in section [ sdiscl ] , and now we show how this definition reduces to the expression for the burgers vector ( [ eburg2 ] ) obtained within elasticity theory . if the curvature tensor for the @xmath0 connection is zero , then , according to theorem [ tlotrs ] , the connection is locally trivial , and there exists such @xmath0 rotation such that @xmath57 . in this case , we return to expression ( [ eburg2 ] ) . if the @xmath0 connection is zero and vielbein is a smooth function , then the burgers vector corresponds uniquely to every contour . it can then be expressed as a surface integral of the torsion tensor . the surface integral depends only on the boundary contour but not on the surface due to the stokes theorem . .45 we have shown that the presence of linear defects results in a nontrivial torsion tensor . in the geometric theory of defects , the equality of the torsion tensor to zero @xmath58 is naturally considered the criterion for the absence of dislocations . then , under the name dislocation fall not only linear dislocations but , in fact , arbitrary defects in elastic media . for example , point defects ; vacancies and impurities , are also dislocations . in the first case we cut out a ball from the euclidean space @xmath4 and then shrink the boundary sphere to a point ( fig . [ fpoide ] ) . in the case of impurity , a point of the euclidean space is blown up to a sphere and the produced cavity is filled with the medium . point defects are characterized by the mass of the removed or added media , which is also defined by the vielbein @xcite @xmath59 where @xmath49 are the transition functions to a cartesian coordinate frame in @xmath4 and @xmath60 is the density of the medium which is supposed to be constant . the mass is defined by the difference of two integrals , each of them being separately divergent . the first integral equals to the volume of the medium with defects and the second is equal to the volume of the euclidean space . the torsion tensor for a vacancy or impurity is zero everywhere except at one point , where it has a @xmath48-function singularity . for point defects , the notion of the burgers vector is absent . according to the given definition , the mass of an impurity is positive because the matter is added to the media , and the mass of a vacancy is negative because part of the medium is removed . the negative sign of the mass causes serious problems for a physical interpretation of solutions of the equations of motion or the schrdinger equation . hence , we make a remark . strictly speaking , the integral ( [ emass ] ) should be called the `` bare '' mass because this expression does not account for elastic stresses arising around a point dislocation . the effective mass of such a defect must contain at least two contributions : the bare mass and the free energy coming from elastic stresses . the question about the sign of the effective mass is not solved yet and demands a separate analysis . surface defects may also exist in three - dimensional space , in addition to point and line dislocations . in the geometric approach , all of them are called dislocations because they correspond to a nontrivial torsion . in the preceeding section , we related dislocations to a nontrivial torsion tensor . for this , we introduced an @xmath0 connection . now we show that the curvature tensor for the @xmath0 connection defines the surface density of the frank vector characterizing other well - known defects disclinations in the spin structure of media @xcite . let a unit vector field @xmath61 @xmath62 be given at all points of the medium . for example , @xmath63 has the meaning of the magnetic moment located at each point of the medium for ferromagnets ( fig . [ fspstr]_a _ ) . for nematic liquid crystals , the unit vector field @xmath63 with the equivalence relation @xmath64 describes the director field ( fig . [ fspstr]_b _ ) . examples of media with the spin structure : ( _ a _ ) ferromagnet , ( _ b _ ) liquid crystal.,title="fig : " ] examples of media with the spin structure : ( _ a _ ) ferromagnet , ( _ b _ ) liquid crystal.,title="fig : " ] + we fix some direction in the medium @xmath65 . then the field @xmath61 at a point @xmath8 can be uniquely defined by the field @xmath66 taking values in the rotation algebra @xmath67 ( the rotation angle ) , @xmath68 where @xmath69 is the rotation matrix corresponding to the algebra element @xmath70 . here , we use the following parameterization of the rotation group @xmath0 by elements of its algebra ( see , e.g. , @xcite ) @xmath71 where @xmath72 and @xmath73 is the modulus of the vector @xmath74 . the pseudovector @xmath75 , where @xmath76 is the totally antisymmetric third - rank tensor , @xmath77 , is directed along the rotation axis and its length equals the rotation angle . we call the field @xmath70 spin structure of the media . if a media has a spin structure , then it may have defects called disclinations . for linear disclinations parallel to the @xmath34 axis , the vector field @xmath78 lies in the perpendicular plain @xmath79 . the simplest examples of linear disclinations are shown in fig . [ fdiscl ] . the vector field distributions on the plane @xmath79 for the linear disclinations parallel to the @xmath34 axis . ( _ a _ ) @xmath80 . ( _ b _ ) @xmath81.,title="fig : " ] the vector field distributions on the plane @xmath79 for the linear disclinations parallel to the @xmath34 axis . ( _ a _ ) @xmath80 . ( _ b _ ) @xmath81.,title="fig : " ] + every linear disclination is characterized by the frank vector @xmath82 where @xmath83 and the integral is taken along closed contour @xmath40 surrounding the disclination axis . the length of the frank vector is equal to the total angle of rotation of the field @xmath63 as it goes around the disclination . the vector field @xmath63 defines a map of the euclidean space to a sphere @xmath84 . for linear disclinations parallel to the @xmath34 axis , this map is restricted to a map of the plane @xmath85 to a circle @xmath86 . in this case , the total rotation angle must obviously be a multiple of @xmath87 . for nematic liquid crystals , we have the equivalence relation @xmath64 . therefore , for linear disclinations parallel to the @xmath34 axis , the director field defines a map of the plane into the projective line @xmath88 . in this case , the length of the frank vector must be a multiple of @xmath89 . the corresponding examples of disclinations are shown in fig . [ fdisc2 ] . the director field distribution in the @xmath79 plane for the linear disclinations parallel to the @xmath34 axis . ( _ a _ ) @xmath90 and ( _ b _ ) @xmath91.,title="fig : " ] the director field distribution in the @xmath79 plane for the linear disclinations parallel to the @xmath34 axis . ( _ a _ ) @xmath90 and ( _ b _ ) @xmath91.,title="fig : " ] + as for the displacement field , the field @xmath92 , taking values in the algebra @xmath67 , is not a smooth function in @xmath4 in the presence of disclinations . we make a cut in @xmath4 bounded by the disclination axis . then the field @xmath92 may be considered smooth in the whole space except the cut . we assume that all partial derivatives of @xmath92 have the same limit as it approaches the cut from both sides . then we define the new field @xmath93 the functions @xmath94 are smooth everywhere by construction except , may be , on the disclination axis . then the frank vector may be expressed as the surface integral @xmath95 where @xmath50 is an arbitrary surface having the contour @xmath40 as the boundary . if the field @xmath94 is given , then the integrability conditions for the system of equations @xmath96 are @xmath97 this noncovariant equality yields the criterion for the absence of disclinations . in the geometric theory of defects , we identify the field @xmath94 with the @xmath0 connection . in the expression for the curvature in ( [ ecucav ] ) , the first two terms coincide with ( [ eomcon ] ) , and we therefore postulate the covariant criterion of the absence of disclinations as the equality of the curvature tensor for the @xmath0 connection to zero : @xmath98 simultaneously , we give the physical interpretation of the curvature tensor as the surface density of the frank vector @xmath99 this definition reduces to the previous expression for the frank vector ( [ efrank ] ) in the case where rotation of the vector @xmath78 occures in a fixed plane . in this case , rotations are restricted by the subgroup @xmath100 . the quadratic terms in the expression for the curvature in ( [ ecucav ] ) disappear because the rotation group @xmath101 is abelian , and we obtain the previous expression for the frank vector ( [ efrank ] ) . thus , we described the media with dislocations ( defects of elastic media ) and disclinations ( defects in the spin structure ) in the framework of riemann cartan geometry , the torsion and curvature tensors being identified with the surface density of dislocations and disclinations , respectively . the relations between physical and geometrical notions are summarized in the table [ tdegeo ] . .[tdegeo ] the relation between physical and geometrical notions in the geometric theory of defects . [ cols="<,>,>",options="header " , ] the same physical interpretation of torsion and curvature was considered in @xcite . several possible functionals for the free energy were also considered . in the next section , we propose a new expression for the free energy . until now , we discussed only the relation between physical and geometrical notions . to complete the construction of the geometric theory of defects , we have to postulate equations of equilibrium describing static distribution of defects in media . the vielbein @xmath55 and @xmath0 connection @xmath94 are basic and independent variables in the geometric approach . in contrast to previous geometric approaches , we completely abandon the displacement field @xmath11 and spin structure @xmath70 as the fields entering the system of equilibrium equations . in a general case of a continuous distribution of defects , they simply do not exist . nevertheless , at some level and under definite circumstances they can be reconstructed , and we discuss this in the following section . the expression for the free energy was derived in @xcite . we assume that equations of equilibrium must be covariant under general coordinate transformations and local rotations , be at most of the second order , and follow from a variational principle . the expression for the free energy leading to the equilibrium equations must then be equal to a volume integral of the scalar function ( the lagrangian ) that is quadratic in torsion and curvature tensors . there are three independent invariants quadratic in the torsion tensor and three independent invariants quadratic in the curvature tensor in three dimensions . it is possible to add the scalar curvature and a `` cosmological '' constant @xmath102 . we thus obtain a general eight - parameter lagrangian @xmath103 where @xmath104 , @xmath105 and @xmath106 are some constants , and we have introduced the trace of torsion tensor @xmath107 and the ricci tensor @xmath108 . here and in what follows , transformation of the greek indices into the latin ones and vice versa is always performed using the vielbein and its inverse . for example , @xmath109 the particular feature of three dimensions is that the full curvature tensor is in a one - to - one correspondence with ricci tensor ( [ ecurir ] ) and has three irreducible components . therefore , the lagrangian contains only three independent invariants quadratic in curvature tensor . we do not need to add the hilbert einstein lagrangian @xmath110 , also yielding second - order equations , to the free energy ( [ egenla ] ) because of identity ( [ eidcur ] ) . thus , the most general lagrangian depends on eight constants and leads to very complicated equations of equilibrium . at present , we do not know precisely what values of the constants describe this or that medium . therefore , we make physically reasonable assumptions to simplify matters . namely , we require that equations of equilibrium must admit the following three types of solutions . 1 . there are solutions describing the media with only dislocations , @xmath111 . 2 . there are solutions describing the media with only disclinations , @xmath112 . 3 . there are solutions describing the media without dislocations and disclinations , @xmath113 . it turns out that these simple assumptions reduce the number of independent parameters from eight to two . we now prove this statement . the lagrangian ( [ egenla ] ) yields the equilibrium equations @xmath114 where the covariant derivative acts with the @xmath0 connection on the latin indices and with the christoffel symbols on the greek ones . for example , @xmath115 the first condition on the class of solutions of the equilibrium equations is that they permit solutions describing the presence of only dislocations in media . this means the existence of solutions with zero curvature tensor corresponding to the absence of disclinations . substitution of the condition @xmath116 into eqn ( [ econne ] ) for the @xmath117 connection yields @xmath118 here @xmath119 , @xmath120 , and @xmath121 are the irreducible components of the torsion tensor , @xmath122 where @xmath123 in a general case of dislocations all irreducible components of torsion tensor differ from zero ( @xmath119 , @xmath120 , @xmath124 ) and eqns ( [ ecotor ] ) have a unique solution @xmath125 for these coupling constants , the first four terms in lagrangian ( [ egenla ] ) are equal to the hilbert einstein lagrangian @xmath126 up to a total divergence due to identity ( [ eidcur ] ) . equation ( [ etriad ] ) then reduces to the einstein equations with a cosmological constant @xmath127 in this way , the first condition is satisfied . according to the second condition , the equations of equilibrium must allow solutions with zero torsion @xmath58 . in this case , the curvature tensor has additional symmetry @xmath128 , and eqn ( [ econne ] ) becomes @xmath129 here , we decompose the ricci tensor onto irreducible components , @xmath130 where @xmath131 note that for zero torsion , the ricci tensor is symmetrical : @xmath132 . contraction of eqn ( [ econre ] ) with @xmath133 leads to the equation @xmath134 in the general case of nonvanishing curvature , the covariant derivatives @xmath135 and @xmath136 differ from zero and are independent . therefore , we obtain two equations for the coupling constants , @xmath137 which have a unique solution @xmath138 in this case , eqn ( [ etriad ] ) for the vielbein corresponding to a nonzero torsion also reduces to einstein equations ( [ eeinst ] ) . the last requirement for the existence of solutions with zero curvature and torsion is satisfied only for the zero cosmological constant @xmath139 . therefore , the simple and physically reasonable requirements define the two parameter lagrangian @xcite @xmath140 which is the sum of the hilbert einstein lagrangian for the vielbein and the square of the antisymmetric part of the ricci tensor . we note that @xmath141 and @xmath142 are constructed for different metrical connections . other quadratic lagrangians were considered , for example , in @xcite . we note that they also contain the displacement vector as an independent variable along with the vielbein . expression ( [ elagfi ] ) defines the free energy density in the geometric theory of defects and leads to the equilibrium equations ( the euler lagrange equations ) . in our geometric approach , the displacement vector and spin structure do not enter expression ( [ elagfi ] ) for the free energy . in the geometric approach , the vielbein @xmath55 and @xmath0 connection @xmath143 are the only variables . the displacement field @xmath11 and the spin structure @xmath144 can be introduced only in those regions of media where defects are absent . as the consequence of the absence of disclinations @xmath145 , the @xmath0 connection is actually a pure gauge ( [ eloctr ] ) , i.e. , the spin structure @xmath144 exists . if , in addition , dislocations are absent ( @xmath58 ) then there is the displacement field such that the vielbein equals its partial derivatives ( [ erepad ] ) . in this and only in this case can we introduce the displacement field and spin structure . we show below that this can be done such that the equations of nonlinear elasticity theory and the principal chiral @xmath0 field are fulfilled . for free energy in ( [ elagfi ] ) , the euler lagrange equations are covariant under general coordinate transformations in @xmath4 and local @xmath0 rotations . this means that any solution of the equilibrium equations is defined up to diffeomorphisms and local rotations . for the geometric theory of defects to make predictions , we have to fix the coordinate system ( to fix the gauge in the language of gauge field theory ) . this allows us to choose a unique representative from each class of equivalent solutions . we say afterwards that this solution of the euler lagrange equations describes the distribution of defects in the laboratory coordinate system . we start with gauge - fixing the diffeomorphisms . for this , we choose the elastic gauge proposed in @xcite . this question is of primary importance , and we discuss it in detail . equations ( [ estati ] ) , ( [ ehook ] ) of elasticity theory for @xmath146 yield the second - order equation for the displacement vector @xmath147 where @xmath148 is the poisson ratio , ( @xmath149 ) , and @xmath150 is the laplace operator . it can be rewritten in terms of the induced metric ( [ emetri ] ) , for which we obtain a first - order equation . we choose precisely this equation as the gauge condition fixing the diffeomorphisms . we note that the gauge condition is not uniquely defined because the induced metric is nonlinear in the displacement vector , and different equations for the metric may have the same linear approximation . we give two possible choices , @xmath151 where we introduced the notation for the trace of metric @xmath152 . gauge conditions ( [ eonead ] ) and ( [ etwoad ] ) are understood in the following way . the metric @xmath153 is the euclidean metric written in an arbitrary coordinate system , for example , in the cylindrical or spherical coordinate system . the covariant derivative @xmath154 is built from the christoffel symbols corresponding to the metric @xmath153 , and @xmath155 as a consequence . the metric @xmath156 is the metric describing dislocations [ an exact solution of the equilibrium equations for the free energy ( [ elagfi ] ) ] . the gauge conditions differ because in the first and second cases the contraction is performed with the metric of dislocation @xmath157 and the euclidean metric @xmath158 , respectively , without changing the linear approximation . both gauge conditions yield eqn ( [ eqsepu ] ) in the linear approximation in the displacement vector ( [ emetri ] ) . this is most easily verified in cartesian coordinates . from the geometric standpoint , we have the following . the medium with dislocations is diffeomorphic to the euclidean space @xmath4 equipped with two metrics @xmath153 and @xmath156 . the metric @xmath153 is a flat euclidean metric written in an arbitrary coordinate system . the metric @xmath156 is not flat and describes the distribution of dislocations in the same coordinate system . in fact , the metric @xmath153 is used only to fix the coordinate system in which the metric @xmath156 is measured . if the solution of the equilibrium equations satisfies one of the gauge conditions ( [ eonead ] ) , ( [ etwoad ] ) , written , for example , in cylindrical coordinate system then we say that the solution is found in the cylindrical coordinates . we suppose here that the distribution of dislocations in elastic media in the laboratory cylindrical coordinate system is described by this particular solution . analogously , we may seek solutions in a cartesian , spherical , or any other coordinate system . gauge conditions may also be written for the vielbein @xmath55 , which is defined by eqn ( [ etetra ] ) . this involves additional arbitrariness because the vielbein is defined up to local rotations . this invariance leads to different linear approximations for the vielbein in terms of the displacement vector . we consider two possibilities in cartesian coordinates : @xmath159 where the index is lowered with the help of the kronecker symbol . for these possibilities and gauge condition ( [ etwoad ] ) we have two gauge conditions for the vielbein @xmath160 where @xmath161 . these conditions differ in the coefficient before the second term . we note that in a curvilinear coordinate system , the covariant derivative @xmath154 must also include the flat @xmath0 connection acting on indices @xmath162 . one can also write other possible gauge conditions having the same linear approximation . the question of the correct choice is unanswered at present and outside the scope of this review . at the moment , we want only to demonstrate that the system of coordinates must be fixed , and that the gauge condition depends on the poisson ratio , which is the experimentally observed quantity . gauge conditions ( [ ethrad])([efouad ] ) are first order equations by themselves and have some arbitrariness . therefore , to fix a solution uniquely , we must impose additional boundary conditions on the vielbein for any given problem . if the defect - free case , @xmath58 , @xmath163 , and the equilibrium equations are satisfied because the euler lagrange equations for ( [ elagfi ] ) are satisfied . in this and only in this case , we can introduce a displacement vector , and the elastic gauge reduces to the equations of nonlinear elasticity theory . in the presence of defects , a displacement field does not exist , and the elastic gauge simply defines the vielbein . in choosing the free energy functional , we required that the conditions @xmath116 and @xmath58 satisfy the euler lagrange equations . this is important because otherwise we would obtain an additional condition on the displacement vector ( the euler lagrange equations ) besides the elasticity theory equations following from the elastic gauge . we stress an important point once again . in the geometric theory of defects , we assume that there is a preferred laboratory coordinate system in which measurements are made . this coordinate system is related to the medium without defects and elastic stresses and corresponds to the flat euclidean space @xmath4 . gauge conditions ( [ eonead ] ) , ( [ etwoad ] ) and ( [ ethrad ] ) , ( [ efouad ] ) are written precisely in this euclidean space @xmath4 and contain a measurable quantity , the poisson ratio @xmath164 . this property essentially distinguishes the geometric theory of defects from the models of gravity in which all coordinate systems are considered equivalent . the elastic gauge is used to fix diffeomorphisms . the expression for the free energy in ( [ elagfi ] ) is also invariant under local @xmath0 rotations , and they must also be fixed . for this , we recently proposed the lorentz gauge for the connection @xcite @xmath165 this gauge is written in the laboratory cartesian coordinate system and has deep physical meaning . that is , let disclinations be absent ( @xmath163 ) . then the @xmath0 connection is a pure gauge : @xmath166 in this case , the lorentz gauge reduces to the principal chiral @xmath0-field equations @xmath167 for the spin structure @xmath92 . principal chiral field models ( see , e.g. , [ 6468 ] ) for different groups and in a different number of dimensions attract much interest in mathematical physics because they admit solutions of topological soliton types and find broad applications in physics . thus , the lorentz gauge ( [ elorga ] ) means the following . in the absence of disclinations , equations of equilibrium are identically satisfied , and there exists a field @xmath70 that satisfies equations for the principal chiral field . by this we mean that the spin structure of the medium is described by the model of the principal chiral field in the defect - free case . the principal chiral field model is not the only one that can be used for fixing local rotations . the skyrme model @xcite can also be used for this purpose . the euler lagrange equations for this model are not difficult to rewrite in terms of the @xmath0 connection and use as the gauge conditions . there are other models for spin structures . for describing the distribution of magnetic moments in ferromagnets or the director field in liquid crystals , one uses the expression for the free energy depending on the vector @xmath78-field itself @xcite . lately , much attention is paid to the faddeev model of the @xmath78-field @xcite . the question whether there are gauge conditions on the @xmath0 connection that yield these models in the absence of disclinations is unanswered at present . thus , we pose the following problem in the geometric theory of defects : to find the solution of the euler lagrange equations for free energy ( [ elagfi ] ) that satisfies the elastic gauge for the vielbein and the lorentz gauge for the @xmath0 connection . in sections [ swedie ] and [ swegeo ] , we solve this problem for the wedge dislocation in the framework of the classical elasticity theory and the geometric theory of defects , respectively , and afterwards compare the obtained results . in the preceeding section , we used the elasticity theory and the principal chiral @xmath0-field model to fix the invariance of free energy ( [ elagfi ] ) in the geometric theory of defects . this is not a unique possibility , because other models may be used for gauge fixing . in the present section , we show how another model asymmetric elasticity theory can be used for fixing diffeomorphisms and local rotations . at the beginning of the last century , the cosserat brothers developed the theory of elastic media , every point of which is characterized not only be its position but also by its orientation in space @xcite , i.e. , a vielbein is specified at every point ( fig.[fcosme ] ) . .45 from physical standpoint , this means that every atom in the crystalline structure is not a point but an extended object having orientation . in this case , the stress tensor is no longer a symmetric tensor , and the corresponding theory is called the asymmetric elasticity theory . contemporary exposition of this approach is given in @xcite . in the present section , we show that the asymmetric theory of elasticity is naturally incorporated into the geometric theory of defects . the main variables in the asymmetric theory of elasticity are the displacement vector @xmath42 and the rotation angle @xmath168 . the direction of the pseudovector @xmath74 coincides with the rotation axis of the medium element and its length is equal to the angle of rotation . the angle of rotation discussed in section [ sdiscl ] is dual to the spin structure field @xmath92 : @xmath169 . the cosserat medium is characterized by the stress tensor @xmath170 ( the density of forces acting on the surface with normal @xmath171 in the direction @xmath172 ) and the torque stress tensor @xmath173 ( the density of torques acting on the surface with normal @xmath171 in the direction @xmath172 ) . the cosserat medium is in equilibrium if forces and torques are balanced at every point , @xmath174 where @xmath20 and @xmath175 are the densities of nonelastic external forces and torques . as a consequence of eqn ( [ emofor ] ) , the stress tensor is symmetric if and only if the condition @xmath176 is satisfied . the displacement field and rotation angle uniquely define the deformation tensor @xmath177 and the twist tensor @xmath178 : @xmath179 in general , the deformation and twist tensors have no symmetry in their indices . hook s law in the cosserat medium is changed to two linear relations connecting the stress and torque tensors with the deformation and twist tensors , @xmath180}+{\lambda}{\delta}_{ij}\e_k{}^k , \\ \label{esizkr } \mu_{ij}&=2\g\kappa_{\lbrace ij\rbrace}+2\e\kappa_{[ij ] } + { \beta}{\delta}_{ij}\kappa_k{}^k,\end{aligned}\ ] ] where @xmath181 are the lam coefficients , and @xmath182 are four new elastic constants characterizing the medium . braces and square brackets denote symmetrization and antisymmetrization of indices , respectively . in the case where @xmath183 the deformation tensor is symmetric and has the previous form ( [ edefte ] ) . equation ( [ emofor ] ) , together with ( [ estrte ] ) and ( [ esizkr ] ) , then reduces to the equation @xmath184 the first term vanishes as a consequence of eqn ( [ eqsepu ] ) . thus , for spin structure ( [ eresye ] ) and @xmath185 we return to the symmetric elasticity theory . equations ( [ enewfo ] ) , ( [ emofor ] ) , ( [ estrte ] ) , and ( [ esizkr ] ) , together with the boundary conditions , define the equilibrium state of cosserat media . we now show how this model is included in the geometric theory . first , we note that in the absence of defects ( @xmath58 , @xmath116 ) , the fields @xmath11 and @xmath70 exist . then the vielbein and the @xmath0 connection are defined by the deformation and twist tensors in the linear approximation : @xmath186 we note that relations ( [ edeizk ] ) can be regarded as equations for the displacement vector and rotation angle . the corresponding integrability conditions were obtained in @xcite . these integrability conditions are the linear approximations of equalities @xmath58 and @xmath116 defining the absence of defects . if nonelastic forces and torques are absent ( @xmath146 , @xmath185 ) , then the asymmetric theory of elasticity reduces to second - order equations for the displacement vector and rotation angle : @xmath187 we rewrite these equations for the vielbein and @xmath0 connection @xmath188 of course , these are not the only equations that coincide with eqns ( [ eqtrfi ] ) and ( [ eqstco ] ) in the linear approximation . at present , we do not have arguments for the unique choice . the derived nonlinear equations of the asymmetric elasticity theory can be used as gauge conditions in the geometric theory of defects . here , we have six equations for fixing diffeomorphisms ( three parameters ) and local @xmath0 rotations ( three parameters ) . thus , the asymmetric theory of elasticity is naturally embedded in the geometric theory of defects . in section [ sfixga ] , we considered the elastic gauge for the vielbein and the lorentz gauge for the @xmath0 connection . in this case , the spin structure variables do not interact with elastic deformations when defects are absent . in the asymmetric elasticity theory , the elastic stresses directly influence the spin structure and vice versa . .45 by wedge dislocation , we understand an elastic medium that is topologically the euclidean space @xmath4 without the @xmath189 axis the core of dislocation , obtained in the following way . we consider the infinite elastic medium without defects and cut out the infinite wedge of angle @xmath190 . for definiteness , we assume that the edge of the wedge coincides with the @xmath191 axis ( fig.[fwedis ] ) . the edges of the cut are then moved symmetrically one to the other and glued together . after that , the medium moves to the equilibrium state under the action of elastic forces . if the wedge is cut out from the medium , then the angle is considered negative : @xmath192 . for positive @xmath193 , the wedge is added . thus , the elastic medium initially occupies a domain greater or lesser than the euclidean space @xmath4 , depending on the sign of the deficit angle @xmath193 ; in the cylindrical coordinates @xmath194 this domain is described by the inequalities @xmath195 we note that the wedge dislocation is often called disclination . in our approach , this term seems unnatural because the wedge dislocation is related to a nontrivial torsion . moreover , the term disclination is used for defects in the spin structure . we now proceed with the mathematical formulation of the problem for the wedge dislocation in the framework of elasticity theory . to avoid divergent expressions arising for infinite media , we suppose that the wedge dislocation is represented by the cylinder of a finite radius @xmath196 . this problem has translational symmetry along the @xmath191 axis and rotational symmetry in the @xmath197 plane . therefore , we use cylindrical coordinate system . let @xmath198 be the components of the displacement covector with respect to orthonormal basis in the cylindrical coordinate system . this covector satisfies the equilibrium equation following the substitution of ( [ ehook ] ) in eqn ( [ estati ] ) in domain ( [ ecylco ] ) , @xmath199 where @xmath200 is the covariant derivative for the flat euclidean metric in the considered coordinate system . for references , we write expressions for the divergence and laplacian for a covector field in the cylindrical coordinate system : @xmath201 taking the symmetry of the problem into account , we seek the solution of eqn ( [ qeeste ] ) in the form @xmath202 where @xmath203 and @xmath204 are two unknown functions depending only on the radius . we impose the boundary conditions : @xmath205 the first four equations are geometrical and correspond to the process of dislocation creation . the last condition has simple physical meaning : the absence of external forces at the boundary of the medium . the unknown function @xmath204 is found from the second to the last boundary condition ( [ ebocow ] ) @xmath206 straightforward substitution then shows that @xmath207 and @xmath191 components of equation of equilibrium ( [ qeeste ] ) are identically satisfied , and the radial component reduces to the equation @xmath208 the general solution of this equation is @xmath209 the constant of integration @xmath210 due to the boundary condition at zero . the constant @xmath211 is found from the last boundary condition in ( [ ebocow ] ) . finally , we obtain the known solution of the considered problem @xcite @xmath212 the letter @xmath213 in the expression for @xmath214 denotes the base of the natural logarithm . we note that the radial component of the displacement vector diverges as @xmath215 . this means that the description of the wedge dislocation requires considering a finite - radius cylinder . the linear elasticity theory is applicable for small relative displacements , which for a wedge dislocation are equal to @xmath216 this means that we are able to expect correct results for the displacement field for small deficit angles ( @xmath217 ) and near the boundary of the cylinder ( @xmath218 ) . we find the metric induced by the wedge dislocation in the linear approximation in the deficit angle @xmath193 . calculations can be performed using the general formulas ( [ emetri ] ) or the known expression for the variation of the form of the metric @xmath219 after simple calculations , we obtain the expression for the metric in the @xmath197 plane : @xmath220 this metric is compared with the metric obtained as the solution of three - dimensional einstein equations in section [ swegeo ] . wedge dislocations are relatively rarely met in nature because they require a much amount of a medium to be added or removed , resulting in a vast quantity of energy expenses . nevertheless , their study is of great importance because other linear dislocations can be expressed as a superposition of wedge dislocations . in this respect , wedge dislocations are elementary . we show this for an edge dislocation one of the most widely spread dislocations , as an example . an edge dislocation with the core coinciding with the @xmath191 axis is shown in fig.[feddis]_a_. directed to the dislocation axis ( _ a _ ) . edge dislocation as the dipole of two wedge dislocations with positive and negative deficit angles ( _ b_).[feddis],title="fig : " ] + it appears as the result of cutting the medium over the half - plane @xmath221 , @xmath222 , moving the lower edge of the cut towards the @xmath191 axis on a constant ( far from the core of the dislocation ) burgers vector @xmath33 , and subsequently gluing the edges . to find the displacement vector field for the edge dislocation , we may solve the corresponding boundary value problem for the equilibrium equations ( [ qeeste ] ) @xcite . however , we follow another way , knowing the explicit form of the displacement vector for a wedge dislocation . the edge dislocation is represented by the dipole of two wedge dislocations with positive , @xmath223 , and negative , @xmath190 , deficit angles as shown in fig.[feddis]_b_. we assume that the axes of the first and second wedge dislocations are parallel to the @xmath191 axis and intersect the @xmath197 plane at points with the respective coordinates @xmath224 and @xmath225 . the distance between the wedge dislocation axes is equal to @xmath226 . it follows from the expression for the displacement field ( [ ewedis ] ) that far away from the origin ( @xmath227 ) , the displacement field for the wedge dislocations has the following form in the first order in small @xmath193 and @xmath228 : @xmath229 , \\ u_y^{(1)}&\approx-\theta\left [ \frac{1 - 2\s}{2(1-\s)}(y - a){{\sf\,ln\,}}\frac{r - a{{\sf\,sin\,}}\vf}{{{\sf\,e}}r } + x\left(\vf-\frac{a{{\sf\,cos\,}}\vf}r\right)\right ] , \end{split } \\ \label{esewdi } \begin{split } u_x^{(2)}&\approx\phantom{-}\theta\left [ \frac{1 - 2\s}{2(1-\s)}x{{\sf\,ln\,}}\frac{r+a{{\sf\,sin\,}}\vf}{{{\sf\,e}}r } -(y+a)\left(\vf+\frac{a{{\sf\,cos\,}}\vf}r\right)\right ] , \\ u_y^{(2)}&\approx\phantom{-}\theta\left [ \frac{1 - 2\s}{2(1-\s)}(y+a){{\sf\,ln\,}}\frac{r+a{{\sf\,sin\,}}\vf}{{{\sf\,e}}r } + x\left(\vf+\frac{a{{\sf\,cos\,}}\vf}r\right)\right ] . \end{split}\end{aligned}\ ] ] it is sufficient to sum displacement fields ( [ efiwdi ] ) and ( [ esewdi ] ) to find the displacement field for the edge dislocation because the elasticity theory equations are linear . after simple calculations , up to the translation of the whole medium by a constant vector along the @xmath230 axis , we obtain @xmath231 , \\ u_y&=-b\left[\frac{1 - 2\s}{2(1-\s)}{{\sf\,ln\,}}\frac r{{{\sf\,e}}r } + \frac1{2(1-\s)}\frac{x^2}{x^2+y^2}\right ] , \end{split}\ ] ] where we have introduced the notation for the modulus of the burgers vector @xmath232 this result coincides with the expression for the displacement field obtained by direct solution of the elasticity theory equations @xcite . thus , we have shown that an edge dislocation is the dipole of two parallel wedge dislocations . we next find the metric induced by the edge dislocation . using formulas ( [ einmec ] ) , we obtain the metric in the @xmath197 plane in the linear approximation in @xmath193 and @xmath228 : @xmath233 we note that the induced metric for an edge dislocation does not depend on @xmath196 . in the absence of disclinations ( @xmath145 ) , the @xmath0 connection is a pure gauge , and equations of equilibrium for the @xmath0 connection ( [ econne ] ) are identically satisfied . in this case , the explicit form of the @xmath0 connection is uniquely determined by the spin structure @xmath70 . the field @xmath70 satisfies equations for the principal chiral field as a consequence of the lorentz gauge in ( [ elorga ] ) . solution of this system of equations defines the trivial @xmath0 connection . thus in the absence of disclinations , the problem is reduced to solution of the einstein equations for the vielbein in the elastic gauge and solution of the principal chiral field model for the spin structure . after that , we can compute the torsion tensor through the formulas ( [ ecurcv ] ) , which defines the surface density of the burgers vector . because disclinations are absent ( the curvature tensor is equal to zero ) , we have the space of absolute parallelism . the whole geometry is then defined by the vielbein @xmath55 , which uniquely defines the torsion tensor via ( [ ecurcv ] ) for a vanishing @xmath0 connection . here , we assume that a trivial @xmath0 connection is equal to zero . the vielbein @xmath55 satisfies the three - dimensional einstein equations with a euclidean signature metric , which follow from the expression for the free energy ( [ elagfi ] ) for @xmath163 , @xmath234 here , we have added the source of dislocations @xmath235 to the right hand side of einstein equations ( it is the energy - momentum tensor in gravity ) . we note that without a source of dislocations , the model would be trivial . indeed , the scalar curvature and ricci tensor are equal to zero : @xmath236 , @xmath237 for @xmath238 as a consequence of einstein equations ( [ eindis ] ) . then , the full curvature tensor without sources is also equal to zero because in three - dimensional space , it is in a one - to - one correspondence with ricci tensor ( [ ecurir ] ) . the vanishing of the full curvature tensor means the triviality of the model because defects are absent in this case . the similar statement in three - dimensional gravity is well known . it is usually formulated as : `` three - dimensional gravity does not describe dynamical , i.e. , propagating degrees of freedom '' . for our purposes , we have to find the solution of einstein equations ( [ eindis ] ) describing one wedge dislocation . the einstein equations are a system of nonlinear second - order partial differential equations . not too many exact solutions are known at present , even in a three - dimensional space . the remarkable exact solution describing an arbitrary static distribution of point particles is well known in three - dimensional gravity for the lorentz signature metric @xmath239 [ 7577 ] . we find this solution for the euclidean signature metric and show that it describes an arbitrary distribution of parallel wedge dislocations in the geometric theory of defects . hence , we first consider the more general case of an arbitrary number of wedge dislocations and then analyze in detail one wedge dislocation which is of interest to us . we do this deliberately because the solution in a more general case does not involve essential complications . at the same time , an arbitrary distribution of wedge dislocations is much more interesting for applications . for example , the edge dislocation was shown in the preceeding section to be represented by a dipole of two wedge dislocations of different signs . we consider elastic medium with arbitrary distributed but parallel wedge dislocations . we choose the coordinate system such that the @xmath189 axis is parallel to dislocations axes , and axes @xmath240 , @xmath241 are perpendicular to the @xmath191 axis . then the metric has the block - diagonal form @xmath242 where @xmath243 is a two - dimensional metric on the @xmath197 plane . a two - dimensional metric @xmath244 and a function @xmath245 are independent of @xmath191 due to translational symmetry along the @xmath191 axis . we can say this more simpler dropping the physical arguments . we consider the block - diagonal metric of form ( [ eqmetr ] ) , which has translational invariance along the @xmath191 axis . afterwards , we show that the corresponding solution of the einstein equations indeed describes an arbitrary distribution of parallel wedge dislocations . the curvature tensor for metric ( [ eqmetr ] ) has the components : @xmath246 where @xmath247 is the curvature tensor for the two - dimensional metric @xmath248 and @xmath249 is the two - dimensional covariant derivative with the christoffel symbols . we choose the source of dislocations as @xmath250 where @xmath251 is the two - dimensional @xmath48-function on the @xmath197 plane with the support at a point @xmath252 . the factor @xmath253 in front of the sum sign is due to the property of the @xmath48-function , which is not a function but a tensor density with respect to general coordinate transformations . we show later that the solution of the einstein equations with such a source describes @xmath254 parallel wedge dislocations with deficit angles @xmath255 , which intersect the @xmath197 plane at the points @xmath256 . in three - dimensional gravity this source corresponds to particles of masses @xmath257 , being at rest at the points @xmath258 . einstein equations ( [ eindis ] ) then reduce to four equations , @xmath259 where @xmath260 is the two - dimensional scalar curvature . the metric of form ( [ eqmetr ] ) is still invariant under coordinate transformations in the @xmath197 plane . using this residual symmetry , we fix the conformal gauge on the plane ( this is always possible locally ) @xmath261 where @xmath262 is some function . in the conformal gauge , eqn ( [ eeinon ] ) becomes @xmath263 for constant boundary conditions for @xmath264 on the boundary of the @xmath197 plane this equation has the unique solution @xmath265 . changing the scale of @xmath191 coordinate , we can set @xmath266 without loss of generality . then eqn ( [ eeintw ] ) reduces to the poisson equation @xmath267 which has the general solution @xmath268 thus the metric in the @xmath197 plane is @xmath269 where the polar coordinates @xmath270 cover the whole plane @xmath85 and not more ( this is important ! ) . any solution of the einstein equations is defined up to choosing the coordinate system because the equations are covariant . using this , we set @xmath271 , which is always possible by choosing the scale of @xmath272 . this is indeed the exact solution of the einstein equations describing arbitrary distribution of parallel wedge dislocations . this statement is made clear from the following consideration . we note that transition to a continuous distribution of dislocations in the geometric approach is simple . for this , we have to substitute a continuous distribution of sources in the right hand side of the einstein equations instead of @xmath48-sources . we consider one wedge dislocation with the source at the origin in more detail to show that metric ( [ edisco ] ) indeed describes an arbitrary distribution of wedge dislocations @xmath273 the corresponding metric ( [ edisco ] ) for @xmath271 is @xmath274 we pass to a new coordinate system @xmath275 in which the metric becomes euclidean @xmath276 but the range of the polar angle differs now from @xmath87 : @xmath277 , and covers the @xmath197 plane with removed or added angle @xmath223 . because the metric coincides with the euclidean one in the primed coordinate system @xmath278 , we have the euclidean plane with a removed or added wedge because the angle @xmath279 varies within the interval @xmath280 . the transformation to coordinates @xmath270 in ( [ ekoorv ] ) means the gluing of the edges of the wedge that has appeared , which produces a cone . therefore , both metrics ( [ eonwed ] ) and ( [ equlco ] ) describe the same geometric object conical singularity . the torsion and curvature tensors are obviously zero everywhere except at the origin . creation of a conical singularity coincides exactly with creation of the wedge dislocation in the geometric theory of defects . it is not difficult to show that general solution ( [ edisco ] ) describes an arbitrary distribution of conical singularities with deficit angles @xmath255 at points @xmath258 . thus , this solution describes an arbitrary distribution of parallel wedge dislocations . in the next section , we consider one wedge dislocation in detail . for this , we perform one more coordinate transformation @xmath281 metric ( [ equlco ] ) then becomes @xmath282 this is one more frequently used form of the metric for a conical singularity . we now consider a wedge dislocation from the geometric standpoint . from the qualitative standpoint , the creation of a wedge dislocation coincides with the definition of conical singularity . however , there is a quantitative difference because metric ( [ ecosth ] ) depends only on the deficit angle @xmath193 and can not coincide with the induced metric ( [ emewed ] ) obtained within the elasticity theory . the difference arises because we require the displacement vector in the equilibrium to satisfy equilibrium equations after removing the wedge and gluing its edges ( creating conical singularity ) in the elasticity theory . at the same time the @xmath197 plane for a conical singularity after the gluing can be deformed in an arbitrary way . formally , this manifests itself in that metric ( [ emewed ] ) obtained within the elasticity depends explicitly on the poisson ratio , which is absent in gravity theory . we proposed the elastic gauge for solving this problem @xcite . we choose elastic gauge ( [ efouad ] ) as the simplest one for a wedge dislocation . this problem can be solved in two ways . first , the gauge condition can be inserted in the einstein equations directly . second , we can find the solution in any suitable coordinate system and then find the coordinate transformation such that the gauge condition is satisfied . it is easier to follow the second way because the solution for the metric is known , eqn ( [ ecosth ] ) . the vielbein can be associated with metric ( [ ecosth ] ) as @xmath283 here , a hat over an index means that it corresponds to the orthonormal coordinate system , and an index without a hat is the coordinate one . components of this vielbein are the square roots of the corresponding metric components and therefore have the symmetric linear approximation ( [ etrsep ] ) . we transform the radial coordinate @xmath284 because a wedge dislocation is symmetric under rotations in the @xmath197 plane . after that transformation the vielbein components take the form @xmath285 where the prime denotes differentiation with respect to @xmath272 . we choose the vielbein corresponding to the euclidean metric as @xmath286 it defines the christoffel symbols @xmath287 and @xmath0 connection @xmath288 , which define the covariant derivative . we write only nontrivial components @xmath289 substitution of the vielbein into gauge condition ( [ efouad ] ) yields the euler differential equation for the transition function @xmath290 its general solution depends on two arbitrary constants @xmath291 , @xmath292 where the exponents @xmath293 are defined by the quadratic equation @xmath294 which has real roots for @xmath295 with different signs : positive root @xmath296 and negative @xmath297 . we recall that there are thermodynamical constraints @xmath149 on the poisson ratio @xcite . to fix the constants , we impose boundary conditions on the vielbein : @xmath298 the first boundary condition corresponds to the last boundary condition for displacement vector ( [ ebocow ] ) ( the absence of external forces on the surface of the cylinder ) , and the second one corresponds to the absence of the angular component of the deformation tensor at the core of dislocation . equations ( [ eborep ] ) define the values of integration constants @xmath299 the obtained vielbein defines the metric @xmath300 where @xmath301 this is the solution of the posed problem . the derived solution is valid for all deficit angles @xmath193 and for all @xmath302 . the obtained metric depends on three constants : @xmath193 , @xmath164 , and @xmath196 . the dependence on the deficit angle @xmath193 is due to its occurrence on the right hand side of einstein equations ( [ eindis ] ) . the dependence on the poisson ratio comes from elastic gauge ( [ efouad ] ) , and , finally , the dependence on the cylinder radius comes from boundary condition ( [ eborep ] ) . if a wedge dislocation is absent , then @xmath303 , @xmath304 , @xmath305 , and metric ( [ emeweg ] ) goes to the euclidean one @xmath306 , as expected . we compare metric ( [ emeweg ] ) obtained within the geometric approach with the induced metric from elasticity theory in eqn ( [ emewed ] ) . first , it has a simpler form . second , in the linear approximation in @xmath193 @xmath307 and metric ( [ emeweg ] ) can be easily shown to coincide precisely with metric ( [ emewed ] ) obtained within the elasticity theory . we see that induced metric ( [ emewed ] ) provides only the linear approximation for the metric obtained within the geometric theory of defects , which , in addition , has a simpler form . beyond the perturbation theory , we see essential differences . in particular , metric ( [ emewed ] ) is singular at the origin , whereas metric ( [ emeweg ] ) obtained beyound the perturbation theory is regular . the stress and deformation tensors are related by hook s law ( [ ehook ] ) . there is an experimental possibility to check formulas ( [ emeweg ] ) because the deformation tensor is the linear approximation for the induced metric . for this one has to measure the stress field for a single wedge dislocation . in this way , the geometric theory of defects can be experimentally confirmed or discarded . the problem of reconstruction of the displacement field for a given metric reduces to solution of differential equations ( [ emetri ] ) with metric ( [ emeweg ] ) on the right hand side and boundary conditions ( [ ebocow ] ) . we do not discuss this problem here . we note that in the geometric theory of defects , a complicated stage of finding the displacement vector where it exists is simply absent and is not necessary . two - dimensional metric ( [ emeweg ] ) describes the conical singularity in the elastic gauge . the relation between conical singularities and wedge dislocations was established in [ 22,7880 ] . in these papers , the metric was used in the other gauges ( coordinate systems ) . elastic oscillations in elastic media without defects are described by a time - dependent vector field @xmath308 that satisfies the wave equation ( see , e.g. , @xcite ) @xmath309 where dots denote differentiation with respect to time and @xmath60 is the mass density of medium , which is assumed to be constant . if the medium contains defects , then the metric of the space becomes nontrivial , @xmath310 . we assume that relative displacements for elastic oscillations are much smaller than stresses induced by defects : @xmath311 then , in the first approximation , we assume that elastic oscillations propagate in a riemannian space with a nontrivial metric induced by dislocations . here , we discard changes in the metric due to elastic oscillation themselves . therefore , for elastic oscillations we postulate the following equation , which is a covariant generalization of ( [ efleld ] ) for spatial variables : @xmath312 where @xmath11 denote components of displacement vector field with respect to an orthonormal space basis @xmath313 ( see the appendix ) , @xmath314 is the covariant laplace beltrami operator built for the vielbein @xmath55 , and @xmath315 is the covariant derivative . the explicit form of the covariant derivative of the displacement field is @xmath316 where @xmath317 is the @xmath0 connection built for zero torsion . we note that the displacement vector field describing elastic oscillations is not the total displacement vector field of points of a medium with dislocations . it was already said in section [ sdislo ] that the displacement field for dislocations can be introduced only in those regions of media where defects are absent . if we denote it by @xmath318 , then the total displacement field in these regions is defined by the sum @xmath319 . there , the vielbein is defined only by the displacement field for dislocations @xmath320 . we note that the smallness of relative deformations ( [ edueco ] ) is also meaningful in those regions of space where displacements @xmath318 are not defined . we now decompose the displacement field covariantly into transversal @xmath321 and longitudinal parts @xmath322 , @xmath323 which are defined by the relations @xmath324 the decomposition of a vector field into longitudinal and transversal parts in three - dimensional space with an accuracy up a constant is unique . we recall that latin indices are lowered with the help of kronecker symbols , @xmath325 , and this operation commutes with covariant differentiation . equation ( [ edelop ] ) can be rewritten as @xmath326 because transformation of latin indices into greek ones commutes with the covariant differentiation , and the christoffel symbols are symmetrical in the first two indices . the last equality means that the 1-form @xmath327 is closed . it is easily confirmed that eqn ( [ ephops ] ) for elastic oscillations is equivalent to two independent equations for transverse and longitudinal oscillations , @xmath328 where @xmath329 are the squares of sound velocities for transverse and longitudinal oscillations . particles arising after the secondary quantization of eqns ( [ eelaos ] ) are called phonons in solids . therefore , strictly speaking , the problem of scattering of phonons on dislocations is a quantum mechanical one . in the present review , we consider only classical aspects of this problem . wave equations ( [ eelaos ] ) contain second and first derivatives of the displacement field . the latter are contained in the covariant laplace beltrami operator @xmath330 . the terms with second derivatives can be written in the four - dimensional form @xmath331 where @xmath332 is the inverse metric to @xmath333 where @xmath334 is either the transverse or longitudinal sound velocity . above , we used the following notations . four - dimensional coordinates are denoted by greek letters from the beginning of the alphabet @xmath335 , and letters from the middle of greek alphabet denote only space coordinates @xmath336 . this rule can be easily remembered by the following inclusions @xmath337 and @xmath338 . christoffel symbols ( [ etigam ] ) can be computed for four - dimensional metric ( [ etigas ] ) , which defines a system of ordinary nonlinear equations for extremals @xmath339 ( lines of extremal lengths that coincide with geodesics in riemannian geometry ) , where dots denote differentiation with respect to canonical parameter @xmath340 . for the block - diagonal metric in ( [ etigas ] ) , these equations decompose : @xmath341 where @xmath342 are the three - dimensional christoffel symbols constructed for the three - dimensional metric @xmath156 which , as we recall , depends only on spatial coordinates for a static distribution of defects . let @xmath343 be an arbitrary extremal in four - dimensional space - time . for metric ( [ etigas ] ) , its natural projection on space @xmath344 is also an extremal but now for the spatial part of the metric @xmath156 . equations for extremals ( [ efextr ] ) , ( [ esextr ] ) are invariant under linear transformations of the canonical parameter @xmath340 . therefore , the canonical parameter can be identified with time , @xmath345 , without loss of generality as a consequence of eqn ( [ efextr ] ) . we assume that a particle moves in space along an extremal @xmath346 with velocity @xmath347 . this means that @xmath348 the length of the tangent vector to the corresponding extremal @xmath349 in four - dimensional space - time is then equal to @xmath350 hence , if the particle moves in space along extremal with velocity less than , equal to , or greater than the speed of sound , then its world line in space - time is timelike , null , or spacelike , respectively . we return now to propagation of phonons in media with defects . as in geometric optics @xcite , there are useful notions of wave fronts and rays in the analysis of the asymptotic form of solutions for wave equations ( [ eelaos ] ) . we do not consider mathematical aspects of this approach , which is nontrivial and complicated @xcite , and instead give only a physical description . in the eikonal ( high frequency ) approximation , phonons propagate along rays coinciding with null extremals for the four - dimensional metric @xmath248 . forms of rays , which are identified with trajectories of phonons , are defined by the three - dimensional metric @xmath156 . this means that in the eikonal approximation , trajectories of transverse and longitudinal phonons in a medium with defects are the same and are defined by eqn ( [ esextr ] ) . the difference reduces to the velocities of propagation for transverse and longitudinal phonons being different and equal to @xmath351 and @xmath352 , respectively . calculations in the present section coincide , in fact , with the analysis performed in section 3 of @xcite . the difference is that in what follows , we use the metric written in the elastic gauge . this is important because we assume that trajectories of phonons seen in an experiment coincide with extremals for the metric precisely in this gauge . in the presence of one wedge dislocation , the space metric in the cylindrical coordinates @xmath194 is @xmath353 0 & { \displaystyle\dfrac{\raise-.7ex\hbox{${\alpha}^2$}}{\g^2 } \dfrac{\raise-.7ex\hbox{$r^{2\g}$}}{r^{2\g-2 } } } & 0 \\[2 mm ] 0 & 0 & 1 \end{pmatrix}\ ] ] where the nontrivial part of the metric in the @xmath270 plane was obtain earlier [ see eqn ( [ emeweg ] ) ] . here , we change @xmath296 to @xmath354 for simplicity of notation . the inverse metric is also diagonal , @xmath355 0 & { \displaystyle\dfrac{\raise-.7ex\hbox{$\g^2$}}{{\alpha}^2 } \dfrac{\raise-.7ex\hbox{$r^{2\g-2}$}}{r^{2\g } } } & 0 \\[2 mm ] 0 & 0 & 1 \end{pmatrix}\ ] ] christoffel symbols for metric ( [ ewemep ] ) are calculated according to formulas ( [ etigam ] ) . as a result , only four christoffel symbols differ from zero : @xmath356 in the preceeding section , we showed that in the eikonal approximation , phonons propagate along extremals @xmath346 defined by eqns ( [ esextr ] ) . in the case considered here , these equations are @xmath357 where the dot denotes differentiation with respect to time @xmath358 . it follows from the last equation that phonons move along the @xmath191 axis with constant velocity , which corresponds to translational invariance along @xmath191 . this means that scattering on a wedge dislocation is reduced to a two - dimensional problem in the @xmath270 plane , as could be expected . the system of equations for @xmath359 and @xmath360 in ( [ extrfw ] ) and ( [ extvfw ] ) can be explicitly integrated . for this , we find two first integrals . first , for any metric , there is the integral for equations for extremals @xmath361 we then have the equality @xmath362 second , the invariance of the metric under rotations about the @xmath191 axis results in the existence of an additional integral . it is constructed as follows . there is a killing vector corresponding to the invariance of the metric , which in cylindrical coordinates has the simple form @xmath363 . straightforward verification proves that @xmath364 in the considered case , this results in the identity @xmath365 we analyze the form of an extremal @xmath366 . first derivatives can be found from eqns ( [ efinte ] ) and ( [ einsew ] ) @xmath367 admissible values of the radial coordinate @xmath272 for which the expression under the square root is nonnegative are to be found later . from the above equations , we obtain the equation defining the form of nonradial ( @xmath368 ) extremals @xmath369 this equation can be easily integrated , and we finally obtain explicit formulas defining the form of an extremal : @xmath370=1,\ ] ] where @xmath371 the constant @xmath372 is positive and defines the minimal distance at which an extremal approaches the core of the dislocation , i.e. , @xmath373 . only for these values of @xmath272 is the expression under the root in eqn ( [ eqfexw ] ) nonnegative . the integration constant @xmath374 is arbitrary and corresponds to the invariance of the problem under rotations around the core of dislocation . equations for extremals ( [ extrfw ] ) and ( [ extvfw ] ) also have degenerate solutions : @xmath375 these extremals correspond to the radial motion of phonons . such trajectories are unstable in the sense that there are no nonradial extremals near them . we note that circular extremals @xmath376 are absent as a consequence of eqn ( [ extrfw ] ) , although integrals of motion ( [ efinte ] ) and ( [ einsew ] ) admit such a solution . this is because eqn ( [ extrfw ] ) was multiplied by @xmath377 in deriving first integral ( [ efinte ] ) . we now analyze the form of nonradial extremals ( [ efoexe ] ) . for any extremal , the radius @xmath272 decreases first from infinity to the minimal value @xmath372 and then increases from @xmath372 to infinity . we can assume here without loss of generality that the argument of sine in ( [ efoexe ] ) varies from 0 to @xmath89 . thus , we obtain the range of changes of the polar angle @xmath378 this means that the extremal comes from infinity at the angle @xmath379 and goes to infinity at the angle @xmath380 . this corresponds to the scattering angle @xmath381 we note that the scattering angle depends only on the deficit angle @xmath193 and does not depend on the elastic properties of the medium . the scattering angle has a simple physical interpretation . for positive @xmath193 , the medium is cut and moved apart . a wedge of the same medium without elastic stresses is inserted in the cavity that has appeared . afterwards , gluing is performed , and the wedge is compressed . the compression coefficient for all circles centered at the origin is equal to @xmath382 due to symmetry considerations . therefore , the scattering angle equals half the deficit angle times the compression coefficient , @xmath383 for @xmath303 , the dislocation is absent , and the scattering angle is equal to zero . for positive deficit angles , the scattering angle is positive , which corresponds to repulsion of phonons from the wedge dislocation . corresponding extremals are shown in fig.[fexwpo]_a _ , and they have asymptotes as @xmath384 . _ extremals for the wedge dislocation with a positive ( @xmath385 ) deficit angle . two extremals and their reflections with respect to the @xmath8 axis are shown for the same @xmath385 but different @xmath372 . _ ( b ) _ extremals for the wedge dislocation with a negative ( @xmath386 ) deficit angle . two extremals are shown for the same @xmath372 but different deficit angles @xmath193.,title="fig : " ] + we note that for positive deficit angles , no two points on the @xmath270 plane can be connected by an extremal , i.e. , there is a domain to the right of the wedge dislocation that can not be reached at all by a phonon falling from the left . all extremals shown in the figures in the present section are calculated numerically . the values of the deficit angle @xmath193 , the scattering angle @xmath387 , and the minimal distance @xmath372 to the dislocation axis are shown in the figures . for definiteness , we choose @xmath388 corresponding to the fall of phonons from the left . for negative deficit angles the scattering angle is negative and is defined by the same formulas ( [ escang ] ) . this corresponds to the attraction of phonons to the dislocation axis . in fig.[fexwpo]_b _ , we show two extremals with the same parameter @xmath372 but for two dislocations with different deficit angles . for @xmath389 , the scattering angle varies from 0 to @xmath87 ( fig.[fexwpo]_b _ ) . for @xmath390 , the scattering angle is equal to @xmath87 . we note that for negative deficit angles , phonons have no asymptotes as @xmath384 , i.e. , for @xmath390 , phonons fall from infinity ( @xmath391 ) and return to the infinity ( @xmath392 ) . when the deficit angle is sufficiently small ( @xmath393 ) , a phonon makes one or several rotations around the dislocation and then goes to infinity . examples of such trajectories are shown in figs.[fwede2][fwede4 ] . for @xmath394 , an extremal makes one rotation around the dislocation and then goes forward in the original direction . the blown - up part of the trajectory is shown in the square to the right.,title="fig : " ] + for @xmath395 , an extremal makes two turns around the dislocation and then goes back . the blown - up part of the trajectory is shown in the square to the right.,title="fig : " ] + for @xmath396 , an extremal makes two and a half rotations around the dislocation and then goes forward in the original direction . subsequent blown - up parts of the trajectory are shown in the two squares to the right.,title="fig : " ] + we consider the asymptotic behavior of nonradial extremals as @xmath397 . as a consequence of eqn ( [ erdotf ] ) , far from the core of dislocation , we obtain @xmath398 it follows from this equation that the dependence of the radius on time is the same as for radial extremals ( [ eradex ] ) . because @xmath399 , an infinite value of @xmath272 corresponds to an infinite value of time @xmath358 . this means that the @xmath270 plane with the given metric is complete at @xmath384 . the origin ( the core of dislocation ) is a singular point . only radial extremals fall into it at a finite moment of time . integrals of motion ( [ efinte ] ) and ( [ einsew ] ) have simple physical meaning . equations for extremals ( [ esextr ] ) follow from the variational principle for the lagrangian @xmath400 describing motion of a free massless point particle in a nontrivial metric @xmath401 . here , the metric is considered as a given external field and is not varied . the energy corresponding to this integral is equal to @xmath402 if the metric , as in our case , does not depend on time explicitly , then the energy is conserved ( @xmath403 ) and its numerical value for the motion in the @xmath270 plane is proportional to the integral of motion @xmath404 . for a wedge dislocation , the metric is independent on the polar angle @xmath207 , and the lagrangian is invariant under rotations : @xmath405 . by the noether theorem , the angular momentum conservation law corresponds to this invariance , @xmath406 as a consequence , the constant of integration @xmath407 is proportional to the angular momentum . we note that the behavior of extremals differs qualitatively from trajectories of point particles moving in flat space with euclidean metric @xmath408 in an external potential field @xmath409 . it can be easily shown that trajectories of point particles described by the lagrangian @xmath410 can not coincide with extremals ( [ esextr ] ) for any function @xmath409 . in the present section , we have showed how the problem of scattering of phonons on the simplest wedge dislocation is solved in the geometric approach . the results of calculations can be checked experimentally . the more general problem of scattering of phonons on an arbitrary distribution of wedge dislocations , including the edge dislocation and continuous distribution of dislocations , is solved in @xcite . solution of this problem in the geometric approach is simpler than in the framework of classical elasticity theory , where we have to solve partial differential equations with complicated boundary conditions . in @xcite , calculations were performed in the conformal gauge for the metric . for comparison with experiment , the results must be recalculated in the elastic gauge , which has physical meaning in the geometric approach . we consider elastic media with one wedge dislocation containing one atom of impurity or vacancy . if we consider the influence of impurity on the distribution of elastic stresses small compared with the elastic stresses induced by the dislocation itself , then the motion of the impurity can be considered as taking place in three - dimensional space with the nontrivial metric ( [ emeweg ] ) . in the geometric approach , we assume that the potential energy of the interaction equals zero , and all interactions are due to changes in the kinetic energy , which depends explicitly on the nontrivial metric . we solve the corresponding quantum mechanical problem . we consider bounded states of impurity moving inside a cylinder of radius @xmath196 in the presence of a wedge dislocation . we assume that the cylinder axis coincides with the core of dislocation . the stationary schrdinger equation is @xmath411 where @xmath412 is the plank constant and @xmath254 , @xmath413 , and @xmath414 are mass , wave function , and energy of the impurity . the nontriviality of the interaction with dislocation reduces to the nontrivial laplace beltrami operator @xmath415 where the metric was found earlier [ see eqn ( [ emeweg ] ) ] , and @xmath416 . taking the symmetry of the problem into account , we solve the schrdinger equation ( [ eshrse ] ) in cylindrical coordinates by the separation of variables . let @xmath417 where we have the two following possibilities for the normalization function @xmath418 . if the impurity moves freely along the @xmath191 axis with momentum @xmath419 , then @xmath420 if its motion is restricted by the planes @xmath421 and @xmath422 , then @xmath423 in what follows , we drop the integer - valued subscript @xmath424 indicating the restricted motion , having both possibilities in mind . the condition for the constant @xmath425 ( the eigenvalue of the projection of the momentum on the @xmath191 axis ) to be an integer appears due to the periodicity condition @xmath426 then for the radial wave function , @xmath427 we then obtain the equation @xmath428 we introduce the new radial coordinate @xmath429 this is the transformation in ( [ eftrra ] ) and ( [ ecoifr ] ) up to a constant . the radial equation is then @xmath430 where @xmath431 this is the bessel equation . we solve it with the boundary condition @xmath432 which corresponds to the motion of an impurity inside the cylinder with an impenetrable boundary . the general solution of the bessel equation ( [ erashe ] ) is @xmath433 where @xmath434 and @xmath435 are bessel and neumann functions of order @xmath436 @xcite . the boundedness of the wave function on the axis of the cylinder requires @xmath437 . the constants of integration @xmath438 are found from the normalization condition @xmath439 boundary condition ( [ eboube ] ) yields the equation for @xmath440 @xmath441 which defines the energy levels of bounded states . it is well known that for real @xmath442 and @xmath443 , this equation has only real roots . positive roots form an infinite countable set and all of them are simple @xcite . this provides the inequality @xmath444 we label the positive roots of eqn ( [ eqenlv ] ) by the index @xmath445 ( principle quantum number ) : @xmath446 . then the spectrum of bounded states is @xmath447 for large radii @xmath448 and @xmath449 we have the asymptotic form : @xmath450 as a result , we obtain explicit expression for the spectrum @xmath451 in the absence of defect , @xmath304 , @xmath452 , @xmath453 , and the radial functions @xmath454 are expressed through the bessel functions of integer order @xmath455 . in this case , the spectrum of bounded states depends only on the sizes of the cylinder . in the presence of the wedge dislocation , bessel functions have a noninteger order in general . in this case , the spectrum of energy levels acquires dependence on the deficit angle @xmath193 and the poisson ratio @xmath456 characterizing elastic properties of the medium . if the mass of impurity and vacancy is defined by integral ( [ emass ] ) , then @xmath457 for impurity and the energy eigenvalues are positive . for vacancy , @xmath458 and the energy eigenvalues are negative . in this case , the energy spectrum is not bounded from below , which causes serious problems for physical interpretation . it seems that one has to insert in the schrdinger equation not the bare mass ( [ emass ] ) but the effective mass , with the contribution of elastic stresses arising around a vacancy taken into account . this question presently remains unanswered . the presentation in the present section is close to that in @xcite . in contrast to that paper , we use the elastic gauge for the metric . therefore , our results depend not only on the deficit angle of a wedge dislocation but also on the elastic properties of the medium . calculations of the energy levels of an impurity in the field of a wedge dislocation are actually equivalent to the calculations of bounded - state energies in the aharonov bohm effect @xcite ( see reviews @xcite ) . the difference reduces only to changing the order of the bessel functions , @xmath459 where @xmath460 is the magnetic flux quantum and @xmath461 is the electron charge . the considered example shows how the influence of the presence of defects is taken into account in the geometric approach in the first approximation . if calculations in some problem were performed in elastic media without defects , then to take the influence of defects into account we have to replace the flat euclidean metric with a nontrivial metric describing the given distribution of defects . this problem may appear complicated mathematically because we have to solve the three - dimensional einstein equations to find the metric . however , there are no principal difficulties : the effect of dislocations reduces to a change in the metric . the geometric theory of defects describes defects in elastic media ( dislocations ) and defects in the spin structure ( disclinations ) from a single point of view . this model can be used for describing single defects as well as their continuous distribution . the geometric theory of defects is based on the riemann - cartan geometry . by definition , torsion and curvature tensors are equal to surface densities of burgers and frank vectors , respectively . equations defining the static distribution of defects are covariant and have the same form as equations of gravity models with dynamical torsion . to choose a solution uniquely , one must fix the coordinate system . for this , the elastic gauge for the vielbein and the lorentz gauge for the @xmath0 connection are proposed . in the defect - free case the displacement vector field and the field of the spin structure can be introduced . equations of equilibrium are then identically satisfied , and the gauge conditions reduce to the equations of elasticity theory and the principal chiral @xmath0-field . in this way , the geometric theory of defects incorporates elasticity theory and the model of principal chiral field . in a certain sense , the elastic gauge represents the equations of the nonlinear elasticity theory . nonlinearity is introduced in elasticity theory in two ways . first , the deformation tensor is defined through the induced metric @xmath462 instead of being defined by linear relation ( [ edefte ] ) . then the stress tensor is given by an infinite series in the displacement vector . second , hook s law can be modified assuming nonlinear dependence of the stress tensor on the deformation tensor . hence , the elastic gauge condition is the equations of the nonlinear elasticity theory where the deformation tensor is assumed to be defined through the induced metric and hook s law is kept linear . a generalization to the nonlinear dependence of the deformation tensor on the stress tensor is obvious . as an example , we considered the wedge dislocation from the standpoint of the elasticity theory and the geometric theory of defects . we showed that the elasticity theory reproduces only the linear approximation of the geometric approach . in contrast to the induced metric obtained within the exact solution of the linear elasticity theory , the expression for the metric obtained as the exact solution of the einstein equations in the elastic gauge is simpler and is defined on the whole space and for all deficit angles . the obtained expression for the metric can be checked experimentally . two problems are considered as applications of the geometric theory of defects . the first is the scattering of phonons on a wedge dislocation . in the eikonal approximation , the problem is reduced to the analysis of extremals for the metric describing a given dislocation . equations for extremals are integrated explicitly , and the scattering angle is found . the second of the considered problems is the construction of the wave functions and energy spectrum of impurity in the presence of a wedge dislocation . this requires solving the schrdinger equation . this problem is mathematically equivalent to solving the schrdinger equation for bound states in the aharonov bohm effect @xcite . the explicit dependence of the spectrum on the deficit angle and elastic properties of the medium is found in the presence of a wedge dislocation . the geometric theory of static distribution of defects can also be constructed for membranes , i.e. , on the plane @xmath85 . for this , one has to consider the euclidean version @xcite of two - dimensional gravity with torsion [ 8890 ] . this model is favored by its integrability [ 9194 ] . the developed geometric construction in the theory of defects can be inverted , and we can consider the gravitational interaction of masses in the universe as the interaction of defects in elastic ether . point masses and cosmic strings @xcite then correspond to point defects ( vacancies and impurities ) and wedge dislocations . in this interpretation of gravity , we have a question about the elastic gauge , which has direct physical meaning in the geometric theory of defects . if we take the standpoint of the theory of defects , then the elastic properties of ether correspond to some value of the poisson ratio , which can be measured experimentally . it seems interesting and important for applications to include time in the considered static approach for describing motion of defects in the medium . such a model is lacking at present . from the geometric standpoint this generalization can be easily performed , at least in principle . it is sufficient to change the euclidean space @xmath4 to the minkowski space @xmath463 and to write a suitable lagrangian quadratic in curvature and torsion , which corresponds to the true gravity model with torsion . one of the arising problems is the physical interpretation of the additional components of the vielbein and lorentz connection that contain the time index . the physical meaning of the time component of the vielbein @xmath464 is simple this is the velocity of a point of the medium . this interpretation is natural from the physical standpoint because the motion of continuously distributed dislocations means a flowing of the medium . in fact , the liquid can be imagined as the elastic media with a continuous distribution of moving dislocations . this means that the dynamical theory of defects based on the riemann cartan geometry must include hydrodynamics . it is not clear at present how this can happen . physical interpretation of the other components of the vielbein and the lorentz connection with the time index also remains unclear . the author is sincerely grateful to i v volovich for numerous discussions on the problems considered in this paper . this work is supported by the rfbr ( grant 05 - 01 - 00884 ) and the program of support for leading academic schools ( grant as-1542.2003.1 ) . in the appendix , we briefly present the main facts from differential geometry and introduce the notation used in this review . the description is given in local coordinates , which is sufficient for our purposes . we recommend @xcite as a textbook on differential geometry . if the metric @xmath465 and the affine connection @xmath466 are given , then we say that the geometry is defined on a differentiable manifold @xmath467 , @xmath468 . we assume that all fields on the manifold are given by smooth @xmath469 functions except , possibly , some singular points , and do not stipulate that in what follows . we also assume that the manifold @xmath467 is topologically trivial , i.e. , diffeomorphic to the euclidean space @xmath470 . in a local coordinate system @xmath37 , @xmath471 , the metric is given by a nondegenerate symmetric covariant second rank tensor @xmath401 , which defines the scalar product of vector fields @xmath472 , @xmath473 @xmath474 in general , the scalar product may be not positive definite . if the scalar product is positive definite , we say that a riemannian metric is given on a manifold . by definition , the metric is a covariant tensor field , i.e. , it transforms under coordinate changes @xmath41 by the tensor law @xmath475 this means that the scalar product of vector fields @xmath476 is a scalar field . in a similar way , contractions with the metric @xmath156 and its inverse @xmath157 , @xmath477 , allows one to build scalar fields from higher - rank tensors or to lower their rank . the metric is also used for lowering and raising of tensor indices . an affine connection on a manifold in a local coordinate system is given by the set of coefficients @xmath478 that transform under the diffeomorphisms as @xmath479 these coefficients do not constitute a tensor field because of the presence of an inhomogeneous term in ( [ econts ] ) . an affine connection on a manifold defines covariant derivatives of tensor fields . in particular , covariant derivatives of a vector field and 1-form @xmath480 have the form @xmath481 the covariant derivative of a scalar field coincides with the partial derivative @xmath482 . the covariant derivative of higher - rank tensors is built in a similar way and contains one term with the plus and minus sign for each contravariant and covariant index , respectively . one can easily check that the covariant derivative of a tensor of an arbitrary type @xmath483 is a tensor field of type @xmath484 , i.e. , it has one additional covariant index . we note that the covariant derivative of a product of tensor fields may contain arbitrary contractions of indices . for example , @xmath485 because the inhomogeneous term in ( [ econts ] ) is symmetric in indices @xmath486 and @xmath487 the antisymmetric part of the affine connection @xmath488}{}^\rho$ ] forms a tensor field of type @xmath489 , which is called the torsion tensor @xmath490 in general , the connection @xmath491 has no symmetry in its indices and is not related to the metric @xmath156 in any way because these notions define different geometric operations on a manifold @xmath467 . we then say that the affine geometry is given on @xmath467 . we emphasize that the metric and the affine connection are defined arbitrarily and are completely independent geometric notions . therefore , in the construction of physical models , they can be considered independent fields having different physical interpretations . if the affine geometry is given on a manifold , then we can construct the nonmetricity tensor @xmath492 that is by definition equal to the covariant derivative of the metric : @xmath493 by construction , the nonmetricity tensor is symmetric with respect to the permutation of the last two indices : @xmath494 . we note that we need both objects , the metric and the connection to define the nonmetricity . thus , for a given metric and connection we constructed two tensor fields : the torsion and the nonmetricity tensor . we prove that for a given metric , torsion , and nonmetricity tensor we can uniquely reconstruct the corresponding affine connection . equation ( [ enonme ] ) can always be solved for the connection @xmath466 . indeed , the linear combination @xmath495 yields the expression for the affine connection with all lowered indices : @xmath496 the right hand side of this equality is symmetric in the indices @xmath19 and @xmath436 except one term @xmath497 , and this is in accord with the definition of the torsion tensor ( [ etorsi ] ) . thus , to define the affine geometry on a manifold @xmath467 , it is necessary and sufficient to define three tensor fields : metric , torsion , and nonmetricity . we stress once again that all three object can be specified in a completely arbitrary way , and they can be considered different dynamical variables in models of mathematical physics . it is easy to compute the number of independent components of connection , torsion , and nonmetricity tensors : @xmath498=m^3,~~~~~~ [ t_{\mu\nu}{}^\rho]=\frac{m^2(m-1)}2,~~~~~~ [ q_{\mu\nu\rho}]=\frac{m^2(m+1)}2.\ ] ] this shows that the total number of independent components of the torsion and nonmetricity tensors equals the number of components of the affine connection . in the attempt to unite gravity and electromagnetism , h. weyl considered the nonmetricity tensor of a special type @xcite @xmath499 where @xmath500 is the weyl form identified with the electromagnetic potential ( here , torsion was assumed to be identically equal to zero ) . we say that the _ riemann cartan weyl geometry _ is defined on a manifold if there are given a metric , torsion , and nonmetricity of special type ( [ ewevec ] ) . if the nonmetricity tensor is identically zero , @xmath501 , but the metric and torsion are nontrivial , then we say that the _ riemann cartan geometry _ is given on a manifold . as a consequence of ( [ elicon ] ) , the affine connection is defined uniquely by the metric and torsion in this case : @xmath502 this connection is called metrical because the covariant derivative of the metric is identically equal to zero : @xmath503 the metricity condition provides commutativity of covariant differentiation and the raising and lowering of indices . if both nonmetricity and torsion tensors are identically equal to zero , @xmath501 , @xmath505 , and the metric is nontrivial , then we say that _ riemannian geometry _ is given on a manifold . in this case , the metrical connection is symmetric with respect to the first two indices and uniquely defined by the metric @xmath506 this connection is called the _ levy - civita connection _ or _ christoffel symbols_. we use the tilde to denote the geometrical objects constructed only for the metric and zero torsion and nonmetricity , i.e. , in a ( pseudo-)riemannian geometry . the prefix pseudo- is used when the metric is not positive definite . from the expression for christoffel symbols ( [ etigam ] ) , we see that @xmath507 this means that for the christoffel symbols to be zero in some coordinate system it is necessary and sufficient that the metric components be constant in these coordinates . in another coordinate system , they may be nontrivial because christoffel symbols are not components of a tensor . for example , the christoffel symbols for euclidean space are zero in the cartesian coordinates but differ from zero , e.g. , in spherical or cylindrical coordinate systems . in the case where the nonmetricity tensor and torsion are identically zero , and there is a coordinate system in which the metric is equal to the diagonal unit matrix in the neighborhood of every point , and , consequently , the christoffel symbols are zero , the geometry is called locally euclidean . the corresponding coordinate system is called _ we say that a vector field @xmath472 is transported parallel along a curve @xmath346 , @xmath508 , if @xmath509 where @xmath510 is the tangent vector to the curve . multiplying this equation by @xmath511 we obtain @xmath512 where @xmath513 . we say that under the parallel transport from a point @xmath514 to a neighboring point @xmath515 , the vector @xmath516 acquires a differential @xmath517 , which is linear in @xmath518 and the components of the vector field . in a similar way , we define parallel transport of arbitrary tensor fields along a curve @xmath346 . the result of parallel transport between two points depends in general on a curve connecting these points . we note that parallel transport along a curve is defined only by an affine connection and has no relation to a metric . there are two types of distinguished curves @xmath346 in an affine geometry : geodesics and extremals . a geodesic line is a curve such that a vector tangent to it remains tangent under parallel transport along the curve . with the parameter @xmath358 along the curve chosen canonically , a geodesic line is defined by the system of ordinary nonlinear differential equations @xmath519 although a geodesic line is defined only by the symmetric part of the affine connection @xmath520 , the latter depends nontrivially on torsion and nonmetricity . explicit expression ( [ elicon ] ) yields @xmath521 the second type of distinguished curves in an affine geometry are extremals or lines of extremal length connecting two points . these lines are exclusively defined by the metric . with the parameter along the curve chosen canonically , extremals are defined by an equation similar to ( [ egeoeq ] ) , @xmath522 however , we now have not a general affine connection on the right - hand side but christoffel symbols constructed only from the metric . we see that geodesics and extremals are in general different curves on a manifold . in a riemann cartan geometry , geodesics and extremals coincide if and only if the torsion tensor is antisymmetric in all three indices . in riemannian geometry , geodesics and extremals always coincide . the primary role in differential geometry is played by the curvature tensor of the affine connection , which arises in different contexts . in local coordinates , it is defined by the expression @xmath523 where the parenthesis @xmath524 denote the previous terms with the permuted indices @xmath19 and @xmath436 . it can be easily verified that curvature ( [ ecurva ] ) is indeed a tensor field . we note that the curvature tensor has no relation to a metric and is defined entirely by the connection . in an affine geometry , the curvature tensor with all of its indices lowered has no symmetry under permutations of indices except antisymmetry in the first two indices . contraction of the curvature tensor in the two indices yields the ricci tensor @xmath525 , which is also defined entirely by the connection . in riemannian geometry , the ricci tensor for the levi - civita connection is symmetric with respect to the permutation of indices : @xmath526 . in a riemann cartan geometry , this symmetry is generally absent for a nonzero torsion tensor . the affine connection is called locally trivial if in the neighborhood of every point one can choose the coordinate system in which all components of the connection are zero . now we formulate two important theorems . the proof of this theorem is reduced to an analysis of the transformation rule for the affine connection components in ( [ econts ] ) . if in the new coordinate system the symmetric part of the connection components is equal to zero , then the system of differential equations for the transition functions appears . the local integrability condition for this system of equations is provided by the equality of the curvature tensor to zero . [ tlocev]if for a given affine geometry the torsion , nonmetricity , and curvature tensors are equal to zero on the whole manifold , then this manifold is isometric either to the whole ( pseudo-)euclidean space @xmath470 or to the quotient space @xmath528 , where @xmath529 is a discreet transformation group acting freely . the riemann cartan geometry defined by the metric and torsion provides the basis for the geometric theory of defects . in the analysis of such models , it is more convenient to introduce the cartan variables : the vielbein and @xmath530 connection instead of the metric and torsion . we assume here that a metric on a manifold is positive definite . if a metric were not positively definite then the @xmath531 connection would appear , where @xmath532 . for a riemannian metric , the orthonormal vielbein @xmath533 , @xmath534 , is defined by the system of quadratic equations @xmath535 where @xmath16 is the kronecker symbol . this relation uniquely defines metric for a given vielbein . conversely , system of equations ( [ etetra ] ) defines the vielbein for a given metric up to local @xmath530 rotations . we distinguish greek and latin indices because different transformation groups act on them . from definition ( [ etetra ] ) , we see that @xmath536 . components of the inverse vielbein @xmath537 , @xmath538 , can be considered components of @xmath425 orthonormal vector fields @xmath539 on a manifold @xmath467 with respect to the metric @xmath465 @xmath540 components of the vielbein @xmath55 define @xmath425 orthonormal 1-forms @xmath541 on a manifold . it is known that any manifold @xmath467 can be equipped with a riemannian metric ( see , for e.g. , @xcite ) . at the same time , the global existence of a vielbein provides , in particular , an orientation of the manifold @xmath467 . thus , a vielbein may exist globally only on orientable manifolds . there are also other topological restrictions for the global existence of a vielbein that we do not discuss here . the coordinate basis @xmath542 of the tangent space in every point of @xmath467 is called holonomic . components of tensors of an arbitrary type may also be considered with respect to the unholonomic basis @xmath313 of tangent spaces defined by the vielbein . for example , a vector field has the components @xmath543 we always assume that the transformation of greek indices into latin ones and vice versa is performed with the help of the vielbein . we now define the @xmath530 connection @xmath544 . this is the name of the connection of the principal fiber bundle with the structure lie group @xmath530 and the base @xmath467 . if the riemann - cartan geometry is given on @xmath467 , and the vielbein is defined , then we can define the @xmath530 connection by the relation @xmath545 we see that under coordinate changes , the components of @xmath546 transform as a covector field in the greek index . thus , they define a 1-form on @xmath467 . for a given vielbein , this relation provides a one - to - one correspondence between components of the connections @xmath491 and @xmath546 . the @xmath530 connection defines covariant derivatives for components of tensor fields relative to an unholonomic basis . for example , @xmath547 the covariant derivative is naturally defined for a tensor field having both greek and latin indices . taking the covariant derivative of eqn ( [ etetra ] ) leads to the antisymmetry of the components @xmath548 . this means that the 1-form @xmath549 takes values in the lie algebra @xmath550 , and this indeed corresponds to the @xmath530 connection . in the general case of an affine geometry , the cartan variables can also be defined by relations ( [ etetra ] ) and ( [ etecov ] ) . for this , we have to replace the kronecker symbol on the right - hand side of ( [ etetra ] ) with an arbitrary nondegenerate symmetric matrix @xmath551 . in this case , relation ( [ etecov ] ) defines a linear @xmath552 connection . in the riemann cartan geometry with the lorentzian - signature metric , we would have the lorentz @xmath553 connection . we consider local rotations with a matrix @xmath554 to show that the components @xmath546 do define an @xmath530 connection in the riemann cartan geometry . by definition , the components of vector fields and 1-forms transform under local rotations according to the rule @xmath555 in order that covariant derivatives ( [ ecodve ] ) have the tensor transformation rule under local rotations , it is necessary and sufficient that the components of the @xmath530 connection transform according to the rule @xmath556 this is the transformation law for the @xmath530 connection , indeed . the same transformation law follows from definition ( [ etecov ] ) if the vielbein @xmath55 transforms as a vector with respect to the index @xmath171 , and the christoffel symbols remain unchanged . of course , we can introduce the metrical and @xmath530 connections on @xmath467 in an independent way . if we require afterwards that the @xmath530 connection acts on tensor field components relative to an unholonomic basis defined by the vielbein , we obtain a one - to - one correspondence between the components of the connections ( [ etecov ] ) . we now express the components of the metric connection @xmath491 through the vielbein @xmath55 and @xmath530 connection @xmath546 with the help of relation ( [ etecov ] ) and substitute them into the definitions of torsion ( [ etorsi ] ) and curvature ( [ ecurva ] ) . as a result , we obtain expressions for torsion and curvature in terms of the cartan variables @xmath557 where @xmath558 torsion ( [ ecurcv ] ) and curvature ( [ ecucav ] ) are 2-forms on the manifold @xmath467 with the values in the vector space and lie algebra @xmath550 , respectively . an @xmath530 connection @xmath546 is called locally trivial if every point has a neighborhood containing this point such that it has the form @xmath559 obviously , after local rotation with the matrix @xmath560 , all components of the connection become zero . this connection is also called a pure gauge . one can easily verify that the curvature of a locally trivial connection identically vanishes . the inverse statement is also valid . this theorem is valid for any structure lie group . the proof is reduced to an analysis of transformation rule ( [ elocta ] ) . if the left - hand side of this equation is zero , then we have a system of equations for the field @xmath561 for which vanishing of curvature tensor is the local integrability condition . the space with zero curvature tensor @xmath163 is called the space of absolute parallelism because the parallel displacement of a vector does not depend on the path connecting two fixed points of the manifold . we can perform a local @xmath530 rotation for the locally trivial @xmath530 connection such that it becomes zero : @xmath562 . then , the zero - torsion equality becomes @xmath563 this equation is the local integrability condition for the system of equations @xmath564 for @xmath425 functions @xmath49 . a solution of this system of equations yields the transition functions to cartesian coordinates . thus , the equalities of curvature and torsion tensors to zero are the necessary and sufficient conditions for the existence of the fields @xmath561 and @xmath49 , i.e. , the existence of such a local rotation and coordinate system where the connection vanishes and the metric becomes euclidean . we note that equality of torsion tensor to zero alone is not enough for the existence of a cartesian coordinate system . three - dimensional space is considered in the geometric theory of defects . we make two remarks concerning this . in lower dimensions , the algebraic structure of the curvature tensor with all lowered indexes becomes much simpler . in two dimensions the full curvature tensor in the riemann cartan geometry is in the one - to - one correspondence with its scalar curvature : @xmath565 in three dimensional space , the full curvature tensor is in the one - to - one correspondence with its ricci tensor @xmath566 these formulas are also correct for nonzero torsion . concluding this appendix , we write the identity that is valid in the riemann cartan geometry in an arbitrary number of dimensions : @xmath567 the riemannian scalar curvature on the right - hand side of this equality is constructed only from the vielbein for zero torsion . i. a. kunin and b. i. kunin . gauge theories in mechanics . in _ trends in application of pure mathematics to mechanics . lecture notes in physics , v.249 . _ , pages 246249 , berlin heidelberg , 1986 . verlag . f. w. hehl , j. d. mccrea , e. w. mielke , and y. neeman . metric - affine gauge theory of gravity : field equations , noether identities , world spinors , and breaking of dilaton invariance . , 258(1&2):1171 , 1995 . v. e. zakharov , s. v. manakov , s. p. novikov , and l. p. pitaevskii . . nauka , moscow , 1980 . ; english transl . : s. p. novikov , s. v. manakov , l. p. pitaevskii , and v. e. zakharov _ the inverse scattering method _ plenum , new york ( 1984 ) . b. a. dubrovin , s. p. novikov , a. t. fomenko . . nauka , moscow , 1998 , fourth edition . [ in russian ] ; english transl . prev . ed . : b. a. dubrovin , a. t. fomenko , and s. p. novikov _ modern geometry : methods and applications , _ part 1 , _ the geometry of surfaces , transformation groups , and fields , _ springer , new york ( 1992 ) . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the chemical composition of a galaxy is one of its most fundamental characteristics . here we focus on disk galaxies . because the chemical enrichment depends on various physical processes , such as the star formation history and the mass exchange between the galaxy and its environment , progress in our understanding of galaxy formation and evolution processes depends to a large extent on improving our knowledge of the detailed chemical properties of galaxies , such as the radial distribution of element abundances across galactic disks . establishing the macroscopic properties of spiral and irregular galaxies that likely govern the distribution of heavy elements across their disks is very important in understanding the ( chemical ) evolution of galaxies . accurate abundance determinations in a sample of galaxies are mandatory for such investigations . the classical @xmath1 method , often referred to as the direct method , is generally considered to provide the most reliable oxygen and nitrogen abundances in hii regions . when this method can not be used ( due to the lack of measurements of the required weak auroral lines in the spectra of hii regions ) then combinations of the strong nebular line intensities in spectra of hii regions can be used as indicators of their oxygen abundances , as was first suggested by @xcite and @xcite . this approach is usually referred to as the `` strong - line method '' and has been widely adopted . the establishment of calibrations ( i.e. , of relations between metallicity - sensitive emission - line combinations and metallicity ) was the subject of numerous studies ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * among many others ) . a calibration is defined not only by the adopted indicators but also by the objects that serve as calibrating data points . a sample of hii regions with abundances derived through the @xmath2 method is used to construct an empirical calibration . a set of photoionization models is used to construct a theoretical ( model ) calibration . even if the same indicator is used in the empirical calibration and in the theoretical calibration and if the same spectral measurements in a given hii region are used , those calibrations can produce significantly different abundance estimations . metallicities derived using theoretical calibrations tend to be systematically higher ( up to @xmath30.7 dex ) than those derived using the empirical calibrations ( see reviews by * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . therefore , oxygen abundances in extragalactic hii regions obtained in different studies using different calibrations can be significantly different . spectroscopic measurements of hii regions within and beyond optical radii of galaxies were carried out in many works ( see list of references below ) . in these studies , usually the hii regions in one or several galaxies are measured and the radial distributions of the element abundances across the disks of those galaxies are estimated . since often different methods for abundance determinations are used in different works , the resulting abundances from these studies are not homogeneous and can not be directly compared to each other . therefore , the abundances in a sample of galaxies can be analyzed only after those abundances are homogenized , i.e. , all the abundances are redetermined in a uniform way . this is the first step in our present investigation . it should be noted that there have been several attempts to use uniform abundances for the determination of radial abundance gradients in a sample of galaxies ; e.g. , @xcite for a sample of 30 galaxies , @xcite for 39 galaxies , @xcite for 11 galaxies , @xcite for 54 galaxies , and @xcite for 21 galaxies . however , those samples contain a relatively small number of galaxies ( whereas our present sample includes 130 galaxies ) . so far , little attention has been paid to the radial distributions of nitrogen abundances in the disks of galaxies , despite the fact that this provides several advantages for the study of the chemical evolution of galaxies . indeed , since at 12+log(o / h ) @xmath4 8.3 , secondary nitrogen becomes dominant and the nitrogen abundance increases at a faster rate than the oxygen abundance ( e.g. , * ? ? ? * ) , the change in nitrogen abundances with galactocentric distance should then show a larger amplitude in comparison to oxygen abundances and , as a consequence , should be easier to measure . furthermore , there is a time delay in nitrogen production as compared to oxygen production @xcite . thus the comparison between the radial distributions of oxygen and nitrogen abundances in the disks of galaxies can shed additional light on the chemical evolution of galaxies . therefore we consider here not only the radial distributions of oxygen abundances but also those of nitrogen abundances . our paper is organized in the following way . we describe the method used for the oxygen and nitrogen abundance determinations in the hii regions of our galaxy sample in section 2 . we describe the observational data that were used to determine the abundances in the hii regions in section 3 . we discuss the abundance properties in the disks of nearby galaxies ( within the optical isophotal radii ) in section 4 . we summarize our results in section 5 . throughout this paper , we will use the following standard notations for the line intensities : + @xmath5 = @xmath6 \lambda 3727 + \lambda 3729 } /i_{{\rm h}\beta } $ ] , + @xmath7 = @xmath8 \lambda 6548 + \lambda 6584 } /i_{{\rm h}\beta } $ ] , + @xmath9 = @xmath10 \lambda 6717 + \lambda 6731 } /i_{{\rm h}\beta } $ ] , + @xmath11 = @xmath12 } \lambda 4959 + \lambda 5007 } /i_{{\rm h}\beta } $ ] . + with these definitions , the excitation parameter @xmath13 can be expressed as : @xmath13 = @xmath11/(@xmath5+@xmath11 ) . a new method ( called the `` @xmath14 method '' ) for oxygen and nitrogen abundance determinations has recently been suggested @xcite . the idea of the @xmath14 method is a very simple . we have compared a several combinations of the strong - line intensities in the spectrum of a given hii region with those in the spectra of a sample of reference hii regions with known abundances in order to find the counterpart for the hii region under study . it is assumed that the oxygen and nitrogen abundances in the studied hii region are the same as in its counterpart . a counterpart can be selected by comparison of four combinations of strong - line intensities : @xmath13 = @xmath11/(@xmath5 + @xmath11 ) ( excitation parameter ) , log@xmath11 , log(@xmath7/@xmath5 ) , and log(@xmath9/@xmath5 ) . however , there are recent measurements of spectra of many hii regions where the intensities of [ oii]@xmath153727+@xmath153729 or [ sii]@xmath156717 , [ sii]@xmath156731 ) lines are not available ( e.g. , in @xcite or in the sloan digital sky survey ( sdss ) ; see @xcite ) . it has been argued that the oxygen and nitrogen abundances in hii regions can be estimated even if the [ sii]@xmath156717+@xmath156731 emission line is not measured @xcite or if the [ oii]@xmath153727+@xmath153729 emission line is not available @xcite . the @xmath14 method can be adapted to such cases @xcite . to find the counterpart for the hii region under study , one does not need to compare the four combinations of strong - line intensities , but can instead also use only three combinations : 1 ) log@xmath11 , @xmath13 , and log(@xmath7/@xmath5 ) , or 2 ) log@xmath11 , log@xmath7 , and log(@xmath7/@xmath9 ) . when this first set of combinations of strong - line intensities is used to find the counterpart then the resulting oxygen and nitrogen abundances will be referred to as ( o / h)@xmath16 and ( n / h)@xmath16 . when the second set of combinations of strong - line intensities is used to find the counterpart then the inferred oxygen and nitrogen abundances will be called ( o / h)@xmath17 and ( n / h)@xmath17 . the data for reference hii regions with @xmath2-based abundances are compiled in @xcite . the very recent spectroscopic observations of @xcite have been added to the compilation . using these combined data we select a sample of reference hii regions for which all the absolute differences for oxygen abundances ( o / h)@xmath16 ( o / h)@xmath18 and ( o / h)@xmath17 ( o / h)@xmath18 and for nitrogen abundances ( n / h)@xmath16 ( n / h)@xmath18 and ( n / h)@xmath17 ( n / h)@xmath18 are less than 0.1 dex . this sample of reference hii regions contains 250 objects and will in the following be used for abundance determinations . this sample will be referred to as @xmath19 sample ( etalon sample 2013 ) below . furthermore , only blue spectra were observed for hii regions in a number of galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , i.e. , intensities of [ nii]@xmath156584 and [ sii]@xmath156717 , [ sii]@xmath156731 ) are not available in those cases . the oxygen abundances in those hii regions can be estimated through the @xmath13 calibration where only the oxygen [ oii]@xmath153727+@xmath153729 and [ oiii]@xmath155007 lines are used @xcite . we have constructed a new variant of the @xmath13 calibration . the sample of reference hii regions , @xmath19 , has been used as calibration data points . to enlarge the number of calibration data points , we have added a number of hii regions with ( o / h)@xmath16 abundances that were chosen in the following way . we determined the ( o / h)@xmath16 abundances in hii regions and obtained radial oxygen abundance gradients across the disks of galaxies ( see below ) . the hii regions where the deviations of the ( o / h)@xmath16 abundances from the general radial abundance trend are less than 0.1 dex were added to the sample of reference hii regions . it is well known that the relation between the oxygen abundance and the strong oxygen line intensities is double - valued , with two distinct parts traditionally known as the upper and lower branches of the r@xmath20 o / h diagram . we have delimited the upper and lower branches and the transition zone , adopting 12+log(o / h ) = 8.3 as the boundary between the upper branch and the transition zone and 12+log(o / h ) = 8.0 as the boundary between the transition zone and the lower branch . these boundaries are somewhat arbitrary , but were chosen so as to give the best calibrations with the existing data . two distinct relations between the oxygen abundance and the strong oxygen line intensities will be established in the following , one for the upper branch ( the high - metallicity calibration ) and one for the lower branch ( the low - metallicity calibration ) . the relation between the oxygen abundance @xmath21 @xmath22 12 + log ( o / h)@xmath23 and r@xmath24 and p can be fitted by a polynomial of the form @xcite @xmath25 where we have used the notation z @xmath22 12 + log ( o / h ) for brevity . to take into account the dependence on the excitation parameter @xmath13 , the coefficients of eq . ( [ equation : kr3 ] ) are chosen to have the form @xmath26 the coefficients @xmath27 , @xmath28 , @xmath29 , @xmath30 , @xmath31 , and @xmath32 can then be determined by looking for the best fit to our sample of hii regions . we wish to derive a set of coefficients in eq . ( [ equation : kr3 ] ) which gives the minimum value of @xmath33 = @xmath34 . here @xmath35 is equal to log(o / h)@xmath36 log(o / h)@xmath37 for each hii region in our sample . the quantity @xmath33 is the average value of the differences between the oxygen abundances determined through the @xmath13 calibration and the original ones . a few data points with large deviations , in excess of 0.15 dex , are rejected , and are not used in the determination of the final relation . the obtained upper - branch @xmath13 calibration ( for 12 + log(o / h ) @xmath4 8.3 ) is @xmath38 the obtained lower - branch @xmath13 calibration ( for 12 + log(o / h ) @xmath39 8.0 ) is @xmath40 hence , here we use the empirical metallicity scale defined by the hii regions with abundances derived through the direct method ( @xmath2 method ) . we have carried out a fairly comprehensive compilation of published spectra of hii regions in late - type galaxies . only those galaxies where radial abundance gradients can be estimated were taken into consideration . our final list includes 130 galaxies . table [ table : sample ] lists the general characteristics of each galaxy . the first column gives its name . we have used the most widely used name for each galaxy . the galaxies are listed in order of name category , with the following categories in descending order : + ngc new general catalogue , + ic index catalogue , + ugc uppsala general catalog of galaxies , + pgc catalogue of principal galaxies . + the morphological type of the galaxy and morphological type code @xmath41 from leda are reported in columns 2 and 3 . the right ascension ( r.a . ) and declination ( decl . ) ( j2000.0 ) of each galaxy are given in columns 4 and 5 . the right ascension and declination are taken from the nasa / ipac extragalactic database ( ned ) . the isophotal radius @xmath0 in arcmin of each galaxy is reported in column 6 . unless otherwise stated , the @xmath0 values are taken from ( * ? ? ? * thereafter rc3 ) . the position angle and inclination are listed in columns 7 and 8 , and the sources for these values are given in column 9 . the adopted distance and its reference are reported in columns 10 and 11 . the ned distances use flow corrections for virgo , the great attractor , and shapley supercluster infall . the isophotal radius in kpc , estimated from the data in columns 6 and 10 , is listed in column 12 . we have carried out an extensive search of the literature and compiled a sample of measurements of hii regions in nearby late - type galaxies . we have searched for spectra of hii regions with the requirement that they include the [ oii]@xmath153727+@xmath153729 , [ oiii]@xmath155007 , [ nii]@xmath156584 , and [ sii]@xmath156717+@xmath156731 lines . while we have tried to include as many sources as possible , we do not claim our search to be exhaustive . thus , for each listed spectrum , we record the measured values of [ oii]@xmath153727+@xmath153729 , [ oiii]@xmath155007 , [ nii]@xmath156584 , [ sii]@xmath156717 , and [ sii]@xmath156731 . the intensities of all lines are normalised to the h@xmath42 line flux . the predicted values of the flux ratio of oxygen [ oiii]@xmath155007/[oiii]@xmath154959 and nitrogen [ nii]@xmath156584/[nii]@xmath156548 lines are very close to three @xcite . the measurements of the [ oiii]@xmath155007 and @xmath154959 lines in sdss spectra confirm this value of the flux ratio ( e.g. , * ? ? ? therefore , the value of @xmath11 can be estimated without [ oiii]@xmath154959 line as @xmath43[oiii]@xmath155007 , and , similarly , the values of @xmath7 are estimated without the lines [ nii]@xmath156548 as @xmath7 = 1.33[nii]@xmath156584 . we have taken the de - reddened line intensities as reported by the authors . in some papers only the measured fluxes are reported . in these cases , the measured emission - line fluxes were corrected for interstellar reddening using the theoretical h@xmath44 to h@xmath42 ratio ( i.e. , the standard value of h@xmath44/h@xmath42 = 2.86 ) and the analytical approximation to the whitford interstellar reddening law from @xcite . the spectroscopic data so assembled form the basis of the present study . our total list contains 3904 spectra including 162 spectra of hii regions beyond the isophotal radius @xmath0 . here we will examine the radial oxygen and nitrogen abundance distributions within the isophotal radius in every galaxy . the radial oxygen and nitrogen abundances beyond the isphotal radius will be discussed elsewere . when measurements of the lines [ oii]@xmath153727+@xmath153729 , [ oiii]@xmath155007 , and [ nii]@xmath156584 were available then the oxygen ( o / h)@xmath16 and nitrogen ( n / h)@xmath16 abundances in the hii regions were estimated and used in our study . if the intensity of the line [ oii]@xmath153727+@xmath153729 was not measured but the measurements of the lines [ oiii]@xmath155007 , [ nii]@xmath156584 , [ sii]@xmath156717 , and [ sii]@xmath156731 ) were available then the oxygen ( o / h)@xmath17 and nitrogen ( n / h)@xmath17 abundances of the hii regions were estimated and used . if the measurements of only the oxygen lines [ oii]@xmath153727+@xmath153729 and [ oiii]@xmath155007 were available ( blue spectrum was observed only ) then the oxygen ( o / h)@xmath45 abundance of the hii regions was estimated and adopted . the deprojected radii of the hii regions were computed using their coordinates ( or offsets from the nucleus ) as reported in the original papers , as were the position angle and inclination listed in table [ table : sample ] . in some publications , the positions of the observed hii regions ( or their identifications in catalogues ) were not reported , but the deprojected radii were listed instead . in these cases these deprojected radii were used ( after correction for the galaxy distance adopted here , if necessary ) . the fractional radii ( normalized to the optical isophotal radius @xmath0 ) were obtained with isophotal radii from table [ table : sample ] . the radial oxygen abundance distribution within the isophotal radius in every galaxy was fitted by the following equation : @xmath46 where 12 + log(o / h)@xmath47 is the oxygen abundance at @xmath48 = 0 , i.e. , the extrapolated central oxygen abundance , c@xmath49 , is the slope of the oxygen abundance gradient expressed in terms of dex @xmath50 , and @xmath51/@xmath52 is the fractional radius ( the galactocentric distance normalized to the disk s isophotal radius @xmath0 ) . we also determined the oxygen abundance gradient expressed in terms of dex kpc@xmath53 . if there were data points with large deviations , in excess of 0.2 dex , those points were rejected , and were not used in the determination of the final relation . the derived parameters of the oxygen abundance distributions are presented in table [ table : grad ] . the name of the galaxy is listed in column 1 . the extrapolated central 12 + log(o / h)@xmath47 oxygen abundance and the gradient ( the coefficient c@xmath49 in eq . ( [ equation : grado ] ) ) expressed in terms of dex @xmath54 are listed in columns 2 and 3 . the oxygen abundance gradient expressed in terms of dex kpc@xmath53 is listed in column 4 . the scatter of oxygen abundances around the general radial oxygen abundance trend is reported in column 5 . as in the case of the oxygen abundance , the radial nitrogen abundance distribution in every galaxy was fitted by the following equation : @xmath55 again , if there were data points with large deviations , in excess of 0.3 dex , those points were rejected , and were not used in the determination of the final relation . these different rejection criteria for oxygen abundances , 0.2 dex , and nitrogen abundances , 0.3 dex , are used for the following reason . there is no one - to - one correspondence between nitrogen and oxygen abundances , but instead there is a scatter in n / h at a given o / h . therefore one can expect that the natural scatter in n / h at a given galactocentric distance can be larger than that in o / h assuming similar uncertainties in both abundance determinations . the derived parameters of the nitrogen abundance distributions are presented in table [ table : grad ] . the extrapolated central 12 + log(n / h)@xmath47 nitrogen abundance and the gradient ( the coefficient c@xmath56 in eq . ( [ equation : gradn ] ) ) expressed in terms of dex @xmath50 are listed in columns 6 and 7 . the nitrogen abundance gradient expressed in terms of dex kpc@xmath53 is listed in column 8 . the scatter of nitrogen abundances around the general radial nitrogen abundance trend is reported in column 9 . a list of references for the emission line flux measurements in the hii regions is given in table [ table : reference ] . in the case of ( o / h)@xmath23 abundances , there is the following problem . the relationship between oxygen abundance and strong oxygen line intensities , the @xmath13 calibration , is double - valued with two distinct parts usually known as the `` lower '' and `` upper '' branches of the @xmath57 o / h relationship . the expression for the oxygen abundance determination in high metallicity hii regions , eq . ( [ equation : ohph ] ) , is valid only for hii regions that belong to the upper branch , with 12+log(o / h ) @xmath4 8.3 . thus , one has to know a priori on which of the two branches the hii region lies . we can overcome this problem in the way suggested in @xcite . it has been known for a long time @xcite that disks of spiral galaxies show radial oxygen abundance gradients , in the sense that the oxygen abundance is higher in the central part of the disk and decreases with galactocentric distance . we thus start from the hii regions in the central part of disks and move outward until the radius @xmath58 where the oxygen abundance decreases to 12+log(o / h ) @xmath3 8.3 . it should be noted that it is difficult to establish the exact value of @xmath58 due to the scatter in oxygen abundance values at any fixed radius . an unjustified use of eq . ( [ equation : ohph ] ) in the determination of the oxygen abundance in low - metallicity hii regions beyond @xmath58 would result in overestimated oxygen abundances , and would produce a false turnover in the slope of the abundance gradients @xcite . therefore , hii regions with galactocentric distances larger than @xmath58 , i.e. , those with 12+log(o / h ) @xmath39 8.3 were rejected . the derived radial distributions of the oxygen and nitrogen abundances in 130 galaxies are presented in figure [ figure : gs1 ] . the oxygen abundances for individual hii regions are indicated by points . all the data points ( including points with large deviations , which are rejected and are not used in determination of the final relation ) are shown . the linear best fits ( derived via the least squares method ) to the points are represented by solid lines . the galactocentric distances are normalized to the isophotal radius . the nitrogen abundances for individual hii regions are shown by the plus signs . the linear best fits to the points are indicated by dashed lines . the values of the gradients in a number of galaxies ( e.g. , in ngc 12 , the first galaxy in our list ) are rather small and are comparable ( or even lower ) to the uncertainties of gradients . it can be assumed that there is no abundance gradient in such a galaxy , and its abundance can be specified by the mean of the central abundance and the abundance at the isophotal @xmath0 radius of the galaxy . thus , the radial oxygen and nitrogen abundance distributions across the optical disk in every galaxy are individually fitted by a single relation . it looks like a rather good approximation for the majority of galaxies . however , a small change in slope in the abundance distribution can not be excluded in the disks of several galaxies ( e.g. , ngc 925 , ngc 3184 , ngc 5457 ) . it is interesting to note the following . @xcite studied the surface brightness profiles of a sample of late - type ( sb to sdm ) spiral galaxies using imaging data from the sdss survey . they found that the surface brightness profiles can be divided into three classes . a small fraction of galaxies ( around 10 % ) belong to type i , which comprises those galaxies that have a normal ( standard ) exponential disk down to the noise limit . the surface brightness distribution of the rest of the galaxies is better described as a broken exponential . about 60 % of the galaxies belong to type ii , which means that they show a down - bending profile with steeper outer part . about 30 % of the galaxies belong to type iii , implying that they show an up - bending profile with shallower outer part . thus , the abundance distribution profile does not seem to follow strictly the slope of the surface brightness profile . it should be noted , however , that the number of measured hii regions in many galaxies is too small to allow one to detect a small bend in the radial abundance distribution . * ngc 1512 . * the five hii regions that are located in the `` bridge '' between the interacting galaxies ngc 1512 and ngc 1510 @xcite were excluded from our analysis . @xcite found that there is a significant discrepancy between the kinematically derived inclination of ngc 1512 and its photometric value ( @xmath59 = 65@xmath60 and @xmath61 = 35@xmath60 ) . they noted that the explanation is quite clear : the photometric parameter is mainly representative of the bar which contributes a large part of the light resulting in more edge - on values . however , the outer isophotes are much more face - on . therefore , the kinematic value of the inclination is adopted here . * ngc 2442 . * the peculiar spiral galaxy ngc 2442 is tidally distorted * and references therein ) . * ngc 3227 . * ngc 3227 is a nearby seyfert galaxy that is interacting with its gas - poor dwarf elliptical companion ngc 3226 . @xcite discovered a cloud of hi close to , but physically and kinematically distinct from , the galactic disk of ngc 3227 . they suggested that this cloud ( j1023 + 1952 ) might be a dwarf galaxy that is either preexisting and being accreted by ngc 3227 , or a newly created tidal dwarf galaxy @xcite . however , the oxygen abundance of the cloud ( hii regions 24 and 25 from @xcite and of the hii region measured by @xcite ) follows the general trend of the radial abundance distribution in the disk of ngc 3227 . the oxygen abundances of the hii region in the cloud estimated in different ways using the emission line measurements by @xcite are in satisfactory agreement to each other : 12+log(o / h)@xmath62 = 8.40 , 12+log(o / h)@xmath63 = 8.43 and 12+log(o / h)@xmath45 = 8.34 . hii regions 10 and 12 from @xcite follow also the general trend of the radial abundance distribution in the disk of ngc 3227 although those hii regions may be associated with ngc 3226 . * ngc 3239 . * the irregular galaxy ngc 3239 is a candidate for a merging system @xcite . the oxygen abundances determined in four hii regions from sdss spectra as well as the global abundance determined using the integrated spectra from @xcite are 12+log(o / h ) @xmath3 8.0 @xmath64 8.1 , i.e. , those abundances correspond to the transition zone between the upper and lower branches of the o / h @xmath65 diagram . therefore , the ( o / h)@xmath45 abundances we derived from measurements by @xcite are not reliable and were not used . * ngc 3310 . * ngc 3310 is believed to be an advanced merger and its unusual outer structure is probably the result of a recent merger with a smaller galaxy @xcite . spectroscopic observations of six hii regions in ngc 3310 were presented by @xcite . the auroral line [ oiii]@xmath154363 is detected in four hii regions . the @xmath2-based abundances in those hii regions are in the range from 12+log(o / h ) = 8.13 to 8.25 . however , the coordinates of the hii regions are not published , which prevents us from using those hii regions in radial gradient determinations . * ngc 3359 . * the ngc 3359 is a giant , very strongly barred spiral galaxy . @xcite estimated the radial gradient using the [ oiii]/h@xmath42 and [ nii]/[oiii ] indicators ( one - dimensional calibrations ) . they found the radial abundance gradient break near the corotation radius . @xcite carried out new measurements and determined abundances through the @xmath66 and @xmath67 calibrations following @xcite . they concluded that , with a high degree of confidence , a model with a break fits the data significantly better than one without a break . the @xmath14-based abundances show no gradient across the entire galaxy . the @xmath2-based abundances in three hii regions in ngc 3359 are in agreement with no gradient . * ngc 3718 . * this galaxy is so peculiar that it is difficult to categorize it morphologically @xcite . * ngc 4559 . * the hii regions in ngc 4559 with galactocentric radii larger than 0.4@xmath0 measured by @xcite were excluded since they are in the transition zone . * ngc 4625 . * this is a magellanic spiral with @xmath0 @xmath3 3 kpc and an extended faint disk reaching four times the optical @xmath0 radius of the galaxy in the ultraviolet . * ngc 5668 . * @xcite have found that within @xmath68 of the nucleus the oxygen abundance o / h follows an exponential profile while the outer abundance trend flattens out to an approximately constant value and could even reverse . * ngc 7518 . * this is a hydrogen - deficient galaxy in the pegasus cluster . following @xcite , oxygen lines from `` blue '' spectra ( 360 nm 560 nm ) have been used in the abundance determinations . * ngc 7529 . * this galaxy of the pegasus cluster has a normal hi content . there is no determination of the position angle , but the inclination is small . again following @xcite , oxygen lines from `` blue '' spectra have been used in the abundance determinations . * ngc 7591 . * this galaxy of the pegasus cluster has a normal hi content . as before , following @xcite , oxygen lines from `` blue '' spectra have been used in the abundance determinations . * ic 10 . * due to uncertainties in the position angle and in the inclination of ic 10 , the galactocentric distances of hii regions have been computed without any de - projection ( following * ? ? ? * ic 5309 . * a hydrogen - deficient galaxy of the pegasus cluster . following @xcite , oxygen lines from `` blue '' spectra have been used in the abundance determinations . * ugc 9562 = ii zw 71 . * the blue compact dwarf galaxy ugc 9562 is a probable polar - ring galaxy , and there are a several luminous hii regions along its major axis . the oxygen abundances in four hii regions are around 12+log(o / h ) @xmath3 8.2 , i.e. , those abundances correspond to the transition zone between the upper and lower branches of the o / h @xmath65 diagram . therefore , the ( o / h)@xmath45 abundances obtained with measurements from @xcite were not used . the global oxygen abundances determined from the integrated spectra from @xcite are 12+log(o / h ) @xmath3 8.3 . * pgc 029167 . * the dwarf galaxy `` garland '' or pgc 029167 ( a tidal dwarf candidate ; * ? ? ? * ) lies within the tidal bridges of neutral hydrogen connecting m 81 , m 82 and ngc 3077 . this alignment has led to the suggestion that this dwarf formed recently , as a result of tidal interactions within the group . hii regions in and near garland exhibit enhanced metallicities compared to other galaxies at similar luminosities . notably , the oxygen abundances are similar to abundances measured in m 81 and ngc 3077 ( * ? ? ? * and references therein ) . figure [ figure : ohnh ] shows the nitrogen abundance as a function of oxygen abundance in our sample of galaxies . the plus signs are the nitrogen ( n / h)@xmath47 and oxygen abundances ( o / h)@xmath47 at @xmath48 = 0 , i.e. , the central nitrogen and oxygen abundances . the points show the nitrogen ( n / h)@xmath69 and oxygen ( o / h)@xmath69 abundances at the @xmath0 radius of the galaxies . our data show a well - defined sequence in the o / h n / h plot . this sequence exhibits the well - known turnover in the sense that the slope at low metallicity is shallower than that at high metallicity . the commonly accepted explanation of this change of slope is that nitrogen can be interpreted as having both primary and secondary components . nitrogen production is primary at low metallicity , but for 12 + log(o / h ) @xmath4 8.3 , secondary nitrogen becomes prominent , and nitrogen increases at a faster rate than oxygen @xcite . figure [ figure : ohnh ] shows that at high metallicity , 12 + log(o / h ) @xmath4 8.2 , the relation between the logarithms of nitrogen and oxygen abundances can be approximated by a linear expression . for ( n / h)@xmath47 and ( o / h)@xmath47 abundances , we found the following relation through the least squares method : @xmath70 a similar relation @xmath71 was obtained for the ( n / h)@xmath69 and ( o / h)@xmath69 abundances . a comparison between eq . ( [ equation : nhro ] ) and eq . ( [ equation : nhrg ] ) shows that the relations n / h = @xmath72(o / h ) for the ( n / h)@xmath47 and ( o / h)@xmath47 abundances and for the ( n / h)@xmath69 and ( o / h)@xmath69 abundances agree with each other within the errors . the n / h = @xmath72(o / h ) relation given by eq . ( [ equation : nhro ] ) is presented in figure [ figure : ohnh ] by the solid line . the dashed lines indicate shifts along the y - axis by @xmath730.15 dex . at low metallicity , 12 + log(o / h ) @xmath39 8.0 , the single relation @xmath74 was found using both central abundances and abundances at the @xmath0 radii of the galaxies . this relation is presented in figure [ figure : ohnh ] by the dotted line . figure [ figure : ohnh ] shows that there is an appreciable spread in n / h at a given o / h . the variation in n / h is around a factor of 2 for a given o / h . the scatter can be partially attributed to the errors in the abundance determinations but part of it seems to be true abundance scatter . two major mechanisms have been proposed for generating a spread . one mechanism invokes a time delay between ejections of the freshly manufactured oxygen and nitrogen into the interstellar medium by a given stellar generation . the n / o ratios may be an indicator of the age of a galactic system , indicating the time that has passed since the bulk of star formation activity @xcite . thus , the n / o ratio in a galaxy would then depend on its star formation history . the second mechanism for causing n / h variations at a given o / h is a variation in the efficiency of enriched galactic winds @xcite or / and in the inflow of gas into the galaxy @xcite . it is believed that galactic winds do not play a significant role in the chemical evolution of large spiral galaxies @xcite . the enhanced n / o ratio in individual hii regions can be caused by the local pollution in nitrogen by wolf - rayet stars ( * ? ? ? * and references therein ) . since we consider `` average '' n and o abundances based on the abundances of different hii regions this origin of the scatter in our diagrams seems to be unlikely . it has been known for a long time that galaxies of different morphological types have different star formation histories , i.e. , spiral galaxies with early morphological types have a larger fraction of old stars @xcite . one can then expect that n / h at a given o / h may depend on the morphological type of galaxy expressed in terms of @xmath41 type @xcite . on the other hand , the star formation history of a galaxy also strongly depends on its mass ( or luminosity ) as is epitomized in the galaxy downsizing effect , where the star formation acitvity shifts from high - mass galaxies at early cosmic times to lower - mass galaxies at later epochs @xcite . again , one can expect then that the n / h ratio at a given o / h will depend on the galaxy mass @xcite . there is a relatively tight linear correlation between the absolute magnitudes and the logarithms of the linear diameters of nearby galaxies @xcite . therefore , the linear diameter of a galaxy can be used as indicator of its luminosity ( and mass ) . thus , if the spread in n / h at a given o / h is caused by the time delay between nitrogen and oxygen enrichment and the different star formation histories in different galaxies then the n / h at a given o / h should correlate with morphological @xmath41 type or / and with linear radius of the galaxy . figure [ figure : residual ] shows the residual of eqs . ( [ equation : nhro]),([equation : nhrg ] ) as a function of the morphological @xmath41 type and linear radius of the galaxy @xmath0 . figure [ figure : residual ] indicates that there may exist some correlation between the n / h at a given o / h and morphological type or / and linear radius of the galaxy . to verify this , we found a relation between n / h and o / h for spiral galaxies ( with @xmath41 @xmath75 7.5 ) where @xmath41 and log@xmath0 are `` secondary parameters '' , n / h = @xmath72(o / h,@xmath41,log@xmath0 ) . it should be noted that @xmath41 and log@xmath0 are not a perfectly independent parameters since there is some correlation between them . the relation obtained for central abundances is @xmath76 the values of the coefficients in the terms containing @xmath41 and log@xmath0 are similar . however , the @xmath41 value is a more important second parameter than the log@xmath0 value since the variation in @xmath41 values ( from 1 to 7.5 ) is much larger than the variation of log@xmath0 values ( from @xmath30.6 to @xmath31.5 ) . therefore , the variation in n / h due to variation in the morphological type is larger than that due to variation in the linear radii of galaxies . for example , the sab galaxies ( @xmath41 = 2 ) have nitrogen abundances larger than on average 0.1 dex ( i.e. , by around 30% ) than sd ( @xmath41 = 7 ) galaxies with the same oxygen abundances . the relation n / h = @xmath72(o / h,@xmath41,log@xmath0 ) obtained for abundances at the optical @xmath0 radii of the galaxies is @xmath77 in this case the log@xmath0 value is the more important second parameter than the @xmath41 value . although the morphological type is the more important second parameter in the relation for central abundances while log@xmath0 is the more important second parameter in the relation for abundances at the optical @xmath0 radii of the galaxies , it is difficult to draw a solid conclusion whether this is physically meaningful since there is a correlation between @xmath41 and log@xmath0 . that the relation between n / h and o / h depends on the additional parameter(s ) @xmath41 and/or log@xmath0 suggests that the scatter in n / h at a given o / h can be caused , at least partly , by the time delay between nitrogen and oxygen enrichment and the different star formation histories in different galaxies . the upper left panel of figure [ figure : xhrof ] shows the central oxygen ( o / h)@xmath47 abundance in a galaxy as a function of its morphological @xmath41 type ( data from table [ table : sample ] and table [ table : grad ] ) . the ( o / h)@xmath47 @xmath41 diagram shows that there is a trend in central oxygen abundance with morphological type for galaxies later than sc ( @xmath41 @xmath4 5 ) such that the central oxygen abundances are lower in galaxies of later types . this trend disappears for early - type spiral galaxies ( @xmath41 @xmath39 5 ) . the upper right panel of figure [ figure : xhrof ] shows the central oxygen abundance in a galaxy as a function of its isophotal radius @xmath0 . since there is a relatively tight linear correlation between the absolute magnitudes and the logarithms of the linear diameters of nearby galaxies @xcite , this diagram can be considered as some kind of analog of the standard `` luminosity metallicity '' diagram . the ( o / h)@xmath47 @xmath0 diagram shows that there is a weak correlation between the central oxygen abundance and optical radius for small galaxies ( @xmath0 @xmath39 10 kpc ) in the sense that the smaller galaxies have on average lower oxygen abundances . this correlation disappears for large galaxies ( @xmath0 @xmath4 10 kpc ) , i.e. , the most oxygen - rich galaxies of different radii have similar central oxygen abundances . different versions of the luminosity metallicity diagram have been constructed in earlier studies . @xmath78-band luminosity characteristic oxygen abundance diagrams were considered in @xcite , where the characteristic oxygen abundance is defined as the abundance at @xmath51 = 0.4@xmath0 . @xcite have used the global abundances ( in the sense that their abundances do not correspond to the abundances at a fixed galactocentric distance , but instead are some kind of mean abundance for a fraction of a galaxy within the fiber aperture ) estimated from sdss spectra in constructing the luminosity metallicity diagram . the luminosity central metallicity diagram was examined in @xcite . the flattening of the luminosity metallicity relation at high luminosities ( essentially a plateau ) can be seen in all versions of the diagram . thus the plateau in our ( o / h)@xmath47 @xmath0 diagram at large radii is consistent with previous results . it has been advocated that the constant maximum value of the observed central oxygen abundance in the most oxygen - rich galaxies suggests that the observed oxygen abundance in the centers of those galaxies represents the maximum attainable value of the gas - phase oxygen abundance @xcite . the upper - row panels of figure [ figure : xhrof ] show that the observed central oxygen abundance in the most oxygen - rich galaxies in our sample is 12 + log(o / h)@xmath47 @xmath3 8.85 . the dashed lines in the upper row of panels of figure [ figure : xhrof ] show this value of oxygen abundance , which seems to correspond to the maximum attainable value of the gas - phase oxygen abundance in galaxies . the observed central oxygen abundance in the most oxygen - rich galaxies from our sample is a factor @xmath32 higher than the gas - phase oxygen abundance in the solar neighbourhood , 12 + log(o / h ) @xmath3 8.5 ( e.g. , * ? ? ? the maximum attainable value of the oxygen abundance in galaxies obtained here is in agreement with the value from @xcite . because some fraction of the oxygen ( about 0.1 dex ) is expected to be locked in dust grains ( e.g. , * ? ? ? * ) , the maximum value of the true ( gas + dust ) oxygen abundances in hii regions of spiral galaxies is 12 + log(o / h)@xmath47 @xmath3 8.95 . the lower left panel of figure [ figure : xhrof ] shows the central nitrogen ( n / h)@xmath47 abundance in a galaxy disk as a function of its morphological @xmath41 type . the comparison of the panels in the left column of figure [ figure : xhrof ] shows that the changes in the central oxygen and nitrogen abundances with morphological type of a galaxy are rather similar . the lower right panel of figure [ figure : xhrof ] shows the central nitrogen abundance in a galaxy disk as a function of its optical isophotal radius @xmath0 . again , the comparison between the right column panels of figure [ figure : xhrof ] indicates that the general behaviour of the central nitrogen abundances as a function of optical radius is similar to that for oxygen . according to the relation between nitrogen and oxygen abundances given in eq . ( [ equation : nhro ] ) , the value of 12 + log(n / h)@xmath47 @xmath3 8.42 corresponds to the maximum attainable value of the gas - phase oxygen abundance in galaxies with 12 + log(o / h)@xmath47 @xmath3 8.85 . this value of nitrogen abundance is indicated in the lower row of panels of figure [ figure : xhrof ] by the dashed lines . one can see that the value of 12 + log(n / h)@xmath47 @xmath3 8.42 can be adopted as the maximum value of the observed central nitrogen abundance in the most nitrogen - rich galaxies at the present - day epoch . however , in contrast to the case of oxygen , it is not necessary that this value corresponds to the the maximum attainable value of the gas - phase oxygen abundance in galaxies because of the time - delay between nitrogen and oxygen enrichment of the interstellar medium . the panels in the left column of figure [ figure : xhrgf ] show the oxygen ( o / h)@xmath69 ( upper panel ) and nitrogen ( n / h)@xmath69 ( lower panel ) abundances measured at the isophotal radius , @xmath0 , as a function of morphological @xmath41 type . a comparison between the panels in the left column of figure [ figure : xhrof ] and figure [ figure : xhrgf ] shows that the changes in the central oxygen ( nitrogen ) abundances and in the abundances at the optical edge of the disk along the hubble sequence ( or morphological type ) are more or less similar ( at least qualitatively ) . however , the oxygen and nitrogen abundances at the @xmath0 radius in four of the late - type galaxies ( ngc 4625 , ic 10 , ugc 223 , and pgc 29167 ) are high and those galaxies show a large deviation from the general trends . ngc 4625 is a magellanic spiral with @xmath0 @xmath3 3 kpc and a very extended faint disk ( see below ) . the dwarf irregular galaxy ic 10 has a very small optical radius , @xmath0 = 0.6 kpc , and a positive radial abundance gradient . one may suggest though that any radial gradient in such small galaxies is hard to define based on hii regions . it remains a mystery why the oxygen abundance changes by a factor of about two on the scale of 0.6 kpc in this galaxy . ic 10 is one of several dwarf galaxies in which evidence for localized , inhomogeneous enrichment has been found @xcite perhaps this is a common mode of enrichment in these small objects . in contrast , the irregular galaxy ugc 223 is rather large with @xmath0 = 9.3 kpc . the peculiar dwarf galaxy pgc 29167 ( or `` garland '' ) is commonly considered a tidal dwarf galaxy ( see comment on this galaxy in section 4.2 ) . the panels in the right column of figure [ figure : xhrgf ] show the oxygen ( o / h)@xmath69 ( upper panel ) and nitrogen ( n / h)@xmath69 ( lower panel ) abundances at the optical edge of the disk as defined by the @xmath0 radius as a function of isophotal radius @xmath0 . the positions of the galaxies in the ( o / h)@xmath69 @xmath0 and ( n / h)@xmath69 @xmath0 diagrams do not show any obvious trends . in particular , the ( o / h)@xmath69 and ( n / h)@xmath69 abundances do not show any appreciable correlation with isophotal radius @xmath0 . however , one feature in these diagrams should be noted : the ( o / h)@xmath69 and ( n / h)@xmath69 abundances have a maximum value in galaxies with an isophotal radius @xmath0 @xmath3 10 kpc and decrease when moving from this value both towards smaller or larger optical radii . figure [ figure : gradf ] shows the radial oxygen and nitrogen abundance gradients in units of dex kpc@xmath53 as a function of morphological @xmath41 type ( left panels ) and of isophotal radius @xmath0 of a galaxy ( right panels ) . inspection of the left column panels of figure [ figure : xhrgf ] shows that the values of the abundance gradients in units of dex kpc@xmath53 do not correlate with the morphological type of a galaxy . according to @xcite , the lack of a correlation between gradients in units of dex kpc@xmath53 and the macroscopic properties of late - type galaxies may suggest that the relationship between these parameters is more complex than a simple correlation . indeed , @xcite have concluded that a correlation for non - barred galaxies is seen . the panels in the right column of figure [ figure : xhrgf ] show that shallow gradients can be found both in small and large galaxies while steep gradients are seen only in small galaxies in the sense that the smaller a galaxy the steeper its gradient . we compiled published spectra of hii regions in 130 nearby galaxies . our list contains 3904 spectra including 162 spectra of hii regions beyond the isophotal radius @xmath0 . the oxygen and nitrogen abundances in hii regions were determined on the metallicity scale defined by hii regions with @xmath2-based abundances . the radial gradients of oxygen and nitrogen abundances across the disks of the galaxies were estimated . at the centers of metal - rich galaxies ( i.e. , ( 12 + log(o / h ) @xmath4 8.2 ) , we found the relation between n and o abundances to be ( n / h)@xmath79 @xmath80 ( o / h)@xmath81 . the ( n / h)@xmath69 = @xmath72(o / h)@xmath69 relation between n and o abundances at the @xmath0 isophotal radii of high metallicity galaxies is similar to that for the abundances at their centers . the variation in ( n / h ) at a given ( o / h ) is around 0.3 dex . to test whether the scatter in n / h at a given o / h can be explained by the time delay between nitrogen and oxygen enrichment and the different star formation histories in galaxies of different morphological types and dimensions ( masses ) , we derived a more complex relation between n and o abundances ( n / h ) = @xmath72((o / h),@xmath41,@xmath0 ) . we found that the morphological type , @xmath41 , is a more important `` second parameter '' in the relation for central abundances , while the log@xmath0 is a more important second parameter in the relation for abundances at the @xmath0 radii of our galaxies . since there is a correlation between @xmath41 and log@xmath0 it is as yet unclear whether this difference is meaningful the fact that the relation between n / h and o / h depends on additional parameter(s ) , namely @xmath41 and/or log@xmath0 suggests that the scatter in n / h at a given o / h can be caused , at least partly , by the time delay between nitrogen and oxygen enrichment and the different star formation histories in different galaxies . the best fit to n / h as a function of o / h is close to a linear relation at low metallicity ( 12 + log(o / h ) @xmath39 8.0 ) . the central oxygen abundances ( o / h)@xmath79 show a trend along the hubble sequence of galaxies of late morphological types ( @xmath41 @xmath4 5 ) such that the oxygen abundances are lower in galaxies of later types . this trend disappears for early morphological types . the central oxygen abundance also correlates with optical galaxy radius for small galaxies , @xmath0 @xmath39 10 kpc , being lower in galaxies of smaller radii . the trend disappears for galaxies with large radii . the maximum gas - phase oxygen abundance in large ( 10 kpc @xmath39 r@xmath82 @xmath39 30 kpc ) galaxies ( or in galaxies of early ( 1 @xmath39 @xmath41 @xmath39 5 ) morphological types ) is constant , 12 + log(o / h ) @xmath3 8.85 . this implies that the observed central oxygen abundance of the most oxygen - rich galaxies in our sample is a factor of @xmath32 higher than the gas - phase oxygen abundance in the solar neighbourhood . the central nitrogen abundances ( n / h)@xmath47 show a similar behaviour . the observed central nitrogen abundance in the most nitrogen - rich galaxies of our sample is 12 + log(o / h ) @xmath3 8.42 . the radial o and n abundance gradients ( in units of dex kpc@xmath53 ) within the optical radius do not show any significant correlation with the morphological type and optical radius . however , the spread in the gradients increases with decreasing galaxy radius in the sense that shallow gradients are seen both in small and large galaxies while steep gradients occur only in a small galaxies . the smaller a galaxy the steeper is the gradient that it may show . the abundance data set presented in this paper serves as the foundation for other investigations we are carrying out . in a forthcoming paper @xcite , we examine relations between the radial abundance distribution across the disk and the disk surface brightness profiles in the optical @xmath78 and infrared @xmath83 bands for a sample of nearby galaxies . we are grateful to the referee for his or her constructive comments . l.s.p . and e.k.g . acknowledge support within the framework of sonderforschungsbereich ( sfb 881 ) on `` the milky way system '' ( especially subproject a5 ) , which is funded by the german research foundation ( dfg ) . thanks the astronomisches rechen - institut at heidelberg university where this investigation was carried out for the hospitality . a.y.k . acknowledges the support from the national research foundation ( nrf ) of south africa . + we thank h.j . zahid and f. bresolin for supporting us with some unpublished details of their observations of hii regions in ngc 3359 . + the authors acknowledge the work of the sdss collaboration . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy office of science . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . + we acknowledge the usage of the hyperleda database ( http://leda.univ-lyon1.fr ) . llcrrrrccrcr ngc 0012 & sc & 4.9 & 2.486458 & 4.612528 & 0.85 & 125 & 27 & 1 & 53.20 & 2 & 13.14 + ngc 0055 & sbm & 8.8 & 3.723333 & -39.196639 & 16.18 & 110 & 79 & 3 & 1.91 & 4 & 8.99 + ngc 0099 & sc & 5.9 & 5.997470 & 15.770428 & 0.71 & 0 & 24 & 1 & 77.00 & 5 & 15.82 + ngc 0224 & sb & 3.0 & 10.684792 & 41.269065 & 95.27 & 35 & 77 & 6 & 0.74 & 7 & 20.51 + ngc 0234 & sabc & 5.3 & 10.884977 & 14.342564 & 0.81 & 0 & 0 & 1 & 59.60 & 2 & 14.06 + ngc 0253 & sabc & 5.1 & 11.888000 & -25.288219 & 13.77 & 50 & 76 & 8 & 3.46 & 9 & 13.86 + ngc 0300 & scd & 6.9 & 13.722833 & -37.684389 & 10.94 & 106 & 42 & 10 & 1.94 & 4 & 6.17 + ngc 0428 & sabm & 8.6 & 18.232125 & 0.981557 & 2.04 & 120 & 41 & 1 & 14.80 & 11 & 8.77 + ngc 0450 & sabc & 5.8 & 18.876840 & -0.860974 & 1.54 & 72 & 41 & 1 & 18.70 & 11 & 8.40 + ngc 0493 & sc & 5.9 & 20.537458 & 0.945362 & 1.69 & 58 & 72 & 1 & 23.20 & 11 & 11.43 + ngc 0575 & sc & 5.3 & 22.694333 & 21.440417 & 0.85 & 0 & 21 & 1 & 41.80 & 2 & 10.32 + ngc 0598 & sc & 5.9 & 23.462042 & 30.660222 & 35.40 & 23 & 56 & 12 & 0.76 & 4 & 7.83 + ngc 0628 & sc & 5.2 & 24.173946 & 15.783661 & 5.24 & 25 & 6 & 13 & 9.91 & 14 & 15.09 + ngc 0753 & sabc & 4.9 & 29.425833 & 35.916125 & 1.26 & 128 & 50 & 15 & 67.10 & 16 & 24.52 + ngc 0783 & sc & 5.3 & 30.277542 & 31.882472 & 0.79 & 35 & 27 & 1 & 69.30 & 2 & 15.98 + ngc 0925 & scd & 7.0 & 36.820333 & 33.579167 & 5.24 & 101 & 58 & 17 & 9.29 & 18 & 14.15 + ngc 1055 & sbb & 3.1 & 40.438458 & 0.443168 & 3.79 & 105 & 69 & 1 & 14.40 & 11 & 15.89 + ngc 1058 & sc & 5.1 & 40.874988 & 37.341339 & 1.51 & 145 & 15 & 19 & 10.60 & 20 & 4.66 + ngc 1068 & sb & 3.0 & 40.669629 & -0.013281 & 3.54 & 70 & 32 & 1 & 10.10 & 11 & 10.40 + ngc 1090 & sbc & 3.8 & 41.641401 & -0.247158 & 1.99 & 102 & 64 & 1 & 35.90 & 16 & 20.79 + ngc 1097 & sbb & 3.3 & 41.579375 & -30.274889 & 4.67 & 134 & 46 & 21 & 16.50 & 2 & 22.40 + ngc 1232 & sabc & 5.0 & 47.439625 & -20.579306 & 3.71 & 90 & 30 & 22 & 21.00 & 2 & 22.64 + ngc 1313 & sbcd & 7.0 & 49.566875 & -66.498250 & 4.56 & 4 & 38 & 23 & 4.39 & 24 & 5.82 + ngc 1365 & sb & 3.2 & 53.401546 & -36.140403 & 5.61 & 42 & 46 & 25 & 19.60 & 18 & 32.00 + ngc 1512 & sa & 1.1 & 60.976167 & -43.348861 & 4.46 & 80 & 35 & 26 & 9.50 & 27 & 12.32 + ngc 1598 & sbc & 4.9 & 67.140292 & -47.782556 & 0.72 & 123 & 55 & 1 & 55.80 & 16 & 11.73 + ngc 1637 & sc & 5.0 & 70.367408 & -2.857961 & 1.99 & 31 & 40 & 19 & 12.00 & 18 & 6.95 + ngc 1642 & sc & 5.2 & 70.728792 & 0.618583 & 0.91 & 175 & 29 & 1 & 62.10 & 2 & 16.43 + ngc 1672 & sb & 3.3 & 71.427125 & -59.247278 & 3.30 & 170 & 34 & 1 & 16.70 & 2 & 16.05 + ngc 2336 & sbc & 4.0 & 111.766875 & 80.178083 & 2.51@xmath84 & 175 & 55 & 28 & 34.20 & 2 & 24.95 + ngc 2403 & sabc & 6.0 & 114.214167 & 65.602556 & 10.94 & 126 & 58 & 29 & 3.06 & 18 & 9.74 + ngc 2441 & sabb & 3.3 & 117.978083 & 73.015694 & 1.00 & 0 & 29 & 1 & 50.90 & 2 & 14.77 + ngc 2442 & sbc & 3.7 & 114.099333 & -69.530833 & 2.75 & 27 & 29 & 30 & 17.10 & 31 & 13.67 + ngc 2541 & sabc & 6.0 & 123.667148 & 49.061709 & 3.16 & 165 & 60 & 1 & 12.60 & 18 & 11.56 + ngc 2805 & sabc & 6.9 & 140.085042 & 64.102778 & 3.16 & 125 & 41 & 1 & 28.00 & 31 & 25.70 + ngc 2835 & sc & 5.0 & 139.470458 & -22.354667 & 3.30 & 160 & 45 & 32 & 10.10 & 16 & 9.71 + ngc 2841 & sb & 3.0 & 140.510975 & 50.976519 & 4.06 & 150 & 66 & 33 & 14.10 & 18 & 16.67 + ngc 2903 & sabb & 4.0 & 143.042125 & 21.500833 & 6.30 & 22 & 65 & 34 & 8.90 & 35 & 16.30 + ngc 2997 & sabc & 5.1 & 146.411625 & -31.191083 & 4.46 & 107 & 32 & 36 & 12.20 & 36 & 15.82 + ngc 3020 & sc & 5.9 & 147.527500 & 12.813639 & 1.58 & 105 & 60 & 1 & 21.90 & 11 & 10.07 + ngc 3023 & sabc & 5.5 & 147.469112 & 0.618167 & 1.44 & 70 & 60 & 1 & 29.40 & 2 & 12.33 + ngc 3031 & sab & 2.4 & 148.888221 & 69.065295 & 13.46 & 150 & 59 & 37 & 3.63 & 18 & 14.21 + ngc 3184 & sabc & 5.9 & 154.570258 & 41.424053 & 3.71 & 135 & 21 & 1 & 12.60 & 14 & 13.59 + ngc 3198 & sc & 5.2 & 154.978966 & 45.549623 & 4.26 & 35 & 67 & 1 & 14.50 & 18 & 17.95 + ngc 3227 & saba & 1.5 & 155.877412 & 19.865050 & 2.68 & 155 & 47 & 1 & 20.60 & 31 & 16.09 + ngc 3239 & ib & 9.8 & 156.270375 & 17.163583 & 2.51 & 80 & 49 & 38 & 8.10 & 31 & 5.90 + ngc 3310 & sabb & 4.0 & 159.691083 & 53.503383 & 1.54 & 163 & 39 & 39 & 19.20 & 2 & 8.63 + ngc 3319 & sbc & 6.0 & 159.789414 & 41.686675 & 3.08 & 37 & 57 & 1 & 14.10 & 18 & 12.64 + ngc 3344 & sbc & 4.0 & 160.879792 & 24.922222 & 3.54 & 156 & 25 & 40 & 6.10 & 31 & 6.28 + ngc 3351 & sb & 3.1 & 160.990417 & 11.703806 & 3.71 & 13 & 42 & 26 & 10.50 & 18 & 11.32 + ngc 3359 & sc & 5.2 & 161.653596 & 63.224236 & 3.62 & 172 & 51 & 41 & 14.40 & 42 & 15.17 + ngc 3486 & sc & 5.2 & 165.099452 & 28.975137 & 3.54 & 80 & 42 & 1 & 9.16 & 2 & 9.43 + ngc 3521 & sabb & 4.0 & 166.452421 & -0.035864 & 5.48 & 160 & 73 & 37 & 10.70 & 43 & 17.06 + ngc 3621 & sbcd & 6.9 & 169.568792 & -32.814056 & 4.89@xmath85 & 165 & 65 & 37 & 7.24 & 18 & 10.29 + ngc 3631 & sc & 5.2 & 170.261976 & 53.169569 & 2.51 & 152 & 20 & 86 & 22.2 & 2 & 16.18 + ngc 3718 & sa & 1.1 & 173.145221 & 53.067922 & 3.77@xmath86 & 15 & 65 & 44 & 17.00 & 31 & 18.63 + ngc 3820 & sbc & 3.6 & 175.520442 & 10.384253 & 0.33 & 26 & 42 & 45 & 91.30 & 2 & 8.77 + ngc 3893 & sabc & 5.2 & 177.159125 & 48.710833 & 2.23 & 165 & 52 & 1 & 18.20 & 11 & 11.82 + ngc 3938 & sc & 5.2 & 178.206042 & 44.120722 & 2.68 & 15 & 24 & 1 & 17.90 & 46 & 13.98 + ngc 4030 & sbc & 4.0 & 180.098445 & -1.100095 & 2.08 & 27 & 44 & 1 & 27.20 & 11 & 16.49 + ngc 4088 & sabc & 4.7 & 181.392458 & 50.539028 & 2.88 & 43 & 67 & 1 & 14.30 & 11 & 11.97 + ngc 4109 & sa & 1.1 & 181.713008 & 42.995642 & 0.50 & 147 & 44 & 45 & 102.80 & 2 & 14.95 + ngc 4254 & sc & 5.2 & 184.706683 & 14.416508 & 2.68 & 59 & 29 & 19 & 14.30 & 16 & 11.17 + ngc 4258 & sabb & 4.0 & 184.739600 & 47.303972 & 9.31 & 150 & 72 & 47 & 7.20 & 48 & 19.50 + ngc 4303 & sbc & 4.0 & 185.478729 & 4.473647 & 3.23 & 137 & 27 & 19 & 13.60 & 2 & 12.77 + ngc 4321 & sabb & 4.1 & 185.728462 & 15.821817 & 3.71 & 30 & 32 & 1 & 17.20 & 18 & 18.55 + ngc 4395 & sm & 8.9 & 186.453592 & 33.546928 & 6.59 & 147 & 46 & 49 & 4.30 & 50 & 8.24 + ngc 4490 & sbcd & 7.0 & 187.650996 & 41.643898 & 3.16 & 125 & 61 & 1 & 7.80 & 31 & 7.16 + ngc 4501 & sb & 3.3 & 187.996503 & 14.420387 & 3.46 & 141 & 63 & 51 & 14.10 & 52 & 14.19 + ngc 4535 & sc & 5.0 & 188.584625 & 8.197750 & 3.54 & 0 & 45 & 1 & 17.80 & 18 & 18.33 + ngc 4559 & sc & 6.0 & 188.990195 & 27.959992 & 5.36 & 143 & 67 & 53 & 7.87 & 16 & 12.27 + ngc 4625 & sabm & 8.8 & 190.469671 & 41.273964 & 1.09 & 130 & 27 & 54 & 9.20 & 18 & 2.93 + ngc 4631 & sbcd & 6.6 & 190.533375 & 32.541500 & 7.74 & 86 & 80 & 1 & 7.62 & 55 & 17.17 + ngc 4651 & sc & 5.2 & 190.927625 & 16.393389 & 1.99 & 78 & 48 & 19 & 23.60 & 56 & 13.67 + ngc 4654 & sc & 5.9 & 190.985750 & 13.126681 & 2.45 & 122 & 57 & 19 & 16.10 & 16 & 11.47 + ngc 4656 & sbm & 9.0 & 190.990531 & 32.168150 & 7.57 & 33 & 79 & 1 & 8.20 & 2 & 18.05 + ngc 4713 & scd & 6.8 & 192.491139 & 5.311405 & 1.35 & 50 & 35 & 57 & 14.90 & 11 & 5.83 + ngc 4725 & saba & 2.2 & 192.610755 & 25.500805 & 5.36 & 35 & 45 & 1 & 13.50 & 18 & 21.04 + ngc 4736 & sab & 2.4 & 192.721088 & 41.120458 & 5.61 & 116 & 37 & 58 & 4.66 & 24 & 7.60 + ngc 4861 & sm & 8.9 & 194.759754 & 34.859444 & 1.99 & 23 & 69 & 59 & 7.59 & 11 & 4.40 + ngc 5033 & sc & 5.1 & 198.364441 & 36.569936 & 5.36 & 170 & 62 & 1 & 18.70 & 16 & 29.14 + ngc 5055 & sbc & 4.0 & 198.955542 & 42.029278 & 6.30 & 101 & 64 & 33 & 7.94 & 11 & 14.54 + ngc 5068 & sc & 6.0 & 199.728375 & -21.039111 & 3.62 & 110 & 29 & 1 & 5.40 & 60 & 5.69 + ngc 5194 & sbc & 4.0 & 202.469629 & 47.195172 & 5.61 & 163 & 52 & 1 & 8.90 & 61 & 14.52 + ngc 5236 & sc & 5.0 & 204.253958 & -29.865417 & 6.44 & 46 & 24 & 58 & 4.61 & 18 & 8.64 + ngc 5248 & sabb & 4.0 & 204.383433 & 8.885178 & 3.08 & 110 & 44 & 1 & 22.90 & 2 & 20.54 + ngc 5457 & sabc & 6.0 & 210.802267 & 54.348950 & 14.42 & 37 & 18 & 62 & 6.85 & 18 & 28.73 + ngc 5474 & sc & 6.1 & 211.256708 & 53.662222 & 2.39 & 159 & 21 & 63 & 6.80 & 35 & 4.73 + ngc 5668 & scd & 6.9 & 218.351408 & 4.450458 & 1.66 & 164 & 21 & 64 & 26.90 & 31 & 12.96 + ngc 6384 & sabb & 3.6 & 263.101250 & 7.060278 & 3.08 & 32 & 48 & 19 & 25.90 & 11 & 23.23 + ngc 6691 & sbc & 3.6 & 279.801042 & 55.641806 & 0.79 & 0 & 21 & 1 & 85.10 & 2 & 19.62 + ngc 6744 & sbc & 4.0 & 287.442125 & -63.857528 & 9.98 & 20 & 50 & 65 & 9.40 & 16 & 27.28 + ngc 6946 & sabc & 5.9 & 308.718012 & 60.153914 & 5.74 & 62 & 38 & 66 & 6.00 & 67 & 10.02 + ngc 7331 & sbc & 3.9 & 339.266725 & 34.415519 & 5.24 & 169 & 75 & 28 & 15.10 & 68 & 23.00 + ngc 7495 & sabc & 5.3 & 347.238250 & 12.048028 & 0.91 & 5 & 24 & 1 & 70.00 & 69 & 18.53 + ngc 7518 & saba & 1.1 & 348.303042 & 6.321667 & 0.71 & 109 & 42 & 70 & 50.50 & 70 & 10.38 + ngc 7529 & sbc & 4.4 & 348.513292 & 8.992500 & 0.43 & 0 & 17 & 1 & 62.10 & 2 & 7.69 + ngc 7570 & sa & 1.1 & 349.186134 & 13.483011 & 0.77 & 30 & 54 & 1 & 17.70 & 16 & 3.99 + ngc 7591 & sbbc & 3.6 & 349.567833 & 6.585806 & 0.98 & 148 & 68 & 71 & 60.20 & 16 & 17.07 + ngc 7678 & sc & 4.9 & 352.116250 & 22.421194 & 1.17 & 5 & 45 & 1 & 48.90 & 2 & 16.67 + ngc 7793 & scd & 7.4 & 359.457625 & -32.591028 & 4.67 & 99 & 54 & 10 & 3.40 & 72 & 4.62 + ic 0010 & ib & 9.9 & 5.072250 & 59.303778 & 3.16 & 125 & 27 & 45 & 0.66 & 73 & 0.61 + ic 0193 & sc & 5.3 & 30.629083 & 11.093083 & 0.93 & 154 & 32 & 1 & 61.60 & 2 & 16.68 + ic 0208 & sbc & 3.6 & 32.115583 & 6.394917 & 0.91 & 0 & 12 & 1 & 38.30 & 16 & 10.14 + ic 0342 & sabc & 6.0 & 56.702096 & 68.096367 & 10.69 & 37 & 31 & 74 & 3.30 & 68 & 10.26 + ic 1132 & sc & 5.4 & 235.028171 & 20.680606 & 0.59 & 50 & 18 & 75 & 70.60 & 2 & 12.06 + ic 2204 & sab & 2.5 & 115.325435 & 34.232169 & 0.52 & 0 & 0 & 1 & 66.20 & 76 & 10.08 + ic 5201 & sc & 6.1 & 335.239333 & -46.035861 & 4.26 & 33 & 63 & 1 & 12.80 & 2 & 15.85 + ic 5309 & sb & 3.1 & 349.798542 & 8.109306 & 0.67 & 20 & 63 & 77 & 50.00 & 16 & 9.81 + ugc 00223 & i & 9.9 & 6.177915 & 14.824582 & 0.46 & 0 & 0 & 1 & 70.80 & 2 & 9.39 + ugc 01087 & sc & 5.3 & 22.860833 & 14.277500 & 0.74 & 0 & 21 & 1 & 59.60 & 2 & 12.82 + ugc 01862 & sabc & 6.4 & 36.103375 & -2.162361 & 0.83 & 10 & 39 & 1 & 20.10 & 16 & 4.85 + ugc 02023 & i & 9.7 & 37.572417 & 33.491111 & 0.83 & 59 & 26 & 78 & 10.30 & 79 & 2.49 + ugc 02216 & i & 9.9 & 41.087913 & 0.675381 & 0.64 & 15 & 72 & 1 & 36.50 & 2 & 6.84 + ugc 02345 & sm & 8.8 & 42.969583 & -1.172222 & 1.73 & 160 & 30 & 1 & 20.00 & 2 & 10.09 + ugc 03701 & sc & 6.0 & 107.927833 & 72.169306 & 0.91 & 0 & 0 & 1 & 43.20 & 2 & 11.43 + ugc 04107 & sc & 5.3 & 119.257792 & 49.567361 & 0.71 & 0 & 12 & 1 & 51.50 & 2 & 10.58 + ugc 04305 & i & 9.9 & 124.770750 & 70.720027 & 3.97 & 177 & 40 & 80 & 3.38 & 9 & 3.90 + ugc 05100 & sbb & 3.0 & 143.661008 & 5.841439 & 0.60 & 30 & 53 & 1 & 80.60 & 2 & 14.09 + ugc 06410 & sabc & 5.3 & 171.024500 & 45.811072 & 0.55 & 5 & 39 & 1 & 82.50 & 2 & 13.16 + ugc 08091 & i & 9.8 & 194.668512 & 14.217511 & 0.54 & 170 & 20 & 81 & 2.08 & 9 & 0.32 + ugc 09562 & sd & 8.0 & 222.810048 & 35.542269 & 0.48 & 30 & 27 & 1 & 23.80 & 31 & 3.31 + ugc 09837 & sabc & 5.3 & 230.965311 & 58.052941 & 0.91 & 21 & 18 & 82 & 42.00 & 82 & 11.12 + ugc 10445 & sc & 6.0 & 248.448417 & 28.984778 & 1.38 & 142 & 46 & 83 & 20.00 & 2 & 8.01 + ugc 12709 & sabm & 8.7 & 354.350001 & 0.391670 & 1.51 & 145 & 47 & 1 & 37.00 & 2 & 16.25 + pgc 29167 & i & 10.0 & 150.965417 & 68.690278 & 2.70@xmath87 & 0 & 0 & & 3.82 & 84 & 3.00 + pgc 41318 & s ? & 4.3 & 187.589848 & 56.791542 & 0.29 & 0 & 21 & 45 & 70.10 & 2 & 6.00 + pgc 44772 & s ? & 1.9 & 195.138542 & 10.129969 & 0.40 & 128 & 32 & 45 & 107.80 & 2 & 12.39 + pgc 45195 & sd & 7.7 & 196.129979 & -3.572419 & 1.77 & 63 & 44 & 1 & 24.10 & 2 & 12.44 + pgc 46182 & sc & 4.9 & 199.061225 & 41.494460 & 0.54@xmath88 & 34 & 52 & 85 & 89.40 & 76 & 14.15 + ; lccrrrrcc ngc 0012 & 8.56 @xmath73 0.02 & 0.015 @xmath73 0.042 & 0.0011 @xmath73 0.0032 & 0.033 & 7.74 @xmath73 0.04 & -0.008 @xmath73 0.087 & -0.0006 @xmath73 0.0066 & 0.068 + ngc 0055 & 8.04 @xmath73 0.05 & -0.053 @xmath73 0.094 & -0.0059 @xmath73 0.0104 & 0.075 & 6.50 @xmath73 0.07 & -0.037 @xmath73 0.131 & -0.0042 @xmath73 0.0145 & 0.091 + ngc 0099 & 8.55 @xmath73 0.04 & -0.319 @xmath73 0.072 & -0.0202 @xmath73 0.0045 & 0.055 & 7.70 @xmath73 0.07 & -0.666 @xmath73 0.137 & -0.0422 @xmath73 0.0087 & 0.105 + ngc 0224 & 8.72 @xmath73 0.02 & -0.352 @xmath73 0.039 & -0.0172 @xmath73 0.0019 & 0.077 & 8.17 @xmath73 0.04 & -0.996 @xmath73 0.072 & -0.0486 @xmath73 0.0035 & 0.136 + ngc 0234 & 8.65 @xmath73 0.01 & -0.154 @xmath73 0.022 & -0.0109 @xmath73 0.0015 & 0.027 & 7.91 @xmath73 0.02 & -0.329 @xmath73 0.046 & -0.0234 @xmath73 0.0033 & 0.058 + ngc 0253 & 8.55 @xmath73 0.13 & -0.125 @xmath73 0.164 & -0.0090 @xmath73 0.0119 & 0.055 & 7.60 @xmath73 0.35 & -0.241 @xmath73 0.441 & -0.0174 @xmath73 0.0319 & 0.149 + ngc 0300 & 8.51 @xmath73 0.02 & -0.519 @xmath73 0.040 & -0.0842 @xmath73 0.0065 & 0.069 & 7.47 @xmath73 0.04 & -0.955 @xmath73 0.080 & -0.1548 @xmath73 0.0129 & 0.130 + ngc 0428 & 8.20 @xmath73 0.06 & -0.035 @xmath73 0.109 & -0.0040 @xmath73 0.0124 & 0.059 & 6.95 @xmath73 0.11 & -0.118 @xmath73 0.184 & -0.0135 @xmath73 0.0210 & 0.099 + ngc 0450 & 8.35 @xmath73 0.03 & -0.285 @xmath73 0.057 & -0.0340 @xmath73 0.0067 & 0.046 & 7.24 @xmath73 0.06 & -0.589 @xmath73 0.112 & -0.0702 @xmath73 0.0133 & 0.092 + ngc 0493 & 8.51 @xmath73 0.04 & -0.450 @xmath73 0.070 & -0.0393 @xmath73 0.0061 & 0.053 & 7.53 @xmath73 0.07 & -0.893 @xmath73 0.124 & -0.0780 @xmath73 0.0109 & 0.093 + ngc 0575 & 8.58 @xmath73 0.01 & -0.195 @xmath73 0.026 & -0.0189 @xmath73 0.0025 & 0.026 & 7.82 @xmath73 0.03 & -0.530 @xmath73 0.061 & -0.0514 @xmath73 0.0059 & 0.060 + ngc 0598 & 8.48 @xmath73 0.02 & -0.359 @xmath73 0.041 & -0.0459 @xmath73 0.0053 & 0.073 & 7.47 @xmath73 0.04 & -0.691 @xmath73 0.084 & -0.0883 @xmath73 0.0107 & 0.145 + ngc 0628 & 8.78 @xmath73 0.01 & -0.572 @xmath73 0.027 & -0.0379 @xmath73 0.0018 & 0.060 & 8.17 @xmath73 0.03 & -1.292 @xmath73 0.064 & -0.0856 @xmath73 0.0042 & 0.137 + ngc 0753 & 8.67 @xmath73 0.08 & -0.212 @xmath73 0.106 & -0.0087 @xmath73 0.0043 & 0.065 & 8.01 @xmath73 0.20 & -0.399 @xmath73 0.290 & -0.0163 @xmath73 0.0118 & 0.177 + ngc 0783 & 8.68 @xmath73 0.07 & -0.189 @xmath73 0.104 & -0.0118 @xmath73 0.0056 & 0.031 & 7.91 @xmath73 0.30 & -0.274 @xmath73 0.434 & -0.0171 @xmath73 0.0272 & 0.130 + ngc 0925 & 8.48 @xmath73 0.02 & -0.473 @xmath73 0.037 & -0.0334 @xmath73 0.0026 & 0.067 & 7.45 @xmath73 0.04 & -0.941 @xmath73 0.071 & -0.0665 @xmath73 0.0050 & 0.128 + ngc 1055 & 8.62 @xmath73 0.04 & -0.140 @xmath73 0.100 & -0.0088 @xmath73 0.0063 & 0.059 & 7.89 @xmath73 0.05 & -0.414 @xmath73 0.118 & -0.0261 @xmath73 0.0074 & 0.070 + ngc 1058 & 8.62 @xmath73 0.01 & -0.352 @xmath73 0.027 & -0.0757 @xmath73 0.0058 & 0.055 & 7.72 @xmath73 0.03 & -0.751 @xmath73 0.069 & -0.1613 @xmath73 0.0149 & 0.137 + ngc 1068 & 8.64 @xmath73 0.02 & -0.121 @xmath73 0.051 & -0.0116 @xmath73 0.0049 & 0.070 & 7.94 @xmath73 0.04 & -0.141 @xmath73 0.100 & -0.0136 @xmath73 0.0096 & 0.126 + ngc 1090 & 8.61 @xmath73 0.03 & -0.284 @xmath73 0.103 & -0.0137 @xmath73 0.0050 & 0.026 & 7.89 @xmath73 0.04 & -0.775 @xmath73 0.131 & -0.0373 @xmath73 0.0063 & 0.033 + ngc 1097 & 8.74 @xmath73 0.01 & -0.357 @xmath73 0.054 & -0.0159 @xmath73 0.0024 & 0.030 & 8.25 @xmath73 0.05 & -0.997 @xmath73 0.209 & -0.0445 @xmath73 0.0093 & 0.114 + ngc 1232 & 8.78 @xmath73 0.04 & -0.573 @xmath73 0.068 & -0.0253 @xmath73 0.0030 & 0.040 & 8.12 @xmath73 0.10 & -1.166 @xmath73 0.172 & -0.0515 @xmath73 0.0076 & 0.103 + ngc 1313 & 8.21 @xmath73 0.03 & -0.159 @xmath73 0.051 & -0.0273 @xmath73 0.0087 & 0.071 & 6.84 @xmath73 0.05 & -0.269 @xmath73 0.096 & -0.0462 @xmath73 0.0165 & 0.132 + ngc 1365 & 8.64 @xmath73 0.02 & -0.253 @xmath73 0.033 & -0.0079 @xmath73 0.0010 & 0.064 & 7.91 @xmath73 0.04 & -0.622 @xmath73 0.080 & -0.0193 @xmath73 0.0024 & 0.137 + ngc 1512 & 8.79 @xmath73 0.03 & -0.461 @xmath73 0.048 & -0.0374 @xmath73 0.0039 & 0.066 & 8.42 @xmath73 0.05 & -1.410 @xmath73 0.083 & -0.1144 @xmath73 0.0067 & 0.115 + ngc 1598 & 8.74 @xmath73 0.03 & -0.402 @xmath73 0.062 & -0.0343 @xmath73 0.0053 & 0.042 & 8.20 @xmath73 0.09 & -1.014 @xmath73 0.178 & -0.0865 @xmath73 0.0152 & 0.122 + ngc 1637 & 8.64 @xmath73 0.01 & -0.115 @xmath73 0.023 & -0.0165 @xmath73 0.0034 & 0.062 & 7.78 @xmath73 0.03 & -0.170 @xmath73 0.067 & -0.0236 @xmath73 0.0096 & 0.160 + ngc 1642 & 8.64 @xmath73 0.01 & -0.154 @xmath73 0.024 & -0.0094 @xmath73 0.0015 & 0.030 & 7.86 @xmath73 0.02 & -0.244 @xmath73 0.050 & -0.0148 @xmath73 0.0031 & 0.062 + ngc 1672 & 8.67 @xmath73 0.02 & -0.216 @xmath73 0.070 & -0.0135 @xmath73 0.0044 & 0.055 & 8.00 @xmath73 0.05 & -0.464 @xmath73 0.158 & -0.0290 @xmath73 0.0098 & 0.119 + ngc 2336 & 8.80 @xmath73 0.04 & -0.390 @xmath73 0.056 & -0.0156 @xmath73 0.0022 & 0.041 & 8.14 @xmath73 0.08 & -0.725 @xmath73 0.121 & -0.0290 @xmath73 0.0049 & 0.088 + ngc 2403 & 8.48 @xmath73 0.02 & -0.524 @xmath73 0.043 & -0.0538 @xmath73 0.0044 & 0.079 & 7.47 @xmath73 0.03 & -1.005 @xmath73 0.069 & -0.1032 @xmath73 0.0071 & 0.128 + ngc 2441 & 8.54 @xmath73 0.01 & 0.017 @xmath73 0.040 & 0.0011 @xmath73 0.0027 & 0.036 & 7.67 @xmath73 0.03 & 0.100 @xmath73 0.080 & 0.0068 @xmath73 0.0054 & 0.072 + ngc 2442 & 8.64 @xmath73 0.05 & -0.079 @xmath73 0.072 & -0.0058 @xmath73 0.0052 & 0.030 & 7.87 @xmath73 0.14 & -0.119 @xmath73 0.204 & -0.0087 @xmath73 0.0150 & 0.087 + ngc 2541 & 8.45 @xmath73 0.02 & -0.608 @xmath73 0.053 & -0.0526 @xmath73 0.0046 & 0.030 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 2805 & 8.51 @xmath73 0.03 & -0.432 @xmath73 0.045 & -0.0168 @xmath73 0.0017 & 0.040 & 7.53 @xmath73 0.06 & -0.869 @xmath73 0.096 & -0.0338 @xmath73 0.0037 & 0.087 + ngc 2835 & 8.50 @xmath73 0.04 & -0.314 @xmath73 0.060 & -0.0324 @xmath73 0.0062 & 0.057 & 7.51 @xmath73 0.10 & -0.633 @xmath73 0.149 & -0.0652 @xmath73 0.0154 & 0.142 + ngc 2841 & 8.99 @xmath73 0.08 & -0.607 @xmath73 0.172 & -0.0363 @xmath73 0.0103 & 0.022 & 8.81 @xmath73 0.20 & -1.488 @xmath73 0.450 & -0.0890 @xmath73 0.0269 & 0.052 + ngc 2903 & 8.82 @xmath73 0.03 & -0.523 @xmath73 0.072 & -0.0321 @xmath73 0.0044 & 0.061 & 8.51 @xmath73 0.06 & -1.588 @xmath73 0.152 & -0.0975 @xmath73 0.0093 & 0.121 + ngc 2997 & 8.80 @xmath73 0.06 & -0.502 @xmath73 0.121 & -0.0318 @xmath73 0.0076 & 0.060 & 8.44 @xmath73 0.13 & -1.503 @xmath73 0.274 & -0.0950 @xmath73 0.0173 & 0.129 + ngc 3020 & 8.39 @xmath73 0.02 & -0.558 @xmath73 0.041 & -0.0554 @xmath73 0.0041 & 0.011 & 7.27 @xmath73 0.07 & -1.053 @xmath73 0.167 & -0.1045 @xmath73 0.0167 & 0.045 + ngc 3023 & 8.34 @xmath73 0.08 & -0.444 @xmath73 0.201 & -0.0361 @xmath73 0.0163 & 0.054 & 7.20 @xmath73 0.14 & -0.889 @xmath73 0.372 & -0.0722 @xmath73 0.0302 & 0.100 + ngc 3031 & 8.58 @xmath73 0.02 & -0.156 @xmath73 0.045 & -0.0110 @xmath73 0.0032 & 0.043 & 7.84 @xmath73 0.06 & -0.464 @xmath73 0.116 & -0.0327 @xmath73 0.0081 & 0.110 + ngc 3184 & 8.66 @xmath73 0.02 & -0.176 @xmath73 0.037 & -0.0130 @xmath73 0.0027 & 0.066 & 7.86 @xmath73 0.04 & -0.338 @xmath73 0.086 & -0.0249 @xmath73 0.0063 & 0.137 + ngc 3198 & 8.60 @xmath73 0.04 & -0.425 @xmath73 0.076 & -0.0237 @xmath73 0.0042 & 0.063 & 7.94 @xmath73 0.11 & -1.145 @xmath73 0.317 & -0.0638 @xmath73 0.0176 & 0.091 + ngc 3227 & 8.64 @xmath73 0.05 & -0.254 @xmath73 0.077 & -0.0158 @xmath73 0.0048 & 0.056 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 3239 & 8.12 @xmath73 0.01 & -0.153 @xmath73 0.012 & -0.0260 @xmath73 0.0021 & 0.005 & 6.78 @xmath73 0.04 & -0.275 @xmath73 0.064 & -0.0467 @xmath73 0.0108 & 0.026 + ngc 3310 & 8.37 @xmath73 0.01 & -0.101 @xmath73 0.023 & -0.0117 @xmath73 0.0027 & 0.040 & 7.27 @xmath73 0.02 & -0.254 @xmath73 0.060 & -0.0295 @xmath73 0.0070 & 0.104 + ngc 3319 & 8.50 @xmath73 0.02 & -0.516 @xmath73 0.043 & -0.0408 @xmath73 0.0034 & 0.024 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 3344 & 8.72 @xmath73 0.02 & -0.618 @xmath73 0.039 & -0.0984 @xmath73 0.0063 & 0.029 & 8.08 @xmath73 0.17 & -1.413 @xmath73 0.252 & -0.2254 @xmath73 0.0407 & 0.044 + ngc 3351 & 8.82 @xmath73 0.01 & -0.210 @xmath73 0.031 & -0.0186 @xmath73 0.0027 & 0.038 & 8.51 @xmath73 0.03 & -0.678 @xmath73 0.094 & -0.0599 @xmath73 0.0083 & 0.116 + ngc 3359 & 8.40 @xmath73 0.06 & -0.271 @xmath73 0.108 & -0.0179 @xmath73 0.0071 & 0.099 & 7.43 @xmath73 0.09 & -0.688 @xmath73 0.154 & -0.0453 @xmath73 0.0102 & 0.130 + ngc 3486 & 8.60 @xmath73 0.02 & -0.421 @xmath73 0.043 & -0.0447 @xmath73 0.0046 & 0.018 & 7.80 @xmath73 0.04 & -0.928 @xmath73 0.088 & -0.0985 @xmath73 0.0094 & 0.036 + ngc 3521 & 8.83 @xmath73 0.04 & -0.666 @xmath73 0.096 & -0.0390 @xmath73 0.0056 & 0.048 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 3621 & 8.68 @xmath73 0.02 & -0.460 @xmath73 0.035 & -0.0446 @xmath73 0.0034 & 0.050 & 8.05 @xmath73 0.05 & -1.298 @xmath73 0.095 & -0.1260 @xmath73 0.0092 & 0.133 + ngc 3631 & 8.71 @xmath73 0.03 & -0.414 @xmath73 0.077 & -0.0256 @xmath73 0.0048 & 0.024 & 7.94 @xmath73 0.05 & -0.719 @xmath73 0.145 & -0.0445 @xmath73 0.0090 & 0.046 + ngc 3718 & 8.60 @xmath73 0.19 & -0.221 @xmath73 0.253 & -0.0119 @xmath73 0.0136 & 0.066 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 3820 & 8.64 @xmath73 0.01 & -0.102 @xmath73 0.019 & -0.0116 @xmath73 0.0022 & 0.019 & 7.92 @xmath73 0.05 & -0.342 @xmath73 0.093 & -0.0389 @xmath73 0.0105 & 0.092 + ngc 3893 & 8.73 @xmath73 0.16 & -0.437 @xmath73 0.254 & -0.0371 @xmath73 0.0215 & 0.028 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 3938 & 8.79 @xmath73 0.05 & -0.577 @xmath73 0.092 & -0.0413 @xmath73 0.0066 & 0.031 & 8.15 @xmath73 0.07 & -1.204 @xmath73 0.141 & -0.0861 @xmath73 0.0101 & 0.048 + ngc 4030 & 8.77 @xmath73 0.02 & -0.258 @xmath73 0.036 & -0.0156 @xmath73 0.0022 & 0.010 & 8.05 @xmath73 0.08 & -0.344 @xmath73 0.177 & -0.0208 @xmath73 0.0107 & 0.052 + ngc 4088 & 8.71 @xmath73 0.04 & -0.366 @xmath73 0.093 & -0.0305 @xmath73 0.0078 & 0.025 & 7.97 @xmath73 0.06 & -0.698 @xmath73 0.116 & -0.0582 @xmath73 0.0097 & 0.032 + ngc 4109 & 8.63 @xmath73 0.03 & -0.402 @xmath73 0.099 & -0.0269 @xmath73 0.0066 & 0.056 & 7.87 @xmath73 0.05 & -1.152 @xmath73 0.154 & -0.0770 @xmath73 0.0103 & 0.081 + ngc 4254 & 8.77 @xmath73 0.02 & -0.353 @xmath73 0.034 & -0.0316 @xmath73 0.0031 & 0.036 & 8.09 @xmath73 0.05 & -0.716 @xmath73 0.100 & -0.0641 @xmath73 0.0090 & 0.106 + ngc 4258 & 8.54 @xmath73 0.03 & -0.161 @xmath73 0.073 & -0.0083 @xmath73 0.0037 & 0.031 & 7.59 @xmath73 0.13 & -0.232 @xmath73 0.282 & -0.0119 @xmath73 0.0144 & 0.120 + ngc 4303 & 8.78 @xmath73 0.03 & -0.577 @xmath73 0.073 & -0.0452 @xmath73 0.0057 & 0.062 & 8.24 @xmath73 0.08 & -1.370 @xmath73 0.189 & -0.1073 @xmath73 0.0148 & 0.156 + ngc 4321 & 8.74 @xmath73 0.03 & -0.219 @xmath73 0.048 & -0.0118 @xmath73 0.0026 & 0.038 & 8.04 @xmath73 0.06 & -0.355 @xmath73 0.104 & -0.0192 @xmath73 0.0056 & 0.083 + ngc 4395 & 8.19 @xmath73 0.05 & -0.164 @xmath73 0.082 & -0.0200 @xmath73 0.0100 & 0.079 & 6.89 @xmath73 0.08 & -0.346 @xmath73 0.143 & -0.0420 @xmath73 0.0174 & 0.136 + ngc 4490 & 8.29 @xmath73 0.08 & -0.318 @xmath73 0.254 & -0.0444 @xmath73 0.0355 & 0.058 & 7.09 @xmath73 0.15 & -0.560 @xmath73 0.449 & -0.0781 @xmath73 0.0629 & 0.102 + ngc 4501 & 8.92 @xmath73 0.12 & -0.424 @xmath73 0.242 & -0.0299 @xmath73 0.0171 & 0.054 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 4535 & 8.71 @xmath73 0.03 & -0.299 @xmath73 0.064 & -0.0163 @xmath73 0.0035 & 0.025 & 7.97 @xmath73 0.03 & -0.594 @xmath73 0.055 & -0.0324 @xmath73 0.0030 & 0.022 + ngc 4559 & 8.53 @xmath73 0.03 & -0.573 @xmath73 0.056 & -0.0467 @xmath73 0.0046 & 0.047 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 4625 & 8.58 @xmath73 0.01 & -0.130 @xmath73 0.044 & -0.0443 @xmath73 0.0151 & 0.042 & 7.84 @xmath73 0.05 & -0.680 @xmath73 0.143 & -0.2318 @xmath73 0.0488 & 0.127 + ngc 4631 & 8.39 @xmath73 0.06 & -0.333 @xmath73 0.125 & -0.0194 @xmath73 0.0073 & 0.077 & 7.18 @xmath73 0.09 & -0.480 @xmath73 0.169 & -0.0280 @xmath73 0.0098 & 0.093 + ngc 4651 & 8.68 @xmath73 0.04 & -0.469 @xmath73 0.096 & -0.0343 @xmath73 0.0070 & 0.049 & 8.02 @xmath73 0.10 & -1.133 @xmath73 0.233 & -0.0829 @xmath73 0.0170 & 0.119 + ngc 4654 & 8.66 @xmath73 0.04 & -0.364 @xmath73 0.083 & -0.0318 @xmath73 0.0073 & 0.044 & 7.90 @xmath73 0.11 & -0.797 @xmath73 0.229 & -0.0695 @xmath73 0.0200 & 0.120 + ngc 4656 & 8.06 @xmath73 0.16 & -0.551 @xmath73 0.280 & -0.0305 @xmath73 0.0155 & 0.084 & 6.62 @xmath73 0.24 & -0.634 @xmath73 0.409 & -0.0351 @xmath73 0.0227 & 0.123 + ngc 4713 & 8.53 @xmath73 0.09 & -0.771 @xmath73 0.250 & -0.1325 @xmath73 0.0428 & 0.095 & 7.60 @xmath73 0.14 & -1.548 @xmath73 0.397 & -0.2662 @xmath73 0.0679 & 0.150 + ngc 4725 & 8.83 @xmath73 0.37 & -0.472 @xmath73 0.853 & -0.0223 @xmath73 0.0404 & 0.070 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 4736 & 8.57 @xmath73 0.03 & -0.066 @xmath73 0.170 & -0.0086 @xmath73 0.0224 & 0.065 & 0.00 @xmath73 0.00 & 0.000 @xmath73 0.000 & 0.0000 @xmath73 0.0000 & 0.000 + ngc 4861 & 8.00 @xmath73 0.06 & -0.013 @xmath73 0.092 & -0.0030 @xmath73 0.0210 & 0.090 & 6.54 @xmath73 0.13 & -0.058 @xmath73 0.200 & -0.0132 @xmath73 0.0454 & 0.193 + ngc 5033 & 8.64 @xmath73 0.06 & -0.278 @xmath73 0.094 & -0.0095 @xmath73 0.0032 & 0.090 & 8.00 @xmath73 0.10 & -0.599 @xmath73 0.130 & -0.0206 @xmath73 0.0045 & 0.072 + ngc 5055 & 8.87 @xmath73 0.02 & -0.490 @xmath73 0.031 & -0.0337 @xmath73 0.0021 & 0.007 & 8.42 @xmath73 0.10 & -1.259 @xmath73 0.186 & -0.0867 @xmath73 0.0129 & 0.044 + ngc 5068 & 8.57 @xmath73 0.04 & -0.355 @xmath73 0.098 & -0.0626 @xmath73 0.0173 & 0.050 & 7.53 @xmath73 0.11 & -0.451 @xmath73 0.263 & -0.0794 @xmath73 0.0463 & 0.134 + ngc 5194 & 8.88 @xmath73 0.03 & -0.324 @xmath73 0.053 & -0.0223 @xmath73 0.0037 & 0.062 & 8.52 @xmath73 0.05 & -0.704 @xmath73 0.102 & -0.0484 @xmath73 0.0070 & 0.117 + ngc 5236 & 8.78 @xmath73 0.01 & -0.221 @xmath73 0.028 & -0.0256 @xmath73 0.0032 & 0.050 & 8.34 @xmath73 0.03 & -0.649 @xmath73 0.058 & -0.0752 @xmath73 0.0068 & 0.104 + ngc 5248 & 8.64 @xmath73 0.03 & -0.142 @xmath73 0.076 & -0.0069 @xmath73 0.0037 & 0.072 & 7.74 @xmath73 0.05 & -0.132 @xmath73 0.155 & -0.0064 @xmath73 0.0075 & 0.146 + ngc 5457 & 8.71 @xmath73 0.01 & -0.840 @xmath73 0.029 & -0.0293 @xmath73 0.0010 & 0.076 & 8.01 @xmath73 0.03 & -1.760 @xmath73 0.056 & -0.0613 @xmath73 0.0020 & 0.142 + ngc 5474 & 8.19 @xmath73 0.02 & -0.008 @xmath73 0.086 & -0.0019 @xmath73 0.0182 & 0.100 & 6.82 @xmath73 0.03 & -0.261 @xmath73 0.144 & -0.0553 @xmath73 0.0303 & 0.156 + ngc 5668 & 8.52 @xmath73 0.03 & -0.471 @xmath73 0.061 & -0.0364 @xmath73 0.0047 & 0.075 & 7.48 @xmath73 0.05 & -1.032 @xmath73 0.127 & -0.0796 @xmath73 0.0098 & 0.147 + ngc 6384 & 8.87 @xmath73 0.04 & -0.531 @xmath73 0.068 & -0.0228 @xmath73 0.0029 & 0.029 & 8.41 @xmath73 0.12 & -1.172 @xmath73 0.197 & -0.0504 @xmath73 0.0084 & 0.083 + ngc 6691 & 8.66 @xmath73 0.02 & -0.152 @xmath73 0.052 & -0.0077 @xmath73 0.0026 & 0.050 & 7.98 @xmath73 0.06 & -0.414 @xmath73 0.157 & -0.0211 @xmath73 0.0080 & 0.142 + ngc 6744 & 8.88 @xmath73 0.03 & -0.726 @xmath73 0.060 & -0.0266 @xmath73 0.0022 & 0.034 & 8.66 @xmath73 0.07 & -1.996 @xmath73 0.157 & -0.0731 @xmath73 0.0057 & 0.087 + ngc 6946 & 8.72 @xmath73 0.06 & -0.337 @xmath73 0.099 & -0.0337 @xmath73 0.0099 & 0.061 & 8.05 @xmath73 0.15 & -0.751 @xmath73 0.237 & -0.0751 @xmath73 0.0237 & 0.148 + ngc 7331 & 8.67 @xmath73 0.08 & -0.222 @xmath73 0.193 & -0.0097 @xmath73 0.0084 & 0.044 & 7.75 @xmath73 0.09 & -0.056 @xmath73 0.213 & -0.0024 @xmath73 0.0092 & 0.048 + ngc 7495 & 8.61 @xmath73 0.01 & -0.260 @xmath73 0.027 & -0.0140 @xmath73 0.0015 & 0.028 & 7.83 @xmath73 0.02 & -0.616 @xmath73 0.062 & -0.0333 @xmath73 0.0033 & 0.064 + ngc 7518 & 8.74 @xmath73 0.02 & -0.127 @xmath73 0.045 & -0.0122 @xmath73 0.0043 & 0.034 & 8.30 @xmath73 0.07 & -0.514 @xmath73 0.136 & -0.0496 @xmath73 0.0131 & 0.104 + ngc 7529 & 8.64 @xmath73 0.04 & -0.232 @xmath73 0.067 & -0.0303 @xmath73 0.0087 & 0.040 & 7.97 @xmath73 0.13 & -0.752 @xmath73 0.223 & -0.0979 @xmath73 0.0290 & 0.133 + ngc 7570 & 8.58 @xmath73 0.03 & -0.086 @xmath73 0.045 & -0.0217 @xmath73 0.0114 & 0.029 & 7.78 @xmath73 0.06 & -0.219 @xmath73 0.106 & -0.0549 @xmath73 0.0265 & 0.068 + ngc 7591 & 8.71 @xmath73 0.01 & -0.146 @xmath73 0.029 & -0.0085 @xmath73 0.0017 & 0.024 & 8.13 @xmath73 0.06 & -0.400 @xmath73 0.114 & -0.0234 @xmath73 0.0066 & 0.093 + ngc 7678 & 8.55 @xmath73 0.04 & -0.079 @xmath73 0.078 & -0.0048 @xmath73 0.0047 & 0.030 & 7.64 @xmath73 0.14 & -0.081 @xmath73 0.248 & -0.0049 @xmath73 0.0149 & 0.094 + ngc 7793 & 8.50 @xmath73 0.02 & -0.305 @xmath73 0.048 & -0.0662 @xmath73 0.0104 & 0.071 & 7.52 @xmath73 0.05 & -0.752 @xmath73 0.095 & -0.1631 @xmath73 0.0207 & 0.135 + ic 0010 & 8.14 @xmath73 0.08 & 0.252 @xmath73 0.185 & 0.4294 @xmath73 0.3025 & 0.094 & 6.90 @xmath73 0.15 & 0.355 @xmath73 0.338 & 0.6035 @xmath73 0.5534 & 0.165 + ic 0193 & 8.59 @xmath73 0.02 & -0.139 @xmath73 0.064 & -0.0083 @xmath73 0.0039 & 0.039 & 7.80 @xmath73 0.05 & -0.306 @xmath73 0.151 & -0.0183 @xmath73 0.0090 & 0.092 + ic 0208 & 8.58 @xmath73 0.01 & -0.168 @xmath73 0.035 & -0.0165 @xmath73 0.0034 & 0.036 & 7.78 @xmath73 0.03 & -0.422 @xmath73 0.083 & -0.0416 @xmath73 0.0082 & 0.086 + ic 0342 & 8.83 @xmath73 0.04 & -0.513 @xmath73 0.087 & -0.0500 @xmath73 0.0085 & 0.045 & 8.46 @xmath73 0.07 & -1.465 @xmath73 0.131 & -0.1426 @xmath73 0.0128 & 0.068 + ic 1132 & 8.65 @xmath73 0.03 & -0.390 @xmath73 0.059 & -0.0323 @xmath73 0.0049 & 0.077 & 7.69 @xmath73 0.06 & -0.598 @xmath73 0.106 & -0.0495 @xmath73 0.0087 & 0.128 + ic 2204 & 8.62 @xmath73 0.01 & -0.160 @xmath73 0.033 & -0.0159 @xmath73 0.0033 & 0.024 & 7.89 @xmath73 0.03 & -0.387 @xmath73 0.086 & -0.0384 @xmath73 0.0086 & 0.064 + ic 5201 & 8.35 @xmath73 0.07 & -0.501 @xmath73 0.147 & -0.0316 @xmath73 0.0093 & 0.074 & 7.19 @xmath73 0.12 & -0.908 @xmath73 0.242 & -0.0573 @xmath73 0.0153 & 0.122 + ic 5309 & 8.70 @xmath73 0.03 & -0.058 @xmath73 0.066 & -0.0059 @xmath73 0.0067 & 0.059 & 8.08 @xmath73 0.09 & -0.211 @xmath73 0.206 & -0.0215 @xmath73 0.0210 & 0.170 + ugc 00223 & 8.53 @xmath73 0.02 & -0.099 @xmath73 0.053 & -0.0105 @xmath73 0.0057 & 0.029 & 7.72 @xmath73 0.06 & -0.295 @xmath73 0.152 & -0.0313 @xmath73 0.0162 & 0.084 + ugc 01087 & 8.60 @xmath73 0.01 & -0.246 @xmath73 0.023 & -0.0192 @xmath73 0.0018 & 0.026 & 7.83 @xmath73 0.03 & -0.609 @xmath73 0.058 & -0.0475 @xmath73 0.0045 & 0.066 + ugc 01862 & 8.43 @xmath73 0.02 & -0.094 @xmath73 0.039 & -0.0194 @xmath73 0.0081 & 0.040 & 7.42 @xmath73 0.04 & -0.316 @xmath73 0.090 & -0.0651 @xmath73 0.0186 & 0.092 + ugc 02023 & 8.13 @xmath73 0.19 & -0.088 @xmath73 0.280 & -0.0358 @xmath73 0.1128 & 0.075 & 6.84 @xmath73 0.29 & -0.197 @xmath73 0.434 & -0.0800 @xmath73 0.1752 & 0.115 + ugc 02216 & 8.08 @xmath73 0.02 & -0.068 @xmath73 0.052 & -0.0099 @xmath73 0.0077 & 0.024 & 6.72 @xmath73 0.04 & -0.124 @xmath73 0.094 & -0.0181 @xmath73 0.0137 & 0.041 + ugc 02345 & 8.04 @xmath73 0.03 & -0.893 @xmath73 0.085 & -0.0885 @xmath73 0.0084 & 0.041 & 6.63 @xmath73 0.11 & -0.846 @xmath73 0.255 & -0.0838 @xmath73 0.0253 & 0.131 + ugc 03701 & 8.41 @xmath73 0.01 & -0.087 @xmath73 0.046 & -0.0076 @xmath73 0.0040 & 0.042 & 7.41 @xmath73 0.03 & -0.329 @xmath73 0.102 & -0.0288 @xmath73 0.0089 & 0.092 + ugc 04107 & 8.56 @xmath73 0.01 & -0.231 @xmath73 0.041 & -0.0219 @xmath73 0.0039 & 0.045 & 7.72 @xmath73 0.03 & -0.498 @xmath73 0.087 & -0.0470 @xmath73 0.0083 & 0.096 + ugc 04305 & 7.89 @xmath73 0.07 & -0.225 @xmath73 0.137 & -0.0573 @xmath73 0.0352 & 0.075 & 6.41 @xmath73 0.13 & -0.307 @xmath73 0.246 & -0.0784 @xmath73 0.0629 & 0.139 + ugc 05100 & 8.58 @xmath73 0.05 & -0.078 @xmath73 0.075 & -0.0056 @xmath73 0.0053 & 0.047 & 7.90 @xmath73 0.13 & -0.220 @xmath73 0.204 & -0.0156 @xmath73 0.0145 & 0.119 + ugc 06410 & 8.58 @xmath73 0.02 & -0.148 @xmath73 0.040 & -0.0113 @xmath73 0.0031 & 0.057 & 7.78 @xmath73 0.05 & -0.454 @xmath73 0.093 & -0.0345 @xmath73 0.0071 & 0.123 + ugc 08091 & 7.58 @xmath73 0.04 & 0.177 @xmath73 0.067 & 0.5487 @xmath73 0.2003 & 0.055 & 6.03 @xmath73 0.08 & 0.255 @xmath73 0.129 & 0.7891 @xmath73 0.3905 & 0.108 + ugc 09562 & 8.22 @xmath73 0.05 & -0.193 @xmath73 0.143 & -0.0585 @xmath73 0.0433 & 0.080 & 6.80 @xmath73 0.06 & -0.206 @xmath73 0.144 & -0.0626 @xmath73 0.0437 & 0.068 + ugc 09837 & 8.45 @xmath73 0.02 & -0.412 @xmath73 0.076 & -0.0370 @xmath73 0.0068 & 0.078 & 7.32 @xmath73 0.04 & -0.820 @xmath73 0.140 & -0.0738 @xmath73 0.0126 & 0.141 + ugc 10445 & 8.27 @xmath73 0.04 & -0.082 @xmath73 0.077 & -0.0103 @xmath73 0.0096 & 0.050 & 7.03 @xmath73 0.08 & -0.197 @xmath73 0.141 & -0.0246 @xmath73 0.0176 & 0.091 + ugc 12709 & 8.47 @xmath73 0.08 & -0.706 @xmath73 0.218 & -0.0434 @xmath73 0.0134 & 0.082 & 7.30 @xmath73 0.15 & -1.089 @xmath73 0.418 & -0.0669 @xmath73 0.0257 & 0.157 + pgc 29167 & 8.34 @xmath73 0.10 & 0.152 @xmath73 0.169 & 0.0507 @xmath73 0.0565 & 0.040 & 7.25 @xmath73 0.01 & 0.005 @xmath73 0.020 & 0.0017 @xmath73 0.0065 & 0.005 + pgc 41318 & 8.45 @xmath73 0.04 & -0.168 @xmath73 0.143 & -0.0281 @xmath73 0.0239 & 0.033 & 7.28 @xmath73 0.17 & -0.563 @xmath73 0.568 & -0.0941 @xmath73 0.0945 & 0.130 + pgc 44772 & 8.59 @xmath73 0.03 & -0.268 @xmath73 0.045 & -0.0216 @xmath73 0.0036 & 0.043 & 7.79 @xmath73 0.07 & -0.705 @xmath73 0.115 & -0.0569 @xmath73 0.0093 & 0.109 + pgc 45195 & 8.14 @xmath73 0.04 & -0.152 @xmath73 0.059 & -0.0122 @xmath73 0.0047 & 0.035 & 6.81 @xmath73 0.08 & -0.277 @xmath73 0.104 & -0.0223 @xmath73 0.0084 & 0.061 + pgc 46182 & 8.59 @xmath73 0.03 & -0.126 @xmath73 0.078 & -0.0089 @xmath73 0.0055 & 0.035 & 7.92 @xmath73 0.11 & -0.462 @xmath73 0.313 & -0.0327 @xmath73 0.0221 & 0.141 + ll ngc 0012 & @xcite + ngc 0055 & @xcite , @xcite , @xcite + ngc 0099 & @xcite , sdss + ngc 0224 ( m 31 ) & @xcite , @xcite , @xcite , @xcite , + & @xcite , @xcite , @xcite + ngc 0234 & @xcite , sdss + ngc 0253 & @xcite + ngc 0300 & @xcite , @xcite , @xcite , @xcite + ngc 0428 & sdss + ngc 0450 & @xcite , sdss + ngc 0493 & sdss + ngc 0575 & @xcite + ngc 0598 ( m 33 ) & @xcite , @xcite , @xcite , @xcite , @xcite , + & @xcite , @xcite , @xcite , @xcite , + & @xcite , @xcite , @xcite + ngc 0628 ( m 74 ) & @xcite , @xcite , @xcite , @xcite , + & @xcite , @xcite + ngc 0753 & @xcite + ngc 0783 & @xcite + ngc 0925 & @xcite + ngc 1055 & sdss + ngc 1058 & @xcite , @xcite + ngc 1068 ( m 77 ) & @xcite , @xcite , @xcite , sdss + ngc 1090 & sdss + ngc 1097 & @xcite , @xcite + ngc 1232 & @xcite , @xcite + ngc 1313 & @xcite , @xcite , @xcite + ngc 1365 & @xcite , @xcite , @xcite , @xcite + ngc 1512 & @xcite + ngc 1598 & @xcite + ngc 1637 & @xcite , @xcite + ngc 1642 & @xcite + ngc 1672 & @xcite + ngc 2336 & @xcite + ngc 2403 & @xcite , @xcite , @xcite , @xcite , @xcite , + & @xcite , @xcite , @xcite , sdss + ngc 2441 & @xcite + ngc 2442 & @xcite + ngc 2541 & @xcite , sdss + ngc 2805 & @xcite , sdss + ngc 2835 & @xcite + ngc 2841 & @xcite , @xcite + ngc 2903 & @xcite , @xcite , @xcite , sdss + ngc 2997 & @xcite , @xcite , @xcite + ngc 3020 & sdss + ngc 3023 & sdss + ngc 3031 ( m 81 ) & @xcite , @xcite , @xcite , @xcite , + & @xcite , sdss + ngc 3184 & @xcite , @xcite , @xcite , sdss + ngc 3198 & @xcite , sdss + ngc 3227 & @xcite , @xcite , @xcite + ngc 3239 & sdss + ngc 3310 & @xcite , @xcite + ngc 3319 & @xcite , sdss + ngc 3344 & @xcite , @xcite , @xcite , sdss + ngc 3351 ( m 95 ) & @xcite , @xcite , @xcite , @xcite , sdss + ngc 3359 & @xcite , sdss + ngc 3486 & sdss + ngc 3521 & @xcite , @xcite + ngc 3621 & @xcite , @xcite + ngc 3631 & sdss + ngc 3718 & @xcite + ngc 3820 & @xcite , sdss + ngc 3893 & @xcite + ngc 3938 & sdss + ngc 4030 & sdss + ngc 4088 & sdss + ngc 4109 & @xcite + ngc 4254 ( m 99 ) & @xcite , @xcite , @xcite , sdss + ngc 4258 ( m 106 ) & @xcite , @xcite , sdss + ngc 4303 ( m 61 ) & @xcite , @xcite , sdss + ngc 4321 ( m 100 ) & @xcite , @xcite , sdss + ngc 4395 & @xcite , @xcite , @xcite , sdss + ngc 4490 & sdss + ngc 4501 & @xcite + ngc 4535 & sdss + ngc 4559 & @xcite , sdss + ngc 4625 & @xcite , @xcite , sdss + ngc 4631 & @xcite , sdss + ngc 4651 & @xcite , sdss + ngc 4654 & @xcite , sdss + ngc 4656 & sdss + ngc 4713 & @xcite , sdss + ngc 4725 & @xcite + ngc 4736 ( m 94 ) & @xcite , @xcite , @xcite + ngc 4861 & @xcite , @xcite , @xcite , @xcite , + & @xcite , @xcite , sdss + ngc 5033 & @xcite , sdss + ngc 5055 ( m 63 ) & @xcite + ngc 5068 & @xcite + ngc 5194 ( m 51 ) & @xcite , @xcite , @xcite , @xcite , sdss , + & @xcite + ngc 5236 ( m 83 ) & @xcite , @xcite , @xcite , @xcite , + & @xcite , @xcite + ngc 5248 & @xcite , sdss + ngc 5457 ( m 101 ) & @xcite , @xcite , @xcite , @xcite , @xcite , + & @xcite , @xcite , @xcite , @xcite , + & @xcite , sdss , @xcite , @xcite , @xcite , + & @xcite , @xcite , @xcite + ngc 5474 & @xcite , sdss + ngc 5668 & @xcite , sdss + ngc 6384 & @xcite , @xcite + ngc 6691 & @xcite + ngc 6744 & @xcite + ngc 6946 & @xcite , @xcite , @xcite + ngc 7331 & @xcite , @xcite + ngc 7495 & @xcite + ngc 7518 & @xcite + ngc 7529 & @xcite + ngc 7570 & @xcite + ngc 7591 & @xcite + ngc 7678 & @xcite + ngc 7793 & @xcite , @xcite , @xcite , @xcite + ic 0010 & @xcite , @xcite , @xcite + ic 0193 & @xcite + ic 0208 & @xcite + ic 0342 & @xcite + ic 1132 & @xcite + ic 2204 & @xcite , sdss + ic 5201 & @xcite + ic 5309 & @xcite + ugc 00223 & @xcite , sdss + ugc 01087 & @xcite + ugc 01862 & @xcite + ugc 02023 & @xcite + ugc 02216 & sdss + ugc 02345 & sdss + ugc 03701 & @xcite + ugc 04107 & @xcite , sdss + ugc 04305 & @xcite + ugc 05100 & @xcite + ugc 06410 & @xcite , sdss + ugc 08091 ( gr 8) & @xcite , @xcite , sdss + ugc 09562 & @xcite , sdss + ugc 09837 & @xcite , sdss + ugc 10445 & @xcite , sdss + ugc 12709 & sdss + pgc 029167 & @xcite + pgc 041318 & @xcite + pgc 044772 & @xcite + pgc 045195 & sdss + pgc 046182 & sdss , @xcite + Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: this year marks the @xmath0 anniversary of two seminal papers by slonczewski @xcite and berger @xcite on the spin transfer torque ( stt ) phenomenon . this refers to the transfer of angular momentum from a spin - polarized charge current , or a pure spin current , to a local magnetization . while berger had already reported on stt - driven domain wall motion using 45 a of current in millimeter - wide films in 1985 @xcite , based on his predictions from the late 1970s @xcite , and slonczewski had developed the theory for spin - current - driven exchange coupling in magnetic tunnel junctions in 1989 @xcite , the two papers in 1996 predicted that a current flowing perpendicularly to the plane of a magnetic metal multilayer should be able to reorient the magnetization direction in one of the layers . these papers were published at a time when magnetic thin film technology had just become capable of beginning to implement these concepts in earnest , with giant magnetoresistance ( gmr ) @xcite multilayers being prepared for hard drive read - head production and the first high - quality magnetic tunnel junctions having been reported the year before @xcite . it nevertheless took a few more years before the first experimental demonstration of stt - driven magnetization precession was reported @xcite , followed by reports of stt - driven magnetic switching @xcite . in 1999 , hirsch revisited @xcite a different phenomenon , the so - called spin hall effect ( she ) , first described by dyakonov and perel in 1971 @xcite . the she can produce a pure spin current in a direction perpendicular to a charge current @xcite , which can in turn exert substantial stt on an adjacent magnetic layer . it is thus possible to use the she in a nonmagnetic metal to achieve , in principle , all the stt - related functions such as magnetic switching @xcite and driven magnetization precession @xcite that previously required a separate magnetic layer to spin - polarize the charge current . from a technological standpoint , the two most important consequences of stt are _ i _ ) the possibility of programming a magnetic memory , such as magnetoresistive random - access memory ( mram ) @xcite , using a small direct current through the bit , rather than larger currents through adjacent programming lines , and _ ii _ ) the possibility of creating nanoscopic and ultratunable microwave signal generators without the need for any semiconductor materials . our paper deals with the latter technology and reviews the fundamentals and state of the art of so - called spin transfer torque and spin hall effect nano - oscillators ( stnos and shnos ) . for more in - depth reviews of stt - based mram and stnos , see @xcite and @xcite . the term _ spintronics _ is often used to refer to electronics in which both the charge and the spin of the carriers in a metal or semiconductor are utilized ( for reviews of spintronics , see @xcite ) . however , this excludes the very active research currently being undertaken in magnetic insulators ; these can not transfer any charge current , but are excellent conductors of spin currents @xcite . a broader , more general and , we would argue , better definition of modern spintronics is any device and technology in which _ spin currents _ @xcite , with or without additional charge or heat currents , are directly employed and utilized for function . as spin is a vector in three - dimensional spin space , and the flow of the spin carriers takes place in three - dimensional real space , the resulting spin current is a @xmath1 tensor ( for illustrative purposes , the spin current may instead be pictured as a flow , in real space , of tiny magnetic moments , also in real space ) . the tensorial nature of the spin current where the direction of the flowing spins is , in the general case , independent of the real - space direction of the spin current provide for very rich physics and a tremendous potential for new device functionality . in the original formulation of stt , the spin current was assumed to be carried by a spin - polarized charge current , and a number of simplifying assumptions ( e.g. , device geometry and uniformity of magnetization ) were made such that the flow direction of the charge and spin currents were identical . most analysis and modelling of stnos still relies on such assumptions , which tend to be highly accurate in most cases . in this case , the spin - polarized current ( and hence the spin current ) arises directly from the spin polarization of the electrons at the fermi surface of the ferromagnetic metal . when an unpolarized current , say in cu , enters a ferromagnetic layer , it becomes spin - polarized in the so - called spin accumulation zone , which extends both into the cu and the ferromagnet in proportion to the spin diffusion lengths in each material @xcite . as a result , the current acquires spin polarization well inside the cu , reaches about half its full spin polarization at the nm / fm interface , and then , over a few nanometers in the fm , rapidly reaches the spin polarization of the bulk of the ferromagnet . the same processes govern how the spin polarization of the current decays as the current leaves the ferromagnet and again enters a nonmagnetic metal . as the spin diffusion length is short in ferromagnets , but can be several hundred nanometers even at room temperature in most ordinary metals , thin magnetic films are highly effective spin filters and the resulting spin - polarized currents can act over quite long distances . while spin filtering is a straightforward technique for generating spin - polarized currents , spin currents do not necessarily require any charge current to flow . in the case of the she , the charge and spin currents flow in perpendicular directions with respect to each other that is , the charge current is unpolarized and the spin current is free of net charge . while we do not discuss this further here , pure spin currents can also be created through magnetodynamic means , as in so - called spin batteries @xcite . the she can be viewed as a limiting case of the anomalous hall effect ( ahe ) @xcite , which was proposed by edwin hall in 1881 when he applied his earlier discovery of the ordinary hall effect ( he ) @xcite to ferromagnetic metals . as is well known , the he in nonmagnetic metals or semiconductors refers to the build - up of a so - called hall voltage across an electrical conductor , transverse both to the electric current in the conductor and to an external magnetic field perpendicular to the current . it is widely used for material characterization and for sensing magnetic fields . in ferromagnets , however , hall found that this effect could be ten times greater than in nonmagnetic metals . whereas the he arises from the magnetic field s lorentz force on the charge carriers , the ahe is a consequence of asymmetric spin - dependent scattering of the charge carriers , such that a much larger hall voltage can build up . the she is then a limiting case of this phenomenon in materials without any spontaneous magnetization but with substantial spin - orbit coupling . as there is no spin imbalance in a nonmagnetic material , the asymmetric spin - dependent scattering does not result in any hall voltage . what it does result in , however , is a net spin current transverse to the charge current and , in a steady state , the build - up of spin accumulation zones at the edges of the conductor , which can be directly observed by optical means @xcite and which is used in spin - current - driven devices . equipped with two different techniques for generating spin currents , we now briefly explain how spin currents can generate stt and control the magnetic states and the magnetodynamics of nanomagnets and magnetic thin films ( for an in - depth review of stt , see @xcite ) . since each electron carries angular momentum , a spin current can be thought of as a flow of angular momentum . when a spin - polarized current enters ( or reflects at ) a nonmagnetic ferromagnetic interface , the spin precession becomes incoherent over a very short distance . this results in a loss in its average spin angular momentum in the direction _ transverse to the local magnetization_. since the total angular momentum has to be conserved in a closed system , whenever the flow of angular momentum changes , a torque acts on the local magnetization . this torque is often referred to as the `` in - plane '' torque , where the plane is defined by the local magnetization and incoming spins . this torque may act gradually , as when spin - polarized electrons flow through a domain wall and have to rotate so as to follow the local magnetization , or very abruptly , as when a spin current enters a ferromagnet with its magnetization in a direction different to that of the spins in the spin current ; this is almost always the case in stnos and shnos . such an in - plane torque manifests itself as an energy - nonconserving force , similar to damping . furthermore , depending on the sign of the current , this in - plane torque is either parallel or antiparallel to the damping torque . when antiparallel to the damping torque , and for sufficient current magnitude , the current acts to destabilize the local magnetization , thus exciting large angle dynamics . there is also the possibility of an additional stt term that is oriented perpendicular to the plane defined by the local magnetization . this `` out - of - plane '' torque is often referred to as an `` effective field '' contribution , as it has the same symmetry as is typically used in field - driven dynamics . while the perpendicular contribution is small for all - metallic multilayers @xcite , it can be substantial in devices with tunneling barriers @xcite . it was not until sufficient advances in nanofabrication techniques had been achieved that magnetic devices with nanoscopic dimensions could be readily fabricated and the currents needed to induce magnetodynamics be decreased to the ma range . one of the first techniques relied on mechanical point - contacts , where a highly sharpened metallic tip is used to make electrical contact , typically with an area on the order of 10@xmath2 nm@xmath2 , to an extended multilayer film @xcite . soon after , lithographically defined contacts were employed in devices that are now termed nanocontacts , to distinguish them from their mechanical point - contact predecessors @xcite . the fabrication of nanocontacts involves opening a small electrical contact of either circular or elliptical @xcite cross - section through an insulating layer using either electron - beam lithography ( ebl ) or hole - colloid lithography ( hcl ) @xcite techniques . in both point - contact and nanocontact devices , it is only the current injection site that is confined to a nanoscopic region . however , in oscillators based on the nanopillar @xcite geometry , all of the magnetic layers are patterned into a wire with either circular or elliptical cross - section . fabrication is most often accomplished by a combination of ebl and ion milling . however , stencil @xcite and electrodeposition @xcite techniques have also been pursued . the nanopillar geometry ensures that all of the electric current passes through the entire multilayer stack , typically resulting in lower current densities ( < 10@xmath3 a / cm@xmath2 ) being needed to excite stt - induced magnetization switching or oscillations than in the case of nanocontact - based devices ( 10@xmath510@xmath6 a / cm@xmath2 ) . it should be noted that nanopillar fabrication is typically more involved and prone to deleterious edge effects that can increase oscillator linewidths and magnify sample - to - sample variations . due to the smaller magnetic volumes involved , nanopillar - based devices are also more sensitive to thermally induced fluctuations , which results in larger linewidths than in nanocontact devices @xcite . another important consideration is that , as all of the magnetic layers are patterned , significant stray fields are present in nanopillars and need to be considered in analyzing them . additionally , hybrid device architectures have also been pursued in which a nanopillar is fabricated on top of an extended film @xcite . such hybrid structures are useful if , for example , optical access to the extended magnetic layer is desired . generally speaking , the transport of spin current , and therefore of the associated stt , does not need to accompany a charge current . such nonlocal devices are engineered to promote different paths for the spin and charge currents . this geometry is particularly useful in tunnel - junction based devices where large voltages can cause electrical breakdown of the tunnel barrier . such devices typically require at least three terminals for both spin current generation and electrical detection . both stt - induced nonlocal switching @xcite and oscillations @xcite of the magnetization have been demonstrated . devices where the stt is provided by the she that is , where the charge and pure spin currents naturally flow in orthogonal directions , are prime examples of nonlocal oscillators @xcite . by and large , most material stacks utilized in stno devices are based on a ferromagnet / spacer / ferromagnet ( fm / s / fm ) trilayer . one fm is considered free and is most susceptible to the influence of stt . it is therefore advantageous for the free layer to be relatively thin and to have a low intrinsic gilbert damping . ni@xmath7fe@xmath8 , which goes by the trade name `` permalloy '' and is often referred to simply as nife , is the canonical choice for the free layer in the majority of stnos and shnos . in contrast , the other fm layer is fixed , either through pinning to an adjacent magnetic layer or by virtue of its intrinsically larger anisotropy or thickness . the primary role of the fixed layer is to spin - polarize the initially unpolarized electron current . the spacer layer may be metallic ( most typically cu ) , forming what is referred to as a pseudo spin valve ( psv ) , or insulating , as in the case of magnetic tunnel junctions ( mtjs ) . as the tunneling magnetoresistance ( tmr ) effect is much larger than the giant magnetoresistance ( gmr ) effect in all - metallic psvs , mtjs are becoming standard in most nanopillar devices @xcite . the use of mtjs in conventional nanocontacts is challenging , as the topmost fm layer would act to shunt the current . nevertheless , there has been some success in creating hybrid nanocontact mtjs , often termed `` sombrero '' structures , which aim to marry the low linewidths inherent in nanocontact devices with the large output powers found in mtjs @xcite . the materials that constitute the fm layers are diverse and can be broadly classified and distinguished by their equilibrium magnetization configurations , as described next . the vast majority of stno devices studied in the early stages were based on free and fixed fm layers with an intrinsic easy - plane anisotropy meaning that the magnetization prefers to lie in the plane of the film if no external magnetic field is applied @xcite . easy - plane materials , such as co and co@xmath9fe@xmath10 for the fixed layer and nife for the free layer , have most often been pursued . one drawback of utilizing materials with easy - plane anisotropy is that large ( 1 t ) fields are often required to overcome the demagnetization fields needed to pull the magnetization out - of - plane . owing to their improved thermal stability , scalability , and potential for low to zero field operation , materials with perpendicular magnetic anisotropy ( pma ) are extremely attractive . one of the first _ all perpendicular _ nanopillar stt devices combined a [ co / pt]@xmath11 multilayer as a fixed layer with a [ co / ni]@xmath11 multilayer free layer @xcite and was primarily devoted to stt - induced switching . later stnos with perpendicular free and fixed layers have shown zero - field operation @xcite . in these types of devices , the equilibrium magnetization direction of the fixed and free layers are orthogonal to each other . one of the first devices employed a perpendicular fixed layer based on a co / pt multilayer and a planar nife free layer @xcite . more recent devices have tended to deal with the opposite situation , with in - plane fixed layers using either co , co@xmath9fe@xmath10 , or even nife , and a perpendicular co / ni multilayer free layer @xcite . a tilted magnetization will have both in - plane and out - of - plane components ; its use in stnos is predicted to provide additional degrees of freedom to manipulate both static and dynamic states . relative to conventional in - plane or pma magnetic films , the use of tilted fixed layers can optimize microwave signal generation and enhance spin - torque efficiency while maintaining high output powers with low to zero field operation @xcite . to date , experimental work has primarily focused on the fabrication of tilted anisotropy exchange - coupled films by taking advantage of the competition between the pma of a co / pd or co / ni multilayer and the in - plane shape anisotropy of a nife of cofeb layer . @xcite . subsequent measurements have then focused on the fundamental spin wave excitations in such tilted anisotropy exchange springs @xcite . zero field operation in cofeb / mgo / cofeb mtjs with tilted cofeb free layers has been experimentally demonstrated @xcite . departing from the trilayer configurations discussed above , theory @xcite predicts that spin filtering by a single fm layer can generate a sizable spin accumulation at each interface . however , it is only when these interfaces are asymmetric that an imbalance of spin accumulation will lead to a significant stt . experiments on nanopillars with a single co layer showed that sufficient stt could be generated to produce spin - wave excitations @xcite and even switch the co magnetization @xcite . the latest work on single layer devices has been in the nanocontact geometry , where evidence of a periodic vortex / antivortex creation / annihilation mode has been found @xcite . stnos exhibit a diversity of fundamental spin wave modes that depend greatly on , for example , the device architecture , constituent fm materials , applied magnetic field strength , and applied magnetic field angle . here we summarize some of the most prevalent stt - induced excitations in both nanopillar and nanocontact stnos . the particular spin - wave mode generated in a nanopillar geometry depends most sensitively on the diameter or thickness of the free fm layer , as well as on the strength of the external magnetic field . frequently , the dimensions of the free layer are such that a magnetic vortex characterized by an in - plane curling of the magnetization with a 10-nm sized out - of - plane vortex core at the center is a stable equilibrium configuration @xcite . if so , then the stt will excite the fundamental gyration of the vortex core with frequencies typically less than 2 ghz @xcite . however , in a large enough external field , the vortex state will no longer be energetically favorable @xcite , the magnetization will become more or less uniform , and stt will typically excite highly elliptical clam - shell precession in the case of a free layer magnetized in - plane . similarly , if either the diameter or thickness of the free layer is sufficiently small that exchange interactions dictate a uniform magnetization , then such uniform precessions will also be generated . for nanocontact stnos where the equilibrium magnetizations of both the free and fixed layers prefer to lie in - plane , the type of spin wave excitation depends strongly on the external applied magnetic field angle . as described by both analytical @xcite and micromagnetic simulations @xcite , the initial experiments @xcite using in - plane applied magnetic fields excited a strongly nonlinear , self - localized solitonic bullet mode . however , when the free layer is magnetized perpendicularly to the film plane , as described by slonczewski @xcite , an exchange dominated propagating spin wave with a wave vector inversely proportional to the nanocontact radius is excited , as directly proven using brillouin light - scattering experiments @xcite . for intermediate oblique angles , the behavior becomes much more complex ; both simulations @xcite and experiments @xcite provide evidence of mode hoping @xcite between propagating and localized bullet modes @xcite . the use of elliptical nanocontacts has been useful in trying to disentangle the coupling mechanisms between these modes @xcite . special attention must be given in the nanocontact geometry to the critical role played by the current - induced oersted field . for typical drive currents and nanocontact dimensions , the oersted field is on the order of 0.1 t at the nanocontact edge ; this is a significant fraction of the external applied magnetic field and can not therefore be considered an insignificant perturbation . for example , the oersted field results in a localization of the propagating mode for applied field angles near the film plane , which in turn promotes mode coexistence with the solitonic bullet mode @xcite . furthermore , at large applied field angles , the oersted field induces an asymmetric fmr field landscape which results in asymmetric spin wave propagation @xcite . in short , in regions where the addition of the external field and the oersted field produce a local field maximum , spin waves can no longer propagate , resulting in beam - like spin - wave propagation towards ( and beyond ) the local field minimum on the other side of the nanocontact @xcite . finally , the oersted field also plays an important role in the formation of vortex oscillations in nanocontacts @xcite . in small external magnetic fields , the circulating oersted field favors a vortex state in the free layer via the zeeman interaction . this interaction in turn generates a confining potential for the vortex core . the action of stt is then to drive the vortex core away from the center of the nanocontact , whereas the natural damping tends to pull the core back in . in equilibrium , these two forces result in a self - sustained vortex gyration around the nanocontact . in stark contrast to nanopillars , where the vortex motion is physically constrained , these vortex orbits can even occur in regions outside of the nanocontact itself . in such cases , this results in a very large time - varying magnetoresistance signal and thus large output powers . as with vortices in nanopillars , the oscillation frequencies are usually less than 2 ghz and the emission spectrum usually contains several higher harmonics due to elliptical orbits . when driven past a critical velocity , the vortex core can flip polarity , leading to chaotic dynamics @xcite . significant changes in the fundamental magnetodynamical modes are found when the free layer exhibits pma , such as occurs in the oft - employed co / ni multilayers . when the applied field is perpendicular to the film plane , three distinct operational modes can be identified . at small applied magnetic fields , a small - angle precessional fmr - like mode is observed with relatively small power @xcite . as the magnitude of the drive current is increased ( decreased ) , there is a gradual increase ( decrease ) in mode power ( frequency ) consistent with an increasing precession angle . however , at larger applied fields , the oscillator behavior changes dramatically . as the applied field is increased , a dramatic drop in oscillator frequency and an increase in oscillator power are observed , consistent with the formation of a magnetic droplet soliton @xcite . the theoretical basis for the droplet soliton was initiated by ivanov and kosevich @xcite , who showed that the landau lifshitz equation can sustain a family of conservative magnon drop solitons in the absence of damping . while initially a theoretical curiosity , it was then shown that the balance between stt and damping underneath the nanocontact can provide the zero damping needed to sustain what are now called dissipative droplet solitons @xcite . finally , at very large external fields , the zeeman energy will dominate and the droplet will collapse @xcite . the influence of the spin current generated from she - induced spin accumulation on the ferromagnetic resonance of an adjacent magnetic layer has been initially shown on micrometer - sized permalloy discs @xcite . while the linewidth could be reduced by a factor of two , full compensation of the damping , and thus auto - oscillations , were not achieved in these micrometer - sized films , as the uniform application of the torque does not favor any auto - oscillation mode . shnos have thus been realized by nanopatterning of various parts of the bilayer to achieve the nonuniform stt , and can be categorized into four primary device geometries . in the three - terminal nanopillar geometry , the magnetic free layer is a nanosized element on top of an extended slab of the she - generating normal metal @xcite . the electrical read - out is carried out here through an mtj element including the oscillating free layer , which is how appreciable output powers can be achieved . another approach is to utilize a nanogap between two highly conductive electrodes , injecting the necessary high current densities into a nanosized region of an extended fm / pt bilayer @xcite . in the more recent nanoconstriction - based shnos , both the fm and the pt layer are patterned into a nanoscopic constriction @xcite . similar confinement of the she - induced spin current has been demonstrated in a 190 @xmath12 nanowire geometry @xcite . the benefits of the latter two shno device geometries over their stno counterparts include easier fabrication ( for example , they require the deposition of only two layers and fewer lithographic steps ) and direct optical access to the oscillating region . in contrast to stnos , which rely on the gmr or tmr effects , these shnos must however rely on the smaller anisotropic magnetoresistance ( amr ) effect in a single fm layer , and therefore typically exhibit lower output powers . the generation of a sufficient she - induced pure spin - current requires a material with a large spin - orbit coupling . to date , only results on shnos using pt and ta have been published . however , a few different fm materials have been studied in addition to nife , including co@xmath13fe@xmath13b@xmath14 @xcite , co / ni multilayers with pma @xcite , and the ferrimagnetic insulator yttrium iron garnet ( yig ) @xcite . for in - plane magnetized thin films , two different auto - oscillation modes have been observed in nanogap shnos . as described in @xcite , these devices have been studied for in - plane applied fields with microfocused brillouin light scattering and from _ ( i ) _ the sub - fmr frequency and _ ( ii ) _ the spatial size , which was smaller than the instrument resolution , the existence of a self - localized spin wave bullet was inferred @xcite . electrical measurements as a function of the temperature , however , showed the existence of the low - frequency and low - linewidth bullet mode only for cryogenic temperatures with a second high - frequency mode appearing at higher temperatures , and even being dominant at room temperature @xcite . micromagnetic simulations suggest that this high - frequency mode is generated within local field minima caused by the bullet mode which implies the simultaneous excitation of both @xcite . however , since the oersted field generated from the nonuniform current in the pt layer always counteracts the applied in - plane field , a similar field minimum in the nanogap is always present , allowing the high - frequency mode to form . using a 1 @xmath15 m large gap between the electrodes , and thus generating rather uniform current distributions , @xcite showed that the high - frequency mode most likely resembles a coherent excitation of the fmr mode . in the nanowire shnos , the existence of both bulk and edge modes has been identified @xcite while the linear high - frequency excitation shows an increasing threshold current with the strength of the applied field , the spin wave bullet has a relatively constant threshold current as long as the internal magnetization angle stays below a critical value @xcite . the output characteristics can be thus tuned between these two regimes via the application of a tilted magnetic field , where the electrically measured linewidth is lowest for high fields , since only the spin - wave bullet exists there @xcite . moreover , the interplay can be varied by the nanogap size , favoring the bullet mode for smaller sizes . when both modes are energized at the same time for intermediate field strengths , their linewidths are broadened due to mode coupling events @xcite . propagating spin waves with frequencies above fmr are important in , for example , mutual synchronization , and are predicted by micromagnetic simulation in nanogap shnos for out - of - plane fields @xcite . however , in this study , the current densities required are about twice as large as for the in - plane field modes , which inhibits the experimental observation of the propagating spin wave modes . nanoconstriction - based shnos show a mode that evolves from the thermal fmr , as shown by micro - focused bls measurements and electrical detection @xcite , and is therefore comparable to the high - frequency mode of nanogap shnos . however , with the constricted geometry around the oscillation center , the presence of the bullet is hindered and the single - mode excitation can be detected with lower linewidths . synchronization to external sources , as described later in regards to injection locking , or mutual phase locking of individual stos , as described here , are techniques often used to improve oscillator power and linewidth . there are several coupling mechanisms that have been investigated to synchronize stnos to each other , each with a varying degree of experimental success . synchronization of serially connected nanopillar stnos using their shared microwave electric current @xcite has been theoretically proposed , but to date has shown no clear success . another route to synchronize nanopillars is to utilize the relatively long - range dipolar stray microwave fields generated during precession . recent simulations show this to be a feasible solution for vortex - based oscillators @xcite . to date , however , the most successful experimental works on synchronizing stnos have occurred in the nanocontact geometry . in such devices , the nanocontacts are patterned on top of a shared free layer , which in turn provides the coupling medium . the first works to show that two stnos could be synchronized were published in 2005 @xcite it was soon shown experimentally @xcite , as well as investigated theoretically @xcite and micromagnetically @xcite , that the coupling mechanism is mediated by propagating spin waves . since these early experiments , progress has remained relatively slow . only in 2009 was it demonstrated that four vortex - based stnos could synchronize , this time by direct exchange interactions mediated by an antivortex . @xcite . then , in 2013 , the successful synchronization of three high - frequency stnos was shown for the first time @xcite . recent simulations have shown robust synchronization of an arbitrary number of nanocontact vortex oscillators mediated by the strong magnetodipolar interaction @xcite . the most recent work on synchronization has focused on the use of highly directional spin - wave beams . while it has been shown that the oersted field generated in the vicinity of the nc can dramatically alter the emission pattern of sws @xcite , its role in the synchronization behavior of multiple ncs has been largely ignored . synchronization is promoted ( impeded ) by the oersted field landscape when the nanocontacts are oriented vertically ( horizontally ) with respect to the in - plane component of the external field , due to the highly anisotropic spin - wave propagation . not only is robust synchronization between two oscillators observed for separations greater than 1000 nm , but synchronization of up to five oscillators has been observed in the vertical array geometry . furthermore , the synchronization must be considered driven , rather than mutual , in nature as the final frequency is enforced by the stno from which the sw beam originates @xcite . the precession frequency of the stnos is tunable over a large range , both through the direct current and the applied magnetic field . the precession frequency generally increases as the external magnetic field increases . its behavior is generally governed by the kittel equation , according to which the precession frequency increases as the square root of the applied field at low fields and linearly at high fields . theoretical works predict a very high frequency of 0.2 thz for the common ferromagnetic material ni@xmath16fe@xmath14 . @xcite in the very first work by kiselev _ et al . _ , @xcite a highest frequency of 25 ghz was demonstrated . the devices used were nanopillars of co / cu / co/ multilayer . a year later , rippard _ et al . _ @xcite demonstrated a frequency of 38 ghz in a nanocontact device based on a ni@xmath16fe@xmath14 free layer . the field tunability of frequency is 18.5 mhz / mt and the current tunability is about -200 to 400 mhz / ma for nanocontact stnos . @xcite bonetti _ et al . _ , @xcite reported a frequency as high as 46 ghz for high magnetic fields applied normally to the film plane . the stno frequency was limited by instrumentation . it was reported that it would be possible to achieve a frequency beyond 65 ghz , which would be potentially attractive for a range of millimeter - wave applications , such as short - range high - speed radio links ( ieee 802.15.3c ) near 60 ghz and vehicle radar for active cruise control at 77 ghz . @xcite an stno may be viewed as a time - dependent resistor with its resistance given by : @xmath17 where @xmath18 is the average resistance of the stno , @xmath19 is the magnetoresistance of the material stack , and @xmath20 is the precession frequency . for a gmr - based device , @xmath19 is low , and hence the power generated by the gmr - based stnos is also low , typically in the sub - nw range . @xcite higher powers have been achieved in magnetic tunnel junction based material stacks , for which @xmath19 is significantly larger . in magnetic tunnel junctions , powers of up to 3.6 @xmath15w have already been reported for spin - torque vortex oscillators with an feb free layer @xcite . however , the frequency of operation of these vortex oscillators is typically low . for uniform mode oscillation , the highest power achieved so far has been 2 @xmath15w , obtained from a sombrero - shaped nanocontact magnetic tunnel junction based stno with a perpendicular anisotropy free layer @xcite . for in - plane magnetized mtj - stnos , the highest power reported was 140 nw @xcite for a frequency around 5 ghz . while the frequency can be increased using the magnetic field , the output power usually drops with increasing field strength ; obtaining higher power at higher frequencies is thus a challenge . a novel method of achieving this involves frequency doubling in mtj - stnos @xcite . in that work , powers of up to 1.25 nw were demonstrated at a frequency of @xmath2110 ghz . modulation is a very important requirement for the communication applications of stnos . modulation studies of stnos are , however , surprisingly limited despite the issue s critical importance for communications applications . the first experimental demonstration of frequency modulation in an stno was reported in 2005 @xcite , with the demonstration that the addition of an alternating current to the direct current results in frequency - modulated spectral output , generating sidebands spaced at the modulation frequency . this phenomena is demonstrated in fig.[fig - mod1 ] ( b ) , where equally spaced multiple sidebands are produced upon the addition of an rf current ( 1 ma , 100 mhz ) . the device is shown in fig.[fig - mod1 ] ( a ) and corresponds to that of @xcite . as can be seen in this figure , the sidebands have unequal amplitudes and the carrier frequency shifts with the addition of the rf current . this phenomena was first reported in @xcite and was attributed to nonlinear frequency modulation . however , in 2010 , it was shown that the unequal sideband amplitudes result from a combination of nonlinear frequency and amplitude modulation ( nfam ) . it was shown that the combined nonlinear frequency and amplitude modulation model can accurately describe the experimental data without any adjustable parameters . the frequency of modulation was 40 mhz in @xcite and 100 mhz in @xcite . later , it was shown that a nanocontact - based stno can be successfully modulated to a modulation frequency of 3.2 ghz @xcite . the agreement of experiment with the nfam model is maintained until 1 ghz @xcite . it was also shown that modulation can be used to narrow the linewidth of stnos by as much as 85 % @xcite . this linewidth reduction is due to a general averaging effect of modulation on the nonlinearity of the stno signal and was explained on the basis of nonlinear auto - oscillator theory @xcite and nonlinear frequency modulation . in 2011 , modulation of a pair of nanocontact stnos was successfully demonstrated@xcite . in fig.[fig - mod2 ] , we reproduce the results from @xcite which show that the synchronized state of a pair of nanocontact stnos ( at 57 ma ) can be maintained while modulating . furthermore , the modulation behavior can be explained by assuming a single oscillator model , as shown in fig.[fig - mod2](b ) . this finding was very promising for applications involving a large array of synchronized stnos . many electronic circuits use injection - locked phase - locked loops ( plls ) as one of the building blocks of rf transceivers and so it is also crucial for applications to demonstrate the modulation of the injection - locked state of the stno . more recently , modulation of injection - locked mtj - stnos was demonstrated and the modulation - mediated nonresonant unlocking mechanism was described @xcite . interestingly , the unlocking phenomena sets a lower limit on the modulation frequency but no upper limit on the modulation frequency for injection - locked mtj - stnos . experimental studies of the modulation of vortex - based stnos are rather limited . the first digital modulation of vortex - based stnos was demonstrated in @xcite . using frequency shift keying ( fsk ) , it was shown that the oscillator can switch between two stabilized frequencies differing by @xmath22 in 20 ns . this performance is difficult to achieve in a voltage - controlled oscillator , and vortex - based stnos hence hold much promise for fsk modulation - based applications . more recently , analog frequency modulation has been demonstrated in a magnetic tunnel junction nanopillar in in - plane fields @xcite . in this work , the authors show modulation for modulation frequencies that are up to 10% of the oscillator frequency . in these experiments , however , the modulation period is comparable to the characteristic time of the vortex dynamics , and hence the frequency modulation data was modelled using `` deviation sensitivity , '' which differs significantly from the tunability measured in a quasistatic regime . when the modulation frequency of an stno is increased , the stno may not be perfectly modulated . a recent study has shown that the upper limit of the frequency of modulation in an mtj - stno is 200 mhz @xcite . in this work , the authors showed that the bandwidth of modulation is limited by the amplitude relaxation rate @xmath23 and that the power level of amplitude modulation strongly decreases beyond @xmath24 , where @xmath25 . this finding is in agreement with the nonlinear auto - oscillator theory of stnos @xcite . although the original nonlinear auto - oscillator theory does not include amplitude modulation , a recent theoretical study that includes an nfam scheme also suggests that the modulation bandwidth of stnos is determined by @xmath23@xcite . the miniature size of stnos and shnos compared to industry - standard oscillators inevitably makes their oscillation more susceptible to noise . the stability of any oscillator can be characterized in terms of its amplitude noise and phase noise , where the width of the spectral peak ( commonly referred to as the linewidth ) is usually governed by the phase noise as has also been found for stnos . the phase noise of shnos has not been as extensively studied as that of stnos , but one initial report @xcite provides a lowest linewidth on par with what has been achieved for stnos . a prerequisite for stable oscillation is that the oscillation is not intermittent . one type of intermittency is on off toggling , which has been observed for low drive currents @xcite . another type is mode - hopping between different distinct frequencies . stnos do in many cases show spectra with the presence of several modes represented by distinct frequencies , where time - domain measurements can often reveal a random mode - hopping between these frequencies . the mode - hopping rate can range from milliseconds down to nanoseconds , meaning that , depending on the bandwidth of the characterization setup , it might not always be observed . it should be noted that in some cases there is the possibility of actual coexistence ( in time ) , a clear indicator of which is the appearance of second - order intermodulation products @xcite . if the application is low bandwidth compared to the mode - hopping rate , the mode - hopping itself need not constitute too large a problem , since the fast fluctuations might be washed out . however , the mode - hopping has a large negative effect on the coherence of the oscillation . in a mode - hopping event from one frequency to another and then back again , the phase will be lost ( introducing phase noise ) . this becomes visible in the spectrum in the form of a dramatic linewidth increase @xcite . the occurrence of multiple simultaneously excitable modes is most evident in nanopillar stnos @xcite , where the modes emerge as various edge and center modes on account of its confined geometry . in nanocontact stnos , with their far - extending magnetic films , mode - hopping within the propagating mode has been experimentally observed only at very high drive currents ( close to the breakdown of the oscillation ) , @xcite while simultaneous excitability of the propagating and localized bullet mode can also induce mode - hopping @xcite . moreover , the propagating mode in nanocontact stnos generally exhibits discontinuous mode transitions as a function of the bias point @xcite . close to these transitions , the linewidths of the individual modes typically increase @xcite , which could be theoretically explained by rapid hopping between the modes @xcite , though this has not been experimentally confirmed yet . apart from control over the existence of multiple oscillation modes ( and the instability they introduce to each other ) , a second fundamental question is that of the stability of a single mode . this stability is primarily considered to be governed by thermal magnetic noise which is usually modelled as random fluctuations in the magnetic field . this has a straightforward implementation in the macrospin approximation . micromagnetically , it is usually implemented as fluctuations that are uncorrelated in both time and space @xcite . it has been noted @xcite , that no rigorous theory has been presented to explain the effects of spatial fluctuations on the spin - wave eigenmodes . another aspect of stability arises from the complexity of micromagnetic systems , where spatial inhomogeneity can be induced by locally varying fields even at zero temperature @xcite . such spatially varying fields may stem from the geometry through the fabricated shape of the free layer or the roughness in the nanometer - scale film thickness , or through local anisotropy fields in the metal grains . the noise properties have been found to be highly dependent on the strong amplitude frequency coupling @xcite which acts to transfer amplitude noise into frequency noise @xcite . as the magnetization is precessing around its steady - state orbit , fluctuations act to displace it along two orthogonal directions : in parallel with and perpendicularly to the orbit . parallel displacement introduces a single phase shift whereas a perpendicular perturbation ( fig . [ fig - phasenoise](a ) ) provides no initial phase difference . however , since a perpendicular perturbation places the magnetization away from the steady - state orbit at an oscillation amplitude different from the steady - state amplitude , the magnetization will momentarily take on another frequency associated with this new amplitude . during the time of precessional relaxation of the amplitude back to the steady - state orbit , the frequency will gradually approach the steady - state frequency , but before this frequency has been reached , the oscillator will have accumulated a phase shift as compared to the nominal frequency ( fig . [ fig - phasenoise](b ) ) . quantitatively , both theoretical calculations @xcite and experiments @xcite show that the main effect of perpendicular perturbations and of amplitude - frequency coupling is by far the dominating mechanism for the phase noise in stnos . in the framework of the general nonlinear auto - oscillator theory , it has also been demonstrated how , in the vicinity of the oscillation threshold , the linewidth increases due to spectral line asymmetry induced by several statistical solutions to the noise process @xcite . the high impact of the strong amplitude frequency coupling on the resulting phase noise means that the oscillator will be least sensitive to fluctuations when its amplitude frequency coupling is nonexistent , but can also be improved by achieving a higher amplitude relaxation rate @xcite . from a fundamental thermal noise perspective , the frequency stability should be increased by the use of materials with low thermal damping of the magnetization precession i.e . materials that exhibit a low value for the gilbert damping parameter . on the contrary , however , the amplitude relaxation rate is proportional to the gilbert damping , meaning that the amplitude perturbations will be damped out more slowly . since the amplitude fluctuations are the major source of frequency noise , a high value for the gilbert damping parameter is preferable . @xcite . several experimental studies report the presence of low - frequency @xmath26 frequency noise in both mtj nanopillar @xcite and gmr nanocontact stnos @xcite . from a characterization point of view , the high level of @xmath26 frequency noise is challenging in the sense that the standard phase noise measurement modes of spectrum analyzers are unable to lock onto the stno electric rf signal . the research community has been able to overcome the characterization problem through time - domain measurement of the waveform and subsequent digital signal processing @xcite , but from an application point of view , @xmath26 noise remains an issue . this @xmath26 frequency noise has been reported to increase in the presence of multiple oscillation modes @xcite , but is nonetheless of considerable strength in the apparently single - mode case . there is currently no theory of stnos that can account for the presence of low - frequency @xmath26 frequency noise in single - mode oscillations , but it has been suggested that this might be related to the existence of unexcited oscillation modes or @xmath26-type amplitude noise possibly introduced by microscopic material inhomogeneity @xcite . two major paths are currently discernible for stabilizing the generation of stno and shno frequencies by applying electrical signals to the device . the first is injection - locking , in which the stno @xcite or shno @xcite is locked to an injected rf signal at the same nominal frequency or at an integer or integer fraction multiple thereof . the locking can be achieved through direct electrical injection of the external rf signal ( superposed on the dc drive current ) or through an ac magnetic field from an adjacent microstrip line . one compelling system is that in which the stno / stho itself acts as the injection source , thereby acting to stabilize itself . such a system has been investigated @xcite analytically and , for the case of a vortex mtj stno , also numerically . with the correct choice of delay time for the reinjected signal , the linewidth for this vortex mtj stno case could be decreased by a factor of 4 . self - injection locking has recently been achieved also experimentally , showing a maximum decrease in linewidth by a factor of 2.5 @xcite . a second path is to use a phase - locked loop ( pll ) for the stno / shno . in a pll , the oscillator s rf signal is compared with a reference signal and the resulting phase difference is used to instantly rebias the oscillator , taking advantage of its frequency - tuning mechanism . pll - locking through an external reference signal has been experimentally achieved for stno vortex oscillations below 150 mhz @xcite and , very recently , for a mtj stno at 5 ghz @xcite . the limited number of reported studies is probably related to the lack of pll circuits with bandwidths high enough to track the stno signal . a unique feature of stnos and shnos is their ability to provide multiple functionalities , from microwave source to microwave detection and signal processing . this versatility opens up the possibility of employing stnos and shnos in a wide variety of applications . the most common applications include microwave sources , microwave detectors , noncoherent transceiver , magnonics , neuromorphic computing , magnetic field sensing , and magnetic field generation . these applications will be introduced in this section . moreover , the issues , challenges , and developments in stno- and shno - based applications will be discussed . the miniature size , very wide tunability , high integration level with cmos technology , and low power consumption of stnos make them good candidates for current - controlled oscillators in multiband multistandard radio systems @xcite . table i summarizes the performance of different stnos in the literature . as can be seen from table i , a single mtj stno @xcite can be used as a local oscillator ( lo ) covering the entire bandwidth of ultrawideband ( uwb ) applications . gmr stnos typically have higher and broader tunable frequency ranges . the operating frequency of gmr stnos has been experimentally observed up to 46 ghz and extrapolated to 65 ghz @xcite . it has been theoretically estimated that the operating frequency of stnos may approach 200 ghz if stno nanocontacts can be made much smaller @xcite . more recently , a very low gilbert damping and ultrafast spin precession with frequencies of up to 280 ghz has been demonstrated @xcite , which makes it possible to further extend the operating frequencies of stnos . the extremely high and broad operating frequency range of stnos also opens up the possibility of utilizing them in millimeter wave application for instance , in wideband data transmission systems . [ cols="<,<,<,<,<,<,<,<",options="header " , ] the phase noise of the mtj stno was measured by nanosc ab at about -65 dbc / hz at 1 mhz offset . the issues of spectrum impurity and the limited output power must be overcome before stnos enter widespread use as microwave sources . recently , significant progress has been achieved in understanding the fundamentals and in overcoming these two limitations of stnos @xcite . as can be seen in table i , a sombrero - shaped nanocontact geometry for mtjs has been developed @xcite that remedies the incompatibility of conventional nanocontact geometry with mtjs . this geometry enables the optimization of either the q factor ( narrower linewidth ) @xcite or the output power @xcite . specifically , a very narrow linewidth of 3.4 mhz @xcite and an output power of -26.2 dbm @xcite were obtained . these newly developed mtj stnos show significant improvement in either linewidth or output power in comparison with the conventional mtj stnos @xcite . in addition , recent research has shown that by using a tilted cofeb fl @xcite , the bias field can be removed , which opens up more possibilities for mtj stno - based applications . for gmr stnos , which typically have a narrower linewidth than mtj stnos , the output power has been enhanced by using cofemnsi ( cfms ) layers as magnetic layers @xcite . by further combining the cfms with nanocontact geometry , a large output power of -46 dbm and a narrow linewidth of 3 mhz can be achieved simultaneously at zero applied field @xcite . in addition , the phase noise of stnos has been reduced by using mutual synchronization and current modulation as an extrinsic tool @xcite . the limited output power has been improved by employing amplifiers dedicated to stnos @xcite . the achieved improvements of phase noise and output power bring stnos closer to being used as microwave sources . in addition to these issues , the device - to - device performance variation of stnos is significant . furthermore , the mode structure and temperature dependence of stnos need to be well understood so as to suppress the unwanted modes and to estimate the performance variation that results from changes in temperature . other issues that need to be taken into consideration in applications include the irreversible changes of stnos and thus reliability issues at high biasing current @xcite . the stt provides many useful phenomena as detailed in @xcite . one important phenomenon is the spin - torque diode effect discovered in mtj - based devices @xcite . the spin - torque diode effect can be described as follows : when a microwave signal with a frequency close to the natural fmr frequency of one of the mtj s electrodes is incident on the device , the induced oscillating tunnel current can excite magnetic precession via spin transfer . the resistance oscillation that results from this precession is combined with the oscillating current so as to produce a measurable dc voltage component across the tunnel junction @xcite . recently , an mtj microwave detector that does not require any external bias fields , and that possesses a sensitivity of about 75400 mv / mw has been demonstrated at room temperature @xcite . in addition , this microwave detector can provide very high sensitivity even at a low input microwave powers of 10 nw ( -50 dbm ) . its sensitivity is about one order of magnitude higher than that of a semiconductor diode , such as a schottky diode . the large sensitivity of mtj stnos makes them promising candidates as frequency - tunable resonant microwave detectors for a wide range of applications in telecommunications , radars , and smart networks @xcite . taking advantage of the fast switching speed of stnos , an stno - based wireless communication system with binary amplitude shift keying ( ask ) has been demonstrated in @xcite . in this system , the stno performs the ask modulation at the transmitter and a simple envelope detector recovers the data at the receiver . for this noncoherent communication system , the considerable phase noise of stnos is not an issue . an stno with 1 ns turn - on time ( or switching speed ) allows a communication system that can achieve a data rate as high as 1.48 gbps @xcite . considering also the spin - torque diode effect , stnos can be utilized as both modulators and demodulators in transceiver systems , as shown in fig . [ sto_based_transceiver ] . as modulators , stnos translate different currents representing the binary data `` 0 '' and `` 1 '' directly into signals at different frequencies . as demodulators , stnos enable the direct conversion of the received signals to high and low output voltages representing `` 0 '' and `` 1 '' . these stno - based transceivers may enable the development of low - complexity , low - cost , high data - rate , and noncoherent communication systems . stnos with a point contact ( fig . 2(a ) ) on an extended free layer allow spin - torque - driven emission as well as propagation of spin waves outward from the nanocontact @xcite . the spin wave emitted from an stno with a nanoscale electrical contact has been directly observed and can propagate for several micrometers @xcite . the emission of propagating spin waves is the basis of synchronizing arrays of stnos for microwave applications . moreover , stnos are very attractive as wave emitters for future use in magnonic devices . magnonics is an emerging and rapidly growing research field that connects magnetism , spintronics , and electronics @xcite . it investigates and utilizes magnetic phenomena connected with spin waves , in a broad sense . magnonics employs different devices to deal with spin waves . the key devices are spin - wave emitters , spin - wave manipulators ( modulators ) , and spin - wave detectors . stnos , as previously discussed , can play all the roles of these key devices . as a consequence , stnos have the potential to realize all the fundamental functions that are necessary for magnonics @xcite . neurons can be modelled as nonlinear oscillators that vary their rhythms based on incoming signals @xcite . to form a network of neurons coupled by synapses , the nonlinear oscillators need to be able to synchronize with each other in frequency or phase . the neural network can therefore be realized using the non - boolean nature of these nonlinear oscillators @xcite . stnos are able to act as nonlinear oscillators because they can synchronize when coupled via a mutual electrical or magnetic interaction . furthermore , it has been demonstrated that other spintronic devices including spin - torque memristors and spin - torque memories are very suitable in playing the roles of synapses and memory processes in mimicking neural networks @xcite . consequently , stnos , together with other spintronic devices , pave the way to full spintronic implementations of neural networks , including both polychronous wave computation and information storage @xcite . non - boolean computing based on stnos shows great potential in data - processing applications , particularly in neural networks . stno - based non - boolean computing is discussed in @xcite and its physical processes are reviewed in @xcite . particularly , the concept of a dual - pillar stno that decouples bias and read paths for non - boolean energy - efficient computing has been proposed , and a simulation framework has been presented in @xcite . the dual - pillar stno is composed of a gmr stno and an mtj stno . the gmr stno is employed for the bias path so as to achieve a low - resistance interface and hence to allow low bias - voltage , as well as high energy efficiency . the high - resistance mtj stno is used to sense a small current and then to produce an output signal with large amplitude . the magnetoresistive effect , which is necessary in the operation of stnos , is also widely used to translate detected magnetic field strength to resistance . specifically , gmr - based sensors have been commercialized for sensing magnetic fields which range in strength from the earth s field up to a couple of hundred oersted @xcite . in addition to this range , larger magnetic fields can also be measured by stnos , by virtue of measuring the change in their oscillation frequency caused by the difference between the bias field and the magnetic field being tested . this stno - based magnetic field sensor can provide high sensitivity up to 180 ghz / t @xcite . moreover , the output of the stno is the frequency or frequency change , rather than the signal amplitude , thus relaxing the requirement for the system to have a large signal - to - noise ratio . for this reason , stnos are very attractive for sensing large magnetic fields . the concept of employing stnos to generate localized ac magnetic fields for microwave - assisted magnetic recording ( mamr ) applications has been proposed in @xcite , theoretically analyzed in @xcite , and extensively studied in @xcite . toshiba recently demonstrated a new technology that uses the localized ac magnetic field to reverse the magnetization direction , allowing the excitement of a magnetization oscillation in only one specific magnetic layer of a multilayer magnetic recording system @xcite . this technology could enable the development of 3-d magnetic recording systems , where independent data are written and read from overlapping layers of a multilayer recording medium . the applications to which stnos are suited are not limited to the above suggestions . for instance , stnos have been found very useful as high - density massively parallel microwave signal processors and small phased array transmitters @xcite . it has also been demonstrated in @xcite that stnos can be employed to develop novel readers with high data transfer rates for hard disk drives ( hdds ) . other applications that would make full use of the nanoscale aspects and of the high - speed broad tunability of stnos need further exploration . with the recent emergence of the spin hall effect , a new type of spintronic oscillator , the shno , has been developed @xcite . this provides a new route for the development of microwave and magnonic devices @xcite . shnos exhibit several advantages over stnos , including easier nanofabrication , lower required dc current , direct optical access to the magnetodynamically active area , smaller radiation losses , and suppressed nonlinear damping process @xcite . however , the linewidth of shnos is on the same order as that of stnos , and their output power is currently lower than that of stnos . despite the issues with shnos , their advantages offer an opportunity to implement novel nanoscale microwave sources and emitters for wireless communications and magnonics applications @xcite . in addition to the need to improve stno and shno performance , stno electrical models , stno and shno integration into integrated circuits ( ic ) , and dedicated circuitry are also necessary for the development of applications based on stnos and shnos . electrical models of stnos and shnos are critical , since they can efficiently reduce the time , cost , and risk of errors involved in building prototypes of the entire system of the stno or shno with the ic . integration of the stno or shno with the ic is important , since it should enable better performance , miniaturization , and even single - chip solutions . dedicated circuitry is essential for stno / shno - based applications , since it can perform necessary functions such as biasing and amplification , and can extrinsically improve the performance of stnos or shnos . the issues , challenges , and development of the electrical models , integration , and dedicated circuitry are discussed in the following . the basic procedure for modeling stnos or shnos involves solving the landau - lifshitz - gilbert equation with a slonczewski spin - transfer term ( llgs ) . one possible approach to solving the llgs is based on micromagnetics @xcite . such a model presents the magnetization dynamics of stnos and shnos on a microscopic scale . it typically considers the anisotropy energy , the zeeman energy of the external field , the ferromagnetic exchange energy and the magnetostatic energy @xcite . however , the micromagnetic simulation typically takes days , which is not efficient for the development of stno / shno - based applications . additionally , the output dataset of this model can not be directly used in a circuit simulator . the complicated numeric computation involved in the micromagnetics - based model prevents it from being implemented in a hardware description language . a hardware description language is a specialized computer language used to describe both structural and behavioral aspects of designs including , but not limited to , electronic components , circuits , and systems @xcite . it allows the automatic analysis , simulation , and evaluation of designs , through the use of electronic design automation ( eda ) tools . the micromagnetics - based model is therefore not suited to evaluating stnos and shnos under different bias conditions together with other circuits , as would be required in stno / shno - based applications . an alternative approach to solving the llgs is based on the macrospin approximation , which assumes that only a spatially uniform magnetization precession is excited and that the spin - polarized current is uniform across the area of the free layer @xcite . a macrospin - based model is able to provide an analytical solution with acceptable accuracy , while allowing implementation in a hardware description language . as a result , a macrospin - based analytical model is needed to design and develop stno - based systems . several macrospin - based models of stnos @xcite have been developed and implemented in verilog - a , a commonly used hardware description language for analog systems . however , the models in @xcite require the aid of matlab to compute the effective magnetic field , which is then manually imported into the verilog - a model . consequently , they can not be considered fully implemented in verilog - a . moreover , these models @xcite use equations that are not fully validated . additionally , only one specific device has been used to verify the models , which is not sufficient . a comprehensive mtj stno model that can overcome the issues in @xcite , has been presented in @xcite . this further allows accurate replication of the phase noise and hence of the generated signal . moreover , this model can be used for various mtj stnos to estimate their performance together with the ics . nonetheless , this model is not suitable for simulations that involve momentary variations in the operating conditions , such as current or field modulation . to enable comprehensive analysis and evaluation of stnos or shnos together with ics , future models need to be divergence - free , while allowing all the necessary simulations and analyses . future models also need to include temperature dependence . to avoid the losses due to external components and connections , as well as to eliminate wave reflections , stnos and shnos need to be integrated with other technologies , such as silicon - cmos and gallium - arsenide ( gaas ) , which allow the circuits required for stnos and shnos to be implemented . additionally , integration is vital if the nanoscale of stnos and shnos is to be capitalized upon and to achieve compact stno / shno - based systems . the integration approach must be considered before implementing the circuits and developing the system , since it has a significant impact on the system topology and circuit design . the advantages and disadvantages of different possible integration approaches are discussed in @xcite . the wire - bonding approach has been successfully used in @xcite , mainly thanks to its independent preintegration characterization and optimization of the ic and stno . in @xcite , an mtj stno has been integrated with a dedicated cmos ic amplifier and an off - chip bias - tee using wire bonding . however , the use of the off - chip bias - tee limits both the integration level and the bandwidth of the system . the cmos ic proposed in @xcite includes an on - chip bias - tee and has been successfully integrated with an stno by bonding wires . the bonding wire has also been used as part of the bias - tee design , since it introduces extra high - frequency impedance into the ac decoupling path of the bias - tee . furthermore , the parasitics of the bonding wires in the signal path , as well as the on - chip bias - tee , determine the requirements of the input and output filters for achieving a flat frequency response over a large bandwidth . the wire - bonding approach is appropriate for the current level of maturity of stno technology , as it allows easy implementation , measurement , and replacement . monolithic and flip - chip - based approaches can ultimately result in a higher level of integration and better performance at high frequencies . the monolithic integration approach has been successfully used in hybrid cmos / mtj logic circuits ( fig . [ sto_ic_integration](a ) ) targeting the st - mram technology @xcite , which utilizes the same stt effect as that in stnos and shnos . for instance , a fully functional 64 mb st - mram built on a standard 90 nm process has been developed in @xcite . nevertheless , monolithic integration has not yet been implemented so as to handle the rf signals generated by stnos or shnos . in order to use monolithic integration , the top - layer surface roughness of the ic stacks needs to be very well controlled , as this layer acts as the cu buffer layer for fabricating stnos and shnos . the roughness of this layer has significant impact on the spatial variation in the multilayer structure , and hence on the dynamics of stnos and shnos @xcite . an rms roughness lower than @xmath27 is necessary to avoid unpredictable device performance and significant performance degradation @xcite . flip - chip - based integration is one of the principal methods of interconnecting ics @xcite . as the cost of flip - chip - based integration decreases @xcite , its application space broadens further . [ sto_ic_integration](b ) shows a possible integration between an stno / shno and a cmos ic using the flip - chip - based approach . compared to monolithic integration , flip - chip - based integration allows easier implementation and measurement and retains a low - loss interconnection @xcite . however , neither monolithic nor flip - chip - based integration will be suitable until the device - to - device performance has improved to the extent that no major adjustments ( such as to the topology ) of the circuitry are required for different devices . the stno / shno - based applications described above could be possibly achieved by integrating stnos or shnos into the existing system architectures and by employing the existing circuits implemented in other technologies @xcite . for instance , a system architecture similar to the simple noncoherent transceiver in @xcite has been used to develop an stno - based wireless communication system @xcite . an amplifier using the existing topology of a cascode differential pair with a resistive load has been implemented in @xcite to provide amplification for the stno . additionally , considering the device - to - device performance variation of stnos and shnos , dedicated circuitry may need to allow adjustable performance so as to avoid multiple designs for different stnos or shnos . for instance , the variable gain amplifier ( vga ) @xcite and the tunable lna @xcite , whose gain and bandwidth , respectively , can be tuned to some extent , may be suitable for providing adjustable gain and bandwidth for different stnos or shnos . however , several dedicated circuits need to be designed for stnos and shnos . the first important circuit is the bias - tee , which is necessary for applying the required dc bias current into stnos and shnos without disturbing the rf signals generated . current mirrors have been considered in @xcite for biasing stnos . however , the resistance ( biasing voltage ) of the stno changes as the angle of the magnetic field is varied , so the current mirror can not accurately copy the current from a current source to the stno under all circumstances . accordingly , an on - chip bias - tee that can provide accurate current for stnos and shnos is necessary . factors that also need to be taken into account during the design of the bias - tee include the maximum applied dc current , the applied integration approach , the resistance of stnos and shnos , the power supply voltage of ic technology , the clamp circuit for providing esd protection , and the parasitics . a dedicated sample bias - tee design is given in @xcite . this design considers the above factors and takes care of the trade - offs involved in selecting the number of turns , the separation distance between each turn , and the width of metalization needed for the inductor to act as the ac decoupling component . to implement the dedicated on - chip bias - tee , electromagnetic simulations may be required when custom inductors are necessary @xcite . the second important circuit for stnos and shnos is the amplifier , which needs to provide sufficient gain and bandwidth for a given application while featuring low noise and proper input impedance . for example , to use stnos or shnos as los , the amplifier should typically enhance the output power of the stnos and shnos to 0 dbm so as to drive an rf mixer to perform frequency conversion . to take full advantage of stnos or shnos in multiband multistandard radios , the bandwidth of the amplifier needs to cover multiple bands of interest . additionally , the amplifier must introduce as little noise as possible to the entire system . the input impedance of the amplifier needs to match the impedance of stnos or shnos for maximizing the transferred power . nevertheless , for stnos or shnos with a small resistance ( a few ohms ) , wideband impedance matching to such a resistance is not feasible . this is because any parasitic will introduce an impedance mismatch much larger than a few ohms . in this case , rather than maximizing the transferred power , an amplifier with large input impedance is preferable , since it can maximize the ac voltage delivered to the amplifier as well as the signal - to - noise ratio at the output @xcite . other important circuits include those that handle the phase noise of stnos and shnos , such as phase - locked loops ( plls ) and injection - lock ring oscillators ( ilros ) for frequency synthesis . this is due to the fact that the frequency fluctuation of stnos occurs every several nanoseconds or even few tens of nanoseconds @xcite . the fast frequency change requires a rapid lock time ( or settling time ) of at most a few nanoseconds , which is difficult to achieve in either plls or ilros . recently , the phase of an mtj stno , which generates a 5.12 ghz microwave signal , has been successfully locked by a pll to that of an 80 mhz reference clock , resulting in a significant increase in q from 800 to 5@xmath28 @xcite . this is the first demonstration of phase synchronization of stno output to a lower frequency external signal , and it may be a milestone toward practical applications of stnos . state - of - the - art stnos and shnos feature very wide tunability , high levels of integration with cmos , high operating frequencies , fast switching speeds , and miniature size . their unique blend of features suggests a wide range of applications , including microwave sources , microwave detectors , noncoherent transceivers , magnonics , neuromorphic computing , magnetic field sensing , and magnetic field generation . over the past decade , extensive research has taken place to understand and further exploit the nonlinear physical phenomena of stnos and shnos . in this paper , we have introduced the fundamentals , geometries , and functional properties of stnos and shnos . we have additionally reviewed the state of the art in stno and shno technologies . furthermore , we have presented the potential for these oscillators in their popular applications . however , challenges still remain on both the device and system level . these need to be overcome before the use of stnos and shnos can become widespread . this work is partially supported by the swedish research council ( vr ) , the swedish foundation for strategic research ( ssf ) , the knut and alice wallenberg foundation ( kaw ) , the european commission fp7-ict-2011-contract no . 317950 `` mosaic '' , and the european research council ( erc ) under the european community s seventh framework programme ( fp/2007 - 2013)/erc grant 307144 `` mustang '' . m. n. baibich , j. m. broto , a. fert , f. n. van dau , f. petroff , p. etienne , g. creuzet , a. friederich , and j. chazelas , `` giant magnetoresistance of ( 001)fe/(001)cr magnetic superlattices , '' _ phys . _ , vol . 61 , no . 21 , p. 2472 , 1988 . j. a. katine , f. j. albert , r. a. buhrman , e. b. myers , and d. c. ralph , `` current - driven magnetization reversal and spin - wave excitations in co @xmath29cu @xmath29co pillars , '' _ phys . _ , vol . 84 , pp . 31493152 , apr 2000 . j. wunderlich , b. kaestner , j. sinova , and t. jungwirth , `` experimental observation of the spin - hall effect in a two - dimensional spin - orbit coupled semiconductor system , '' _ phys . rev . _ , vol . 94 , no . 4 , pp . 14 , 2005 . i. m. miron , k. garello , g. gaudin , p .- j . zermatten , m. v. costache , s. auffret , s. bandiera , b. rodmacq , a. schuhl , and p. gambardella , `` perpendicular switching of a single ferromagnetic layer induced by in - plane current injection , '' _ nature _ , vol . 7359 , pp . 189193 , jul . 2011 . l. liu , c .- f . pai , d. c. ralph , and r. a. buhrman , `` magnetic oscillations driven by the spin hall effect in 3-terminal magnetic tunnel junction devices , '' _ phys 109 , no . 18 , p. 186602 , oct . 2012 . v. e. demidov , s. urazhdin , h. ulrichs , v. tiberkevich , a. slavin , d. baither , g. schmitz , and s. o. demokritov , `` magnetic nano - oscillator driven by pure spin current , '' _ nature mater . _ , vol . 11 , no . 12 , pp 10281031 , dec . 2012 . b. n. engel , j. kerman , b. butcher , r. w. dave , m. deherrera , m. durlam , g. grynkewich , j. janesky , s. v. pietambaram , n. d. rizzo , j. m. slaughter , k. smith , j. j. sun , and s. tehrani , `` a 4-mb toggle mram based on a novel bit and switching method , '' _ ieee trans . _ , vol . 41 , no . 1 , 132136 , jan . 2005 . r. l. stamps , s. breitkreutz , j. kerman , a. v. chumak , y. otani , g. e. w. bauer , j .- u . thiele , m. bowen , s. a. majetich , m. klui , i. l. prejbeanu , b. dieny , n. m. dempsey , and b. hillebrands , `` the 2014 magnetism roadmap , '' _ j. phys . _ , vol . 47 , no . 33 , p. 333001 , 2014 . m. zwierzycki , y. tserkovnyak , p. j. kelly , a. brataas , and g. e. w. bauer , `` first - principles study of magnetization relaxation enhancement and spin transfer in thin magnetic films , '' _ phys . b _ , vol . 71 , p. 064420 , feb 2005 . r. matsumoto , a. chanthbouala , j. grollier , v. cros , a. fert , k. nishimura , y. nagamine , h. maehara , k. tsunekawa , a. fukushima , and s. yuasa , `` spin - torque diode measurements of mgo - based magnetic tunnel junctions with asymmetric electrodes , '' _ appl . , vol . 4 , no . 6 , p. 063001 , 2011 . m. tsoi , a. g. jansen , j. bass , w. c. chiang , v. tsoi , and p. wyder , `` generation and detection of phase - coherent current - driven magnons in magnetic multilayers , '' _ nature _ , vol . 6791 , pp . 4648 , jul . 2000 . w. h. rippard , m. r. pufall , and t. j. silva , `` quantitative studies of spin - momentum - transfer - induced excitations in co / cu multilayer films using point - contact spectroscopy , '' _ appl . phys _ , vol . 82 , no . 8 , p. 1260 , 2003 . m. r. pufall , w. h. rippard , and t. j. silva , `` materials dependence of the spin - momentum transfer efficiency and critical current in ferromagnetic metal / cu multilayers , '' _ appl . _ , vol . 83 , no . 2 , pp . 323325 , 2003 . w. h. rippard , m. r. pufall , s. kaka , s. e. russek , and t. j. silva , `` direct - current induced dynamics in co@xmath9fe@xmath10/ni@xmath16fe@xmath14 point contacts , '' _ phys . _ , vol . 92 , p. 027201 , jan 2004 . r. k. dumas , s. r. sani , s. m. mohseni , e. iacocca , y. pogoryelov , p. k. muduli , s. chung , p. drrenfeld , and j. kerman , `` recent advances in nanocontact spin - torque oscillators , '' _ ieee trans . _ , vol . 50 , no . 6 , p. 4100107 , jun . 2014 . s. sani , j. persson , s. mohseni , y. pogoryelov , p. muduli , a. eklund , g. malm , m. kll , a. dmitriev , and j. kerman , `` mutually synchronized bottom - up multi - nanocontact spin - torque oscillators , '' _ nat . _ , vol . 4 , p. 2731 , nov . 2013 . s. i. kiselev , j. c. sankey , i. n. krivorotov , n. c. emley , r. j. schoelkopf , r. a. buhrman , and d. c. ralph , `` microwave oscillations of a nanomagnet driven by a spin - polarized current , '' _ nature _ , vol . 425 , no . 6956 , pp . 380383 , sep . 2003 . wegrowe , a. fabian , p. guittienne , x. hoffer , d. kelly , j ansermet , and e. olive , `` exchange torque and spin transfer between spin polarized current and ferromagnetic layers , '' _ appl . _ , vol . 80 , no . 37753777 , 2002 . f. abreu araujo , m. darques , k. a. zvezdin , a. v. khvalkovskiy , n. locatelli , k. bouzehouane , v. cros , and l. piraux , `` microwave signal emission in spin - torque vortex oscillators in metallic nanowires : experimental measurements and micromagnetic numerical study , '' _ phys . b _ , vol . 86 , p. 064424 , aug 2012 . f. abreu araujo , l. piraux , v. a. antohe , v. cros , and l. gence , `` single spin - torque vortex oscillator using combined bottom - up approach and e - beam lithography , '' _ appl . 102 , no . 22 , p. 222402 , 2013 . w. h. rippard , m. r. pufall , and s. e. russek , `` comparison of frequency , linewidth , and output power in measurements of spin - transfer nanocontact oscillators , '' _ phys . b _ , vol . 74 , p. 224409 , dec 2006 . j. z. sun , m. c. gaidis , e. j. osullivan , e. a. joseph , g. hu , d. w. abraham , j. j. nowak , p. l. trouilloud , y. lu , s. l. brown , d. c. worledge , and w. j. gallagher , `` a three - terminal spin - torque - driven magnetic switch , '' _ appl . _ , vol . 95 , no . 8 , p. 083506 , 2009 . l. xue , c. wang , y .- t . cui , l. liu , a. swander , j. z. sun , r. a. buhrman , and d. c. ralph , `` resonance measurement of nonlocal spin torque in a three - terminal magnetic device , '' _ phys . 108 , p. 147201 , apr 2012 . v. e. demidov , s. urazhdin , a. zholud , a. v. sadovnikov , a. n. slavin , and s. o. demokritov , `` spin - current nano - oscillator based on nonlocal spin injection , '' _ sci . rep . _ , vol . 5 , no . 8578 , pp . 15 , 2015 . v. e. demidov , s. urazhdin , h. ulrichs , v. tiberkevich , a. slavin , d. baither , g. schmitz , and s. o. demokritov , `` magnetic nano - oscillator driven by pure spin current , '' _ nature mater . _ , vol . 11 , pp . 10281031 , 2012 . z. duan , a. smith , l. yang , b. youngblood , j. lindner , v. e. demidov , s. o. demokritov , and i. n. krivorotov , `` nanowire spin torque oscillator driven by spin orbit torques , '' _ nat . commun . _ , vol . 5 , p. 5616 a. m. deac , a. fukushima , h. kubota , h. maehara , y. suzuki , s. yuasa , y. nagamine , k. tsunekawa , d. d. djayaprawira , and n. watanabe , `` bias - driven high - power microwave emission from mgo - based tunnel magnetoresistance devices , '' _ nature phys . _ , vol . 4 , no . 10 , 803809 , oct . h. maehara , h. kubota , y. suzuku , t. seki , k. nishimura , y. nagamine , k. tsunekawa , a. fukushima , a. m. deac , and s. yuasa , `` large emission power over 2 @xmath15w with high q factor obtained from nanocontact magnetic - tunnel - junction - based spin torque oscillator , '' _ appl . , vol . 6 , no . 11 , p. 113005 , 2013 . h. maehara , h. kubota , y. suzuki , t. seki , k. nishimura , y. nagamine , k. tsunekawa , a. fukushima , h. arai , t. taniguchi , h. imamura , k. ando , and s. yuasa , `` high q factor over 3000 due to out - of - plane precession in nano - contact spin - torque oscillator based on magnetic tunnel junctions , '' _ appl . 7 , no . 2 , p. 023003 , 2014 . s. mangin , d. ravelosona , j. a. katine , m. j. carey , b. d. terris , and e. e. fullerton , `` current - induced magnetization reversal in nanopillars with perpendicular anisotropy , '' _ nature mater . _ , vol . 5 , no . 3 , pp . 210215 , 2006 . d. houssameddine , u. ebels , b. delaet , b. rodmacq , i. firastrau , f. ponthenier , m. brunet , c. thirion , j .- p . michel , l. prejbeanu - buda , m .- c . cyrille , o. redon , and b. dieny , `` spin - torque oscillator using a perpendicular polarizer and a planar free layer , '' _ nature mater . _ , vol . 6 , no . 6 , pp . 447453 , jun 2007 . w. h. rippard , a. m. deac , m. r. pufall , j. m. shaw , m. w. keller , s. e. russek , g. e. w. bauer , and c. serpico , `` spin - transfer dynamics in spin valves with out - of - plane magnetized coni free layers , '' _ phys . rev . b _ , vol . 81 , p. 014426 , jan 2010 . s. m. mohseni , s. r. sani , j. persson , t. n. anh nguyen , s. chung , y. pogoryelov , and j. kerman , `` high frequency operation of a spin - torque oscillator at low field , '' _ phys . status solidi rapid res . _ , vol . 5 , no . 12 , pp . 432434 , 2011 . s. m. mohseni , s. r. sani , j. persson , t. n. a. nguyen , s. chung , y. pogoryelov , p. k. muduli , e. iacocca , a. eklund , r. k. dumas , s. bonetti , a. deac , m. a. hoefer , and j. kerman , `` spin torque - generated magnetic droplet solitons , '' _ science _ 6125 , pp . 12951298 , mar . s. m. mohseni , s. r. sani , r. k. dumas , j. persson , t. n. anh nguyen , s. chung , y. pogoryelov , p. k. muduli , e. iacocca , a. eklund , and j. kerman , `` magnetic droplet solitons in orthogonal nano - contact spin torque oscillators , '' _ physica b : condensed matter _ , 435 , pp . 84 87 , 2014 . s. chung , s. m. mohseni , s. r. sani , e. iacocca , r. k. dumas , t. n. anh nguyen , y. pogoryelov , p. k. muduli , a. eklund , m. hoefer , and j. kerman , `` spin transfer torque generated magnetic droplet solitons ( invited ) , '' _ j. appl 115 , no . 17 , p. 172612 , may 2014 . y. zhou , s. bonetti , c. l. zha , and j. kerman , `` zero - field precession and hysteretic threshold currents in a spin torque nano device with tilted polarizer , '' _ new j. phys . _ , vol . 11 , no . 10 , p. 103028 , 2009 . t. n. a. nguyen , y. fang , v. fallahi , n. benatmane , s. m. mohseni , r. k. dumas , and j. kerman , `` [ co / pd]-nife exchange springs with tunable magnetization tilt angle , '' _ appl . _ , vol . 98 , no . 17 , p. 172502 , 2011 . t. a. nguyen , n. benatmane , v. fallahi , y. fang , s. mohseni , r. dumas , and j. kerman , `` [ co / pd]4-co - pd - nife spring magnets with highly tunable and uniform magnetization tilt angles , '' _ j. magn . 324 , no . 3929 3932 , 2012 . s. chung , s. m. mohseni , v. fallahi , t. n. anh nguyen , n. benatmane , r. k. dumas , and j. kerman , `` tunable spin configuration in [ co / ni]-nife spring magnets , '' _ j. phys . _ , vol . 46 , no . 12 , p. 125004 , 2013 . s. tacchi , t. n. a. nguyen , g. carlotti , g. gubbiotti , m. madami , r. k. dumas , j. w. lau , j. kerman , a. rettori , and m. g. pini , `` spin wave excitations in exchange - coupled [ co / pd]-nife films with tunable tilting of the magnetization , '' _ phys . b _ , vol . 87 , p. 144426 , apr 2013 . s. tacchi , t. n. a. nguyen , g. gubbiotti , m. madami , g. carlotti , m. g. pini , a. rettori , v. fallahi , r. k. dumas , and j. kerman , `` [ co / pd]-cofeb exchange spring magnets with tunable gap of spin wave excitations , '' _ j. phys . _ , vol . 47 , no . 49 , p. 495004 , 2014 . w. skowroski , t. stobiecki , j. wrona , g. reiss , and s. van dijken , `` zero - field spin torque oscillator based on magnetic tunnel junctions with a tilted cofeb free layer , '' _ appl . , vol . 5 , no . 6 , p. 063005 , 2012 . s. sani , p. drrenfeld , s. mohseni , s. chung , and j. kerman , `` microwave signal generation in single - layer nano - contact spin torque oscillators , '' _ ieee trans . _ , vol . 49 , no . 7 , pp . 43314334 , july 2013 . v. s. pribiag , i. n. krivorotov , g. d. fuchs , p. m. braganca , o. ozatay , j. c. sankey , d. c. ralph , and r. a. buhrman , `` magnetic vortex oscillator driven by d.c . spin - polarized current , '' _ nature phys . _ , vol . 3 , no . 7 , pp . 498503 , jul . 2007 . v. sluka , a. kkay , a. m. deac , d. e. brgler , r. hertel , and c. m. schneider , `` spin - transfer torque induced vortex dynamics in fe / ag / fe nanopillars , '' _ j. phys . _ , vol . 44 , no . 38 , p. 384002 , 2011 . n. locatelli , v. v. naletov , j. grollier , g. de loubens , v. cros , c. deranlot , c. ulysse , g. faini , o. klein , and a. fert , `` dynamics of two coupled vortices in a spin valve nanopillar excited by spin transfer torque , '' _ appl . _ , vol . 98 , no . 6 , pp . , a. v. khvalkovskiy , j. grollier , a. dussaux , k. a. zvezdin , and v. cros , `` vortex oscillations induced by spin - polarized current in a magnetic nanopillar : analytical versus micromagnetic calculations , '' _ phys . b _ , vol . 80 , p. 140401 , oct 2009 . p. bortolotti , e. grimaldi , a. dussaux , j. grollier , v. cros , c. serpico , k. yakushiji , a. fukushima , h. kubota , r. matsumoto , and s. yuasa , `` parametric excitation of magnetic vortex gyrations in spin - torque nano - oscillators , '' _ phys . b _ , vol . 88 , p. 174417 , nov 2013 . e. grimaldi , a. dussaux , p. bortolotti , j. grollier , g. pillet , a. fukushima , h. kubota , k. yakushiji , s. yuasa , and v. cros , `` response to noise of a vortex based spin transfer nano - oscillator , '' _ phys . b _ , vol . 89 , p. 104404 , mar 2014 . r. lehndorff , d. e. brgler , s. gliga , r. hertel , p. grnberg , c. m. schneider , and z. celinski , `` magnetization dynamics in spin torque nano - oscillators : vortex state versus uniform state , '' _ phys . b _ , vol . 80 , p. 054412 , aug 2009 . g. consolo , b. azzerboni , g. gerhart , g. a. melkov , v. tiberkevich , and a. n. slavin , `` excitation of self - localized spin - wave bullets by spin - polarized current in in - plane magnetized magnetic nanocontacts : a micromagnetic study , '' _ phys . rev . b _ , vol . 76 , p. 144410 , oct 2007 . s. bonetti , v. tiberkevich , g. consolo , g. finocchio , p. muduli , f. mancoff , a. slavin , and j. kerman , `` experimental evidence of self - localized and propagating spin wave modes in obliquely magnetized current - driven nanocontacts , '' _ phys . 105 , p. 217204 , nov 2010 . m. madami , s. bonetti , g. consolo , s. tacchi , g. carlotti , g. gubbiotti , f. b. mancoff , m. a. yar , and j. kerman , `` direct observation of a propagating spin wave induced by spin - transfer torque , '' _ nature nanotech . _ , vol . 6 , p. 635 , august 2011 . s. bonetti , v. puliafito , g. consolo , v. s. tiberkevich , a. n. slavin , and j. kerman , `` power and linewidth of propagating and localized modes in nanocontact spin - torque oscillators , '' _ phys . b _ , vol . 85 , p. 174427 , may 2012 . n. c. krivorotov , i. n. andemley , j. c. sankey , s. i. kiselev , d. c. ralph , and r. a. buhrman , `` time - domain measurements of nanomagnet dynamics driven by spin - transfer torques , '' _ science _ , vol . 5707 , pp . 228231 , 2005 . r. k. dumas , e. iacocca , s. bonetti , s. r. sani , s. m. mohseni , a. eklund , j. persson , o. heinonen , and j. kerman , `` spin - wave - mode coexistence on the nanoscale : a consequence of the oersted - field - induced asymmetric energy landscape , '' _ phys . rev . lett . 110 , p. 257202 , jun 2013 . m. madami , e. iacocca , s. sani , g. gubbiotti , s. tacchi , r. k. dumas , j. kerman , and g. carlotti , `` propagating spin waves excited by spin - transfer torque : a combined electrical and optical study , '' _ phys . b _ , vol . 92 , p. 024403 , jul 2015 . q. mistral , m. van kampen , g. hrkac , j .- v . kim , t. devolder , p. crozat , c. chappert , l. lagae , and t. schrefl , `` current - driven vortex oscillations in metallic nanocontacts , '' _ phys . 100 , p. 257201 , jun 2008 . a. ruotolo , v. cros , b. georges , a. dussaux , j. grollier , c. deranlot , r. guillemet , k. bouzehouane , s. fusil , and a. fert , `` phase - locking of magnetic vortices mediated by antivortices , '' _ nature nanotech . _ , vol . 4 , no . 8 , 528532 , aug . m. manfrini , j .- v . kim , s. petit - watelot , w. van roy , l. lagae , c. chappert , and t. devolder , `` propagation of magnetic vortices using nanocontacts as tunable attractors , '' _ nature nanotech . _ , vol . 9 , no . 20 , pp . 121125 , feb 2014 , letter . s. petit - watelot , j .- v . kim , a. ruotolo , r. m. otxoa , k. bouzehouane , j. grollier , a. vansteenkiste , b. van de wiele , v. cros , and t. devolder , `` commensurability and chaos in magnetic vortex oscillations , '' _ nature phys . _ , vol . 8 , no . 9 , pp . 682687 , sep . 2012 . v. e. demidov , s. urazhdin , e. r. j. edwards , m. d. stiles , r. d. mcmichael , and s. o. demokritov , `` control of magnetic fluctuations by spin current , '' _ phys . 107 , p. 107204 , sep 2011 . m. ranjbar , p. drrenfeld , m. haidar , e. iacocca , m. balinskiy , t. q. le , m. fazlali , a. houshang , a. a. awad , r. k. dumas , and j. kerman , `` cofeb - based spin hall nano - oscillators , '' _ ieee magn . _ , vol . 5 , p. 3000504 , 2014 . m. collet , x. de milly , o. dallivy kelly , v. v. naletov , r. bernard , p. bortolotti , v. e. demidov , s. o. demokritov , j. l. prieto , m. mu noz , v. cros , a. anane , g. de loubens , and o. klein , `` generation of coherent spin - wave modes in yttrium iron garnet microdiscs by spin - orbit torque , '' _ arxiv:1504.01512 [ cond-mat.mtrl-sci]_ , 2015 . g. gerhart , e. bankowski , g. a. melkov , v. s. tiberkevich , and a. n. slavin , `` angular dependence of the microwave - generation threshold in a nanoscale spin - torque oscillator , '' _ phys . b _ , vol . 76 , no . 2 , p. 024437 2007 . f. abreu araujo , a. d. belanovsky , p. n. skirdkov , k. a. zvezdin , a. k. zvezdin , n. locatelli , r. lebrun , j. grollier , v. cros , g. de loubens , and o. klein , `` optimizing magnetodipolar interactions for synchronizing vortex based spin - torque nano - oscillators , '' _ phys . b _ , vol . 92 , p. 045419 , jul 2015 . s. kaka , m. r. pufall , w. h. rippard , t. j. silva , s. e. russek , and j. a. katine , `` mutual phase - locking of microwave spin torque nano - oscillators , '' _ nature _ , vol . 7057 , pp . 389392 , sep . 2005 . m. r. pufall , w. h. rippard , s. e. russek , s. kaka , and j. a. katine , `` electrical measurement of spin - wave interactions of proximate spin transfer nanooscillators , '' _ phys . _ , vol . 97 , p. 087206 , aug 2006 . t. kendziorczyk , s. o. demokritov , and t. kuhn , `` spin - wave - mediated mutual synchronization of spin - torque nano - oscillators : a micromagnetic study of multistable phase locking , '' _ phys . b _ , vol . 90 , p. 054414 , aug 2014 . a. houshang , p. drrenfeld , e. iacocca , s. r. sani , j. kerman , and r. k. dumas , `` spin wave beam driven synchronisation of multiple nanocontact spin torque oscillators , '' _ nature nanotech . _ , under review . m. hoefer , m. ablowitz , b. ilan , m. pufall , and t. silva , `` theory of magnetodynamics induced by spin torque in perpendicularly magnetized thin films , '' _ phys . _ , vol . 95 , no . 26 , p. 267206 s. bonetti , p. muduli , f. mancoff , and j. kerman , `` spin torque oscillator frequency versus magnetic field angle : the prospect of operation beyond 65 ghz , '' _ appl . _ , vol . 94 , no . 10 , p. 102507 , 2009 . k. muduli , y. pogoryelov , y. zhou , f. mancoff , and j. kerman , `` spin torque oscillators and rf currents - modulation , locking , and ringing , '' _ integr . 125 , no . 1 , pp . 147154 , 2011 . j. hasch , e. topak , r. schnabel , t. zwick , r. weigel , and c. waldschmidt , `` millimeter - wave technology for automotive radar sensors in the 77 ghz frequency band , '' _ ieee trans . theory tech . _ , vol . 60 , no . 3 , pp . 845860 , march 2012 . s. tsunegi , h. kubota , k. yakushiji , m. konoto , s. tamaru , a. fukushima , h. arai , h. imamura , e. grimaldi , r. lebrun , j. grollier , v. cros , and s. yuasa , `` high emission power and q factor in spin torque vortex oscillator consisting of feb free layer , '' _ appl . , vol . 7 , no . 6 , p. 063009 , 2014 . k. muduli , y. pogoryelov , s. bonetti , g. consolo , f. mancoff , and j. kerman , `` nonlinear frequency and amplitude modulation of a nanocontact - based spin - torque oscillator , '' _ phys . b _ , vol . 81 , no . 14 , p. 140408 k. muduli , y. pogoryelov , f. mancoff , and j. kerman , `` modulation of individual and mutually synchronized nanocontact - based spin torque oscillators , '' _ ieee trans . _ , vol . 47 , pp . 15751579 , jun . 2011 . p. drrenfeld , e. iacocca , j. kerman , and p. k. muduli , `` modulation - mediated unlocking of a parametrically phase - locked spin torque oscillator , '' _ appl . 105 , no . 25 , p. 252404 , 2014 . m. manfrini , t. devolder , j .- v . kim , p. crozat , n. zerounian , c. chappert , w. van roy , l. lagae , g. hrkac , and t. schrefl , `` agility of vortex - based nanocontact spin torque oscillators , '' _ appl . _ , vol . 95 , no . 19 , p. 192507 , nov . 2009 . m. quinsat , f. garcia - sanchez , a. s. jenkins , v. s. tiberkevich , a. n. slavin , l. d. buda - prejbeanu , a. zeltser , j. a. katine , b. dieny , m .- c . cyrille , and u. ebels , `` modulation bandwidth of spin torque oscillators under current modulation , '' _ appl . 105 , no . 15 , p. 152401 , 2014 . e. iacocca and j. kerman , `` analytical investigation of modulated spin - torque oscillators in the framework of coupled differential equations with variable coefficients , '' _ phys . b _ , vol . 85 , p. 184420 , may 2012 . d. houssameddine , u. ebels , b. dieny , k. garello , j .- p . michel , b. delaet , b. viala , m .- c . cyrille , j. a. katine , and d. mauri , `` temporal coherence of mgo based magnetic tunnel junction spin torque oscillators , '' _ phys . 102 , no . 25 , p. 257202 , jun . 2009 . s. bonetti , v. puliafito , g. consolo , v. tiberkevich , a. slavin , and j. kerman , `` power and linewidth of propagating and localized modes in nanocontact spin - torque oscillators , '' _ phys . b _ , vol . 85 , no . 17 , pp . 18 , may 2012 . a. eklund , s. sani , s. mohseni , j. persson , b. malm , and j. @xmath30kerman , `` triple mode - jumping in a spin torque oscillator , '' in _ noise and fluctuations ( icnf ) , 2013 22nd international conference on _ , june 2013 , pp . m. r. pufall , w. h. rippard , s. e. russek , and e. r. evarts , `` anisotropic frequency response of spin - torque oscillators with applied field polarity and direction , '' _ phys . b _ , vol . 86 , p. 094404 , sep 2012 . k. muduli , o. g. heinonen , and j. kerman , `` temperature dependence of linewidth in nanocontact based spin torque oscillators : effect of multiple oscillatory modes , '' _ phys . b _ , vol . 86 , p. 174408 , nov 2012 . j. c. sankey , i. n. krivorotov , s. i. kiselev , p. m. braganca , n. c. emley , r. a. buhrman , and d. c. ralph , `` mechanisms limiting the coherence time of spontaneous magnetic oscillations driven by dc spin - polarized currents , '' _ phys . b _ , vol . 72 , p. 224427 , dec 2005 . d. v. berkov , `` magnetization dynamics including thermal fluctuations : basic phenomenology , fast remagnetization processes and transitions over high - energy barriers , '' in _ handbook of magnetism and advanced magnetic materials _ , h. kronmller and s. parkin , eds.1em plus 0.5em minus 0.4emjohn wiley & sons , ltd , 2007 , vol . 3 . l. bianchini , s. cornelissen , j .- v . kim , t. devolder , w. van roy , l. lagae , and c. chappert , `` direct experimental measurement of phase - amplitude coupling in spin torque oscillators , '' _ appl . _ , vol . 97 , no . 3 , p. 032502 , 2010 . kim , v. tiberkevich , and a. n. slavin , `` generation linewidth of an auto - oscillator with a nonlinear frequency shift : spin - torque nano - oscillator , '' _ phys . 100 , p. 017207 , jan 2008 . b. georges , j. grollier , v. cros , a. fert , a. fukushima , h. kubota , k. yakushijin , s. yuasa , and k. ando , `` origin of the spectral linewidth in nonlinear spin - transfer oscillators based on mgo tunnel junctions , '' _ phys . b _ , vol . 80 , p. 060404 , aug 2009 . m. quinsat , d. gusakova , j. f. sierra , j. p. michel , d. houssameddine , b. delaet , m .- c . cyrille , u. ebels , b. dieny , l. d. buda - prejbeanu , j. a. katine , d. mauri , a. zeltser , m. prigent , j .- c . nallatamby , and r. sommet , `` amplitude and phase noise of magnetic tunnel junction oscillators , '' _ appl . _ , vol . 97 , no . 18 , p. 182507 , 2010 . j. f. sierra , m. quinsat , f. garcia - sanchez , u. ebels , i. joumard , a. s. jenkins , b. dieny , m .- c . cyrille , a. zeltser , and j. a. katine , `` influence of thermal fluctuations on the emission linewidth in mgo - based spin transfer oscillators , '' _ appl . _ , vol . 101 , no . 6 , p. 062407 , 2012 . kim , q. mistral , c. chappert , v. s. tiberkevich , and a. n. slavin , `` line shape distortion in a nonlinear auto - oscillator near generation threshold : application to spin - torque nano - oscillators , '' _ phys . 100 , p. 167201 , apr 2008 . o. j. lee , p. m. braganca , v. s. pribiag , d. c. ralph , and r. a. buhrman , `` quasilinear spin - torque nano - oscillator via enhanced negative feedback of power fluctuations , '' _ phys . b _ , vol . 88 , p. 224411 , dec 2013 . r. sharma , p. drrenfeld , e. iacocca , o. g. heinonen , j. kerman , and p. k. muduli , `` mode - hopping mechanism generating colored noise in a magnetic tunnel junction based spin torque oscillator , '' _ appl . 105 , no . 13 , p. 132404 , 2014 . a. eklund , s. bonetti , s. r. sani , s. m. mohseni , j. persson , s. chung , s. a. h. banuazizi , e. iacocca , m. @xmath31stling , j. @xmath30kerman , and b. g. malm , `` dependence of the colored frequency noise in spin torque oscillators on current and magnetic field , '' _ appl . 104 , no . 9 , p. 092405 , 2014 . w. h. rippard , m. r. pufall , s. kaka , t. j. silva , s. e. russek , and j. a. katine , `` injection locking and phase control of spin transfer nano - oscillators , '' _ phys . _ , vol . 95 , p. 067203 , aug 2005 . j. c. sankey , p. m. braganca , a. g. f. garcia , i. n. krivorotov , r. a. buhrman , and d. c. ralph , `` spin - transfer - driven ferromagnetic resonance of individual nanomagnets , '' _ phys . _ , vol . 96 , p. 227601 , jun 2006 . b. georges , j. grollier , m. darques , v. cros , c. deranlot , b. marcilhac , g. faini , and a. fert , `` coupling efficiency for phase locking of a spin transfer nano - oscillator to a microwave current , '' _ phys . _ , vol . 101 , p. 017201 , jul 2008 . a. dussaux , a. v. khvalkovskiy , j. grollier , v. cros , a. fukushima , m. konoto , h. kubota , k. yakushiji , s. yuasa , k. ando , and a. fert , `` phase locking of vortex based spin transfer oscillators to a microwave current , '' _ appl . _ , vol . 98 , no . 13 , p. 132506 , 2011 . r. lebrun , a. jenkins , a. dussaux , n. locatelli , s. tsunegi , e. grimaldi , h. kubota , p. bortolotti , k. yakushiji , j. grollier , a. fukushima , s. yuasa , and v. cros , `` understanding of phase noise squeezing under fractional synchronization of a nonlinear spin transfer vortex oscillator , '' _ phys . 115 , p. 017201 , jun 2015 . v. e. demidov , h. ulrichs , s. v. gurevich , s. o. demokritov , v. s. tiberkevich , a. n. slavin , a. zholud , and s. urazhdin , `` synchronization of spin hall nano - oscillators to external microwave signals , '' _ nat . _ , vol . 5 , p. 3179 , jan . 2014 . g. khalsa , m. d. stiles , and j. grollier , `` critical current and linewidth reduction in spin - torque nano - oscillators by delayed self - injection , '' _ appl . 106 , no . 24 , p. 242402 , 2015 . s. tsunegi , e. grimaldi , r. lebrun , h. kubota , a. s. jenkins , k. yakushiji , a. fukushima , p. bortolotti , j. grollier , s. yuasa , and v. cros , `` self - injection locking of a vortex spin torque oscillator by delayed feedback , '' _ arxiv:1509.05583v1 _ , 2015 . s. tamaru , k. yakushiji , b. wang , a. fukushima , and h. kubota , `` generation of highly stable 5 ghz microwave from a spin torque oscillator by phase locked loop referenced to a 80 mhz clock , '' in _ magnetics conference ( intermag ) , 2015 ieee _ , may 2015 , pp . 11 . p. villard , u. ebels , d. houssameddine , j. katine , d. mauri , b. delaet , p. vincent , m. c. cyrille , b. viala , j. p. michel , j. prouve , and f. badets , `` a ghz spintronic - based rf oscillator , '' _ ieee j. solid - state circuits _ , vol . 45 , no . 1 , pp . 214223 , 2010 . t. chen , p. drrenfeld , s. rodriguez , j. kerman , and a. rusu , `` a highly tunable microwave oscillator based on mtj sto technology , '' _ microw . _ , vol . 56 , no . 9 , pp . 20922095 , 2014 . k. muduli , o. g. heinonen , and j. kerman , `` bias dependence of perpendicular spin torque and of free - and fixed - layer eigenmodes in mgo - based nanopillars , '' _ phys . b _ , vol . 83 , no . 18 , p. 184410 , 2011 . s. mizukami , f. wu , a. sakuma , j. walowski , d. watanabe , t. kubota , x. zhang , h. naganuma , m. oogane , y. ando , and t. miyazaki , `` long - lived ultrafast spin precession in manganese alloys films with a large perpendicular magnetic anisotropy , '' _ phys . 106 , no . 11 , p. 117201 , 2011 . d. houssameddine , s. h. florez , j. a. katine , j. p. michel , u. ebels , o. mauri , d. ozatay , b. delaet , b. viala , l. folks , b. d. terris , and m. c. cyrille , `` spin transfer induced coherent microwave emission with large power from nanoscale mgo tunnel junctions , '' _ appl . _ , vol . 93 , no . 2 , p. 022505 , 2008 . k. mufuli , s. bonetti , y. zhou , j. @xmath30kerman , and o. g. heinonen , `` mgo - based spin torque oscillators with above 0.1 @xmath15w output power , '' in _ ieee international magnetics conference _ , 2009 . h. kubota , k. yakushiji , a. fukushima , s. tamaru , m. konoto , t. nozaki , s. ishibashi , t. saruya , , s. yuasa , t. taniguchi , h. arai , and h. imamura , `` spin - torque oscillator based on magnetic tunnel junction with a perpendicularly magnetized free layer and in - plane magnetized polarizer , '' _ appl . , vol . 6 , no . 10 , p. 103003 , 2013 . p. drrenfeld , s. m. mohseni , f. gerhard , m. ranjbar , f. qejvanaj , s. r. sani , s. chung , c. gould , l. w. molenkamp , and j. @xmath30kerman , `` low- current , narrow - linewidth microwave signal generation in single - layer - nimnsb based nano - contact spin torque oscillators , '' _ unpublished manuscript _ , 2015 . w. h. rippard , m. r. pufall , s. kaka , t. j. silva , and s. e. russek , `` current - driven microwave dynamics in magnetic point contacts as a function of applied field angle , '' _ phys . b _ , vol . 70 , no . 10 , p. 100406 , 2004 . q. mistral , j .- v . kim , t. devolder , p. crozat , c. chappert , j. a. katine , m. j. carey , and k. ito , `` current - driven microwave oscillations in current perpendicular - to - plane spin - valve nanopillars , '' _ appl . _ , vol . 88 , no . 19 , p. 192507 , 2006 . j. sinha , m. hayashi , y. k. takahashi , t. taniguchi , m. drapeko , s. mitani , and k. hono , `` large amplitude microwave emission and reduced nonlinear phase noise in co@xmath32fe(ge@xmath33ga@xmath33 ) heusler alloy based pseudo spin valve nanopillar , '' _ appl . _ , vol . 99 , no . 16 , p. 162508 , 2011 . t. seki , y. sakuraba , r. okura , and k. takanashi , `` high power radio frequency oscillation by spin transfer torque in a co@xmath34mnsi layer : experiment and macrospin simulation , '' _ j. appl 113 , no . 3 , p. 033907 , 2013 . t. seki , y. sakuraba , h. arai , m. ueda , r. okura , h. imamura , and k. takanashi , `` high power all - metal spin torque oscillator using full heusler co@xmath34(fe , mn)si , '' _ appl . 105 , no . 9 , p. 092406 , 2014 . t. yamamoto , t. seki , t. kubota , h. yako , and k. takanashi , `` zero - field spin torque oscillation in co@xmath34(fe , mn)si with a point contact geometry , '' _ appl . 106 , no . 9 , p. 092406 , 2015 . r. k. dumas , s. sani , s. m. mohseni , e. iacocca , y. pogoryelov , p. k. muduli , s. chung , p. drrenfeld , and j. kerman , `` recent advances in nanocontact spin - torque oscillators , '' _ ieee trans . _ , vol . 50 , no . 6 , 2014 . a. houshang , m. fazlali , s. r. sani , p. drrenfeld , e. iacocca , j. kerman , and r. k. dumas , `` effect of excitation fatigue on the synchronization of multiple nanocontact spin - torque oscillators , '' _ ieee magn . _ , vol . 5 , p. 3000404 , 2014 . a. a. tulapurkar , y. suzuki , a. fukushima , h. kubota , h. maehara , k. tsunekawa , d. d. djayaprawira , n. watanabe , and s. yuasa , `` spin - torque diode effect in magnetic tunnel junctions , '' _ nature _ , vol . 438 , pp . 339342 , 2005 . s. ishibashi , t. seki , t. nozaki , h. kubota , s. yakata , a. fukushima , s. yuasa , h. maehara , k. tsunekawa , d. d. djayaprawira , and y. suzuki , `` large diode sensitivity of cofeb / mgo / cofeb magnetic tunnel junctions , '' _ appl . , vol . 3 , no . 7 , , 2010 . c. wang , y .- t . cui , j. z. sun , j. a. katine , r. a. buhrman , and d. c. ralph , `` sensitivity of spin - torque diodes for frequency - tunable resonant microwave detection , '' _ j. appl . 106 , no . 5 , p. 053905 , 2009 . b. fang , m. carpentieri , x. hao , h. jiang , j. a. katine , i. n. krivorotov , b. ocker , j. langer , k. l. wang , b. zhang , b. azzerboni , p. k. amiri , g. finocchio , and z. zeng , `` giant spin - torque diode sensitivity at low input power in the absence of bias magnetic field , '' _ arxiv preprint arxiv:1410.4958 _ , 2014 . h. s. choi , s. y. kang , s. j. cho , i. y. oh , m. shin , h. park , c. jang , b. c. min , s. i. kim , s. y. park , and c. s. park , `` spin nano - oscillator - based wireless communication , '' _ sci . _ , vol . 4 , p. 5486 , 2014 . m. madami , s. bonetti , g. consolo , s. tacchi , g. carlotti , g. gubbiotti , f. b. mancoff , m. a. yar , and j. kerman , `` direct observation of a propagating spin wave induced by spin - transfer torque , '' _ nature nanotech . _ , vol . 6 , pp . 635638 , 2011 . m. pufall , w. rippard , g. csaba , d. nikonov , g. bourianoff , and w. porod , `` physical implementation of coherently - coupled oscillator networks , '' _ ieee j. explor . solid - state computat . devices circuits _ , vol . pp , no . 99 , pp . 18 , 2015 . p. m. braganca , b. a. gurney , b. a. wilson , j. a. katine , s. maat , and j. r. childress , `` nanoscale magnetic field detection using a spin torque oscillator , '' _ nanotechnology _ , vol . 21 , no . 23 , p. 235202 , 2010 . s. okamoto , n. kikuchi , m. furuta , o. kitakami , and t. shimatsu , `` microwave assisted magnetic recording technologies and related physics , '' _ j. phys . _ , vol . 48 , no . 35 , p. 353001 , 2015 . ( 2015 , july ) demonstration of a new technology for a multilayer magnetic recording to realize high - density hard disk drives . [ online ] . available : https://www.toshiba.co.jp/rdc/rd/detail_e/e1507_01.html r. sato , k. kudo , t. nagasawa , h. suto , and k. mizushima , `` simulations and experiments towards high - data - transfer - rate st - mram composed of a spin - torque oscillator , '' _ ieee trans . _ , vol . 48 , no . 5 , 2012 . i. firastrau , d. gusakova , d. houssameddine , u. ebels , m .- c . cyrille , b. delaet , b. dieny , o. redon , j .- c . toussaint , and l. d. buda - prejbeanu , `` modeling of the perpendicular polarizer - planar free layer spin torque oscillator : micromagnetic simulations , '' _ phys . rev . b _ , vol . 78 , no . 2 , p. 024437 , 2008 . t. chen , a. eklund , e. iaccoca , b. g. malm , s. rodriguez , j. kerman , and a. rusu , `` comprehensive and macrospin - based magnetic tunnel junction spin torque oscillator model - part i : analytical model of the mtj sto , '' _ ieee trans . on electron devices _ 62 , no . 3 , pp . 10371044 , 2015 . t. chen , a. eklund , e. iaccoca , b. g. malm , s. rodriguez , j. kerman , and a. rusu , `` comprehensive and macrospin - based magnetic tunnel junction spin torque oscillator model- part ii : verilog - a model implementation , '' _ ieee trans . on electron devices _ 62 , no . 3 , pp . 10451051 , 2015 . h. lim , s. ahn , m. kim , s. lee , and h. shin , `` a new circuit model for spin - torque oscillator including perpendicular torque of magnetic tunnel junction , '' _ adv . matter phys . 2013 , p. 169312 , 2013 . m. kim , h. lim , s. ahn , s. lee , and h. shin , `` advanced circuit - level model of magnetic tunnel junction - based spin - torque oscillator with perpendicular anisotropy field , '' _ j. semicond _ , vol . 13 , pp . 556561 , 2013 . t. chen , a. eklund , s. sani , s. rodriguez , b. g. malm , j. kerman , and a. rusu , `` integration of gmr - based spin torque oscillators and cmos circuitry , '' _ solid state electron . 111 , pp . 9199 , 2015 . ( 2010 , february ) `` fujitsu and university of toronto develop high - reliability read - method for spin - torque - transfer mram , next - generation non - volatile memory '' . [ online ] . available : http://www.fujitsu.com/global/about/resources/news/press-releases/2010/0210-03.html k. ikegami , h. noguchi , c. kamata , m. amano , k. abe , k. kushida , e. kitagawa , t. ochiai , n. shimomura , a. kawasumi , h. hara , j. ito , and s. fujita , `` a 4ns , 0.9 v write voltage embedded perpendicular stt - mram fabricated by mtj - last process , '' in _ proceedings of technical program-2014 international symposium on vlsi technology , systems and application ( vlsi - tsa)_.1em plus 0.5em minus 0.4emieee , 2014 , pp . n. d. rizzo , d. houssameddine , j. janesky , r. whig , f. b. mancoff , m. l. schneider , m. deherrera , j. j. sun , k. nagel , s. deshpande , h. j. chia , s. m. alam , t. andre , s. aggarwal , and j. m. slaughter , `` a fully functional 64 mb ddr3 st - mram built on 90 nm cmos technology , '' _ ieee trans . _ , vol . 49 , no . 7 , pp . 44414446 , 2013 . j. m. shaw , h. t. nembach , and t. j. silva , `` roughness induced magnetic inhomogeneity in co / ni multilayers : ferromagnetic resonance and switching properties in nanostructures , '' _ j. appl . 108 , no . 9 , p. 093922 , 2010 . j. wagner , r. wolf , n. joram , and f. ellinger , `` 0.59 ghz cmos variable gain amplifier with control linearization , low phase variations and self - matching , '' in _ european microwave integrated circuit conference ( eumic ) , 2014 9th _ , 2014 , pp . 100103 . tingsu chen ( s11 ) received an m.sc . degree in system - on - chip design from the kth royal institute of technology , sweden , 2011 . she is currently working toward a ph.d . degree at kth in the research area of modeling spin - torque oscillators and designing high - frequency circuits for spin - torque oscillator technology . randy k. dumas ( member ieee ) received his ph.d . in physics from the university of california , davis on the topic of reversal mechanisms in magnetic nanostructures in 2009 . from 20092012 , he was a postdoctoral research fellow at the university of gothenburg , where he is currently a permanent researcher . his research interests include the fabrication and characterization of magnetic nanostructures with a current focus on magnonics . anders eklund ( s13 ) received an m.sc . degree in engineering physics from kth royal institute of technology , sweden , in 2011 . he is currently working towards a ph.d . degree in physics at kth , experimentally investigating the frequency stability and mode structure of spin - torque oscillators by means of electrical characterization and synchrotron x - ray measurements . pranaba k. muduli ( member ieee ) received his ph.d . degree in physics from the humboldt university , berlin , germany in 2005 . he did postdoctoral research work at the university of north carolina , usa , the royal institute of technology , sweden , and the university of gothenburg , sweden . since july 2012 , he has been an assistant professor at the indian institute of technology delhi , india . he is also a researcher at the university of gothenburg , sweden . his current research interests include spintronics and nanomagnetism with an emphasis on spin - torque induced magnetization dynamics . afshin houshang ( s12 ) received his m.sc . in electroceramics engineering from shiraz university , shiraz , iran , 2010 . he is currently working towards a ph.d . degree in physics at the university of gothenburg in the area of synchronization phenomena in spin - torque and spin hall oscillators . ahmad a. awad ( member ieee ) received his ph.d . in physics from the autonomous university of madrid , spain , on high frequency magnetization and vortex dynamics in both magnetic and superconducting nanostructures in 2012 . since 2014 , he has been a postdoctoral research fellow at the university of gothenburg , sweden . his research interests include the characterization of magnetic nanostructures and spintronic devices with a focus on spin - torque induced magnetization dynamics . philipp drrenfeld ( s11 ) has a dipl .- ing . univ . degree ( 2010 ) from the university of wrzburg , germany . in 2015 , he received his ph.d . in physics from the university of gothenburg , sweden on the topic of spin - torque and spin hall nano - oscillators with single magnetic layers . his research interests include the nanofabrication and electrical characterization of spintronics and magnonics devices . b. gunnar malm ( m98 - sm10 ) was born in stockholm , sweden , in 1972 . he received an m.s . from uppsala university , sweden in 1997 and a phd in solid - state electronics from the royal institute of technology ( kth ) , stockholm in 2002 . he has been an associate professor at the school of ict , kth since 2011 . his recent work includes silicon photonics , silicon - carbide technology for extreme environments , and spintronics . he also serves on the kth sustainability council . ana rusu ( m92 ) received an m.sc . degree in electronics and telecommunications from the technical university of iasi , romania in 1983 , and a ph.d . degree in electronics from the technical university of cluj - napoca , romania , in 1998 . she has been with kth royal institute of technology , stockholm , sweden , since 2001 , where she is professor in electronic circuits for integrated systems . her research interests include low / ultralow - power high - performance cmos circuits and systems , sto - based systems , rf graphene , and high temperature sic circuits . johan @xmath30kerman ( m06 ) gained an ing . dipl . degree in 1994 from epfl , switzerland , an m.sc . in physics in 1996 from lth , sweden , and a ph.d . in materials physics in 2000 from kth royal institute of technology , stockholm . in 2008 , he was appointed full professor at the university of gothenburg and is a guest professor at kth royal institute of technology . he is also the founder of nanosc ab and nanosc instruments ab . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: studying the properties of resolved stellar populations through color - magnitude diagrams ( cmds ) can provide important constraints on stellar evolution and on the formation and evolution of stellar systems . for example , the apparent magnitude of the horizontal branch ( hb ) stars in a population , or the magnitude and color of the main sequence turn - off ( msto ) , constrain or allow determination of the distance to and age and metallicity of the population . the most detailed and most objectively quantifiable information can be obtained by a numerical cmd fitting analysis ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . over the last seven years , the sloan digital sky survey ( sdss , * ? * ; * ? ? ? * ) , has performed a large wide - field photometric and spectroscopic survey in the northern galactic cap . because sdss , unlike previous all - sky surveys , is a ccd - based survey , the photometric depth , accuracy , and homogeneity achieved is unprecedented over such a large area . sdss data have proven to be of great value for the study of the galactic stellar halo and nearby stellar systems with resolved populations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? they have also led to the discovery of a host of previously unknown , faint local group members @xcite . the study of dwarf galaxies and the milky way halo can provide important clues to the formation and assembly of the local group of galaxies and its individual constituents , as well as test cosmological theories . in the cold dark matter ( cdm ) paradigm , an abundance of subhaloes should be present around milky way type galaxies . the observed number of satellites is , however , one or two orders of magnitudes lower than what cdm theory predicts @xcite . the recently discovered satellites partially alleviate this discrepancy , but if the cdm predictions are accurate , many more new satellite galaxies are to be discovered . another prediction of cdm theories is that galaxies are formed hierarchically , i.e. built up by accreting smaller systems ( e.g. * ? ? ? * ; * ? ? ? * ) . among the population of satellite galaxies surrounding the mw , some should have been formed within the mw halo while others should have a different origin studying the stellar populations , masses and abundance patterns of these satellites can provide important constraints for galaxy formation and cdm theory . our goals for this paper are two - fold . first , we adapt the cmd - fitting software match @xcite for use on sdss data , making use of the theoretical isochrones provided by @xcite , and describe several methods of applying this software to the data . a set of globular clusters taken from the sdss data is used to test both the algorithmic approach and the theoretical isochrones . second , we apply match to the sdss data for the new milky way satellites ( that were recently discovered using the same sdss data ) , to provide a uniform analysis of these objects and constrain their population parameters . in section [ sec : methods ] we describe the data and the different cmd - fitting methods . the tests on globular clusters are done in section [ sec : gctests ] and in section [ sec : newsats ] we analyze the properties of the new galactic satellites . we conclude and summarize our results in section [ sec : summary ] . the sdss is a large photometric and spectroscopic survey , primarily covering the north galactic cap . as of the data release 5 ( dr5 , * ? ? ? * ) , sdss photometry covers almost a quarter of the sky in five passbands ( @xmath0 , @xmath1 , @xmath2 , @xmath3 , and @xmath4 ; * ? ? ? * ; * ? ? ? * ; * ? ? ? for the work presented in this paper we use photometry in the two most sensitive bands , namely @xmath1 and @xmath2 . a fully automated data reduction pipeline produces accurate astrometric and photometric measurements for all detected objects @xcite . photometric accuracy is on the order of 2% down to @xmath522.5 and @xmath622 @xcite , and comparison with deeper surveys has shown that the data are complete to @xmath622 . we use stellar data extracted from the sdss archive with artifacts removed and we apply the extinction corrections from @xcite provided by dr5 . in order for the cmd fitting with match to work , a model describing the completeness and the photometric accuracy as a function of magnitude is needed . completeness has been checked for @xmath2-band by comparing detection of stars and galaxies by sdss with the deeper combo-17 survey . for stellar sources the completeness is 100% to @xmath2=22 and then drops to 0% at r=23.5 . since the stellar classification of combo-17 starts to become incomplete at approximately one magnitude fainter than in sdss , this completeness is not extremely accurate , but amply sufficient for our purposes . the photometric errors given by the automated reduction pipeline have been found to be very accurate @xcite . comparison of the photometric accuracy in @xmath1 and @xmath2 as a function of apparent magnitude shows similar behavior , but in @xmath1 the rise in the uncertainties occurs some 0.5 magnitudes fainter . for the @xmath1-band completeness we assume similar behavior as in @xmath2 , but also shifted half a magnitude fainter . using this information we create artificial star test files that are given as input to our cmd fitting code . from this , the code creates the photometric accuracy and completeness model with which the theoretical population models are then convolved . figure [ fig : fakedetails ] shows the completeness and average photometric errors of our artificial star files for the @xmath1- and @xmath2-band separately . even though differences in , for example , seeing cause differences in photometric accuracy and completeness between different parts of the sky , we use this same average model in the remainder of this paper . several software packages have been developed for quantitative and algorithmic cmd ( or hess diagram ) fitting ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the strength of these methods lies in the fact that all photometric information from imaging in two bands can be used simultaneously in a robust and numerically well - defined way . the software package we use is match , which was written and successfully used for the cmd analysis of globular clusters and dwarf galaxies @xcite . the software works by converting an observed cmd into a hess diagram ( a two - dimensional histogram of stellar density as a function of color and magnitude ; * ? ? ? * ) and comparing it with synthetic hess diagrams of model populations . to create these synthetic hess diagrams , theoretical isochrones are convolved with a model of the photometric accuracy and completeness . the use of hess diagrams enables a pixel - by - pixel comparison of observations to models and a maximum - likelihood technique is used to find the best - fitting single model , or best linear combination of models . to account for contamination by fore- and background stars , a `` control field '' cmd needs to be provided . match will then use the control field hess diagram as an extra ` model population ' that can be scaled up or down . the resulting output will be the best - fitting linear combination of population models plus the control field . to construct star formation histories ( sfhs ) and age - metallicity relations ( amrs ) , a set of theoretical isochrones is needed that spans a sufficient range of ages , masses and metallicities . when match was written , the only published set of isochrones with full age and metallicity coverage available was provided by @xcite . since we are specifically interested in the application to sdss data , we need isochrones in the sdss photometric system . the only set available to date is one from @xcite and has the same parameter space coverage . some worries exist about the accuracy of the isochrone calibration ( e.g. * ? ? ? * ) , which was limited to three systems available in the first data release ( dr1 ) of sdss . this unavoidably adds an extra source of uncertainty to all cmd fitting results . one of the main motivations for the various tests that are described in this paper is to see whether reasonable results can be obtained using the current @xcite isochrones . we use match for slightly different purposes than originally described by @xcite ; the new `` modes '' are described briefly below . match was originally developed to solve for the sfh and amr of a stellar system for which all stars can be assumed to be at the same distance , such as globular clusters and dwarf galaxies . in this original mode of match the distance and the foreground extinction to all stars are fixed a priori for each fit . from a set of age bins and metallicity bins , the combination that best fits the observed cmd is found , resulting in an estimate of the sfh and amr . by doing the same fit with different values of the foreground extinction and distance , these two latter parameters can also be determined by comparing the goodness - of - fit of the individual fits . in principle , parameters like the steepness of the imf and the binary fraction can be constrained in the same fashion , although in most cases the data quality and the degeneracies between different parameters will prevent the determination of all of these parameters simultaneously . analyzing the parameter range of acceptable fits also gives robust error estimates for the sfh and amr ; this approach we refer to as match s `` standard mode '' . to apply cmd fitting techniques to nearby and extended stellar systems ( for example substructure within the milky way as in @xcite ) , the assumption that all stars are practically equidistant need not hold . to fit populations at differing distances simultaneously we have implemented a new way in which to run match that keeps the same number of free parameters , or parameter dimensions , to be explored . while in the standard mode the distance is fixed and age and metallicity are independent variables , here the distance becomes a free parameter , but age and metallicity are being limited via a specified age metallicity relation ( amr ) . for any given amr there is only one `` population parameter '' , e.g. age with a certain metallicity linked to each age bin . of course , different age - metallicity relations can be explored , potentially covering a wide range of possibilities . it seems sensible to choose an amr that is appropriate for the studied stellar population on astrophysical grounds , such as the expected or known metallicity evolution . for example , for the study of structures in the galactic thin disk one would consider a different amr than for the study of a metal - poor dwarf galaxy . however , within this new mode of running match the choice of amr is completely flexible and a flat amr ( no dependence of metallicity on age ) or even an amr where metallicity decreases with time is possible . when studying low - luminosity or low surface brightness contrast systems , such as some of the recently discovered dsphs , or systems at low galactic latitudes , for example substructures in the milky way , the contamination of the cmds of these systems by fore- and background stars will be very high . if the ratio of ` member ' stars to field stars becomes low , the s / n of the cmd features to be fit consequently becomes very low . in such cases , all free parameters used to constrain distance , age , and metallicity in the match mode of sections [ subsec : sfhamr ] and [ subsec : distsol ] can no longer be constrained by the data alone . to test a more restricted set of hypotheses in the case of low s / n systems we have developped a single component ( sc ) fitting mode for match . in this mode , template stellar population models are created that consist of a single component with given , fixed distance , foreground extinction , age range and metallicity range . such a sc template is fit to the data together with a control field . since the template consists of a single component , the only free parameters during the fit are the absolute scaling of ( i.e. the number of stars in ) the template and the control field . for each individual sc ( i.e. for each combination of distance , foreground extinction , age range and metallicity range ) , a value of the goodness - of - fit is obtained . comparing the goodness - of - fit values for all the sc s we can then find the best - fitting templates and thus constrain the overall properties of the system . the quality of each fit is expressed by the goodness - of - fit parameter q , determined using the maximum - likelihood technique described in @xcite . this statistic is meant to measure how likely it is that the observed distribution of stars would result from a random drawing from the fitted model . in the sc fitting scheme we know that our fits will not be perfect since an sc model will not be a perfect representation of a real stellar population , but we are primarily interested in a likelihood - ratio test for a range of sc models . the expected variance in the parameter @xmath7 for a certain cmd can be calculated from poisson statistics and also obtained from monte carlo simulations using random drawings from the models . all fits with a value of @xmath7 less than 1 @xmath8 away from that of the best - fit sc model can be considered statistically acceptable , under the assumption that the best fit is a reasonable representation of the data . taking all these fits together , error estimates on the fit parameters can be determined from the encountered spread of the parameter values . this approach of using relative values of @xmath7 should only be used in cases where the `` best fit '' is a reasonable approximation of the data . in cases where there are several very different prominent stellar populations this will not be the case and results will break down . studying different parts of the cmd separately can help to circumvent such problems , as is done in @xcite and also in section [ subsec : nsscfits ] of this paper for the dwarf galaxy leo t. the same cmd fitting techniques can also be used to detect or verify , faint stellar over - densities in large datasets like that of the sdss survey . in case of an overdensity at a specific distance , the stars will not only cluster around a certain spatial position , but also along a certain locus in the cmd . the method we have developped works by fitting the cmd of a small target patch of sky both with a control cmd obtained from a large area around the target patch and with the same control field plus a simple stellar population model . if no over - density is present , the fit with the control cmd should be good and no significant improvement in fit quality will result by including an extra stellar population model . however , if there is an over - density present and the parameters of the population model , such as distance , age and metallicity , are close enough to those of the over - density , the fit including the model will result in a significant improvement in the fit quality . by doing this for a range of models the presence of a stellar over - density can be assessed by looking at the difference in the fit quality parameter , @xmath9 , between the control cmd fit and the control cmd plus model fit . moreover , by comparing the values of @xmath9 for different models , a rough estimate of the properties of the overdensity can be obtained at the same time . an important prerequisite for this method is of course that the over - density should contribute a larger fraction of the total number of stars to the target cmd than to the control cmd , and ideally should be contained completely within the target patch . this means that over - densities will be detected with the highest s / n if their size matches the size of the target patch . cccc 1 & 1013 & @xmath101.5 & 14.5 + 2 & 1013 & @xmath101.7 & 14.5 + & 22.5 & @xmath101.3 & 14.5 + 3 & 1013 & @xmath101.5 & 20.5 + 4 & 1013 & @xmath101.7 & 20.5 + & 22.5 & @xmath101.3 & 20.5 + [ tab : fakegcprops ] before turning to real data it is useful to get a feel for the degree of precision that we can expect cmd fits to achieve . to do this , we create a few realistic artificial cmds and apply sfh fitting , sc fitting and distance fitting methods to these to try and recover their properties . using the theoretical isochrones from @xcite , a salpeter initial mass function @xcite , and our model for the photometric errors and completeness of the sdss data , we generate artificial stellar populations , to which field stars taken from the sdss data are added . in total we create four cmds , shown in figure [ fig : fakecmds ] . the number of stars in each cmd is chosen to be the same , to make for a fair comparison . cmds 1 and 3 contain a simple , single burst stellar population , located at different distances , so that either the ms and msto or the rgb and hb are prominent features . artificial cmds 2 and 4 contain two different age and metallicity populations , again with the cmds corresponding to different distances . the detailed parameters of the artificial populations are listed in table [ tab : fakegcprops ] . from the sdss archives all stellar sources in a 5@xmath115 patch of sky close to the globular cluster m13 were extracted and a small fraction used to create the field star contamination in these cmds . the remaining sources will be used as a `` control field '' population when fitting these background contaminated cmds . we recover the sfhs of the artificial cmds by running match with nine age bins and twelve metallicity bins . the age bins are defined in log @xmath12 because the isochrones are spaced more evenly in log @xmath12 than in @xmath12 , with the oldest running from 10.0 to 10.2 ( 10 gyr to 15.8 gyr ) and the youngest from 8.6 to 8.8 ( 400 myr to 630 myr ) . the metallicity range of [ fe / h]@xmath132.4 to 0.0 is covered by twelve bins of 0.2 dex width . thus , the artificial cmds are fit with 108 model cmds plus the `` control field '' stars , for a total of 109 partial cmds . the best fitting linear combination of model cmds will form the resulting sfh and metallicity evolution . to get a sense of the uncertainties on the sfh , each cmd is fit thrice , each time with a slightly different assumed distance . the distance moduli used are 14.3 , 14.4 and 14.5 mag . the resulting sfhs and metallicities are presented in figure [ fig : fakesfhamr ] as the relative star formation rate ( sfr ) and as [ fe / h ] as function of the age of the stars . for the sc fits models are created for a grid of parameters with a distance modulus range of 2.0 with 0.1 magnitude intervals centered on the the actual distance modulus ( table [ tab : fakegcprops ] ) . ages used run from 9.0 to 10.2 in log @xmath12 with bin widths of 0.1 dex and metallicities go from [ fe / h]@xmath132.1 to 0.0 in steps of 0.1 dex and bin widths of 0.2 dex . each model plus the `` control field '' is fit to an artificial cmd , after which the resulting goodness - of - fit values are compared . the resulting contour plots showing the best - fitting parameter combinations are shown in figure [ fig : fakescfit ] . using the new distance solution option of match we solve for the sfh of the artificial cmds again while fixing the metallicity to a single , uniform value . the same age bins are used as for the sfh fits described above , and a distance range of 3.0 is probed with 0.1 magnitude resolution . including the `` control field '' , 271 partial cmds are used in these fits . the fits are done at three different values of the metallicity : [ fe / h]@xmath131.6 , @xmath101.5 , and @xmath101.4 to probe the uncertainties . figure [ fig : fakesfhdist ] shows the resulting sfhs in relative sfrs and distance determinations . the sfhs in figures [ fig : fakesfhamr ] and [ fig : fakesfhdist ] show that the software succesfully recovers the single burst in cmds 1 and 2 , and the two distinct episodes of star formation in cmds 3 and 4 . some bleeding of star formation in adjacent bins will occur even when the fitted age bins exactly match the input sfh @xcite . therefore it is to be expected that this also happens here , since the duration of the input star forming episodes do not match any age bins perfectly . it should be noted that the error bars in the sfhs represent the range of sfrs encountered in each age bin within the three fits on which the sfhs are based , which means that the errors between adjacent bins are correllated . for example , in the sfh of cmd 4 in figure [ fig : fakesfhamr ] the two bins at ages of 2 to 4 gyr are not independent ; when the sfr in one of them is low , it is high in the other . that the age found for the young population in cmd 4 in figure [ fig : fakesfhdist ] is the natural result of the metallicities in the fit being lower than the actual metallicity of the younger population . in all cases , the average metallicities and distances found in figures [ fig : fakesfhamr ] and [ fig : fakesfhdist ] agree well with the input values . turning to the sc fits in figure [ fig : fakescfit ] , we see that for cmds 1 and 3 , the input properties ( indicated by stars ) are recovered very well . because there is a degeneracy between distance , age and metallicity when only the ms and msto are seen , the properties of cmd 1 are , however , constrained more poorly than for cmd 3 . since the sc fits use a single model population , the results for cmds with multiple populations , such as cmds 2 and 4 , can not be expected to properly recover the properties of all stars . rather , the population representing the majority of stars in the cmd will have the strongest weight . this is clearly visible in the sc fitting results for cmd 4 , that recover the properties of the older population , that dominates the star counts in the rgb and hb . for cmd 2 , however , the upper ms and msto of the younger population has such high s / n that this population does influence the sc fits significantly and the favored age is relatively young ( @xmath144 gyr ) . this shows that for high s / n cmds with a complex stellar population make - up , the sc fits is not well suited ; they were , however , developed for application to low s / n cmds . what distinguishes these artificial cmds from the real data , is that the artificial stellar populations are based on the exact same isochrones , imf , and photometric error and completeness model as used by match . inaccuracies in any of these will degrade the fit results one obtains for real data compared to the results for the artificial data . lccccc m 13 & 7.7 , 7.5@xmath15 & 14.38 , 14.43 & @xmath101.54 & 12@xmath15 & 0.019 + m 15 & 10.3 , 10.0@xmath16 & 15.37 , 15.0@xmath16 & @xmath102.26 & 13.2@xmath16 & 0.11 + m 92 & 8.2 & 14.6 & @xmath102.28 & 14@xmath17 & 0.025 + ngc 2419 & 84.2 , 98.6@xmath18 & 19.63 , 19.97@xmath18 & @xmath102.12 & 14@xmath18 & 0.060 + ngc 6229 & 30.4 & 17.44 & @xmath101.43 , @xmath101.1@xmath19 & @xmath1411.5@xmath20 & 0.023 + pal 14 & 73.9 & 19.34 & @xmath101.52 & @xmath1410@xmath21 & 0.035 + [ tab : gcproperties ] globular clusters are ideal test cases for the sdss implementation of match since in most of them their stars have been found to be of a single age and nearly single metallicity . here we will use six gcs present in the sdss data to test how well match , using the isochrones from @xcite , can recover their known population properties . the systems in question , , , , , , and , cover a large range in distance , background contrast ( i.e. ratio of member stars versus field stars ) and have metallicities between [ fe / h]@xmath22 and @xmath23 dex . for the following tests , all stars within a 5@xmath115 field centered on each gc were obtained from the sdss catalog archive server ( cas ) . it should be noted that in the cases of m15 , m92 and ngc 2419 these fields were not completely covered by the dr5 footprint . flags were used to exclude objects with poor photometry and all stars were individually corrected for foreground extinction using the correction factors provided by dr5 and based on the maps by @xcite . any contamination of the samples of stars obtained by misclassified galaxies or other contaminants do not pose problems for our analysis because of the use of control fields that should contain the same populations of these contaminants . `` control field '' cmds were constructed from all stars outside a radius of 0.5 from the center of each gc . `` target fields '' were constructed by selecting stars within a certain radius from the center . the radii used for m13 , m15 , m92 , ngc 2419 , ngc 6229 , and pal 14 are 0.25 , 0.2 , 0.25 , 0.15 , 0.15 , and 0.1 , respectively . these radii are similar to the tidal radii of the gcs but are not strictly related to them , since in all clusters and especially in the densest ( m13 and m15 ) the sdss photometry in the central parts is missing due to the high degree of crowding . with the radii used , a significant number of field stars is present in each target cmd , so that match has a good handle on the proper scaling of the control cmd . the sdss cmds of the gcs are shown in figure [ fig : gccmds ] and their previously measured parameters ( distance , age , [ fe / h ] ) are listed in table [ tab : gcproperties ] ; the mean reddening @xmath24 in magnitudes according to the @xcite maps in the target area for each gc is also listed in the table . the gcs in the three top panels in figure [ fig : gccmds ] have metallicities of around [ fe / h]@xmath25 dex , and the gcs in the lower panels are more metal - poor ( [ fe / h]@xmath26 ) . for the more nearby systems ( m13 , m15 , m92 and ngc 6229 ) the main cmd features that will drive the fits are the main sequence ( ms ) and main sequence turn off ( msto ) ; for the more distant systems ( ngc 2419 and pal 14 ) only the red giant branch ( rgb ) and horizontal branch ( hb ) are observed . pal 14 in particular represents the case of a very sparsely populated system with high contamination from field stars . in the following subsections , we discuss the results of applying match to these gcs in the four different modes described in section [ sec : modes ] . in fitting the cmds of these objects we will use a color range of @xmath27 , and limit the magnitude range to @xmath28 and @xmath29 ; the color cut includes the msto and the tip of the rgb , but excludes a large fraction of the numerous low - mass foreground stars . for all the fits in this section we will assume a slope for the stellar initial mass function that is close to salpeter ; the binary fraction is assumed to be 0.1 , in line with recent observational and theoretical evidence for low binary fractions in globulars ( e.g. , * ? ? ? * ; * ? ? ? to measure the sfh and amr of the clusters we run match fits with nine age bins and twelve metallicity bins . because the isochrones are spaced more evenly in log @xmath12 than in @xmath12 , the age bins are defined to have equal width in log @xmath12 , with the oldest running from 10.0 to 10.2 ( 10 gyr to 15.8 gyr ) and the youngest from 8.6 to 8.8 ( 400 myr to 630 myr ) . the metallicity bins cover the range from [ fe / h]@xmath132.4 to 0.0 in twelve bins of 0.2 dex width . together with the control cmd this yields 109 model cmds for each gc . the fitting software then finds the linear combination of these model cmds that best fits the observed cmd . each object is fit several times with slightly different fixed distances to determine the impact of distance errors on the recovered solutions . in the cases of m15 and ngc 2419 , where values from the literature for the distance modulus are quite discrepant , we run fits for distance moduli of 15.0 to 15.4 ( m15 ) and 19.6 to 20.0 ( ngc 2419 ) in 0.1 magnitude steps . for the other clusters we run three fits each , one at the literature value and at 0.1 magnitude further and nearer . figure [ fig : gcsfh ] shows the results of these fits , with the fractional sfr as a function of time in the left column , and the metallicity as a function of time in the right column . metallicities are only plotted for age bins with a significant sfr . the plotted sfr for a given age bin is the average of the values given by the fits at the different distances , with the error bars showing the complete range of sfr values found . the metallicity in a bin is the average of the fits weighted by the sfr , and the error bars correspond to the standard deviation . looking at the three intermediate metallicity gcs in the top half of figure [ fig : gcsfh ] we see that in all cases match infers a purely old ( @xmath307 gyr ) population , as expected . for m13 and ngc 6229 practically all stars are found to be older than 10 gyrs ; some bleeding into the second oldest age bin is always likely to occur , cf . section [ subsec : fakegc ] and @xcite . in the case of pal 14 the sfr is found to be constant over the two oldest age bins , but this only confirms the age estimate of @xmath1410 gyr @xcite , which implies that star formation would likely have been spread over these two bins . the intermediate metallicities of m13 , ngc 6229 and pal 14 are reproduced well and our recovered values are within 0.1 to 0.2 dex of the literature values . for ngc 6229 our fits favor the slightly lower metallicity from @xcite rather than the higher value from @xcite . turning to the bottom half of figure [ fig : gcsfh ] with the three metal - poor clusters , we see that here also the results are consistent with a purely old population , although the sfhs of m15 and ngc 2419 show a non - zero sfr in a few younger age bins . in the case of m15 this is due to the blue horizontal branch ( bhb ) stars that are not properly reproduced by the theoretical isochrones , causing the code to fit them as a younger turn - off . this demonstrates that cmd fitting results always need to be looked at in detail in order to understand their implications . the age distribution in ngc 2419 is not very well constrained because no msto is present in the cmd . the metallicities of m15 and m92 are recovered well for the old stars . while the estimated metallicity of @xmath101.9 for ngc 2419 is on the high side , for these metal - poor gcs the recovered metallicities are within 0.2 dex of the literature values . as described in section [ subsec : scfit ] above , in the single component ( sc ) fitting mode an individual model cmd is fit to the data together with the control cmd , rather than a whole set of models . the results therefore provide information on the characteristic properties of the dominant stellar population in the cmd . we sample the 3d parameter space of distance , age and metallicity by using one of nine age bins , one of 21 distance bins , and one of 23 metallicity bins . the age bins have a width of 0.1 log - years , going from 10.1 - 10.2 log - years ( 12.5 gyr to 15.8 gyr ) to 9.7 - 9.8 log - years ( 5 gyr to 6.3 gyr ) in steps of 0.05 log - years . this age range more than suffices for these gcs . we use a distance modulus range of two magnitudes centered on the literature value with a resolution of 0.1 magnitude . metallicity bins are centered at [ fe / h]@xmath132.2 to 0.0 with an inter - bin distance of 0.1 dex and a width of 0.2 dex . after matching each of the 4347 sc options to each gc and determining the expected variance in the goodness - of - fit parameter @xmath7 , we take the best fit as our measurement and compare the other fits with respect to these best values . the resulting likelihood contour plots for metallicity , distance and age measurements are shown in figure [ fig : gcscfit ] . note that in these contour plots , the third parameter is unconstrained , rather than fixed at its known value . if the third parameter would be kept fixed , the contours would shrink significantly . in general the agreement with the literature values ( indicated with asterisks in figure [ fig : gcscfit ] ) is very good . the small offset for m13 is most likely due to inaccuracies in the isochrones , which illustrates the added uncertainty in cmd fitting results that depend on a single set of theoretical isochrones . our distance modulus measurement from the sc fits for m15 lies exactly in between the two literature values , while for ngc 2419 our results clearly favor the smaller distance value from @xcite . from the two metallicity determinations in the literature for ngc 6229 , the fit results agree significantly better with the lower metallicity of [ fe / h]@xmath311.4 dex . in some cases , it looks as if the peak of the likelihood lies outside of the parameter space probed , at lower metallicities or older ages . unfortunately , these limits of the parameter space are dictated by the parameter range of the available isochrones . also shown in the right column of figure [ fig : gcscfit ] are the observed cmds of the gcs , with contours representing the model populations . in most cases we used the literature parameters for these models , except for m15 , where we chose a distance modulus of 15.2 , the mean of the two different available values . for ngc 6229 and ngc 2419 we use the literature metallicities and distance moduli that best correspond to our results . the agreement of model and data is generally acceptable , although inaccuracies , especially in the sub - giant branch , can be seen in m13 and m15 . important to note is the influence of the degeneracies between age , distance , and metallicity . for the nearby systems , m13 , m15 and m92 , where the fit is driven by the ms , there is a clear distance - metallicity covariance ; the left - most panels in figure [ fig : gcscfit ] show that the distance estimate increases with increasing metallicity . for the distant gc pal 14 , where the fit is driven by the rgb , the covariance is the other way around . another important , yet not wholly surprising , point is that higher s / n in the cmd features results in much tighter contours , illustrated by the fact that m13 , m15 and ngc 2419 give the best constrained results . ngc 2419 has the additional advantage that the well - populated hb provides a very strong constraint on the distance , making this system the best - constrained of the lot . as mentioned in section [ subsec : distsol ] , we explore fits for different distances and different ages , with the metallicity linked to the age . the result of each fit is a sfh with a ( potentially different ) distance for each age bin ; differing distances for different sub - populations are not astrophysically relevant for globular clusters , but serve as a test case . for fitting the gcs it seems reasonable to use the same metallicity in all age bins . we use the same nine age bins as in section [ subsec : gcsfh ] . in distance space we use a distance modulus range of three magnitudes centered on the known distance of each system with a resolution of 0.1 magnitudes , giving a total of 31 distance bins . for each gc we carry out three fits with different metallicities , centered on the literature values ( see table [ tab : gcproperties ] ) with offsets of 0.1 dex . for ngc 6229 we assume the lower metallicity value [ fe / h]@xmath131.43 , which is favored by our sfh and single component fitting results . in figure [ fig : gcdfit ] we present the age - distance distributions that result from these fits . compared to the fits in section [ subsec : gcsfh ] the number of free parameters per fit is the same , with now metallicity fixed instead of distance . thus , in general similar quality sfhs should be expected , although not necessarily exactly the same . while in the `` standard '' sfh fits the star formation within an age bin can be spread over several metallicity bins , it can now be spread over several distance bins . these spread will translate differently into the color - magnitude plane , and therefore can result in differences between the sfhs in figures [ fig : gcsfh ] and [ fig : gcdfit ] . as in figure [ fig : gcsfh ] all gcs are found to have a single peak in their sfr in the oldest age bin , in some cases with some bleeding into the adjacent bin . the extension of the sfh of pal 14 to more recent times has large uncertainties and is not significant . it should be taken into account that the exact age of this system is difficult to constrain with sdss data , since the msto is not present . interesting to see is that m13 and m92 show almost no bleeding into the second oldest age bin here , while they did in figure [ fig : gcsfh ] ; for m15 and ngc 6229 the exact opposite is the case . apparently , the former are fit better with some distance spread , while the latter benefit more from a metallicity spread . this does not necessarily reflect an actual distance or metallicity spread , but rather shows that inaccuracies in the isochrones are sometimes better accounted for one way and sometimes another way . the distances we find for the oldest age bin , and therefore for the majority of the stars , show very good agreement with the literature values . comparing figures [ fig : gcsfh ] and [ fig : gcdfit ] shows that the sfhs obtained when fixing either the distance or the metallicity are very similar . the main motivation for enabling this distance fitting mode was , however , not to analyze single - distance systems , but to be able to cope with cases where stars are located at different distances ( cf . canis major , * ? ? ? * ) . to demonstrate this application , we combine the cmds of m13 and ngc 6229 , resulting in a cmd containing two populations with similar metallicities and ages , but located at significantly different distances . the cmds of m13 , ngc 6229 and the combined cmd are shown in panels a ) , b ) and c ) of figure [ fig : m13ngc6229 ] . we run distance solution fits to these cmds in the same way as before , with the same age bins , but now using a larger distance modulus range , from 12.0 to 19.0 magnitudes with a resolution of 0.2 magnitudes . three fits are run , for uniform metallicities of [ fe / h]@xmath131.4 , @xmath101.5 , and @xmath101.6 . we include also a control cmd in the fit which is the control cmd of m 13 ; since m 13 and ngc 6229 are only 10 degrees apart and at similar galactic latitudes , this gives good results . the averaged best - fitting distributions of stars over the grid of age and distance bins are shown in panels d ) , e ) and f ) of figure [ fig : m13ngc6229 ] . both ages and distances of the two populations are recovered extremely well , with all stars in the oldest age bin ( @xmath3010 gyr ) , m13 stars centered at a distance modulus of 14.4 and ngc 6229 stars at a distance modulus of 17.4 . moreover , the result for the combined cmd is almost exactly the sum of the results for the individual cmds , clearly illustrating the power of match to decompose more complex cmds . the final application we describe and test here is aimed at algorithmically detecting stellar populations which are confined to a limited distance and area on the sky , i.e. that form `` overdensities '' . of course , the gcs used in this section are very obvious stellar overdensities and should be detected at very high s / n . following the approach described in section [ subsec : detect ] we define a grid of 20@xmath1120 boxes in each 5@xmath115 field that was obtained from the sdss catalog server , with the central box centered on the gc . the cmd of each box is fit with the control field cmd and with the control field plus a simple population model . for each gc we use a model with the appropriate distance , age and metallicity . figure [ fig : gcareas ] shows the difference in the goodness - of - fit parameter @xmath7 between the two fits for areas around each gc . note that for m15 and m92 , the 5@xmath115 field is not completely observed by sdss . clearly , all gcs produce high peaks in @xmath9 . since @xmath9 is roughly proportional to the number of stars attributed to the population model in the model fit , the objects with the largest numbers of stars produce the strongest detections . only in the plot for pal 14 does the vertical scale allow the random variations in @xmath9 to be visible in the area around the object . the s / n of these detections should be measured with respect to this random noise . using the 216 boxes around the central 9 boxes we obtain an estimate of the rms in @xmath9 from random variations in the cmds . negative values of @xmath9 are suppressed because a negative number of stars in a population is unphysical , which has to be taken into account when calculating the rms . even pal 14 , a rather sparsely populated and distant cluster , is detected with high s / n , as @xmath32 . in summary , section [ sec : gctests ] has illustrated the robustness and precision with which match can interpret high - quality cmds in a variety of circumstances . over the past few years the importance of deep wide - field surveys for the study of the stellar components of the milky way galaxy has been demonstrated by 2mass ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and the sdss . specifically , the sdss has enabled a detailed picture of the galactic stellar halo ( e.g. , * ? ? ? * ; * ? ? ? * ) and led to the discovery of at least fourteen previously unknown dwarf galaxies and globulars : ( umai , * ? ? ? * ) , ( wil1 , * ? ? ? * ) , ( cvni , * ? ? ? * ) , ( booi , * ? ? ? * ) , ( umaii , * ? ? ? * ) , , , , , ( com , cvnii , her , leoiv , seg1 , * ? ? ? * ) , @xcite , ( booii , * ? ? ? * ) , and two very low - luminosity globulars and @xcite . while these discoveries are important for our understanding of the so - called `` missing satellite problem '' ( e.g. , * * ; * ? ? ? * ) , these new systems might also shed new light on the mechanisms of star formation in low - mass halos with small baryon content . why these systems were only discovered with sdss data can be explained by their low surface brightness , as shown by @xcite . apart from the promise that more systems of even lower surface brightness might be waiting to be discovered by future generations of deep imaging surveys , this fact raises intriguing questions . do these systems show that star formation can occur under conditions generally considered to be incapable of supporting such activity , or have these systems evolved to their current state through , for example , tidal interactions with the galaxy ? a thorough and systematic study of the properties of these objects is needed to understand their formation and evolution and the possible cosmological implications . several groups have been working on photometric and spectroscopic follow - up of the new mw companions in order to constrain their properties . a compilation of these results is given in table [ tab : newprops ] . although individual measurements of properties might be less precise in some cases , the advantage of studying these objects using the sdss photometry is that this highly uniform dataset allows a systematic approach to the complete set of objects and fair comparisons between them . in the following , we apply our single component ( sc ) fitting techniques to constrain the overall distances , metallicities and ages of the objects listed in table [ tab : newprops ] . we also constrain the sfh of a subset of objects , since the majority contain too few stars to make this a worthwhile exercise . the objects we use are boo , cvn i , and uma i because of their relatively well - populated cmds , and leo t because of its clearly complex star formation history . for all the fits in this section we assume a salpeter - like initial mass function and a binary fraction of 0.5 , similar to the value for disk stars @xcite . lccccc bootes i & 60@xmath336 & 18.9@xmath330.2 , 19.1@xmath330.1@xmath15 , 19.0@xmath330.1@xmath16 & @xmath14 - 2 , @xmath14 - 2.5@xmath34 , @xmath14 - 2.1@xmath35 & 12.8@xmath330.7 & -5.8@xmath330.5 + bootes ii & 60@xmath3310 & 18.9@xmath330.5 & @xmath14 - 2 & 4@xmath332 & -3@xmath331 + canes venatici i & 220@xmath3320 & 21.75@xmath330.20 & @xmath14 - 2 , -2.5 -1.5@xmath18 , -2.0@xmath35 , -2.09@xmath19 & 8.5@xmath330.5 & -7.9@xmath330.5 + canes venatici ii & 150@xmath3315 & 20.9@xmath330.2 & -2.3@xmath19 & @xmath143.2 & -4.8@xmath330.6 + coma berenices & 44@xmath334 & 18.2@xmath330.2 & @xmath14 - 2 , -2.00@xmath19 & @xmath145.5 & -3.7@xmath330.6 + hercules & 140@xmath3313 & 20.7@xmath330.2 & -2.27@xmath19 & @xmath148.2 & -6.0@xmath330.6 + leo iv & 160@xmath3315 & 21.0@xmath330.2 & -2.3@xmath19 & @xmath143.4 & -5.1@xmath330.6 + leo t & @xmath14420 & @xmath1423.1 & @xmath14 - 1.6 , -2.3@xmath19 & @xmath141.4 & @xmath14 - 7.1 + segue 1 & 23@xmath332 & 16.8@xmath330.2 & - & @xmath144.5 & -3.0@xmath330.6 + ursa major i & @xmath14100 & @xmath1420 & @xmath14 - 2 , -2.2@xmath35 , -2.1@xmath19 & @xmath147.8 & @xmath14 - 6.5 + ursa major ii & 32@xmath335 & 17.5@xmath330.3 & @xmath14 - 1.3 , -1.6@xmath35 , -2.0@xmath19 & @xmath1412 & -3.8@xmath330.6 + willman 1 & 38@xmath337 & 17.9@xmath330.4 & @xmath14 - 1.3 , -1.6@xmath35 & @xmath141.9 & @xmath14 - 2.5 + [ tab : newprops ] as for the gcs we obtained data in 5@xmath115 fields around each object from the sdss cas as outlined in section [ sec : gctests ] . for the target cmds we used the stars within one half - light radius ( @xmath36 , table [ tab : newprops ] ) from the center ( with a minimum of 3 ) , with all stars outside a radius of 60 used for the control cmds . figure [ fig : nscmds ] shows the target cmds for the twelve objects . in all cases except leo t a color range of @xmath27 is used for the cmd fits . the population parameter space we sample consists of 12 age bins , 24 metallicity bins and 31 distance bins . the age bins have a 0.1 log - year width and separations of 0.1 log - year , with the oldest running from 10.110.2 log - years ( 12.5 gyr to 15.8 gyr ) and the youngest from 9.09.1 ( 1 gyr to 1.26 gyr ) . metallicities run from [ fe / h]=@xmath102.3 to 0.0 in steps of 0.1 dex , and each metallicity bin has a width of 0.2 dex . the range in distance modulus probed is 3 magnitudes wide , with a resolution of 0.1 magnitudes , and is centered on the literature value for each object . for leo t we use a slightly different approach . the group of blue stars in the cmd is too bright and does not have the right morphology to be interpreted as hb stars belonging to the same population as the rgb stars . this means there are two populations present in two different parts of the cmd , with the rgb corresponding to the old population and the stars at @xmath370.3 to a much younger population . as demonstrated in section [ subsec : fakegc ] , in such a case the sc fits will give very poor results . however , since the two populations in leo t are only present on one side of the @xmath38 line , they can be easily separated . thus , to determine the properties of the two populations separately we divide the leo t cmd into a red ( -1.0@xmath390.3 ) and a blue ( 0.3@xmath391.5 ) part and fit these separately . the red part of the cmd is fit with the same parameter space as the other objects . for the blue part the same metallicities and distance moduli are probed , but a younger set of age bins is used . here we use nine bins , the four oldest of which have a width of 0.1 log - years centered on 9.05 , 9.15 , 9.25 and 9.35 ; the five youngest have a width of 0.2 log - years centered on 8.1 , 8.3 , 8.5 , 8.7 and 8.9 log - years . in figures [ fig : nsscfit1 ] and [ fig : nsscfit2 ] the sc fitting results for the new mw satellites , except leo t , are presented . the right - most panels show the cmd of each object again , with the best - fit sc model overplotted as contours . monte carlo runs were used to determine the values of @xmath8 . except for boo i and cvn i , solid contours are shown for the case of a fixed metallicity or age , where such a measurement is available in the literature . figure [ fig : leotscfit ] shows the same results for the red and blue part of the leo t cmd . the fact that low s / n hinders a robust determination of the population properties provides a serious problem for the majority of the targets . although for most objects a finite region of parameter space lies within 1 @xmath8 from the best fit ( the white area in the contour plots ) , reasonable measurements of any individual parameter can only be obtained for boo i , cvn i , cvn ii , com and her . especially for boo ii , leo iv and leo t , parameters are very poorly constrained , and @xmath7 space shows a complicated morphology with multiple minima . to constrain the sfhs of boo i , cvn i , leo t and uma ii we follow the same approach as before for the gcs . a grid composed of nine age bins and twelve metallicity bins is used , the age bins going from 10.2 to 8.4 log - years with an 0.2 log - year width , and the metallicities from [ fe / h]=-2.4 to 0.0 with a 0.2 dex width . a fit is done at three different distance moduli , 0.1 magnitude apart , with the middle value equal to the literature value . the resulting sfhs and amrs are plotted in figure [ fig : nssfh ] . as with the gcs , we also test our overdensity detection scheme , described in section [ subsec : detect ] , on these new milky way satellites . the 5@xmath115 field centered on each object is divided in a square grid of 225 20@xmath1120 boxes , the cmds of which are fit both with a control field and with the control field plus a single component population model . as the control field the whole 5@xmath115 field is used . for the sc model we consider two cases . first we use a typical old population model , with age range 10.0 to 10.1 log - year and [ fe / h]@xmath40 at the best - fit distance obtained with the sc fits from section [ subsec : nsscfits ] . second we use a model corresponding to the best - fit properties obtained for each object . the former case serves to simulate a realistic search of all sdss data for unknown stellar overdensities . when performing such a search , an old , metal - poor model would be a likely choice since most mw satellites are dominated by such populations . another factor that would influence the results of a search for unknown stellar overdensities is the size of the box used . as a test , we also did the same exercise using different box sizes for boo i. in addition to the 20@xmath1120 boxes , we also used 40@xmath1140 boxes and 1@xmath111boxes . to enable the use of these coarser grids , we retrieved a larger , 10@xmath1110 field from the sdss dr5 archive centered on boo i. in figure [ fig : nsarea ] we present the results of the fits with the fine - tuned stellar population models to the twelve objects . each panel shows the improvement in the goodness - of - fit , @xmath41 , as function of the position in the field centered on a satellite . figure [ fig : booiareas ] shows the resulting @xmath41 for the three different box sizes in a 10@xmath1110 field centered on boo i. the significances of all detections , expressed in @xmath42 , are listed in table [ tab : detsig ] . lcc boo i@xmath15 & 106.9 & 121.6 , 58.2 , 27.8 + boo ii & 0.86 & 0.93 + cvn i & 447.7 & 589.4 + cvn ii & 2.61 & 13.3 + com & 57.5 & 53.9 + her & 15.2 & 28.2 + leo iv & 13.8 & 38.2 + leo t & 5.8 & 11.3 + seg 1 & 9.9 & 7.8 + uma i & 12.9 & 11.2 + uma ii & 1156.2 & 865.3 + wil 1 & 48.8 & 47.8 + [ tab : detsig ] we have performed a systematic and uniform numerical cmd analysis of the recently discovered mw satellites . how well the properties of an individual object can be determined depends on the number of member stars present in the cmd and the observed cmd features . for some objects , this type of analysis is hard - pressed to provide strong statistical constraints due to the sparseness of their cmds . in table [ tab : newresults ] we summarize the results of our single component and star formation history fits . the distance moduli , metallicities and ages listed are the unweighted averages of all sc fits that lie within 1 @xmath8 of the best sc fit and are therefore considered to be acceptable fits . the errors are the standard deviations of the parameter values of these fits , combined with an added systematic uncertainty of 0.1 dex in [ fe / h ] and 0.1 mag in distance modulus to account for isochrone uncertainties . in cases where these properties can not be sufficiently constrained we list the metallicities and ages we get when assuming the distance modulus from the literature , as this is usually the best - determined property . the star formation history results , where available , are summarized in table [ tab : newresults ] by the main star formation episodes . below follows an object - by - object discussion of these and the star formation history results . _ boo i._ this dwarf spheroidal is relatively nearby at @xmath1460 kpc and an rgb , hb and msto are all present in the sdss data , albeit sparsely sampled . the overall properties can therefore be determined with relatively good accuracy , and the star formation history is the best - constrained of all . especially the distance of boo i is very well constrained thanks to the hb which is detected at high s / n due to the absence of field stars at @xmath430 . the msto and slope of the rgb provide constraints on the age and metallicity , respectively . our distance , metallicity and age measurements agree very well with the findings of other authors based on spectroscopy and deeper imaging data . the star formation history shows one significant epoch of star formation in the oldest ( @xmath44 gyr ) bin , with no other bins containing a significant sfr . considering the ability of these fits to distinguish between a single and a multiple component system , demonstrated by the fits to the artificial cmds in section [ subsec : fakegc ] , we conclude , despite the sparseness of the cmd , that boo i very probably only contains old stars . thus , our results are consistent with the picture of a purely old , metal - poor stellar population . _ boo ii._ the number of boo ii member stars in the sdss data is very small and traces only the rgb . unfortunately , the resulting very low s / n combined with the degeneracies between the different parameters make it impossible for match to constrain the stellar population parameters in a statistically significant way . although @xcite show that the sdss cmd of boo ii is _ consistent _ with boo ii having similar metallicity , age , and distance as boo i , the sdss data alone do not support any significant constraints on any of these properties . the only way to get such constraints is by obtaining deeper follow - up data . _ cvn i._ being the brightest of the twelve objects analyzed here , cvn i has a well - populated rgb and hb . this helps to increase the robustness of the constraints we obtain from the sc fits . although at the edge of the sensitivity limits of sdss , the hb enables the determination of the distance . moreover , the extent of the hb to very blue colors requires the stars in cvn i to be both old and metal - poor . this constraint on the metallicity is further enhanced by the slope of the rgb . together , the extended hb and rgb thus can provide robust constraints on the properties of the dominant stellar population . the distance modulus we find is slightly smaller than that found in the literature , but not significantly so . the metallicity of [ fe / h]=-1.8@xmath330.2 falls nicely within the range of metallicities observed by @xcite . although an age of 14@xmath332 gyr fits the cmd best , the star formation history shows two peaks . the oldest star forming episode in the 10 - 16 gyr bin is dominant in terms of the number of stars and drives the sc fits , but the peak in the 2.5 - 4 gyr bin is equally significant . for the old stars the metallicity is found to be close to [ fe / h]=-2 , while the younger stars have metallicities around [ fe / h]=-1.5 . it should be noted that the young population found here is not caused by the algorithm attempting to fit bhb stars , as is the case for m15 in figure [ fig : gcsfh ] . stars belonging to this young population only appear in the regions of the cmd where the rgb and the red part of the hb are located . the presence of two populations is therefore purely based on the rgb morphology and the density of stars along the hb . this detection of an old and intermediate age population with these metallicities confirms the findings of @xcite and @xcite . furthermore , @xcite note the presence of a carbon star very close to the center of cvn i and its heliocentric radial velocity of 36@xmath3320 km / s is consistent with the radial velocities that @xcite and @xcite measure for cvn i , making it is likely that this star belongs to the cvn i dwarf . since carbon stars are often of intermediate age , this is another indication of the presence of an intermediate age population in this system . _ cvn ii._ although very faint and distant , our distance measurement for cvn ii is very accurate and in very good agreement with the value from @xcite due to the presence of hb stars . from the extension of the hb to very blue colors ( @xmath45 ) we can infer that the stellar population is both old and metal - poor . despite the sparseness of the cmd we thus get relatively good constraints on these parameters and we find a metallicity of [ fe / h]@xmath14 - 2 and a very high age . cvn ii thus appears to be a very old and metal - poor object . _ com._ an absence of a clear hb makes constraining the distance of com difficult and leads to a strong degeneracy between age , distance and metallicity . similar degeneracies occur for artificial cmd 1 in section [ subsec : fakegc ] and the gc ngc 6229 ( see figures [ fig : fakescfit ] and [ fig : gcsfh ] ) , even though their cmds are much less sparsely populated . the rgb and msto alone do not allow a robust determination of the individual population parameters . as a consequence , the distance modulus we recover has rather large error bars but is consistent with the findings of @xcite . the sc fits favor an ancient population ( @xmath44 gyr ) and a metallicity of [ fe / h]@xmath312 , but also here the uncertainties are large . _ her._ in the case of her the contamination of the cmd with fore- and background stars is severe , which is due to its location at lower galactic latitudes compared to the other objects . this high contamination means a low s / n of the cmd features of her , apart from the hb , which is in the relatively clean part of the cmd at @xmath46 . our distance modulus of 20.4@xmath330.2 , which largely is determined by the hb , is somewhat lower than estimated by @xcite , but is consistent within the error bars . this object too is consistent with an ancient and metal - poor ( [ fe / h]@xmath14 - 2 ) stellar population . even though the region of parameter space that lies within 1 @xmath8 of the best fit values is well constrained , the complete parameter space lies within 2 @xmath8 , which is due to the low s / n of the cmd . these results based only on sdss data are also consistent with the results obtained by applying the same methods to much deeper @xmath47 photometry from the large binocular telescope @xcite . in fact , the uncertainties on the values of distance , metallicity and age from the deeper data are very similar to the 1@xmath8 values we obtain here . lcccr bootes i ( booi ) & 18.8@xmath330.2 & -2.2@xmath330.2 & 14@xmath332 & @xmath1410 - 16 gyr + bootes ii ( booii ) & - & - & - & - + canes venatici i ( cvni ) & 21.5@xmath330.2 & -1.8@xmath330.2 & 14@xmath332 & @xmath142 - 4 gyr , @xmath1410 - 16 gyr + canes venatici ii ( cvnii ) & 20.8@xmath330.2 & -2.1@xmath330.3 & 14@xmath332 & - + coma berenices ( com ) & 18.4@xmath330.4 & -1.9@xmath48 & 11@xmath49 & - + hercules ( her ) & 20.4@xmath330.2 & -2.1@xmath330.2 & 14@xmath333 & - + leo iv ( leoiv ) & - & _ -2.1@xmath330.3 _ & _ 14@xmath332 _ & - + leo t ( leot ) & 23.1@xmath330.6 & -1.6@xmath330.6 & - & @xmath50 gyr , @xmath146 - 10 gyr + segue 1 ( seg1 ) & - & _ -1.6@xmath330.5 _ & _ 11@xmath334 _ & - + ursa major i ( umai ) & - & _ -1.8@xmath330.4 _ & _ 10@xmath335 _ & - + ursa major ii ( umaii ) & - & _ -1.5@xmath330.5 _ & _ 11@xmath334 _ & @xmath142 - 6 gyr , @xmath1410 - 16 gyr + willman 1 ( wil1 ) & 17.9@xmath330.4 & -1.6@xmath330.5 & 10@xmath335 & - + [ tab : newresults ] _ leo iv._ the cmd of leo iv is so sparse that a statistically significant determination of the population properties is not possible based on the sc fits alone . the best fit is given by a population of @xmath1414 gyr old stars with [ fe / h]=@xmath102.2 at a distance modulus of 20.8 , but a large number of models gives a goodness - of - fit that is only marginally worse . assuming a distance modulus of 21.0 @xcite , gives the solid contours in the age - metallicity panel for leo iv in figure [ fig : nsscfit2 ] . using this prior , the sc fits allow the age and metallicity to be constrained to 14@xmath332 gyr and [ fe / h]@xmath132.1 @xmath330.3 , respectively . _ leo t._ this object is very interesting and differs from the others analyzed here , in that it is the least luminous system known which shows signs of on - going star formation and has hi gas associated with it @xcite . because of the presence of two clearly very different populations the sc fitting method is not very well suited . even when fitting the two populations separately , by dividing the cmd in two distinct parts as we have done , the results are poor , due to the small numbers of stars in each part . taking the results of the two populations together we find a distance modulus of 23.1@xmath330.6 magnitudes . both the old and young populations are best fit with low metallicities , [ fe / h]@xmath51 - 1.5 . the star formation history we find ( figure [ fig : nssfh ] ) clearly shows two peaks , confirming the presence of two distinct populations . the age of the old population , not well - constrained with the sc fits , is put in the 6.5 - 10 gyr age bin . comparing this with the sfhs of the gcs in figure [ fig : gcsfh ] and the other sfhs in figure [ fig : nssfh ] , where in all cases the peak of star formation is in the oldest age bin , this implies that the old population in leo t is on average younger than the old population in these other objects . the young stars are clearly found in the three youngest age bins , where they are fit as young ( @xmath50 gyr ) blue loop stars . @xcite interpreted the nature of these stars similarly . _ seg1._ in seg 1 only a ms and msto are populated in the very sparse cmd . as with the artificial cmd 1 in section [ subsec : fakegc ] and the gcs m13 , m15 and m92 , where the sc fits also solely relie on the ms and msto , strong degeneracies between age , distance and metallicity are expected . indeed , the contour plots for seg 1 in figure [ fig : nsscfit2 ] show a similar morphology , but due to the sparseness of the cmd with even poorer constraints . this makes any determination of the population properties of seg 1 , based solely on the sc fits , impossible . even when using the distance modulus from @xcite as a prior , the age - metallicity degeneracy prevents strong constraints on these parameters , resulting in very large error bars on the values for seg 1 listed in table [ tab : newresults ] . _ uma i._ this object also has a very sparsely populated cmd . an rgb is readily seen , with some probable hb stars that , however , are too few to produce strong statistical constraints on the distance . because of this and the usual degeneracies , the significance of the sc fit results is very low , causing all parameters to be poorly constrained . assuming a distance modulus of 20 @xcite , our results are consistent with an old ( @xmath1410 gyr ) and metal - poor population , but with large error bars . _ uma ii._ although the msto of uma ii is well - populated , the scarcity of stars in other parts of the cmd provides problems for this object . there are some rc stars , but the s / n of this cmd feature is not high enough to make for a reliable distance determination . added to that there are strong degeneracies between the different solutions ; similar degeneracies are seen in section [ sec : gctests ] for the gc ngc 6229 where also the msto drives the sc fit results . our results are , however , consistent with previous findings . with a prior on the distance modulus , @xmath5217.5 , the fits agree with the conclusion of @xcite that uma ii is not very metal - poor , but seems to have a more intermediate metallicity of [ fe / h]@xmath531.5 , although the error bars are large ( see table [ tab : newresults ] ) . the spectroscopic metallicity measurement from @xcite is consistent with our value , but the one from @xcite is more metal - poor , but still consistent within the errors . while the sc fits imply that the majority of stars is likely old , @xmath1411 gyr , the sfh ( figure [ fig : nssfh ] ) hints at more complexity . according to the sfh results , there is a population of very old ( @xmath3010 gyr ) and metal - poor ( [ fe / h]@xmath54 - 2 ) , as well as a younger ( @xmath145 gyr ) , close to solar metallicity population . since the sc fits measure the average properties , this is consistent with the intermediate metallicity we find there . @xcite report that the upper main sequence of uma ii in their deeper subaru data is wider than expected just from observational errors , one explanation for which could be the presence of a mix of populations . also in our fit , the double peak results from the width of the msto , which is less well - fit by a single - age population . the exact ages and metallicities of the populations should be studied with deeper and spectroscopic data , since the sdss cmd does not support such detailed conclusions . however , taking our sfh fit and these other results together it does seem likely that uma ii , like cvn i , is a system with a complex star formation history , which might also be ( part of ) the explanation for the discrepancy between the two available spectroscopic metallicity measurements . _ wil1._ for this system only the msto is observed , which is also sparsely populated . this leads to a strongly degenerate solution space , similar to the cases of uma ii and seg 1 . the distance can be determined slightly better here , but the range of distance moduli for which fits within 1@xmath8 from the best fit can be obtained is still very large : 17.2@xmath5518.5 . using the distance modulus from @xcite ( m - m=17.9 ) as a prior , the fit results imply an old population ( @xmath1410 gyr ) with an intermediate to low metallicity ( [ fe / h]@xmath14 - 1.6 ) , but also here the error bars on these numbers are large . figure [ fig : nssummary ] summarizes our results from the match fits to the cmds for the sample of new milky way satellites . in the left panel we plot our metallicity estimates for all objects versus their absolute magnitudes , together with the spectroscopic metallicities from @xcite and @xcite . especially for the fainter objects our metallicity estimates are very poorly constrained ( figures [ fig : nsscfit1 ] and [ fig : nsscfit2 ] ) and because of the available isochrones our metallicity range is restricted to [ fe / h]@xmath132.4 at the metal - poor end . in light of this , our photometric results agree remarkably well with the spectroscopic values . still , care has to be taken when drawing conclusions from these metallicity estimates . we can say at best that we find no indication for a corellation between metallicity and the luminosity of the objects , which is observed for brighter dwarf spheroidals ( e.g. , * ? ? ? * ; * ? ? ? the mean age of the stars versus absolute magnitude is shown in the middle panel of figure [ fig : nssummary ] . for objects in which the sfh fits showed evidence for two populations , both populations are plotted separately . the right panel shows the inferred ages of the satellites versus distance modulus , to probe possible environmental influences . looking at the middle panel of figure [ fig : nssummary ] there is no clear evidence for a luminosity dependence of the ability of a dwarf galaxy to form stars recently . although the two most luminous objects in this sample , leo t and cvn i , both have young or intermediate age populations , the much fainter uma ii does as well . while the environment might play a role , the right panel of figure [ fig : nssummary ] shows no obvious relation between age or the presence of multiple star formation epochs with distance from the sun ( since these objects are at high galactic latitude and at large distances , the heliocentric and galactocentric distances are not significantly different ) . this is contrary to the trend suggested by @xcite for the brighter local group dwarf spheroidals , in which systems with a larger fraction of young or intermediate age stars are systematically more distant . since most of these recently detected objects are intrinsically faint and/or distant , they are of course much harder to detect by any overdensity detection scheme than the gcs . the improvement in the goodness - of - fit , @xmath41 , between a fit with a control field only and a control field plus sc model is roughly proportional to the number of stars in the cmd that can be fit by the sc model . this means that the objects with sparsely populated cmds are the most difficult to detect . comparing the overdensity detection results in figure [ fig : nsarea ] with the cmds of the objects ( figure [ fig : nscmds ] ) confirms this , as boo , cvn i and uma ii give by far the highest @xmath41 . similarly , the rms noise in @xmath41 increases with increasing numbers of field stars in the cmds , and at the low s / n end the noise level starts to play an important role . for example , comparing the panels for boo ii and cvn ii in figure [ fig : nsarea ] , which have the same @xmath41 scale , reveals very different noise levels . this can be traced back to the positions of these objects on the sky , with bootes ii lying behind the sagittarius stream and cvn ii in a much ` cleaner ' region of the halo . the sagittarius stream also produces a high noise level around segue 1 . while most objects produce clear peaks in @xmath41 , the detectability of several is rather low . for example , cvn ii , leo t , segue 1 and uma i have a @xmath41 only a factor of a few higher than the most prominent noise peaks ; boo ii is not clearly detected at all . when inspecting the values in table [ tab : detsig ] , it turns out that the use of a fine - tuned population model as compared to the generic old , metal - poor model has a varied effect . in some cases the detection significance is improved , in some cases the change has little effect and in a few cases the detectability is actually decreased . in these last cases , even though the peak @xmath41 is the same to or higher for the fine - tuned model , the change in the rms causes the detection to become less significant . this indicates that , while sampling a range of population models would be preferable when doing an overdensity search , a relatively small grid of models will not severely compromise the outcome . to test the significance of the @xmath42 measure , we tested the overdensity detection scheme on a presumably ` empty ' patch of sdss sky with 170@xmath56ra@xmath56200and 40@xmath56dec@xmath5665 . this patch of sky was divided into 20@xmath1120 boxes with some overlap , giving a total of 7088 boxes , equalling an effective search area of almost 800 square degrees . a control field ( the cmd from the surounding 5@xmath115area ) was fit to the cmd of each box , as well as this control field plus a typical old stellar population model with an age bin of 10.0 to 10.1 log - year and [ fe / h]=-2 . this was done seven times , with the stellar population model shifted to different distance moduli : 17.0 , 18.0 , 19.0 , 20.0 , 21.0 , 22.0 and 23.0 magnitudes . thus , for each box seven values of @xmath42 are obtained , where the rms is obtained from the boxes in the surounding 5@xmath115 . the methodology closely resembles the approach used for the mw satellites . figure [ fig : pksig ] shows the histogram of all @xmath42 values for all boxes ( solid line ) , as well as for two individual distance moduli , 18.0 ( dotted ) and 22.0 ( dashed ) mag . all three histograms show the same distribution , implying that the noise behaves similarly for all distance moduli . the tail of the distribution extends to @xmath57 . an inspection of the locations with values higher than 7 reveals no obvious overdensities , implying that this tail is indeed caused by noise . in order to obtain a clean sample of stellar overdensity detections , a @xmath42 threshold of @xmath149 seems appropriate . this means that when using a sufficiently sampled grid of population models and distances , all new mw satellites would be detected , with the exception of boo ii ( cf . table [ tab : detsig ] ) . the results of the fits to boo i with different box sizes , listed in table [ tab : detsig ] and shown in figure [ fig : booiareas ] , illustrate the effect of the spatial resolution of the cmd grid . here , the detectability is also strongly affected by the rms of the background noise . since the @xmath41 due to random variations in the field population increases with increasing numbers of field stars , the rms is higher for larger box sizes , favoring the smallest boxes used here . the fact that the @xmath58 decreases roughly by a factor of 2 as we move to larger box sizes is caused by the number of background stars increasing by a factor 4 , while the number of boo i stars in the central box does not change . it is also interesting to note that here the noise is dominated by the sagittarius stream , which runs accross the lower left corner of the panels in figure [ fig : booiareas ] . the significance of the detections in table [ tab : detsig ] can be boosted by decreasing the box sizes , especially for the objects with the smallest ( projected ) size . we have shown that application of match @xcite to hess diagrams or cmds of sdss data successfully recovers the known stellar population properties of globular clusters ( section [ sec : gctests ] ) , within the limitations of the data . this gives confidence that despite inaccuracies in the theoretical isochrones in isolated parts of the cmd , the overall cmd - fitting technique still provides good and trustworthy results . apart from the standard star formation history fitting method , we have also demonstrated three other ways of applying match to sdss data : + 1 ) identifying the best fitting ` single component ' stellar population , a mode useful for cmds of very faint or distant objects with limited s / n in their cmd ; + 2 ) fitting the line - of - sight distribution of stars to keep the fit s degrees of freedom constant , we fix the metallicity at each age bin ; and + 3 ) algorithmic detection of objects , i.e. of a localized stellar population overdensity in a particular ( unknown ) direction and at a particular ( unknown ) distance . this is done by comparing the difference in the goodness - of - fit between a cmd - fit to a control field and a cmd - fit to a control field plus model population . these tests show that the application of cmd - fitting techniques to wide - field survey data is a very promising endeavor in the light of sdss and other surveys that will start yielding data over the next few years . further improvements in software and theoretical isochrones can only increase the use and accuracy of such work . on the basis of these tests and the application to known globular clusters , we have applied our single component fitting technique to twelve recently discovered milky way satellites . the sparseness of several of the cmds of these objects inhibits a robust determination of the population parameters and the best fit solutions are often not unique . in order to obtain good constraints at least some cmd features such as the msto , hb and rgb should be observed with sufficiently high s / n . for the least luminous objects this is inherently problematic because of their lack of ( evolved ) stars . for only four of the satellites , the cmds had sufficiently high s / n to warrant a star formation history fit . the results of the population fits to the new milky way satellites , summarized in table [ tab : newresults ] , are consistent with the individual discovery and follow - up studies . when looking at the sample as a whole , we find that the majority of these objects are likely dominated by old ( @xmath59 10 gyr ) and metal - poor ( [ fe / h]@xmath14 - 2 ) stellar populations . for boo , cvn ii , com , and her the single component fits converge to this picture , and for boo the full fit of a star formation history also confirms this . seg 1 , uma i , uma ii , and wil 1 are not well constrained by the sc fits , but using the distances from the literature as priors , they are also consistent with similar ages and metallicities , although slightly higher metallicities ( [ fe / h]@xmath311.6 ) can not be excluded . the single component fits show cvn i to be old as well , but not extremely metal - poor . however , the sfh reveals the presence of both an old , metal - poor , as well as a younger ( @xmath513 gyr ) and more metal - rich population . this is consistent with the findings of @xcite , who detect two kinematically distinct populations in cvn i which differ in age and metallicity , and the detection of a young component by @xcite in deeper photometry . the sfh of uma ii also shows a peak at an age of @xmath145 gyr , which supports suggestion of multiple populations in this object by @xcite , who find the msto width of uma ii to be too large to be explained by photometric errors alone . finally , in leo t the presence of two distinct populations is readily seen in the cmd and reproduced in the star formation history . the rgb seen in this system belongs most likely to an intermediate age ( @xmath145 - 10 gyr ) population with a metallicity of [ fe / h]@xmath14 - 1.5 to -2 , while the blue stars in the cmd are most likely very young ( @xmath561 gyr ) blue loop stars with similar metallicities . the objects studied here are all intrinsically faint and sparse , and in some cases the population parameters can be constrained only poorly . however , their star formation histories do not appear uniform , nor do all of them appear to be exclusively old and metal poor . we thank matthew coleman for helpful discussions and the anonymous referee for his useful comments that helped to clarify this paper . j.t.a.d.j . acknowledges support from dfg priority program 1177 . d.b.z . was supported by a pparc - funded rolling grant position . thanks the deutsche forschungsgemeinschaft for their support through the emmy noether program . v.b . was supported by a pparc fellowship . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . aparicio , a. , gallart , c. & bertelli , g. 1997 , , 114 , 680 , j. , catelan , m. , ferraro , f. r. , spassova , n. , buonanno , r. , iannicola , g. , richtler , t. , & sweigart , a. v. 1999 , , 343 , 813 clem , j. l. 2006 , phd th . , univ . of victoria dallora , m. , et al . 2006 , , 653 , l109 de jong , j. t. a. , butler , d. j. , rix , h - w . , dolphin , & a. e. , martnez - delgado , d. 2007 , , accepted , astro - ph/0701140 dolphin , a. e. 1997 . , 2 , 397 dolphin , a. e. 2001 , , 332 , 91 duquennoy , a. & mayor , m. 1991 , , 248 , 485 fukugita , m. , ichikawa , t. , gunn , j. e. , doi , m. , shimasaku , k. , & schneider , d. p. 1996 , , 111 , 1748 gallart , c. , aparicio , a. , bertelli , g. , & chiosi , c. 1996 , , 112 , 1950 , l. , bertelli , g. , bressan , a. , chiosi , c. , groenewegen , m. a. t. , marigo , p. , salasnich , b. , & weiss , a. 2002 , , 391 , 195 girardi , l. , grebel , e. k. , odenkirchen , m. , & chiosi , c. 2004 , , 422 , 205 gunn , j.e . 1998 , , 116 , 3040 gunn , j.e . 2006 , , 131 , 2332 harris , w. e. 1996 , , 112 , 1487 harris , w. e. , et al . 1997 , , 114 , 1030 harris , j. & zaritsky , d. 2001 , , 136 , 25 hernandez , x. , gilmore , g. & valls - gabaud , d. 2000 , , 317 , 831 hogg , d.w . , finkbeiner , d.p . , schlegel , d.j . , & gunn , j.e . 2001 , aj , 122 , 2129 holtzman , j. a. et al . 1999 , , 118 , 2262 ibata , r. a. , chapman , s. , irwin , m. j. lewis , g. , & martin , n. f. 2006 , , 373 , l70 ivezi , . 2004 , an , 325 , 583 kroupa , p. , tout , c. a. & gilmore , g. 1993 , , 262 , 545 lupton , r. , gunn , j. , & szalay , a. 1999 , , 118 , 1406 muoz , r. r , carlin , j. l. , frinchaboy , p. m. , nidever , d. l. , majewski , s. r. , & patterson , r. j. 2006 , , 650 , l51 olsen , k. a. g. 1999 , , 117 , 2244 sarajedini , a. 1997 , , 113 , 682 schlegel , d. , finkbeiner , d. & davis , m. 1998 , , 500 , 525 smith , j. a. , et al . 2002 , , 123 , 2121 stoughton , c. et al . 2002 , aj , 123 , 485 tolstoy , e. & saha , a. 1996 , , 462 , 672 york , et al . , 2000 , , 120 , 1579 Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: causality is a topic which nowadays receives much attention . statisticians , epidemiologists , biostatisticians , social scientists , computer [ especially those in artificial intelligence , see , e.g. , @xcite ( @xcite ) ] , econometricians and philosophers are investigating questions like `` what would have happened if '' and `` what would happen if . '' this article discusses estimating the effect of a time - varying treatment . as a recurring example , this article focuses on the effect of a medical treatment which is adapted to a patient s state during the course of time . large observational studies have become widely used in medical research when data from randomized experiments are not available . randomized clinical trials are often expensive , impractical , and sometimes unfeasible for ethical reasons because treatment is withheld from some patients regardless of medical considerations . also , in some instances , exploratory investigations using nonexperimental data are used before conducting a randomized trial . in observational studies there is no pre - specified treatment protocol . data are collected on patient characteristics and treatments in the course of the normal interaction between patients and doctors . obviously , it is considerably more difficult to draw correct causal conclusions from observational data than from a randomized experiment . the main reason is the so - called confounding by indication or selection bias . for example , doctors may prescribe more medication to patients who are relatively unhealthy . thus , association between medication dose and health outcomes may arise not only from the treatment effect but also from the way the treatment was assigned . if this confounding by indication only takes place at the start of the treatment , one can condition on initial patient characteristics or covariates , such as blood pressure or number of white blood cells , in order to remove the effect of the confounding , and get meaningful estimates of the treatment effect . linear regression , logistic regression or cox regression can be used for this purpose . however , estimating treatment effects is more difficult if treatment decisions after the start of the treatment are adapted to the state of the patients in subsequent periods . treatment might be influenced by a patient s state in the past , which was influenced by treatment decisions before ; thus , simply conditioning on a patient s state in the past means disregarding information on the effect of past treatment . in such a case , even the well - known time - dependent cox model does not answer the question of whether , or how , treatment affects the outcome of interest . the time - dependent cox model studies the rate at which some event of interest happens ( e.g. , the patient dying ) , given past treatment- and covariate history . however , under time - dependent confounding , past covariate values may have been influenced by previous treatment . the net effect of treatment can thus not be derived from just this rate ; see , for example , @xcite , @xcite or @xcite . structural nested models , proposed in @xcite , lok , gill , van der vaart and robins ( @xcite ) and robins ( @xcite ) to solve practical problems in epidemiology and biostatistics , effectively overcome these difficulties and estimate the effect of time - varying treatments . the main assumption underlying these models is that all information the doctors used to make treatment decisions , and which is predictive of the patient s prognosis with respect to the final outcome , is available for analysis . this assumption of `` no unmeasured confounding '' makes it possible to distinguish between treatment effect and selection bias . what data have to be collected to satisfy this assumption of no unmeasured confounding is for subject matter experts to decide . all of the past treatment- and covariate information which both ( i ) influences a doctor s treatment decisions and ( ii ) is relevant for a patient s prognosis with respect to the outcome of interest , has to be recorded . in section 5 of @xcite and in section 8.1 of robins , rotnitzky and scharfstein ( @xcite ) , a sensitivity analysis methodology for estimation of structural nested models is developed that does not assume no unmeasured confounders . beyond treatment and covariates , the data requirements also include the measure of an outcome of interest ; for example , survival time , time to clinical aids or cd4 count after the treatment period . lok et al . ( @xcite ) study structural nested models in discrete time . these models assume that changes in the values of the covariates and treatment decisions take place at finitely many deterministic times , which are the same for all patients and known in advance . lok et al . ( @xcite ) also assume that covariates and treatment take values in a discrete space . they indicate why it is reasonable to expect consistency and asymptotic normality in discrete time , and they refer to @xcite for the proofs . @xcite generalize lok et al . ( @xcite ) to covariates and treatment taking values in @xmath0 . in this article we consider structural nested models in continuous time , proposed in robins ( @xcite ) . structural nested models in continuous time allow for both changes in the values of the covariates and treatment decisions to take place at arbitrary times for different patients . as noted in @xcite , structural nested models in continuous time assume that a short duration of treatment has only a small effect on the distribution of the outcome of interest . the effect of the treatment on an individual patient may be large , but then the probability of such effect has to be small for any particular short duration of treatment ( see page 7 , bottom ) . this article provides a conceptual framework and mathematical formalization of these practical methods , solving important outstanding problems and contributing to the causality discussion , especially for the time ordered and continuous time case . in particular , this article proves the conjectures in @xcite that structural nested models in continuous time lead to estimators which are both consistent and asymptotically normal . the proof simplifies considerably for structural nested models in discrete time ( see our discussion , section [ discu ] ) . this article also proves that a test related to the score test can be used to investigate whether treatment affects the outcome of interest without specifying a model for the treatment effect . the setting to which structural nested models in continuous time apply is as follows . the outcome of interest , from now on called @xmath1 , is a continuous real variable . for example , the survival time of a patient , time to clinical aids , or cd4 count after the treatment period . we wish to estimate the effect of treatment on the outcome @xmath1 . there is some fixed time interval @xmath2 $ ] , with @xmath3 a finite time , during which treatment and patient characteristics are observed for each patient . we suppose that after time @xmath3 treatment is stopped or switched to some kind of baseline treatment . in this article we assume that there is no censoring , and that the outcome @xmath1 is observed for every patient in the study . see , for example , @xcite , hernn et al . ( @xcite ) and @xcite for ideas about dealing with censoring . we denote the probability space by @xmath4 . the covariate process describes the course of the patient characteristics , for example , the course of the blood pressure and the white blood cell count . we assume that a realization of this covariate process is a function from @xmath2 $ ] to @xmath5 , and that such a _ sample path _ is continuous from the right with limits from the left ( cadlag ) . the covariates which _ must _ be included are those which both ( i ) influence a doctor s treatment decisions _ and _ ( ii ) possibly predict a patient s prognosis with respect to the outcome of interest . if such covariates would not be observed , the assumption of no unmeasured confounding , mentioned in the introduction , will not hold . for the moment consider one single patient . we write @xmath6 for the covariate- _ and _ treatment values at time @xmath7 . we assume that @xmath6 takes values in @xmath8 , and that @xmath9 is measurable for each @xmath10 $ ] . moreover , we assume that @xmath11 , seen as a function on @xmath2 $ ] , is cadlag . we write @xmath12 for the covariate- and treatment history until time @xmath7 , and @xmath13 for the space of cadlag functions from @xmath14 $ ] to @xmath8 in which @xmath15 takes it values . similarly , we write @xmath16 for the whole covariate- and treatment history of the patient on the interval @xmath2 $ ] , and @xmath17 for the space in which @xmath16 takes its values . in this article we choose the projection @xmath18-algebra as the @xmath18-algebra on @xmath13 and @xmath17 ; measurability of @xmath19 for each @xmath20 is then equivalent with measurability of the random variable @xmath15 . for technical reasons , we include in @xmath11 a counter of the number of jump times of the measured treatment- and covariate process . we suppose that observations on different patients are independent . structural nested models are models for relations between so - called counterfactuals . consider for a moment just one patient . in reality this patient received a certain treatment and had final outcome @xmath1 . if his or her actual treatment had been stopped at time t , the patient s final outcome would possibly have been different . the outcome he or she would have had in that case we call @xmath21 . of course , @xmath21 is generally not observed , because the patient s actual treatment after @xmath7 is usually different from no treatment ; it is a counterfactual outcome . instead of stopping treatment , one can also consider switching to some kind of baseline treatment , for example , standard treatment . figure [ figuur1 ] illustrates the nature of counterfactual outcomes . we suppose that all counterfactual outcomes @xmath21 , for @xmath10 $ ] and for all patients , are random variables on the probability space @xmath4 . to formalize the assumption of no unmeasured confounding , consider the history of a particular patient . decisions of the doctors at time @xmath7 may be based , in part , on recorded information on the state of the patient and treatment before @xmath7 , that is , on @xmath22 , but not on other features predicting the outcome of the patient . in particular , given @xmath23 , changes of treatment at time @xmath7 should be independent of @xmath21 , the outcome of the patient in case he or she would not have been treated after time @xmath7 , given @xmath23 . note that @xmath21 is an indication of the prognosis of the patient which does not depend on treatment decisions at or after time @xmath7 , since it is the counterfactual outcome which we would have observed if treatment would have been stopped at time @xmath7 . only if treatment would have no effect , the observed outcome @xmath1 could play this role . this is why robins assumption of no unmeasured confounding demands the independence , given @xmath23 , of treatment decisions at time @xmath7 and @xmath21 . similar conditions , though without time - dependence , can be found in , for example , @xcite . the statement `` changes of treatment at @xmath7 should be independent of @xmath21 , the outcome of the patient in case he or she would not have been treated after time @xmath7 , given @xmath23 '' is not a formal statement : it includes conditioning of null events ( since the probability that treatment changes at @xmath7 may be @xmath24 for every fixed @xmath7 ) on null events ( @xmath23 ) . to overcome this difficulty , we assume that the treatment process can be represented by or generates a ( possibly multivariate ) counting process @xmath25 . for instance , @xmath26 registers the number of treatment changes until time @xmath7 and/or the number of times treatment reached a certain level until time @xmath7 . a counting process constructed this way may serve as @xmath25 in the following . more about counting processes can be found in , for example , andersen et al . ( @xcite ) . we assume that the treatment process @xmath25 has an intensity process . formally , such an intensity process @xmath27 is a predictable process such that @xmath28 is a martingale . the intensity @xmath27 with respect to @xmath29 can be interpreted as the rate at which the counting process @xmath25 jumps given the past treatment- and covariate history @xmath23 . [ int ] @xmath25 has an intensity process @xmath27 on @xmath2 $ ] with respect to the filtration @xmath29 . this intensity process satisfies the following regularity conditions : @xmath30 is bounded by a constant which does not depend on @xmath31 , @xmath27 is continuous from the left . according to this assumption , @xmath32 is a martingale on @xmath2 $ ] with respect to the filtration @xmath29 . since most counting process martingale theory deals with filtrations @xmath33 which satisfy the usual conditions ( @xmath34 contains all null sets and @xmath35 ) , we mention that , under , @xmath36 is also a martingale with respect to @xmath37 , the usual augmentation of @xmath29 . this follows from lemma 67.10 in @xcite , since @xmath38 is cadlag . often , @xmath25 will be chosen to count the number of events of a certain type concerning the treatment process ( e.g. , the number of times treatment changed ) . at @xmath3 , the time the study ends , treatment is stopped or switched to baseline treatment , so a natural choice of @xmath25 will often jump at @xmath3 with positive probability . however , jumps of @xmath25 at @xmath3 are not useful for estimation , and we wish to avoid modeling jumps of @xmath25 at @xmath3 . therefore , we assume that , with probability @xmath39 , @xmath25 does not jump at @xmath3 , and if a natural choice of @xmath25 _ does _ jump at @xmath3 with positive probability , then we just adapt it , only at @xmath3 , so that it does not jump there . we also make the following assumption . [ ycadlag ] @xmath40 is a cadlag process . within this framework , the assumption of no unmeasured confounding could be operationalized as follows . the rate at which the counting process @xmath25 jumps given past treatment- and covariate history is also the rate at which @xmath25 jumps given past treatment- and covariate history and @xmath41 . that is , the following : [ intconf ] the intensity process @xmath27 of @xmath25 with respect to @xmath29 is also an intensity process of @xmath25 with respect to @xmath42 . this can be interpreted as conditional independence ( given @xmath23 ) of treatment decisions at time @xmath7 and @xmath43 . this assumption is stronger than just conditional independence of treatment decisions at @xmath7 and @xmath21 as assumed in robins ( @xcite ) . however , also @xmath44 for @xmath45 is an indication of the patient s prognosis upon which treatment decisions at time @xmath7 ( @xmath46 ) should not depend . assumption [ intconf ] allows us to use the usual counting processes framework . under assumption [ intconf ] , @xmath47}\lambda(s)\ , ds$ ] is a martingale also with respect to @xmath42 and its usual augmentation @xmath48 . this formalization of the assumption of no unmeasured confounding in terms of compensators with respect to the filtration @xmath42 is a novel feature of this paper relative to the previous literature on structural nested models . robins et al . ( @xcite ) , @xcite and @xcite use a cox model for initiation and/or changes in treatment . however , none of them formalized the assumption of no unmeasured confounders in terms of compensators with respect to @xmath42 . as a consequence , they could not use the extensive theory on counting process martingales to show the asymptotics of their estimators , which then remained without proof . structural nested models in continuous time model distributional relations between @xmath21 and @xmath49 , for @xmath50 small , through a so - called infinitesimal shift - function @xmath51 . write @xmath52 for the cumulative distribution function and @xmath53 for its generalized inverse @xmath54 then the infinitesimal shift - function @xmath51 is defined as @xmath55 the ( right - hand ) derivative of the quantile quantile transform which moves quantiles of the distribution of @xmath21 to quantiles of the distribution of @xmath49 ( @xmath56 ) , given the covariate- and treatment history until time @xmath7 , @xmath15 . notice that for differentiability of @xmath57 with respect to @xmath58 we need that a short duration of treatment has only a small effect on the distribution of the outcome of interest ( see page 3 , second paragraph ) , since @xmath59 must be equal to @xmath60 . [ daexa ] robins , blevins , ritter and wulfsohn ( @xcite ) describe an aids clinical trial to study the effect of azt treatment on survival in hiv - infected subjects . embedded within this trial was an essentially uncontrolled observational study of the effect of prophylaxis therapy for pcp on survival . pcp , pneumocystis carinii pneumonia , is an opportunistic infection that afflicts aids patients . the aim of robins et al . ( @xcite ) was to study the effect of this prophylaxis therapy on survival . thus , the outcome of interest @xmath1 is the survival time , and the treatment under study is prophylaxis for pcp . once treatment with prophylaxis for pcp started , it was never stopped . suppose that @xmath61 then ( see section [ cmim ] for details ) , for @xmath62 , withholding treatment from @xmath7 onward leads to ( with @xmath63 meaning `` is distributed as '' ) @xmath64\\[-8pt ] & = & e^{\psi}\cdot \mathit{dur}(t , y)+1\cdot \bigl(y - t-\mathit{dur}(t , y)\bigr)\qquad \mbox{given } \overline{z}_t,\nonumber\end{aligned}\ ] ] with @xmath65 the duration of treatment in the interval @xmath66 . thus , treated residual survival time ( @xmath7 until @xmath1 ) is multiplied by @xmath67 by withholding treatment ; compare this with accelerated failure time models , see , for example , @xcite . this multiplication factor @xmath67 should be interpreted in a distributional way . one of the models studied in robins et al . ( @xcite ) assumes that ( [ dasim ] ) is true even with @xmath63 replaced by @xmath68 ( though only for @xmath69 ) . notice that supposing ( [ dasim ] ) to be true with @xmath63 replaced by @xmath68 would be much stronger . [ daexa2 ] consider the situation from example [ daexa ] again . in another model mentioned in robins , blevins , ritter and wulfsohn ( @xcite ) the factor with which treated residual survival time is multiplied when treatment is withheld can depend on the azt treatment the patient received and whether or not the patient had a history of pcp prior to start of pcp prophylaxis . since this was a clinical trial for azt treatment , the azt treatment was described by a single variable @xmath70 indicating the treatment arm the patient was randomized to ( @xmath70 is @xmath24 or @xmath39 ) . whether or not the patient had a pcp history prior to start of prophylaxis is described by an indicator variable @xmath71 . @xmath71 equals @xmath39 if the patient had pcp before or at @xmath7 _ and _ before prophylaxis treatment started ; otherwise @xmath71 equals @xmath24 . if @xmath72 then ( see section [ cmim ] for details ) withholding prophylaxis treatment from @xmath7 onward leads to @xmath73 for @xmath62 . [ prior ] following @xcite , again consider survival as the outcome of interest . suppose that it is known that treatment received at time @xmath7 only affects survival for patients destined to die by time @xmath74 if they would receive no further treatment . an example would be a setting in which failure is death from an infectious disease , the treatment is a preventive antibiotic treatment which is of no benefit unless the subject is already infected and , if death occurs , it always does within five weeks from the time of initial unrecorded subclinical infection . in that case , as remarked in @xcite , the natural restriction on @xmath51 is that @xmath75 more biostatistical examples of models for @xmath51 can be found in , for example , @xcite , witteman et al . ( @xcite ) , @xcite and keiding et al . ( @xcite ) . @xmath76 can be interpreted as the infinitesimal effect on the outcome @xmath1 of the treatment actually given in the time - interval @xmath77 ( relative to baseline treatment ) . to be more precise , from the definition of @xmath51 we have @xmath78 in figure [ dfig ] this is sketched . @xmath79 in the picture is the @xmath80th quantile of the distribution of @xmath21 given @xmath15 . for @xmath50 , the @xmath80th quantile of the distribution of @xmath49 given @xmath15 is @xmath81 . thus , to shift from quantiles of the distribution of @xmath21 to the distribution of @xmath49 given @xmath15 ( @xmath50 ) is approximately the same as to just add @xmath82 to those quantiles . for example , if @xmath83 lies to the right of @xmath84 for @xmath50 , then treatment between @xmath7 and @xmath85 increases the outcome ( in distribution ) , and @xmath86 is greater than @xmath24 . . ] consider again this interpretation of @xmath51 as the infinitesimal effect of treatment given in @xmath77 . if the outcome of interest is survival , then @xmath76 should be zero if @xmath15 indicates the patient is dead at time @xmath7 . indeed , in that case @xmath83 and @xmath84 should be almost surely the same for every @xmath56 , since withholding treatment after death does not change the survival time . thus , @xmath87 is constant in @xmath58 for @xmath56 and , therefore , @xmath88 . however , this reasoning is not precise because of the complication of null sets . we will therefore just formally define @xmath76 to be zero if the outcome of interest is survival and @xmath15 indicates the patient is dead at time @xmath7 . it can be shown that @xmath89 if treatment does not affect the outcome of interest , as was conjectured in @xcite . to be more precise , @xcite shows that , for example , @xmath89 if and only if , for every @xmath50 and @xmath7 , @xmath49 has the same distribution as @xmath21 given @xmath15 . that is , @xmath89 if and only if `` at any time @xmath7 , whatever patient characteristics are selected at that time ( @xmath15 ) , stopping ` treatment as given ' at some fixed time after @xmath7 would not change the distribution of the outcome in patients with these patient characteristics . '' in the rest of this article @xmath90 will always indicate a correctly specified parametric model for @xmath51 , with @xmath89 if @xmath91 . define @xmath92 as the continuous solution to the differential equation @xmath93 with final condition @xmath94 , the observed outcome ( see figure [ dtfig ] ) . then @xmath92 mimics @xmath21 in the sense that it has the same distribution as @xmath21 given @xmath15 . this rather surprising result was conjectured in @xcite and proved in lok ( @xcite ) . to prove this result , we need the following consistency assumption . to the differential equation @xmath95 with final condition @xmath94 in case the outcome is survival time . ] [ inst ] @xmath96 has the same distribution as @xmath1 given @xmath97 . notice that since by assumption no treatment was given after time @xmath3 and since treatment is right - continuous , there is no difference in treatment between @xmath96 and @xmath1 . we suppose that this assumption holds , and we also suppose that a short duration of treatment has only a small effect on the distribution of the outcome of interest ( @xmath98 ) . under these assumptions and regularity conditions only , lok ( @xcite ) proved that indeed equation ( [ xdef ] ) has a unique solution for every @xmath99 , and that this solution @xmath92 mimics @xmath21 in the sense that @xmath92 has the same distribution as @xmath21 given @xmath15 ( see appendix [ xtap ] ) . throughout this article we will assume that this result holds true . [ daexai ] survival of aids patients ( continuation of example ) . suppose that @xmath100 then @xmath101 if @xmath102 , and @xmath103 for @xmath104 . suppose now that one has a correctly specified parametric model for the infinitesimal shift - function @xmath51 , @xmath90 . then one can calculate `` @xmath105 , '' the solution to @xmath106 with final condition @xmath107 . for the true @xmath108 , @xmath105 has the same distribution as @xmath21 , the outcome with treatment stopped at @xmath7 , even given all patient - information at time @xmath7 , @xmath15 . so instead of the unobservable @xmath21 s , we have the observable @xmath109 s which for the true @xmath108 mimic the @xmath21 s . although we do not know the true @xmath108 , this result turns out to be very useful , both for estimating @xmath108 ( sections [ ees ] and [ asnorm ] ) and for testing [ section [ test ] ; notice that when testing whether treatment affects the outcome ( i.e. , whether @xmath89 ) , @xmath110 can simply be calculated from the data ( @xmath111 ) under the null hypothesis of no treatment effect ] . in this appendix we present conditions under which @xmath92 mimics @xmath21 in the sense that it has the same distribution as @xmath21 given @xmath15 . this result is used heavily in this article . for the proofs , which are lengthy and use discretization , we refer to lok ( @xcite ) . section [ xtapt ] deals with survival outcomes , section [ xtapy ] with other outcomes . survival outcomes require a different set of assumptions , as will become clear below . the conditions here are somewhat more restrictive than the ones in lok ( @xcite ) , but they are simpler . this section contains a sufficient set of regularity conditions to have existence and uniqueness of a solution @xmath92 to ( [ xdef ] ) , @xmath507 with final condition @xmath94 , the observed outcome ( see figure [ dtfig ] ) . furthermore , together with assumption [ inst ] ( consistency ) , they imply that @xmath92 has the same distribution as @xmath21 given @xmath15 . the regularity conditions below should be read as the following : there exist conditional distribution functions @xmath83 such that all these assumptions are satisfied . they can be relaxed to @xmath58 in a neighborhood of @xmath24 , if this neighborhood does not depend on @xmath16 . we only consider @xmath56 , so the derivative with respect to @xmath58 at @xmath508 is always the right - hand derivative . [ sup ] ( _ support _ ) . there exist finite numbers @xmath509 and @xmath510 such that all @xmath83 have the same bounded support @xmath511 $ ] . all @xmath512 have a continuous nonzero density @xmath513 on @xmath514 $ ] . there exists an @xmath216 such that @xmath515 for all @xmath514 $ ] , @xmath99 and @xmath10 $ ] . ( _ smoothness _ ) . for every @xmath99 , @xmath516 is differentiable with respect to @xmath7 , @xmath79 and @xmath58 with continuous derivatives on @xmath511\times[t_1,t_2)\times { \mathbb r}$ ] if @xmath11 does not jump in @xmath239 , with a continuous extension to @xmath511\times[t_1,t_2]\times [ 0,\infty)$ ] . the derivatives of @xmath512 with respect to @xmath79 and @xmath58 are bounded by constants @xmath315 and @xmath317 , respectively . @xmath517 and @xmath518 have derivatives with respect to @xmath79 which are bounded by constants @xmath519 and @xmath520 , respectively . the support conditions may be restrictive for certain applications . nevertheless , most real - life situations can be approximated this way , since @xmath509 and @xmath510 are unrestricted and @xmath216 is unrestricted , too . although the support conditions may well be stronger than necessary , they simplify the analysis considerably and , for that reason , they are adopted here . the smoothness conditions allow for nonsmoothness where the covariate- and treatment process @xmath11 jumps . this is important , since if the covariate- and treatment process @xmath11 jumps , this can lead to a different prognosis for the patient and thus to nonsmoothness of the functions concerned . [ thm1 ] suppose that regularity condition is satisfied . then @xmath76 exists . furthermore , for every @xmath99 , there exists exactly one continuous solution @xmath92 to @xmath95 with final condition @xmath94 . if also assumption ( consistency ) is satisfied and there are no more than finitely many times @xmath7 for which the probability that the covariate- and treatment process jumps at @xmath7 is greater than @xmath24 , then this @xmath92 has the same distribution as @xmath21 given @xmath15 for all @xmath10 $ ] . for a proof we refer to lok ( @xcite ) . this section contains a sufficient set of regularity conditions to have existence and uniqueness of a solution @xmath92 to equation ( [ xdef ] ) , @xmath95 with final condition @xmath94 , the observed outcome ( see figure [ dtfig ] ) . furthermore , together with assumption [ inst ] ( consistency ) and assumptions [ ttcons ] and [ ttinst ] below , they imply that @xmath92 has the same distribution as @xmath21 given @xmath15 . the conditions here are natural conditions if the outcome of interest @xmath1 is a survival time . as compared to section [ xtapy ] , we make two extra assumptions . the first is a consistency assumption , stating that stopping treatment after death does not change the survival time . the second assumption states that there is no instantaneous effect of treatment at the time the patient died [ notice that the difference between @xmath521 , the outcome with treatment stopped at the survival time @xmath1 , and @xmath1 is in treatment _ at _ @xmath1 ] . [ ttcons ] @xmath523 on @xmath524 . [ ttinst ] @xmath525 under these assumptions , treatment in the future does not cause or prevent death at present or before : [ conslem ] under assumptions and : for all @xmath56 : @xmath526 on @xmath527 . for all @xmath528 with @xmath529 and @xmath530 . for a proof we refer to @xcite . if the outcome is survival , the support condition in assumption [ xtapy ] , saying that all @xmath83 have the same bounded support @xmath511 $ ] , will not hold . the reason for this is as follows . @xmath15 includes the covariate - measurements and treatment until time @xmath7 . if covariates and treatment were measured at time @xmath7 , it can not be avoided to include in @xmath15 whether or not a patient was alive at time @xmath7 . given that a patient is dead at time @xmath7 and given his or her survival time , the distribution of this survival time can not have the fixed support @xmath511 $ ] , which is independent of @xmath7 . also , given that a patient is alive at time @xmath7 , this is hardly ever the case ; one often expects that @xmath7 is the left limit of the support . thus , in case the outcome is survival , the support condition for theorem [ thm1 ] has to be slightly changed . [ tsup ] there exists a finite number @xmath531 such that : for every @xmath99 and @xmath7 with @xmath102 , all @xmath83 for @xmath56 have support @xmath532 $ ] . for every @xmath99 and @xmath7 with @xmath102 , all @xmath512 for @xmath56 have a continuous nonzero density @xmath513 on @xmath533 $ ] . there exists a number @xmath216 such that , for all @xmath99 and @xmath7 with @xmath102 , @xmath534 for @xmath535 $ ] . next we look at the differentiability conditions in assumption [ sup ] . it does not seem reasonable to assume that @xmath512 is continuously differentiable with respect to @xmath58 and @xmath79 on @xmath536 $ ] since , for @xmath529 , @xmath537 [ lemma [ conslem](b ) ] . therefore , the derivative of @xmath512 with respect to @xmath58 is likely not to exist at @xmath538 ( and is equal to zero for @xmath539 ) . also , the derivative of @xmath512 with respect to @xmath79 may not exist at @xmath538 , because of the different treatment before and after @xmath85 . for survival outcomes , we replace the smoothness conditions of assumption [ sup ] by the following : [ tsmooth ] for every @xmath99 : if @xmath11 does not jump in @xmath239 and @xmath540 , the restriction of @xmath541 to @xmath542\times[t_1,t_2)\times { \mathbb r}_{\geq 0}\dvtx y\geq t+h\}$ ] is @xmath543 in @xmath528 . the derivatives of @xmath512 with respect to @xmath79 and @xmath58 are bounded by constants @xmath315 and @xmath317 , respectively , for @xmath533 $ ] . @xmath517 and @xmath518 have derivatives with respect to @xmath79 which are bounded by constants @xmath519 and @xmath520 , respectively , for @xmath533 $ ] . the smoothness condition above concentrates on @xmath544 . for @xmath545 we can choose @xmath537 because of lemma [ conslem](b ) . because of assumption [ inst ] ( consistency ) , @xmath546 has the same support as @xmath547 , so @xmath546 has support @xmath532 $ ] if @xmath102 [ assumption [ tsup](a ) ] . assume the following [ ytsup ] for all @xmath99 and @xmath7 with @xmath102 , @xmath548 is continuous and strictly increasing on its support @xmath532 $ ] . [ tthm1 ] suppose that regularity conditions and are satisfied . then @xmath76 exists . furthermore , for every @xmath99 , there exists exactly one continuous solution @xmath92 to @xmath95 with final condition @xmath94 . if also assumptions and ( consistency and no instantaneous treatment effect at time of death ) are satisfied , then this @xmath92 has the same distribution as @xmath21 given @xmath15 for all @xmath10 $ ] . for a proof we refer to lok ( @xcite ) . previous applications of structural nested models [ see , e.g. , robins et al . ( @xcite ) , @xcite ( @xcite ) , witteman et al . ( @xcite ) and keiding et al . ( @xcite ) ] have assumed the so - called local rank preservation condition . local rank preservation states that @xmath21 is a local solution to ( [ xdef ] ) . however , if @xmath21 is locally a solution to ( [ xdef ] ) , it is usually also globally a solution to ( [ xdef ] ) ; see , for example , theorem [ isun ] in the . hence , if one knew the parameter @xmath108 , every @xmath21 would be a deterministic function of the observed data . deterministic dependence of counterfactuals on the observed data is a very strong condition , which , though untestable , is generally considered implausible . the previous literature [ see , e.g. , @xcite ( @xcite ) , robins et al . ( @xcite ) , @xcite ( @xcite ) and keiding et al . ( @xcite ) ] acknowledged this problem , and conjectured that the assumption of local rank preservation could be relaxed in continuous time [ since it is known that the assumption of local rank preservation can be relaxed for structural nested models in discrete time ; this was pointed out by @xcite ( @xcite ) , and lok et al . ( @xcite ) provided a proof ] . see @xcite for a more elaborate discussion . the following example describes a setting where the assumption of local rank preservation is implausible . in the situation of example [ daexa ] , consider the following thought experiment . suppose that two patients had the same covariate history until time @xmath7 , and both received the same constant treatment to prevent pcp until time @xmath7 ( equal @xmath15 ) . suppose , furthermore , that both patients received no treatment after time @xmath7 , that they did not have pcp before time @xmath7 and both died at the same time @xmath112 ( for both , @xmath113 ) . possibly , the first patient would have had pcp at some time @xmath45 and would have died from it before @xmath114 in case he or she would not have been treated . possibly , the other patient would not have had pcp in case he or she would not have been treated , and would have died at the same time @xmath112 as without treatment . thus , it is easy to imagine that these patients would have had different outcomes under no treatment ( different @xmath115 ) . however , the assumption of local rank preservation excludes this possibility . local rank preservation is a very strong condition , for which structural nested models have previously been attacked . in fact , this article shows that the assumption of local rank preservation is not needed for structural nested models . however , proofs would be much easier under rank preservation ; for details , see the remarks before the proofs of theorems [ thec ] and [ mgc ] . see also @xcite for a more informal reasoning . to estimate the infinitesimal shift - function @xmath51 , @xcite proposes to use a ( semi-)parametric model to predict future treatment ( @xmath25 in our case ) on the basis of past treatment- and covariate history @xmath23 . this may seem odd , since prediction of treatment is not what we are interested in . however , we will show that such a model to predict treatment changes can indeed be a tool to get unbiased estimating equations for the parameter @xmath108 in the model for @xmath51 . moreover , often doctors may have a better understanding , at least qualitatively , about how decisions about treatment were made than about the effect of the treatment . in what follows we will assume that @xmath116 is a correctly specified parametric model for the intensity @xmath30 of @xmath25 . recall from section [ nounsec ] that , under no unmeasured confounding ( assumption [ intconf ] ) , @xmath21 does not contain information about treatment changes given past treatment- and covariate history @xmath23 . since @xmath92 has the same distribution as @xmath21 given @xmath15 , one could expect that also @xmath92 does not contain information about treatment changes given @xmath23 . unfortunately , this reasoning is not precise : we have to somehow deal with null sets since the probability that treatment changes at @xmath7 given past covariate- and treatment history is often equal to @xmath24 for each @xmath7 . in section [ mgs ] we will show how this can be dealt with . in the current section we present a class of unbiased estimating equations for @xmath117 and @xmath108 . these will be used for the proof in the next section , but they are also of interest in their own . in section [ mgs ] we will see that these estimating equations are in fact martingales , for the true parameters @xmath118 and @xmath119 . recall from section [ nounsec ] that , under no unmeasured confounding , we have the martingale @xmath47}\lambda(s)\,ds$ ] with respect to the filtration @xmath42 and its usual augmentation . from this martingale we can construct a whole family of martingales . if @xmath120 is a @xmath48-predictable process , then , under regularity conditions , @xmath121 is a martingale with respect to @xmath48 . for a more formal statement , we first make sure that @xmath120 is predictable . we put the following restriction on the functions @xmath122 we consider here : [ hreg ] when in this section we consider functions @xmath122 from @xmath123 , we assume that they are measurable and satisfy the following : @xmath122 is bounded by a constant which does not depend on @xmath7 and @xmath16 , for all @xmath124 $ ] , @xmath125 and @xmath99 , @xmath126 when @xmath127 and @xmath128 . for such @xmath122 , @xmath120 is a @xmath48-predictable process : [ hpred ] suppose that @xmath40 is cadlag assumption . then @xmath129 is a @xmath48-predictable process for any @xmath122 satisfying restriction . proof @xmath120 is adapted . it is also left - continuous : because @xmath40 is cadlag , @xmath130 exists . thus , we come to the following lemma : [ lemec ] under assumptions bounded intensity process @xmath40 cadlag and [ intconf ] ( no unmeasured confounding ) , @xmath131 is a martingale on @xmath2 $ ] with respect to @xmath48 for all @xmath122 satisfying restriction . proof @xmath132 is a martingale on @xmath2 $ ] with respect to @xmath48 , because of assumption [ intconf ] . it is of integrable variation . because of lemma [ hpred ] , @xmath133 is a @xmath48-predictable process . it is also bounded [ restriction [ hreg](a ) ] . hence,@xmath134 is an integral of a bounded predictable process with respect to a martingale of integrable variation , and , therefore , a @xmath48-martingale . to construct unbiased estimating equations for @xmath135 , we need to assume that the probability that @xmath136 and @xmath40 jump at the same time is zero . this assumption is a formalization of the assumption of no instantaneous treatment effect as proposed in @xcite , which can be seen as follows . given @xmath23 and @xmath137 , @xmath25 jumps at @xmath7 with rate @xmath27 ( assumption [ intconf ] , no unmeasured confounding ) . @xmath40 is a cadlag process ( assumption [ ycadlag ] ) , which thus for every @xmath99 jumps at most countably many times on the finite time interval @xmath2 $ ] . therefore , if @xmath40 and @xmath25 would jump at the same time with positive probability , this would imply a dependence of these jumps ; the obvious interpretation of this dependence would be that a change of treatment instantaneously affects the outcome of interest . [ jump ] the probability that there exists a @xmath7 such that @xmath136 and @xmath40 both jump at time @xmath7 is @xmath24 . notice that this excludes estimation of the effect of point exposures . for example , if treatment is surgery or another point exposure given at some time @xmath7 , the outcome under `` treatment stopped at time @xmath7 '' will typically jump at time @xmath7 if treatment affects the outcome of interest , at the same time as the treatment itself . however , this assumption does not exclude the possibility that the outcome differs depending on whether a patient is treated or not at a certain point in time . for example , @xmath138 and @xmath139 may be different when a virus is contacted at time @xmath7 . the model in this article can accommodate differences between @xmath138 and @xmath139 , as long as the probability that the observed treatment changes is @xmath24 at that precise time . or , in more generality , as long as the probability that @xmath25 jumps at the same time is @xmath24 . this was previously noticed in @xcite , section 8 . the estimating procedures in this article do not deal with instantaneous treatment effects . suppose that the above conditions hold and that @xmath140 for @xmath10 $ ] ( see section [ cmim ] ) . then if @xmath141 and @xmath116 are correctly specified ( parametric ) models for @xmath51 and @xmath30 , respectively , each choice of @xmath122 satisfying restriction [ hreg ] leads to an unbiased estimating equation for both the parameter of interest @xmath108 and the ( nuisance ) parameter @xmath117 : [ thec ] suppose that assumptions bounded intensity process@xmath40 cadlag ( no unmeasured confounding ) and ( no instantaneous treatment effect ) are satisfied . suppose also that , for every @xmath10 $ ] , @xmath92 has the same distribution as @xmath21 given @xmath15 . then @xmath142 for each @xmath122 satisfying restriction . thus , if @xmath90 and @xmath116 are correctly specified parametric models for @xmath51 and @xmath30 , respectively , @xmath143 with @xmath144 the empirical measure @xmath145 , is an unbiased estimating equation for @xmath146 , for each @xmath122 satisfying restriction . @xmath122 here is allowed to depend on @xmath108 and @xmath117 , as long as it satisfies restriction for @xmath146 . as before , @xmath109 here is the continuous solution of ( [ xpsi ] ) , @xmath147 with boundary condition @xmath107 . moreover , as before , we assume that for all @xmath90 we have existence and uniqueness of such solutions ; theorem [ isun ] in the appendix provides sufficient conditions for that . under local rank preservation ( see section [ loc ] ) , @xmath148 for each @xmath7 . in that case theorem [ thec ] follows immediately from lemma [ lemec ] . however , as argued in section [ loc ] , local rank preservation is generally considered implausible . proof of theorem [ thec ] we have to show that @xmath149 has expectation zero for all @xmath122 satisfying restriction [ hreg ] . to do that , we prove that it has the same expectation as @xmath150 which has expectation zero because of lemma [ lemec ] . we will first show that the terms with @xmath151 have the same expectation , that is , @xmath152 after that we show that the terms with @xmath153 have the same expectation , that is , @xmath154 as we will see below , ( [ firste ] ) and ( [ seconde ] ) have to be proved separately , since we do not have or expect that @xmath155 for @xmath45 ; we only have this for @xmath156 . therefore , the approximations below have to be chosen carefully . at first we prove ( [ firste ] ) , by approximating these sums and showing that the approximations have the same expectation . next we show that the approximations converge and that ( [ firste ] ) follows with lebesgue s dominated convergence theorem . define @xmath157 , @xmath158 , etc . , the jump times of the counting process @xmath25 in the interval @xmath2 $ ] . they are measurable [ e.g. , because of @xcite ( @xcite ) , lemma 74.4 ] . note that the number of jumps in @xmath2 $ ] is almost surely finite because @xmath25 is integrable ( it has a bounded intensity process ) . in the following read @xmath159 if there is no @xmath160th jump of @xmath25 in the interval @xmath2 $ ] . next split up the interval @xmath2 $ ] in intervals of equal length : for @xmath161 fixed , put @xmath162 , @xmath163 . fix @xmath164 for the moment . the right - hand side of equation ( [ firste ] ) is harder to approximate than the left - hand side , both because @xmath21 does not need to be continuous in @xmath7 while @xmath92 does and because knowing @xmath21 and @xmath15 does not imply knowing @xmath44 for @xmath45 and we do not have or expect @xmath155 for @xmath45 . the approximations we choose are @xmath165\\[-8pt ] \qquad & \approx & \sum_{j=1}^\infty\sum_{k=0}^{k-1 } 1_{(\tau_k,\tau_{k+1}]}(t_j ) h_{\tau_{k}}\bigl(y^{(\tau_{k+1})},\overline{z}_{\tau_{k}-}\bigr)\nonumber\end{aligned}\ ] ] and @xmath166\\[-8pt ] \qquad & \approx & \sum_{j=1}^\infty\sum_{k=0}^{k-1 } 1_{(\tau_k,\tau_{k+1}]}(t_j ) h_{\tau_{k}}\bigl(x(\tau_{k+1}),\overline{z}_{\tau_{k}-}\bigr).\nonumber\end{aligned}\ ] ] to show that these approximations have the same expectation , we use that @xmath167 . therefore , also @xmath168}(t_j ) h_{\tau_{k}}(x(\tau_{k+1}),\overline{z}_{\tau_{k-}})\sim 1_{(\tau_k,\tau_{k+1}]}(t_j ) h_{\tau_{k}}\bigl(y^{(\tau_{k+1})},\overline{z}_{\tau_{k-}}\bigr)\ ] ] [ notice that @xmath169}(t_j)$ ] is a function of @xmath170 . hence , the expectation of each of the terms on the right - hand side of ( [ apy ] ) is equal to the expectation of the corresponding term on the right - hand side of ( [ apx ] ) . since @xmath122 is bounded [ restriction [ hreg](a ) ] and the expected number of jump times @xmath171 is finite ( @xmath25 is integrable ) , this implies that the expectation of the right hand - side of equation ( [ apy ] ) is equal to the expectation of the right - hand side of equation ( [ apx ] ) . equation ( [ firste ] ) follows if the expectation of the approximations in ( [ apy ] ) and ( [ apx ] ) converges to the right - hand side and left - hand side of equation ( [ firste ] ) , respectively . this convergence is harder to show for ( [ apy ] ) than for ( [ apx ] ) , since @xmath40 may jump [ while @xmath172 does not , by construction ] . fix @xmath160 for a moment . define @xmath173 and @xmath174 as the grid points such that @xmath175 $ ] . as @xmath176 , @xmath177 , so that since @xmath178 is cadlag , @xmath179 . moreover , as @xmath180 , @xmath181 , so that because of restriction [ hreg ] on @xmath58 , @xmath182 . combining this for all @xmath160 leads to @xmath183}(t_j ) h_{\tau_{k}}\bigl(y^{(\tau_{k+1})},\overline{z}_{\tau_{k-}}\bigr ) \rightarrow \sum_{j=1}^\infty h_{t_j}\bigl(y^{(t_{j})},\overline{z}_{t_j-}\bigr)\ ] ] as @xmath184 for every @xmath31 for which the number of jumps of @xmath25 is finite , so for almost every @xmath99 . @xmath122 is bounded [ restriction [ hreg](a ) ] and the left - hand side is bounded by the number of jumps of @xmath25 times this bound . the expectation of that is finite because @xmath25 is integrable . thus , lebesgue s dominated convergence theorem can be applied , and @xmath185}(t_j ) h_{\tau_{k}}\bigl(y^{(\tau_{k+1})},\overline{z}_{\tau_{k}-}\bigr)\biggr ) \rightarrow e\biggl(\sum_{j=1}^\infty h_{t_{j}}\bigl(y^{(t_{j})},\overline{z}_{t_j-}\bigr)\biggr)\ ] ] as @xmath184 . because with probability one @xmath40 and @xmath25 do not jump at the same time ( assumption [ jump ] of no instantaneous treatment effect ) , @xmath186 so that we can replace @xmath187 by @xmath188 on the right - hand side of ( [ naarplus ] ) . therefore , indeed , the expectation of the approximation in ( [ apy ] ) converges to the expectation of the left - hand side of ( [ apy ] ) . the same reasoning shows this for ( [ apx ] ) . here less caution is necessary since @xmath92 is continuous in @xmath7 . that concludes the proof of equation ( [ firste ] ) . next we prove ( [ seconde ] ) , also by approximation . here , too , we show that the approximations have the same expectation and that ( [ seconde ] ) follows with lebesgue s dominated convergence theorem . divide the interval @xmath2 $ ] as above . the approximations we choose here are @xmath189 and @xmath190 because @xmath191 and @xmath192 are a measurable function of @xmath193 ( assumption [ int ] , bounded intensity process ) , the expectation of each of the terms in ( [ apy2 ] ) is equal to the expectation of the corresponding term in ( [ apx2 ] ) . thus , the expectations of these approximations are equal . equation ( [ seconde ] ) follows if the expectation of the approximations in ( [ apy2 ] ) and ( [ apx2 ] ) converge to the right - hand side and left - hand side of equation ( [ seconde ] ) , respectively . this convergence is also harder to show for ( [ apy2 ] ) than for ( [ apx2 ] ) because of possible discontinuities of @xmath40 . first notice that as @xmath184 , for @xmath7 fixed , @xmath194}(t ) h_{\tau_k}\bigl(y^{(\tau_k)},\overline{z}_{\tau_k-}\bigr ) \lambda(\tau_k)\rightarrow h_{t}\bigl(y^{(t-)},\overline{z}_{t-}\bigr ) \lambda(t)\ ] ] for every @xmath99 fixed and for every @xmath195 : @xmath40 has limits from the left ( assumption [ ycadlag ] ) , so that as @xmath196 , restriction [ hreg](b ) on @xmath58 can be used , and @xmath30 is continuous from the left [ assumption [ int](b ) ] . taking integrals and applying lebesgue s dominated convergence theorem [ @xmath122 and @xmath30 are bounded because of restriction [ hreg](a ) and assumption [ int](a ) , resp . ] leads to @xmath197 for every @xmath99 . as both @xmath58 and @xmath30 are bounded , lebesgue s dominated convergence theorem guarantees that indeed the expectation of the approximation in ( [ apy2 ] ) converges to the expectation of the left - hand side of ( [ apy2 ] ) . the same reasoning shows this for ( [ apx2 ] ) , which concludes the proof of equation ( [ seconde ] ) and theorem [ thec ] . @xcite shows that if the rest of the conditions in this section are satisfied , assumption [ jump ] ( treatment does not instantaneously affect the outcome of interest ) is a necessary condition for theorem [ thec ] . [ wei ] consider the setting of examples [ daexa ] and [ daexa2 ] and define @xmath198 if prophylaxis treatment started at or before time @xmath7 and @xmath24 otherwise . suppose that initiation of prophylaxis treatment can be correctly modeled with the time - dependent weibull proportional hazards model @xmath199 where @xmath200 equals @xmath39 if the patient had pcp before time @xmath7 and @xmath24 otherwise , and @xmath201 and @xmath202 are greater than zero [ for more about the weibull proportional hazards model and its applications see , e.g. , @xcite ( @xcite ) ] . if the patient died before @xmath7 or prophylaxis treatment already started before , the patient is not `` at risk '' for initiation of treatment and , thus , @xmath30 equals @xmath24 . then the ( partial ) score equations for estimation of @xmath203 are @xmath204 such estimating equations can also be written down for the model including @xmath205 , @xmath206 @xcite proposes to estimate the parameters in a model like this by choosing those parameters @xmath207 which maximize the likelihood when @xmath208 is considered fixed and known , and for which @xmath209 : for the true @xmath108 , @xmath210 does not contribute to the model for treatment changes ( under no unmeasured confounding ) . to make the connection with the estimators in the current article , notice that this leads to the same estimators as the ones that solve the estimating equations arising from the likelihood when @xmath208 is considered fixed and known , with @xmath211 put to zero . more precise , since we know that the true @xmath211 is equal to @xmath24 , we put @xmath211 equal to @xmath24 and get the estimating equations @xmath212 for the parameter @xmath108 ( and thus also for @xmath51 ) and the ( nuisance ) parameters @xmath203 . these estimating equations are of the form of theorem [ thec ] , @xmath213 but the function @xmath122 here is not bounded and @xmath30 need not be bounded ( if @xmath214 ) , so unbiasedness does not follow immediately from theorem [ thec ] . however , we could restrict the interval @xmath2 $ ] to @xmath215 $ ] for @xmath216 ( to assure that @xmath30 is bounded ) and @xmath217 can be approximated by the bounded functions @xmath218 ( @xmath219 ) ( to make @xmath122 bounded ) , which all lead to unbiased estimating equations because of theorem [ thec ] . the above estimating equations are then also unbiased because of lebesgue s dominated convergence theorem [ the dominating function is integrable since @xmath220 which is finite since @xmath221 . under the model for @xmath51 of example [ daexa ] , @xmath222 if the patient did not die before time @xmath7 . in that case these are five unbiased estimating equations for five unknown parameters . if the parameter @xmath108 is of dimension greater than @xmath39 , more unbiased estimating equations can be constructed by adding more terms of the form @xmath223 . we show that , under no unmeasured confounding , just as @xmath21 , @xmath92 does not predict treatment changes , given past treatment- and covariate history @xmath23 . we could hope for that since @xmath224 given @xmath15 ( see section [ cmim ] ) . the formal statement is ( compare with assumption [ intconf ] , no unmeasured confounding ) the following : the intensity process @xmath27 of @xmath25 with respect to @xmath29 is also the intensity process of @xmath25 with respect to @xmath225 . then @xmath226 is also a martingale with respect to @xmath225 . that will be useful later when we study the behavior of estimators @xmath227 and @xmath228 which are constructed with estimating equations of the form of theorem [ thec ] , @xmath229 . for example , we can use the fact that usually @xmath230}h(s)\,dm(s)$ ] is a martingale if @xmath38 is a martingale and @xmath231 a predictable process ; a sufficient condition for this is that @xmath232 . before going on , we first clarify why @xmath233 is indeed a filtration . for @xmath45 , @xmath234 is a deterministic ( though unknown ) function of @xmath235 ( i.e. , if solutions to the differential equation with @xmath51 are unique ; see , e.g. , theorem [ isun ] in the ) . similarly , for @xmath45 , @xmath92 is a deterministic function of @xmath236 . in the rest of this article we will assume that these functions are measurable functions on @xmath237 ( sufficient conditions for that are that the infinitesimal shift - function @xmath51 satisfies regularity assumption [ dreg ] below and that for each @xmath99 , @xmath11 only jumps finitely many times ; see appendix [ smtb ] , lemma [ mtb ] ) . thus , @xmath238 we will use the filtration @xmath233 below , keeping in mind that it is indeed a filtration and satisfies equation ( [ fil ] ) . in the rest of this section we assume that the infinitesimal shift - function @xmath51 satisfies the following regularity condition : [ dreg ] ( _ continuity between the jump times of _ @xmath11 ) . if @xmath11 does not jump in @xmath239 , then @xmath76 is continuous in @xmath240 on @xmath241 and can be continuously extended to @xmath242 $ ] . ( _ boundedness _ ) . for each @xmath99 , there exists a constant @xmath243 such that @xmath244 for all @xmath10 $ ] and all @xmath79 . ( _ lipschitz continuity _ ) . for each @xmath99 , there exist constants @xmath245 and @xmath246 with @xmath247 for all @xmath10 $ ] and all @xmath248 and @xmath249 if @xmath45 and @xmath11 does not jump in @xmath250 $ ] . most regularity conditions on @xmath51 here are satisfied for the @xmath51 s from appendix [ xtap ] [ see also lok ( @xcite ) ] . only the second lipschitz condition is extra . the lipschitz conditions are satisfied , for example , if , in between the jump times of @xmath11 , @xmath51 is continuously differentiable with respect to @xmath79 and @xmath7 with derivatives which are bounded for every fixed @xmath99 . the next theorem states that @xmath38 is indeed also a martingale with respect to @xmath225 : [ mgc ] suppose that the conditions of theorem hold : assumptions ( bounded intensity process ) , ( @xmath40 cadlag ) , ( no unmeasured confounding ) , ( no instantaneous treatment effect ) and for every @xmath10 $ ] , @xmath92 has the same distribution as @xmath21 given @xmath15 . suppose , furthermore , that for each @xmath99 , @xmath11 jumps at most finitely many times , and that @xmath51 satisfies regularity condition [ dreg ] . then the intensity process @xmath27 of @xmath25 with respect to @xmath29 is also the intensity process of @xmath25 with respect to the filtration @xmath225 . recall that in section [ cmim ] we already mentioned that , under regularity conditions , @xmath92 mimics @xmath21 in the sense that it has the same distribution as @xmath21 given @xmath15 . under local rank preservation ( see section [ loc ] ) , @xmath148 for each @xmath7 . in that case theorem [ mgc ] would be the same as the assumption of no unmeasured confounding [ intconf ] . however , as argued in section [ loc ] , local rank preservation is generally considered implausible . proof of theorem [ mgc ] because of assumption [ int ] , @xmath251 is predictable with respect to @xmath29 , so then it is also predictable with respect to the larger filtration @xmath225 . we still need to prove that @xmath38 is a martingale with respect to @xmath225 . since a cadlag martingale with respect to some filtration is also a martingale with respect to its usual augmentation [ see @xcite ( @xcite ) , lemma 67.10 ] , it suffices to prove that @xmath38 is a martingale with respect to @xmath233 . thus we need to prove that , for @xmath252 , @xmath253 = 0.\ ] ] this is not immediate , since we do not have or expect that @xmath254 if @xmath255 . by the definition of conditional expectation , the above is the same as @xmath256 for all @xmath257 . because of theorem 34.1 in @xcite , it is sufficient to consider @xmath258 s forming a @xmath259-system generating @xmath260 . with @xmath261 the @xmath18-algebra on @xmath262 , @xmath263 is such a @xmath259-system : it is closed under the formation of finite intersections and generates @xmath260 . therefore , we only consider @xmath258 s of this form . we prove ( [ enul ] ) for any @xmath264 . let @xmath265 be any approximation of @xmath266 which is continuous for every fixed @xmath267 , with @xmath268 for every @xmath269 as @xmath270 and @xmath271 for all @xmath269 and @xmath267 . then @xmath272}\ , dm(t)\biggr)\\ & & \qquad = e\int 1_{(t_1,t_2]}(t)\ , 1_a(\overline{z}_{t_1 } ) 1_{(x_1,x_2)}(x(t_1))\ , dm(t)\\ & & \qquad = e \int 1_{(t_1,t_2]}(t)\ , 1_a(\overline{z}_{t_1 } ) \lim_{n\rightarrow \infty } 1^{(n)}_{(x_1,x_2)}(x(t_1))\ , dm(t)\\ & & \qquad = e\lim_{n\rightarrow \infty } \int 1_{(t_1,t_2]}(t)\ , 1_a(\overline{z}_{t_1 } ) 1^{(n)}_{(x_1,x_2)}(x(t_1 ) ) \bigl(dn(t)-\lambda(t)\,dt\bigr)\\ & & \qquad=\lim_{n\rightarrow \infty } e \int 1_{(t_1,t_2]}(t)\ , 1_a(\overline{z}_{t_1 } ) 1^{(n)}_{(x_1,x_2)}(x(t_1))\,dm(t).\end{aligned}\ ] ] the last two equalities follow from lebesgue s dominated convergence theorem [ the prior to last equality since , for @xmath273 fixed , the integral is bounded since @xmath25 is finite and @xmath30 is bounded ; the last equality since the integrals are all bounded by @xmath274 , whose expectation is bounded by @xmath275 times the upper bound of @xmath30 ] . equation ( [ enul ] ) and the result of the theorem would follow from theorem [ thec ] if @xmath277}(t)\ , 1_a(\overline{z}_{t_1 } ) 1^{(n)}_{(x_1,x_2)}(x(t_1 ) ) = h^{(n)}_t(x(t),\overline{z}_{t-})\ ] ] for some @xmath278 from @xmath279 satisfying restriction [ hreg ] for each fixed @xmath267 . in principle , this seems possible , since @xmath280 is a function of @xmath92 and @xmath23 for every @xmath281 . indeed , under the conditions above on @xmath51 and @xmath11 , it is possible to find such an @xmath278 , as follows . write @xmath282 for the solution of the differential equation @xmath283 with ( final or initial , depending on @xmath7 ) condition @xmath284 . existence and uniqueness of @xmath282 on @xmath2 $ ] for every fixed @xmath99 follows from theorem [ isun ] in appendix [ difeqap ] . in this notation , @xmath285}(t)\ , 1_a(\overline{z}_{t_1 } ) 1^{(n)}_{(x_1,x_2)}(x(t_1))\nonumber\\ & & \qquad=1_{(t_1,t_2]}(t)\ , 1_a(\overline{z}_{t_1 } ) 1^{(n)}_{(x_1,x_2)}(x(t_1;t , x(t ) ) ) = h^{(n)}_t(x(t),\overline{z}_{t-})\end{aligned}\ ] ] with @xmath286}(t)\ , 1_a(\overline{z}_{t_1 } ) 1^{(n)}_{(x_1,x_2)}(x(t_1;t , y)).\ ] ] we have to show that ( [ ht ] ) satisfies restriction [ hreg ] . first we show that , for @xmath7 fixed , @xmath287 is measurable . from ( [ ht ] ) we see that this is the case if @xmath288 is measurable , which follows immediately from lemma [ mtb ] . restriction [ hreg](a ) is immediate , since @xmath278 is bounded by 1 . for restriction [ hreg](b ) , we have to prove that , for all @xmath99 , @xmath289 when @xmath127 and @xmath128 . fix @xmath99 . we consider three different kinds of @xmath290 . if @xmath291 and @xmath128 , @xmath292 , so that the convergence follows immediately . if @xmath293 and @xmath128 , eventually @xmath294 , so that the convergence also follows immediately . if @xmath295 $ ] , convergence of the first two factors is immediate . for the last factor , we need differential equation theory . @xmath296 is continuous . thus , to prove that the last factor in equation ( [ ht ] ) converges , it suffices to show that @xmath297 as @xmath128 and @xmath127 . fix @xmath99 . for @xmath7 close enough to @xmath290 , we compare the solution of the differential equation with final condition @xmath79 at @xmath7 with the solution of the differential equation with final condition @xmath298 at @xmath290 ; we look at the value of the solution at the time point @xmath299 before both @xmath7 and @xmath290 . first , notice that because of existence and uniqueness of solutions ( theorem [ isun ] ) , the solution of the differential equation with final condition @xmath79 at @xmath7 takes a unique value @xmath300 at @xmath290 . since @xmath269 is differentiable with respect to its first argument with derivative @xmath51 and @xmath51 is bounded by @xmath243 [ assumption [ dreg](b ) ] , @xmath301 is not far from @xmath79 if @xmath7 is not far from @xmath290 : @xmath302 next , notice that , again because of existence and uniqueness of solutions , the value at @xmath299 of the solution of the differential equation with final condition @xmath79 at @xmath7 is the same as the value at @xmath299 of the solution of the differential equation with final condition @xmath300 at @xmath290 . this observation implies that @xmath303 for the first inequality , we use corollary [ difbc1 ] and assumption [ dreg ] [ notice that possible jumps of @xmath51 at the jump times of @xmath11 do not matter here since one can split up the interval , so if , e.g. , there is just one jump at @xmath304 $ ] , one gets a factor @xmath305 etc . ( a formal proof can be given with induction since , with @xmath99 still fixed , there are only finitely many jumps of @xmath11 ) ] . for the last inequality , we use equation ( [ yyt ] ) . if @xmath127 and @xmath128 , the bound in equation ( [ xafs ] ) converges to @xmath24 for every fixed @xmath99 . thus , indeed , if @xmath127 and @xmath128 , @xmath306 converges to @xmath307 . this finishes the proof . the estimating equations for @xmath308 from section [ ees ] were all of the form @xmath309 . in the current section we choose the dimension of @xmath310 the same as the dimension of @xmath308 . estimating equations of this form are well known . theorem [ consas ] below is an example of asymptotic theory in the setting of this article , with conditions in terms of @xmath58 and the intensity process @xmath30 . notice , however , that these conditions are in fact stronger than necessary . for more theory about these types of estimating equations and less restrictive conditions , see @xcite , chapter 5 . in particular , conditions could be weakened by considering the estimating equations as a whole instead of looking at @xmath58 and @xmath30 separately ( see , e.g. , example [ weilem ] ) . we only consider smooth @xmath311 : [ hregpc ] the functions @xmath312 are measurable and : every component of @xmath313 satisfies restriction [ hreg ] . @xmath314 is bounded by a constant @xmath315 not depending on @xmath117 , @xmath108 , @xmath7 , @xmath79 and @xmath99 . for each @xmath10 $ ] and @xmath99 , @xmath316 is continuous . there exists a neighborhood of @xmath135 such that , for each @xmath10 $ ] and @xmath99 , @xmath314 is continuously differentiable with respect to @xmath117 , @xmath108 and @xmath79 , and these derivatives are all bounded by a constant @xmath317 not depending on @xmath117 , @xmath108 , @xmath7 , @xmath79 and @xmath99 . every component of @xmath318 satisfies restriction [ hreg ] . [ consas ] suppose that assumptions ( bounded intensity process ) , [ @xmath178 cadlag ] , ( no unmeasured confounding ) and no instantaneous treatment effect are satisfied . suppose also that , for every @xmath10 $ ] , @xmath92 has the same distribution as @xmath21 given @xmath15 . from theorem we know that , for @xmath58 satisfying restriction , @xmath135 is a zero of @xmath319 suppose now that @xmath135 is the only zero . suppose , furthermore , that we know that @xmath320 with @xmath321 compact , that @xmath322 is continuous for each @xmath7 and bounded by a constant @xmath323 which does not depend on @xmath324 , and that @xmath325 is continuous for each @xmath7 . then any sequence of ( almost ) zeros @xmath326 of @xmath327 that is , any sequence of estimators @xmath326 such that @xmath328 converges in probability to zero , is a consistent estimator for @xmath135 for each @xmath122 satisfying restriction . suppose , moreover , that @xmath329 is differentiable with the respect to @xmath117 with derivative bounded by a constant @xmath330 in a neighborhood of @xmath118 , and @xmath331 is differentiable with respect to @xmath108 with the derivative bounded by a constant @xmath332 in a neighborhood of @xmath119 . then for each @xmath58 satisfying restriction [ hregpc ] there is a neighborhood of @xmath135 such that @xmath333 is continuously differentiable with respect to @xmath308 . suppose , moreover , that the matrix @xmath334 is nonsingular . then there exists a sequence of ( almost ) zeros @xmath326 to @xmath335 . furthermore , any such sequence is asymptotically normal : @xmath336 with @xmath337 the matrix above and , with @xmath338 , @xmath339 proof consistency follows from theorem 5.9 of @xcite . repeatedly applying lebesgue s dominated convergence theorem shows that our conditions imply the conditions of the first and second paragraph after this theorem 5.9 . the existence of ( almost ) zeros follows from @xcite , section 3.9 , problem 9 , whose solution is practically given by the hint below it . this problem 9 states that if @xmath340 is a homeomorphism of a neighborhood of @xmath341 onto a neighborhood of @xmath342 , then every continuous @xmath343 for which @xmath344 is sufficiently small has at least one zero . in our case @xmath345 is continuously differentiable in a neighborhood of @xmath135 by restriction [ hregpc](d ) and the assumptions on @xmath116 and @xmath105 , under which differentiation and integration can be exchanged ( twice ) . the derivative of this @xmath346 at @xmath135 is nonsingular by assumption , and hence , @xmath310 is a homeomorphism of a neighborhood of @xmath341 onto a neighborhood of @xmath347 . @xmath335 is continuous in @xmath308 and close enough to @xmath346 for large @xmath267 with probability approaching @xmath39 because of the second paragraph below theorem 5.9 in @xcite . hence , @xmath335 has a zero with probability approaching @xmath39 . asymptotic normality follows from theorem 5.21 of @xcite , as follows . define @xmath348 , which is continuously differentiable with respect to @xmath308 in a neighborhood @xmath349 of @xmath135 under our conditions . making @xmath349 smaller so that all boundedness conditions hold on @xmath349 , we define @xmath350 , which is bounded by @xmath351 , a constant plus a constant times @xmath352 . this @xmath353 is square integrable since @xmath352 is square integrable : it is well known that counting processes with bounded intensity processes are square integrable [ it follows from , e.g. , proposition ii.4.1 of andersen et al . ( @xcite ) ] . for the same reason , @xmath354 . the remaining conditions of theorem 5.21 from @xcite were checked before , so , indeed , @xmath326 is asymptotically normal with asymptotic covariance matrix ( [ asvar ] ) . the asymptotic variance ( [ asvar ] ) is often estimated by replacing @xmath135 by their estimates and @xmath355 by @xmath144 . thus , confidence intervals for @xmath119 can be constructed . also , tests for whether @xmath119 has a specific value can be constructed that way . for more about testing , see section [ test ] . one can often simplify the expression for the asymptotic variance in equation ( [ asvar ] ) using corollary [ casvar ] below . we use the following lemma : [ caslem ] suppose that the conditions of theorem hold : assumptions bounded intensity process [ @xmath40 cadlag ] , no unmeasured confoundingno instantaneous treatment effect , for every @xmath10 $ ] , @xmath92 has the same distribution as @xmath21 given @xmath15 ( see section ) , for each @xmath99 , @xmath11 jumps at most finitely many times , and @xmath51 satisfies regularity condition . write @xmath356 and @xmath338 . then for @xmath357 every component of which satisfies restriction @xmath358 if , furthermore , @xmath359 is a correctly specified model for @xmath30 such that @xmath360 exists and @xmath141 is a correctly specified model for @xmath51 such that , for each @xmath7 , @xmath109 is differentiable with respect to @xmath108 at @xmath361 , then , for @xmath362 satisfying restriction [ hregpc ] , @xmath363 if the left- or right - hand side exists and @xmath364 if the left- or right - hand side exists , where @xmath365 . proof for the first statement , we use counting process theory from andersen et al . ( @xcite ) , chapter 2 . if @xmath366 is a martingale , @xmath367 ( if it exists ) is defined as a predictable process such that @xmath368 is a ( local ) martingale . if @xmath369 is another martingale , @xmath370 ( if it exists ) is defined as a predictable process such that @xmath371 is a ( local ) martingale . @xmath367 is called the predictable variation process of @xmath366 and @xmath370 is called the predictable covariation process of @xmath366 and @xmath369 . for vector - valued @xmath366 , @xmath367 is defined as a predictable process such that @xmath372 is a ( local ) martingale . hence , it is a matrix with @xmath373 at the @xmath374th row , @xmath160th column . as shown in theorem [ mgc ] , @xmath132 is a martingale with respect to the filtration @xmath225 . counting process martingales like this have compensators : @xmath375 [ see , e.g. , proposition ii.4.1 in andersen et al . ( @xcite ) ] . moreover , if @xmath376 and @xmath377 are ( locally ) bounded @xmath225-predictable processes with values in @xmath378 , then @xmath379 exists and @xmath380 [ proposition ii.4.1 or ( 2.4.9 ) in andersen et al . ( @xcite ) ] . because @xmath381 satisfies restriction [ hreg ] , @xmath382 is a bounded @xmath225-predictable process ( proof just as in lemma [ hpred ] ) . therefore , the theory above leads to @xmath383 for the second statement , notice that , under the conditions of the lemma , @xmath384 and the expectation of the second term here is equal to zero because of theorem [ thec ] . for the third statement , notice that , under the conditions of the lemma , @xmath385 because of the chain rule , and the expectation of the second term is equal to zero because of theorem [ thec ] . this lemma simplifies the asymptotic variance formula of the estimators in equation ( [ asvar ] ) : [ casvar ] suppose that the conditions of theorem hold : assumptions ( bounded intensity process ) , [ @xmath40 cadlag ] , no unmeasured confounding ( no instantaneous treatment effect ) , for every @xmath10 $ ] , @xmath92 has the same distribution as @xmath21 given @xmath15 ( see section ) , for each @xmath99 , @xmath11 jumps at most finitely many times , and @xmath51 satisfies regularity condition . suppose also that @xmath359 is a correctly specified model for @xmath30 such that @xmath360 exists and that @xmath141 is a correctly specified model for @xmath51 such that , for each @xmath7 , @xmath109 is differentiable with respect to @xmath108 at @xmath361 . then if @xmath386 and @xmath362 satisfies restriction [ hregpc ] , the asymptotic variance ( [ asvar ] ) is equal to @xmath387 with @xmath388 and @xmath389 with @xmath390 and , with @xmath365 , @xmath391 we conclude this section with an example to see the machinery work in practice . notice that the boundedness conditions of theorem [ consas ] are somewhat too restrictive for this example , but that the results hold true under these weaker restrictions , too . [ ] [ weilem ] consider the setting of example , and suppose that the assumptions of section [ mgs ] are satisfied . in example @xmath392 where at risk means at risk for initiation of prophylaxis treatment . @xmath109 is the solution to the differential equation @xmath393 with final condition @xmath394 and @xmath395 , so @xmath396 in example [ wei ] we already saw that @xmath397 is a zero of @xmath398 suppose now that @xmath397 is the only zero . suppose , furthermore , that the survival time @xmath1 takes values in a compact space @xmath399\subset\mathbb{r}$ ] and that we know that @xmath400 , with @xmath401 , @xmath402 , @xmath403 and @xmath404 all four compact ( note that this implies that @xmath201 and @xmath202 are bounded away from @xmath24 ) . then any sequence of ( almost ) zeros @xmath405 of @xmath406 that is , any sequence of estimators @xmath405 such that @xmath407 converges in probability to zero , is a consistent estimator for @xmath397 . moreover , @xmath389 as in corollary [ casvar ] exists , and @xmath408 and @xmath409 is a five - dimensional vector with zeros in the first four positions and @xmath410 in the fifth , with @xmath411 if this @xmath337 is a nonsingular matrix , then there exists a sequence of ( almost ) zeros @xmath405 of ( [ weiboh ] ) . furthermore , any such sequence is asymptotically normal : @xmath412 with @xmath337 the matrix above and @xmath413 moreover , @xmath227 and @xmath228 are asymptotically independent . the asymptotic independence here turns out to be no coincidence ; see lok ( @xcite ) . proof of lemma [ weilem ] the findings here are similar to the findings in theorem [ consas ] , but the boundedness conditions fail to hold here . consistency follows from @xcite , theorem 5.9 . existence of a sequence of ( almost ) zeros follows from @xcite , section 3.9 , problem 9 , whose solution is practically given by the hint below it . asymptotic normality follows from @xcite , theorem 5.21 . the asymptotic variance equals @xmath414 because of corollary [ casvar ] ( @xmath30 and @xmath58 are not bounded here , but we can restrict the interval to @xmath215 $ ] for @xmath216 and let @xmath415 ) . we leave checking the conditions of these theorems to the reader [ or see @xcite ( @xcite ) , section 7.7 ] . asymptotic independence of @xmath227 and @xmath228 follows by direct calculation , after noticing that @xmath416 with @xmath417 and @xmath418 define @xmath419 this concludes the proof : @xmath420 we show that one can often test whether treatment affects the outcome of interest without specifying a model @xmath141 for @xmath51 . this was conjectured , but not proved , in @xcite . if one does not have to specify a model for @xmath51 in order to test whether treatment affects the outcome , false conclusions caused by misspecification of the model for @xmath51 can be avoided . under the null hypothesis of the no treatment effect , @xmath89 and @xmath421 ( see section [ model ] ) . if there is no unmeasured confounding , @xmath92 does not predict @xmath26 given @xmath23 ( see theorem [ mgc ] ) . hence , if there is no unmeasured confounding and no treatment effect , adding the _ observed outcome _ @xmath1 to the prediction model for treatment effect should not help the prediction . this idea , presented in @xcite for the case of local rank preservation ( see section [ loc ] ) , can be proven to be correct as follows . technically , the tests in this section are similar to the score test [ for more about the score test see , e.g. , @xcite ( @xcite ) ] . suppose that the conditions of section [ mgs ] are satisfied , and that we have a correctly specified parametric model @xmath359 for @xmath30 . define @xmath422 with @xmath423 satisfying the regularity condition restriction [ hreg ] . the key idea of this procedure is that if treatment does not affect the outcome , @xmath89 , so @xmath421 , and @xmath424 has expectation zero because of theorem [ thec ] . since @xmath118 is unknown , we base the test on the limiting behavior under @xmath89 of @xmath425 , where @xmath227 is an estimator of the nuisance parameter @xmath118 . we will show that if @xmath89 , @xmath425 converges to a normal random variable with expectation zero , which leads to a test for whether @xmath89 in the usual way . the nuisance parameter @xmath118 will be estimated using some set of estimating equations @xmath426 with @xmath427 , @xmath428 differentiable in @xmath117 and @xmath429 . a natural choice would be a maximum ( partial ) likelihood estimator for @xmath118 . we suppose throughout this section that the resulting estimator @xmath227 is consistent and asymptotically normal with @xmath430 as will usually follow from , for example , @xcite , theorem 5.21 . if @xmath116 and @xmath431 are sufficiently smooth , @xmath432 is differentiable with respect to @xmath117 , and a taylor expansion around @xmath118 leads to @xmath433 with @xmath434 the derivative of @xmath435 with respect to @xmath117 and @xmath436 between @xmath118 and @xmath227 . since @xmath227 converges in probability to @xmath118 , so does @xmath436 . therefore , usually , @xmath437 . sufficient conditions under which this holds are given in appendix [ gcap ] , lemma [ pgc ] . because of ( [ thetaas ] ) and the central limit theorem , @xmath438 converges in distribution . therefore , an application of slutzky s lemma leads to @xmath439 with @xmath440 . if @xmath89 , @xmath421 , so that theorem [ thec ] implies that also the expectation of @xmath441 is equal to zero . therefore , the central limit theorem can be applied on the vector with @xmath442 on the right - hand side ; it converges to a normal random variable with expectation zero . because of the continuous mapping theorem [ see , e.g. , @xcite ( @xcite ) , chapter 18 ] , @xmath425 then converges to a normal random variable with expectation zero , too . calculation of its limiting covariance matrix is standard [ see , e.g.,@xcite ( @xcite ) , chapter 18 ] . to save space , we omit that calculation here . if desirable , one can use theorem [ thec ] and lemma [ caslem ] to simplify the expression . notice that a test for whether @xmath443 for any specific @xmath444 can be constructed in exactly the same way . if we have a correctly specified model @xmath141 for @xmath51 , this thus also leads to a confidence region for @xmath119 in the usual way , using the duality between testing and confidence regions : include those @xmath108 for which the null hypothesis @xmath445 is not rejected . the proof of consistency and asymptotic normality of the estimators presented in this article applies to continuous - time structural nested models . a similar proof applies to structural nested models in discrete time ( when covariates are only measured at finitely many fixed times @xmath446 , which are the same for all patients and known in advance ) . @xcite argue without proof that consistency and asymptotic normality should hold for discrete time structural nested models under reasonable assumptions ; the proof is completed with the current article , as follows . it is easy to see that in discrete time , @xmath447 is the compensator of @xmath25 with respect to @xmath29 [ see , e.g. , @xcite ( @xcite ) , section 7.4 ] . the assumption of no unmeasured confounding can be formalized as @xmath448 since @xmath449 given @xmath193 , this implies that also @xmath450 the discrete - time counterparts of theorem [ thec ] and theorem [ mgc ] follow immediately [ see @xcite ( @xcite ) ] . consistency and asymptotic normality follow in the same way as for continuous time models . the tests for treatment effect in this article can be carried out without specifying a model for treatment effect ; that is , no model for @xmath51 is needed . this is an important feature of the tests because it allows one to avoid false conclusions caused by misspecification of the model for @xmath51 . in practice , it may be hard to specify a correct parametric model for the infinitesimal shift - function , @xmath51 . thus , it is good that specification of a model for @xmath51 is not needed to test for treatment effect . the estimators in this article require the correct specification of a model for treatment effect and of a model for prediction of treatment changes . for the discrete - time setting , @xcite has recently proposed estimators which are doubly robust . doubly robust estimators are consistent and asymptotically normal if ( i ) the model for prediction of treatment changes ( @xmath30 in the current article ) is correctly specified or if ( ii ) a regression model of a blipped down outcome [ @xmath109 in the current article ] on past treatment- and covariate history @xmath23 is correctly specified . in any case , the model for treatment effect ( @xmath90 in the current article ) has to be well specified . in this article estimation started with the specification of a model for the infinitesimal shift - function , @xmath51 . interpretation of results may be easier when one starts with a model like ( [ dasim ] ) , @xmath451 , given @xmath452 . here , @xmath65 is the duration of treatment in the interval @xmath66 . the main results in the current article apply also to this model . the proofs for theorems [ thec ] and [ consas ] do not depend on @xmath110 being the solution to @xmath453 . the proof of theorem [ mgc ] does depend on @xmath453 , but it simplifies considerably if ( [ dasim ] ) or ( [ dasim2 ] ) is used as a starting point . let us show this for ( [ dasim ] ) . define @xmath454 . using the first part of the proof of theorem [ mgc ] , for @xmath455 , @xmath285}(t ) 1_{a}(\overline{z}_{t_1})1^{(n)}_{(x_1,x_2)}(\tilde{x}(t_1))\\ & & \qquad=1_{(t_1,t_2]}(t ) 1_{a}(\overline{z}_{t_1})1^{(n)}_{(x_1,x_2)}\bigl(\tilde{x}(t)+\mathit{dur}(t_1,t ) ( -1+e^\psi)\bigr)\\ & & \qquad = h_t^{(n)}(\tilde{x}(t),\overline{z}_{t-}),\end{aligned}\ ] ] where @xmath456 is the duration of treatment in the interval @xmath457 . @xmath458 is measurable if the duration of treatment until @xmath7 is in included in @xmath6 . moreover , if @xmath459 and @xmath127 , then @xmath460 . hence , it follows immediately that @xmath458 satisfies restriction [ hreg ] , which concludes the proof . depending on the specific application , it will be more appropriate to start with @xmath461 given @xmath452 or with a model for @xmath51 ; see , for example , example [ prior ] . in the previous literature [ see , e.g. , robins et al . ( @xcite ) , @xcite , witteman et al . ( @xcite ) , keiding et al . ( @xcite ) and hernn et al . ( @xcite ) ] applications have been carried out under the assumption of local rank preservation , where @xmath148 for each @xmath7 . as pointed out in these papers , and as discussed in section [ loc ] , the assumption of local rank preservation is generally considered implausible . this article relaxes the assumption of local rank preservation in structural nested models . the estimators and tests applied in the previous literature are specific cases of the estimators and tests studied in this article , with the only difference that some of the estimators in the previous literature allow for censoring of @xmath1 . aside from the issue of censoring , this article provides a mathematical foundation behind previous estimators , relaxes the specification of the counterfactual outcomes as deterministic variables , and allows for a distributional interpretation of the estimators . @xcite conjectures that one can often use standard software to test whether treatment affects the outcome of interest ( without specifying a model @xmath90 for @xmath51 ) , and to estimate @xmath108 . lok ( @xcite ) shows that both testing and estimation can also be considered from a partial likelihood point of view . as shown in lok ( @xcite ) , this approach leads to a subclass of the estimators and tests studied in this article which can indeed be calculated with standard software . example [ wei ] is a specific case of that . the possibility to use standard software may be a good reason to choose these estimators in practice . see @xcite and lok ( @xcite ) for a more elaborate discussion . the approach adopted in the current article leads to a large class of estimators and tests . when treatment and covariates change at finitely many fixed times only , robins ( @xcite ) proposes , without proof , an optimal procedure for survival and nonsurvival outcomes , respectively . the optimal choice of estimators or tests under the framework of this article is another intriguing topic for future research . the current article assumes a parametric model @xmath116 for the prediction of treatment changes . in practice , applications have often used a semi - parametric cox model for @xmath116 . @xcite shows that specifying @xmath116 using a semiparametric cox model leads to unbiased estimating equations , which just as in this article are martingales for the true parameters . consistency and asymptotic normality of the resulting estimators for @xmath51 still remain to be shown and constitute interesting topics for future research . in many applications , observations are censored . @xcite and hernn et al . ( @xcite ) have proposed methods to deal with censoring that could potentially be adapted to the results in this paper . for @xmath51 of the form of example [ daexa ] , @xcite includes proofs with censoring due to the fact that the study ends , so - called administrative censoring , using ideas from these previous papers . the estimators in this paper depend on solving a differential equation for each observation . in the examples , these equations are simple enough to be solved analytically . if that is not possible , these equations should be solved numerically . it might be worth investigating how a small contamination of the solution to the differential equation @xmath92 affects the estimates of treatment effect . i conclude with a discussion of the assumptions used in this article . the most important assumption in this article is the assumption of no unmeasured confounding ( assumption [ intconf ] ) . as discussed before , this assumption is valid if all information has been recorded which both ( i ) predicts treatment decisions and ( ii ) is an independent risk factor for the outcome of interest . the validity of the assumption of no unmeasured confounding can not be tested statistically , and depends on the quality of the recorded information . therefore , it is for subject matter experts to decide about the plausibility of the assumption of no unmeasured confounding . second , we only estimate the effect of treatment for which a short duration of treatment has only a small effect on the distribution of the outcome of interest . the effect of the treatment on an individual patient may be large , as long as the probability of such an effect is small for any small duration of treatment . third , the assumption of no instantaneous treatment effect ( assumption [ jump ] ) is also restrictive : it excludes the estimation of the effect of treatments that have instantaneous effects , such as surgery or other point exposures . the remaining assumptions in this article are mostly benign . the assumption that the covariate- and treatment process can be represented by a cadlag process is generally accepted for most medical situations [ see , e.g. , andersen et al . ( @xcite ) ] . the functions @xmath122 and @xmath90 can be chosen such that the regularity conditions on these functions are satisfied . even if @xmath122 is not bounded , it can often be approximated by bounded functions , and results may follow by a simple application of lebesgue s dominated convergence theorem [ see , e.g. , example [ wei ] ] . the same is true for the boundedness condition ( assumption [ int ] ) of the intensity process @xmath30 . assumption [ ycadlag ] that the counterfactual process @xmath21 is cadlag is impossible to verify , but it is a plausible and convenient regularity condition . [ isun ] suppose that a function @xmath76 satisfies the following : ( ) . if @xmath11 does not jump in @xmath239 , then @xmath76 is continuous in @xmath240 on @xmath241 and can be continuously extended to @xmath242 $ ] . ( ) . for each @xmath99 , there exists a constant @xmath462 such that @xmath463 for all @xmath10 $ ] and all @xmath248 . suppose , furthermore , that , for each @xmath99 , there are no more than finitely many jump times of @xmath11 . then , for each @xmath124 $ ] and @xmath125 , there is a unique continuous solution @xmath464 to @xmath465 with boundary condition @xmath466 and this solution is defined on the whole interval @xmath2 $ ] . this theorem follows from well - known results about differential equations ; see , for example , @xcite , chapter 2 . for the next theorem , we also refer to @xcite , chapter 2 . it is a consequence of gronwall s lemma . [ difb ] suppose that @xmath467 is an open or closed interval in @xmath468 , @xmath469 is continuous and @xmath470 is continuous , and suppose that @xmath471 for all @xmath472 and @xmath473 . then , for every @xmath474 and @xmath125 , there is a unique solution @xmath475 of @xmath476 with @xmath477 , and this solution is defined for all @xmath472 . if @xmath478 is continuous and @xmath479 is a solution of @xmath480 , then @xmath481 for all @xmath482 with @xmath483 . in @xcite the interval is always an open interval , but as is generally known , this can be overcome by extending both @xmath484 and @xmath310 outside the closed interval @xmath467 by taking the values at the boundary of @xmath467 . this preserves the lipschitz- and continuity conditions . existence and uniqueness on all of finitely many intervals implies global existence and uniqueness . we have a differential equation with end condition at @xmath3 , so we are interested in @xmath485 with @xmath486 : [ difbc1 ] suppose that the conditions of theorem are satisfied . then , for every @xmath474 and @xmath487 , there is a unique solution @xmath475 of @xmath476 with @xmath477 , and this solution is defined for all @xmath472 . if @xmath478 is continuous and @xmath479 is a solution of @xmath480 , then @xmath488 for all @xmath485 with @xmath486 . proof put @xmath489 . then @xmath490 where @xmath491 . thus , @xmath489 is a solution of the differential equation @xmath492 with boundary condition @xmath493 . define also @xmath494 . applying theorem [ difb ] on @xmath301 concludes the proof , as follows : @xmath495 with @xmath496 notice that , because of equation ( [ lc2 ] ) , @xmath497 with @xmath498 . hence , theorem [ difb ] implies that @xmath499 for the first term , we do a change of variables ; @xmath201 from @xmath24 to @xmath7 , put @xmath500 ; @xmath501 . @xmath502 ; @xmath503 from @xmath504 to @xmath505 . we conclude that the first term is equal to @xmath506 for the second term , similar changes of variables can be done , resulting in corollary [ difbc1 ] . in most of this article we assume that the function which maps @xmath549 to @xmath550 , with @xmath551 , is a measurable function on @xmath123 , with the projection @xmath18-algebra on @xmath552 ( see section [ sets ] ) . moreover , we sometimes assume that the function which maps @xmath553 to @xmath92 , with @xmath551 , is a measurable function on @xmath554 . in this appendix we give sufficient conditions for this . if these two functions are measurable , @xmath233 is a filtration , and , moreover , @xmath233 is the same as @xmath555 [ see equation ( [ fil ] ) in section [ mgs ] ] . [ mtb ] suppose that @xmath51 satisfies regularity assumption and that , for each @xmath99 , @xmath11 jumps at most finitely many times . then the function which maps @xmath549 to @xmath550 , with @xmath551 , is a measurable function from @xmath123 to @xmath378 . also , the function which maps @xmath553 to @xmath92 , with @xmath551 , is a measurable function from @xmath123 to @xmath378 . for the proof of this result , which is quite technical since many results on differential equations are nonconstructive , we refer to @xcite . the proof uses the idea behind euler s forward method to approximate the solution to the differential equation . the following lemma is a worked - out case of theory from @xcite , chapter 19 . [ pgc ] let @xmath556 be i.i.d . random variables with values in a measurable space @xmath557 . let @xmath558 be a collection of measurable functions from @xmath557 to @xmath559 indexed by a subset @xmath560 which contains an open neighborhood @xmath561 of @xmath118 . suppose that @xmath562 is continuous on @xmath561 for every @xmath563 . suppose also that there exists a measurable function @xmath52 on @xmath557 such that @xmath564 for every @xmath565 and such that @xmath566 exists . then if @xmath227 converges in probability to @xmath118 , @xmath567 where @xmath144 indicates the empirical distribution of @xmath568 . proof notice that @xmath569 we show that both terms converge to zero in probability . choose @xmath570 compact and such that it contains an open neighborhood of @xmath118 . example 19.8 from @xcite implies that , under the conditions above , @xmath571 since @xmath572 and @xmath573 contain an open neighborhood of @xmath118 , this implies that the first term converges in probability to zero . for the second term , notice that , on @xmath561 , @xmath574 is continuous in @xmath117 and that each of the components of @xmath575 is bounded by the integrable function @xmath52 , so that lebesgue s dominated convergence theorem implies that @xmath576 is continuous in @xmath117 on @xmath561 . thus , since @xmath572 , also the second term converges in probability to zero . an earlier version of this article was a chapter of my ph.d . dissertation at the free university of amsterdam . i am indebted to richard gill and aad van der vaart for their support , insight and encouragement on this project . i also thank james robins for fruitful discussions . i thank both referees for helpful suggestions . besides that , i thank the netherlands organization for scientific research ( nwo ) for a scholarship . hernn , m. a. , cole , s. r. , margolick , j. , cohen , m. and robins j. m. ( 2005 ) . structural accelerated failure time models for survival analysis in studies with time - varying treatments . _ pharmacoepidemiology and drug safety _ * 14 * 477491 . keiding , n. , filiberti , m. , esbjerg , s. , robins , j. m. and jacobsen , n. ( 1999 ) . the graft versus leukemia effect after bone marrow transplantation : a case study using structural nested failure time models . _ biometrics _ * 55 * 2328 . lok , j. j. , gill , r. d. , van der vaart , a. w. and robins , j. m. ( 2004 ) . estimating the causal effect of a time - varying treatment on time - to - event using structural nested failure time models . _ statist . neerlandica _ * 58 * 271295 . mark , s. d. and robins , j. m. ( 1993 ) . estimating the causal effect of smoking cessation in the presence of confounding factors using a rank preserving structural failure time model . _ statistics in medicine _ * 12 * 16051628 . robins , j. m. ( 1989 ) . the analysis of randomized and non - randomized aids treatment trials using a new approach to causal inference in longitudinal studies . in _ health service research methodology : a focus on aids _ ( l. sechrest , h. freeman and a. bailey , eds . ) 113159 . nchsr , u.s . public health service , washington , dc . robins , j. m. ( 1993 ) . analytic methods for hiv treatment and cofactor effects . in _ methodological issues of aids behavioral research _ ( d. g. ostrow and r. kessler , eds . ) 213287 . plenum press , new york . robins , j. m. ( 2000 ) . robust estimation in sequentially ignorable missing data and causal inference models . in _ proceedings of the american statistical association section on bayesian statistical science _ * 1999 * 610 . robins , j. m. , blevins , d. , ritter , g. and wulfsohn , m. ( 1992 ) . g - estimation of the effect of prophylaxis therapy for pneumocystis carinii pneumonia on the survival of aids patients . _ epidemiology _ * 3 * 319336 . robins , j. m. , rotnitzky , a. and scharfstein , d. o. ( 2000 ) . sensitivity analysis for selection bias and unmeasured confounding in missing data and causal inference models . in _ statistical models in epidemiology the environment and clinical trials _ ( m. e. halloran and d. berry , eds . ) springer , new york . robins , j. m. and wasserman , l. ( 1997 ) . estimation of effects of sequential treatments by reparameterizing directed acyclic graphs . in _ proceedings of the thirteenth conference on uncertainty in artificial intelligence _ ( d. geiger and p. shenoy , eds . ) 409420 . morgan kaufmann , san francisco , ca . witteman , j. c. m. , dagostino , r. b. , stijnen , t. , kannel , w. b. , cobb , j. c. , de ridder , m. a. j. , hofman , a. and robins , j. m. ( 1998 ) . g - estimation of causal effects : isolated systolic hypertension and cardiovascular death in the framingham study . _ american j. epidemiology _ * 148 * 390401 . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: in the long theoretical preparation for the lhc , a wide spectrum of options for the new physics at the tev scale has been considered . a major role has been played by supersymmetric models . several schemes have been investigated in which supersymmetry is broken in a hidden sector with no renormalizable interactions with the observable sector and is communicated to the latter by a variety of mechanisms @xcite . below the scale @xmath4 at which supersymmetry breaking is communicated , sfermion masses are typically described by the model - independent effective lagrangian operator @xmath5 where @xmath6 is a standard model ( sm ) singlet hidden chiral superfield whose @xmath7-term vev breaks supersymmetry , @xmath8 , and @xmath9 is a generic light , observable chiral superfield , for example an mssm one . such an effective , supersymmetric description holds if @xmath10 . different models are characterized by different origins for the above operator . surprisingly enough , a simple and attractive possibility has been neglected in the almost three decades of phenomenological studies of supersymmetry : the possibility that the operator in eq . ( [ eq : sfermioneffective ] ) arises from the renormalizable , tree level exchange of heavy vector superfields , as in fig . [ fig : diagram ] . this is the communication mechanism that we call tree level gauge mediation ( tgm ) @xcite . besides its simplicity ( compare for example with the cumbersome set of two loop diagrams generating the operator in eq . ( [ eq : sfermioneffective ] ) in ordinary , loop gauge mediation ) , this framework is also motivated by the necessary presence of superheavy vector fields in grand unified theories ( guts ) . on the verge of the lhc era , we believe it is worth filling this lacuna and spell out the consequences of tgm , also in the light of its peculiar predictions . one may wonder how such a simple possibility could have been missed . the reason might be that well known arguments seem to prevent it . the main obstacle is represented by the supertrace formula @xcite and its consequences . in the context of the tree level , renormalizable spontaneously broken supersymmetric theory underlying fig . [ fig : diagram ] , we must have @xmath11 for the supertrace of the squared masses of the fields in the model . eq . ( [ eq : str ] ) holds separately for each set of conserved quantum numbers @xcite . if the action of the gauge generators on the full set of chiral superfields , @xmath12 , is traceless , as in the case we are going to consider , the supertrace vanishes . this represents a potential phenomenological problem . still , tgm leads to a viable spectrum , as we will see . while the explicit construction in the next sections is all we need to get to our results , we find useful , in this introduction , to review what the potential problem is and illustrate how tree level gauge mediation solves it . we can see the potential problem at two different levels . first , eq . ( [ eq : str ] ) holds in particular when applied to all fields with the quantum numbers of the sm fermions . let us consider then the case of the mssm . in this case the fields with the quantum numbers of the sm fermions are the sm fermions themselves , @xmath13 , and their supersymmetric partners , the sfermions @xmath14 . from @xmath15 we then conclude that the sum of the squared masses of fermions and sfermions should coincide . this is in clear contradiction with the experimental bounds on the sfermion masses , giving @xmath16 . this is however far from being the end of the story , as any realistic complete theory of supersymmetry breaking is likely to involve additional fields on top of the mssm ones . this is the case of our tgm framework , where the positive contribution to the supertrace from the mssm fermions and sfermions is compensated by an opposite contribution from extra fields with quantum numbers within the ones of the sm fermions , @xmath17 , so that @xmath18 , in agreement with the supertrace formula . the extra chiral superfields will get heavy supersymmetric mass terms . their negative contribution to the supertrace is due to the fact that their scalar components get negative @xmath19 soft masses , which however represent only negligible corrections to their much larger , positive supersymmetric mass term . as we will see , this can be obtained without ad hoc model building efforts . the supertrace formula has stronger implications than the ones outlined above , which should also be addressed . let us consider the fields with the @xmath20 quantum numbers of the @xmath21 quarks only . the latter set will contain at least the three down - type sm quarks and their scalar partners and possibly the extra fields we need to compensate the supertrace formula . let us project the supertrace formula in the flavor space of down - type fields along the direction corresponding to the lightest @xmath21-quark mass eigenstate , the down quark . note that when restricted to a given set of quantum numbers , the trace on the right - hand side of eq . ( [ eq : str ] ) can be non - vanishing . assume now that the only u(1 ) factor in the gauge group is the sm hypercharge , @xmath22 . we then obtain @xcite @xmath23 where @xmath24 is the lightest @xmath21-sfermion mass eigenstate , @xmath25 is the down quark mass , @xmath26 , @xmath27 is the hypercharge gauge coupling and @xmath28 the hypercharge @xmath29-term . ( [ eq : constraint ] ) represents a serious phenomenological problem , even in the presence of the extra fields invoked above . if @xmath30 , in fact , eq . ( [ eq : constraint ] ) would force a down sfermion mass to be smaller than about @xmath31 , in contrast with the lowest experimental limits of a few hundreds gev . if @xmath32 , the constraint would be even stronger . if @xmath33 , the constraint would be loosened , but one could repeat the argument for the up quarks and squarks . for which the @xmath28 contribution to the relation analogous to eq . ( [ eq : constraint ] ) would have opposite sign , leading to an even stronger bound for the lightest up squark . in order to bypass this problem , an extra u(1 ) factor , giving the same sign on both down and up fields , is needed . such an extra u(1 ) factor is present `` by definition '' in the tgm scheme . it is the u(1 ) factor associated to the heavy vector exchange in fig . [ fig : diagram ] ( as @xmath6 is a sm singlet , the heavy vector must also be a sm singlet ) . we therefore have all the ingredients needed to overcome the potential problem set by the supertrace formula . as we will see , those ingredients naturally combine in phenomenologically viable schemes . another potential problem is represented by the fact that gaugino masses arise at the loop level and are therefore potentially suppressed with respect to the sfermion masses by a large loop factor , thus pushing the sfermions out of the reach of the lhc and introducing a significant fine - tuning in the determination of the higgs mass . we will list in section [ sec : gauchi ] a number of gaugino mass enhancement factors that can compensate fully or partially that loop factor . a minimal model of tree - level gauge mediation has been presented in @xcite , solving the supersymmetric flavor problem and predicting the ratio of different sfermion masses to be different from msugra and other schemes . in this paper we would like to take a broader point of view and study the general implementation of tgm . this will allow to establish the general properties of tgm , to set up the guidelines for model building and to identify what are the hypotheses under which the peculiar predictions on soft masses of the minimal model hold . moreover , we would like to present a few new approaches to the @xmath3-problem , both in the context of well known ( giudice - masiero , nmssm ) and new solutions . in particular , in section [ sec : tree ] we will discuss what are the conditions under which heavy vector superfields can act as tree - level messengers of supersymmetry breaking and obtain a general expression for the tree level contribution to the supersymmetry breaking lagrangian , in particular to the sfermion soft masses . in section [ sec : oneloop ] , we will consider the one - loop contributions to soft masses , concentrating mostly on gaugino masses and the enhancement factors compensating their loop suppression . in section [ sec : guidelines ] we consider the possibility to obtain a phenomenologically viable model from the general formalism previously introduced . we will see that clear model building guidelines emerge , leading to peculiar predictions for the pattern of mssm sfermion masses and we will identify the assumptions underlying such predictions . in section [ sec : mu ] , we will discuss a few new approaches to the @xmath3-problem , before summarizing in section [ sec : conclusions ] . the paper also contains two appendixes . in appendix a , we outline the procedure to integrate out vector superfields and address a few minor issues , such as the generalization to the non - abelian case and the role of gauge invariance in a consistent supersymmetric generalization of the expansion in the number of derivatives . in appendix b , we provide an example of a superpotential achieving supersymmetry breaking , so(10 ) breaking to the sm , ensuring that only the mssm fields survive at lower energy ( in particular providing doublet - triplet splitting ) and solving the @xmath3-problem . such a superpotential is not aimed at being simple or realistic , but it represents a useful existence proof . in this section we discuss the conditions under which heavy vector superfields can act as tree - level messengers of supersymmetry breaking in the context of a generic , renormalizable , @xmath34 globally supersymmetric gauge theory in four dimensions . then we recover the general expression for the tree level contribution to the sfermion soft masses . we discuss their origin both in the context of the full , renormalizable theory , and in an effective theory approach . we start from a lagrangian described by a canonical khler @xmath35 and gauge kinetic function and by a generic superpotential @xmath36 function of the chiral superfields @xmath37 , with no fayet - iliopoulos term . we follow the conventions in @xcite . we will denote by @xmath38 , @xmath39 , @xmath40 the scalar , spinor , and auxiliary component of @xmath41 and by @xmath42 , @xmath43 , @xmath44 the vector , spinor , and auxiliary component of @xmath45 . the gauge group @xmath46 ( assumed for simplicity to be simple with a single gauge coupling @xmath47 ) is broken by the scalar component vev @xmath48 to the subgroup @xmath49 at a scale @xmath50 , at which the theory is approximately supersymmetric . in the phenomenological applications we have in mind , @xmath49 contains the sm gauge group @xmath51 , @xmath46 is a grand - unified group ( for example so(10 ) or @xmath52 ) , and the breaking scale is of the order of the gut scale . correspondingly , the vector superfields split into light and heavy ones , associated to the orthonormalized generators @xmath53 and @xmath54 respectively : @xmath55 , @xmath56 , @xmath57 . the heavy vector superfields acquire a squared mass matrix given by @xmath58 we choose the basis of heavy generators @xmath59 in such a way that the above mass matrix is diagonal , @xmath60 the heavy vector superfields become massive by eating up a corresponding number of goldstone chiral superfields . it is then convenient to split the chiral superfields as follows @xmath61 where @xmath62 , @xmath63 are the goldstone superfields associated to the generators @xmath59 and @xmath64 , @xmath65 is an orthonormal basis in the space of the `` physical '' chiral fields @xmath66 , @xmath67 . in the supersymmetric limit , @xmath68 is orthogonal to @xmath69 and @xmath69 does not mix with the physical superfields . the physical components of the massive vector superfield @xmath45 are @xmath42 , @xmath43 , @xmath70 , @xmath71 , all with mass @xmath72 . the imaginary part of @xmath73 , the goldstone boson , becomes as usual the longitudinal component of the massive gauge boson @xmath42 and the spinors @xmath70 and @xmath43 pair up in a dirac mass term . this spectrum can be split by supersymmetry breaking corrections , as we will see in section [ sec : gauginovector ] . as for the physical chiral superfields @xmath74 , their supersymmetric mass matrix is given by @xmath75 again , we choose the basis @xmath76 in such a way that the above mass matrix is diagonal and positive , @xmath77 the scalar and fermion components of @xmath66 can be split by supersymmetry breaking corrections , which can also induce a mixing with the scalar and fermion components of the heavy vector superfields . supersymmetry is supposed to be broken at a much lower scale than @xmath78 , where some of the fields @xmath66 get an @xmath7-term , @xmath79 , @xmath80 . as a consequence , @xmath68 satisfies with good approximation the @xmath7-term and @xmath29-term conditions at the scale @xmath78 , @xmath81 and @xmath82 ( see eq . ( [ eq : dh ] ) ) for each @xmath83 . the @xmath7-terms induce a non vanishing vev for the @xmath29-terms @xmath84 of the heavy vector superfields . the stationary condition for the scalar potential @xmath85 , @xmath86 , together with the gauge invariance of the superpotential give @xmath87 with the light @xmath29-terms still vanishing . clearly , only generators @xmath59 that are singlets under the unbroken group @xmath49 can contribute to such @xmath29-term vevs . note also the condition @xmath88 which implies that the goldstone superfields @xmath69 do not get @xmath7-term vevs ( the @xmath29-term condition implies that in the supersymmetric limit they do not get scalar vev either ) . the latter relation also follows from the gauge invariance of the superpotential . in turn , the @xmath29-terms above give rise to tree level soft masses for the scalar components @xmath89 of the chiral superfields @xmath74 @xmath90 provided that both @xmath91 and the scalars @xmath92 are charged under the ( broken ) gauge interaction associated to @xmath59 and provided that @xmath59 is a singlet under @xmath49 . the complete list of tree level soft terms obtained from the gauge dynamics can be more conveniently recovered in the effective theory below @xmath78 . before discussing it , let us observe that this theory must necessarily satisfy the supertrace formula @xmath93 . in the case of the soft terms in eq . ( [ eq : mmd ] ) , this simply follows here from @xmath94 . in particular , the tracelessness condition implies that positive soft masses are accompanied by negative ones in eq . ( [ eq : dhsfermion ] ) . this is a potential phenomenological problem , which has long been considered as an obstacle to models in which supersymmetry breaking terms are generated , as here , at the tree , renormalizable level . however , it has recently been shown @xcite that such a potential problem can be easily solved by adding a large positive supersymmetric mass term to the chiral superfields whose tachyonic nature would be problematic . as mentioned , the generation of the sfermion masses can be conveniently seen in the effective theory below @xmath78 , where the heavy vector and the goldstone chiral superfields have been integrated out . in this theory , the chiral degrees of freedom are the @xmath66 . the gauge group is @xmath49 and it is unbroken ( we neglect electroweak symmetry breaking ) . as a consequence , there is no @xmath29-term contribution to supersymmetry breaking . the scalar masses arise in this context from @xmath7-terms vevs through an effective khler operator , as we will see in a moment . the vector superfields can be integrated out by solving the equations of motion @xmath95 @xcite . in appendix [ sec : eom ] we illustrate the details of such a procedure in a general case , we explicitly write the resulting effective theory at the leading order , and we make a few general remarks on the approximations involved in using @xmath95 and on the role of gauge invariance in a consistent supersymmetric generalization of the expansion in the number of derivatives @xcite . for the present purposes , we are only interested in the terms in the effective lagrangian relevant to ( sizable ) soft supersymmetry breaking . those are the ones following from the effective tree level contribution to the khler potential in eq . ( [ eq : keff2]a ) : @xmath96 where we remind that @xmath66 has no vev in its scalar component . the operator in eq . ( [ eq : dk ] ) can be seen to arise from the diagram on the left - hand side in fig . [ fig : diagram ] . ( 400,130)(-80,-20 ) ( -60,50)(-10,50)46 ( -90,75)(-60,50 ) ( -90,25)(-60,50 ) ( 20,75)(-10,50 ) ( 20,25)(-10,50 ) ( -95,80 ) [ ] @xmath66 ( 27,80 ) [ ] @xmath97 ( -30,62 ) [ ] @xmath85 ( -98 , 28 ) [ ] @xmath98 ( 29 , 28 ) [ ] @xmath66 ( 53,50 ) [ ] @xmath99 ( 120,50)(180,50)46 ( 90,75)(120,50 ) ( 90,25)(120,50 ) ( 210,75)(180,50 ) ( 210,25)(180,50 ) ( 86,80 ) [ ] @xmath6 ( 83 , 28 ) [ ] @xmath100 ( 150,62 ) [ ] @xmath85 ( 218,80 ) [ ] @xmath101 ( 220 , 28 ) [ ] @xmath102 ( 250 , 50 ) [ ] @xmath103 ( 300,20)(300,80)46 ( 330,105)(300,80 ) ( 270,105)(300,80 ) ( 330,-5)(300,20 ) ( 270,-5)(300,20 ) ( 266,108 ) [ ] @xmath6 ( 264 , -3 ) [ ] @xmath100 ( 313,50 ) [ ] @xmath85 ( 339,108 ) [ ] @xmath104 ( 340 , -3 ) [ ] @xmath9 as mentioned , the only possible source of supersymmetry breaking in the effective theory are the @xmath7-term vevs of the chiral superfields @xmath66 . we remind that such @xmath7-term vevs must belong to non - trivial representations of the full group @xmath46 , in order to play a role in tgm . the only terms in the lagrangian containing such @xmath7-term vevs , at the tree level and up to second order in @xmath91 , @xmath105 , and @xmath106 , arise from the superpotential and from the operator in eq . ( [ eq : dk ] ) : @xmath107 where @xmath108 is the superpotential in the effective theory , @xmath109 let us consider the different terms in eq . ( [ eq : treesoft ] ) in turn . the first term in the second line reproduces the contribution to the soft scalar masses in eq . ( [ eq : mmd ] ) . the second term gives rise to an additional contribution , only relevant to superfields that are gauge partners of the goldstino superfield ( and have the same quantum numbers under @xmath49 as some of the generators of @xmath46)-term contribution to the scalar potential using eq . ( [ eq : mixing ] ) below . ] . all in all , we have @xmath110 .\ ] ] note that the soft terms do not actually depend on the gauge coupling or on the normalization of the generators @xmath111 , as @xmath112 is also proportional to @xmath113 . the second term in the first line of eq . ( [ eq : treesoft ] ) is a gauge - generated yukawa interaction with coupling @xmath114 , usually absent in models of supersymmetry breaking . from a phenomenological point of view , such tiny yukawa couplings might play a role in neutrino physics , where they could represent naturally small dirac neutrino yukawa couplings @xcite . finally , the first term in eq . ( [ eq : treesoft ] ) , has to do with the existence of a hidden sector in the effective theory . in the phenomenological applications we have in mind , the light spectrum will contain the mssm chiral superfields , as part of a light , `` observable '' sector . the latter will be charged under the residual gauge group @xmath115 . on the other hand , the supersymmetry breaking superfields do not feel the residual gauge interactions . in the effective theory , therefore , the supersymmetry breaking sector is hidden from the observable sector from the point of view of gauge interactions . in order for the supersymmetry breaking sector to be hidden also from the point of view of superpotential interactions , it is sufficient to make sure that the first term in eq . ( [ eq : treesoft ] ) does not induce a direct coupling between the two sectors . to be more precise , we can write the chiral superfields of the effective theory , @xmath66 , as @xmath116 the superfield @xmath6 is the only one getting an @xmath7-term vev , @xmath117 . its fermion component is the goldstino and therefore @xmath6 is a massless eigenstate of the mass matrix @xmath118 in eq . ( [ eq : mc0 ] ) . the remaining mass eigenstates are divided in two groups , the heavy ones , @xmath119 with masses @xmath120 , and the light , or observable , ones @xmath121 , with masses @xmath122 . in order to hide supersymmetry breaking from the observable sector also from the point of view of superpotential interactions , we require that @xmath123 ( at least for the renormalizable part of the superpotential ) . we can then see supersymmetry breaking as arising in a hidden sector and then communicated from the to the observable sector by the diagrams on the right - hand side of fig . [ fig : diagram ] . this can perhaps be considered as the simplest way to communicate supersymmetry breaking : through the tree level renormalizable exchange of a heavy gauge messenger . since heavy gauge messengers at a scale not far from the planck scale are automatically provided by grand - unified theories , this possibility is not only simple but also well motivated . the reason why it has not been pursued in the past is an apparent obstacle arising from the supertrace theorem that , as mentioned , can be easily evaded by providing heavy , supersymmetric masses to some of the superfields . such mass terms can naturally arise in the context of grand - unified theories , as we will see . we end this section with some comments on integrating out heavy chiral superfields and on the corresponding possible tree level contributions to @xmath124-terms and soft scalar masses . the heavy vector superfields may not be the only fields living at the scale @xmath78 , as chiral superfields could have mass terms of similar size or get it after gauge symmetry breaking . such chiral fields should also be integrated out in order to get the effective theory below the scale @xmath78 . in general , we want to integrate out all the heavy chiral superfields @xmath125 . since their masses @xmath126 are assumed to be much larger than the supersymmetry breaking scale , it will still be possible to write the effective theory in a manifestly supersymmetric way . in order to integrate them out , let us write the superpotential as @xmath127 where @xmath128 is at least trilinear in its argument . the equations of motion @xmath129 give @xmath130 the effective superpotential for the light fields @xmath6 and @xmath9 is therefore @xmath131 a contribution to the effective khler is also induced @xmath132 the effective contributions to the superpotential and to the khler in eqs . ( [ eq : weff ] ) and ( [ eq : keffchiral ] ) may give rise to `` chiral - mediated '' tree - level @xmath124-terms and ( negative ) additional contributions to soft scalar masses respectively . the latter should be subleading with respect to the ( positive ) vector mediated contributions in eq . ( [ eq : mm ] ) , at least in the case of the mssm sfermions . such tree level contributions could only arise in the presence of trilinear superpotential couplings in the form @xmath133 . in the following we will consider the case in which such a coupling is absent , @xmath134 so that the chiral - mediated tree tree level contributions also vanish . this is often the case , as illustrated by the model in @xcite . in this section we consider the one loop contributions to soft masses , focusing mostly on gaugino masses and the enhancement factors compensating their loop suppression . gaugino masses do not arise at the tree level . they are however generated at the one - loop level , as in standard , `` loop '' gauge mediation models . the suppression of gaugino masses by a loop factor with respect to scalar masses represents a potential phenomenological problem . given the present experimental limits on gaugino masses , a loop factor enhancement would make the sfermions heavier than @xmath135 , beyond the reach of the lhc and heavy enough to introduce a serious fine - tuning problem , thus approaching the split supersymmetry regime @xcite . however , it turns out that the loop hierarchy between gaugino and scalar soft masses is typically reduced or eliminated , as we will see in this section . we calculate gaugino masses in the full theory above @xmath78 . there are two types of one loop diagrams contributing to gaugino masses , depending on whether the degrees of freedom running in the loop are components of the heavy vector superfields ( including the goldstone superfields ) , as in fig . [ fig : mg]a , or physical chiral superfields , as in fig . [ fig : mg]b . correspondingly , we will distinguish a `` vector '' and a `` chiral '' contribution to the light gaugino masses , @xmath136 the latter may easily dominate on the former , as we will see . the source of supersymmetry breaking entering the diagrams of fig . [ fig : mg]a and 2b is a tree level splitting among the components of the heavy vector and chiral superfields respectively . we now examine the two contributions in eq . ( [ eq : gauginos ] ) in turn and write the known results @xcite in a form general enough to be suitable for the following discussion of their quantitative importance compared to the tree level scalar soft terms . ( 200,100)(50,5 ) ( 150,50)(40,180,360)38 ( 150,50)(40,0,45 ) ( 150,50)(40,90,45 ) ( 150,50)(40,180,135 ) ( 150,50)(40,90,135 ) ( 70,50)(110,50 ) ( 230,50)(190,50 ) ( 68,50)[r]@xmath137 ( 232,50)[l]@xmath138 ( 152,25)[t]@xmath139 ( 146,94)(154,86 ) ( 154,94)(146,86 ) ( 110,65)[r]@xmath140 ( 192,65)[l]@xmath140 ( 130,90)[r]@xmath141 ( 174,90)[l]@xmath141 ( 152,98)[b]@xmath142 ( 151,-10)[b]@xmath143 ( 200,100)(50,5 ) ( 150,50)(40,180,360)3 ( 150,50)(40,90,0 ) ( 150,50)(40,90,180 ) ( 70,50)(110,50 ) ( 230,50)(190,50 ) ( 68,50)[r]@xmath137 ( 232,50)[l]@xmath137 ( 150,25)[t]@xmath144 ( 146,94)(154,86 ) ( 154,94)(146,86 ) ( 115,75)[r]@xmath141 ( 190,75)[l]@xmath141 ( 152,98)[b]@xmath142 ( 151,-10)[b]@xmath143 + ( 300,100)(0,-10 ) ( 150,50)(40,180,360)3 ( 150,50)(40,90,0 ) ( 150,50)(40,90,180 ) ( 70,50)(110,50 ) ( 230,50)(190,50 ) ( 68,50)[r]@xmath137 ( 232,50)[l]@xmath137 ( 146,14)(154,6 ) ( 154,14)(146,6 ) ( 115,75)[r]@xmath141 ( 190,75)[l]@xmath141 ( 113,25)[r]@xmath145 ( 188,25)[l]@xmath145 ( 151,28)[t]@xmath146 ( 151,-10)[b]@xmath147 in the supersymmetric limit , the fields @xmath42 , @xmath43 , @xmath70 , @xmath71 form a massive vector multiplet with mass @xmath72 . once supersymmetry is broken , this spectrum is split by corrections to the fermion and scalar masses , which may also mix them with the components of the physical chiral superfields . here , we are interested to the supersymmetry breaking fermion mass term in the form @xmath148 , which is the source of the vector contribution to gaugino masses through the diagrams in fig . [ fig : mg]a . the mass term @xmath149 vanishes in the supersymmetric limit because of the gauge invariance of @xmath150 . the situation is different in the presence of supersymmetry breaking , when the gauge invariance of @xmath150 gives @xmath151 note also the more general expression for the mixed supersymmetry breaking terms @xmath152 before showing the expression for the gaugino masses induced by @xmath153 , let us remind that the heavy vector representation is in general reducible , under the unbroken gauge group @xmath49 , to a set of irreducible components , each with a single value of the mass . let us call @xmath154 the value of the mass in the representation @xmath155 and denote @xmath156 if @xmath59 , @xmath54 belong to the representation @xmath155 . in the limit @xmath157 , the supersymmetry breaking source @xmath153 can be treated as a perturbation in the one loop computation of gaugino masses . at the leading order in @xmath153 , the diagram in fig . [ fig : mg]a generates a contribution to light gaugino masses given by @xmath158 where @xmath159 is the dynkin index of the representation @xmath160 of the generator @xmath111 . the above contribution to gaugino masses arises at the scale @xmath78 where the heavy vectors live . let us now discuss the relevance of the above contribution to gaugino masses . first , let us note that in order for @xmath161 to be non - vanishing we need the following two conditions to be verified at the same time @xmath162 as it can be seen from eqs . ( [ eq : dmvector ] ) and ( [ eq : fgauge ] ) . in particular , we need at least one irreducible ( under the full group @xmath46 ) chiral superfield multiplet to get vev in both its scalar and @xmath7 components . at the same time , we need @xmath163 in order for the tree level contribution to scalar masses to be generated . the conditions in eqs . ( [ eq : con1],[eq : con2 ] ) may force the vector contribution to gaugino masses to vanish . on top of that , @xmath161 is always suppressed by a loop factor @xmath164 compared to the typical scalar mass in eq . ( [ eq : mm ] ) . if @xmath161 was the only contribution to gaugino masses , this would lead to an hierarchy between gaugino and scalar soft masses . moreover , the present experimental lower limits on gaugino masses , @xmath165 , would force the sfermion masses to be heavier than @xmath135 . however , as we will see in a moment , the chiral contribution to gaugino masses can be significantly larger than the vector contribution , thus reducing or even eliminating the loop suppression with respect to soft scalar masses . in this case , the vector contribution to gaugino masses typically ends up to be subdominant . the chiral contribution to gaugino masses arises from the one loop diagram in fig . [ fig : mg]b , as in ordinary loop gauge mediation . the scalar and fermion components of the chiral superfields entering the loop are split by a supersymmetry breaking scalar mass term @xmath166 . as a consequence of eqs . ( [ eq : hide ] ) and ( [ eq : pure ] ) , the supersymmetry breaking couples directly only to the heavy chiral fields and @xmath146 can be treated as a perturbation in the calculation of gaugino masses . the mass term @xmath146 is then given by @xmath167 which adds to the supersymmetric scalar mass term @xmath168 ( in the notation of eq . ( [ eq : mc0diag ] ) ) . the physical chiral superfield representation under the unbroken gauge group @xmath49 is in general reducible to a set of irreducible components , each with its own mass @xmath169 . let us denote @xmath170 at the leading order in @xmath146 , the diagram in fig . [ fig : mg]b generates a contribution to light gaugino masses given by @xmath171 each of the contributions in the sum in the rhs of eq . ( [ eq : gauginochiral ] ) arises at at the scale @xmath169 at which the corresponding chiral superfield lives . if the heavy chiral superfields split in two conjugated representations @xmath172 with a mass term in the form @xmath173 , eq . ( [ eq : gauginochiral ] ) still holds with @xmath174 and a factor 2 multiplying the rhs . let us now discuss the size of the typical chiral contribution to gaugino masses @xmath175 and compare it with the typical size of the tree level scalar soft masses @xmath176 in eq . ( [ eq : mm ] ) . let us consider for simplicity the case in which the scalar masses are due to the exchange of a single heavy vector and the irreducible ( under @xmath49 ) components of the physical chiral superfields have definite charges @xmath177 under the corresponding generators . as for the dynamics giving rise to gaugino masses , let us assume that there are no bare mass terms in the superpotential , i.e. @xmath178 . then both @xmath179 and @xmath180 arise from the same trilinear term in @xmath36 . under the above assumptions , we have @xmath181 where @xmath9 is the charge of the scalar acquiring the mass @xmath182 . while the loop factor @xmath164 suppresses @xmath175 compared to @xmath182 by a @xmath183 factor , the expressions in eqs . ( [ eq : comparison ] ) may give rise to several enhancements of @xmath184 reducing or even eliminating the loop hierarchy : * in the context of grand unified theories the heavy vectors contributing to the soft scalar masses are either one ( as in the case of the minimal possibility so(10 ) , see section [ sec : guidelines ] ) or a few , unless the unified group is very large . on the other hand , the gaugino masses may always get contribution from several chiral messengers . * sfermion and gaugino masses depend on different group factors . sfermions can get a mild suppression if @xmath185 . this is indeed what turns out to happen in simple models , as we will see in section [ sec : guidelines ] . * the heavy vector masses whose exchange generates @xmath182 collect all the vevs breaking the corresponding charge @xmath9 . the scalar mass @xmath182 is therefore suppressed by all such vevs . on the other hand , gaugino masses are only suppressed by the vevs related to supersymmetry breaking by superpotential interactions @xmath186 . unless some of then have @xmath187 , the vevs suppressing gaugino masses will be a subset of the vevs suppressing scalar masses , thus leading to an enhancement of gaugino masses . in the presence of an hierarchy between the vevs related to supersymmetry breaking and some of the other , @xmath9-breaking vevs , this enhancement can be quite large . * different couplings @xmath186 can appear in the numerator and denominator of the expression @xmath188 . this is likely to be the case as a consequence of the relation @xmath189 , which can be satisfied without the need of cancellations only in the case in which the fields charged under @xmath9 do not have vevs in both the @xmath7 and scalar components . if the couplings appearing in the numerator and the denominator are hierarchical , gaugino masses can be sizably enhanced ( or , in this case , further suppressed ) . the study of simple models shows that indeed the enhancement factors above can naturally arise ( see section [ sec:1610 ] below and @xcite ) . in particular , the first two factors reduce the hierarchy @xmath184 by factors 3 and @xmath190 respectively . the third factor gives at least a factor @xmath191 enhancement . the @xmath183 hierarchy arising from the loop factor is thus reduced by a factor 10 , which is enough to bring the sfermions within the reach of the lhc . such a milder hierarchy may even be necessary in the light of the bounds on the higgs mass , which require the stops not to be too light . still , the residual @xmath192 hierarchy can then be easily fully eliminated by the remaining two factors in the above list . besides gaugino masses , which can be seen to arise from one loop corrections to the gauge kinetic function , a number of soft terms can be generated or get a contribution from the one - loop corrections to the khler . the latter can be computed by using the general results in @xcite , which give @xmath193 - 2 \operatorname{tr}\left [ m^2_v \left ( \log \frac{m^2_v}{\lambda^2 } - 1 \right ) \right ] \right ] , \ ] ] where @xmath194 are functions of the chiral superfields , @xmath195 is our canonical khler @xmath196 and the indexes run on the heavy vector and chiral superfields . as in the case of gaugino masses , the soft terms might get a contribution from both heavy vector and chiral superfields running in the loop . as the contribution to one loop soft terms are highly model dependent , we just remind and collect their general expression in terms of @xmath197 . let us expand @xmath197 in terms of powers of @xmath9 and @xmath6 around @xmath68 . the relevant terms are @xmath198 where @xmath199 , @xmath200 , are hermitian , @xmath201 , @xmath202 symmetric and all are dimensionful . we have omitted @xmath203 terms , which are well - known to destabilize the hierarchy @xcite . their absence can be ensured for example by requiring that there are no light chiral fields with the same quantum numbers as @xmath6 . the first term @xmath199 gives rise to the following `` @xmath124-terms '' @xmath204 ( and to a two loop contribution to scalar soft masses ) , where @xmath205 is the scalar component of @xmath9 . the second term @xmath201 generates a contribution to the `` @xmath3-term '' in the superpotential @xmath206 and the fourth term @xmath202 a contribution to the `` @xmath207-term '' @xmath208 a more comprehensive discussion of the @xmath3-term and the @xmath3 problem can be found in section [ sec : mu ] . finally , @xmath200 gives 1-loop contributions to soft scalar masses @xmath209 that add to the tree level contributions in eq . ( [ eq : mm ] ) . additional one - loop contributions to soft scalar masses can come from an induced 1-loop fayet - iliopoulos term @xcite associated for example to the heavy @xmath49-singlet generators , in particular to those involved in the mediation of supersymmetry breaking at the tree level . such terms vanish if the heavy chiral mass matrix and the matrix of their couplings to the spurion @xmath6 are diagonal in the same basis ( in which case the condition in eq . ( [ eq : pure ] ) is also automatically satisfied ) or if the latter matrix of couplings is hermitian in one basis in which the mass matrix is diagonal @xcite . this completes the list of the soft terms arising at one loop . two loop corrections to the scalar soft masses can also arise , of course , as in standard loop gauge mediation , and be sizable in the presence of an enhancement of one - loop gaugino masses @xcite . we now consider the possibility to obtain a phenomenologically viable model from the general formalism discussed so far . we will see that clear model building guidelines emerge from this analysis , leading , in economical schemes , to peculiar predictions for the pattern of mssm sfermion masses . in particular , we will identify the assumptions underlying such predictions . in a phenomenologically viable model , the unbroken gauge group @xmath49 should contain the sm group , @xmath210 , and the light superfield content should contain the mssm spectrum , @xmath211 , in standard notations , where @xmath212 is the family index . we assume that the full gauge group @xmath46 is a simple , grand - unified group , motivated by the well known successful predictions of the sm fermion gauge quantum numbers , of the strong coupling in the mssm , and of the unification scale in the phenomenologically allowed region . the candidates for the unified group @xmath46 in a four - dimensional theory are su(@xmath213 ) , @xmath214 , so(4@xmath215 + 2 ) , @xmath216 , and the exceptional group @xmath52 @xcite . in the following we will focus on the smallest ( or unique ) representatives of each class , su(5 ) , so(10 ) , and @xmath52 . we want the mssm sfermions to get a positive , @xmath217 mass through tree level gauge mediation . the general form of such mass terms is given in eq . ( [ eq : mm ] ) . the latter contains two contributions , corresponding to the two diagrams on the right - hande side in fig . [ fig : diagram ] . in order for the second contribution to play a role for sfermion masses , the corresponding chiral superfields should live in the same unified multiplet as the supersymmetry breaking source @xmath6 . this will not be the case in the models we consider ( as a consequence , for example , of a matter parity telling the supersymmetry breaking multiplet from the matter ones ) . on the other hand , the second contribution might contribute to the higgs masses , if some of the gauge generators have the same quantum numbers ( which is not the case in so(10 ) , the unified group we will consider in greater detail ) . the mssm sfermions then get their tree level soft masses from the first term in eq . ( [ eq : mm ] ) only . in order for @xmath218 to be non - vanishing , the heavy generator @xmath59 must be a sm singlet , since @xmath91 is . we therefore need a group @xmath46 with rank 5 at least . this means that su(5 ) can not give rise to tree level gauge mediation , while so(10 ) and @xmath52 are in principle suitable . let us first consider the `` minimal '' option , so(10 ) , which has also the well known virtue to be able to accommodate a whole mssm family in a single irreducible spinorial representation . we will make a few considerations on the @xmath52 option at the end of this section . in so(10 ) there is exactly one ( up to a sign ) ortho - normalized heavy sm - singlet generator , @xmath219 , where @xmath220 is the su(5 ) invariant so(10 ) generator . the quantum numbers of the so(10 ) representations with dimension @xmath221 under @xmath222 are given in table [ tab : x ] . @xmath223 the values of the @xmath222 quantum numbers are crucial because the soft terms turn out to be proportional to those charges . from eq . ( [ eq : mm ] ) we obtain in fact @xmath224 where @xmath225 is the @xmath222-charge of the sfermion @xmath14 and @xmath78 is the mass if the vector superfield associated to the generator @xmath222 ( note that the gauge coupling and the normalization of the generator @xmath226 cancel in eq . ( [ eq : mmf ] ) ) . in order to predict the pattern of the tree level sfermion masses , we then just need to specify the embedding of the three mssm families into so(10 ) , which we will do through their su(5 ) embedding into three light @xmath227 , @xmath212 . we use two constraints to determine the embedding of the @xmath228 into so(10 ) representations . the first one is related to quite a nice feature eq . ( [ eq : mmf ] ) : the soft terms turn out to be family - universal , thus neatly solving the supersymmetric flavor problem . provided , of course , that the three families of each of the mssm matter multiplets are embedded in the same type of so(10 ) representation , which we will assume in order to ensure that family - universality indeed holds . on top of that , we want the mssm sfermion soft masses in eq . ( [ eq : mmf ] ) to be positive in order to avoid spontaneous symmetry breaking of color , electric charge , or lepton number at the scale @xmath182 . clearly , the standard embedding of a whole family into a 16 of so(10 ) would not work , as it would lead to negative masses for the sfermions in either the @xmath229 or the 10 of su(5 ) . this is in turn related to the tracelessness of the so(10 ) generators , and in particular of @xmath222 . as a consequence , whatever is the so(10 ) representation in which we choose to embed a given mssm matter multiplet with positive soft mass , that representation will necessarily contain extra fields with negative soft masses . this apparent obstacle can be easily overcome by splitting the so(10 ) representation containing the mssm multiplet through so(10 ) breaking , in such a way that the extra fields with negative soft masses acquire a large supersymmetric mass term . the negative soft mass will then represent a negligible supersymmetry breaking correction to that large ( positive ) mass . it turns out that such a splitting is actually expected to arise , as will see , a fact that reinforces the logical consistency of tree level gauge mediation . we are now ready to discuss the embeddings of the three @xmath230 and @xmath231 of su(5 ) containing the light mssm families in so(10 ) . as @xmath232 is positive , the possible choices depend on the sign of @xmath233 . we limit ourselves to the so(10 ) representations with @xmath221 , as in table [ tab : x ] . there are then only two possibilities : * @xmath234 . in this case we need to embed the @xmath230 s and @xmath231 s into so(10 ) representations containing @xmath235 and @xmath236 of su(5 ) with positive charges under @xmath222 . from table [ tab : x ] we see that the only possibility is to use three @xmath237 and three @xmath238 , @xmath212 , where we have explicitly indicated the su(5 ) decomposition , and to embed the @xmath231 s into the @xmath239 s , @xmath240 , and the @xmath230 s into the @xmath241 s , @xmath242 . the spare components @xmath243 , @xmath244 get negative soft masses and need to acquire a large supersymmetric mass term . * @xmath245 . in this case we need the @xmath230 s and @xmath231 s to have negative charges under @xmath222 . the only possibility is then to use three @xmath239 s as before and three @xmath246 , @xmath212 , with @xmath247 and @xmath248 . the spare components @xmath249 , @xmath250 , get negative or vanishing soft masses and need to acquire a large supersymmetric mass term . in both cases the chiral content of the theory is still given by three 16 of so(10 ) . we have implicitly neglected the possibility of mixed embeddings in which , for example , the @xmath251 s of su(5 ) are a superposition of the @xmath251 s in the @xmath241 s and @xmath239 s of so(10 ) . while this possibility is in principle not excluded , it would in general introduce a dependence of the sfermion soft masses on mixing parameters that are in general flavor violating , thus possibly spoiling the flavor universality result . the two possibilities above give rise to two definite predictions for the patter of sfermion soft masses at the scale @xmath78 : @xmath252 m^2_{\overline{5 } } = 2 m^2_{10 } \quad\text{if } \quad f_0^\dagger x f^{\phantom{\dagger}}_0>0 \\ m^2_{\overline{5 } } = \frac{3}{4 } m^2_{10 } \quad\text{if } \quad f_0^\dagger x f^{\phantom{\dagger}}_0<0 . \notag\end{gathered}\ ] ] to summarize , the latter predictions are based on the following hypotheses : `` minimal '' unified gauge group so(10 ) , embedding of the mssm families in the so(10 ) representations with dimension @xmath221 not containing the goldstino , and absence of mixed embeddings to automatically preserve flavor - universality . the predictions on the ratios @xmath253 in eq . ( [ eq : prediction ] ) are peculiar enough to make a possible experimental test at the lhc a strong hint for tree level gauge mediation . as for the source of supersymmetry breaking , @xmath117 , we need @xmath6 to have a non - vanishing charge under @xmath222 . if we limit ourselves again to representations with @xmath221 , the only possibility is that @xmath6 has a component in the `` right - handed neutrino '' direction of a 16 or a @xmath254 . with the sign conventions we adopted , a component in a 16 gives a positive contribution to @xmath233 , while a component in a @xmath254 gives a negative contribution . we now want to show that the two embeddings of the light mssm families described above can be obtained in a natural way . we have to show that it is possible to split the so(10 ) representations in which the mssm fields are embedded in such a way that the extra fields ( with negative soft masses ) get a heavy supersymmetric mass term from so(10 ) breaking . it will turn out that the so(10 ) breaking vevs of a @xmath255 , essential to break so(10 ) to the sm ( unless representations with @xmath256 are used to reduce the rank ) just provide the needed splitting . the fact that such vevs make heavy precisely the components of the so(10 ) representations that get a negative soft supersymmetry breaking mass reinforces the logical consistency of this framework . in the following , we first discuss the @xmath257 embedding in a general , top - bottom perspective , obtaining a generalization of the model in @xcite , and discuss the conditions for a pure ( non mixed ) embedding . we then discuss the possibility of a @xmath258 embedding . let us consider the embedding associated to the case @xmath259 . we assume the existence of a matter parity symmetry that tells matter superfields from higgs superfields . let 16 , @xmath254 be the so(10 ) multiplets breaking so(10 ) to su(5 ) ( we can always choose the basis in the space of the 16 ( @xmath254 ) representations in which a single 16 ( @xmath254 ) gets a vev in its scalar component ) . the most general renormalizable superpotential involving 16 , @xmath254 , @xmath239 , @xmath241 , @xmath212 , and invariant under a matter parity under which the so(10 ) higgs fields 16 , @xmath254 are even and the the matter fields are odd is @xmath260 where @xmath261 takes care of providing a vev to the 16 , @xmath254 in the sm - singlet direction and does not depend on the matter fields ( but can involve additional even fields , where @xmath222 is an so(10 ) singlet . ] ) . the term @xmath262 is just what needed to split the su(5 ) components of the @xmath237 and of the @xmath238 and make heavy the unwanted components @xmath243 and @xmath263 . once 16 acquires a vev @xmath85 in its singlet neutrino component , in fact , a mass term is generated for those components , @xmath264 the singlet neutrinos @xmath265 remain light at the renormalizable level but can get a mass at the non - renormalizable level through the operator @xmath266 . it is remarkable that the components acquiring a large mass are precisely those that get a negative soft mass term . on the other hand , this is only true in the limit in which the @xmath267 mass term in eq . ( [ eq : w1 ] ) can be neglected . in the presence of a non negligible @xmath267 , in fact , the full mass term would be @xmath268 which would give rise to a mixed embedding of the light @xmath230 s in the @xmath239 s and @xmath241 s . in order to abide to our assumptions , which exclude the possibility of mixed embeddings , such a @xmath267 term should be absent . this can be easily forced by means of an appropriate symmetry . let us however relax for a moment that assumption in order to quantify the deviation from universality associated to a small , but non - negligible @xmath267 . the mssm sfermions in the @xmath235 of su(5 ) receive in this case two contributions to their soft mass , a positive one associated to the components in the @xmath241 s , proportional to @xmath269 , and a negative one associated to the components in the @xmath239 s , proportional to @xmath270 . the soft mass matrix for the light sfermions in the @xmath235 of su(5 ) can be easily calculated in the limit in which the @xmath267 mass term can be treated as a perturbation . in this limit , the light mssm fields in the @xmath235 of su(5 ) are in fact @xmath271 and their soft scalar mass matrix at the scale @xmath78 is @xmath272 where @xmath176 is defined below . the mixed embedding induced by the mass term @xmath267 leads to flavor - violating soft - terms . setting @xmath273 allows to preserve the flavor blindness of the soft terms and to satisfy the fcnc constraints without the need of assumptions on the structure of the flavor matrices @xmath274 and @xmath267 . we therefore assume that @xmath267 is vanishing or negligible . we then have @xmath275 , @xmath276 , with the extra components @xmath243 and @xmath244 obtaining a large supersymmetric mass term @xmath277 , as desired . the soft masses for the light sfermions are [ eq : prediction1 ] @xmath278 as anticipated in eq . ( [ eq : prediction ] ) . the reason for the factor @xmath279 will become clear in a moment . we now need to identify the embedding of the mssm higgs superfields and obtain the mssm superpotential for them , in particular the mssm yukawa interactions . it is useful to discuss the yukawa interaction in su(5 ) language . the up quark yukawa interactions arise from the su(5 ) operator @xmath280 where @xmath281 contains the mssm up higgs . as @xmath276 , the operator in eq . ( [ eq : l1 ] ) can arise at the renormalizable level from a so(10 ) invariant operator only if @xmath281 has a component into a @xmath282 of so(10 ) , @xmath283 , with @xmath284 where @xmath285 measures the size of the @xmath281 component from 10 representations of so(10 ) ( a basis in the space of the 10 representations can always be chosen such that @xmath281 is contained in a single one , the @xmath282 ) . the operator in eq . ( [ eq : l1 ] ) will then originate as @xmath286 the down quark and charged lepton yukawa interactions arise at the renormalizable level from the su(5 ) operator @xmath287 where @xmath288 contains the mssm down higgs . as @xmath276 and @xmath275 , the operator in eq . ( [ eq : l2 ] ) can arise at the renormalizable level from a so(10 ) invariant operator only if @xmath288 has a component into a @xmath289 of so(10 ) , @xmath290 , with @xmath291 where @xmath292 measures the size of the @xmath288 component from 16 representations of so(10 ) . the operator in eq . ( [ eq : l2 ] ) will then originate as @xmath293 it is tempting ( and economical ) to identify the @xmath289 with 16 , the field whose vev breaks so(10 ) to su(5 ) , in which case @xmath294 and the mass of the heavy extra components @xmath243 and @xmath244 in eq . ( [ eq : unwanted1 ] ) turns out to be proportional to the corresponding light fermion masses ( up to non - renormalizable corrections needed to fix the light fermion mass ratios ) . having introduced the mssm higgs fields , let us now discuss their soft mass terms . to summarize the previous discussion , with our @xmath221 representation content , the up ( down ) higgs superfield @xmath295 ( @xmath296 ) can be embedded in either 10 s or 16 s ( @xmath254 s ) of so(10 ) , in both cases through the embedding into a @xmath281 ( @xmath288 ) of su(5 ) . we have denoted by @xmath285 ( @xmath297 ) the overall size of the @xmath295 ( @xmath296 ) component in the 10 s . the overall size of the component in the 16 s ( @xmath254 s ) is then measured by @xmath298 ( @xmath292 ) . correspondingly , the higgs soft masses get two contributions from the first term in eq . ( [ eq : mmf ] ) proportional to two different @xmath222 charges : [ eq : mmh ] @xmath299 let us now consider gaugino masses . a general discussion of all possible contributions to gaugino masses in the embedding we are considering and in the presence of an arbitrary number of so(10 ) representation with @xmath221 would be too involved . we then consider a few examples meant to generalize the case considered in @xcite and to illustrate the general properties discussed in section [ sec : oneloop ] . let us begin by illustrating in more detail the structure of supersymmetry breaking . with the representation content of table [ tab : x ] , supersymmetry breaking can be associated to the @xmath7-term vevs of superfields in 16 , @xmath254 , 45 , 54 representations ( the ones containing sm singlets ) . however , only the 16 , @xmath254 , whose singlets have non - vanishing @xmath222-charges , can contribute to tree level soft masses . let us call @xmath300 , @xmath301 the matter parity even superfields in the 16 and @xmath254 representations of so(10 ) . in a generic basis , we can parametrize the vevs of their singlet components as @xmath302 the @xmath29-term condition for the @xmath222 generator requires @xmath303 while gauge invariance gives @xmath304 sfermion masses are proportional to @xmath305 ( due to the factor 5 in the definition of @xmath176 ) , where @xmath306 by definition in the case we are considering . note that @xmath176 is suppressed by _ all _ vevs contributing to @xmath222 breaking . let us now comment on the vector contribution to gaugino masses . let us assume to begin with that the @xmath254 s do not break supersymmetry . without loss of generality we can then assume that supersymmetry breaking is only associated to @xmath307 . the gauge invariance condition then gives @xmath308 , i.e. a vev for both the @xmath7-term and scalar components is not allowed . since the @xmath7-term and scalar components belong to different irreducible representations , no vector contribution to gaugino masses is generated by the 16 s . a vector contribution can still be generated by the @xmath7-term vev of a 45 , for example , for which the gauge invariance condition does now prevent a vev in both the scalar and @xmath7-term component . or , it can be generated by the @xmath7-terms of the 16 s if some of the @xmath254 also breaks supersymmetry and cancels the contribution of the 16 to eq . ( [ eq : gaugeinv ] ) . let us next consider the chiral contribution to gaugino masses . the massive components @xmath243 and @xmath263 of the matter superfields will act as chiral messengers if they are coupled to supersymmetry breaking . let us then consider as before the case in which the @xmath254 s do not break supersymmetry , supersymmetry breaking is provided by the @xmath7-term vev @xmath7 of the singlet component of the @xmath309 and is felt by the chiral messengers through the @xmath310 interaction . let @xmath311 be the field whose vev gives mass to the @xmath243 , @xmath263 through the @xmath262 interaction , as in eq . ( [ eq : w1 ] ) . and let us assume that additional @xmath300 s and @xmath301 s get vevs in their scalar components . the chiral messengers @xmath243 , @xmath263 have therefore a supersymmetric mass @xmath312 and their scalar components get a supersymmetry breaking term mass term @xmath313 . the induced one loop chiral contribution to gaugino masses is then @xmath314 the tree level soft mass of the stop ( belonging to the 10 of su(5 ) ) is @xmath315 we can then compare stop and gaugino masses ( before radiative corrections ) . their ratio is particularly interesting , as the gaugino mass @xmath175 is at present bounded to be heavier than about @xmath316 , while @xmath317 enters the radiative corrections to the higgs mass . therefore , the ratio @xmath318 should not be too large in order not to increase the fine - tuning and not to push the stops and the other sfermions out of the lhc reach . from the previous equations we find @xmath319 eq . ( [ eq : ratio ] ) illustrates all the enhancement factors discussed in section [ sec : oneloop ] that can compensate the loop suppression of gaugino masses . the factor 3 corresponds to the number of chiral messenger families ( @xmath320 for @xmath321 ) contributing to gaugino masses , to be compared to the single vector messenger generating sfermion masses at the tree level . the factor @xmath190 comes from the ratio of charges @xmath322 suppressing the stop mass in eq . ( [ eq : comparison ] ) . the factor @xmath323 is the ratio of the vev suppressing gaugino masses ( the one related to supersymmetry breaking though superpotential interactions , @xmath324 ) , and the combination of vevs suppressing sfermion masses ( all of them ) . note that in the presence of hierarchies of vevs , the factor @xmath325 can be large . finally @xmath326 represents a combination of couplings that can further enhance ( or suppress , in this case ) gaugino masses . all in all , we see that the loop factor separating @xmath317 and @xmath175 is partially compensated by a combination of numerical factors : @xmath327 ( leading to @xmath328 for @xmath329 ) becomes at least @xmath330 ( leading to @xmath331 for @xmath329 ) . a largish value of the factors @xmath325 or @xmath326 can then further reduce the hierarchy and even make @xmath332 , if needed . let us now consider the second type of embedding identified above , corresponding to @xmath333 . the most general renormalizable superpotential involving 16 , @xmath254 and @xmath239 , @xmath334 , @xmath212 and invariant under matter parity is @xmath335 the term @xmath336 is just what needed to split the su(5 ) components of the @xmath237 and of the @xmath246 and make heavy the unwanted components @xmath249 and @xmath337 . once 16 acquires a vev @xmath85 , in fact , a mass term is generated for those components , @xmath338 it is remarkable that also in this case the components acquiring a large mass are precisely those that get a negative soft mass term . on the other hand , this is only true in the limit in which the @xmath267 mass term in eq . ( [ eq : w2 ] ) can be neglected . in order to abide to our pure embedding assumption , we will neglect such a term . let us note , however , that such a term should arise at some level in order to make the @xmath339 s components heavy . note that the @xmath340 s do not affect gauge coupling unification at one loop and can therefore be considerably lighter than the gut scale , consistently with the required smallness of @xmath267 . the soft masses for the light sfermions are now @xmath341 unfortunately , the embedding we are discussing can not be implemented with renormalizable interactions and @xmath342 representations only . the problem is obtaining the yukawa interactions . let us consider the up quark yukawas , arising as we saw from the su(5 ) operator in eq . ( [ eq : l1 ] ) . given its size , we expect at least the top yukawa coupling to arise at the renormalizable level . as in the present case @xmath343 , the operator in eq . ( [ eq : l1 ] ) can arise at the renormalizable level from a so(10 ) invariant operator only if @xmath281 has a component in a so(10 ) representation coupling to @xmath344 . and the lowest dimensional possibility containing the 5 of su(5 ) is the 210 . for this reason , we do not pursue this possibility further here , although models with large representations are not a priori excluded . we close this section with a few considerations about the possibility to identify the unified group with @xmath52 . such a possibility looks particularly appealing in the light of what above . we have seen in fact that the most straightforward possibility to realize tree level gauge mediation in so(10 ) requires the matter superfield content to include three @xmath345 , @xmath212 . this is precisely what @xmath52 predicts . the fundamental of @xmath52 , in fact , a representation of dimension 27 , decomposes as @xmath346 the matter content needed by the @xmath345 embedding can therefore be provided in the context of @xmath52 by three matter @xmath347 , @xmath212 , and the @xmath289 and @xmath282 needed to accommodate the higgs fields can also be provided by a single higgs @xmath348 . all yukawas can then in principle follow from the single @xmath52 interaction @xmath349 we postpone the analysis of this promising possibility to further study . in this section , we discuss a few approaches to the @xmath3-problem in the context of tree level gauge mediation . let us remind what the @xmath3-problem is . any supersymmetric extension of the sm must contain two higgs doublet chiral superfields @xmath350 , @xmath351 , with hypercharges @xmath352 , within the light spectrum @xmath9 . moreover , the lagrangian must contain a mass term for their higgsino ( fermion ) components , @xmath353 . it should also contain a corresponding term for the scalar components , @xmath354 , where @xmath207 is a dimension two parameter . the higgsino mass @xmath3 is constrained to be in the window @xmath355 by the present bounds on chargino masses and by naturalness considerations . this coincides with the window for the supersymmetry breaking scale in the observable sector , @xmath356 . it is then tempting try to establish a connection between these two a priori independent scales , in such a way that @xmath357 when @xmath358 , thus making the coincidence of the two scales not accidental . this is the @xmath3-problem . in the absence of such a connection , there would be no reason why @xmath3 should not be of the order of a much larger , supersymmetry conserving scale such as the gut or the planck scale . or , if a symmetry or some other independent principle suppressed @xmath3 , there would be no reason why @xmath3 should not be much smaller . fermion mass terms such as @xmath353 belong to the list of possible soft supersymmetry breaking mass terms @xcite . the reason why they are usually omitted from the mssm effective soft supersymmetry breaking lagrangian is that they can be always reabsorbed in the superpotential ( through appropriate additions to the scalar soft lagrangian ) . moreover , most models of supersymmetry breaking , including the ones we are considering , do not generate such supersymmetry breaking fermion mass terms . we can then assume that the higgsino mass term arises from a corresponding term in the superpotential . the problem is then to relate the coefficient of that ( supersymmetric ) superpotential term , @xmath359 , to the supersymmetry breaking scale in the observable sector , which in our case is given by @xmath360 . we discuss in the following three possible connections . one is peculiar of tree level gauge mediation , the other two have been considered in other contexts , but have specific implementations in tree level gauge mediation . we classify them according to the dimension @xmath29 of the so(10 ) operator from which the @xmath3 term arises . note that we are not addressing the origin of the smallness of @xmath182 and @xmath3 compared to the plank scale , just their connection . the three options we consider are : @xmath361 : : : @xmath3 comes from the operator @xmath362 . it is the supersymmetry breaking scale to be derived from @xmath3 , and not viceversa : @xmath363 , where @xmath364 , and @xmath365 . @xmath366 : : : @xmath3 comes from the operator @xmath367 . the light sm singlet @xmath368 gets a vev from a potential whose only scale is @xmath182 , so that @xmath369 . @xmath370 : : : @xmath3 comes from the operator @xmath371 , so that @xmath372 . let us discuss each of those possibilities in turn . such a possibility was anticipated in @xcite , where however no concrete implementation was given . let us consider the @xmath257 embedding . as discussed in section [ sec:1610 ] , @xmath350 is a superposition of the `` up higgs - type '' components in the @xmath254 s and @xmath236 s ( with @xmath374 ) in the model . analogously , @xmath351 will be a superposition of the `` down higgs - type '' components in the ( @xmath375 ) @xmath376 s and @xmath236 s . the only possible @xmath373 origin of the @xmath3-term in the context of the full so(10 ) theory are then @xmath19 mass terms for the above @xmath254 s , @xmath377 s , and @xmath236 s . as said , we do not address the origin of such a small parameter in the superpotential , as we do not address here the smallness of the supersymmetry breaking scale . the latter can for example be explained by a dynamical mechanism . we want however to relate such mass parameters , in particular the coefficient of a @xmath378 mass term , to the supersymmetry breaking scale . this is actually pretty easy , as the tree level gauge mediation embedding we are considering provides all the necessary ingredients and the result arises from their simple combination . we have seen in fact that the model needs a @xmath377 , @xmath254 pair to get a vev in the sm singlet direction of the scalar component , in order to break so(10 ) to the sm . moreover , we have seen that an independent @xmath309 , @xmath379 pair is required to break supersymmetry trough the @xmath7-term vev of the sm singlet component in the @xmath309 . the simplest way to achieve such a pattern is through a superpotential like @xmath380 where @xmath222 , @xmath6 are so(10 ) singlets and @xmath381 . this is a generalization of an example in @xcite . finally , we have just reminded that the light higgses may have a component in @xmath377 , @xmath309 , @xmath254 , @xmath379 . let @xmath382 be the coefficient of the @xmath296 component in the @xmath309 and @xmath383 the coefficient of the @xmath295 component in the @xmath254 . then a @xmath3 parameter is generated in the form @xmath384 from the @xmath385 term in eq . ( [ eq : wvev ] ) . the parameter @xmath386 is therefore required to be in the window @xmath387 . in the limit @xmath388 , supersymmetry is unbroken and @xmath377 , @xmath254 acquire a vev that can be rotated in the sm singlet component @xmath389 . a non - vanishing @xmath3 , on the other hand , triggers supersymmetry breaking and induces an @xmath7-term vev for the singlet component of the @xmath309 , @xmath390 , with @xmath391 . we therefore have @xmath392 providing the desired connection between @xmath3 and the supersymmetry breaking scale . tree level gauge mediation plays a crucial role not only in providing the ingredients ( and no need to stir ) but also because it is the very so(10 ) structure providing the heavy vector messengers to relate in a single irreducible representation ( the 16 ) supersymmetry breaking ( the @xmath7-term vev of its sm singlet component ) and the down higgs entering the @xmath3-term ( the lepton doublet - type component of the 16 ) . in the appendix [ sec : w ] we provide an existence proof of a ( perturbative ) superpotential that i ) implements the mechanism above , thus breaking supersymmetry and so(10 ) to su(5 ) , ii ) further breaks su(5 ) to the sm , iii ) makes all the fields that are not part of the mssm spectrum heavy , in particular achieves doublet - triplet splitting . this is an implementation of the nmssm solution of the @xmath3-problem ( see e.g. @xcite and references therein ) . as we will see , the implementation of such a solution in the context of tree level gauge mediation avoids some of the problems met in ordinary gauge mediation . in order to implement the nmssm solution of the @xmath3-problem , an explicit term @xmath359 should be forbidden , for example by a symmetry ; the light fields @xmath9 should include a sm singlet @xmath368 , coupling to the higgses through the superpotential interaction @xmath393 ; and @xmath368 should develop a non - zero vev . a @xmath3 parameter will then be generated , @xmath394 . in the absence of terms linear or quadratic in @xmath395 in the superpotential , the scale of a vev for @xmath368 can only be provided by the supersymmetry breaking terms in the soft lagrangian , @xmath396 , in which case @xmath397 , as desired . in order to generate a non - zero vev for @xmath368 , one would like to have a negative soft mass for @xmath368 at the weak scale , along with a stabilization mechanism for large values of the fields . in ordinary gauge mediation this is not easy to achieve . while the stabilization can be simply provided by a @xmath398 term in @xmath150 , as in the nmssm ( or by a quartic term in @xmath399 extensions of the mssm @xcite ) , the soft mass term of @xmath368 vanishes at the messenger scale because @xmath368 is typically a complete gauge singlet . a non - vanishing negative mass term is generated by the rge running but it is typically too small . another problem is that the higgs spectrum can turn out to be non - viable @xcite . a sizable soft mass can still be generated by coupling @xmath368 to additional heavy fields . such possibilities can be implemented in our setup by promoting @xmath368 to an so(10 ) singlet and coupling it to the higgses through a @xmath400 or a @xmath401 coupling to the so(10 ) representations containing ( a component of ) the higgs fields . tree level gauge mediation offers a different avenue . a sizable , negative soft mass term for @xmath368 can in fact be generated by embedding @xmath368 in a @xmath254 of so(10 ) ( this is the only choice within the fields in table [ tab : x ] ) . on the other hand , the stabilization of the potential for @xmath368 is not straightforward . a sizable @xmath398 term is not expected to arise , as it should involve a so(10 ) operator with three @xmath254 . however , the @xmath398 term can be replaced by a term involving a second light singlet @xmath213 , @xmath402 the latter can come from a @xmath403 coupling , if @xmath213 is in the 126 singlet , or from a @xmath404 coupling , where @xmath213 is the @xmath405 singlet and @xmath406 gets a vev . the scalar potential for @xmath407 can be written as @xmath408 where @xmath409 is the mssm scalar potential with @xmath410 , @xmath411 , and @xmath412 or @xmath176 depending on whether @xmath213 comes from a 126 or a 16 . we have neglected the @xmath124-terms , which play a role in explicitly breaking @xmath413-symmetries that could lead to massless states . the potential above has a minimum with a sizable @xmath414 , and a @xmath3 parameter whose size is controlled by @xmath326 . finally , let us consider the possibility to generate the @xmath3 parameter through a @xmath370 correction to the khler in the form @xmath415 , as in the giudice masiero mechanism @xcite . the @xmath7-term vev @xmath416 of @xmath6 would give in this case @xmath417 . we show first that the operator above can not arise at the tree level from integrating out heavy vector or chiral superfields . the corrections to the khler obtained by integrating out heavy vector superfields are given in eqs . ( [ eq : keff2 ] ) . all terms are at least of second order in @xmath418 and no trilinear term is present . moreover , no sizable trilinear term can be obtained through the vev of @xmath66 , as by definition the scalar components of @xmath66 do not get a vev ( and an @xmath7-term vev would give an additional @xmath419 suppression ) . a similar conclusion can be obtained for the corrections one obtains by integrating out chiral superfields @xmath119 with mass @xmath420 . we have seen in section [ sec : tree ] that the equations of motion allow to express @xmath119 in terms of the light fields as in eq . ( [ eq : phih ] ) . since @xmath128 contains terms at least trilinear in the fields , the expression for @xmath119 is at least quadratic in the light fields . when plugging eq . ( [ eq : phih ] ) in the canonical khler for @xmath119 one gets again terms that contain at least four light fields , with none of them getting a vev in the scalar component . therefore , no operator @xmath421 can be generated at the tree level by integrating out heavy fields . let us now consider the possibility that the @xmath370 operator above is obtained at the one loop level . this possibility raises two issues . first , @xmath3 would be suppressed compared to , say , the stop mass @xmath317 by a loop factor @xmath422 . as for the case of gaugino masses vs sfermion masses , such a large hierarchy would lead to sfermions beyond the reach of the lhc and a significant fine - tuning . however , as we will see , this problem can be overcome in the same way as for the gaugino masses . we will see in fact in an explicit model that @xmath3 and @xmath423 get a similar enhancement factor . the second issue is the well known @xmath3-@xmath207 problem . @xmath207 is a dimension two parameter generated , as @xmath3 , at the one loop level . therefore , we expect an order of magnitude separation between @xmath424 and @xmath3 : @xmath425 . this is however tolerable in a scheme in which @xmath426 , with @xmath427 and @xmath428 . the explicit model will show that the above pattern can be achieved in the large @xmath429 regime . in turn , the large @xmath429 regime raises a new issue . the minimization of the mssm potential shows in fact that large @xmath429 corresponds to small @xmath430 , while in the situation we want to reproduce , @xmath431 , we expect @xmath432 . in order to make @xmath429 large we therefore need to cancel the contribution to @xmath207 we get at one loop with an additional contribution , at least in the specific example we consider . such a cancellation may not be required in different implementations of the one - loop @xmath370 origin of the @xmath3 parameter . that is why we believe it is worth illustrating the example below despite the cancellation that needs to be invoked . let us consider as before a model involving the following @xmath374 fields : @xmath377 , @xmath254 , @xmath309 , @xmath379 , @xmath236 , with @xmath389 , @xmath433 , @xmath434 . let us denote the coefficients of the @xmath295 and @xmath296 components in the above so(10 ) representations as follows : @xmath435 , @xmath436 , @xmath437 , @xmath438 , @xmath439 , @xmath440 , where @xmath441 , @xmath442 , @xmath443 , @xmath444 , etc . the notation is in agreement with the definition of @xmath445 , @xmath446 in section [ sec:1610 ] . the @xmath3 and @xmath207 parameters , as the gaugino masses , get a vector and a chiral one - loop contribution , see eqs . ( [ eq : dkloop],[eq : dkloop2],[eq : mu],[eq : bmu ] ) . the vector contribution turns out to be [ eq : muvec ] @xmath447 as in the case of gaugino masses , the vector contribution to @xmath3 is suppressed with respect to the sfermion masses by a full loop factor . we therefore need a larger chiral contribution in order to reduce the hierarchy between @xmath3 and @xmath317 . let us then consider the one - loop chiral contribution associated to the superpotential @xmath448 that is easily found to be vanishing because of a pq symmetry of the superpotential . such a pq symmetry can however be broken by adding a term @xmath449 to the above superpotential , coming for example from the non - renormalizable so(10 ) operator @xmath450 after @xmath254 gets its vev ( note that @xmath451 would give @xmath452 ) . en passant , the singlet mass term in eq . ( [ eq : wpq2 ] ) is nothing but the right - handed neutrino majorana mass term entering the see - saw formula for light neutrino masses . note however that no light neutrino mass is generated here , as the light lepton doublets do not have yukawa interactions with the `` right - handed neutrinos '' , @xmath265 . once the pq symmetry is broken by the mass term in eq . ( [ eq : wpq2 ] ) , the @xmath3 and @xmath207 parameters get a chiral one - loop contribution given by [ eq : muchi ] @xmath453 where @xmath454 , @xmath455 are the top and bottom yukawa couplings respectively and the functions @xmath13 , @xmath47 are given by @xmath456 we have assumed the yukawa couplings @xmath274 , @xmath457 to be hierarchical in the basis in which the down yukawa matrix is diagonal . we can see from eq . ( [ eq : muchi ] ) that the one loop chiral contribution to @xmath3 is comparable to the corresponding contribution to @xmath423 if i ) @xmath458 , which corresponds to the large @xmath429 regime ( remember that the bottom mass is given by @xmath459 , where @xmath460 ) ; ii ) @xmath461 , @xmath462 ; iii ) @xmath463 ; iv ) @xmath464 . if the above conditions are satisfied , @xmath465 and both parameters can easily be enhanced , as explained in section [ sec : gauchi ] , for example because @xmath466 . the only non - trivial condition is the large @xmath429 one . remember in fact that @xmath429 is determined by @xmath207 through the minimization of the mssm potential , which gives @xmath467 therefore large @xmath429 , i.e. small @xmath468 , requires a small @xmath207 . this is in contrast with the situation we want to reproduce , @xmath431 . the rge evolution of @xmath207 from the scale at which it is generated ( @xmath469 ) down to the electroweak scale can reduce the value of @xmath207 but not enough to make it as small as we need . a significant rge contribution would in fact require @xmath470 , in contrast with the @xmath471 we are trying to reproduce . we are then forced to invoke a cancellation between the one - loop contribution to @xmath207 in eq . ( [ eq : muchi]b ) and an additional contribution . for example , a tree level contribution to @xmath207 can be obtained as in appendix b or in @xcite . in this paper we have considered what may be regarded as one of the simplest ways to communicate supersymmetry breaking from a hidden to the observable sector , through the tree level , renormalizable exchange of superheavy gauge ( gut ) messengers , and we have studied the general properties of such a tree level gauge mediation ( tgm ) scheme . we have first of all obtained the general structure of the tree - level soft terms arising from a supersymmetry breaking source that is part of a non - trivial gauge ( gut ) multiplet . this is most conveniently done in the effective theory in which the heavy vector superfields associated to the broken generators are integrated out at the tree level ( en passant , in appendix a , we summarized the procedure to integrate out vector superfields and addressed a few minor issues , such as the generalization to the non - abelian case and the role of gauge invariance in a consistent supersymmetric generalization of the expansion in the number of derivatives ) . the scalar soft terms then obtain the two contributions in eq . ( [ eq : mm ] ) , corresponding to the two diagrams on the right - hand side of fig . [ fig : diagram ] . only the first contribution is relevant for scalars that are not in the same gauge multiplet as the scalar partner of the goldstino ( or have not the same sm quantum numbers as some of the gut generators ) . because of the tracelessness condition , such a contribution gives both positive and negative soft masses . this potential phenomenological problem , which has long been considered as an obstacle to tree level supersymmetry breaking , is automatically solved in the models we consider because the fields getting a @xmath19 negative soft mass also get an @xmath472 positive , supersymmetric mass . gaugino masses do not arise at the tree level , but can be generated at the one - loop level , as in ordinary gauge mediation . they receive two contributions , from loops involving heavy vector or chiral superfields . the loop factor suppression of gaugino compared to sfermion masses must be at least partially compensated if the sfermions are to be within the lhc reach and the split - supersymmetry regime is to be avoided . we calculated in full generality the vector and chiral contributions to gaugino masses corresponding to the diagrams in fig . [ fig : mg ] . we have seen that the vector contribution is always suppressed by a full loop factor and is typically subdominant ( often vanishing ) . on the other hand , the chiral contribution is typically larger . we listed four potential enhancement factors that can ( do ) compensate , at least partially , the loop suppression : a larger number of ( chiral ) messengers contributing to gaugino masses than ( vector ) messengers contributing to sfermion masses ; group theoretical factors that in practice turn out to enhance gaugino masses ; the fact that sfermion masses are suppressed by all the vevs with non - vanishing gauge coupling to the vector messengers , while gaugino masses are suppressed only by the vevs that are related to supersymmetry breaking through superpotential interactions ; ratios of yukawa couplings appearing in the expression for the gaugino masses . in minimal models the first two factors partially compensate the @xmath422 loop factor , reducing it to the level of a tolerable ( and possibly necessary ) one order of magnitude hierarchy between gauginos and sfermions . the last two factors are more model - dependent but can give rise to larger enhancements . the general analysis of the tgm scheme allowed us to define the guidelines to obtain phenomenologically viable models from the general formalism and to identify the assumptions underlying the peculiar predictions one obtains . clear model building guidelines emerge , identifying so(10 ) and @xmath52 as the `` minimal '' grand - unified groups , while su(5 ) is found not to have the necessary structure ( rank @xmath473 ) to realize the tgm scheme . the so(10 ) possibility turns out to be quite appealing . it turns out in fact that the so(10 ) breaking vevs of a @xmath255 , important to break so(10 ) to the sm , typically make heavy precisely the components of the so(10 ) representations that need to be made heavy because of their negative soft supersymmetry breaking masses . this reinforces the logical consistency of the tgm framework . in so(10 ) , the tree level sfermion soft masses turn out to be proportional to their charges under the su(5)-invariant so(10 ) generator @xmath222 . we find two possible embeddings of the mssm superfields into so(10 ) representations , depending on whether @xmath474 is positive or negative . in the first case , @xmath475 , the three mssm families are embedded in three @xmath239 and three @xmath241 , @xmath212 . the quark doublets , the up quark singlets , and the lepton singlets , unified in 10 s of su(5 ) , are embedded in the @xmath239 s , while the lepton doublet and down quark singlets , unified in @xmath235 s of su(5 ) , are embedded in the @xmath241 s . they all get positive soft masses . the spare components in the @xmath239 s and @xmath241 s get superheavy , positive , supersymmetric mass terms ( and tev scale negative soft masses ) . in the second case , @xmath476 , the three mssm families are embedded in three @xmath239 and three @xmath334 , @xmath212 . the mssm fields in 10 s of su(5 ) are embedded in the @xmath334 s , while the ones unified in @xmath235 s of su(5 ) , are embedded in the @xmath239 s . as before , they all get positive soft masses . in both cases the chiral content of the theory is still given by three 16 of so(10 ) . an important property of the tgm soft terms is that they turn out to be family universal , thus solving the supersymmetric flavor problem . this property only depends on the hypothesis that the three mssm families are embedded in the same so(10 ) representations . mixed embeddings , in which the mssm fields are superpositions of fields in inequivalent so(10 ) representations , are also possible , but can spoil the flavor universality property . each of the two possible flavor - universal embeddings leads to specific and peculiar predictions for the soft masses at the gut scale : @xmath1 in the @xmath234 case and @xmath2 in the @xmath245 case , where @xmath477 and @xmath478 are common and family - independent soft masses for the fields in the @xmath235 and @xmath236 of su(5 ) respectively . the latter predictions are only based on i ) the use of the `` minimal '' unified gauge group so(10 ) , ii ) the embedding of the mssm families in the so(10 ) representations with dimension @xmath221 not containing the goldstino , and iii ) the absence of mixed embeddings to automatically preserve flavor - universality . the predictions on the ratios @xmath253 in eq . ( [ eq : prediction ] ) are determined by group theory factors and are peculiar enough to make a possible experimental test at the lhc a strong hint for tree level gauge mediation . the embedding into three @xmath257 s has the advantage that the large top yukawa coupling can be accounted for by a renormalizable superpotential interaction involving only low - dimensional ( @xmath479 ) representations for the chiral superfields . in the @xmath258 case , a @xmath480 representation of so(10 ) must be used to reproduce the top yukawa coupling at the renormalizable level . the @xmath52 option is also quite appealing , as the matter superfield content of the @xmath257 embedding is precisely the one obtained from three fundamentals @xmath347 of @xmath52 . the latter decompose in fact as @xmath481 under so(10 ) . we have postponed the investigation of this promising possibility to further study . finally , we have illustrated three possible approaches to the @xmath3-problem in tgm , which we classify according to the dimension @xmath29 of the so(10 ) operator from which the @xmath3-term arises . the @xmath373 option provides a new approach to the @xmath3-problem , peculiar of tgm . the idea is that the supersymmetry breaking scale turns out to coincide with the @xmath3 scale because supersymmetry is triggered by the same @xmath373 so(10 ) operator from which the @xmath3-term arises . we have provided an explicit realization of such a possibility in appendix b. the superpotential shown there also achieves supersymmetry breaking , so(10 ) breaking to the sm , and ensures that only the mssm fields survive below the breaking scale ( in particular it provides doublet - triplet splitting ) . while it is not meant to be simple or realistic , that superpotential represents a useful existence proof . the @xmath366 option is nothing but the nmssm solution of the @xmath3 problem , in which the @xmath3-term is obtained from the vev of a sm singlet superfield stabilized at the supersymmetry breaking scale . we pointed out that the above singlet can easily get a sizable , predictable , negative soft mass term in tgm . this makes giving a vev to the singlet easier than in ordinary gauge mediation ( where its soft mass usually vanishes before rge running ) , provided that the singlet potential can be made stable . the @xmath370 option is nothing but the giudice - masiero mechanism realized at the loop level , as in gauge mediation . the consequent loop hierarchy between the @xmath3-term and the sfermion masses can be reduced exactly as for the gaugino masses . we provided an explicit example , which however needs an extra contribution to the @xmath207 parameter in order to give rise to the necessary large @xmath429 . we thank andrea brignole and claudio scrucca for useful discussions and acknowledge partial support from the rtn european program `` universenet '' ( mrtn - ct-2006 - 035863 ) . we thank the galileo galilei institute for theoretical physics , where part of this work was done , for the hospitality . in this appendix , after a few general comments , we write the effective theory one obtains by integrating heavy vector superfields at the tree level and in unitary gauge in a generic , non - abelian , @xmath34 globally supersymmetric theory with renormalizable khler @xmath195 and gauge - kinetic function ( the superpotential @xmath150 is allowed to be non - renormalizable ) . the general prescription has been studied in @xcite . in particular , it has been shown in @xcite that the usual expansion in the number of derivatives @xmath482 can be made consistent with supersymmetry by generalizing @xmath482 to the parameter @xmath483 where @xmath484 is the number of fermion bilinears and @xmath485 the number of auxiliary fields from chiral superfields . with such a definition , a chiral superfield @xmath486 has @xmath487 and @xmath488 integrations and supercovariant derivatives have @xmath489 . such an expansion makes sense when supersymmetry breaking takes place at a scale much smaller than the heavy superfield mass @xmath4 and in particular when the @xmath7-terms and fermion bilinears from heavy superfields being integrated out are much smaller than @xmath4 . in the presence of vector superfields one should further assume that the @xmath29-terms and gaugino bilinears are small and should generalize eq . ( [ eq : n ] ) to account for the number @xmath490 of gauginos and the number @xmath491 of vector auxiliary fields . we claim that the correct generalization is @xmath492 according to which a vector superfield @xmath85 has @xmath487 . note that the double weight of @xmath29-terms compared to @xmath7-terms is consistent with eq . ( [ eq : dh ] ) . with such a definition , the initial lagrangian has @xmath493 , except for the gauge kinetic term , which has @xmath494 . chiral and vector superfields can then be integrated out at the tree level by using the supersymmetric equations of motion @xmath495 up to terms with @xmath496 when integrating out chiral superfields and @xmath497 when integrating vector superfields ( with the missing terms originating from the gauge - kinetic term having @xmath498 ) . from a physical point of view , we are interested not only in the expansion in @xmath215 but also , and especially , in the expansion in the power @xmath386 of @xmath499 . it is therefore important then to remark that using eqs . ( [ eq : susyeom ] ) amounts to neglecting terms with @xmath500 when integrating chiral superfields and @xmath501 when integrating out vector superfields . we are now ready to present out results on the effective theory obtained integrating out the heavy vector superfields in a generic supersymmetric gauge theory as above . we are interested in operators with dimension up to 6 ( @xmath502 ) in the effective theory . we can then use the equation @xmath503 . neglecting higher orders in @xmath386 , the latter equation can be rewritten as @xmath504 where @xmath66 is defined in eq . ( [ eq : goldstones ] ) , @xmath505 , the indices run over the broken generators , and @xmath506 is a function of the light vector superfields : @xmath507 ( m^2_{v2})_{ab } = \frac{g^4}{3 } \phi_0^ * t_a^h v^l v^l t^h_b \phi_0 + ( a \leftrightarrow b ) . \end{gathered}\ ] ] in order to solve eq . ( [ eq : vh ] ) for @xmath508 , we need to invert the field - dependent matrix @xmath506 . in the wess - zumino gauge for the light vector superfields , we get @xmath509 the effective contribution to the khler potential is @xmath510 where [ eq : keff2 ] @xmath511 \delta k^1_\text{eff } & = - 2 g^3 ( m_{v0}^2)^{-1}_{ab } ( \phi'^\dagger t^h_a \phi ' ) ( \phi'^\dagger \{v^l , t^h_b\ } \phi ' ) { \stepcounter{mysubequation}\tag{\theequation\protect{{\alph{mysubequation}}}}}\\ \delta k^2_\text{eff } & = - \frac{4}{3 } g^4 ( m_{v0}^2)^{-1}_{ab } ( \phi'^\dagger t^h_a \phi ' ) \phi'^\dagger ( t^h_b v^l v^l + v^l t^h_b v^l + v^l v^l t^h_b ) \phi ' { \stepcounter{mysubequation}\tag{\theequation\protect{{\alph{mysubequation}}}}}\\ & - g^4 ( m_{v0}^2)^{-1}_{ab } \left ( ( \phi'^\dagger \ { t^h_a , v^l \ } \phi ' ) ( \phi'^\dagger \ { t^h_b , v^l \ } \phi ' ) + \frac{1}{3 } ( \phi'^\dagger [ t^h_a , v^l ] \phi ' ) ( \phi'^\dagger [ t^h_b , v^l ] \phi ' ) \right ) . \notag\end{aligned}\ ] ] in recovering eq . ( [ eq : keff2]c ) we have used the identity @xmath512 where @xmath513 are the structure constants of the gauge group , the latin indices refer to broken generators and the greek one refers to an unbroken one . we are interested to soft supersymmetry breaking terms arising from eqs . ( [ eq : keff2 ] ) when some of the auxiliary fields get a vev . the relevant terms should contain up to two @xmath7-terms and one @xmath29-term ( eq . ( [ eq : dh ] ) ) . the only relevant terms are therefore those in ( [ eq : keff2]a ) . in this appendix we provide an example of a superpotential achieving supersymmetry breaking , so(10 ) breaking to the sm , ensuring that below the scale of this breaking only the mssm fields survive ( in particular providing doublet - triplet splitting ) and solving of the @xmath3-problem . we do not aim at being simple or realistic , we just aim at providing an existence proof . we include in this example representations with dimension @xmath514 . it would be interesting to obtain a dynamical supersymmetry breaking , in particular the @xmath7-term vev of a 16 of so(10 ) . so(10 ) will broken to the sm at a scale @xmath515 . below this scale only the mssm fields survive , in particular the higgs triplets are made heavy via a generalization of the dimopoulos - wilczek mechanism @xcite . the @xmath3-term is present in the theory in the form of a @xmath373 operator present at the gut scale and triggers supersymmetry breaking . @xmath516 is generated at the tree - level and turns out to be of the same order as the sfermion masses . the superpotential we use is @xmath517 where @xmath518 here we denote the fields according to their so(10 ) representation , except the so(10 ) singlet fields @xmath519 . the mass parameter @xmath386 is of the order of the tev scale ( we do not discuss the origin of such a small parameter here ) , while all other mass parameters are near the gut scale @xmath520 let us discuss the role of the different contributions to the superpotential and anticipate the vacuum structure and the spectrum . @xmath521 is responsible for supersymmetry breaking and the breaking of so(10 ) to su(5 ) : as we are going to show below , this part of the superpotential generates @xmath472 vevs for the scalar components of @xmath377 and @xmath522 along the su(5 ) singlet direction @xmath523 and @xmath524 and a supersymmetry breaking vev for the @xmath7-term component of @xmath525 along the su(5 ) singlet direction @xmath526 . it also provides small supersymmetry breaking vevs for the @xmath7-term component of @xmath222 @xmath527 and for the @xmath29-term of the vector superfield corresponding to the u(1)@xmath528 generator of so(10 ) @xmath529 this @xmath29-term vev will generate sfermion masses along the lines of section [ sec:1610 ] . this superpotential appears in eq . ( [ eq : wvev ] ) and is a generalization of an example in @xcite . @xmath530 contains the mssm yukawa couplings and provides supersymmetry breaking masses for heavy chiral superfields that will generate gaugino masses at 1-loop as in ordinary gauge mediation . the mssm matter is embedded in both the @xmath239 and the @xmath241 , as explained in section [ sec:1610 ] . the mssm higgs fields are linear combinations of different fields and have components in different representations , @xmath531 therefore the first term in @xmath530 contains the up - type yukawas , while the second and third terms provide down - type and charged lepton yukawas . the second term also gives a large mass to the additional fields @xmath532 and @xmath533 the latter are also the only fields that couple to the @xmath7-term vev in the @xmath525 and act as one - loop messengers of supersymmetry breaking . while this gives a subleading contribution to sfermion masses , it is the only source of gaugino masses in this model . + the role of @xmath534 is the breaking of su(5 ) to the standard model gauge group . it provides a large vev for the @xmath535 along the @xmath536 direction @xmath537 as needed for the dimopoulos - wilzcek mechanism . also @xmath538 and the sm singlet in the @xmath539 take large vevs . @xmath128 merely gives large masses to components in the @xmath525 and @xmath540 . note that since the @xmath541 does not contain su(5 ) singlets , the neutrino component in the @xmath525 stays massless as it should , being the dominant component of the goldstino superfield . @xmath542 takes care of the higgs sector : it keeps the mssm higgs doublets light and gives a large mass to the corresponding triplets . its last term provides the @xmath207 term because @xmath6 gets a small supersymmetry breaking vev and both @xmath543 and @xmath544 have components in the @xmath236 . the @xmath3 term is contained in @xmath521 , because @xmath544 has a component also in the @xmath525 . we are interested in a vacuum that does not break the sm gauge group . thus only that part of the superpotential which involve su(5 ) singlets is relevant for the determination of the ground state . we denote the singlets in @xmath545 by @xmath546 ( which is different than the notation used in the main text ) and the singlets in the @xmath547 by @xmath548 , where @xmath549 are the properly normalized fields corresponding to the @xmath536 and @xmath550 generators in so(10 ) . the relevant part of the superpotential is @xmath551 & + { \overline{s}^{\prime \prime}}\left ( -\frac{{\lambda_{3 } \,}}{2 } t + \frac{{\lambda_{3 } \,}}{2 } \sqrt{\frac{3}{2 } } b + { \lambda_{4 } \ , } u\right ) s + { \overline{s}}\left ( -\frac{{\lambda_{5 } \,}}{2 } t + \frac{{\lambda_{5 } \,}}{2 } \sqrt{\frac{3}{2 } } b + { \lambda_{6 } \ , } u \right ) { s^{\prime \prime}}\\[-1 mm ] & + { m_{45}}(b^2 + t^2 ) + { m_{54}}v^2 + { \lambda_{7 } \ , } v \left ( \frac{1}{2 } \sqrt{\frac{3}{5 } } t^2 - \frac{1}{\sqrt{15 } } b^2 \right ) . \notag\end{aligned}\ ] ] the @xmath7-term and @xmath552-term equations show that susy is broken ( @xmath553 ) and that all vevs are determined except @xmath554 , for which there exist three solutions , all yielding @xmath555 . this tree - level degeneracy is lifted by one - loop corrections which select the solution with @xmath556 . one can check that the vevs are given by @xmath557 s = m - \frac{m^2}{4 m } \left(\frac{1}{\lambda_1 ^ 2 } - \frac{1}{50 g^2}\right ) \qquad { \overline{s}}= m - \frac{m^2}{4 m } \left(\frac{1}{\lambda_1 ^ 2 } + \frac{1}{50 g^2}\right ) \notag \\ u = - \frac{3 \sqrt{5 } } { 2 } \frac{{\lambda_{5 } \,}}{{\lambda_{6 } \ , } { \lambda_{7 } \ , } } \sqrt{{m_{45}}{m_{54 } } } \qquad { u^\prime}= - \frac{3 \sqrt{5 } } { 2 } \frac{{\lambda_{3 } \,}}{{\lambda_{4 } \ , } { \lambda_{7 } \ , } } \sqrt{{m_{45}}{m_{54 } } } \\[1 mm ] v = \frac { \sqrt{15 } { m_{45}}}{{\lambda_{7 } \ , } } \qquad b = \frac{\sqrt{30}}{{\lambda_{7 } \ , } } \sqrt{{m_{45}}{m_{54 } } } \notag \\ f_{{s^\prime } } = - m m \qquad f_z = \frac{m^2}{2 { \lambda_{1 } \ , } } \qquad d_x = - \frac{m^2}{10 g}. \notag \end{gathered}\ ] ] in order to identify the light ( with respect to @xmath558 ) we can set @xmath559 and consider the supersymmetric limit . most fields are at the gut scale , with the light ones being the mssm ones , the goldstino superfield @xmath560 , and the right - handed neutrinos in the @xmath239 , which can easily be made heavy through a non - renormalizable superpotential operator @xmath561 . the mssm matter fields are embedded in the @xmath249 and in the @xmath562 , as desired . the higgs doublets are embedded into the @xmath563 and @xmath564 according to @xmath565 with normalization factors @xmath566 and @xmath567 , where @xmath568 , @xmath569 denote the sm component with the quantum numbers of @xmath296 , @xmath296 in the so(10 ) representation @xmath570 . after switching on @xmath386 the soft supersymmetry breaking terms and @xmath3-term are generated . the @xmath3-term is already present in the high energy lagrangian and is of order @xmath386 , the vev of @xmath552 generates sfermion and higgs masses of order @xmath571 and the vev of @xmath572 gives rise to a @xmath207 term of order @xmath571 . the heavy fields @xmath244 and @xmath243 act as messengers of susy breaking to the gauginos who get masses of order @xmath573 . the goldstino will be mainly the fermion in @xmath560 but gets also small contributions from the gaugino corresponding to the @xmath574 generator and the fermion in @xmath6 . the corresponding scalar will get a mass of order @xmath571 . h. p. nilles , phys . b * 115 * ( 1982 ) 193 ; a.h . chamseddine , r. arnowitt and p. nath , phys . * 49 * , 970 ( 1982 ) ; r. barbieri , s. ferrara and c. a. savoy , phys . b * 119 * , 343 ( 1982 ) ; l.e . ibez , phys . b * 118 * , 73 ( 1982 ) ; l.j . hall , j.d . lykken and s. weinberg , phys . d * 27 * , 2359 ( 1983 ) ; n. ohta , prog . phys . * 70 * , 542 ( 1983 ) . m. dine , w. fischler and m. srednicki , nucl . b * 189 * , 575 ( 1981 ) ; m. dine and w. fischler , phys . b * 110 * , 227 ( 1982 ) ; m. dine and a. e. nelson , phys . d * 48 * , 1277 ( 1993 ) [ arxiv : hep - ph/9303230 ] ; m. dine , a. e. nelson and y. shirman , phys . rev . d * 51 * , 1362 ( 1995 ) [ arxiv : hep - ph/9408384 ] . l. randall and r. sundrum , nucl . b * 557 * , 79 ( 1999 ) [ arxiv : hep - th/9810155 ] ; g. f. giudice , m. a. luty , h. murayama and r. rattazzi , jhep * 9812 * , 027 ( 1998 ) [ arxiv : hep - ph/9810442 ] . d. e. kaplan , g. d. kribs and m. schmaltz , phys . d * 62 * , 035010 ( 2000 ) [ arxiv : hep - ph/9911293 ] ; z. chacko , m. a. luty , a. e. nelson and e. ponton , jhep * 0001 * , 003 ( 2000 ) [ arxiv : hep - ph/9911323 ] . m. nardecchia , a. romanino and r. ziegler , jhep * 0911 * ( 2009 ) 112 [ arxiv:0909.3058 [ hep - ph ] ] . s. ferrara , l. girardello and f. palumbo , phys . d * 20 * ( 1979 ) 403 . s. dimopoulos and h. georgi , nucl . b * 193 * ( 1981 ) 150 . s. weinberg , _ the quantum theory of fields . 3 : supersymmetry _ , cambridge university press . j. wess and j. bagger , _ supersymmetry and supergravity _ , princeton university press . l. brizi , m. gomez - reino and c. a. scrucca , nucl . b * 820 * ( 2009 ) 193 [ arxiv:0904.0370 [ hep - th ] ] . i. affleck , m. dine and n. seiberg , nucl . b * 256 * ( 1985 ) 557 ; a. pomarol and s. dimopoulos , nucl . b * 453 * ( 1995 ) 83 [ arxiv : hep - ph/9505302 ] ; r. rattazzi , phys . b * 375 * , 181 ( 1996 ) [ arxiv : hep - ph/9507315 ] ; n. arkani - hamed , m. dine and s. p. martin , phys . b * 431 * , 329 ( 1998 ) [ arxiv : hep - ph/9803432 ] . k. a. intriligator and n. seiberg , nucl . * 45bc * ( 1996 ) 1 [ arxiv : hep - th/9509066 ] ; k. a. intriligator and n. seiberg , class . quant . grav . * 24 * ( 2007 ) s741 [ arxiv : hep - ph/0702069 ] . n. arkani - hamed , l. j. hall , h. murayama , d. tucker - smith and n. weiner , phys . d * 64 * ( 2001 ) 115011 [ arxiv : hep - ph/0006312 ] ; d. a. demir , l. l. everett and p. langacker , phys . lett . * 100 * , 091804 ( 2008 ) [ arxiv:0712.1341 [ hep - ph ] ] . n. arkani - hamed and s. dimopoulos , jhep * 0506 * ( 2005 ) 073 [ arxiv : hep - th/0405159 ] ; g. f. giudice and a. romanino , nucl . b * 699 * ( 2004 ) 65 [ erratum - ibid . b * 706 * ( 2005 ) 65 ] [ arxiv : hep - ph/0406088 ] . j. hisano , h. murayama and t. goto , phys . d * 49 * , 1446 ( 1994 ) . i. l. buchbinder , s. kuzenko and z. yarevskaya , nucl . b * 411 * ( 1994 ) 665 ; b. de wit , m. t. grisaru and m. rocek , phys . b * 374 * ( 1996 ) 297 [ arxiv : hep - th/9601115 ] ; a. pickering and p. c. west , phys . b * 383 * ( 1996 ) 54 [ arxiv : hep - th/9604147 ] ; m. t. grisaru , m. rocek and r. von unge , phys . b * 383 * ( 1996 ) 415 [ arxiv : hep - th/9605149 ] ; a. brignole , nucl . b * 579 * ( 2000 ) 101 [ arxiv : hep - th/0001121 ] . j. polchinski and l. susskind , phys . d * 26 * ( 1982 ) 3661 . s. dimopoulos and g. f. giudice , phys . b * 393 * ( 1997 ) 72 [ arxiv : hep - ph/9609344 ] . g. r. dvali , g. f. giudice and a. pomarol , nucl . b * 478 * ( 1996 ) 31 [ arxiv : hep - ph/9603238 ] . r. slansky , phys . rept . * 79 * ( 1981 ) 1 . m. frigerio , p. hosteins , s. lavignac and a. romanino , nucl . b * 806 * ( 2009 ) 84 [ arxiv:0804.0801 [ hep - ph ] ] . l. calibbi , m. frigerio , s. lavignac and a. romanino , arxiv:0910.0377 [ hep - ph ] . l. girardello and m. t. grisaru , nucl . b * 194 * ( 1982 ) 65 . n. arkani - hamed , s. dimopoulos , g. f. giudice and a. romanino , nucl . b * 709 * , 3 ( 2005 ) [ arxiv : hep - ph/0409232 ] . u. ellwanger , c. hugonie and a. m. teixeira , arxiv:0910.1785 [ hep - ph ] . p. langacker , n. polonsky and j. wang , phys . d * 60 * ( 1999 ) 115005 [ arxiv : hep - ph/9905252 ] . a. de gouvea , a. friedland and h. murayama , phys . d * 57 * ( 1998 ) 5676 [ arxiv : hep - ph/9711264 ] . g. f. giudice and a. masiero , phys . b * 206 * ( 1988 ) 480 . s. dimopoulos and f. wilczek , nsf - itp-82 - 07 ( unpublished ) ; m. srednicki , nucl . b * 202 * ( 1982 ) 327 . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: scattering experiments are amongst the most powerful tools to obtain information on the microscopic structure of matter . in solid state physics , conventional elastic scattering of photons , electrons , and neutrons are the standard methods to study the precise spacing and the location of atoms in solids . one obtains this information by measuring the angular distribution of the diffracted particles which depends on the microscopic lattice structure of the studied solid . in conventional x - ray and electron diffraction experiments , the incident particles interact with all the electrons in the sample and , since the majority of the electrons is located close to the nucleus , these experiments yield the averaged position of the atoms . the situation is very similar for neutron diffraction , for which scattering occurs from the nuclei directly . in addition to the sensitivity to the lattice structure , neutron and non - resonant x - rays also interact with the magnetic moments and therefore information on the magnetic structures in solids can be derived . however , in many of the currently most intensively studied systems not only the behaviour of lattice and spins needs to be studied . also the charge and orbital degrees of freedom often play an essential role . in some materials a collective electronic ordering of spins , charges , and orbitals occurs which typically only affects a small fraction of the valence electrons . these phenomena can lead to novel exciting ground states . for instance , complex electronic phenomena involving the cooperative ordering of various electronic degrees of freedom are discussed intensively in relation with such outstanding phenomena like high - temperature superconductivity in cuprates@xcite or the colossal magneto - resistance in manganites@xcite . as we will describe later on , charge and orbital orders are very difficult to observe with the aforementioned traditional scattering techniques . a new experimental probe , which enables to observe complex electronic order was therefore urgently needed . resonant ( elastic ) soft x - ray scattering ( rsxs ) provides exactly this : a highly sensitive probe for spacial modulations of spins , charges , and orbitals in complex materials . this unique sensitivity is achieved by merging diffraction and x - ray absorption spectroscopy ( xas ) into a single experiment , where the scattering part provides the information of spatial modulations and the xas part provides the sensitivity to the electronic structure . more precisely , rsxs close to an absorption edge involves virtual transitions from core levels into unoccupied states close to the fermi level and these virtual transitions depend strongly on the spin , charge and orbital configuration of the resonant scattering centers . the ordering of spins , charges , and orbitals typically results in electronic superlattices with periodicities of several nanometers , which matches very well with the wavelengths of soft x - rays lying between between @xmath0 6 to 0.6 nm corresponding to photon energies between @xmath0 200 ev to 2000 ev . moreover , since the virtual excitations in rsxs are related to specific core level excitations , the excitation energy of which changes from element to element , the method is element specific like xas . thus rsxs , different from magnetic neutron scattering can probe magnetic structures related to specific elements . the price paid for the unique sensitivity offered by rsxs is a very limited ewald sphere and a rather short photon penetration depth limiting studies to only the topmost 100 atomic layers or , depending on the resonance used , even less . but these limitations are very often compensated by the gain in sensitivity , which is extremely important when turning to magnetic and electronic properties of samples characterized by a very small amount of contributing material , i.e. , for studying thin films , nanostructures as well as surfaces and interfaces . as a consequence of the broken translational symmetry at the surface , nanosystems can show ordering phenomena which differ locally or macroscopically from those of the respective bulk systems , and even completely new phenomena can arise . for such systems , rsxs has been established as a very powerful and unique tool to study complex electronic ordering phenomena on a microscopic scale . the need for such studies has strongly increased within the last decade since the fabrication of high quality samples with macroscopic properties tunable by composition , strain , size and dimensionality has become possible , raising the hope for future multifunctional heterostructures characterized by so far unexpected and unexplored novel material properties and functionality . although rsxs offers many important ingredients to study condensed matter , it has only recently developed its full power . the reason for this is that intense soft x - ray sources with tunable energy only became available with the advent of 2nd and 3rd generation synchrotron radiation facilities . furthermore those soft x - rays are absorbed in air and therefore the diffraction stations have to operate in vacuum , which poses another complication and made the development of rsxs challenging . finally , as mentioned above , the penetration depth into the solids is of the order of nanometers which means that surface effects may become important . thus in many surface sensitive systems the surfaces should be prepared under ultra - high vacuum ( uhv ) conditions or at least should be kept under uhv during the measurements . this review covers the recent progress in rsxs and gives a survey of the application of the technique . after the introduction , the principles of rsxs are presented , followed by a description of the experimental development . the main part is then devoted to the application of rsxs in the study of magnetic structures , charge order , and orbital order in thin layers , artificial structures , interfaces , and correlated systems . the review concludes with a summary and an outlook . elastic resonant x - ray scattering combines x - ray spectroscopy and x - ray diffraction in one single experiment . roughly speaking , x - rax diffraction provides the information about the spatial order , while the spectroscopic part provides the sensitivity to the electronic states involved in the ordering . a first qualitative understanding of this strongly enhanced sensitivity to electronic order can be gleaned from the schematic illustration of the resonant scattering process shown in figure[fig : rsxs_scheme ] : the incoming photon virtually excites a core electron into the unoccupied states close to the fermi level , thereby creating the so - called intermediate state @xmath1 of the resonant scattering process . this virtual transition depends very strongly on the properties of the valence shell and , therefore , results in the tremendously enhanced sensitivity of resonant x - ray scattering to electronic ordering . the state @xmath1 then decays back into the ground state @xmath2 and a photon with the same energy as the incoming one is re - emitted . this combination of spectroscopy and diffraction will be described in the following . in this section , aspects of the diffraction of x - rays by a crystal will be summarized . for more detailed and extensive descriptions of this topic the reader is referred to the literature@xcite . + + in the following we will consider a crystal that is formed by a perfectly periodic arrangement of lattice sites , which act as scattering centers for the incident x - ray field . the scattering from site @xmath3 in the crystal is described in terms of a scattering length @xmath4 , which is also called the form factor , and can be represented as@xcite @xmath5 the first two terms @xmath6 and @xmath7 represent the non - resonant charge and magnetic scattering , respectively , where the scattering described by @xmath6 , which is proportional to the total number of electrons of the scatterer , is called thomson scattering . @xmath8 denotes the so - called dispersion correction , which is not only a function of the photon energy @xmath9 , but also of the polarisation of the incoming ( @xmath10 ) and scattered beam ( @xmath11 ) . this correction becomes very important close to absorption edges and describes the resonant scattering processes illustrated in figure[fig : rsxs_scheme ] . and @xmath12 , respectively . the whole crystal can be described by discrete translations of the unit cell with its basis . the relative phase of x - rays scattered by the two sites of the basis is @xmath13 ( right panel).,width=283 ] physically the scattering length @xmath4 describes the change in amplitude and phase suffered by the incident wave during the scattering process : a scatterer with scattering length @xmath4 exposed to an incident plane wave @xmath14 causes a scattered radial wave @xmath15 , which , far away from the scattering center , can be approximated by a plane wave . in order to calculate the intensity of the total scattered wave field with wave vector @xmath16 , all the radial waves have to be summed with the correct relative phases . as illustrated in figure[fig : scatter_scheme ] , the geometric relative phase of two scatterers is given by @xmath17 , where @xmath18 is the scattering vector and @xmath19 is the difference in position . the total intensity detected in a distant detector is therefore @xmath20 where @xmath21 is the vector pointing to the origin of unit cell @xmath22 and @xmath23 is the position of the scatterer @xmath4 measured from that origin . @xmath24 is called the unit cell structure factor and describes the interference of the waves scattered from the different sites within a unit cell . the lattice sum @xmath25 is due to the interference of the scattering from the different unit cells at @xmath21 . its @xmath26-dependence therefore provides information about the number of sites scattering coherently , i.e. , it is related to the correlation length of the studied order . for an infinite number of coherently scattering unit cells , @xmath27 , where @xmath28 represents the reciprocal lattice . this corresponds to the well - known laue condition , which states that a reflection can only be observed if @xmath29 is a reciprocal lattice vector . the treatment described above is known as the kinematic approximation . this approximation is valid as long as the intensity of the scattered wave is much weaker than that of the incidence wave , which means that the interaction between the incoming and outgoing waves or multiple scattering events can be neglected . since in resonant x - ray diffraction experiments one usually deals with weak superlattice reflections , this approximation is valid in most cases . if , however , the scattered wave becomes too strong , then one has to resort to the so - called dynamical theory of diffraction . this description will not be presented here and the interested reader is referred to the literature@xcite . and @xmath16 , is vertical in this example . incoming and outgoing polarisations are denoted as @xmath30 and @xmath31 , respectively , where @xmath32 indicates a polarisation perpendicular to the scattering plane and @xmath33 refers to a polarisation parallel to the scattering plane . the polarisation of the scattered radiation can be determined using the polarisation analyzer . @xmath34 define the reference frame in which the polarisations , wave vectors and spin - directions will be expressed.,width=283 ] + + a typical geometry of a resonant scattering experiment is shown in figure[fig : exp_scheme ] . an incoming photon beam with defined wave vector @xmath35 impinges on the sample and is scattered elastically into the direction defined by @xmath16 , corresponding to a scattering vector @xmath18 . in the so - called specular geometry the incoming and the outgoing beam are at an angle @xmath36 with the sample surface . in this case , the scattering angle between @xmath35 and @xmath35 is @xmath37 . the wave vectors of the incident and scattered beam define the scattering plane , which is vertical in the present example . polarisation directions parallel or perpendicular to this scattering plane are referred to as @xmath33- and @xmath32-polarisation , respectively . correspondingly , the incoming and outgoing polarisations in figure[fig : exp_scheme ] are denoted as @xmath30 and @xmath31 . an important aspect of resonant x - ray scattering is given by the polarisation dependence of @xmath38 at resonance . this can cause the polarisation of the outgoing beam to be different from that of the incoming beam . a typical example , which is frequently encountered in resonant scattering experiments , is a change of the incoming polarisation from @xmath39 to an outgoing polarisation @xmath40 ; so - called @xmath41-scattering . such changes in the polarisation contain important information about the symmetry of the studied order , as will be discussed in more detail later on . controlling both the incoming and the outgoing polarisation can therefore be a big advantage . the incoming polarisation can be routinely controlled by modern insertion devices like elliptical undulators . the determination of the outgoing polarisation in rsxs is currently less common , but can be achieved using a polarisation analyzer as sketched in figure[fig : exp_scheme ] . by switching between the two configurations shown in the figure , for example , it is possible to observe either the @xmath40- or the @xmath42-component of the scattered beam . in the following discussion of the various case studies , a number of different scan - types will be described , which we will introduce briefly here : * radial scan or @xmath43 scan . this is a scan along the direction defined by the scattering vector @xmath44 , which therefore provides information about the correlations along @xmath44 . in practice this scan is done by rotating the sample and the detector in steps of @xmath45 and @xmath46 about @xmath47 , respectively , which is equivalent to changing @xmath36 of the incoming and outgoing beam by the same amount . * transverse scan . this is a scan perpendicular to the direction defined by @xmath44 . it hence contains information about correlation in the plane perpendicular to @xmath44 , but is also often affected by the sample mosaic . a transverse scan can be done by rotating the sample about @xmath48 or @xmath47 , while keeping the detector position fixed . * azimuth scan . this is another way of investigating the polarisation dependence of the resonant scattering . in this type of scan , the resonant scattered intensity is recorded while rotating the sample by an angle @xmath49 about the scattering vector @xmath44 ( cf . figure [ fig : exp_scheme ] ) . azimuthal scans will be described in more detail in sec.[sec : pol_dep ] . * energy scan . here the photon energy dependence of a given reflection at @xmath44 is measured . in many cases this measurement is performed by scanning the photon energy , while recording the scattered intensity at @xmath18 . since both @xmath35 and @xmath16 change with the photon energy , this means that for each photon energy the scattering geometry has to be adjusted such as to keep @xmath44 constant . in the following sections[sec : lightmatter]-[sec : rsxsandxas ] the resonant scattering length and its relation to x - ray absorption will be described in a rather formal way . readers who are mainly interested in the application of the method , can skip this theoretical part and continue with sec.[sec : hannon ] . readers who would like to learn more about the theoretical description of the interaction between light and matter can find a very nice treatment with more details in ref.@xcite . + + in general , the scattering of photons caused by matter is due to the interactions between the electromagnetic wave and the particles in a solid . in the present case only the interactions between the electromagnetic wave and the electrons in a solid are important . the corresponding coupling term @xmath50 is obtained by replacing the momentum operator @xmath51 in the free electron hamiltonian @xmath52 by @xmath53 , which gives @xmath54 here , @xmath51 and @xmath55 correspond to the _ l_th electron of the atom . furthermore , @xmath56 for a transversal radiation field @xmath57 has been used . note also , that the spin dependent part of @xmath50 , which is proportional to and , has been neglected , because these relativistic terms scale as @xmath58 and hence are very small compared to the terms in eq.[hint ] . the quantized electromagnetic field that couples to the electrons can be expressed as @xmath59 where @xmath60}$ ] and @xmath61}$ ] . the operators @xmath62 and @xmath63 , respectively , create and annihilate a photon with polarisation @xmath64 and wave vector @xmath65 . since the first term in eq.[hint ] is proportional to @xmath66 it is linear in @xmath62 and @xmath63 and therefore describes processes involving the emission and absorption of one photon . in other words , the @xmath66 term changes the number of photons by @xmath67 . the second term in eq.[hint ] is proportional to @xmath68 and is therefore quadratic in the @xmath62 and @xmath63 . as a result , this term changes the number of photons by 0 or @xmath69 . + + as introduced above , the scattering of a photon from a specific site @xmath3 in the crystal is described by the scattering length @xmath4 , which is determined by the interactions given by eq.[hint ] . the most important terms entering @xmath4 can be calculated by means of time dependent perturbation theory up to second order in @xmath50 . in the following we will briefly sketch the main steps of this calculation and will present the final result . a detailed derivation of the expression for @xmath4 can be found , for example , in ref.@xcite . since in the following we will consider the scattering from a single site , we will drop the site index @xmath3 . the following scattering event will be considered : a single photon with wave vector @xmath65 and polarisation @xmath10 impinges on a lattice site in the initial state @xmath2 and is scattered into a state with wave vector @xmath70 and polarisation @xmath11 . since we are dealing with elastic scattering , the lattice site will remain in state @xmath2 ; i.e. latexmath:[\[\label{scatter } only one photon exists and , therefore , the only non - vanishing contributions of @xmath50 ( eq.[hint ] ) to the scattering process must contain one creation and one annihilation operator corresponding to @xmath72 and @xmath73 , respectively . the terms first order in @xmath50 are given by the matrix element @xmath74 . since the interaction term @xmath66 is linear in the @xmath63 and @xmath75 , it changes the net number of photons and therefore does not contribute to @xmath76 . only the term proportional to @xmath77 contains products of @xmath78 and @xmath63 , which do not change the number of photons and , hence , only this term gives non - vanishing contributions to @xmath76 . while the @xmath66 term does not contribute to the first order matrix element @xmath76 , there are contributions of this term in second order perturbation theory , which are of the same order of magnitude as @xmath76 . these second order contributions have to be considered as well . the second order matrix element @xmath79 involves two successive @xmath80 interactions . the dominant contributions to @xmath79 are due to terms , where the incoming photon is annihilated first and then the scattered photon created : @xmath81 . here , @xmath1 denotes an intermediate state of the system without a photon . taken together , the terms given by @xmath76 and @xmath79 yield the following expression for the differential cross - section @xmath82 , i.e. , the probability that a photon is scattered into the solid angle @xmath83 : @xmath84 in this equation , @xmath85 is the classical electron radius , @xmath86 corresponds to the fourier amplitude at @xmath87 of the charge density , @xmath88 ( @xmath89 ) is the ground state ( intermediate state ) of the scatterer ( cf . figure[fig : rsxs_scheme ] ) , @xmath90 ( @xmath91 ) is the energy of @xmath88 ( @xmath89 ) , @xmath92 is the life time of @xmath1 and @xmath93 is the current operator that describes the virtual transitions between @xmath88 and @xmath89 . equation[kheqn ] is the famous kramers - heisenberg formula applied to elastic scattering of photons . by definition the differential cross - section is related to the scattering length by @xmath94 , i.e. , the above equation also provides the expression for @xmath4 . the first term proportional to @xmath95 describes the non - resonant thomson scattering of photons from the total charge density @xmath96 . the scattering due to the thomson term is given by the first order matrix element @xmath76 , which describes the direct scattering of a photon caused by the @xmath77 interaction . this non - resonant scattering scales with the total number of electrons and , hence , usually does not provide high sensitivity to electronic ordering phenomena , which typically affect only a very small fraction of @xmath96 . the second order matrix element @xmath79 results in the second term in eq.[kheqn ] . this term describes the resonant scattering processes of the kind illustrated in figure[fig : rsxs_scheme ] . the matrix elements in the nominators describe the virtual @xmath2@xmath97 @xmath1@xmath97 @xmath2 processes , which correspond to the two successive @xmath80 interactions . close to an absorption edge , the photon energy @xmath98 is close to some of the @xmath99 , in which case the corresponding contribution to the sum becomes large and can completely dominate the scattering . according to the above discussion , the scattering length @xmath100 can be written as @xmath101 where @xmath102 represents the non - resonant thomson scattering and @xmath103 is the anomalous dispersion correction . equation[eq : scattkh ] corresponds exactly to eq.[eq : f_general ] given above with the magnetic scattering neglected . + + in many cases , the current operator @xmath104 is treated in the dipole approximation , which means that for @xmath105 one can use @xmath106 and therefore @xmath107 . in the dipole approximation the dependence of @xmath103 on @xmath65 and @xmath70 is therefore neglected . furthermore , since @xmath108=-2i\hbar { \bf p}$ ] , one can replace the momentum operators @xmath51 appearing in @xmath104 by the commutator @xmath109 $ ] . this together with the resonant term in eq.[kheqn ] then yields the following expression for the resonant scattering length : @xmath110 here , @xmath111 is the dipole operator . the polarisation dependence in the above equation can also be expressed using the tensor @xmath112 , which is defined by @xmath113 . this tensor formalism is often found in the literature and will also be used in the following . + + as can be already realized in figure[fig : rsxs_scheme ] , the excitation from the ground state into the intermediate state corresponds to an x - ray absorption process . therefore , there should be a relation between the resonant scattering length @xmath103 and the x - ray absorption cross section . that such a direct relation indeed exists can be seen in the following way : firstly , @xmath103 can also be written more compactly as @xmath114^+ \ , \mathcal{g}(\hbar \omega)\ , [ { \bf e}.{\bf j}(-{\bf k})]\,|g\rangle\ ] ] using the greens function @xmath115 note that @xmath116 is an operator , which is related to the hamiltonian @xmath117 describing the intermediate state , i.e. , the state with the core - hole . secondly , the transition probability per unit time @xmath118 for the x - ray absorption is given by fermi s golden rule : @xmath119 as explained in sec.[sec : lightmatter ] , only the @xmath66 is linear in the photon annihilators and , hence , it is the only term active in the x - ray absorption process . the effective vector field @xmath120 , which couples @xmath2 to @xmath121 can be expressed as@xcite . @xmath122 the action of the operator @xmath63 on the multiphoton state yields the factor @xmath123 , where @xmath124 is the number of photons with wave vector @xmath35 and polarisation @xmath125 . furthermore , @xmath118 together with the photon flux @xmath126 gives the the absorption cross - section @xmath127 . this relation together with eqs.[eqn : fermi ] and [ aabsorp ] gives the following result : @xmath128^+ \ , \mathcal{g}(\hbar \omega)\ , [ { \bf e}.{\bf j}(-{\bf k})]\,|g\rangle\,\right]\end{aligned}\ ] ] since the greens function entering @xmath129 is also given by eq.[eqn : green ] , the comparison of eqs.[eqn : dfgreen ] and [ eqn : abs ] shows that @xmath130\ ] ] for @xmath131 and @xmath132 . this important relation is known as the optical theorem . it enables one to determine the imaginary part of the resonant scattering length from x - ray absorption . once @xmath133 $ ] is known , the corresponding real part can be calculated by means of the kramers - kronig relations , which read @xmath134&=&\frac{2}{\pi } \mathcal p \int_0^{\infty } \frac{\epsilon'{\rm im}[\delta f(\epsilon')]}{\epsilon'^2-\epsilon^2 } d\epsilon'\quad { \rm and}\nonumber\\ { \rm im}[\delta f(\epsilon)]&= & -\frac{2\epsilon}{\pi } \mathcal p \int_0^{\infty } \frac{{\rm re}[\delta f(\epsilon')]}{\epsilon'^2-\epsilon^2 } d\epsilon'\ , , \end{aligned}\ ] ] with @xmath135 the principal part @xmath136 of the integral and @xmath137 the photon energy . edge for yba@xmath138cu@xmath139o@xmath140 . the xas spectra exhibit a characteristic polarisation dependence , which is caused by the local symmetries of the different cu sites in the lattice . the right panel shows the resonant structure factors as a function of energy and polarisation , which were deduced from a kramers - kronig analysis . it can be seen that the energy dependent lineshape of @xmath103 depends strongly on the polarisation . ( reprinted from @xcite . copyright 2011 , american physical society.),width=321 ] using these relations the energy dependence of @xmath141 can be calculated , where @xmath125 is the polarisation of the absorbed photon . this is demonstrated in figure[fig : kkanalysis ] , were such an analysis is shown for the high - temperature cuprate superconductor yba@xmath138cu@xmath139o@xmath140 . it is important to realize however , that the above analysis allows one to obtain only matrix elements of the form @xmath142 . the matrix elements @xmath143 with @xmath144 can therefore not directly be determined by x - ray absorption . in principle it is possible , however , to measure the x - ray absorption with 3 different polarisations with respect to the crystal axes and 6 other linear combinations of those . this enables to retrieve all 9 matrix elements of @xmath112 , by solving a set of linear equations . the scattering length can also be related to the index of refraction @xmath3 . we will only briefly discuss the origin of this relation . readers interested in the detailed derivation are referred to the literature@xcite . the relation between @xmath3 and @xmath100 is essentially given by the fact that the transmission of a beam through a plate of some material can be described in two equivalent ways : ( i ) behind the plate there will be an additional phase shift that was suffered by the wave during the passage through the material of the plate . at normal incidence , this phase shift is simply given by @xmath145 , where @xmath146 and @xmath147 are the the wave vector of the light and the thickness of the plate , respectively . ( ii ) the plate can also be considered as a set of scatterers , which cause an additional scattered wave behind the plate . the total wave field behind the plate is therefore a superposition of the incident wave and the scattered wave , where the latter is related to the scattering length @xmath100 of the scatterers within the plate . only identical scattering centers are assumed . since both descriptions must give the same result for the wave field behind the plate one can deduce that @xmath148 where @xmath149 is the number density of the material . since @xmath103 is a complex number , the above eq.[eq : nfrelation ] relates the real and imaginary part of @xmath100 to the dispersion terms @xmath150 and @xmath151 of @xmath152 . + + the general expressions derived in the previous sections are useful to understand the general process of resonant diffraction . however , for the analysis of experimental data it is necessary to evaluate @xmath103 for a specific situation . the simplest case for which explicit expressions can be derived , is the one of a free atom with a magnetic moment . in this situation only the magnetic moment breaks the spherical symmetry of the free atom . for this special case the resonant scattering length in the dipole approximation can be expressed as@xcite @xmath153 where @xmath154 is the unit vector defining the local spin moment and the @xmath155 ( @xmath156 ) are photon energy dependent resonance factors . these factors depend on the products @xmath157 appearing in eq.[eqn : fdipole ] and define the strength of the resonant scattering . there are 3 different terms in eq.[eqn : hannon ] : the first term does not depend on the spin direction and corresponds to non - magnetic resonant scattering . this term simply adds to the thomson scattering in eqs.[eq : f_general ] and [ kheqn ] , which has the same polarisation dependence . the second term is proportional to @xmath158 and is linear in @xmath154 and therefore describes magnetic resonant scattering . by virtue of the optical theorem this part of @xmath103 can be related to the x - ray magnetic circular dichroism in absorption@xcite . the third term is proportional to @xmath159 , i.e. , this term is quadratic in @xmath154 and hence also corresponds to resonant magnetic x - ray scattering . this scattering is related to the linear dichroism in x - ray absorption@xcite . equation[eqn : hannon ] is very often used for the analysis of resonant scattering from magnetically ordered materials@xcite . however , it is important to remember that this expression has been derived for a free atom , where only the magnetic moment breaks the otherwise spherical symmetry . + + whenever an atom is embedded in a crystal , the chemical environment will break the spherical symmetry that was used to derive eq.[eqn : hannon ] . this equation has therefore to be considered as an approximation for describing magnetic scattering in a crystal lattice . for instance , the x - ray absorption of a non - magnetic and non - cubic material will change with the polarisation @xmath10 pointing along the inequivalent lattice directions . the optical theorem therefore immediately implies that @xmath160 also varies with @xmath125 . this polarisation dependence is not captured by eq.[eqn : hannon ] , because in the non - magnetic case , this formula implies @xmath161 , which is isotropic and the same for all @xmath162 . in the following we will provide a simplified description of the effects described by carra and thole@xcite , and by haverkort@xcite . in particular , we will restrict ourselves to the dipole approximation and to magnetic scattering terms that are linear in the spin direction @xmath163 . this is sufficient to illustrate the effects of the local environment , but in general the higher order terms can be relevant as well . for the complete expressions including higher order terms in @xmath164 the reader is referred to ref.@xcite . in the most general case of a magnetic site in a low symmetry environment , the tensor @xmath165 ( cf . [ sec : dipoleapprox ] ) , describing the polarisation and energy dependence of the resonant scattering has the form : @xmath166 with complex energy dependent matrix elements @xmath167 . however , depending on the local point group symmetry of the scatterer , this tensor can be simplified and especially in high symmetries the simplifications are significant : in the extreme case of a free non - magnetic atom with spherical symmetry one has @xmath168 with the identity matrix @xmath169 . if the spherical symmetry is broken only by the spin moment of the scatterer , the first two terms of eq.[eqn : hannon ] translate to @xmath170 it is this expression that is very often used for the analysis of resonant x - ray scattering data . it is important to know in which situations this approximate expression can be used and in which situation it yields wrong results . we therefore discuss two other cases to clarify this point . first , we consider a local environment of @xmath171-symmetry . in this situation @xmath112 can be expressed as @xmath172 comparing eqs.[eqn : fsph ] and [ eqn : foh ] it can be seen , that the spherical approximation may still be used , as long as the magnetic term linear in @xmath164 dominates the resonant scattering signal . we mention that already for @xmath171-symmetry the terms quadratic in @xmath164 differ from eq.[eqn : hannon ] . as a second example we consider a local @xmath173 symmetry . depending on the deviation from @xmath171 , the resonant scattering length can now deviate significantly from the spherical approximation : @xmath174 as can be seen in the above equation , the energy dependence of the magnetic resonant scattering now depends on the spin direction , i.e. , depending on the direction of @xmath164 , @xmath175 and @xmath176 contribute with different weights . this effect is therefore given by the difference between @xmath175 and @xmath176 . it is also important to note that even the non - magnetic scattering in eq.[eqn : fd4h ] is no longer isotropic . this means that for local symmetries lower than cubic the non - magnetic resonant scattering also becomes polarisation dependent , in agreement with the qualitative arguments given in the beginning of this section . we already mentioned that the polarisation dependence of resonant x - ray scattering provides important information about the symmetry of the studied ordering . in this section we will provide a simple example for this , namely the resonant magnetic scattering of a bcc antiferromagnet . in this case the polarisation dependence of the resonant scattering allows one to determine the directions of the ordered moment . + + before we discuss our example , it is useful to introduce an efficient way to describe polarisation dependent resonant scattering . the resonant scattering length as a function of the incoming and outgoing polarisation can be expressed , using a @xmath177 scattering matrix @xmath178@xcite : @xmath179 the matrix elements of @xmath180 therefore correspond to @xmath181- , @xmath41- , @xmath182- and @xmath183-scattering of lattice site @xmath3 . using these @xmath180 , the corresponding structure factor matrix @xmath184 can be calculated according to eq.[eq : kinematic ] : @xmath185 . for the special case described by eq.[eqn : hannon ] , the scattering matrix as a function of the local spin - direction reads@xcite : @xmath186 where the spin - direction @xmath187 is expressed with respect to the @xmath34 and @xmath36 denotes the scattering angle ( cf . figure[fig : exp_scheme ] ) . - and @xmath183-scattering , where the spin directions are @xmath188 and @xmath49 is the rotation angle about the ( 00l)-scattering vector . the lower two panels display the azimuthal dependences of the @xmath41- ( left ) and @xmath183-scattering ( right ) for the spins pointing along @xmath189 . in this example , the polarisation dependence of the resonant scattering enables the spin - directions to be determined . @xmath190 in all cases.,width=321 ] . + + in order to illustrate the polarisation dependence of resonant magnetic x - ray scattering , we consider the example of a bcc antiferromagnet . the unit cell contains two inequivalent lattice sites : one at @xmath191 with @xmath192 and one at @xmath193 with @xmath194 . for the sake of simplicity we assume that the two lattice sites differ only in the local spin moment and are identical otherwise . note that the above coordinates refer to the so - called crystal frame , which is defined by the crystal axes . we will consider the ( 001)-reflection for which @xmath195 . since the system is cubic and only the terms linear in @xmath196 contribute to resonant @xmath197 , we can use the expression given in eq.[eqn : foh ] which is proportional to @xmath198 , i.e. , the @xmath199 are simply given by the term proportional to @xmath200 in eq.[eqn : soh ] . this then yields : @xmath201 in this equation @xmath49 is the azimuthal angle and describes a rotation of the sample about @xmath202 as indicated in figure[fig : exp_scheme ] . when the sample is rotated about @xmath202 , the crystal frame and with it the spin directions rotate within the @xmath203-plane . in the current setting the spin directions are along @xmath47 for @xmath204 . the square modulus of the matrix elements of @xmath184 is proportional to the intensities in the different scattering channels . for instance , the intensities in the @xmath41- and the @xmath183-channels are @xmath205 and @xmath206 , respectively . the calculated azimuth - dependences for @xmath207 and @xmath208 are shown in the top panel figure[fig : azi ] . the azimuthal dependence of @xmath207 is due to the fact that @xmath209 , i.e. , this intensity is maximal if the spin directions are parallel to the scattering plane . similarly , it is easy to see that @xmath210 , which means that @xmath208 maximal for spins aligned perpendicular to the scattering plane . figure[fig : azi ] also shows the azimuthal dependences of @xmath207 and @xmath208 for spins along @xmath211 . it can be seen that the azimuthal dependence changes drastically with the inclination angle @xmath212 between @xmath213 and @xmath154 . in this simple example of a cubic antiferromagnetic system , the polarisation dependence of the resonant scattering provides a powerful means to identify magnetic scattering and to determine spin - directions : ( i ) in a cubic system , the observation of @xmath41-scattering directly implies magnetic scattering . ( ii ) analyzing the azimuthal dependences enables to determine the direction of the spins . however , great care must be taken when dealing with non - cubic systems . in this case , @xmath41-scattering no longer unambiguously identifies magnetic scattering . a very prominent example is the orbitally ordered and magnetically disordered phase of lamno@xmath139 , where the @xmath41-scattering is caused by orbital ordering and not by spin order@xcite . the @xmath41-scattering in this case is related to the polarisation dependence of the first term in eq.[eqn : fd4h ] . in materials with low local symmetries , the analysis of the azimuthal dependences is also more involved and the cubic approximation might lead to wrong results@xcite . one of the most outstanding properties of resonant x - ray diffraction is its greatly enhanced sensitivity to electronic ordering phenomena . this sensitivity is a spectroscopic effect , which is related to the resonance factors @xmath214 of the resonant scattering length . it should be noted , however , that resonant scattering involves transitions into unoccupied states , i.e. , these experiments probe modulations of the unoccupied states close to the fermi level . besides the high sensitivity , the inherent element selectivity is a second advantage of resonant x - ray scattering : by tuning the photon energy to a specific absorption edge , the scattering is usually dominated solely by the scatterers at resonance . in the following we very briefly summarize some of the main spectroscopic effects at the o @xmath215 , the tm @xmath216 and the re @xmath217 edges , which will be relevant for the results presented in section[sec : casestudies ] . a much more complete account of spectroscopic effects in core - level spectroscopies , including the core - levels described below , can be found in @xcite . + + in the tm oxides the o @xmath218 states play an important role in the chemical bonding . it is therefore of great interest to investigate their role for electronic ordering phenomena . by performing a scattering experiment at the o @xmath215 edge , the scattering process involves @xmath219 transitions , i.e. , virtual excitations directly into the @xmath218 valence shell . in this way , the resonant scattering at the o @xmath215 edge becomes very sensitive to spatial modulations of the o @xmath218 states . there is no spin - orbit interaction of the @xmath220 core - hole and the spin - orbit interaction is usually weak in the @xmath218 final state also . resonant scattering at the o @xmath215 edge is therefore typically rather insensitive towards the ordering of spin moments . however , the hybridisation of the o @xmath218 levels with the tm @xmath221 states can induce magnetic sensitivity also here as discussed in section[sec : multiferroics ] . in addition to this , and in contrast to the tm @xmath216 and re @xmath217 edges described below , the @xmath220 core hole does not produce a strong multiplet splitting in the intermediate state . up to date a number of experiments on different materials , including the cuprates and the manganites , have demonstrated a strong resonant enhancement of superlattice reflections at the o @xmath215 edge , revealing the active role of the oxygen @xmath218 states in the electronic order ( cf . section[sec : chargeorder ] ) . nearly all the resonances observed so far are peaked at the low - energy side of the oxygen @xmath215 edge , showing that the probed electronic modulations mainly involve the electronic states close to the fermi level . + + at the transition metal @xmath216 edges the resonant scattering process involves @xmath223 transitions . this makes it possible to study ordering phenomena related to the @xmath221 valence states of the tm . the strong spin - orbit interaction of the @xmath218 core - hole splits the x - ray absorption spectra into the so - called @xmath224 edge at lower energies ( @xmath225 core hole ) and the @xmath226 edge at higher energies ( @xmath227 core hole ) . this is shown for the cu @xmath216-edge in figure[fig : kkanalysis ] . furthermore , the coupling of the @xmath218 core - hole with a partially filled @xmath221-shell in the intermediate state , results in a very pronounced multiplet structure of the @xmath216 edges not only in absorption , but also in resonant scattering , as can be inferred from the optical theorem . the spin - orbit interaction causes a greatly enhanced sensitivity towards spin - order@xcite , which , by means of the optical theorem , is directly related to the magnetic circular and linear dichroism in absorption@xcite . rsxs at the tm @xmath216 edges therefore allows the study of magnetic order of the tm - sites . in addition other electronic degrees of freedom of the @xmath221 electrons , like orbital and charge can be probed at these edges in a very direct way , since the virtual transitions depend directly on the configuration of the @xmath221 shell . + + as will be described in detail in section[sec : magneticstructureofthinfilms ] , rsxs at the re @xmath217 edges plays an important role in determining the magnetic structure of thin films containing res . at the re @xmath217 edge , the resonant scattering process involves @xmath228 transitions . one therefore probes the @xmath229 states of the re elements , which play the most important role for their magnetism . in the @xmath230 final state of xas there is a strong spin - orbit coupling that splits the absorption edge into the @xmath231 edge at lower energies ( @xmath232 core hole ) and the @xmath233 edge at higher energies ( @xmath234 core hole ) . furthermore , the coupling between the core - hole and the valence shell , and the spin - orbit interaction acting on the 4f - electrons , causes a multiplet structure . both the @xmath217 splitting and the multiplet structure are also reflected in rsxs as dictated by the optical theorem . similar to the case of the tm @xmath216 edge , the spin - orbit splitting in the intermediate state of the rsxs process again results a greatly enhanced magnetic sensitivity , which enables the study of the magnetic order in extremely small sample volumes . a particular feature of the re spectra is the small crystal field splitting . hence , the spectral shape at the @xmath217 resonances is largely determined by the atomic multiplet structure independent of the crystal symmetry . effects of the local symmetry , as discussed in section[sec : haverkort ] are therefore not so pronounced as in case of @xmath221 tm , which permits an interpretation of rsxs data without recourse to detailed multiplet calculations@xcite . rsxs experiments require photon energies in the range between 200 ev and 2000 ev and substantially benefit from a control of the incident light polarisation , in particular linear in the scattering plane as well as in the perpendicular direction ( @xmath32 and @xmath33 in figure[fig : exp_scheme ] ) . but also circular polarisation is important , as will be discussed in connection with cycloidal magnetic structures in section[sec : multiferroics ] . these conditions are met by a number of beamlines at third - generation synchrotron radiation sources worldwide . they typically have elliptical undulators as radiation sources , often of apple - ii type . beamlines are designed using grating monochromators and focusing mirrors operated in grazing incidence . an account of the beamline layout at the berlin synchrotron bessy ii is given in@xcite . as pointed out already , soft x - rays are subject to strong absorption by ambient atmosphere . unlike diffractometers for hard x - ray resonant scattering , apparatus for resonant scattering in this energy range must therefore be designed as vacuum instruments , while permitting scattering geometries as shown in figure[fig : exp_scheme ] . for many applications , even ultra - high vacuum ( uhv ) conditions are preferred , because the x - ray absorption length in the materials studied can be as short as 200 in the vicinity of soft x - ray resonances@xcite , rendering the method quite surface sensitive at small scattering angles . therefore , cleanliness of the sample surface can become an issue , and particularly thin - film reflectivities at low temperatures will be modified by growing overlayers due to gas absorption . while resonant diffraction using hard x - rays could make use of conventional diffractometers , instrumentation for soft x - ray diffraction therefore has only recently become available . one of the first soft x - ray diffractometers was developed for studies of thin polymer films and worked in a vacuum of 10@xmath235 torr@xcite . motivated by the high magnetic sensitivity of the soft x - ray resonances , and following pioneering experiments@xcite , the development of vacuum instrumentation for diffraction experiments was initiated . reflectometers developed at synchrotron radiation sources for the characterization of optical elements were modified for the study of magnetic thin - film materials@xcite . as soft x - ray resonances were found to be very useful for the study of complex ordering in correlated electron systems like charge or orbital order@xcite dedicated diffractometers were developed that are now operational at various synchrotron radiation sources@xcite . [ scale=0.8 , trim= 0 0 0 0 , angle= 0 ] hawthorn2011 - 1 for the design of these instruments , two approaches are employed . in order to maintain the full flexibility in terms of scattering geometries and sample rotation , diffractometers as known from hard x - ray applications were built from uhv - compatible materials and mounted in a uhv chamber . this requires vacuum - compatible motors and sophisticated mechanical design but maintains the full functionality of a conventional diffractometer . the first instrument of this type was commissioned at the national synchrotron light source@xcite . recently , a new diffractometer with a horizontal scattering geometry was launched at beamline 10id-2 of the canadian light source@xcite . the instrument is shown in figure[hawthorn2011 - 1 ] and represents a 4-circle diffractometer , with sample ( @xmath236 ) and detector ( @xmath237 ) rotation as well as sample tilt ( @xmath238 ) and azimuthal rotation ( @xmath239 ) . sample manipulation is complemented by @xmath240 , @xmath241 , and @xmath242 stages . situated inside a vacuum chamber , it operates at a base pressure of @xmath243 torr . samples can be cooled with a closed - cycle he cryostat connected to the sample holder by flexible cu braids . avoiding a rigid mechanical connection to the cryostat has the advantage that temperature - dependent studies can be carried out without substantial movement of the sample due to the thermal contraction of the sample holder . sample position was reported to be stable within 100@xmath244m@xcite . the base temperature achieved in this instrument is 18 k. the @xmath237 rotation carries various photon detectors , such as a axuv 100 silicon photodiode , which works very reliably over a large dynamic range of many orders of magnitude , which is particularly required for reflectivity measurements . for the detection of weak signals a channeltron typically coated with a low workfunction material such as kbr can be used . the instrument also has a multi channel plate ( mcp ) with resistive anode readout for the two - dimensional detection of scattering signals . [ scale=1 , trim= 0 0 0 0 , angle= 0 ] beale2010a-1 other designs at least partially avoid in - vacuum motors and complex mechanics and have the drives at atmosphere . figure[beale2010a-1 ] shows the experimental station rasor operated with a vertical scattering geometry at beamline i10 of the diamond light source . here sample and detector rotation are driven by motors and gears outside the vacuum using a double differentially pumped rotary feed - through . also sample tilt as well as the translational alignment @xmath245 are actuated from outside the vacuum , while sample azimuth involves an in - vacuum drive@xcite . a new in - vacuum spectrometer has been installed on beamline x1a2 at the nsls , brookhaven national laboratory . this spectrometer employs a 6-circle geometry with both horizontal and vertical detector rotations , enabling the detector to be placed anywhere within the hemisphere of the vacuum vessel . this detector configuration dispenses with the need for a @xmath238 rotation of the sample and greatly simplifies the sample rotation hardware enabling a much shorter cryostat to aid in stability and the future use of more complicated sample environments . unique to this instrument is the capability of performing both surface soft x - ray scattering measurements and , through the use of zone plate optics , soft x - ray nano - diffraction by raster scanning the beam over the sample to measure a real space image of the scattering@xcite . [ scale=0.5 , trim= 0 0 0 0 , angle= 0 ] bessydiffractometer-1 other designs of soft x - ray diffractometers do not use a conventional four - circle geometry and in this way benefit from avoiding in - vacuum motors and mechanics . such an instrument is operated at beamline ue46-pgm1 of the berlin synchrotron bessy ii ( see figure[bessydiffractometer-1 ] ) . the basis of the diffractometer is a large differentially pumped two - circle rotary feedthrough driven outside the vacuum by two modified huber rotation stages . working in horizontal scattering geometry , the instrument combines very high mechanical precision with the capability of carrying high loads , particularly on the detector circle . the latter consists of a large disc that comprises several uhv flanges up to a bore size of 100 mm diameter . detectors of various degrees of complexity can be assembled on dn 100 cf flanges and can then be easily transferred and removed from the diffractometer ( see figure[bessydiffractometer-2 ] ) . presently , the instrument is equipped with a conventional axuv 100 si photodiode , a slit carousel allows in - situ adaption of the resolution . the whole detector is mounted on a linear feed - through , which resembles an @xmath246 rotation of the detector and readily compensates for a missing @xmath150 tilt of the sample . the instrument can operate under uhv conditions . pressures in the low @xmath247 mbar range are reached even without baking the chamber by operating the sample cryostat together with a liquid nitrogen cryogenic trap for a few days . various sample holders can be introduced via a cf flange with a bore of 90 mm diameter ( see figure[bessydiffractometer-2 ] ) . for standard experiments , samples are mounted on a janis he flow cryostat . here , base temperatures of @xmath248k are easily achieved and with proper shielding 3 k is reliably obtained . it is to be noted , however , that in this temperature region , heating by the x - ray beam is non - negligible and must be taken into account , e.g. , by adequately reducing the photon flux . [ scale=0.5 , trim= 0 0 0 0 , angle= 0 ] bessydiffractometer-2 in rsxs experiments , the requirements for sample environment do not only include vacuum conditions and low temperatures . an important parameter for the study of magnetic materials is an external magnetic field . while fields of the order of 100 gauss can be obtained by small coils in the vicinity of the sample , larger fields require substantial effort . several apparatus were developed in recent years that are equipped with normal - conducting coil systems , achieving magnetic fields of @xmath249 tesla . the alice instrument operated at bessy ii has water - cooled coils outside the vacuum , and the magnetic field is guided into the chamber by an iron yoke@xcite . fields at the sample position can reach 0.8 tesla . the resoxs instrument now operated at soleil has a four - pole magnet arrangement that can be rotated and allows high flexibility in the direction of the field with respect to the sample@xcite . here , @xmath250 tesla are achieved . a recent reflectometer for magnetic measurements developed at the max - planck - institut stuttgart provides a magnetic field of 0.65 tesla@xcite . dedicated soft x - ray diffractometers equipped with superconducting coils that provide magnetic fields considerably larger than one tesla are presently not available . however , a superconducting magnet that is rotatable inside a vacuum chamber and allows for selected scattering geometries with magnetic fields up to 7tesla is operated at beamline ue46-pgm1 of bessy ii . while sample environment is one challenge of soft x - ray diffraction experiments , x - ray detection is now also a focus of instrumental development . two - dimensional detectors can enhance the efficiency of measurements substantially and some of the instruments are already equipped with vacuum - compatible x - ray ccd cameras that are mounted on the @xmath251-arm of the diffractometer and can be rotated inside the vacuum . instruments with that option are operated at the swiss light source@xcite and the national synchrotron light source@xcite . in order to obtain the full information that can be obtained from a soft x - ray diffraction experiment , polarisation analysis of the scattered x - ray beam is also desirable . the method is routinely used in hard x - ray non - resonant and resonant diffraction@xcite and uses single crystals with a suitable lattice spacing as analyzers ( cf . figure[fig : exp_scheme ] ) . with a bragg diffraction angle corresponding to the brewster angle ( @xmath252 ) at the given photon energy , x - rays with @xmath33 linear polarisation are ( ideally ) completely suppressed . rotating the analyzer crystal around the x - ray beam scattered from the sample allows to characterize the polarisation of the scattered x - rays . [ scale=1 , trim= 0 0 0 0 , angle= 0 ] staub2008 - 1 using the same method with soft x - rays usually requires artificial multilayer structures with suitable lattice spacings . the technique was initially applied in polarimeters to characterize the polarisation state of synchrotron radiation@xcite . recently , it was also integrated in soft x - ray diffractometers . the first instrument to be equipped with a polarisation analysis was reported by staub et al.@xcite . polarisation analysis is performed by an artificial w / c multilayer with graded layer spacings . by linear translation the layer spacing can be chosen to meet the bragg condition at the brewster angle for a given resonance energy . the efficiency of such a multilayer at the first - order diffraction peak is typically of the order of a few percent , as can be inferred from the reflectivity curve shown in figure[staub2008 - 1 ] . despite this substantial loss in intensity , polarisation analysis in soft x - ray diffraction is feasible and can be employed to study complex phenomena such as orbital currents@xcite . recently , an advanced polarisation analysis has become available with the previously mentioned rasor instrument@xcite . here , even a full rotation of the analyzer multilayer around the scattered x - ray beam is possible ( see figure[beale2010a-2 ] ) . [ scale=1 , trim= 0 0 0 0 , angle= 0 ] beale2010a-2 following the prediction of very strong magnetic scattering cross sections at the lanthanide @xmath217 and 3@xmath253-transition metal @xmath216 absorption thresholds @xcite , historically , rsxs was first employed to study magnetic ordering in thin metallic films and multilayers . the first demonstration of the strength of magnetic rsxs has been reported by kao et al . , who observed changes in the specular reflectivity of a ferromagnetic 35 - thin single - crystalline fe film upon magnetization reversal@xcite . exploiting linear polarised x - rays , they found asymmetry ratios @xmath254 as large as 13% at the fe @xmath216 resonance and even stronger effects have been reported for scattering of circular polarised x - rays from a 37thin co film with asymmetry ratios up to 80% close to the co @xmath216 resonance as shown in figure[p_kao94 ] @xcite . here @xmath255 and @xmath256 are the measured specular reflectivity for the two opposite in - plane magnetization directions of this experiment . the asymmetry ratios strongly depend on the scattering angle , which represents additional structural information on the magnetic order . reflectivity of a 37 thin co film as a function of grazing angle measured at the maximum of the co @xmath257 edge . the solid and dashed lines denote the two opposite magnetization directions , the dots represent the resulting asymmetry ratio , which shows a strong angular dependence and amounts up to 80% . ( reprinted with permission from @xcite . copyright 1994 , american physical society.),width=264 ] these two pioneering experiments revealed the huge potential of rsxs combining the spectroscopic power of x - ray magnetic circular dichroism ( xmcd ) with structural information . even without exploiting the latter aspect , a large number of studies used the strong dichroic effects of the scattered intensities to extract information usually obtained by x - ray absorption techniques @xcite , like monitoring element - specific magnetic hysteresis loops by measuring the field dependence of the circular dichroism of the reflected intensity for fixed photon energies and angular positions . such studies avoid the experimental difficulties of xmcd performed in total electron yield ( tey ) or fluorescence yield ( fy ) mode , i.e. , a limited probing depth and the influence of external fields and charging in the tey mode or weak signals connected with the fy mode . while these first experiments focused on thin single films , tonnerre et al . presented the first demonstration of magnetic rsxs from a multilayer system consisting of a repetition of ag(11)/ni(17.5 ) double layers with antiferromagnetic ( afm ) or ferromagnetic ( fm ) coupling of the ni layers depending on external magnetic field . from this sample either pure half - order magnetic bragg reflections ( afm coupling ) or huge magnetic circular dichroism on the corresponding chemical superstructure reflections ( fm coupling ) could be observed . by analyzing the magnetic scattering intensities , a magnetic scattering strength at the ni @xmath216 edge of the order of ten times the electron radius @xmath258 was deduced ( see figure[p_tonnerre95 ] ) , i.e. , a value very close to the non - resonant charge scattering strength @xcite . imaginary part of magnetic rsxs at the ni @xmath216 edge ( open circles ) obtained from a ni / ag multilayer compared to ni magnetic circular dichroism data ( solid line ) in electron units . ( reprinted with permission from @xcite . copyright 1995 , american physical society . ) ] the strong enhancement of the scattering cross section at resonance is intrinsically accompanied by a huge change of the optical constants @xmath152 ( see section[sec : rsxsandxas ] ) , with @xmath151 and @xmath150 being of the order of @xmath259 . as a consequence , dispersion and absorption effects can no longer be neglected but often have to be included in a quantitative analysis of rsxs data . consequently , a large number of studies have been dedicated to a quantitative characterization of soft x - ray resonances , where the optical parameters are typically not well known and the tabulated values @xcite can not be applied . experimentally , the easiest access to the optical constants is by measuring the absorption coefficient by transmission @xcite , tey , or fy methods and subsequent calculation of the real part of the index of refraction by applying the kramers - kronig relation ( cf . section[sec : rsxsandxas ] ) . however , the accuracy of the individual methods of measuring the absorption coefficient are often limited by various experimental uncertainties such that it is advantageous to acquire additional information independently as has been done by measuring the faraday rotation of linear polarised light passing through an fe / cr multilayer @xcite or the energy - dependent bragg - peak displacement of fe / v superlattice reflection @xcite . in particular , it has been shown that the analysis of polarisation- and energy - dependent position , width and intensity of a superstructure reflection can be used to determine the full set of optical parameters at resonance @xcite . while these first studies focused on the @xmath221 tm @xmath216 resonances , detailed characterization of the lanthanide @xmath217 resonances has been performed later revealing even stronger resonance effects @xcite . figure [ p_peters2004 ] shows the resonant scattering length through the gd @xmath217 resonance calculated on the basis of polarisation dependent absorption obtained in transmission through a thin gd@xmath260fe@xmath261 film . here , @xmath262 represents the resonant charge scattering contribution of the order of 500@xmath263 , which is almost 10 times stronger than off - resonant charge scattering ( cf . eq.[eqn : hannon ] in which the photon energy dependent resonant factors are called @xmath264 , i.e. , @xmath265=@xmath264 ) . even more impressive , the circular dichroic scattering contribution @xmath266 ( @xmath200 in eq.[eqn : hannon ] ) is as large as 200@xmath263 , i.e. , a magnetic scattering strength 3 times stronger than off - resonant charge scattering . resonant amplitudes at the gd @xmath217 edges . shown are the complex charge @xmath262 , circular magnetic @xmath267 , and linear magnetic @xmath268 , atomic scattering factors as function of energy in units of @xmath263 . top : imaginary parts , bottom : real parts . right axis : approximate atomic cross sections in @xmath269 using a fixed wavelength for e=1200 ev . dash - dotted line : high energy limit of the atomic scattering amplitude z=64 . ( reprinted with permission from @xcite . copyright 2004 , american physical society . ) ] with this extraordinarily high magnetic sensitivity , the study of magnetic structures of thin films down to only a few magnetic layers became feasible . + + the power of magnetic rsxs to study afm bragg reflections from thin single films of only a few magnetic layers has been demonstrated for thin holmium metal films @xcite . metallic holmium possesses the largest magnetic moment of all elements coupled by oscillatory long - range rkky interactions . below the bulk ordering temperature of 132k , ho develops ferromagnetic order within the closed - packed planes of the hexagonal closed packed ( hcp ) structure but with a certain angle between the moments of neighbouring planes resulting in a long - period afm structure along the [ 001 ] direction with an incommensurate temperature - dependent period of about 10 monolayers at 40k . in a magnetic diffraction experiment , this magnetic structure causes magnetic superstructure reflections well separated from the crystallographic reflections ( 0 0 2l @xmath270 ) with @xmath271 at 40k as observed from ho single crystals by neutron scattering@xcite . the potential of magnetic off - resonant and magnetic hard x - ray resonant scattering has been demonstrated in a series of pioneering magnetic structure studies on ho by gibbs et al.@xcite . in rsxs experiments at the ho @xmath272 soft x - ray resonance , due to the long period of the magnetic structure , the @xmath273 magnetic reflection of ho is well within the ewald sphere . exploiting the huge magnetic scattering strength of rsxs , this magnetic structure could be readily studied in thin films down to the thickness of a single magnetic period @xcite as can been seen from the intense and well developed magnetic superstructure reflections obtained from a 11 monolayer thin ho metal film shown in figure[p_weschke2004 ] . top : ( a ) magnetic reflection of an 11 ml - thin ho film at various temperatures , recorded at the ho @xmath272 resonance . ( b ) square root of the integrated intensity ( solid squares ) and half widths of the rocking curves of the magnetic reflection ( open squares ) . the solid lines represent the result of a simultaneous power - law fit analysis to determine @xmath274 . bottom : @xmath275 as a function of ho film thickness @xmath253 , including data from mbe films ( solid circles ) and films grown in situ on w(110 ) ( open circles).the inset shows a comparison to the behaviour of ferromagnet gd . ( reprinted from @xcite . copyright 2004 , american physical society . ) ] while for rsxs the relation between magnetic ordering parameters and integrated diffraction peak intensity can be more involved , the integrated superstucture peak intensity of ho has been shown to be in good approximation a measure of the helical afm order parameter@xcite and thus allows to determine the ordering temperatures for various film thicknesses @xmath253 . for small @xmath253 , a finite - size scaling of @xmath275 would have been anticipated and was indeed observed . however , this study found a much faster decrease of @xmath275 with @xmath253 ( see figure[p_weschke2004 ] ) than known from simple antiferromagnets and ferromagnets @xcite . the observed modified finite - size effect in the ordering temperature is characterized by an offset thickness @xmath276 representing a minimum sample thickness below which no helical order can be established . this finding has been related to the long period of the afm structure@xcite , in line with results obtained for the long - period spin - density wave of cr in cr / fe superlattices @xcite . as shown in figure[p_weschke2004 ] , even above @xmath275 , finite magnetic intensity could be observed characterized by a diverging peak width with increasing temperature . these remaining broad intensity distributions are caused by short - range magnetic correlations persisting above @xmath275 , with the peak width being an inverse measure of the magnetic correlation length . this observation shows the huge potential of rsxs for studying magnetic short - range correlations in films of only a few monolayer thickness , even with the perspective to employ coherent scattering as discussed later . the helical magnetic structure of bulk ho was already well established by neutron scattering . for some materials , however , no large single crystals can be synthesized , which limits the capability of neutron scattering for magnetic structure determination . in such cases , rsxs can provide detailed information on the magnetic moment directions exploiting the polarisation dependence of the scattered intensity as discussed in section[sec : pol_dep ] . here , the example of thin epitaxial ndnio@xmath277 ( nno ) films @xcite is presented . in this work the element specificity of rsxs to study the magnetic structure of ni and nd moments in nno separately is exploited . nno shows a metal - to - insulator transition at about 210k . at the same temperature afm ordering occurs . from the observed ( 0.5 0 0.5 ) magnetic reflection , initially , an unusual collinear up - up - down - down magnetic structure connected with orbital ordering of the ni@xmath278 @xmath279 electrons has been proposed on the basis of neutron powder diffraction data @xcite . these results have been challenged by the later observation of charge disproportionation without any indication for orbital order using hard x - ray diffraction @xcite . however , off - resonant and resonant scattering at the ni @xmath215 edge can suffer from weak signals and only indirect sensitivity to orbital order . therefore the ( 0.5 0 0.5 ) reflection has been studied by rsxs in the vicinity of the ni @xmath216 and nd @xmath280 resonance . figure [ p_scagnoli2008 ] shows the observed azimuthal dependence of the scattered intensity compared to model calculations based on the formalism developed by hill and mcmorrow@xcite . analogous to the case of the simple bcc antiferromagnet ( cf . figure[fig : azi ] ) , either a collinear up - up - down - down or a non - collinear magnetic structure was assumed . azimuthal angle dependence of the ( 0.5 0 0.5 ) magnetic reflection of ndnio@xmath139 measured at the ni @xmath224 and nd @xmath280 edge . the solid and dotted lines correspond to model calculations assuming a non - collinear and a collinear up - up - down - down magnetic structure , respectively . ( reprinted with permission from @xcite . copyright 2008 , american physical society . ) ] the results clearly reveal a non - collinear magnetic structure of the ni and nd moments at low temperatures . in addition , the observed polarisation dependence of scattered intensity , including polarisation analysis of the scattered light at the ni @xmath216 resonance , has been found to be inconsistent with orbital order and the energy - dependent line shape of the reflection could only be reproduced assuming charge order . thus these studies fully support a picture of combined charge and non - collinear magnetic order of the ni sites but are inconsistent with orbital order suggesting that the metal - insulator transition is driven by charge disproportionation in nno . at this point , it is appropriate to stress again a caveat : the model calculations for nno used eq.[eqn : hannon ] , which strictly applies to the case , in which only the spin breaks the otherwise spherical local symmetry . this approach is commonly used and yields a reasonable data interpretation in many cases . nonetheless , it has to be kept in mind that this is an approximation which might not always be valid in a real material ( cf . section[sec : haverkort ] ) . in particular , for systems with symmetry lower than cubic , the standard treatment can yield completely wrong results . this is demonstrated in a study on nacu@xmath138o@xmath138 by leininger et al . nacu@xmath138o@xmath138 belongs to the class of edge - sharing copper - oxide chain compounds which have attracted considerable interest in the past years due to their diversity of ground states . the related compound @xmath281 , e.g. has been found to exhibit ferroelectricity . at present it is not clear whether this ferroelectricity is caused by the complex magnetic structure or the strong tendency of li and cu intersubstitution . a non - collinear spiral antiferromagnetic structure was derived by neutron diffraction on polycrystalline samples and nmr measurements on nacu@xmath138o@xmath138@xcite in line with results from susceptibility measurements @xcite . serving as an isostructural and isoelectronic reference compound without complicated intersubstitution of alkaline and @xmath221 tm ions@xcite , nacu@xmath138o@xmath138 is only available in form of tiny single crystals . therefore , rsxs is the method of choice to study details of the magnetic structure . however , according to the simple treatment of magnetic rsxs data , the azimuthal angle dependence of the magnetic reflection suggested a magnetic structure made of only one magnetic moment direction in stark contrast to the results of the susceptibility measurements and neutron powder diffraction data . this inconsistency could be explained on the basis of the strong anisotropy observed in the polarisation dependent absorption probability . the anisotropy is caused by a selection rule precluding excitations into the partially occupied planar @xmath282 orbitals of the @xmath283 ions which modifies the polarisation dependent magnetic scattering strength according to the kramers - heisenberg formula and hence masks the azimuthal angle dependence of the magnetic signal . these findings are in excellent agreement with the theoretical predictions by haverkort et al.@xcite , which shows that in general the local orbital symmetry has to be taken into account for magnetic structure determination by magnetic rsxs . the first application of rsxs to study long - period afm structures in bulk materials has been reported by wilkins et al.@xcite showing the high potential of rsxs to study magnetic structures and , even more importantly , other electronic ordering phenomena in bulk oxides . for magnetic structure determination , however , neutron scattering will stay an unrivaled method . on the other hand , despite the limited ewald sphere , the potential of magnetic rsxs for exploring magnetic structure has been clearly shown and its importance will increase with growing interest in thin transition metal oxide films and superstructures that can be obtained with very high quality . films of @xmath284mno@xmath139 perovskites represent a prominent example , where long period complex magnetic order is strongly coupled to ferroelectricity . such a thin improper multiferroic has been indeed studied for the first time by rsxs very recently by wadati et al.@xcite . diffraction scans with momentum transfer perpendicular to the magnetic stripes in fepd films for left ( dotted ) and right ( solid ) circularly polarised x - rays and resulting difference signal below . insets : right : asymmetry ratios for the first and second order magnetic reflections . left : magnetic force microscope image of stripe domains and schematic magnetization profile.(reprinted with permission from @xcite . copyright 1999 , american association for the advance of science . ) ] + + while rsxs is particularly useful in case of long - period magnetic structures , rsxs can also provide detailed information on the microscopic magnetic properties of ferromagnets and simple antiferromagnets , in particular when applied to artificial structures that produce an additional periodicity of the order of the wavelength of soft x - rays . such structures are also naturally obtained by the formation of domains in chemically homogeneous materials , as has been observed for thin films of fepd . ferromagnetic fepd films are characterized by perpendicular magnetic anisotropy leading to the formation of well ordered stripe domains as shown in figure[p_durr99 ] . in order to reduce the stray field outside the sample , closure domains can be formed at the surface , which link the magnetizations of the neighbouring stripes , resulting in an overall domain structure of chiral nature . x - ray scattering experiments usually are not sensitive to the phase of the scattered wave and , hence , can not easily distinguish left and right handed magnetic structures . however , for circular polarised incident light , the resonant magnetic structure factor differs for structures of different sense of rotation @xcite giving rise to circular dichroism in the scattered intensity not connected with absorption effects but depending on the sense of rotation of a magnetic structure . this property has been exploited by drr et al.@xcite to study closure domains in thin fepd films by rsxs . here , the regular domain pattern at the surface does not cause superstructure reflections in terms of a single peak but rod - like off - specular magnetic scattering intensity with a maximum intensity at a transverse momentum transfer determined by the inverse of the mean distance of the stripe domains as shown in figure [ p_durr99 ] . the intensity of these off - specular magnetic rods showed pronounced circular dichroism , in this way proving the existence of the above mentioned closure domains . by modeling the q - dependent magnetic rod intensity , shown in figure[p_dudzik00 ] , the depth of the closure domains could be estimated to be about 8.5 nm in a 42 nm thick film @xcite . rod scan of the magnetic satellite peak from stripe domains in fepd thin films together with model calculations showing the influence of closure domain depth : solid line : no closure domains , dashed line : closure depth 110 , dotted line : closure depth 90 .(reprinted with permission from @xcite . copyright 2000 , american physical society . ) ] as shown in the latter example , rsxs is capable of studying ordering phenomena with spatial resolution . the information on spatially varying ordering phenomena is contained in the dependence of scattered intensity on the chosen incident energy and momentum transfer which can be very sensitive even to tiny modifications of order . hence , rsxs has been applied very successfully to obtain magnetization depth profiles of thin films and multilayers with very high depth resolution . in the past years , three different ways have been introduced to extract depth dependent information on magnetic ordering , exploiting 1 . the change of the sample volume probed through the resonance due to strong incident energy - dependence of the photon penetration depth @xcite , 2 . the sensitivity of the shape of a magnetic reflection from very thin samples on the spatial varying magnetic ordering parameter @xcite , and 3 . the incident energy- and momentum - transfer dependent reflectivity @xcite . top : ( a ) and ( b ) magnetic diffraction data characterizing the fm / helical afm phase transition of 180ml dy / w(110 ) upon cooling down ( open symbols ) and warming up ( filled symbols ) for two different photon energies ; ( c ) widths of magnetic peaks in the directions parallel and perpendicular to the film plane recorded at 7780 ev ( squares ) and at 1305 ev ( circles ) . bottom : magnetic depth profile of a 180ml thick dy film during the first order phase transition from the fm to the helical afm phase . ( reprinted from @xcite . copyright 2006 , american institute of physics . ) ] while the first two approaches allow to draw qualitative conclusions from the raw data without involved data analysis they , however , require a magnetic reflection inside the ewald sphere at resonance . approach ( i ) has been applied to study depth - dependent afm ordering in thin dy metal films grown on w(110 ) . metallic dy develops a long - period helical afm superstructure below @xmath275=179 k , which undergoes a first order transition into a ferromagnetic structure at about 80 k. this transition is characterized by a pronounced thickness dependence of its hysteretic behaviour , i.e. , the hysteresis smears out with decreasing film thickness . the long period afm structure gives rise to magnetic superstructure reflections well separated from the structural bragg reflections similar to those observed from ho metal described above , which can be easily measured in the vicinity of the dy @xmath217 resonance . through this resonance not only the magnetic scattering strength varies but also the probing depth of the photons due to the energy dependence of the imaginary part of the index of refraction . figure [ p_ott2006 ] shows the temperature dependent intensity of this afm reflection through the afm - fm transition measured for two different photon energies , i.e. , with different probing depths of the photons . distinctly different hysteretic behaviour was observed , which readily indicates a pronounced depth dependence of the growth of the helical afm order with temperature . from modeling these energy - dependent hysteresis loops , taking the energy - dependent absorption into account , a temperature - dependent depth profile through this first order phase transition could be extracted shown in figure[p_ott2006 ] @xcite . obviously , upon heating the helical order starts to grow in the near surface region of the film and develops deeper towards the w interface with temperature . this behaviour has been attributed to pinning of the fm phase at the interface by strain effects . interestingly , also at the top surface , a phase of modified afm order characterized by a tendency towards ferromagnetism survives over a large range of temperature . this way of achieving depth - dependent information requires a film thickness much larger than the minimum photon penetration depth at resonance and can hence not be applied to ultra - thin samples . in this latter case , however , according to approach ( ii ) , the shape of a superstructure reflection measured at a fixed energy already contains the information on the spatial modulation of magnetic order , as has been shown for magnetic scattering from the magnetic semiconducter eute @xcite . below @xmath275=9.8k , eute develops a simple afm structure with ferromagnetically ordered ( 111 ) planes but alternating magnetization along the [ 111 ] direction . by virtue of the strong magnetic eu @xmath217 resonance and the very high quality of the epitaxially grown [ 111 ] eute films , a very well resolved magnetic half order reflection could be observed from a film of only 20 monolayers , with a peak intensity in the order of almost 1% of the incident intensity . this magnetic reflection is already comparable to the off - resonant charge scattering from the superlattice peak of figure[fig : multilayer ] . at resonance , this magnetic reflections appears very close to the brewster angle where the charge scattering background for @xmath33 polarised incident light is almost completely suppressed . this ideal combination of magnetic period length and resonance energy in eute therefore leads to a magnetic signal - to - background ratio of several orders of magnitude , allowing measurements of the magnetic superstructure reflection with unprecedented quality . as can be seen in figure[p_schierle2008 ] , the intensity of this reflection is not contained in a single peak but distributed over several side maxima caused by the finite size of the system ( so - called laue oscillations ) . the period of these oscillations displays a measure of the inverse number of contributing magnetic layers and the overall envelope functions depends sensitively on the magnetization depth profile . by virtue of the high magnetic contrast in this study , the side maxima could be resolved over a large range of momentum transfer revealing distinct changes with temperature : with increasing temperature the period of the laue oscillations increases and the intensity of the side maxima compared to the central peak decreases . this behaviour signals a non - homogeneous decrease of the magnetic order through the film , such that the effective magnetic thickness decreases with increasing temperatures , i.e. , the order decays faster at the film boundaries . ( colour online ) magnetic ( 0.5 0.5 0.5 ) bragg reflections with magnetic laue oscillations of thin eute films with 44 monolayers ( a ) and 20 monolayers ( b ) thickness . ( c ) data for a 20 monolayer thick film , recorded for various temperatures below ( solid ) and above ( dashed ) @xmath275 ; the photon energy was 1127.5 ev , the eu @xmath231 resonance maximum . ( d ) two selected experimental diffraction patterns ( dots ) and results of calculations , taking magnetization depth profiles into account ( lines ) . ( reprinted from @xcite . copyright 2008 , american physical society.),width=7 ] from a kinematical modeling of the observed intensity distributions , temperature - dependent atomic - layer - resolved magnetization profiles across the entire film could be extracted . even at low temperatures , these profiles displayed reduced order at the film interfaces , which is due to some chemical intermixing . in addition , the profiles reveal a faster decrease of the order close to the film boundaries with temperature , i.e. , a layer - dependent temperature dependence of the magnetic order is observed . due to the half filled 4f shell of eu@xmath285 ions , eute exhibits almost pure spin magnetism caused by strongly localised moments and can therefore be regarded as a heisenberg model system , for which this type of surface - modified order was theoretically predicted@xcite . top : charge intensities ( upper panels ) and magnetic asymmetries ( lower panels ) of the first- and second - order bragg peaks recorded at the co @xmath216 absorption edges from a co@xmath286mnge / au multilayer . the dots represent measured data ; the lines are model calculations as described in @xcite . bottom : structural and magnetic depth profiles of co and mn as determined from the model calculations . ( reprinted with permission from @xcite . copyright 2005 , american physical society . ) ] ( colour online ) simulated xmcd spectra for two lcmo layers on strontium - titanate with thicknesses of ( a ) 6 and ( b ) 50 nm . the presence of a 1 nm nonmagnetic surface layer strongly affects the spectral shape irrespective of the film thickness , making the strong sensitivity of xmcd in reflection geometry to nonmagnetic layers evident . ar - xmcd here denotes the asymmetry value @xmath287 , where @xmath255 and @xmath256 refer to the measured intensities for opposite directions of the external magnetic field . ( reprinted with permission from @xcite . copyright 2007 , american institute of physics . ) ] modeling of q - dependent intensity distributions is not restricted to magnetic superstructure reflections but can also be applied in a more general way to the specular reflectivity from systems without any magnetic or chemical reflection inside the ewald sphere at a soft x - ray resonance , e.g. ferromagnetic films , in this way opening the method to a broader field of applications . this has been successfully demonstrated for hard x - ray resonant scattering@xcite . in the soft x - ray range , in particular at resonance , specular reflectivity typically does not fall off so rapidly and hence can be measured over a large range of momentum transfer due to the strong optical contrast at the interfaces and surfaces . information on ferromagnetic depth profiles usually requires to compare reflectivities measured with different polarisations of the incident x - ray beam or for opposite magnetization directions . interpretation of such reflected intensities is often less intuitive and requires detailed theoretical modeling as well as precise knowledge of the energy - dependent optical parameters . based on an magneto - optical matrix algorithm developed by zak et al.@xcite , several computer codes have been developed within the last decade for this purpose . in the pioneering early experiments , modeling of energy - dependent reflectivity data has been exploited to extract averaged magnetic properties , like the fe magnetic moment in fe@xmath261mn@xmath260 thin films , which undergo a magneto - structural transition at x=0.75 @xcite . later studies improved data analysis to incorporate the effects of magnetic and chemical roughness as well as magnetization depth profiles @xcite . examples of interest are ferromagnetic half - metals , like la@xmath260sr@xmath261mno@xmath277 ( lsmo ) close to @xmath240=0.3 and a number of heusler compounds . half - metals are characterized by a band gap for only one spin direction and therefore are candidates to show 100% spin polarisation at the fermi level , thus being of high interest for spintronic devices . functionality of such materials crucially depends on the magnetic properties at interfaces , which can be strongly altered by extrinsic effects like electronic reconstruction , disorder , roughness , or strain , but also by intrinsic changes caused by the broken translational symmetry . therefore , obtaining depth - dependent information on the ferromagnetic properties of such materials at various interfaces is greatly needed but challenging . the interface magnetization depth profiles for co and mn moments in co@xmath138mnge / au multilayers were obtained from the analysis of element - specific rsxs@xcite . by fitting the xmcd spectra measured at the angular positions of the first three superlattice reflections , the individual magnetization profiles shown in figure[p_grabis2005 ] for the co and mn moments through the magnetic layers could be reconstructed . they are characterized by non - magnetic interface regions of different extension for the upper and lower boundaries and also differ for the two magnetic species leading to a rather complicated behaviour of the magnetization through a magnetic layer . this observation has been attributed to structural disorder caused by strain at the interfaces which affects co and mn spin in a different way since co on a regular mn position keeps its ferromagnetic spin orientation and full moment but mn on a co position has an antiparallel spin orientation and a reduced moment @xcite . with a similar scope , rsxs was applied to thin manganite films . manganites show one of the richest set of phase diagrams among the transition metal oxides and therefore offer a high potential for application in future functional heterostructures . freeland et al.@xcite showed the existence of a thin non - ferromagnetic and insulating surface layer with a thickness of only one mn - o bilayer on a single crystal of layered lsmo by analyzing its energy - dependent reflectivity at the mn @xmath216 resonance combined with xmcd and tunneling probe techniques@xcite . this finding renders the material a natural magnetic tunnel junction . the sensitivity of rsxs to weak magnetic modifications like a thin non - magnetic layer in manganites is nicely demonstrated in a study of la@xmath260ca@xmath261mno@xmath277 ( lcmo ) films for different thicknesses and substrates by valencia et al.@xcite . by analyzing the xmcd spectra , measured on the specular reflectivity for fixed scattering angles , detailed knowledge of the chemical and magnetic surface properties could be obtained . in particular , all samples studied developed a thin surface region ranging from 0.5 to 2 nm with depressed magnetic properties . figure [ p_valencia2007 ] shows simulations of the xmcd spectra depending on the presence of a magnetic dead layer , revealing distinctly different energy profiles irrespective of the film thickness . in the same way , verna et al . very recently reconstructed the magnetization depth profile with high spatial resolution from thin lsmo samples , again , observing a non - ferromagnetic surface region of about 1.5 nm thickness which expands with increasing temperature @xcite . the observation of such non - ferromagnetic dead layers at manganite surfaces for lcmo as well as lsmo single crystals and films of different thickness , independent of the particular epitaxial strain strongly suggests that the observed modified surface magnetization is an intrinsic property of manganite surfaces @xcite . this conclusion is further supported by an xas study exploiting linear dichroism at mn @xmath216 resonance from thin lsmo films which observed an orbital reconstruction , i.e. , a rearrangement of the orbital occupation at the surface consistent with a suppression of the double - exchange mechanism responsible for the ferromagnetic properties @xcite . the latter examples of modified , mostly suppressed magnetic properties at surfaces and interfaces , are important results in view of possible application of materials in heterostructures . however , the proximity of two materials in heterostructures can cause much stronger and more complex changes of magnetic properties at those buried interfaces , which has been studied by rsxs intensively in the last decade . interfaces are presently of high interest since here direct control of material properties is possible by several different mechanisms . in particular , the proximity of two different materials at an interface can generate phenomena not present in the individual materials . while this holds in general for a rich variety of electronic phenomena , it is already true for two adjacent magnetic systems . + + transverse scans normalized to the maximum of the specular reflectivity from a 7 nm co film capped with al recorded at the co @xmath224 resonance . diamonds : average signal from two measurements with opposite magnetization , representing the correlation of the chemical structure . circles : difference signal of the two measurements with opposite magnetization , representing the magnetic contribution . the faster decay with @xmath288 in the latter case yields smaller magnetic roughness . ( reprinted with permission from @xcite . copyright 1996 , american physical society . ) ] ( colour online ) . profile of the optical parameters @xmath150 and @xmath151 across the interface of the mnpd / fe exchange bias system , separated into structural non - magnetic ( red , black ) and various magnetic contributions . rotatable mn moments lead to the dashed green ( @xmath150 ) and dashed blue ( @xmath151 ) curve . dashed - dotted magenta and light - blue curves : same for the pinned mn moments . ( reprinted with permission from @xcite . copyright 2008 , american physical society . ) ] famous examples of interface - generated magnetic phenomena with high technological relevance are giant magneto resistance and exchange bias . the experimental challenge of characterizing such phenomena is due to the extremely small amount of contributing material , the presence of more than one type of magnetic moment , and the buried character of the region of interest . also here , rsxs has been shown to be a very sensitive tool to study such interfaces since its huge charge and magnetic scattering strength is accompanied by a moderate penetration depth , element specificity and access to structural information . one crucial parameter that can influence interface - generated functionality is chemical and/or magnetic roughness , which for a buried interface can be characterized best by scattering techniques measuring the related off - specular scattered intensity distributions . such diffuse scattering contributions , however , are typically several orders of intensity weaker than the specular reflectivity and require an extraordinary magnetic scattering strength to be detectable . the first demonstration of how diffuse rsxs intensities can be exploited to track magnetic roughness was reported by mckay et al . for co films and co / cu / co trilayers @xcite . in this study the diffuse intensity around the specular reflectivity was measured at the co @xmath224 resonance containing contributions from magnetic and chemical roughness . in order to separate the two quantities , the intensity was measured with circular polarised x - rays for two opposite magnetizations of the sample . in this case the sum of the two measurements represents the chemical roughness while the difference was interpreted as a measure of the magnetic roughness . from the intensity and width of such broad intensity distributions , properties like the root mean square roughness ( intensity ) and correlation lengths ( width ) can be extracted independently for the magnetic and the chemical interface . interestingly , the magnetic signal drops much faster with increasing momentum transfer @xmath288 than the chemical intensity as can be seen in figure[p_mckay96 ] . this observation yields a surprising result , namely a much smoother magnetic interface if compared with the chemical roughness . it was explained by a weaker coupling of the co moments in the intermixing region compared to that of the homogeneous part of the film . this means that the moments connected with chemical roughness do not contribute to the signal from the film magnetization and the magnetic interface therefore appears smoother than the chemical one . similar results have been obtained by freeland at al . who studied systematically the relation between magnetic and chemical interfacial roughness by monitoring the diffuse scattering from a series of cofe alloy films with varying chemical roughness@xcite . here , it was also shown that the diffuse intensity can be used to measure magnetic hysteresis loops of bulk and interfacial magnetization independently @xcite . it has been shown later , that the difference signal contains charge - magnetic interference contributions , such that it does not represent the pure magnetic roughness @xcite . the general result , however , that the magnetic and chemical roughness can be very different , is not affected . it was rather supported later by a study of hase et al . , who separated chemical and magnetic roughness in a different way by comparing the diffuse scattering in the vicinities of either a chemical or half - order pure magnetic reflection from an antiferromagnetically coupled [ co / cu]@xmath289 multilayer @xcite . the observation of different roughnesses of the chemical and magnetic interfaces shows that the understanding of macroscopic magnetic phenomena and their relation to roughness can not be simply obtained from the knowledge of the chemical roughness alone . nevertheless , grabis et al . observed almost identical chemical and magnetic roughness parameters in a co@xmath286mnge / au multilayer , showing that different roughness are not a general feature of magnetic interfaces @xcite . theory describing diffuse resonant scattering from chemically and magnetically rough interfaces beyond the kinematical approximation has been published by lee et al.@xcite . the proximity of two different magnetic systems can cause more complex changes of the magnetic properties than just introducing disorder , as discussed in the previous paragraphs . a prominent example , connected with functionality already applied in technological applications is the exchange bias effect . exchange bias is the observation of a shift of the hysteresis along the field axis of a fm material in contact with an antiferromagnet ( af ) after field - cooling through the nel temperature . the general mechanism was identified to be the exchange interaction between the spins of the fm and the af at the interface , which generates an additional unidirectional anisotropy for the fm spins @xcite . however , the original models assuming perfect interfaces failed to describe exchange bias and related phenomena in a quantitative way . rather , the exchange bias strongly depends on details of the magnetic structure at the interface . exploiting xmcd techniques , rich knowledge on the magnetic behaviour of the interface was obtained . in particular the presence of a fm component within the af including a pinned fraction that is related to the exchange bias was observed @xcite . due to the fact that interfaces are buried , techniques not limited to the top sample surface region are of advantageous . therefore , a large number of experiments exploited the xmcd on the specular reflectivity to characterize the element specific behaviour of spins in exchange bias systems @xcite . such studies revealed the complex magnetic situation at the interface connected with exchange bias . in particular , several different types of spins are involved , namely the fm spins , the compensated afm moments as well as uncompensated afm moments that can be divided into a pinned and a rotatable fraction . by exploiting the structural information contained in reflectivity data , detailed characterization of the depth dependence of the different magnetic contributions at the interface could be achieved . by varying the absorption threshold , magnetization direction , photon polarisation and scattering geometry , rsxs can be tuned to be sensitive to almost all the magnetic ingredients of the magnetic interface . in this way roy et al . determined the depth profile of fm and unpinned uncompensated afm spins of the exchange bias system co / fef@xmath286 by analyzing the specular reflectivities measured with constant helicity of the incident photon beam for opposite magnetization directions @xcite . they found the majority of the uncompensated unpinned afm spins in a region of a few nanometers below the interface , characterized by afm coupling to the spins of the ferromagnet . in addition to the magnetization direction , elliptical undulators allow to switch the light helicity , providing a detailed picture of the magnetic structure at the interface . in this way , brck et al . studied the interface of the mnpd / fe exchange bias system@xcite . figure[p_bruck ] shows the scenario resulting from an analysis of respective reflectivity difference curves for either opposite external field directions or light polarisations , separating the pinned and unpinned uncompensated spins within the antiferromagnet . according to this study , the major fraction of the uncompensated afm spins located very close to the ferromagnet is rotatable and antiferromagnetically coupled to the ferromagnet , while the uncompensated afm spins deeper in the afm are predominantly pinned to the af which finally generates the exchange bias effect . similar results were obtained by tonnerre et al . who extended the method to study the interface of a system with perpendicular exchange bias @xcite . ( colour online ) absorption signal and circular dichroic asymmetry @xmath290 observed in specular reflectivity from fe / batio@xmath277 samples through the o @xmath215 ( left ) and ti @xmath216 resonance as well as corresponding element - selected ferromagnetic hysteresis loops ( right ) . they show a magnetic polarisation of the otherwise non - magnetic o and ti ions by the fe at the interface . ( reprinted with permission from @xcite . copyright 2011 , nature publishing group . ) , width=566 ] interesting interface - generated functionality in magnetic heterostructures can emerge already from a magnet in contact with a non - magnetic material . in a very recent study , valencia et al . explored the influence of a ferromagnet in contact with a ferroelectric@xcite . in this experiment , the structural information that can be extracted from scattering experiments was disregarded , still , it can be considered a text - book example showing the sensitivity of xmcd - like experiments exploiting specular reflectivity rather than tey or fy methods to study tiny magnetic effects at buried interfaces . in addition , it represents an important example of interface - generated material properties@xcite . in present solid state research , magnetic control of ferroelectric properties and , in particular , electric control of magnetic order are phenomena with very promising perspectives for applications . unfortunately , most of the known ferroelectrics are either not magnetically ordered or magnetism and ferroelectricity are weakly coupled . exceptions are the so called improper ferroelectrics , where ferroelectricity is induced by complex mostly antiferromagnetic structures , which is discussed for bulk materials in more detail in section[sec : multiferroics ] . these materials , however , are often connected with very low ordering temperatures . in the quest for room - temperature multiferroicity , one way to overcome this limitation is to combine a room - temperature ferromagnet and a room - temperature ferroelectric in a heterostructure . here , the chosen heterostructures consist of fe or co on batio@xmath277 ( bto)@xcite . bto is a robust diamagnetic room - temperature ferroelectric . the study could show that the tunnel - magnetic resistance of fe or co / bto / lsmo trilayers strongly depends on the direction of the ferroelectric polarisation in the insulating diamagnetic bto layer . thus , these heterostructures possess coupled magnetic and ferroelectric , i.e. , multiferroic , properties . element - selective xmcd in reflection geometry could identify the key mechanism : the ferromagnet generates ferromagnetism in the topmost layer of bto by inducing a magnetic moment in the formally non - magnetic o and ti@xmath291 ions as revealed by a seizable asymmetry @xmath290 displayed in figure [ p_valencia2011 ] . the corresponding magnetic interface properties are very weak and hardly detectable by conventional xmcd techniques . exploiting the dichroism in specular reflectivity through the ti @xmath216 and o @xmath215 resonance , valencia et al . have been able to unambiguously prove the existence of ferromagnetic moments at the ti and o ions coupled to the magnetization of the ferromagentic layer . this observation shows the huge advantage of performing xmcd - like experiments exploiting scattered photons to study tiny effects at buried interfaces and , besides these methodological aspects , opens the door for future application of interface - generated room temperature multiferroicity . modified macroscopic and/or microscopic properties at heterostructure interfaces are by far not limited to magnetic behaviour but can be found for other electronic phenomena as well , which offers even more fascinating new perspectives for future nanoscale devices . + + ( colour online ) reflectivity of a 23.2-nm la@xmath286cuo@xmath292 film ( a ) just below and on the mobile carrier peak ( mcp ) at 535 ev , showing the resonant enhancement ; and ( b ) over the full angular range for three different energies . the faster decrease of the red curve compared to the blue one indicates a smoothing of the charge - carrier density at the interface . ( reprinted with permission from @xcite . copyright 2002 , american association for the advance of science . ) ] ( colour online ) energy dependence of ( 0 0 l ) superlattice reflections of smo / lmo multilayers at t=90 k : at l=3 ( line ) , at l=1 ( circles ) , and l=2 ( triangles ) , compared to xas data ( squares , aligned to zero below the edge ) . the strong resonant enhancement in the pre - edge region of the o @xmath215 resonance is indicative of an electronic effect . ( reprinted with permission from @xcite . copyright 2007 , american physical society . ) ] in ultra - thin samples or at heterostructure interfaces completely new phenomena may arise that can not be found in the corresponding bulk materials . an important and fast developing field of research was triggered by the discovery that interfaces of transition metal oxides can be grown with very high structural quality down to a flatness on the atomic level . with these structurally perfect interfaces , fundamental studies of new interface properties have become feasible . examples are the discovery of a highly mobile electron gas @xcite and even superconductivity @xcite at the interface of the two insulating perovskite oxides lao and srtio@xmath277 ( sto ) . a recent example is the observation of insulating and antiferromagnetic properties of lanio@xmath277 ( lno ) in lno / laalo@xmath277 ( lao ) superlattices , as soon as the lno thickness falls below 3 unit cells , while thicker lno samples are metallic and paramagnetic @xcite . finally a major reconstruction of the orbital occupation and orbital symmetry in the interfacial cuo@xmath286 layers at the ( y , ca)ba@xmath286cu@xmath277o@xmath293 / lcmo interface has been observed @xcite . it is a matter of debate to which extent such properties are caused by either electronic reconstruction , strain , oxygen vacancies , or disorder . hence , microscopic understanding of macroscopic interface properties may require detailed knowledge on microscopic electronic properties with high spatial resolution . with its strong sensitivity not only to magnetic but also to charge and orbital order , rsxs is well suited to extract quantitative information on even subtle changes of these electronic degrees of freedom at interfaces . the general strategy here is very similar to the case of magnetic depth profiling : the information on spatially modulated orbital occupation or charge distribution is contained in the reflectivities , measured as a function of momentum transfer and incident photon energy . this has been demonstrated first by abbamonte et al . while studying the carrier distribution in thin films of la@xmath286cuo@xmath292 grown on srtio@xmath277 @xcite . the thin film reflectivities of these samples are characterized by well developed thickness oscillations when measured at the cu @xmath216 resonance , proving high sample homogeneity and surface / interface quality ( see figure[p_abbamonte2002 ] ) . in contrast , with a photon energy corresponding to the mobile carrier peak ( mcd ) , the oscillations vanish at higher angles of incidence . this mobile carrier peak in the o @xmath215 pre - edge region represents a characteristic energy for scattering from the doped carriers with an enhancement of the scattering strength of a single hole by about two orders of magnitude . hence , the reflectivity measured at this photon energy is almost entirely determined by the distribution of the doped carriers . the strong damping of the thickness oscillations at this specific energy is readily explained by assuming a smoothing of the carrier density at the interface towards the substrate . in a very similar way smadici et al . studied the distribution of doped holes in superlattices consisting of double layers of insulating la@xmath286cuo@xmath294 ( lco ) and overdoped la@xmath295sr@xmath296cuo@xmath294 @xcite . while none of these materials is superconducting , the heterostructure shows superconductivity below @xmath297=35k suggesting charge redistribution at the interface . in rsxs with the photon energy tuned to the la @xmath231 resonance , the superstructure gives rise to a series of reflections detectable up to the 5th order . these reflections mainly contain information about the distribution of the sr ions that induce the hole doping . in contrast , at the mobile carrier peak energy only the first superstructure reflection could be observed . this result directly shows that , although being modulated with the superlattice period , the doped hole distribution does not follow that of the sr ions . from a quantitative analysis , an average hole density in the insulating lco layers of 0.18 holes per cu site has been deduced , suggesting that superconductivity occurs in the formally insulating lco layers . in order to clearly separate interface behaviour , often it is of advantage to design a multilayer such that a specific superstructure reflection is interface sensitive , i.e. , its structure factor essentially is given by the difference of the optical properties of an interfacial layer and a bulk - like layer , in that way directly representing the changes at the interface . this approach has been demonstrated first by smadici et al.@xcite who studied the interfacial electronic properties of srmno@xmath277 ( smo ) / lamno@xmath277 ( lmo ) . while the two single materials are a mott insulator ( lmo ) and a band insulator ( smo ) , repectively , ferromagnetic and metallic interface properties had been predicted @xcite and macroscopically observed @xcite . in this experiment , ( 0 0 l ) superstructure reflections were studied in a heterostructure consisting of @xmath298lmo/@xmath299smo double layers . @xmath300 and @xmath3 were chosen such that the l=3 reflection perpendicular to the surface only occurs if scattering from the interfacial and inner mno@xmath286 planes of lmo or smo differ . a pronounced l=3 reflection from this structure could be observed only in the vicinity of the mn @xmath216 resonance and in a very narrow energy region at the onset of the o @xmath215 resonance ( see figure[p_smadici2007 ] ) , i.e. , involving electronic states in the vicinity of the fermi level . the observation of no broken symmetry in the atomic lattice ( i.e. , no l=3 reflection off - resonance ) , but an interfacial reflection induced by the unoccupied density - of - states near @xmath301 gives strong evidence that the interface is characterized by electronic reconstruction . the intensity of the l=3 reflection at the mn @xmath216 resonance was identified to be of magnetic origin from a rough azimuthal dependence of the scattered intensity . comparison of the temperature dependence of the macroscopic properties ( conductivity and magnetization ) with the observed l=3 intensities at the o @xmath215 and mn @xmath222 resonances suggests that metallic and ferromagnetic behaviour is indeed interface generated driven by electronic reconstruction . a very similar approach has been used to characterize also the sto / lao interface in detail very recently by wadati et al.@xcite . ( colour online ) energy scans of the reflectivity from a lno / lao superlattice with constant momentum transfer @xmath302 close to the ( 002 ) superlattice peak using linearly polarised light . the top panel shows the polarisation - dependent experimental data revealing linear dichroism at the ni@xmath224 resonance . the bottom panel shows the corresponding simulated curves for lno layers with ( 1 ) homogeneous orbital occupation within the lno layer stack and ( 2 ) modulated orbital occupation . ( reprinted with permission from @xcite . copyright 2011 , nature publishing group . ) ] while the latter studies revealed interface - driven modifications in a more qualitative way , it has been shown very recently that quantitative information from such superstructure reflections can be derived for each individual atomic plane . this , however , requires detailed modeling of energy- and polarisation - dependent reflected intensities as in the case of magnetic depth profiling discussed above . the studied system is a multilayer made of repeated bilayers of four unit cells of lao and four unit cells of lno . ni in cubic symmetry is characterized by a doubly degenerate @xmath303 level filled with one electron . according to model calculations , epitaxial strain can be utilized to favour the occupation of the in - plane @xmath304 orbital in a superlattice geometry , in this way matching the electronic structure of cuprate high-@xmath305 superconductors , which may be a route to generate high-@xmath305 superconductivity in artificial nanostructures @xcite . while magnetic depth profiling used the contrast obtained by circularly polarised light , light with linear polarisations @xmath33 and @xmath32 ( cf . figure[fig : exp_scheme ] ) was used to measure the linear dichroism of the orbital scattering according to the last term in eq.[eqn : hannon ] . by changing the light polarisation from in - plane to normal to the superlatice plane , the sensitivity to the @xmath306 orbital could be varied . the observed linear dichroism in absorption yields the average difference in occupation of the two @xmath303 orbitals of 5.5% . in addition , the linear dichroism at the second superlattice reflection , sensitive to the difference of the scattering strength from the inner lno unit cell and the interfacial lno unit cell , was monitored . as shown in figure[p_benckiser2011 ] , the observed linear dichroism , here , is much stronger than expected for a homogeneous system with the same orbital occupancy for the inner and interfacial lno layers . hence , an inhomogeneity of the orbital occupation through this 4 unit cells thin lno layer can be readily deduced . modeling the observed energy dependence for the two different linear polarisations , taking the average dichroism as an fixed input , the data could be explained by a distinct higher occupation of the @xmath304 orbitals at the interface ( about 7 % ) compared to the inner layers ( about 4 % ) , which is explained by the reduced gain of kinetic energy for the electrons in the interface layer by hopping across the interface due to the closed shell of the neighbouring al@xmath278 ions . the previous sections demonstrated the capabilities of rsxs for studying ordering phenomena in nanostructures , with particular focus on multilayer samples where interesting physics typically emerge at the interfaces . besides such depth - structured samples , systems can also be lateral structured . this has been discussed for stripe - domains in fepd films above , but can be expanded to artificially designed structures on a nanometer length scale , like regular or irregular line and dot arrays . such regular patterns of magnetic materials are of interest in connection with future data storage technologies . the typical length scales ( nm ) of nanostructures perfectly match the wavelengths of soft x - rays . interestingly , rsxs has not been widely applied to laterally structured samples so far . the first application was presented by chesnel et al . monitoring the magnetization reversal of an regular arrangement of magnetic lines @xcite . these samples have been made of si lines of about 200 nm width and 300 nm height with a line spacing of 75 nm covered by co / pt multilayers . in contrast to the fepd stripe domains discussed above , this structure gives rise to off - specular intensity rods even without magnetism ( see figure[p_chesnel2002 ] ) . analyzing the in - plane momentum transfer @xmath307 as well as the width and the relative intensities of the superstructure - rods of different order yields a detailed characterization of the chemical structure . at the co @xmath216 resonance , additional magnetic information can be extracted from the scattered intensities . here , ferromagnetically aligned neighbouring stripes contribute magnetic intensity to the chemical superstructure rods , while afm aligned lines give rise to pure magnetic satellites at half order positions . measuring hysteresis loops by recording the scattered intensity at various @xmath307 as a function of an external field as shown in figure[p_chesnel2002 ] , yields information about the magnetic coupling of such nanolines and the magnetization reversal process could be modeled in detail . while the hysteresis for @xmath308=0 displays the average magnetic behaviour , as also accessible by macroscopic techniques , the off - specular magnetic rod intensity reflects the magnetic behaviour of the well - ordered part of the sample . the occurrence of significant magnetic intensity at half - order positions for the external field close to the coercive field shows that the demagnetized state of this sample is in fact characterized by an afm alignment of neighbouring lines with a correlation length of about 4 lines . similar studies has been performed on a regular dot array of ferromagnetic permalloy dots @xcite as well as of permalloy rings , where the rsxs intensity distribution has been shown to be sensitive to the possible magnetization states of the single rings , i.e. , vortex or so - called onion states @xcite . resonant scattering at the co @xmath224 resonance from an artificial co / pt multilayer structure on si nanolines separated by 200 nm . upper left corner : chemical superstructure and half - order magnetic superstructure peaks . other panels : evolution of the structural peaks ( a , b , c ) under a perpendicular magnetic field . the bottom panel displays the hysteresis measured on the half - order magnetic peak describing the entire hysteresis loop . ( reprinted with permission from @xcite . copyright 2002 , american physical society . ) ] ordering of mobile charge carriers is a phenomenon observed in a number of transition metal oxides , such as cuprates , nickelates , cobaltates , and manganites . many of the interesting properties seem to be closely related to this kind of order . the most intriguing case is that of the cuprates , as inhomogeneous charge distributions have been discussed for many years in connection with high-@xmath309 superconductivity ( htsc ) in these materials . htsc has fascinated nearly the whole solid state community , and the interest remained even 25 years after the discovery of htsc , since there is no consensus on the mechanism for the superconducting pairing of the charge carriers in these compounds . besides the mechanism of htsc , it is the physics of electron correlations in general , which makes the cuprates so interesting , and charge order is one manifestation . since in many cases , charge ordering involves only a small fraction of the total charge , the high sensitivity of rsxs to scattering from particular electronic states renders the method well - suited to study this phenomenon . + + htsc in cuprates occurs after doping the afm mott - hubbard insulator@xcite with holes or electrons . since the onsite coulomb interaction @xmath310 on the cu sites is considerably larger than the charge transfer energy @xmath311 between cu @xmath221 states and o @xmath218 states , in the undoped system the gap is essentially determined by @xmath311 and the holes are not formed on the cu sites but on the o sites@xcite . the doping - induced mobile holes are shared by four oxygen sites surrounding a divalent cu site with one hole in the @xmath221 shell , in this way forming the antiferromagnetically coupled zhang - rice singlet@xcite state . this state may be considered as an effective lower hubbard band state . the effective upper hubbard band is formed predominantly by cu @xmath312 states hybridised with some o @xmath313 states . the doping , which leads to a controlled metal - insulator transition , occurs by block layers between which the cuo@xmath138 layer are embedded . in la@xmath314sr@xmath315cuo@xmath138 hole doping of the cuo@xmath138 layers occurs via replacing the trivalent la ions in the lao block layers by divalent sr ions . long - range afm order disappears at @xmath240 = 0.05 and the highest superconducting transition temperature @xmath305 is reached at @xmath240= 0.18 . in the overdoped case ( @xmath316 0.18 ) the normal state can be well described by a fermi liquid . for @xmath317 0.15 the normal state properties are far from being understood . there is no agreement what should be the minimal model which contains the basic physics of underdoped cuo@xmath138 layers . several phases are found in this doping regime : an afm insulating phase , the high-@xmath305 superconducting phase , and a charge and spin ordered phase@xcite . regarding the latter two different phases are discussed in the literature : the checkerboard phase and the stripe - like phase . in the stripe - like phase , afm antiphase domains are separated by periodically spaced domain walls along the cu - o directions in which the holes are situated ( see figure[tranquada ] ) . partially motivated by the detection of incommensurate low - energy spin excitations , the existence of a stripe - like state was proposed by theoretical work on the basis of a hartree - fock analysis of the one - band or three - band hubbard model@xcite . the proposed ground state resembles a soliton state in doped conjugated polymers@xcite . the theoretical predictions of a stripe phase were based on a mean - field approximation and the importance of the long range coulomb interaction , not taken into account in the hubbard model , was pointed out@xcite . the phase separation , which seems that the afm background expels holes , was supposed to have a strong influence on the physical properties of doped cuprates : ( i ) a central ingredient to the pairing mechanism in htsc , ( ii ) the non - fermi - liquid behaviour around optimal doping , and ( iii ) the existence of the pseudogap in the underdoped region . the stripe - like structure consists of three concomitant modulations of the spin density , of the charge density , and of the lattice which is coupled to the charge modulation . the spin density modulation has twice the wavelength of the charge and lattice modulation , and in a scattering experiment , these two modulations give rise to superstructure reflections around the afm bragg peaks at ( 0.5@xmath318,0.5,l ) and around the structural reflection at ( @xmath319,0,l ) , respectively , and symmetry related positions ( see figure[tranquada ] ) . assuming half - doped stripes , independent of the average doping concentration , the propagation vector @xmath320 is determined by @xmath321 where @xmath3 is the average stripe distance given in @xmath253-spacings of the lattice . using the same assumption about the constant doping concentration in the stripes , @xmath320 should be equal to 0.5@xmath322 . actually this is observed for @xmath323@xcite , indicating that in real space incommensurate stripes exist , where the stripe distance varies on a local scale and where only the mean distance between the stripes is determined by @xmath322 . in several doped cuprates such as la@xmath314ba@xmath315cuo@xmath138 ( lbco ) , ( la , nd)@xmath314sr@xmath315cuo@xmath138 , and ( la , eu)@xmath314sr@xmath315cuo@xmath138 , the stripe - like order is stabilized near @xmath324 , concomitant with a suppression of superconductivity . it occurs in the so - called low - temperature tetragonal ( ltt ) phase , which is characterized by a corrugated pattern caused by rotation of the cuo@xmath325 octahedra . evidence for static stripes was first detected in an elastic neutron scattering study on the system ( la , nd)@xmath314sr@xmath315cuo@xmath138 , where the superstructure reflection for the spin and the lattice modulation have been detected@xcite . neutron scattering@xcite and non - resonant hard x - ray scattering@xcite monitor the ordering of the charges indirectly by the associated lattice distortion . the reason for this is that these techniques are mainly sensitive to the nuclear scattering and the core electron scattering , respectively . detection of charge order by these methods is hence not fully conclusive , since lattice distortions may also occur with very small or even no charge order of the valence electrons : they may be caused as a result of a bond - length mismatch between different units of a solid ( e.g. planes or chains ) or by a spin density wave alone like , e.g. in chromium metal . on the other hand , rsxs is a method which can directly probe the existence of the charge modulation of the conduction electrons . furthermore , this method enables one to study the wave length of the modulation , the coherence length , the temperature dependence of the order parameter , and in principle also the momentum dependence of the form factor which would give the detailed spatial dependence of the charge modulation . this has been demonstrated for the doped cuprates by rsxs at the o @xmath215 and the cu @xmath216 edges by abbamonte et al.@xcite . in figure[lbco_abba ] we show data from this work on lbco@xcite , comparing x - ray absorption data , measured with the fluorescence method , with the resonance profile , i.e. , the photon energy dependence of the intensity of the charge - order superstructure reflection at energies near a core excitation . edge . green line : xas spectrum for photons polarised parallel to the cuo bond direction . red circles : intensity of the ( @xmath326,0,l ) superstructure reflection at l=0.72 . the enhancement at the mobile charge carrier peak ( mcp ) and the upper hubbard band ( uhb ) demonstrates a significant modulation of the doped hole density . blue line : form factor for scattering from doped holes calculated from xas data . ( b ) data near the cu @xmath224 edge for l=1.47 . ( reprinted with permission from @xcite . copyright 2005 , nature publishing group.),width=283 ] absorption edges at the o @xmath215 and cu @xmath327 edges in cuprates have been extensively studied by electron energy loss spectroscopy ( eels)@xcite and xas@xcite . the o @xmath215 edge spectra of hole - doped cuprates show two pre - peaks : the lower one at 528.6 ev is assigned to transitions into empty o @xmath218 states related to the zhang - rice singlet states . the intensity of this pre - peak , also called the mobile carrier peak ( mcp ) is directly related to the number of the mobile charge carriers . the second pre - peak ( uhb ) at 530.4 ev is related to the upper hubbard band . with increasing doping concentration there is a spectral weight transfer from the upper uhb peak to the mcp peak which is a classical signature of a doped correlated system@xcite , not observed in doped semiconductors . the resonance profile for the ( @xmath328,0,l ) superstructure reflection shown in figure[lbco_abba](a ) for the o @xmath215 edge exhibits a prominent resonance at the mcp , in this way providing a direct link to the doped holes and hence identifying the charge order . interestingly , the reflection also resonates at the uhb energy , which means that also the mottness is spatially modulated . mottness in this context means that the system acts like a mott insulator . the resonance profile for the ( @xmath328,0,l ) superstructure reflection shown in figure[lbco_abba](b ) shows also a strong enhancement near the cu @xmath224 absorption peak . this resonance probably arises from the modulation of the cu lattice@xcite . the charge - order reflections are sharp along h but rod - like along l , indicating quasi - long - range order in the cuo@xmath138 plane but weak coupling between planes . in the work by abbamonte et al.@xcite not only the existence of a charge modulation in lbco has been directly demonstrated but also a quantitative analysis has been presented . analogous to the procedure outlined in section [ sec : rsxsandxas ] , the imaginary part of the form factor can be derived from x - ray absorption data and finally the real part can be derived via a kramers - kronig transformation . the form factor for that part which is related to doped holes is shown in figure[lbco_abba](a ) by the blue curve . one realizes a strong enhancement of the calculated form factor near the two pre - peaks although the line shape does not perfectly agree with the measured resonance profile . according to this analysis , the form factor for the scattering of a doped hole at resonance has a value of 82 , which means that a single hole in this compound scatters like a pb atom off resonance . since the scattered intensity is determined by the form factor squared , this also means that at resonance the scattered intensity of a hole is amplified by a factor of about 7000 . this fact demonstrates the enormous sensitivity of rsxs to the charge modulation of valence electrons which renders this method an ideal tool to study the charge modulation of doped holes in transition metal oxides . the amplitude of the charge modulation was estimated to be 0.063 holes . in a one - band model , which only contains cu sites , the charge modulation for @xmath329 should be close to 0.5 holes . as outlined above , this value is expected for this doping concentration in a simple model in which three cu - o lines are not doped and the fourth line is 50% doped . the relatively small experimental value of 0.063 hole modulation can be explained by the fact that the holes are distributed among many different ( oxygen ) sites . in figure[lbco_abba](b ) xas data at the cu @xmath224 edge together with a resonance profile is presented . here the enhancement of the scattering length was estimated to be about 300 . finally , the temperature dependence of the charge ordering in lbco has been studied yielding a transition temperature @xmath330 = 60 k which is the same as the transition temperature @xmath331 for a lattice transition from the low - temperature orthorhombic ( lto ) to the low - temperature tetragonal ( ltt ) phase . in the latter structure , the cuo@xmath325 octahedra are tilted along the cu - o bond direction stabilizing the stripe formation . a similar rsxs study has been performed on the system la@xmath332eu@xmath333sr@xmath315cuo@xmath334 ( lesco)@xcite . generally , with decreasing ionic radius of the substitutes of la ( here eu ) , the tilt angle @xmath212 in the ltt phase increases leading to a higher lattice transition temperature t@xmath335 . this stabilizes the stripe order and leads to a stronger suppression of high-@xmath305 superconductivity@xcite . charge - order diffraction peaks for lesco could be observed over a range of doping levels @xmath240 . besides information about the phase diagram discussed below , data for different @xmath240 allow to extract more detailed information about the amplitude of the hole doping modulation from the photon energy dependences according to the following considerations . in figure[lesco1_fink ] ( a ) and ( b ) lesco data for @xmath240=0.15 are reproduced , showing xas results , calculated intensities , and a rsxs scattering profile . the calculated intensities were derived from the calculated form factors presented in figure[lesco1_fink ] ( d ) and ( f ) which in turn were evaluated from doping dependent xas data as described above . the resonant scattering intensity of lesco in red as a function of photon energy through the o @xmath336 [ cu @xmath327 ] absorption edge for the stripe superstructure peak . data were taken near the o @xmath215 [ cu @xmath224 ] edges for l=0.75 [ l=1.6 ] . also shown is the x - ray absorption spectrum in black and the calculated scattering intensity in dark blue . ( c ) the real ( solid line ) and the imaginary ( dotted line ) parts of the atomic form factor @xmath100 at the o @xmath215 edge of lsco for @xmath240=0.07 depicted in red and @xmath240=0.15 depicted in blue . ( d ) the real part in red and the imaginary part in blue of the atomic form factor @xmath100 at the cu @xmath327 edge of lsco for @xmath240=0.125 . ( reprinted from @xcite . copyright 2005 , american physical society.),width=7 ] assuming a linear variation of the atomic form factor @xmath339 as a function of the doping concentration @xmath240 one can expand @xmath340= @xmath341 . then the intensity ratio between the lower hubbard band ( lhb ) and the upper hubbard band ( uhb ) is @xmath342_{lhb}/[\delta f(e , x)/\delta x]_{uhb}|^2 $ ] . a remarkable result , typical of a correlated electron system , is that due to a spectral weight transfer , in the xas data the intensity of the uhb decreases proportional to @xmath343 while the intensity of the mcp peak increases proportional to @xmath344@xcite . since the absorption is related to the form factor , in first approximation , the ratio for the form factor of the uhb to the lhb should be near 2 , i.e. , the intensity ratio in the resonant diffraction profile should be four , in agreement with the calculated ratio [ see figure[lesco1_fink ] ( a ) ] , but in clear disagreement with the measured resonance profile presented in the same figure . this discrepancy was explained in terms of a non - linear change of the absorption at the uhb and the mcp observed for doping concentrations @xmath345@xcite . from this analysis it was concluded that the hole doping modulation per cu site was larger than 20 @xmath346 in agreement with the modulation of about 50 @xmath346 derived for lbco mentioned above@xcite . eu@xmath333sr@xmath347cuo@xmath334 near the o @xmath215 edge measured in the fluorescence yield mode in blue . also shown in black is the ( 001 ) superstructure reflection intensity as a function of the photon energy . the inset shows the ( 001 ) reflection measured with a photon energy of 533.2 ev at @xmath348 = 6 k. ( reprinted from @xcite . copyright 2011 , american physical society.),width=283 ] eu@xmath333sr@xmath347cuo@xmath334 normalized to the intensity at @xmath348= 6 k. squares : ( 001 ) reflection measured with photon energies 533.2 ev near the o @xmath215 edge . circles : ( 0.254 0 0.75 ) reflection measured with photon energies 529.2 ev near the o @xmath336 edge . diamonds : ( 0.254 0 1.6 ) reflection measured with photon energies 929.8 ev near the cu @xmath327 edge . ( reprinted from @xcite . copyright 2011 , american physical society.),width=283 ] in figure[lesco2_fink ] we illustrate that other resonances near the o @xmath215 edge can be used to obtain information on the structure of cuprates . there , besides the first two pre - peaks in the xas spectrum , discussed already before , near 533.2 ev a further peak appears due to a hybridisation of o @xmath218 states with rare earth ( @xmath284 = la and eu ) @xmath349 and/or @xmath229 states@xcite . in figure[lesco2_fink ] we also show the photon energy dependence of the ( 001 ) reflection measured in the ltt phase at @xmath348=6 k. in this phase , neighbouring cuo@xmath138 planes are rotated by 90@xmath350 , yielding o sites with different ( rotated ) local environments . as a result , the ( 001 ) reflection becomes allowed at resonance@xcite . the strong resonance at 533.2 ev is due to octahedral tilts , which cause different local environments and affect the hybridisation between the apical o and the @xmath284 orbitals . in the lto phase , neighbouring cuo@xmath138 planes are just shifted , not rotated , with respect to each other . in this case the ( 001 ) reflection remains forbidden even at resonance . using the resonance at 533.2 ev it is possible the detect the lto - ltt phase transition with soft x - rays . this transition is not detectable by the orthorhombic strain ( a - b splitting ) , because the ( 100)/(010 ) reflections can not be reached due to the limited range in momentum space in rsxs . with the help of this resonance feature , it is possible to study the temperature dependence of the structural ltt order and the charge order in one experiment , as shown in figure[lesco3_fink ] . here the temperature dependent rsxs data of lesco @xmath240=0.15 of the stripe superstructure reflection and the ( 001 ) reflection are depicted . these data indicate a first order lto - ltt transition at @xmath331=135 k and a charge order transition at @xmath330=65 k , clearly showing that the two phase transition are well separated . a systematic rsxs study of the doping dependence of charge order yielded the phase diagram for lesco shown in figure[lesco4_fink ] . eu@xmath333sr@xmath315cuo@xmath334 showing transition temperatures for the ltt phase @xmath331 , the antiferromagnetic structure @xmath351 , the magnetic stripe order @xmath352 , the stripe like charge order @xmath330 , and the superconducting transition temperature @xmath305 . closed circles from rsxs experiments@xcite . open circles from ref.@xcite . closed diamond from neutron diffraction data presented in ref.@xcite . ( reprinted from @xcite . copyright 2011 , american physical society.),width=7 ] different from lbco , the lesco results represent the first example in which the lattice transition temperature @xmath331 is so high that the charge order can no longer be stabilized at this temperature and therefore a gap of 55 k exists between @xmath331 and @xmath330 . this is a remarkable result since it demonstrates that the charge order is at least partially electronic in origin . the width of the superstructure reflections , being about five times larger than the experimental resolution , indicate disorder effects and/or glassy behaviour . from the widths the coherence lengths of the charge order have been determined . these lengths are of the order of about 100 lattice constants and increase , at least for smaller @xmath240 , linearly with increasing sr concentration . as a function of doping no maximum of the coherence lengths has been detected for integer numbers of stripe separation in units of the in - plane lattice spacing corresponding to @xmath353 with n being integer . this is different from doped nickelates where a clear maxinum was detected for @xmath354@xcite . the results on the cuprates indicate that in these systems the coherence length is not only related to the commensurability of the stripe lattice with the underlying cuo lattice . furthermore the results point to the fact that the coherence lengths in these systems are not determined by the impurity potential of the doping atoms but is probably related to an increasing tilt angle with increasing sr content . finally the wave vector of the superstructure reflection increases with increasing sr content , at least up to @xmath240=0.125 . since the nesting vector between parallel segments in the fermi surface decreases with increasing doping@xcite , this indicates that stripes are not conventional charge density waves caused by nesting . thus these results point to more strong coupling scenarios for the stripe formation . in a more recent work the stripe modulations in lbco and la@xmath355nd@xmath356sr@xmath357cuo@xmath334 ( lnsco ) were compared using rsxs and hard x - ray scattering@xcite . making use of a two - dimensional detector for rsxs an isotropic coherence length of the charge modulation of the hole density in the cuo@xmath138 planes has been detected . also the strain modulation of the lattice shows an isotropic coherence length . these results are surprising given that the stripes in a single cuo@xmath138 layer are highly anisotropic . in the direction perpendicular to the cuo@xmath138 planes , the stripes are weakly correlated giving rise to the uniform streak of intensity along l. both the charge and the strain modulation is better correlated in lbco than in lnsco . the in - plane hole correlation lengths for lbco and lnsco are @xmath358 and @xmath359 , respectively . similar to lesco , in lnsco the hole density modulation sets in well below the transition into the ltt phase while for lbco @xmath330 is equal to @xmath331 and below @xmath330 the amplitude of the modulation is independent of the temperature . this suggests that the ltt transition occurred at a higher temperature in lbco then it is likely that both the electronic and structural modulations would also have persisted to higher temperatures . a further result from this investigation is that the electronic charge stripe modulation in lbco is 10 times larger in amplitude than in lnsco . this result is surprising since from various other experiments , for lbco , lnsco , and lesco a similar amplitude for the charge modulation is expected . further work is required to solve this puzzle . until now we have reviewed rsxs studies on static stripe - like cdw order in lbco , lnsco and lesco . in all these compounds the cdw is stabilized by an ltt lattice distortion . an important issue in this context is whether static stripes also exist in other two - dimensional cuprates . checkerboard - like static charge order has been reported in ca@xmath314na@xmath315cuo@xmath138cl@xmath138 ( nccoc ) using surface sensitive scanning tunneling experiments@xcite . cdw order could be related to the electron pockets detected in underdoped ybco in quantum oscillation measurements@xcite . the observation of a complete wipe - out of the cu nuclear quadrupole resonance signal at low temperatures in ni substituted ndba@xmath138cu@xmath139o@xmath140@xcite , similar to that of stripe ordered underdoped cuprates , suggests the existence of static stripes in this compound . thus one may think that static stripes are generic to the cuprates and it is very important to look at stripe - like cdw order in other systems which are not stabilized by an ltt lattice distortion . a rsxs study on a tetragonal compound , already discussed in section[sec : newmat ] , has been performed on an excess o doped la@xmath138cuo@xmath292 layer ( @xmath360 ) epitaxially grown on a srtio@xmath139 crystal@xcite . in this work the resonance effects at the pre - edge of the o @xmath215 shell excitation and near the cu @xmath222 edge have been used for the first time to exploit charge modulation in the cuo@xmath138 layers . it was there where it was demonstrated that at the pre - edge of the o @xmath215 shell excitations , the scattering amplitude for mobile holes is enhanced by a factor of 82 which leads to an amplification of diffraction peaks related to these holes by more than 10@xmath361 . extensive non - successful search for superstructure reflections due to a stripe - like cdw lead to the suggestion of an absence of static stripes in this high-@xmath305 superconductor . a further rsxs study on a possible charge modulation in a tetragonal cuprate has been performed on nccoc@xcite . from experiments at the mcp peak at the o @xmath215 edge or at the cu @xmath257 edge no evidence for a checkerboard - like charge modulation has been detected . from this null experiment the authors have concluded that the checkerboard order observed in the sts experiments is either glassy or nucleated by the surface . a further rsxs experiment on ni substituted ndba@xmath138cu@xmath139o@xmath140 was stimulated by the above mentioned cu nuclear quadrupole resonance experiment@xcite . no superstructure reflection due to stripe - like charge order could be detected at low temperatures in this compound@xcite . more rsxs experiments on charge order have been performed on the most studied high-@xmath305 superconductor ybco which contains a cuo@xmath138 double layer and in addition cuo@xmath139 chains along the @xmath362 axis . in a first paper on the ortho - ii yba@xmath138cu@xmath139o@xmath363 phase , in which the cuo@xmath139 chain layers are ordered into alternating full and empty chains , charge order in the planes and in the chains were reported@xcite . this observation was later questioned , partially by the same authors , in a more refined rsxs study on ortho - ii and ortho - viii oxygen ordered ybco@xcite . however , charge order in underdoped ybco was postulated on the basis of nmr studies in high magnetic fields @xcite and was eventually detected in very recent experiments by rsxs @xcite and also with hard x - ray diffraction @xcite . the former study showed by the resonance profile near the cu @xmath224 resonance , that the charge order in fact takes place in the cuo@xmath138 planes rather than the chains . the temperature dependence of the charge order found in both studies clearly revealed competition with superconductivity . this result has important implications for the understanding of the material and the cuprates in general . it readily explains the anomalously low superconducting temperature in underdoped ybco and shows that charge order seems to be a generic feature of cuo@xmath138 planes in layered cuprates . + + among the various mechanisms discussed for high-@xmath305 superconductivity in cuprates , the electronic model based on ladder - like structures in which cu and o atoms are ordered in two chains , coupled by rungs , has been heavily discussed@xcite . in this model , depending on the exchange coupling along the chains and that along the rungs , singlets can exist on the rungs of a doped ladder compound and exchange - driven superconductivity can be formed . depending on the size of parameters or doping concentration , an insulating hole crystal in which the carriers crystallize into a static wigner crystal may also form the ground state . note , this would be an electronic charge density wave ( cdw ) , driven by the coulomb interaction and not by a coupling to the lattice ( peierls transition ) . the competition of the two phases is similar to that believed to occur between ordered stripes and high-@xmath305 superconductivity in two - dimensional cuprates . indeed the only known doped ladder compound sr@xmath364ca@xmath315cu@xmath365o@xmath366 ( scco ) exhibits both phases : superconductivity has been detected for @xmath240= 13.6 below @xmath305=12 k at hydrostatic pressures larger than 3.5 gpa@xcite , while for @xmath240=0 scco exhibits a cdw . therefore and because the interplay between charge and spin degrees of freedom can be easier studied theoretically in quasi one - dimensional systems , experimental studies of charge density modulations in scco by rsxs are extremely important@xcite . scco consists of two different alternating types of copper oxide sheets perpendicular to the @xmath362 axis . one with chains , formed out of edge - sharing cuo@xmath334 plaquettes and one with weakly coupled two - leg ladders , i.e. , two parallel adjacent chains formed out of corner - sharing cuo@xmath334 plaquettes . both the chains and the ladders are aligned along the @xmath367 axis . the sheets are separated by sr / ca ions . since the cuo@xmath334 plaquettes in the chains and those in the ladders are rotated by 45@xmath350 with respect to each other , the ratio of the lattice constant for the ladder @xmath368 to that of the chain @xmath369 is about @xmath370 . this means that one unit cell is composed out of ten cuo@xmath334 plaquettes forming the chains and two times seven equal to 14 cuo@xmath334 plaquettes forming the two - leg ladders . this leads to a `` misfit compound '' which is structurally incommensurate and has a large unit cell along the c axis with the length @xmath367=27.3@xmath371 . according to the chemical formula and assuming that cu is divalent , there are 6 holes per formula unit . the distribution of the holes among chains and ladders is still under debate . earlier polarisation dependent xas measurements for @xmath240=0@xcite suggested that 0.8 holes are on the ladder and the rest is on the chains . replacing the sr ions by the smaller isovalent ca ions , i.e. , by chemical pressure , there is a hole transfer from the chains to the ladders which according to the earlier xas results@xcite , for @xmath240=12 increases the number of holes on the ladders to 1.1 , i.e. , 0.08 holes per cu site . optical spectroscopy@xcite , arpes experiments@xcite and recent xas experiments@xcite showed for high ca replacements values of 0.2 , 0.15 - 0.2 , and 0.31 holes per cu site , respectively , with no sign of convergence with time . in figure[scco_rusyd1 ] we reproduce scco data showing xas spectra for @xmath240=0 and a photon polarisation parallel to @xmath367 together with resonance profiles for various ca concentrations@xcite . in agreement with previous xas data near the o @xmath215 edge@xcite the first pre - peak at 528.4 ev and the shoulder at 528.7 ev were assigned to hole states at the chains and the ladders , respectively . the next peak at 530 ev was interpreted in terms of the uhb . like in other doped cuprates , the first peak at 931.5 ev in the cu @xmath224 xas spectrum is assigned to a transition into empty cu @xmath221 states while the shoulder at 933 ev is related to ligand ( o ) hole states . for @xmath240=0 , a resonance profile measured at a photon energy of 528.7 ev and a wave vector @xmath372 shows a clear resonance for @xmath373=1/5 ( see figures[scco_rusyd1 ] and [ scco_rusyd2 ] ) . this observation together with the fact that away from the ladder hole shoulder no scattering intensity has been detected , signals the existence of a commensurate hole crystal in the ladders with a wave length @xmath374 . at higher ca concentrations ( 10@xmath375 ) a resonance is observed for @xmath373=1/3 ( see figures[scco_rusyd1 ] and [ scco_rusyd2 ] ) . on the other hand , no resonance has been detected for @xmath373=1/4 . this could indicate that the hole crystal is stable for odd , though not even , multiples of the ladder period . the resonance profiles at the cu @xmath224 edge shows , in particular for higher @xmath240 values two peaks , at 930 ev and at 933 ev . the latter peak coincidences with the ligand - hole shoulder and thus indicates the presence of a hole modulation . the former indicates a lattice modulation@xcite which is more pronounced for larger hole ( ca ) concentrations . the hole crystal intensities show both for @xmath373 = 1/5 and 1/3 a strong temperature dependence which points to melting of the hole crystal at higher temperatures and also signals that the modulation is not caused by the misfit between the chains and the ladders . the non - existence of hole crystals with non - integer l values signals that incommensurate hole crystals melt even at very low temperatures . hole crystallization in ladder compounds was predicted by a @xmath376 model@xcite and density matrix renormalization group calculations@xcite . in particular it was found that holes like to pair up on the rungs of the ladder . thus doped rungs are separated by the wave length @xmath377 and the doping concentration @xmath378 per cu site should be then @xmath379 or the number of holes on the ladder per unit cell should be @xmath380 . from these considerations we expect for zero ca concentration @xmath381 @xmath382 and @xmath383 and for @xmath384 @xmath385 and @xmath386 . these values are in good agreement with the values derived in the recent xas evaluation@xcite and thus offer strong support for a pairing of holes along the rungs and thus also support models on the mechanism of high-@xmath305 superconductivity in cuprates based on ladder systems . on the other hand , these results are in striking disagreement with other results on the number of holes an the ladders mentioned above . the absence of a @xmath373=1/4 periodicity was explained in terms of resonant valence bond calculations indicating that the wigner hole crystal is stable for odd , but not for even multiples of the ladder period . further theoretical work showed that the widely used ladder @xmath376 model is not sufficient and has to be supplemented by coulomb repulsion between the holes on neighbouring ladders to explain the existence of a hole crystallization on the ladders . a mean field calculation of the extended model could explain a charge density wave with an odd period@xcite . at the end of the paragraph reviewing charge order in the system scco , we report on rsxs results on the chains of these compounds@xcite . for the undoped compound in which divalent sr was replaced by trivalent la , a superstructure reflection for @xmath387 has been detected for photons with an energy near the cu @xmath224 and @xmath226 edges . since in this compound , no holes can form a charge density wave and since the superstructure reflection was independent of the temperature , it has been interpreted in terms of a pure strain wave formed by the misfit between the chains and the ladders . for doped scco an incommensurate superstructure reflection at @xmath388 has been detected . it shows a remarkable temperature dependence and appears mainly near the ligand - hole shoulder . from this it was concluded that in scco a strain - stabilized charge density wave is formed . + + the first transition metal oxide system in which static stripe - like order was detected was la@xmath314sr@xmath315nio@xmath389 ( lsno ) . in contrast to lsco , upon doping , i.e. , replacing the trivalent la ions by divalent sr ions or by adding oxygen , the nickelates remain insulating except for very high doping concentrations , and superconductivity has not been detected so far . both in the electron diffraction @xcite and neutron diffraction@xcite data , superstructure reflections were observed , which were interpreted in terms of coupled charge and spin - density modulations in the nio@xmath138 planes , similar to the case depicted for the cuprates in figure[tranquada ] . different to cuprates , however , the stripes are rotated by 45@xmath350 . in an orthorhombic unit cell with axes rotated by 45@xmath350 with respect to the ni - o bonds , the magnetic peaks are split about the antiferromagnetic position ( 1,0,0 ) along the [ 100 ] and [ 010 ] directions by @xmath320 , while the charge order are split about the fundamental bragg peaks by 2@xmath320 , as in the case of cuprates . also similar to the cuprates , for hole concentrations @xmath390 , in the nickelates @xmath391 . stripe - like charge and spin order in lsno is observed for 0.15 @xmath392 0.5 . the stripe order is most stable at a doping level @xmath354 where it shows the highest charge and spin ordering temperatures and the longest correlation length . different from the cuprates in la@xmath138nio@xmath334 there are two intrinsic holes on each ni site in the @xmath312 and the @xmath393 state@xcite . edges of la@xmath394sr@xmath333nio@xmath334 for the spin and charge modulation superstructure reflections recorded with @xmath33 and @xmath32 polarised photons . also shown are xas data . the experimental results ( points ) are compared to simulations ( solid line ) . ( reprinted from@xcite . copyright 2005 , american physical society.),width=7 ] rsxs experiments were performed on a la@xmath314sr@xmath315nio@xmath334 single crystal@xcite . the energy profiles at the ni @xmath216 edges , shown in figure[lsno_schuessler ] , exhibit resonances at the charge - order wave vector ( 2@xmath320,0,1 ) and at the spin - order wave vector ( 1-@xmath320,0,0 ) with @xmath320=0.196 in the above notation . a resonance enhancement is also observed at the la @xmath233 edge at 849.2 ev , however , this enhancement is much weaker than at the ni @xmath216 edges indicating that the modulation occurs mainly in the nio@xmath138 planes . the resonances show a strong polarisation dependence , e.g. the intensity of the magnetic superstructure peak for @xmath32 polarisation ) is only 10 @xmath346 of that for @xmath33 polarisation . since only a magnetic moment perpendicular to the polarisation of the incoming light is probed and because in the chosen experimental geometry the @xmath33 polarisation is perpendicular to the stripes , one can conclude that the ni spins are essentially collinear with the stripes . further conclusions can be drawn from a comparison with calculations based on a configuration - interaction model ( see figure[lsno_schuessler ] ) . the difference of about 1 ev for the charge modulation resonances for @xmath33 and @xmath32 polarisation indicates a large energy splitting between the @xmath312 and the @xmath393 levels . the comparison also signals that the holes are going mainly to the o sites on a ligand molecular orbital with @xmath395 symmetry . due to the strong on - site coulomb repulsion of two holes on the ni sites , the ni 3@xmath253 count is not strongly modulated in the stripe structure , ruling out a ni@xmath285/ni@xmath278 charge order scenario . on the contrary , the the situation is very close to that of stripe structures in cuprates , except that in the nickelates there is an additional intrinsic hole in the ni 3@xmath396 states . in summary , the comparison with the rsxs data with the cluster calculations yields interesting results on the stripe structure and the electronic structure of nickelates : ( i ) both the charge order and the spin order resides in the nio@xmath138 layers , ( ii ) the doped holes are mainly located on o sites and the spin of these holes are coupled anti - parallel to the spins of the intrinsic hole on the ni 3@xmath397 states in close analogy to the zhang - rice singlets in the cuprates . + + magnetite , fe@xmath139o@xmath334 , is another prototype correlated transition metal oxide , in which a subtle interplay of lattice , charge , spin , and orbital degrees of freedom determines the physical properties@xcite . it was the first magnetic material known to mankind . verwey discovered in the late 1930 that upon lowering the temperature below the verwey temperature @xmath398=123 k , fe@xmath139o@xmath334 undergoes a first - order transition connected with a conductivity decrease by two orders of magnitude . magnetite has been considered as a mixed valence system . in this compound at high temperatures , the tetrahedral @xmath399 sites are occupied by fe@xmath278 ions while the octahedral @xmath400 sites are occupied by an equal number of randomly distributed fe@xmath285 and fe@xmath278 cations . according to verwey the transition into the low conducting state is caused by an ordering of the fe valency on the @xmath400 sites accompanied by a structural transition . although numerous experimental@xcite and theoretical studies@xcite have been performed on magnetite , no full consensus over all aspects of the order transition exists . with its particular sensitivity to the fe @xmath221 electronic structure , recent rsxs studies at the fe @xmath216 edges contributed significantly to the understanding of the verwey transition@xcite . xas data of fe@xmath139o@xmath334 at the fe @xmath216 edge together with resonace profiles of the ( 0,0,@xmath401 and the ( 0,0,1 ) superstructure reflections are displayed in figure[feo_schlappa ] . xas spectra of fe@xmath139o@xmath334 compared to resonance profiles of the ( 0,0,@xmath402 ) and the ( 0,0,1 ) superstructure reflections . the solid lines are simulations for the charge order scenario while the dashed line is related to the homogeneously mixed - valence scenario . ( reprinted from@xcite . copyright 2008 , american physical society.),width=7 ] based on previous interpretations of the fe @xmath216 absorption spectra and on simulations it was possible to decompose the absorption structure and to assign the different peaks to specific fe sites [ see figure[feo_schlappa ] ( a ) ] . the low - energy features ( red solid line ) were assigned to fe@xmath285 . remarkably , the ( 0,0,@xmath402 ) shows a resonance exactly at these energies which means that the ( 0,0,@xmath402 ) peak is due to an order , which involves only @xmath400-site fe@xmath285 ions . this implies further that there is an orbital order of the @xmath403 electrons that distinguishes different @xmath400-site fe@xmath285 ions . this experimental result on the orbital ordering of the @xmath403 electrons of the fe@xmath285 ions agrees with the theoretical predictions on the basis of lda+@xmath310 calculations@xcite . the ( 0,0,1 ) superstructure reflection shoes maxima at two energies corresponding to the @xmath400-site fe@xmath285 and the @xmath400-site fe@xmath278 ions . this is exactly expected for a charge order involving the two @xmath400-site fe ions . in figure[feo_schlappa](c ) the measured resonance profile is compared with simulations for a charge ordered state ( solid line ) and a homogeneous fe@xmath404 configuration ( dashed ) . the very good agreement with the charge order state and the disagreement with the homogeneous state clearly supports charge order on the @xmath400 sites in magnetite . in this context , one should mention that the charge order of the fe @xmath221 electrons on the @xmath400-site is small and that one should rather speak of a modulation of the @xmath403 count which is partially screened by a charge transfer from the oxygen neighbors to the empty @xmath303 states@xcite . in transition metal compounds the degeneracy of partially occupied electronic states is very often lifted , resulting in a long - range orbital ordering of the occupied orbitals below a transition temperature . therefore orbital ordering phenomena play a fundamental role in determining the electronic and magnetic properties of many transition metal oxides . the origin of the orbital ordering is still under discussion . one scenario explains orbital ordering in terms of a purely electronic mechanism , i.e. , by a super - exchange between the transition metal ions@xcite . the other model explains orbital ordering by a lifting of the degeneracy of the atomic level by a cooperative jahn - teller effect , i.e. , by a net energy gain due to lowering of the electronic state and an energy loss by a local distortion of the ligands around the transition metal , similar to the peierls transition mentioned above . in this second scenario lattice effects are important . possibly in real systems a combination of the two models is realized . + + classical examples of orbitally ordered systems are kcuf@xmath139 @xcite and lamno@xmath139 . in case of manganites , in particular , orbital order has attracted considerable attention and controversy since they exhibit a wide diversity of ground states including phases with colossal magnetic resistance@xcite or charge and orbitally ordered ground states . in many cases small changes in some parameters such as the charge carrier doping or temperature can lead to transitions between disparate ground states . the origin of this rich physics is widely believed to be due to a complex interplay of charge , magnetic , orbital , and lattice degrees of freedom , which results in different strongly competing electronic phases . two prototypical examples are the manganites @xmath405mno@xmath139 and @xmath406mno@xmath334 ( @xmath284=trivalent rare earth ions , @xmath399= divalent alkaline - earth ions ) , which belong to the famous ruddlesden - popper series . in all these compounds each mn ion is surrounded by an o octahedron . the interaction with the o ions leads to a partial lifting of the degeneracy of the @xmath221 states into the lower @xmath403 states and the twofold degenerate @xmath303 states at higher energies . in the undoped case ( @xmath240=0 ) , the mn ions have a formal valence of 3 + with four electrons in the @xmath221 shell . since the hund s rule energy is much larger than the crystal field splitting , the mn @xmath407 is in a very stable high - spin state that corresponds to three spin - up electrons in the @xmath408 states and a single spin - up electron in the @xmath409 state , which results in a @xmath410=2 state of mn . the @xmath403 states , being less hybridised with the o ions than the @xmath411 states , are assumed to be more localised , also by strong correlation effects . the @xmath303 electrons are supposed to be more delocalised due to the stronger hybridisation with the o ions . on the other hand , for a cubic mno@xmath325 octahedron , a single electron in the degenerate @xmath411 levels can occupy any linear combination of the @xmath395 and @xmath412 orbitals . this is referred to as the orbital degree of freedom , which plays a very prominent role for the physics of manganites . the ground state of a cubic octahedron is therefore degenerate and , according to the theorem of jahn and teller , this degeneracy will always be lifted by a symmetry reduction , i.e. , a distortion of the octahedron . the mn @xmath407 is therefore strongly jahn - teller active causing local lattice distortions . when the filling of the @xmath303 states is close to 1 or close to a commensurate value , the distortions of the interconnected octahedra can occur in a cooperative way and one is then talking about a collective jahn - teller distortion . this collective ordering of jahn - teller distorted octahedra also corresponds to an orbital ordering . in other words the collective jahn - teller effect is one way to stabilize an orbitally ordered state . there is however another mechanism that can stabilize orbital order , even in the absence of electron - phonon coupling , which is given by the so - called kugel - khomskii exchange@xcite . it was discussed controversially which of the two mechanisms is the main driving force for orbital order in the manganites . a particular example for this controversy was given by the half - doped manganite la@xmath260sr@xmath315mno@xmath334 , which is composed of mno@xmath138 planes separated by a cubic ( lasr)o layer . upon hole doping , more and more mn ions with a formal valence of 4 + ( @xmath413 ) are created , which suppresses the collective jahn - teller effect and causes an increased itineracy of the @xmath303 electrons . the @xmath411 electrons thus play the role of conduction electrons coupled to the localised @xmath410=3/2 @xmath408 electrons . as discussed first by zener@xcite , the mobility of @xmath303 conduction electrons is strongly affected by the spin degrees of freedom : due to the so - called double exchange mechanism , the @xmath303 electrons can delocalise only for ferromagnetic alignment of neighbouring spins . upon increasing the temperature above the ferromagnetic transition temperature the configuration of the spins gets disordered , which causes a strong spin - charge scattering and thus an enhancement of the resistivity close to the curie temperature . applying an external magnetic field at this temperature can easily align the local spins leading to a strong magneto - resistance . this is a simple qualitative explanation of the colossal magneto - resistance observed e.g. in la@xmath260sr@xmath315mno@xmath139 ( x=0.3)@xcite which is not correct on a quantitative level , since it neglects the orbital and lattice degrees of freedom . the relevance of the latter becomes particular evident near @xmath240=0.5 where an electron ordering takes place which has attracted much attention and controversy . several studies have been performed on the compound la@xmath414sr@xmath415mno@xmath334 which is composed of mno@xmath138 planes separated by a cubic ( lasr)o layer . at room temperature the mn sites in this system are all crystallographically equivalent and have an average valency of + 3.5 . at @xmath416 240k a charge disproportionation occurs creating two inequivalent mn sites . originally the two sites were assigned to mn@xmath278 and mn@xmath291 ions , but later on it turned out that the charge difference of the two sites is much less than 1 . below @xmath274=120 k a long range antiferromagnetic ordering of the mn ions into a ce - type structure is realized , originally detected in the related compound la@xmath260ca@xmath315mno@xmath139 by neutron scattering@xcite . it is believed that this complex electronic order is stabilized by several interactions , but it was discussed controversially whether orbital order really exists in this system and , if it does , whether it is driven by a collective jahn - teller distortions or by superexchange interactions . synchrotron - based investigations of orbital order in the manganites were initiated by a seminal study of la@xmath414sr@xmath415mno@xmath334 using hard x - rays at the mn @xmath215 edge@xcite . superstructure reflections were detected , which could not be explained on the basis of the high - temperature crystal structure and were therefore explained in terms of an orbital ordering of the mn @xmath221 states . while there was consensus about the orbital origin of these reflections , the direct observation by resonant hard x - ray scattering was challenged , as the m @xmath215 edge is dominated by dipole - allowed transitions to the 4p states , which are not sensitive to orbital order but rather lattice distortions , as pointed out in a theoretical study @xcite . another theoretical paper@xcite thus suggested the use of rsxs experiments at the mn @xmath222 edges , which are directly related to virtual excitations into mn @xmath221 states and thus to orbital ordering . the calculations predicted different photon energy dependencies of the scattered intensity for orbital ordering being stabilized by superexchange interaction compared to jahn - teller effect . such rsxs studies on single - layered manganite la@xmath414sr@xmath415mno@xmath334 at the mn @xmath222 edges were subsequently reported in refs . @xcite . the first report was given by wilkins et al.@xcite , who studied the orbital and magnetic reflections at wave vectors of @xmath417 and @xmath418 , respectively . we note that the occurrence of orbital order causes the unit cell to quadruple in the @xmath419 plane . edges of la@xmath414sr@xmath415mno@xmath334 at the orbital - order superstructure reflection @xmath420 measured at 63k ( full black line ) compared with theoretical fits ( dashed red lines ) for ( a ) @xmath421 and ( b ) @xmath422 types of orbital ordering . in panel ( c ) a fit for an orthorhombic crystal field is presented . the inset shows the temperature dependence of the orbital order parameter . ( reprinted with permission from @xcite . copyright 2005 , american physical society.),width=7 ] figure[lsmoo_wilkins ] shows the energy dependence of the scattered intensity at the orbital order wave vector @xmath423 through the mn @xmath224 ( near 640 ev ) and @xmath226 edges ( near 650 ev ) which are related to virtual electric - dipole transitions between the mn @xmath225 and @xmath227 core levels to the unoccupied @xmath221 states . figure[lsmom_wilkins ] shows analogous data for the afm order reflection @xmath424 . it can be clearly observed in these figures that the orbital and magnetic reflection exhibit strong resonances at the mn @xmath216 edges . we note that these measurements provide the most direct prove of orbital ordering in these materials . in figures[lsmoo_wilkins ] and [ lsmom_wilkins ] the insets show the temperature dependences of the orbital and magnetic order parameters , respectively . the orbital order parameter decreases continuously with increasing temperature and disappears at @xmath425=230 k. the intensity of the antiferromagnetic reflection shows a similar behaviour but disappears at a lower temperature of @xmath274=120 k. edges of la@xmath414sr@xmath415mno@xmath334 at the magnetic superstructure reflection @xmath426 measured at 63 k ( full black line ) compared with theoretical fits ( dashed red lines ) for ( a ) @xmath421 and ( b ) @xmath422 types of orbital ordering . in panel ( c ) a fit for an orthorhombic crystal field is presented . the inset shows the temperature dependence of the magnetic order parameter . ( reprinted with permission from @xcite . copyright 2005 , american physical society.),width=7 ] in both figures the experimental data are compared with theoretical calculations of the rsxs spectra based on atomic multiplet calculations in a crystal field . in the calculations the crystal fields were modified in such a way as to fit the experimental data . in the panels ( a ) and ( b ) the cubic and the tetragonal crystal field parameters were adjusted for a @xmath421 and @xmath422 type of orbital ordering , respectively . in the panel ( c ) for the @xmath421 type ordering a small orthorhombic crystal field component was added . from the comparison of the experimental data with the theoretical calculations the authors concluded that the orbital ordering below @xmath425 is predominantly of @xmath421 type . the inclusion of a small orthorhombic crystal field component moderately improved the fits , although many details of the complicated experimental lineshape could not be reproduced by the calculation . notwithstanding these difficulties , calculations of the resonance profiles predicted a drastic reduction of the @xmath427 intensity ratio with decreasing tetragonal crystal field . this theoretical prediction was used for the interpretation of experimental results . the experimental temperature dependent rsxs data shown in figure[lsmot_wilkins ] ( a ) and ( b ) indicate that below the orbital ordering temperature @xmath425 the @xmath427 ratio increases , suggesting an increase of the tetragonal crystal field , i.e. , an increase of the jahn - teller distortion with cooling . approaching @xmath274 the ratio @xmath427 increases further and saturates below @xmath274 . edges of la@xmath414sr@xmath415mno@xmath334 . ( a ) integrated intensities of the main features of the orbital ordering @xmath420 reflection at at the @xmath224 edge ( triangles ) and at the @xmath226 edge ( inverted triangles ) . ( b ) intensity ratio of the integrated intensities @xmath427 . ( c ) temperature dependence of the integrated intensity of the magnetic @xmath426 reflection at 643 ev . ( reprinted with permission from @xcite . copyright 2005 , american physical society.),width=7 ] from these findings the authors concluded that there are two separate contributions causing the orbital ordering : a dominant mechanism related to the cooperative jahn - teller distortion of the o ions around the mn@xmath278 ions and the direct magnetic kugel - khomskii superexchange mechanism which further strengthens orbital ordering by short - range afm order and another increase by long - range afm order below @xmath274 . furthermore the authors concluded a strong interaction between orbital and magnetic correlations . nearly at the same time another group has performed similar rsxs experiments on the same compound la@xmath414sr@xmath415mno@xmath334@xcite . the diffraction profile at the mn @xmath216 edges for the the orbital order @xmath420 superstructure reflection were very close to those shown in figure[lsmoo_wilkins ] . in a following paper@xcite , the experimental data were compared to calculations of a small planar cluster consisting of a central active mn site with a first neighbor shell comprising o and mn sites . thus by allowing a hopping between the mn and the o ions an integer charge ordering of the mn ions could be abandoned . in addition to the jahn - teller distortion of the o ions , the spin magnetization of the inactive mn@xmath291 sites were explored as adjustable parameters . the calculations based on a @xmath428 orbital ordering of the @xmath411 electrons reproduced all spectral features of the resonance profile while calculations on a @xmath429 order did not , although the differences were not dramatic enough to provide a definite answer . a further result of the rsxs studies of dhesi et al.@xcite was the detection of a complicated temperature dependence of the individual features of the @xmath224 edge indicating that the complete @xmath224 edge intensities can not be assigned to one order parameter ( e.g. orbital ordering by a collective jahn - teller effect ) . these findings were supported by a following rsxs study of the same compound by staub et al.@xcite . they realized that the features observed in the resonance profiles of the magnetic reflection show the same temperature dependence while those of the orbital order reflection show a strong energy and temperature dependence . the high - energy features of the @xmath216 edges ( in figure[lsmoo_wilkins ] at 647 and 656 ev ) which were assigned to a jahn - teller ordering parameter saturate below @xmath274 . the other features assigned to a direct super - exchange ordering parameter still increases below @xmath274 . from this experimental result together with a polarisation dependence of the resonance profiles they came to the conclusion that there are two interactions leading to the orbital order . it was concluded that the orbital super - exchange interaction dominates over the jahn - teller distortion strain field interaction and drives the transitions . the rsxs results on the manganites described above , demonstrate that the lineshapes at the mn @xmath216 edges display rich spectral features , which contain a lot of information about the underlying order phenomena . to extract this information and to interpret the spectra is , however , challenging as it requires extensive theoretical modeling . here it should be mentioned that in addition various rsxs studies have been performed on double and multi layer manganites@xcite . although largely discussed in connection with 3d transition metal compounds , orbital order is a phenomenon that also occurs in other materials , such as @xmath284 compounds . here , the shielding of the @xmath229 shell by the outer @xmath349 electrons causes a weaker coupling of the electronic degrees of freedom of the @xmath229 electrons to the lattice and hence , allows studying orbital order mechanisms in a less complex situation . in a series of publications on orbital ordering of @xmath229 electrons in @xmath430 compounds mulders et al . could show that rsxs performed at the @xmath431 resonances is not just able to observe orbital order of the @xmath229 electrons but can in addition quantify the contributions of the different higher order multipole moments to the electronic ordering@xcite . in recent years , the interest in materials with coupled magnetic and ferroelectric order has seen an extraordinary revival@xcite . the incentive for numerous applied as well as fundamental studies of these materials is the perspective to manipulate electric polarisation by a magnetic field@xcite or magnetic order by an electric field@xcite . in this context , the notion multiferroics has been coined , which is , of course , more general and applies to a broader range of materials with coupled order parameters . the renewed interest in magnetoelectric materials was initiated by the discovery of strong coupling between magnetic and electric order in tbmno@xmath139 for which one can flip the electric polarisation by application of a magnetic field@xcite . the ferroelectric order in tbmno@xmath139 is connected to the occurrence of a cycloidal magnetic arrangement of the mn spins@xcite , a property that was also found for other members of the perovskite @xmath284mno@xmath139 series , where @xmath284 denotes a rare earth gd , tb , or dy@xcite . different theoretical models were used to explain the strong coupling of magnetic and electric order , however , arriving at the same expression that links the direction of the electric polarisation @xmath432 to the chirality of the spin structure@xcite : @xmath433 in this equation , @xmath434 and @xmath435 denote spins at the sites @xmath436 and @xmath437,and @xmath438 is the unit vector connecting the two sites . other rare earth manganites of the form @xmath284mn@xmath138o@xmath439 also exhibit multiferroic behaviour , with reversible switching of the ferroelectric polarisation by a magnetic field , as found in tbmn@xmath138o@xmath439@xcite . the coupling of magnetic and electric order in these latter materials is believed to be of a different origin , involving magnetoelastic coupling , ionic displacements or electronic ferroelectric polarisation . the majority of the recently studied magnetoelectrics are manganites , however , also cuprates , among them cupric oxide cuo itself have been found to be multiferroic@xcite , with a particularly high ordering temperature of @xmath440 k. as the multiferroic materials are transition metal oxides with complex order of the magnetic moments , neutrons have initially been the method of choice for their magnetic characterization@xcite . rsxs , on the other hand , is perfectly matching these materials as both the @xmath221 and @xmath229 electronic states of the transition metals as well as the @xmath218 states of oxygen can be addressed via resonant dipole transitions@xcite . thus , the method has provided substantial contributions to the understanding of the materials , also because circularly polarised x - rays are readily available at synchrotron radiation sources . by the magnetic structure factor , this provides direct access to a possible handedness , as present in helical@xcite or cycloidal arrangements@xcite . in addition , frustrated magnetic interactions result in rather long - period superstructures that can thus be accessed at soft x - ray wavelengths . [ scale=0.5 , trim= 0 0 0 0 , angle= 0 ] wilkins2009 - 1 tbmno@xmath139 is the most studied material of the @xmath284mno@xmath139 series . for this material , magnetic neutron diffraction provided the first experimental observation of a cycloidal magnetic structure in connection with ferroelectric polarisation@xcite . from the observation of components of the mn @xmath221 magnetization along both the @xmath362 and the @xmath367 crystallographic axis in the ferroelectric phase a non - collinear structure was concluded that together with a propagation along the @xmath362 axis has the proper chirality to induce a ferroelectric polarisation along the @xmath367 axis according to eq.[eqnpolarisation ] . the tb @xmath229 structure , on the contrary , remained collinear , as only a component along the @xmath441 axis was found , pointing out the essential role of the mn ordering for the ferrolectric properties of the material . the mn spin cycloid leads to substantial circular dichroism in x - ray scattering as first found in a non - resonant hard x - ray study using circularly polarised light , in this way directly linking to the handedness of the magnetic structure@xcite . while it was generally agreed that ferroelectricity is essentially due to a mn @xmath442 cycloid , further details of the magnetic structure could be elucidated taking advantage of the very high sensitivity of resonant magnetic soft x - ray diffraction . a refinement of the mn magnetic structure was obtained from resonant soft x - ray diffraction at the mn @xmath226 resonance@xcite . here , a magnetic f - type reflection along [ 0 k 0 ] was observed that would not be allowed for strictly a - type magnetic ordering in the perovskite structure of these materials@xcite . from modeling the azimuthal dependences of the diffraction intensities measured with linearly polarised x - rays ( figure[wilkins2009 - 1 ] ) it was concluded that a canted magnetic component along the @xmath367 axis must be present in all ordered phases of the material , rendering also the previously assumed sinusoidal phase non - collinear . [ scale=0.8 , trim= 0 0 0 0 , angle= 0 ] jang2011 - 1 further details of the magnetic structure were revealed by the application of circular polarised x - rays at the mn @xmath226 resonance ( see figure[jang2011 - 1 ] ) . the magnetic ( 0 @xmath443 0 ) reflection exhibits pronounced circular dichroism , thus representing an f - type magnetic cycloid that is coupled to the previously observed a - type cycloid@xcite . this clear dependence of the scattered intensity on the helicity of the x - rays is not to be confused with x - ray circular dichroism in absorption , as it is due to the magnetic structure factor@xcite and not an effect of the resonant form factor . the occurrence of an f - type magnetic cycloid is explained by antisymmetric exchange , the so - called dzyaloshinskii - moriya interaction that favours canting of the magnetic moments towards the @xmath367 axis@xcite . figure[jang2011 - 1 ] also indicates another application of resonant soft x - ray diffraction to multiferroic materials . with the help of an external electric field , the electric polarisation of tbmno@xmath139 is switched from pointing along -c to pointing along c. according to eq.[eqnpolarisation ] , this corresponds to magnetic cycloids of opposite chirality as depicted on the right panel of figure[jang2011 - 1 ] . therefore , the asymmetry in the scattering of circular polarised light changes sign , as can be inferred from the changing response to left ( @xmath222 ) and right ( @xmath444 ) circularly polarised light . hence , besides general refinements of the magnetic structures , rsxs can provide a useful contrast mechanism to identify and study microscopic details of the magnetic structure upon the application of external fields . the first demonstration of such an application was provided by bodenthin et al . for the case of ermn@xmath138o@xmath439@xcite . the contrast was obtained at the mn @xmath224 resonance using the commensurate ( 1/2 0 1/4 ) reflection , which is closely connected to the ferroelectric phase in this material . as shown in figure[bodenthin2008 - 1 ] ( upper panel ) , differences in the intensities of this peak are observed with an electric field applied compared to the situation without field . the differences are in fact small but significant , as shown in the inset , and the peak intensity reveals hysteretic behaviour , in this way demonstrating the manipulation of the magnetic structure by an external electric field@xcite . [ scale=1 , trim= 0 0 0 0 , angle= 0 ] bodenthin2008 - 1 the studies reported so far were mainly concerned with the magnetic order of the mn spins . however , also the ordering of the @xmath229 moments is expected to play a role for the ferroelectric properties of the @xmath284mno@xmath139 compounds . this may not be so important for tbmno@xmath139 , but the situation is definitely different in case of dymno@xmath139 . here , the ferroelectric polarisation is much larger than observed for tbmno@xmath139@xcite , and its temperature dependence is closely linked to @xmath229 ordering@xcite . by tuning the photon energy to the @xmath284 @xmath217 resonances , one takes advantage of the element - selectivity of the method that allows to study the @xmath229 magnetic ordering separately . using circularly polarised x - rays , it was shown that the @xmath229 moments in dymno@xmath139 themselves form a magnetic cycloid with the proper chirality to support ferroelectric polarisation along the @xmath367 axis according to eq.[eqnpolarisation ] , the same direction , as also promoted by the mn cycloid in the material@xcite . this is demonstrated in figure[schierle2010 - 1](c ) where the upper panels display the incommensurate ( @xmath445 ) diffraction peak , recorded at the dy @xmath231 resonance using left and right circularly polarised light . obviously , a large intensity difference is observed in the ferroelectric phase at 10 k , which is absent in the paraelectric phase at 20 k. this latter phase is characterized by a collinear sinusoidal magnetic modulation that exhibits no handedness . figure[schierle2010 - 1 ] also suggests an interesting approach to manipulate ferroelectric domains : as the photoelectric effect induces local charging of the sample surface in the highly insulating dymno@xmath139 , a radial electric field is induced that creates two regions of opposite electric polarisation along the crystallographic @xmath367 axis ( cf . figure[schierle2010 - 1](a ) ) . they are characterized by magnetic cycloids of opposite handedness . when the sample is scanned along @xmath242 , i.e. , parallel to the @xmath367 axis , the asymmetry of the magnetic diffraction signal , defined as @xmath446 , changes its sign as seen in figure[schierle2010 - 1](c ) . [ scale=0.4 , trim= 0 0 0 0 , angle= 0 ] schierle2010 - 1 the contrast obtained by the circular dichroic asymmetry in scattering could eventually be used to image multiferroic domains in the material as shown in figure[schierle2010 - 2 ] . the domain pattern was imprinted by the x - ray beam either while cooling through the paraelectric / ferroelectric phase transition ( position 1 ) or during heating close to the transition temperature in order of facilitate switching of the electric polarisation ( positions 2 and 3 ) . using the circular dichroic contrast of the incommensurate ( @xmath445 ) diffraction peak , the resulting domain pattern was imaged by scanning the sample@xcite . [ scale=0.4 , trim= 0 0 0 0 , angle= 0 ] schierle2010 - 2 while it was argued that the electric field generated by the local charging via the photoelectric effect@xcite is inducing the ferroelectric domains , later studies of ymn@xmath138o@xmath439 and ermn@xmath138o@xmath439 provided evidence that small currents flowing through the samples upon the application of the electric fields may also play a role@xcite . after essentially considering the @xmath221 and @xmath229 magnetic ordering in the multiferroic @xmath284 manganites , recent studies have eventually also elucidated the role of oxygen in multiferroics . as discussed before , an incommensurate magnetic cycloid was the driving force for ferroelectricity in @xmath284mno@xmath139 compounds . for tbmn@xmath138o@xmath139 , on the other hand , the magnetic ordering connected to the ferroelectric phase is commensurate@xcite , and almost collinear . nevertheless , strong magnetoelectric effects are observed@xcite . a possible mechanism involving charge transfer between mn@xmath278 and o was suggested to account for the large electric polarisation of the material@xcite . resonant soft x - ray diffraction at the o@xmath215 edge of tbmn@xmath138o@xmath439 provided clear evidence for a magnetic polarisation of the oxygen along with the commensurate antiferromagnetic ordering in the ferroelectric phase@xcite . figure[beale2010 - 1 ] displays the intensity change of the antiferromagnetic ( 1/2 0 1/4 ) reflection when the photon energy is scanned across the o @xmath215 edge , revealing resonant behaviour at energies that correspond to unoccupied oxygen states characterized by hybridisation with mn 3@xmath253 states . model calculations of this resonance behaviour with and without mn spin supported the interpretation that in fact a magnetic polarisation at the o site is observed that is antiferromagnetically correlated@xcite . these results support the above mentioned theoretical models@xcite that the o magnetic polarisation is key to the magnetoelectric coupling mechanism . [ scale=0.8 , trim= 0 0 0 0 , angle= 0 ] beale2010 - 1 a similar result was obtained by partzsch et al.@xcite , who compared the temperature dependence of the ( 1/2 0 1/4 ) reflection in the isostructural compound ymn@xmath138o@xmath439 both at the mn @xmath224 resonance and the o @xmath215 edge ( see figure[partzsch2011 - 1 ] ) . while the behaviour at the mn @xmath224 resonance closely resembles the magnetic order parameter as measured by neutron diffraction ( top ) , the temperature dependence at the o@xmath215 edge is rather different and follows the spontaneous polarisation in the material , i.e. , tracking the ferroelectric order parameter ( bottom ) . this provides evidence that the covalency of the mn and o atoms plays a central role for the ferroelectric polarisation in ymn@xmath138o@xmath439 . density - functional calculations indeed show that the spin order drives a redistribution of the valence band electrons resulting in a purely electronic contribution to the ferroelectric polarisation@xcite . [ scale=0.8 , trim= 0 0 0 0 , angle= 0 ] partzsch2011 - 1 a further result in this context is to be mentioned here : a study of tbmn@xmath138o@xmath439 revealed the existence of an internal field defined by @xmath447 . a quantity that is related to that expression and that was measured by resonant diffraction follows the same temperature dependence as the ferroelectric polarisation@xcite . [ scale=0.8 , trim= 0 0 0 0 , angle= 0 ] wu2010 - 1 the majority of multiferroic materials studied by rsxs are manganites . the recent discovery that cupric oxide cuo is a multiferroic even at comparably high temperatures around 230k@xcite , has triggered magnetic rsxs studies of this compound . rsxs has been used to identify ferroelectric nanoregions by diffuse scattering@xcite ( see figure[wu2010 - 1 ] ) . an anomalous memory effect for the direction of the electric polarisation in the commensurate - incommensurate magnetic transition has been detected that coincides with the ferroelectric transition . from the rsxs results , incorporated with simulations of diffuse scattering , it was proposed that a preserved spin handedness in the multiferroic nanoregions is responsible for this memory effect in the magnetically induced ferroelectric properties of cuo . copper oxide was also recently studied by scagnoli et al.@xcite . rather than the ferroelectric properties , this investigation was focused on the low - temperature phase of cuo . polarisation analysis as well as particular dependencies of diffracted intensities on incident circular polarisation revealed the existence of so - called orbital currents in the material . the increasing number of studies using rsxs is closely connected with the high intensity and the particular properties of the x - ray radiation available at modern synchrotron radiation sources . the control of the incident x - ray polarisation and its use was already discussed throughout this article . another characteristic is the high transverse coherence of the radiation that can be provided by undulator sources . together with the longitudinal coherence given by the high energy resolution , synchrotrons can provide fully coherent x - ray beams for correlation spectroscopy as well as imaging . and finally , synchrotron radiation has a well - defined time structure that can be used for time - resolved experiments . recent development have pushed the resolution into the femtosecond region , which allows to access the time scales involved in the ordering phenomena discussed throughout this article . both , high coherence and femtosecond time resolution is eventually provided by free - electron lasers , which will permit both spatially and time - resolved studies . as the scope of this account is to provide a more general overview of rsxs , it can not discuss these latter developments in depth . coherent scattering using soft x - rays , in particular , is already well - developed , and a complete description of this field is beyond the scope of this article . time - resolved experiments using rsxs , on the other hand , are scarce , even though studies related to the physics of complex order in correlated materials have been carried out with femtosecond lasers already at length . but also here , the instrumental development is on the way , and many exciting result can be expected in the near future . while a complete description of these fields can not be accomplished in this article , still , the last section is devoted to a short outlook to what can be achieved using the high sensitivity to magnetic and electronic ordering in the soft x - ray regime , starting with a classical method of x - ray diffraction that only recently was extended to the soft x - ray regime . resonant magnetic x - ray diffraction is a method essentially applied to single - crystalline material as discussed throughout this article . and in fact , magnetic x - ray diffraction from powders is very difficult because of the weak signal intensity compared to charge scattering and fluorescence . for the determination of magnetic structures , magnetic neutron powder diffraction is the method of choice , because the available momentum transfer allows to measure a large number of reflections required for structural analysis and refinement . nevertheless , resonant x - ray diffraction may be useful in cases , where the synthesis of new materials does not yield single crystals , even if the full potential of the method like azimuth dependent studies can not be used . [ scale=0.8 , trim= 0 0 0 0 , angle= 0 ] garcia - fernandez2008 - 1 even in the hard x - ray regime , resonant magnetic powder diffraction studies are scarce . a recent investigation of gdni@xmath138ge@xmath138 at the comparably weak gd @xmath224 edge used polarisation analysis and fluorescence background suppression to prepare the magnetic signal @xcite . already a decade earlier , resonant magnetic diffraction was demonstrated for the case of uo@xmath138 @xcite , exploiting the huge enhancement at the u @xmath233 resonance , a situation comparable to soft x - ray resonances discussed in this article . however , these two studies essentially demonstrated the feasibility of resonant magnetic x - ray powder diffraction , rather than providing new insights into the microscopic properties of the materials . the situation has changed with recent rsxs studies @xcite that were carried out at the swiss light source . figure [ garcia - fernandez2008 - 1 ] ( left panel ) shows a two - dimensional image of the diffraction pattern obtained from a powder sample of smbamn@xmath138o@xmath325 . the data were obtained at the mn @xmath224 resonance using an in - vacuum x - ray ccd camera and represent a section of the debye - scherrer ring corresponding to orbital order in the material . vertical integration yields a clearly discernible ( 1/4 1/4 0 ) diffraction peak . diffraction from a powder , of course , provides an average over all azimuthal directions . therefore , an important input for the study of the type of order by measuring azimuthal dependences is not available . but the energy dependence of the diffraction peak allows to further characterize the orbital order in this layered manganite material as @xmath448 type , compared to @xmath449 present in single - layer manganites @xcite . further , the temperature dependence of the diffraction peak was studied ( see figure [ garcia - fernandez2008 - 1 ] , right panel ) , which provides a measure of the orbital order parameter . below the ordering temperature of @xmath450k , a second orbital phase transition around 210k is observed that is also reflected in resistivity and magnetization data . the intensity increase of the orbital reflection is quantitatively reproduced by a structure factor calculation that provides a scenario involving a change of the orbital stacking at the second transition @xcite . [ scale=0.8 , trim= 0 0 0 0 , angle= 0 ] bodenthin2011 - 1 another example of resonant powder diffraction was given by a study of rare - earth nickelates @xmath284nio@xmath139 . the interest in these perovskite nickelates stems from metal - insulator transitions with transition temperatures that can be tuned by the ionic radius of the rare - earth ion and that are accompanied by magnetic and charge order . figure [ bodenthin2011 - 1 ] displays the intensity of the ( 1/2 0 1/2 ) magnetic diffraction peak across the ni @xmath216 resonance for a series of @xmath284nio@xmath139 compounds , with the ionic radius of the @xmath284 decreasing from pr to y. again , data were obtained with an x - ray ccd camera and subsequent integration across the debye - scherrer ring . the energy dependencies shown in fig . [ bodenthin2011 - 1 ] reveal a striking similarity , with no energy shifts and little variation of the spectral shape . this shows that the electronic structure of the materials essentially stays the same across the series . more specifically , model calculations of the energy dependencies revealed that the change in the charge disproportionation between inequivalent ni sites connected with charge order in the material can not be larger than 0.05 electron charges despite the substantial change in the metal - insulator transition temperature @xcite . while the application of resonant soft x - ray diffraction to powder materials is still rare , these examples show the usefulness of the method beyond a mere demonstration of feasibility . conventional x - ray diffraction investigates predominantly periodic structures . the diffraction patterns are transformed into real - space atomic maps . the determination of non - periodic nanoscale structures by x - ray scattering is much more difficult since the inversion of the intensity profile suffers from the intrinsic loss of phase information . in standard experiments the diffuse scattering intensity around a bragg peak yields some information on the disorder of a system , as also discussed in section[sec : casestudies ] in connection with magnetic roughness . by limiting the diffraction volume to the coherence volume , extra information about the structure is available . in this case the measured intensity profile has strong fringes and which represent a specific microscopic configuration of the sample , which are called `` speckles '' . speckle patterns can be recorded as a function of time , in this way probing the time - dependence of correlations in a material . in holography , on the other hand , a reference beam is used to interfere with the light scattered from the sample . this allows retrieval of the scattering phase and hence reconstruction of the scattering object by a simple fourier transform . scattering of coherent light is a technique that is well established in the optical regime , as optical lasers providing highly coherent light are available for a long time . recent years have seen a transfer of the method to the x - ray range@xcite , which has become possible with the advent of undulator radiation sources at third - generation synchrotrons that provide x - ray beams with sufficient degree of coherence . a high degree of longitudinal coherence of x - rays is anyway provided by monochromatization , but only undulators with their high brilliance provide sufficient transverse coherence as well , resulting in a highly coherent x - ray beam . the full power of coherent x - ray scattering will be exploited after the installation of x - ray free electron lasers . in the meantime , both , correlation spectroscopy as well as holographic imaging were successfully carried out using soft x - rays , exploiting the magnetic sensitivity at resonance . a first experiment that transferred the method of coherent scattering into the soft x - ray regime was carried out by price et al . @xcite , who studied fluctuations in smectic liquid crystals . these experiments , however , did not exploit electronic resonances in the materials . the power of coherent scattering in rsxs for the imaging of magnetic domains has been demonstrated in a landmark experiment by eisebitt et al.@xcite in which the meandering maze - like magnetic domains in co / pt multilayers with alternating up and down magnetization perpendicular to the layer surface were investigated . the magnetic contrast has been achieved by exploiting the magnetic dichroism in resonance at the co @xmath216 edges using circularly polarised photons . a schematic picture of the experimental set - up is shown in figure[fig : hellwig2006 ] . the coherent part of the undulator radiation is selected by an aperture of micrometer diameter and hits a specially designed sample structure : co / pt multilayers are deposited on a silicon nitride membrane that is transparent at soft x - ray energies . the imaging pinhole is of micrometer diameter . the sample also comprises the nanometer - sized reference pinhole that defines a reference beam . the x - rays scattered by the magnetic structure interfere with the reference beam and form a holographic interference pattern , i.e. , the speckle pattern , which encodes the image of the magnetic nanostructure . the final image derived by a fourier transform and shown in the inset in the upper right corner of figure[fig : hellwig2006 ] demonstrates the detection of the labyrinth stripe domain pattern which is characteristic for magnetic systems with perpendicular anisotropy . magnetic field dependent studies revealed a transformation of the labyrinth domain into isolated domains and , finally near saturation , into isolated bubble domains with increasing field . following this first demonstration of lens - less imaging with soft x - rays , further experiments improved the efficiency of the method @xcite and demonstrated that the resolution of the method can be substantially enhanced beyond the fabrication limit of the reference structure @xcite . it was also shown that lens - less imaging can even be accomplished without reference beam by oversampling methods @xcite . these exploring experiments demonstrated that with photon sources with shorter pulses and higher brightness , such as free electron lasers , there will be a pathway to study dynamics of magnetic structures in real time to foster progress in rapid magnetic reading and writing processes . the availability of free electron lasers in the soft x - ray region with a spectacular increase of some nine order of magnitude in peak brilliance in combination with femtosecond time resolution and high coherence@xcite allow now ground breaking , completely new experiments on the dynamics of magnetic nanoscaled structures . [ width=7cm , angle=90 ] hellwig2006 as one of the first examples , we present the investigation of meandering maze - like domains in co / pt multilayers with alternating up and down magnetization perpendicular to the layer surface ( see above ) . speckle patterns recorded at the free electron laser flash with a time - resolution of 30 fs are depicted in figure[gutt2010 - 1]@xcite . magnetic contrast for the domains was obtained by using photons at the co@xmath451 edge and a geometry which maximizes the second term in eq.[eqn : hannon ] . information on the mean value of the size of the domains could be obtained from a single 30 fs free electron laser pulse . in this work it was also demonstrated that the speckle signal for subsequent pulses provided information on the magnetization dynamics as a function of time . these pioneering results point to the possibility of ultrafast magnetization studies down to the 100 fs timescale . [ scale=0.73 , trim= 0 0 0 0 , angle= 0 ] gutt2010 - 1 all these experiments are carried out in transmission geometry , with sample and reference beam scattered from the same substrate , which inherently provides a rigid coupling between sample and reference . in these cases , samples require a special design with particularly thin films in the nanometer range that remain transparent at resonance energies in the soft x - ray regime . in reflection geometry , the scope of coherent scattering can be substantially broadened towards application to surfaces of bulk materials as discussed throughout this paper . while this is more demanding , a successful demonstration at 500 ev photon energy using a test structure was achieved @xcite . [ scale=1 , trim= 0 0 0 0 , angle= 0 ] konings2011 - 1 another example of coherent scattering carried out in reflection geometry is given by correlation spectroscopy across the phase transition of helical magnetic ordering in a ho thin film@xcite as shown in figure[konings2011 - 1 ] . the helical magnetic structure of ho leads to a pronounced magnetic diffraction peak that can be probed by x - rays at the ho @xmath231 resonance@xcite ( cf . [ sec : long - periodafm ] ) . at 52k , well below the ordering temperature of the film ( @xmath452k ) , the speckle pattern connected with this diffraction peak is static , as shown on the top panel . it shows that magnetic fluctuations at this temperature are practically absent . upon heating to 70k close to the ordering temperature , the speckle patterns change with time , indicating the development of magnetic fluctuations at the phase transition . a close inspection of the speckle patterns as well as the time - averaged intensity distribution on the bottom of figure[konings2011 - 1 ] shows that some parts still remain fixed . hence , fluctuations at the phase transition do not affect the whole sample , revealing a non - ergodic behaviour of the system . an analogous experiment on the orbital order of the half - doped manganite pr@xmath414ca@xmath414mno@xmath139 revealed similar behaviour near the phase transition , with both pinned and slowly fluctuating orbital domains @xcite , the examples of coherent scattering described here open a broad field of applications to the materials and their various electronic ordering phenomena discussed throughout this paper . the development of lasers with pulse duration in the femtosecond range opened a new field to study ultrafast processes . magnetic and other complex ordering phenomena can be studied using photons in the optical range , as changes of the symmetry in a system upon ordering are often reflected in the optical properties of the material , particularly involving second - order processes that are excited using intense laser light . thus , optical wavelengths provide contrast to study the ultrafast dynamics connected with the decay of ferromagnetic or more complex ordering phenomena in transition metal oxides@xcite . however , as soft x - ray resonances provide highest sensitivity to spin , orbital , and charge order in combination with element selectivity , the development of free electron lasers in this energy range represents a large step in this field . in addition to free electron lasers , also slicing facilities at third - generation synchrotron sources provide the ultrashort photon pulses required for time - resolved x - ray diffraction studies with femtosecond time resolution . a setup installed at the swiss light source provides ultrashort x - ray pulses for hard x - ray diffraction . using structural superlattice reflections associated with charge and orbital order , it was possible to study the dynamics in la@xmath453ca@xmath454mno@xmath139@xcite . [ scale= 0.57 , trim= 0 0 0 0 , angle= 0 ] holldack2010 - 1 a slicing beamline covering the soft x - ray region and providing linear as well as circularly polarised light is installed at bessy ii at the helmholtz - zentrum berlin@xcite . here , studies of demagnetization dynamics could be extended to the soft x - ray region , with the particular advantage of element selectivity and the possibility to access spin and orbital moment separately . first experiments were concerned with x - ray absorption only@xcite , but recently , also rsxs experiments with fs time resolution were carried out successfully . for that purpose , eute films were used that were already discussed in section[magneticdepth ] , as they exhibit magnetic bragg peaks of unprecedented quality at the eu - m@xmath439 resonance@xcite . time resolved experiments were carried out using a pump - probe scheme as illustrated in figure[holldack2010 - 1 ] . the system was excited by a laser using a wavelength of 400 nm that is sufficiently small for an excitation across the band gap of the material and corresponds approximately to a @xmath455 excitation . the photon energy of the probe pulse was tuned to the eu @xmath231 resonance and the scattering geometry was chosen to correspond to the ( @xmath456 ) afm bragg peak that occurs almost at a scattering angle of @xmath457 . [ scale=0.73 , trim= 0 0 0 0 , angle= 0 ] holldack2010 - 2 figure[holldack2010 - 2 ] displays the intensity of the afm diffraction peak as a function of the time delay between the optical pump and the x - ray probe pulse . interestingly , a very fast decay of the afm order is observed , which is faster than ( @xmath458 ) fs . this is quite remarkable for a semiconductor , as the fast demagnetization mechanisms involving conduction electrons known from metals are not available . figure[holldack2010 - 2 ] also displays a rocking scan through the afm diffraction peak recorded with a 100 fs probe pulse without laser pumping . this demonstrates that in this type of experiments the capabilities of diffraction beyond recording the mere peak intensity can be exploited to study the temporal evolution of spatial correlations in the ordered phase . while experimental facilities at the free - electron laser in hamburg ( flash ) at desy ( deutsches elektronen - synchrotron ) and the linac coherent light source ( lcls ) at the slac national accelerator laboratory have paved the way to ultrafast studies , time - resolved scattering experiments for the study of correlated materials using soft x - rays are presently still scarce . an experiment recently carried out at flash is concerned with the decay of orbital order in the prototypical correlated material magnetite fe@xmath139o@xmath334@xcite(cf . section[magnetite ] ) . in figure[pontius2011 - 1 ] the time - dependent intensity of the ( 0 0 1/2 ) reflection , measured with photons at the o@xmath215 edge , is shown as a function of time after a femtosecond laser pump pulse , for several pump fluences . since the intensity of the peak in resonance is directly related to the unoccupied density of states , this intensity provides direct information about the electronic order and hence of the melting of charge / orbital order induced with the infrared pulse . an ultrafast melting of the charge / orbital order has been found in the rapid decrease of the intensity of the superstructure peak . the persistent intensity of the peak after 200 ps even for pump fluences that correspond to sample heating above the verwey transition ( inset of figure[pontius2011 - 1 ] ) was interpreted in terms of a transient phase characterized by the existence of partial charge / orbital order which has not been observed in equilibrium . in this way the time - dependent rsxs experiment have demonstrated that important information on the dynamics of phase transitions in correlated systems can be obtained . [ scale=0.4 , trim= 0 0 0 0 , angle= 0 ] pontius2011 - 1 a third time - resolved experiment , albeit only with picosecond resolution , was reported recently from diamond light source@xcite . time resolution was achieved here by gating the synchrotron radiation pulses synchronized with the laser pulses . the material studied was la@xmath414sr@xmath415mno@xmath334 , which is characterized by spin and orbital order ( cf . section[sec : manganites ] ) . these two types of order are represented by the ( @xmath459 ) afm bragg peak , and the ( @xmath460 ) orbital diffraction peak , respectively . figure[ehrke-1 ] displays the evolution upon laser excitation of the two diffraction peaks , revealing that spin order can be quickly destroyed , while the orbital order persists . the results were interpreted in terms of an intermediate transient phase in which the spin order is completely removed by the photoexcitation while the orbital order is only weakly perturbed . in this way the time - dependent rsxs experiments have demonstrated that it is possible to separate the spin dynamics determined by the electronic structure from jahn - teller contributions which are more related to the lattice dynamics . [ scale=0.2 , trim= 0 0 0 0 , angle= 0 ] ehrke2011 - 1 during the last two decades rsxs has established itself as a powerful tool in modern solid state physics to investigate magnetic , orbital and charge ordering phenomena associated with electronic degrees of freedom . since the technique is coupled to specific core excitations it is element specific . the techniques combines diffraction methods with x - ray absorption spectroscopy and therefore delivers structural nanoscale information on the modulation of the valence band electron density , on the bond orientation , i.e. , orbital ordering , and on the spin density . in particular rsxs has provided important structural and spectroscopic information on 3@xmath253 transition metal compounds and 4@xmath100 systems which comprise interesting solids such as high-@xmath305 superconductors , charge density compounds , giant and colossal magneto - resistance systems which may be important for spintronics . since rsxs has a finite probing depth , the method is not only suited for the investigation of bulk properties : rsxs provides also important structural information on small crystals , ultra thin films , on multilayer systems and interfaces , and on all kind of nanostructures . the future of the technique will strongly depend on the instrumental development of the photon sources and the diffractometers . variable polarisation is already realized at various beamlines at synchrotron radiation facilities . fast switching of the polarisation will help to detect small polarisation effects . in several rsxs diffractometers the application of suitable polarisation analyzers will allow to extend investigations of magnetic and orbital ordering in a large class of interesting materials . arrays of channel plate detectors or charge coupled devices ( ccd ) will enable to perform faster diffraction experiments . sample environment will play an important role in future rsxs experiments : the possibility to apply high magnetic fields and low temperatures to the sample will offer interesting studies on new ground states in correlated systems . furthermore , the development of intense coherent soft x - ray sources by the development of fourth generation synchrotron radiation facilities ( energy recovery linacs and free electron lasers ) will allow to reveal , element specific and with nanometer spacial resolution , complicated non - periodic lattice , magnetic , and charge structures . here important issues will be phenomena related to fluctuation near phase transitions , chemical reactions , defects and precipitations in materials etc . finally , with the advent of free - electron laser photon sources in the soft x - ray region , interesting time - dependent pump - probe studies on the structural evolution of photo - excited solids will be feasible on a large scale . possibly movies of chemical reactions , studies on matter under extreme conditions , watching magnetic spin flips in real time , and unraveling the functional dynamics of biological materials will be feasible . the future of rsxs will be not only related to bulk studies of the charge , orbital , and spin order in transition metal compounds . rather the focus will be also on similar studies in thin films , all kind of artificial nanostructures and in particular on the structure and the electronic properties of interfaces . therefore those rsxs experiments will strongly depend on the development of preparation stations for artificial nanostructures , e.g. by laser deposition or molecular beam epitaxy . one focus of future rsxs experiments will still be related to complex bulk properties of correlated materials as a function of composition , temperature , and magnetic field . there , mainly charge , spin , and orbital ordering as well as jahn - teller distortions will be studied in correlated transition metal and rare earth compounds . new bulk experiments could be considered in the fields of vortex lattices of superconductors since the charge inside and outside of vortexes is different . since similar to the o @xmath215 edges in tm oxides , the c @xmath215 edges in soft matter are strongly dependent on the chemical bonding of c atoms , rsxs will develop to a great tool for investigations of nanoscale structures and spacial distributions in polymer blends , multiblock copolymers , polymer solutions , lipid membranes , colloids , micelles , and polymeric biological composite particles . furthermore , using coherent light sources , disordered systems will be studied applying speckle methods . finally rsxs will be extended to the time domain via setting up pump - probe experiments to study the dynamics of phase transitions . the other focus of future rsxs will be certainly related in the analysis of structural and electronic properties of artificial structures on the nanometer scale which have great promise for novel functionality . in particular , rsxs will help to resolve the problem to characterize the electronic structure of oxide hetero - epitaxial devices in which , as already discussed above , at the interfaces between two band insulators metallicity and even superconductivity has been detected . at present it is not decided whether this observation is due to the polar nature of the layers or due to o vacancies . future rsxs experiments will certainly help to resolve open questions in this field . thus one can be confident that rsxs will have a great impact on a vast variety of scientific fields ranging from solid state physics across materials science and chemistry to biology and medicine . the authors thank j. hill , u. staub , and s. wilkins for helpful suggestions . 100 d. g. hawthorn , k. m. shen , j. geck , d. c. peets , h. wadati , j. okamoto , s .- w . huang , d. j. huang , h .- j . lin , j. d. denlinger , ruixing liang , d. a. bonn , w. n. hardy , and g. a. sawatzky . resonant elastic soft x - ray scattering in oxygen - ordered @xmath464 . , 84:075125 , 2011 . m. w. haverkort , n. hollmann , i. p. krug , and a. tanaka . symmetry analysis of magneto - optical effects : the case of x - ray diffraction and x - ray absorption at the transition metal @xmath465 edge . , 82:094403 , 2010 . h. ott , c. schler - langeheine , e. schierle , a. yu . grigoriev , v. leiner , h. zabel , g. kaindl , and e. weschke . magnetic x - ray scattering at the @xmath467 absorption edge of @xmath468 . , 74(9):094412 , 2006 . e. schierle , v. soltwisch , d. schmitz , r. feyerherm , a. maljuk , f. yokaichiya , d. n. argyriou , and e. weschke . cycloidal order of @xmath229 moments as a probe of chiral domains in @xmath470 . , 105:167207 , 2010 . u. englisch , h. rossner , h. maletta , j. bahrdt , s. sasaki , f. senf , k.j.s . sawhney , and w. gudat . the elliptical undulator @xmath471 and its monochromator beam - line for structural research on nanomagnets at @xmath472 . , 467 - 468(0):541 544 , 2001 . h. a. drr , e. dudzik , s. s. dhesi , j. b. goedkoop , g. van der laan , m. belakhovsky , c. mocuta , a. marty , and y. samson . chiral magnetic domain structures in ultrathin @xmath477 films . , 284:21662168 , 1999 . hatton , s.b . wilkins , t.a.w . beale , t.k . johal , d. prabhakaran , and a.t . resonant soft x - ray scattering - a new probe of charge , spin and orbital ordering in the manganites . , 290:891897 , 2005 . t. takeuchi , a. chainani , y. takata , y. tanaka , m. oura , m. tsubota , y. senba , h. ohashi , t. mochiku , k. hirata , and s. shin . an ultrahigh - vacuum apparatus for resonant diffraction experiments using soft x rays ( h@xmath485 ) . , 80 , 2009 . d. g. hawthorn , f. he , l. venema , h. davis , a. j. achkar , j. zhang , r. sutarto , h. wadati , a. radi , t. wilson , g. wright , k. m. shen , j. geck , h. zhang , v. novk , and g. a. sawatzky . an in - vacuum diffractometer for resonant elastic soft x - ray scattering . , 82(7):073104 , 2011 . t. a. w. beale , t. p. a. hase , t. iida , k. endo , p. steadman , a. r. marshall , s. s. dhesi , g. van der laan , and p. d. hatton . rasor : an advanced instrument for soft x - ray reflectivity and diffraction . , 81:073904 , 2010 . s. b. wilkins , m. p. m. dean , jrg fink , markus hcker , j. geck , v. soltwisch , e. schierle , e. weschke , g. gu , s. uchida , n. ichikawa , j. m. tranquada , and j. p. hill . comparison of stripe modulations in @xmath486 and @xmath487 . , 84:195101 , 2011 . l. paolasini , c. detlefs , c. mazzoli , s. wilkins , p. p. deen , a. bombardi , n. kernavanois , f. de bergevin , f. yakhou , j. p. valade , i. breslavetz , a. fondacaro , g. pepellin , and p. bernard . . , 14:301312 , 2007 . franz schfers , hans - christoph mertins , andreas gaupp , wolfgang gudat , marcel mertin , ingo packe , frank schmolla , silvia di fonzo , grard soulli , werner jark , richard walker , xavier le cann , ralf nyholm , and mikael eriksson . soft - x - ray polarimeter with multilayer optics : complete analysis of the polarization state of light . , 38:40744088 , 1999 . v. scagnoli , u. staub , y. bodenthin , r. a. de souza , m. garca - fernndez , m. garganourakis , a. t. boothroyd , d. prabhakaran , and s. w. lovesey . observation of orbital currents in @xmath490 . , 332:696698 , 2011 . kao , c. t. chen , e. d. johnson , j. b. hastings , h. j. lin , g. h. ho , g. meigs , j .- brot , s. l. hulbert , y. u. idzerda , and c. vettier . dichroic interference effects in circularly polarized soft - x - ray resonant magnetic scattering . , 50:95999602 , 1994 . j. w. freeland , v. chakarian , y. u. idzerda , s. doherty , j. g. zhu , j .- park , and c .- c . kao . identifying layer switching in magnetic multilayers with x - ray resonant magnetic scattering . , 71:276278 , 1997 . j. m. tonnerre , l. seve , a. barbara - dechelette , f. bartolome , d. raoux , v. chakarian , c. c. kao , h. fischer , s. andrieu , and o. fruchart . soft x - ray resonant magnetic reflectivity study of thin films and multilayers . , 83(11):62936295 , 1998 . f. radu , s. k. mishra , i. zizak , a. i. erko , h. a. duerr , w. eberhardt , g. nowak , s. buschhorn , h. zabel , k. zhernenkov , m. wolff , d. schmitz , e. schierle , e. dudzik , and r. feyerherm . origin of the reduced exchange bias in an epitaxial @xmath492(111)/@xmath493(111 ) bilayer . , 79 , 2009 . s. tardif , s. cherifi , m. jamet , t. devillers , a. barski , d. schmitz , n. darowski , p. thakur , j. c. cezar , n. b. brookes , r. mattana , and j. cibert . exchange bias in @xmath494 nanocolumns : the role of surface oxidation . , 97:062501 , 2010 . s. valencia , a. crassous , l. bocher , v. garcia , x. moya , r. o. cherifi , c. deranlot , k. bouzehouane , s. fusil , a. zobelli , a. gloter , n. d. mathur , r. gaupp , a. abrudan , f. radu , a. barthlmy , and m. bibes . interface - induced room - temperature multiferroicity in @xmath495 . , 10:753 , 2011 . flavio c. vicentin , stefano turchini , francisco yubero , jan vogel , and maurizio sacchi . absorption cross sections at the @xmath499 edges of rare earths : a soft x - ray transmission experiment . , 74:187 194 , 1995 . maurizio sacchi , coryn f. hague , luca pasquali , alessandro mirone , jean - michel mariot , peter isberg , eric m. gullikson , and james h. underwood . optical constants of ferromagnetic iron via @xmath500 resonant magnetic scattering . , 81:15211524 , 1998 . l. seve , j.m . tonnerre , d. raoux , j.f . bobo , m. piecuch , m. desantis , p. troussel , j.m . brot , v. chakarian , c.c . kao , and c.t . resonant magnetic scattering in soft - x - ray range using multilayers . , 148:6869 , 1995 . mertins , o. zaharko , a. gaupp , f. schfers , d. abramsohn , and h. grimmer . soft x - ray magneto - optical constants at the @xmath501 edge determined by @xmath502ragg scattering and @xmath488araday effect . , 240:451 453 , 2002 . j. f. peters , j. miguel , m. a de vries , o. m. toulemonde , j. b. goedkoop , s. s. dhesi , and n. b. brookes . soft x - ray resonant magneto - optical constants at the @xmath505 and @xmath506 edges . , 70:224417 , 2004 . c. schssler - langeheine , e. weschke , a.yu grigoriev , h. ott , r. meier , d.v . vyalikh , chandan mazumdar , c. sutter , d. abernathy , g. grbel , and g. kaindl . resonant magnetic x - ray scattering from ultrathin @xmath468 metal films down to a few atomic layers . , 114 - 116:953 957 , 2001 . e. weschke , h. ott , e. schierle , c. schussler - langeheine , d.v . vyalikh , g. kaindl , v. leiner , m. ay , t. schmitte , h. zabel , and p.j . finite - size effect on magnetic ordering temperatures in long - period antiferromagnets : holmium thin films . , 93:157204 , 2004 . c. m. schneider , p. bressler , p. schuster , j. kirschner , j. j. de miguel , and r. miranda curie temperature of ultrathin films of fcc - cobalt epitaxially grown on atomically flat @xmath510(100 ) surfaces . , 64:10591062 , 1990 . v. scagnoli , u. staub , a.m. mulders , m. janousch , g.i . meijer , g. hammerl , j.m . tonnerre , and n. stojic . role of magnetic and orbital ordering at the metal - insulator transition in @xmath517 . , 73 , 2006 . v. scagnoli , u. staub , y. bodenthin , m. garcia - fernandez , a. m. mulders , g. i. meijer , and g. hammerl . induced noncollinear magnetic order of @xmath518 in @xmath519 observed by resonant soft x - ray diffraction . , 77:115138 , 2008 . j. l. garca - muoz , j. rodrguez - carvajal , and p. lacorre . neutron - diffraction study of the magnetic ordering in the insulating regime of the perovskites _ @xmath521 ( _ r _ = @xmath522 and @xmath523 ) . , 50:978992 , 1994 . u. staub , g. i. meijer , f. fauth , r. allenspach , j. g. bednorz , j. karpinski , s. m. kazakov , l. paolasini , and f. dacapito . direct observation of charge order in an epitaxial @xmath517 film . , 88:126402 , 2002 . v. scagnoli , u. staub , m. janousch , a. m. mulders , m. shi , g. i. meijer , s. rosenkranz , s. b. wilkins , l. paolasini , j. karpinski , s. m. kazakov , and s. w. lovesey . charge disproportionation and search for orbital ordering in @xmath519 by use of resonant x - ray diffraction . , 72:155111 , 2005 . . leininger , m. rahlenbeck , m. raichle , b. bohnenbuck , a. maljuk , c. t. lin , b. keimer , e. weschke , e. schierle , s. seki , y. tokura , and j. w. freeland . electronic structure , magnetic , and dielectric properties of the edge - sharing copper oxide chain compound @xmath524 . , 81:085111 , 2010 . drechsler , j. richter , a.a . a. gippius , a. vasiliev , a.a.a . bush , a.a.s . moskvin , j. malek , yu . prots , w. schnelle , and h. rosner . helical ground state and weak ferromagnetism in the edge - shared chain cuprate @xmath526 . , 73:83 , 2006 . h. wadati , j. okamoto , m. garganourakis , v. scagnoli , u. staub , y. yamasaki , h. nakao , y. murakami , m. mochizuki , m. nakamura , m. kawasaki , and y. tokura . origin of the large polarization in multiferroic @xmath527 thin films revealed by soft- and hard - x - ray diffraction . , 108:047203 , 2012 . a. m. mulders , s. m. lawrence , a. j. princep , u. staub , y. bodenthin , m. garcia - fernandez , m. garganourakis , j. hester , r. macquart , and c. d. ling . circularly polarized soft x - ray diffraction study of helical magnetism in hexaferrite . , 81:092405 , 2010 . e. dudzik , s. s. dhesi , h. a. drr , s. p. collins , m. d. roper , g. van der laan , k. chesnel , m. belakhovsky , a. marty , and y. samson . influence of perpendicular magnetic anisotropy on closure domains studied with x - ray resonant magnetic scattering . , 62:57795785 , 2000 . s. roy , m. r. fitzsimmons , s. park , m. dorn , o. petracic , igor v. roshchin , zhi - pan li , x. batlle , r. morales , a. misra , x. zhang , k. chesnel , j. b. kortright , s. k. sinha , and ivan k. schuller . depth profile of uncompensated spins in an exchange bias system . , 95:047201 , 2005 . s. valencia , a. gaupp , w. gudat , ll . abad , ll . balcells , and b. martinez . surface degradation of magnetic properties in manganite thin films proved with magneto - optical techniques in reflection geometry . , 90:252509 , 2007 . j. m. tonnerre , m. de santis , s. grenier , h. c. n. tolentino , v. langlais , e. bontempi , m. garcia - fernandez , and u. staub . depth magnetization profile of a perpendicular exchange coupled system by soft - x - ray resonant magnetic reflectivity . , 100:157202 , 2008 . h. ott , c. schler - langeheine , e. schierle , e. weschke , and g. kaindl . depth - resolved magnetic structure across the ferromagnetic to helical - antiferromagnetic phase transition in @xmath529 . , 82:214408 , 2010 . j. grabis , a. bergmann , a. nefedov , k. westerholt , and h. zabel . element - specific characterization of the interface magnetism in @xmath530_{n}$ ] multilayers by x - ray resonant magnetic scattering . , 72:024438 , 2005 . j. geissler , e. goering , m. justen , f. weigand , g. schtz , j. langer , d. schmitz , h. maletta , and r. mattheis . pt magnetization profile in a @xmath531 bilayer studied by resonant magnetic x - ray reflectometry . , 65:020405 , 2001 . n. jaouen , j.m . tonnerre , d. raoux , e. bontempi , l. ortega , m. menzenberg , w. felsch , a. rogalev , h.a . drr , e. dudzik , g. van der laan , h. maruyama , and m. suzuki . ce 5 @xmath532 magnetic profile in @xmath533 multilayers for the @xmath534 and @xmath535-like @xmath536 phases by x - ray resonant magnetic scattering . , 66:134420 , 2002 . s. valencia , a. gaupp , w. gudat , ll . abad , ll . balcells , and b. martinez . x - ray magnetic circular dichroism in reflection geometry : a tool for investigating surface magnetism in thin films . , 104:023903 023908 , 2008 . j. w. freeland , k. e. gray , l. ozyuzer , p. berghuis , e. badica , j. kavich , h. zheng , and j. f. mitchell . full bulk spin polarization and intrinsic tunnel barriers at the surface of layered manganites . , 4:6267 , 2005 . a. verna , bruce a. davidson , y. szeto , a.yu . petrov , a. mirone , a. giglia , n. mahne , and s. nannarone . measuring magnetic profiles at manganite surfaces with monolayer resolution . , 322:1212 1216 , 2010 . a. tebano , c. aruta , s. sanna , p. g. medaglia , g. balestrino , a. a. sidorenko , r. de renzi , g. ghiringhelli , l. braicovich , v. bisogni , and n. b. brookes . evidence of orbital reconstruction at interfaces in ultrathin @xmath540 films . , 100:137401 , 2008 . hase , i. pape , b.k .. tanner , h. drr , e. dudzik , g. van der laan , c.h . marrows , and b.j . hickey . soft - x - ray resonant magnetic diffuse scattering from strongly coupled @xmath542 multilayers . , 61:r3792r3795 , 2000 . d. r. lee , s. k. sinha , c. s. nelson , j. c. lang , c. t. venkataraman , g. srajer , and r. m. osgood . x - ray resonant magnetic scattering from structurally and magnetically rough interfaces in multilayered systems . @xmath543 . diffuse scattering . , 68:224410 , 2003 . serban smadici , peter abbamonte , anand bhattacharya , xiaofang zhai , bin jiang , andrivo rusydi , james n. eckstein , samuel d. bader , and jian - min zuo . electronic reconstruction at @xmath547 superlattice interfaces . , 99:196404 , 2007 . n. reyren , s. thiel , a. d. caviglia , l. fitting kourkoutis , g. hammerl , c. richter , c. w. schneider , t. kopp , a .- s . retschi , d. jaccard , m. gabay , d. a. muller , j .- m . triscone , and j. mannhart . superconducting interfaces between insulating oxides . , 317:11961199 , 2007 . a. v. boris , y. matiks , e. benckiser , a. frano , p. popovich , v. hinkov , p. wochner , m. castro - colin , e. detemple , v. k. malik , c. bernhard , t. prokscha , a. suter , z. salman , e. morenzoni , g. cristiani , h .- u . habermeier , and b. keimer . dimensionality control of electronic phase transitions in nickel - oxide superlattices . , 332:937940 , 2011 . j. chakhalian , j. w. freeland , h .- u . habermeier , g. cristiani , g. khaliullin , m. van veenendaal , and b. keimer . orbital reconstruction and covalent bonding at an oxide interface . , 318(5853):11141117 , 2007 . s. smadici , j. c. t. lee , s. wang , p. abbamonte , g. logvenov , a. gozar , c. deville cavellin , and i. bozovic . superconducting transition at 38 in insulating - overdoped @xmath550 superlattices : evidence for interface electronic redistribution from resonant soft x - ray scattering . , 102:107004 , 2009 . t. koida , m. lippmaa , t. fukumura , k. itaka , y. matsumoto , m. kawasaki , and h. koinuma . effect of _ a _ -site cation ordering on the magnetoelectric properties in @xmath551_{n}$ ] artificial superlattices . , 66:144418 , 2002 . h. wadati , j. geck , d. g. hawthorn , t. higuchi , m. hosoda , c. bell , y. hikita , h. y. hwang , c. schssler - langeheine , e. schierle , e. weschke , and g. a. sawatzky . electronic structure of the @xmath552 interface revealed by resonant soft x - ray scattering . , 24:012012 , 2011 . e. benckiser , m. w. haverkort , s. brck , e. goering , s. macke , a. frano , x. yang , o. k. andersen , g. cristiani , h .- u . habermeier , a. v. boris , i. zegkinoglou , p. wochner , h .- j . kim , v. hinkov , and b. keimer . orbital reflectometry of oxide heterostructures . , 10:189193 , 2011 . k. chesnel , m. belakhovsky , s. landis , j.c . toussaint , s.p . collins , g. van der laan , e. dudzik , and s.s . x - ray resonant magnetic scattering study of the magnetic coupling in @xmath555 nanolines and its evolution under magnetic field . , 66:024435 , 2002 . f. y. ogrin , e. sirotkin , g. van der laan , g. beutier , c. a. ross , w. jung , and r. menon . soft x - ray resonant magnetic scattering investigation of stable magnetic configurations in patterned rings . , 103:07e909 , 2008 . k. yamada , c. h. lee , k. kurahashi , j. wada , s. wakimoto , s. ueki , h. kimura , y. endoh , s. hosoya , g. shirane , r. j. birgeneau , m. greven , m. a. kastner , and y. j. kim . doping dependence of the spatially modulated dynamical spin correlations and the superconducting - transition temperature in @xmath562 . , 57:61656172 , 1998 . j. m. tranquada , j. d. axe , n. ichikawa , y. nakamura , s. uchida , and b. nachumi . neutron - scattering study of stripe - phase order of holes and spins in @xmath563@xmath564@xmath565@xmath566 . , 54:74897499 , 1996 . m. von zimmermann , a. vigliante , t. niemoller , n. ichikawa , t. frello , j. madsen , p. wochner , s. uchida , n. h. andersen , j. m. tranquada , d. gibbs , and j. r. schneider . hard - x - ray diffraction study of charge stripe order in @xmath567 . , 41:629634 , 1998 . p. abbamonte , g. blumberg , a. rusydi , a. gozar , p. g. evans , t. siegrist , l. venema , h. eisaki , e. d. isaacs , and g. a. sawatzky . crystallization of charge holes in the spin ladder of @xmath569 . , 431:1078 , 2004 . c. t. chen , f. sette , y. ma , m. s. hybertsen , e. b. stechel , w. m. c. foulkes , m. schluter , s. w. cheong , a. s. cooper , l. w. rupp , b. batlogg , y. l. soo , z. h. ming , a. krol , and y. h. kao . electronic states in @xmath574 probed by soft - x - ray absorption . , 66:104107 , 1991 . e. pellegrin , n. ncker , j. fink , s. l. molodtsov , a. gutierrez , e. navas , o. strebel , z. hu , m. domke , g. kaindl , s. uchida , y. nakamura , j. markl , m. klauda , g. saemannischenko , a. krol , j. l. peng , z. y. li , and r. l. greene . orbital character of states at the @xmath575 level in @xmath562 and @xmath576 . , 47:33543367 , 1993 . jrg fink , enrico schierle , eugen weschke , jochen geck , david hawthorn , viktor soltwisch , hiroki wadati , hsueh - hung wu , hermann a. drr , nadja wizent , bernd bchner , and george a. sawatzky . charge ordering in @xmath577 studied by resonant soft x - ray diffraction . , 79:100502 , 2009 . d. c. peets , d. g. hawthorn , k. m. shen , young - june kim , d. s. ellis , h. zhang , seiki komiya , yoichi ando , g. a. sawatzky , ruixing liang , d. a. bonn , and w. n. hardy . x - ray absorption spectra reveal the inapplicability of the single - band @xmath559 model to overdoped cuprate superconductors . , 103:087402 , 2009 . h. h. klauss , w. wagener , m. hillberg , w. kopmann , h. walf , f. j. litterst , m. hucker , and b. buchner . from antiferromagnetic order to static magnetic stripes : the phase diagram of @xmath579 . , 85:45904593 , 2000 . m. hcker , g. d. gu , j. m. tranquada , m. von zimmermann , h. h. klauss , n. j. curro , m. braden , and b. bchner . coupling of stripes to lattice distortions in cuprates and nickelates . , 460:170173 , 2007 . a. a. kordyuk , s. v. borisenko , m. s. golden , s. legner , k. a. nenkov , m. knupfer , j. fink , h. berger , l. forr , and r. follath . doping dependence of the @xmath575 surface in @xmath582 . , 66:014502 , 2002 . nicolas doiron - leyraud , cyril proust , david leboeuf , julien levallois , jean - baptiste bonnemaison , ruixing liang , d. a. bonn , w. n. hardy , and louis taillefer . quantum oscillations and the @xmath575 surface in an underdoped high-@xmath570 superconductor . , 447:565568 , 2007 . grafe , f. hammerath , a. vyalikh , g. urbanik , v. kataev , th . wolf , g. khaliullin , and b. bchner . contrasting spin dynamics in @xmath584 and @xmath585 @xmath586 single crystals from @xmath510 nuclear quadrupole resonance : evidence for correlations between antiferromagnetism and pseudogap effects . , 77:014522 , 2008 . d. l. feng , a. rusydi , p. abbamonte , l. venema , i. elfimov , r. liang , d. a. bonn , w. n. hardy , c. schusler - langeheine , s. hulbert , c. kao , and g. a. sawatzky . zooming - in on the charge ordering in @xmath589 . , 2004 . tao wu , hadrien mayaffre , steffen kramer , mladen horvatic , claude berthier , w. n. hardy , ruixing liang , d. a. bonn , and marc - henri julien . magnetic - field - induced charge - stripe order in the high - temperature superconductor @xmath590 . , 477(7363):191194 , september 2011 . g. ghiringhelli , m. le tacon , m. minola , s. blanco - canosa , c. mazzoli , n. b. brookes , g. m. de luca , a. frano , d. g. hawthorn , f. he , t. loew , m. moretti sala , d. c. peets , m. salluzzo , e. schierle , r. sutarto , g. a. sawatzky , e. weschke , b. keimer , and l. braicovich . long - range incommensurate charge fluctuations in @xmath591 . , 2012 . j. chang , e. blackburn , a. t. holmes , n. b. christensen , j. larsen , j. mesot , r. liang , d. a. bonn , w. n. hardy , a. watenphul , m. v. zimmermann , e. m. forgan , and s. m. hayden . direct observation of competition between superconductivity and charge density wave order in @xmath590 . , 2012 . t. nagata , m. uehara , j. goto , j. akimitsu , n. motoyama , h. eisaki , s. uchida , h. takahashi , t. nakanishi , and n. mri . pressure - induced dimensional crossover and superconductivity in the hole - doped two - leg ladder compound @xmath592 . , 81:1090 , 1998 . a. rusydi , m. berciu , p. abbamonte , s. smadici , h. eisaki , y. fujimaki , s. uchida , m. rubhausen , and g. a. sawatzky . relationship between hole density and charge - ordering wave vector in @xmath592 . , 75:104510 , 2007 . a. rusydi , p. abbamonte , h. eisaki , y. fujimaki , s. smadici , n. motoyama , s. uchida , y. j. kim , m. ruebhausen , and g. a. sawatzky . strain amplification of the 4k@xmath593 chain instability in @xmath569 . , 100:036403 , 2008 . n. ncker , m. merz , c. a. kuntscher , s. gerhold , s. schuppler , r. neudert , m. s. golden , j. fink , d. schild , s. stadler , v. chakarian , j. freeland , y. u. idzerda , k. conder , m. uehara , t. nagata , j. goto , j. akimitsu , n. motoyama , h. eisaki , s. uchida , u. ammerahl , and a. revcolevschi . hole distribution in @xmath594 ladder compounds studied by x - ray absorption spectroscopy . , 62:14384 , 2000 . a. koitzsch , d. s. inosov , h. shiozawa , v. b. zabolotnyy , s. v. borisenko , a. varykhalov , c. hess , m. knupfer , u. ammerahl , a. revcolevschi , and b. bchner . observation of the @xmath575 surface , the band structure , and their diffraction replicas of @xmath595 by angle - resolved photoemission spectroscopy . , 81:113110 , 2010 . c. schssler - langeheine , j. schlappa , a. tanaka , z. hu , c.f . chang , e. schierle , m. benomar , h. ott , e. weschke , g. kaindl , o. friedt , g.a . sawatzky , h.j . lin , c.t . chen , m. braden , and l.h . spectroscopy of stripe order in @xmath603 using resonant soft x - ray diffraction . , 95 , 2005 . j. schlappa , c. schler - langeheine , c. f. chang , h. ott , a. tanaka , z. hu , m. w. haverkort , e. schierle , e. weschke , g. kaindl , and l. h. tjeng . direct observation of @xmath606 orbital ordering in magnetite . , 100:026406 , 2008 . wilkins , p.d . spencer , p.d . hatton , s.p . collins , m.d . roper , d. prabhakaran , and a.t . boothroyd . direct observation of orbital ordering in @xmath611 using soft x - ray diffraction . , 91:167205 , 2003 . wilkins , n. stojic , t.a.w . beale , n. binggeli , c.w.m . castleton , p. bencok , d. prabhakaran , a.t . boothroyd , p.d . hatton , and m. altarelli . resonant soft x - ray scattering investigation of orbital and magnetic ordering in @xmath611 . , 71:245102 , 2005 . dhesi , a. mirone , c. de nadai , p. ohresser , p. bencok , n.b . brookes , p. reutler , a. revcolevschi , a. tagliaferri , o. toulemonde , and g. van der laan . unraveling orbital ordering in @xmath611 . , 92:056403 , 2004 . u. staub , v. scagnoli , a.m. mulders , k. katsumata , z. honda , h. grimmer , m. horisberger , and j.m . orbital and magnetic ordering in @xmath611 studied by soft x - ray resonant scattering . , 71:214421 , 2005 . k. j. thomas , j. p. hill , s. grenier , y - j . kim , p. abbamonte , l. venema , a. rusydi , y. tomioka , y. tokura , d. f. mcmorrow , g. sawatzky , and m. van veenendaal . soft x - ray resonant diffraction study of magnetic and orbital correlations in a manganite near half doping . , 92:237204 , 2004 . s. b. wilkins , n. stojic , t. a. w. beale , n. binggeli , p. d. hatton , p. bencok , s. stanescu , j. f. mitchell , p. abbamonte , and m. altarelli . separating the causes of orbital ordering in @xmath615 using resonant soft x - ray diffraction . , 18:l323l329 , 2006 . s. grenier , k. j. thomas , j. p. hill , u. staub , y. bodenthin , m. garca - fernndez , v. scagnoli , v. kiryukhin , s - w . cheong , b. g. kim , and j. m. tonnerre . direct observation of oxygen superstructures in manganites . , 99:206403 , 2007 . t. a. w. beale , s. r. bland , r. d. johnson , p. d. hatton , j. c. cezar , s. s. dhesi , m. v. zimmermann , d. prabhakaran , and a. t. boothroyd . thermally induced rotation of @xmath221 orbital stripes in @xmath616 . , 79:054433 , 2009 . s. y. zhou , y. zhu , m. c. langner , y .- d . chuang , p. yu , w. l. yang , a. g. cruz gonzalez , n. tahir , m. rini , y .- h . chu , r. ramesh , d .- h . lee , y. tomioka , y. tokura , z. hussain , and r. w. schoenlein . ferromagnetic enhancement of @xmath617-type spin ordering in @xmath618 . , 106:186404 , 2011 . lee , c .- c . kao , c. s. nelson , h. jang , k .- ko , s. b. kim , y. j. choi , s .- w . cheong , s. smadici , p. abbamonte , and j .- h . park . fragile magnetic ground state in half - doped @xmath619 . 107:037206 , 2011 . a. m. mulders , u. staub , v. scagnoli , s. w. lovesey , e. balcar , t. nakamura , a. kikkawa , g. van der laan , and j. m. tonnerre . high - order @xmath473 multipole motifs observed in @xmath620 with resonant soft x - ray @xmath621 diffraction . , 18:11195 , 2006 . a. m. mulders , u. staub , v. scagnoli , y. tanaka , a. kikkawa , k. katsumata , and j. m. tonnerre . manipulating @xmath229 quadrupolar pair - interactions in @xmath622 using a magnetic field . , 75:184438 , 2007 . a. j. princep , a. m. mulders , u. staub , v. scagnoli , t. nakamura , a. kikkawa , s.w . lovesey , and e. balcar . triakontadipole and high - order dysprosium multipoles in the antiferromagnetic phase of @xmath620 . , 23:266002 , 2011 . a. j. princep , a. m. mulders , e. schierle , e. weschke , j. hester , w. d. hutchison , y. tanaka , n. terada , y. narumi , and t. nakamura . high - order @xmath468 multipoles in @xmath623 observed with soft resonant x - ray diffraction . , 24:075602 , 2012 . ying - hao chu , lane w. martin , mikel b. holcomb , martin gajek , shu - jen han , qing he , nina balke , chan - ho yang , donkoun lee , wei hu , qian zhan , pei - ling yang , arantxa fraile - rodriguez , andreas scholl , shan x. wang , and r. ramesh . electric - field control of local ferromagnetism using a magnetoelectric multiferroic . , 7:478482 , 2008 . m. kenzelmann , a. b. harris , s. jonas , c. broholm , j. schefer , s. b. kim , c. l. zhang , s .- w . cheong , o. p. vajk , and j. w. lynn . magnetic inversion symmetry breaking and ferroelectricity in @xmath624 . , 95:087206 , 2005 . y. yamasaki , h. sagayama , n. abe , t. arima , k. sasai , m. matsuura , k. hirota , d. okuyama , y. noda , and y. tokura . cycloidal spin order in the @xmath441-axis polarized ferroelectric phase of orthorhombic perovskite manganite . , 101:097204 , 2008 . j. koo , c. song , s. ji , j .- s . lee , j. park , t .- h . jang , c .- h . yang , j .- h . park , y. h. jeong , k .- b . lee , t. y. koo , y. j. park , j .- y . kim , d. wermeille , a. i. goldman , g. srajer , s. park , and s .- w . non - resonant and resonant x - ray scattering studies on multiferroic @xmath627 . , 99:197601 , 2007 . t. r. forrest , s. r. bland , s. b. wilkins , h. c. walker , t. a. w. beale , p. d. hatton , d. prabhakaran , a. t. boothroyd , d. mannix , f. yakhou , and d. f. mcmorrow . ordering of localized electronic states in multiferroic @xmath624 : a soft x - ray resonant scattering study . , 20:422205 , 2008 . y. bodenthin , u. staub , m. garcia - fernandez , m. janoschek , j. schlappa , e. i. golovenchits , v. a. sanina , and s. g. lushnikov . manipulating the magnetic structure with electric fields in multiferroic @xmath629 . , 100:027201 , 2008 . s. b. wilkins , s. di matteo , t. a. w. beale , y. joly , c. mazzoli , p. d. hatton , p. bencok , f. yakhou , and v. a. m. brabers . critical reexamination of resonant soft x - ray @xmath621 forbidden reflections in magnetite . , 79:201102(r ) , 2009 . u. staub , y. bodenthin , c. piamonteze , m. garcia - fernandez , v. scagnoli , m. garganourakis , s. koohpayeh , d. fort , and s. w. lovesey . parity- and time - odd atomic multipoles in magnetoelectric @xmath630 as seen via soft x - ray @xmath502ragg diffraction . , 80 , 2009 . u. staub , y. bodenthin , m. garcia - fernandez , r. a. de souza , m. garganourakis , e. i. golovenchits , v. a. sanina , and s. g. lushnikov . magnetic order of multiferroic @xmath631 studied by resonant soft x - ray @xmath621 diffraction . , 81:144401 , 2010 . s. partzsch , s. b. wilkins , j. p. hill , e. schierle , e. weschke , d. souptel , b. bchner , and j. geck . observation of electronic ferroelectric polarization in multiferroic @xmath632 . , 107:057201 , 2011 . r. a. de souza , u. staub , v. scagnoli , m. garganourakis , y. bodenthin , s .- w . huang , m. garcia - fernandez , s. ji , s .- h . lee , s. park , and s .- w . magnetic structure and electric field effects in multiferroic @xmath633 . , 84:104416 , 2011 . s. w. huang , d. j. huang , j. okamoto , c. y. mou , w. b. wu , k. w. yeh , c. l. chen , m. k. wu , h. c. hsu , f. c. chou , and c. t. chen . magnetic ground state and transition of a quantum multiferroic @xmath634 . , 101 , 2008 . t. a. w. beale , s. b. wilkins , r. d. johnson , s. r. bland , y. joly , t. r. forrest , d. f. mcmorrow , f. yakhou , d. prabhakaran , a. t. boothroyd , and p. d. hatton . antiferromagnetically spin polarized oxygen observed in magnetoelectric @xmath635 . , 105:087203 , 2010 . hoyoung jang , j .- s . ko , w .- s . noh , t. y. koo , j .- y . kim , k .- b . lee , j .- h . park , c. l. zhang , sung baek kim , and s .- w . coupled magnetic cycloids in multiferroic @xmath636 and @xmath637 . , 106:047203 , 2011 . s. b. wilkins , t. r. forrest , t. a. w. beale , s. r. bland , h. c. walker , d. mannix , f. yakhou , d. prabhakaran , a. t. boothroyd , j. p. hill , p. d. hatton , and d. f. mcmorrow . nature of the magnetic order and origin of induced ferroelectricity in @xmath624 . , 103:207602 , 2009 . f. fabrizi , h. c. walker , l. paolasini , f. de bergevin , a. t. boothroyd , d. prabhakaran , and d. f. mcmorrow . circularly polarized x rays as a probe of noncollinear magnetic order in multiferroic @xmath624 . , 102:237205 , 2009 . o. prokhnenko , r. feyerherm , e. dudzik , s. landsgesell , n. aliouane , l. c. chapon , and d. n. argyriou . enhanced ferroelectric polarization by induced @xmath473 spin order in multiferroic @xmath638 . , 98:057206 , 2007 . m. garcia - fernandez , u. staub , y. bodenthin , s. m. lawrence , a. m. mulders , c. e. buckley , s. weyeneth , e. pomjakushina , and k. conder . resonant soft x - ray powder diffraction study to determine the orbital ordering in a - site - ordered @xmath640 . , 77:060402(r ) , 2008 . m. garca - fernndez , u. staub , y. bodenthin , v. scagnoli , v. pomjakushin , s. w. lovesey , a. mirone , j. herrero - martn , c. piamonteze , and e. pomjakushina . orbital order at @xmath643 and @xmath644 sites and absence of @xmath645 polaron formation in manganites . , 103:097205 , 2009 . u. staub , m. garcia - fernandez , y. bodenthin , v. scagnoli , r. a. de souza , m. garganourakis , e. pomjakushina , and k. conder . orbital and magnetic ordering in @xmath646 and @xmath647 manganites near half doping studied by resonant soft x - ray powder diffraction . , 79:224419 , 2009 . y. bodenthin , u. staub , c. piamonteze , m. garcia - fernandez , m. j. martinez - lope , and j. a. alonso . magnetic and electronic properties of @xmath648 perovskites studied by resonant soft x - ray magnetic powder diffraction . , 23:036002 , 2011 . m. garcia - fernandez , v. scagnoli , u. staub , a. m. mulders , m. janousch , y. bodenthin , d. meister , b. d. patterson , a. mirone , y. tanaka , t. nakamura , s. grenier , y. huang , and k. conder . magnetic and electronic @xmath649 states in the layered cobaltate @xmath650 . , 78:054424 , 2008 . stefano marchesini , sebastien boutet , anne e. sakdinawat , michael j. bogan , sasa bajt , anton barty , henry n. chapman , matthias frank , stefan p. hau - riege , abraham szoke , congwu cui , david a. shapiro , malcolm r. howells , john c. h. spence , joshua w. shaevitz , joanna y. lee , janos hajdu , and marvin m. seibert . massively parallel x - ray holography . , 2:560563 , 2008 . diling zhu , manuel guizar - sicairos , benny wu , andreas scherz , yves acremann , tolek tyliszczak , peter fischer , nina friedenberger , katharina ollefs , michael farle , james r. fienup , and joachim sthr . high - resolution x - ray lensless imaging by differential holographic encoding . , 105:043901 , 2010 . c. gutt , s. streit - nierobisch , l .- m . stadler , b. pfau , c. m. gnther , r. knnecke , r. frmter , a. kobs , d. stickler , h. p. oepen , r. r. fustlin , r. treusch , j. feldhaus , e. weckert , i. a. vartanyants , m. grunze , a. rosenhahn , t. wilhein , s. eisebitt , and g. grbel . single - pulse resonant magnetic scattering using a soft x - ray free - electron laser . , 81:100401 , 2010 . s. konings , c. schler - langeheine , h. ott , e. weschke , e. schierle , h. zabel , and j. b. goedkoop . magnetic domain fluctuations in an antiferromagnetic film observed with coherent resonant soft x - ray scattering . , 106:077402 , 2011 . p. beaud , s. l. johnson , e. vorobeva , u. staub , r. a. de souza , c. j. milne , q. x. jia , and g. ingold . ultrafast structural phase transition driven by photoinduced melting of charge and orbital order . , 103:155702 , 2009 . k. holldack , n. pontius , e. schierle , t. kachel , v. soltwisch , r. mitzner , t. quast , g. springholz , and e. weschke . ultrafast dynamics of antiferromagnetic order studied by femtosecond resonant soft x - ray diffraction . , 97:062502 , 2010 . c. stamm , t. kachel , n. pontius , r. mitzner , t. quast , k. holldack , s. khan , c. lupulescu , e. f. aziz , m. wietstruk , h. a. drr , and w. eberhardt . femtosecond modification of electron localization and transfer of angular momentum in nickel . , 6:740743 , 2007 . c. boeglin , e. beaurepaire , v. halte , v. lopez - flores , c. stamm , n. pontius , h. a. drr , and j .- y . bigot . distinguishing the ultrafast dynamics of spin and orbital moments in solids . , 465:458461 , 2010 . n. pontius , t. kachel , c. schler langeheine , w. f. schlotter , m. beye , f. sorgenfrei , c. f. chang , a. fhlisch , w. wurth , p. metcalf , i. leonov , a. yaresko , n. stojanovic , m. berglund , n. guerassimova , s. dsterer , h. redlin , and h. a. drr . time - resolved resonant soft x - ray diffraction with free - electron lasers : femtosecond dynamics across the @xmath497erwey transition in magnetite . , 98(18):182504 , 2011 . h. ehrke , r. i. tobey , s. wall , s. a. cavill , m. frst , v. khanna , th . garl , n. stojanovic , d. prabhakaran , a. t. boothroyd , m. gensch , a. mirone , p. reutler , a. revcolevschi , s. s. dhesi , and a. cavalleri . photoinduced melting of antiferromagnetic order in @xmath466 measured using ultrafast resonant soft x - ray diffraction . , 106:217401 , 2011 . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: an important trend in the condensed matter physics of the last two decades , has been the use of advanced field - theoretical methods to discuss various subtle and fundamental properties of interacting many - particle systems at low temperatures . there are several reasons for this trend . the first reason is of course , that a traditional topic in statistical and condensed matter physics is the study of phase transitions and critical phenomena , for which the universal properties are independent of the microscopic details of the system and can therefore be determined by a field theory describing only the large - scale properties of the system of interest . since the latter are usually solely determined by symmetry considerations , this has led to the important concept of spontaneous symmetry breaking , which has turned out to be not only highly successful in condensed matter physics , but also in high - energy physics and in particular in the standard model of elementary particles @xcite . a second reason is that soon after the development of the renormalization group methods for critical phenomena @xcite , it was realized that the same methods can in fact be used to describe the large - scale properties of many - particle systems at any temperature and not only near the critical one . moreover , application of the renormalization group ideas does not only lead to an understanding of the static behaviour but also of the dynamical properties @xcite . as a result quantum field theories can be used to conveniently determine the dynamics of many - particle systems close to equilibrium , i.e. , for example the linear hydrodynamical equations of motion . in addition , it can even be used in highly nonequilibrium situations where in general also nonlinearities play an important role . this feature , that also the dynamics of the system can be captured by a quantum field theory , has for instance led in recent years to the study of so - called quantum phase transitions @xcite . finally , the importance of field - theoretical methods in condensed matter physics is associated with the observation that also the effects of imperfections , i.e. , disorder , can be treated in this way @xcite . apart from the technological importance of disorder , for example for superconducting magnets , disorder leads also to fundamentally new physics such as the phenomenon of localization @xcite . in quantum hall systems , a combination of disorder and interaction effects give rise to the realization of various peculiar quantum fluids with fractionally charged excitations @xcite . the application of field theories has proven to be highly successful in this case and has led to a theory of the quantum hall effect in terms of edge states that form a chiral luttinger liquid @xcite . in mesoscopic physics , the study of disorder in small electronic structures has resulted in the so - called random matrix theory @xcite , which has also been of much use in the study of the quantization of classically chaotic systems . in this course we aim to show that quantum field theory is also very convenient for obtaining a detailed understanding of the equilibrium and nonequilibrium properties of trapped atomic gases . after the first observations of bose - einstein condensation in 1995 @xcite , degenerate atomic gases have received again a great deal of attention and are presently the main subject of study of a large number of experimental and theoretical groups around the world . the reason for all this excitement is , first of all , that bose - einstein condensation has never before been observed experimentally in a clear - cut manner , even though this phenomenon was already predicted by einstein in 1925 @xcite . second , it is of fundamental interest because it is the only phase transition that occurs also in the absense of interactions and is , therefore , the textbook example for the use of statistical - physics methods . finally , the goal of achieving bose - einstein condensation turned out to be much more difficult than anticipated at first . just before the breakthrough in 1995 , it had even acquired the nature of a ` quest for the holy grail ' , since the pioneering experiments were already performed in 1980 @xcite . from a theoretical point of view , a quantitative understanding of the experiments with cold atomic gases requires that we take into account the following two effects . first , the gas is magnetically trapped in an , usually axially symmetric , harmonic oscillator potential . this is necessary because , in order to obtain the required record low temperatures of @xmath0 nk , the gas can not be allowed to have any contact with material walls . second , the atoms of the gas interact with each other , which in general dramatically affects the behaviour of the condensed gas , even at the relevant low densities of @xmath1 @xmath2 . an accurate description of these degenerate gases thus requires the solution of a highly inhomogeneous many - body problem . the theoretical challenge posed by these new quantum systems lies therein that the density of the gas is sufficiently small that it should be possible to accurately solve this many - body problem from first principles and to compare the outcome of the theory directly with experiment , i.e. , without any adjustable parameters . in our opinion , quantum field theory is the most simple way in which we are able to meet this challenge . we start our development of the quantum field theory of trapped atomic gases by considering first the equilibrium properties of these gases . we consider both bose and fermi gases , and the ultimate aim of this section is to arrive , for both cases , at an accurate description of the normal and superfluid phases of the gas . although mixtures of bose and fermi gases are also of current interest , we do not consider them explicitly here , because they can be treated by a straightforward generalization of the theory . in section [ nft ] we then turn to the nonequilibrium properties , which are perhaps the most interesting and certainly the least understood at present . the reason for organizing the course in this way , is that the development of the equilibrium theory gives us an opportunity to introduce all the necessary tools that are required for a treatment of the more complicated nonequilibrium case . in particular , we present in detail the way in which we can make use of functional methods . to make also a connection with the more familiar operator language , however , we first briefly summarize the outcome of the second quantization formalism . the atoms of interest to us have internal degrees of freedom due to the electron and nuclear spins . in principle this is very important , because it gives the atom a magnetic moment , which is used to trap the atoms in a magnetic field minumum . during this course , however , we restrict ourselves to atomic gases that are a mixture of at most two hyperfine states . without loss of generality , we can then suppose to have @xmath3 identical atoms with mass @xmath4 and effective spin @xmath5 in an external potential @xmath6 . as a result , the time - dependent schrdinger equation we have to solve is @xmath7 where the hamiltonian is @xmath8 @xmath9_- = i\hbar$ ] , and all other commutators of the positions and momenta vanish . the first term in the right - hand side is the sum of the one - particle hamiltonians , which includes an effective zeeman interaction that accounts for a possible difference in the hyperfine energies . the second term represents the interactions . for simplicity , we have assumed that the interaction @xmath10 is independent of the hyperfine states of the atoms @xmath11 and @xmath12 . this is in general not justified for realistic atomic gases , but is valid for the specific applications that we have in mind . moreover , in section [ hft ] we also discuss the general case . finally , we have also neglected possible three - body forces . this is a result of the fact that we are interested in dilute quantum gases , for which it is highly improbable for three atoms to simultaneously interact with each other . without interactions the eigenstates are , of course , given by the states @xmath13 , where the specific quantum state for each atom is exactly known . here @xmath14 and the nonnegative integers @xmath15 , @xmath16 , and @xmath17 denote the three quantum numbers that are required to specify the one - particle eigenstates in the external potential . the wave functions and energies of these eigenstates are @xmath18 and @xmath19 , respectively , and are found from the time - independent schrdinger equation @xmath20 in addition , the internal state @xmath21 is a shorthand notation for @xmath22 . the many - body wave function @xmath23 , however , has to be symmetric or antisymmetric under permutations for bosonic or fermionic atoms , respectively . therefore it is more convenient to use a properly ( anti)symmetrized version of the above basis , i.e. , the states @xmath24 with the occupation numbers @xmath25 for bosons and @xmath26 for fermions . the hilbert space of all these states , without the constraint @xmath27 , is known as the fock space . clearly , the many - body wave function @xmath23 can be expanded in this basis as @xmath28 where @xmath29 is the amplitude for the gas to be in state @xmath24 at time @xmath30 . in this basis the schrdinger equation becomes @xmath31 which shows that we need the matrix elements of the hamiltonian between different states in the fock space . to calculate these most easily we introduce so - called annihilation operators @xmath32 by @xmath33 from which it follows that the creation operators @xmath34 obey @xmath35 as a result , we see that the operator @xmath36 counts the number of atoms in the state @xmath37 , i.e. , @xmath38 we have also that @xmath39_{\mp } = [ \hat{\psi}^{\dagger}_{{\bf n},\alpha } , \hat{\psi}^{\dagger}_{{\bf n}',\alpha'}]_{\mp}=0 $ ] and most importantly that @xmath40_{\mp } = \delta_{{\bf n},{\bf n } ' } \delta_{\alpha,\alpha'}~.\ ] ] in equations ( [ cr ] ) and ( [ com ] ) the upper sign refers to bosons and the lower to fermions . this will be true throughout the course . from these results we can now easily show , first of all , that the basis in the fock space is given by @xmath41 with @xmath42 the vacuum state containing no atoms . second , the hamiltonian is @xmath43 with @xmath44 the one - particle energies . furthermore , @xmath45 is the amplitude for a collision which scatters two atoms out of the states @xmath46 and @xmath47 into the states @xmath37 and @xmath48 , as schematically shown in figure 1(a ) . introducing the field operators @xmath49 and @xmath50 , that annihilate and create atoms in the spin state @xmath21 at position @xmath51 respectively , we can rewrite this result into @xmath52 note that , due to the ( anti)commutation relations of the creation and annihilation operators and the orthogonality of the wave functions @xmath53 , the field operators obey @xmath54_{\mp}= [ \hat{\psi}^{\dagger}_{\alpha}({\bf x } ) , \hat{\psi}^{\dagger}_{\alpha'}({\bf x}')]_{\mp}= 0 $ ] and @xmath55_{\mp}= \delta({\bf x}-{\bf x}')\delta_{\alpha,\alpha'}$ ] . moreover , it is also important for the following to note that the number operator is @xmath56 and , similarly , that the effective total spin operator is @xmath57 the density of atoms in the state @xmath21 is thus simply @xmath58 . this completes our brief discussion of the second quantization formalism . in principle , we could now proceed to develop an operator formulation of the quantum field theory of interest to us . since the experimentally most important observables can be expressed as appropriate products of the field operators , as we have just seen , this would essentially amount to the study of the ( imaginary ) time evolution of the heisenberg operator @xmath59 at a fixed chemical potential @xmath60 @xcite . put differently , the desired quantum field theory would be defined by the heisenberg equation of motion @xmath61_-~,\ ] ] and we would need to solve this equation in a sufficiently accurate approximation . as mentioned previously , however , we here want to develop feynman s ` path - integral ' formulation of the problem , which will turn out to be much more convenient for our purposes . to do so in a manner that is the same for both bosonic and fermionic atomic gases , we first need to introduce some mathematical background . we have seen that in the case of fermions , we need to make use of anticommuting creation and annihilation operators . this automatically builds in the pauli principle in the theory , because it implies that @xmath62 and thus that the occupation numbers @xmath63 are restricted to be either @xmath64 or @xmath65 . for reasons that will become clear in a moment , it is in that case also convenient to introduce anticommuting complex numbers or grassmann variables . the simplest example is to have two such grassmann variables , say @xmath66 and @xmath67 . the set @xmath68 , and linear combinations thereof with complex coefficients , form then a so - called grassmann algebra . by definition we have @xmath69_+ = [ \phi,\phi^*]_+ = [ \phi^*,\phi^*]_+ = 0 $ ] and thus in particular @xmath70 . therefore , the above set is indeed complete . the complex conjugation in this algebra is defined by @xmath71 , @xmath72 , and @xmath73 . moreover , we can also define an analytic function on this algebra by @xmath74 as a result , it is natural to define also a differentiation by @xmath75 to be more precise this is in fact a left differentiation and the minus sign occurs , because we need to permute @xmath67 and @xmath66 before we can differentiate with respect to @xmath66 . so , similarly , we have @xmath76 and @xmath77 next we also need integrations over these grassmann variables . note that since @xmath78 we have only two possible integrals , namely @xmath79 and @xmath80 . we define these by @xmath81 and @xmath82 this means that integration is equivalent to differentation . the main reason for the above definition is that we want the integration to obey the usual rules of partial integration . in particular , this implies that @xmath83 for any function @xmath84 . it is clear that this condition requires that @xmath85 . the result of @xmath86 is then solely a question of normalization . it turns out that we are primarily interested in integrals of the form @xmath87 as we will see in section [ fi ] . clearly , we can now also consider the grassmann algebra based on the variables @xmath88 and @xmath89 with @xmath90 . what we will need in the following are gaussian integrals over these variables . it is not difficult to show with the above definitions that @xmath91}~.\ ] ] if the variables @xmath88 and @xmath89 were just ordinary complex numbers we would in contrast have the result @xmath92}~.\ ] ] these last two results will be used many times in the following . one immediate use of these grassmann variables is that we can now consider eigenstates of the annihilation operator @xmath32 , also when we are dealing with fermions . these eigenstates are called coherent states . consider the state @xmath93 where @xmath94 is a grassmann variable that also anticommutes with the creation and annihilation operators in our fock space . clearly we have that @xmath95 so @xmath96 is indeed an eigenstate of @xmath32 with eigenvalue @xmath94 . in general we can now make the states @xmath97 that obey @xmath98 . introducing the grassmann - valued field @xmath99 , the latter two relations can be rewritten as @xmath100 and @xmath101 . it is important to note that these coherent states are not orthonormal . in contrast , we find that @xmath102 nevertheless they obey a closure relation , as can be seen explicitly from @xmath103 we write this from now on simply as @xmath104d[\phi]~e^{-(\phi|\phi ) } |\phi\rangle\langle\phi| = \hat{1}~.\ ] ] the interesting observation at this point is that essentially the same formulaes also hold for bosons @xcite . we have only a minus sign difference in @xmath105 but then it can still be easily shown that @xmath106 . a convenient way to do so is , for instance , by noting that the commutation relation @xmath107_- = \delta({\bf x}-{\bf x } ' ) \delta_{\alpha,\alpha'}$ ] implies that @xmath108 acts as @xmath109 on these states . furthermore , the same expressions hold for the overlap @xmath110 and the closure relation , if we define the integration measure by @xmath111d[\phi ] \equiv \int \prod_{{\bf n},\alpha } d\phi^*_{{\bf n},\alpha } d\phi_{{\bf n},\alpha}/(2\pi i)$ ] in this case . summarizing , we have thus found for bosons and fermions that @xmath112 @xmath113 and @xmath104d[\phi]~e^{-(\phi|\phi ) } |\phi\rangle\langle\phi| = \hat{1}~.\ ] ] the last ingredient we need is that in terms of these coherent states the trace of an operator @xmath114 over the fock space can be expressed as @xmath115 = \int d[\phi^*]d[\phi]~e^{-(\phi|\phi ) } \langle\pm\phi|\hat{o}|\phi\rangle~.\ ] ] the minus sign in the fermionic case is easily understood from the fact that we then have @xmath116 , due to the anticommuting nature of the grassmann variables . after this mathematical interlude we can now return to physics and to our goal of arriving at a unified treatment of bosonic and fermionic atomic gases . we are interested in determining the equilibrium properties of a trapped atomic gas at some temperature @xmath117 . from statistical physics we know that all these properties can be obtained from the grand - canonical partition function @xmath118~,\ ] ] where @xmath119 and @xmath60 the chemical potential . we thus need to evaluate this quantity . as mentioned previously , we want to do so by making use of quantum field theory and , in addition , of feynman s path - integral approach to quantum mechanics . we thus want to write the partition function as a functional integral over time - dependent fields @xmath120 , just like the partition function of a single particle in an external potential can be written as a functional integral over all time - dependent paths @xmath121 . this can indeed be achieved with the coherent states that we have introduced in the previous section . we start with using our formula ( [ trace ] ) for the trace of an operator , @xmath122d[\phi]~e^{-(\phi|\phi ) } \langle\pm\phi| e^{-\beta(\hat{h}-\mu \hat{n } ) } |\phi\rangle~,\ ] ] and observe that we are thus faced with the task of calculating the matrix elements @xmath123 with @xmath124 and @xmath125 . this is difficult in general but can be simplified in the following way . we first realize that the operator @xmath126 is identical to the quantum mechanical evolution operator @xmath127 evaluated at @xmath128 . put differently , we thus want to calculate the matrix elements of the imaginary - time evolution operator @xmath129 for @xmath130 . to do so , we next split the time interval @xmath131 $ ] into @xmath132 pieces , with @xmath133 and @xmath134 . so @xmath135 . the procedure is summarized in figure 2 . $ ] needed for feynman s path - integral formulation of the partition function . ] at each intermediate time @xmath136 we then apply a closure relation of the coherent states . this gives @xmath137d[\phi_m]~ e^{-(\phi_m|\phi_m ) } \right ) \prod_{m=1}^m \langle\phi_m| e^{-\delta\tau(\hat{h}-\mu \hat{n})/\hbar } |\phi_{m-1}\rangle~. \nonumber\end{aligned}\ ] ] now we can use that in the limit @xmath138 we only need to know the latter matrix elements up to order @xmath139 , because terms of order @xmath140 lead to corrections of order @xmath141 , which vanish in that limit . hence @xmath142/\hbar ) ~ , \nonumber\end{aligned}\ ] ] with the grand - canonical hamiltonian functional resulting from equations ( [ hamilton ] ) and ( [ number ] ) equal to @xmath143 = \sum_{\alpha } \int d{\bf x}~ \phi^*_{\alpha}({\bf x } ) \left\ { - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) + \epsilon_{\alpha } - \mu \right\ } \phi_{\alpha}({\bf x } ) \hspace*{0.4 in } \\ & & \hspace*{0.85 cm } + \frac{1}{2 } \sum_{\alpha,\alpha ' } \int d{\bf x } \int d{\bf x}'~ \phi^*_{\alpha}({\bf x } ) \phi^*_{\alpha'}({\bf x } ' ) v({\bf x}-{\bf x } ' ) \phi_{\alpha'}({\bf x } ' ) \phi_{\alpha}({\bf x})~ , \nonumber\end{aligned}\ ] ] since @xmath144 and @xmath145 . thus , neglecting terms of order @xmath140 , we can again exponentiate the right - hand side of equation ( [ me ] ) , which leads in first instance to @xmath146/\hbar}~,\ ] ] and therefore for the desired matrix element of the imaginary - time evolution operator to @xmath147d[\phi_m]~ e^{-(\phi_m|\phi_m ) } \right ) \\ & & \hspace*{0.83 in } \times \exp \left\ { \sum_{m=1}^{m } ( ( \phi_m|\phi_{m-1})- \delta\tau h[\phi^*_m,\phi_{m-1}]/\hbar ) \right\}~. \nonumber \end{aligned}\ ] ] this can then be manipulated into the suggestive form @xmath148d[\phi_m]~ \right ) \\ & & \times \exp \left\ { - \frac{1}{\hbar } \sum_{m=1}^{m } \delta\tau \left ( \hbar \frac{(\phi_m|\phi_{m})-(\phi_m|\phi_{m-1 } ) } { \delta\tau } + h[\phi^*_m,\phi_{m-1 } ] \right ) \right\}~. \nonumber \end{aligned}\ ] ] taking now the continuum limit @xmath138 and putting @xmath149 , we find that @xmath150d[\phi]~ e^{-s[\phi^*,\phi]/\hbar}~ , \nonumber\end{aligned}\ ] ] where the so - called euclidean action is @xmath151 \\ & & = \int_0^{\hbar\beta } d\tau~ \left\ { \sum_{\alpha } \int d{\bf x}~ \phi^*_{\alpha}({\bf x},\tau ) \hbar \frac{\partial}{\partial \tau } \phi_{\alpha}({\bf x},\tau ) + h[\phi^*(\tau),\phi(\tau ) ] \right\}~. \nonumber\end{aligned}\ ] ] this is essentially the desired functional integral over the complex fields @xmath120 with the boundary conditions @xmath152 and @xmath153 . it is precisely the field theory analogue of the feynman path integral . to obtain the partition function we only need to put @xmath154 equal to @xmath155 and perform a last integration over @xmath156 and @xmath157 . it then finally becomes @xmath158d[\phi]~ e^{-s[\phi^*,\phi]/\hbar}\ ] ] with the boundary conditions @xmath159 , i.e. , the fields are periodic in @xmath131 $ ] for bosons and antiperiodic for fermions . note that in equation ( [ path1 ] ) we have used the same notation for the integration measure as in equation ( [ path2 ] ) , although there is in principle one more integration in the expression for the partition function . the difference is in the continuum limit accounted for in the boundary conditions , which are in practice usually left implicite . having arrived at an exact identity between the partition function and a functional integral , we are now going to familiarize ourselves with this identity , and with how to perform functional integrals in general , by considering the ideal quantum gases . since the partition functions @xmath160 of the ideal quantum gases are known exactly , they are ideal test cases for our field - theoretical methods . moreover , a thorough knowledge of the ideal quantum gases is also an important first step in understanding experiments with trapped atomic gases , because the effects of the interatomic interaction can essentially be included perturbatively . how this perturbation theory is performed is discussed in detail in section [ ifd ] , but before we can do that we need to understand the noninteracting gases first . in that case , we have @xmath161 = \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \\ & & \hspace*{0.2 in } \times~ \phi^*_{\alpha}({\bf x},\tau ) \left\ { \hbar \frac{\partial}{\partial \tau } \right . - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) + \epsilon_{\alpha } - \mu \right\ } \phi_{\alpha}({\bf x},\tau ) \nonumber\end{aligned}\ ] ] and the partition function is a gaussian integral , which explains why we were interested in gaussian integrals in section [ gv ] . it will be illustrative to evaluate this partition function in three different ways . in the first way , we perform the evaluation of the trace involved in the definition of the partition function at the end of the calculation , and start by considering the matrix element @xmath162 as the functional integral @xmath163d[\phi_{{\bf n},\alpha } ] \right)~ \exp \left\ { \sum_{{\bf n},\alpha } \phi^*_{{\bf n},\alpha}(\hbar\beta ) \phi_{{\bf n},\alpha}(\hbar\beta ) \right\ } \nonumber \\ & & \times \exp \left\{- \frac{1}{\hbar } \int_0^{\hbar\beta } d\tau \sum_{{\bf n},\alpha } \phi^*_{{\bf n},\alpha}(\tau ) \left ( \hbar \frac{\partial}{\partial \tau } + \epsilon_{{\bf n},\alpha } - \mu \right ) \phi_{{\bf n},\alpha}(\tau ) \right\}~ , \nonumber\end{aligned}\ ] ] which is the product for each @xmath164 and @xmath165 of the path integral @xmath166d[\phi]~ \exp \left\ { \phi^*(\hbar\beta ) \phi(\hbar\beta ) - \frac{1}{\hbar } \int_0^{\hbar\beta } d\tau~ \phi^*(\tau ) \left ( \hbar \frac{\partial}{\partial \tau } + \epsilon - \mu \right ) \phi(\tau ) \right\ } \nonumber\end{aligned}\ ] ] with the boundary conditions @xmath167 and @xmath168 . it clearly corresponds to the matrix element @xmath169 for one particular value of @xmath164 and @xmath165 . we calculate this matrix element by performing a shift in the integration variables , i.e. , @xmath170 and @xmath171 , where @xmath172 obeys the ` classical ' equations of motion @xmath173}{\delta \phi^*(\tau ) } \right|_{\phi=\phi_{\rm cl } } = \left ( \hbar \frac{\partial}{\partial \tau } + \epsilon - \mu \right ) \phi_{\rm cl}(\tau ) = 0\ ] ] and similarly for @xmath174 . the solutions with the correct boundary solutions are @xmath175 and @xmath176 , leading to the path integral @xmath177d[\xi]~ \exp \left\ { - \frac{1}{\hbar } \int_0^{\hbar\beta } d\tau~ \xi^*(\tau ) \left ( \hbar \frac{\partial}{\partial \tau } + \epsilon - \mu \right ) \xi(\tau ) \right\ } \nonumber\end{aligned}\ ] ] with the boundary conditions @xmath178 . this means that the last path integral is just equal to @xmath179 and that the desired result is just the prefactor . substituting this for each value of @xmath164 and @xmath165 , we apparently have @xmath180 and the partition function becomes the product @xmath181 this is the correct result , because from the usual thermodynamic identity @xmath182 we find @xmath183 as desired . the second way is easier and more common in practice . we immediately start with @xmath184d[\phi]~ e^{-s_0[\phi^*,\phi]/\hbar}\ ] ] and incorporate the boundary conditions by expanding the fields as @xmath185 where @xmath186 for bosons and @xmath187 for fermions . these are known as the even and odd matsubara frequencies , respectively . using this expansion we have @xmath188 if we also take account of the jacobian involved in the change of integration variables . note that the difference between the jacobians in the bosonic and fermionic case , is a consequence of the fact that for grassmann variables we have that @xmath189 instead of the result @xmath190 that we expect on the basis of ordinary complex integration . note also that these are all again gaussian integrals , so we find in first instance @xmath191 to evaluate the sum over matsubara frequencies we need to add a convergence factor @xmath192 and finally take the limit @xmath193 . the precise reason for this particular procedure can not be fully understood at this point but is explained in section [ ifd ] . however , doing so we indeed find that @xmath194 to see that this is correct , we differentiate the latter equation with respect to @xmath195 . this gives us @xmath196 which can be proved by contour integration in the following way . the function @xmath197 has poles at the even and odd matsubara frequencies with residu @xmath198 . hence , by cauchy s theorem the left - hand side is equal to @xmath199 with @xmath200 a contour that fully encloses the imaginary axis in the direction shown in figure 3 . adding to the contour @xmath200 the contour @xmath201 , which gives no contribution to the integral , we can again apply cauchy s theorem to obtain the desired result . why does the integration over the contour @xmath201 vanish ? the reason for this is that the integrant behaves as @xmath202 for @xmath203 and as @xmath204 for @xmath205 . the integrant thus always vanishes much faster than @xmath206 on the contour @xmath201 for any @xmath207 . the third and last method is simplest and , therefore , most common in practice . we first rewrite the partition function as @xmath208d[\phi]~ \exp \left\ { \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x } \right . \\ & & \times \left . \sum_{\alpha ' } \int_0^{\hbar\beta } d\tau ' \int d{\bf x}'~ \phi^*_{\alpha}({\bf x},\tau ) g^{-1}_{\alpha,\alpha'}({\bf x},\tau;{\bf x}',\tau ' ) \phi_{\alpha'}({\bf x}',\tau ' ) \right\ } \nonumber\end{aligned}\ ] ] and see @xmath209 as a ` matrix ' both in spin space as in coordinate space . we then know that this gaussian integral is just @xmath210^{\mp 1 } = \exp \left\ { \mp { \rm tr}[\ln ( -g^{-1 } ) ] \right\}~.\ ] ] clearly , we have from the action @xmath211 $ ] that @xmath212 or equivalently that @xmath213 which means that @xmath214 is a green s function . we come back to its physical meaning shortly . for now we only need to see that the last equation is solved by @xmath215 thus again @xmath216 in principle , we have slightly cheated in the last step of this derivation , because equation ( [ green ] ) shows that @xmath217 the problem with this last way of calculating the functional integral is that it does not correctly account for the jacobian involved in the change of variables from imaginary time @xmath218 to the matsubara frequencies @xmath219 . however , this problem can be avoided by calculating never a single determinant , but always the ratio of two determinants . in this manner the effect of the jacobian is exactly cancelled . we will see some examples of how this works in sections [ hft ] and [ sc ] . we now return to the physical meaning of the above green s function . to see its meaning , we first consider the time - ordered expectation value @xmath220 \rangle \\ & & \equiv \theta(\tau-\tau ' ) \langle \hat{\psi}_{\alpha}({\bf x},\tau ) \hat{\psi}^{\dagger}_{\alpha'}({\bf x}',\tau ' ) \rangle \pm \theta(\tau'-\tau ) \langle \hat{\psi}^{\dagger}_{\alpha'}({\bf x}',\tau ' ) \hat{\psi}_{\alpha}({\bf x},\tau ) \rangle~. \nonumber\end{aligned}\ ] ] here the expectation value is taken in the grand - canonical ensemble and @xmath221 is the imaginary time heisenberg operator , which is defined by @xmath222 and therefore obeys the heisenberg equation of motion @xmath223_-$ ] . for the noninteracting case it reads @xmath224 as a result @xmath225 \rangle = \hbar \delta(\tau-\tau ' ) \langle [ \hat{\psi}_{\alpha}({\bf x},\tau ) , \hat{\psi}^{\dagger}_{\alpha'}({\bf x}',\tau')]_{\mp } \rangle \hspace*{0.4 in } \\ & & \hspace*{1.8 cm } + \left ( \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } - v^{\rm ex}({\bf x } ) - \epsilon_{\alpha } + \mu \right ) \langle t[\hat{\psi}_{\alpha}({\bf x},\tau ) \hat{\psi}^{\dagger}_{\alpha'}({\bf x}',\tau ' ) ] \rangle~. \nonumber\end{aligned}\ ] ] substituting the equal - time relations that we derived in section [ sq ] , i.e. , @xmath226_{\mp } = \delta({\bf x}-{\bf x } ' ) \delta_{\alpha,\alpha'}$ ] , thus suggests that @xmath227 \rangle~.\ ] ] we can actually also prove this important relation , that bridges the gap between the functional formulation of quantum field theory used here and the more familiar operator formalism . first of all , it is clear from the slicing procedure used in our derivation of the functional integral in section [ fi ] , that it automatically leads to time - ordered expectation values . we should , therefore , only be able to prove that @xmath228d[\phi]~ \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) e^{-s_0[\phi^*,\phi]/\hbar}~. \nonumber\end{aligned}\ ] ] this is most easily achieved in the following way . we introduce a partition function in the presence of external currents @xmath229 and @xmath230 , where in the fermionic case these currents are also grassmann variables . the partition function is taken to be @xmath231 = \int d[\phi^*]d[\phi]~ \exp \left\ { - \frac{1}{\hbar } s_0[\phi^*,\phi ] \raisebox{0.25 in } { } \right . \\ & & \hspace*{0.15 in } + \left . \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ ( \phi^*_{\alpha}({\bf x},\tau ) j_{\alpha}({\bf x},\tau ) + j^*_{\alpha}({\bf x},\tau ) \phi_{\alpha}({\bf x},\tau ) ) \right\}~ , \nonumber\end{aligned}\ ] ] since then we must simply prove that @xmath232 using a short - hand notation we have @xmath233 = \int d[\phi^*]d[\phi]~ \exp \left\ { ( \phi|g^{-1}|\phi ) + ( \phi|j ) + ( j|\phi ) \right\}~.\ ] ] the terms in the exponent can be rewritten as @xmath234 , which is usually called completing the square . performing a shift in the integration variables , we then easily see that @xmath235 = z_0[0,0 ] e^{-(j|g|j ) } = z_0[0,0 ] \exp \left\ { - \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x } \right . \hspace*{0.4 in } \\ & & \hspace*{0.08 in } \times \left . \sum_{\alpha ' } \int_0^{\hbar\beta } d\tau ' \int d{\bf x}'~ j^*_{\alpha}({\bf x},\tau ) g_{\alpha,\alpha'}({\bf x},\tau;{\bf x}',\tau ' ) j_{\alpha'}({\bf x}',\tau ' ) \right\}~ , \nonumber\end{aligned}\ ] ] which after differentiation indeed leads to the desired result . we can now in fact calculate the expectation value of the time - ordered product of any number of operators . with the above expression for @xmath236 $ ] one can easily prove that this results in the sum of all possible products of time - ordered expectation values of two operators . for instance @xmath237 this is the famous wick s theorem , which plays a crucial role in the next section where we start to discuss the profound effects that interactions can have on the results obtained thusfar . the green s function , or one - particle propagator , @xmath214 is one of the most important quantities that we want to determine theoretically , because it gives us the possibility to calculate in principle the expectation value of any one - particle observable . it also gives us the elementary excitations of our system of interest . for example , the average density of the spin state @xmath21 is in an ideal gas given by @xmath238 where we introduced the notation @xmath239 for the limit @xmath193 of @xmath240 . note that this procedure is necessary due to the time - ordening involved in the definition of the green s function and the fact that the field operators do not commute at equal times . it also gives a natural explanation for our previous use of the convergence factor @xmath192 . moreover , from the argument of the bose or fermi distribution function , we see that the elementary excitations have an energy of @xmath241 . if we replace in the fourier transform of the green s function , i.e. , in @xmath242 @xmath243 by @xmath244 we also see that @xmath245 has a pole at @xmath246 . this is a general result : poles in @xmath245 correspond to the energies of the elementary excitations . these energies in general can also have a negative imaginary components , which then correspond to the lifetime of the excitation . the question that arises , therefore , is how to determine also the green s function for an interacting system . this can be done in perturbation theory , as follows . we want to calculate the expectation value @xmath247d[\phi]~ \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) e^{-s[\phi^*,\phi]/\hbar } \nonumber\end{aligned}\ ] ] with the action @xmath248=s_0[\phi^*,\phi]+s_{\rm int}[\phi^*,\phi]$ ] . we now expand both the numerator and the denominator in powers of @xmath249 . up to first order we find for the partition function @xmath250d[\phi]~e^{-s_0[\phi^*,\phi]/\hbar } \left ( 1 - \frac{s_{\rm int}[\phi^*,\phi]}{\hbar } \right ) \\ & & \equiv z_0 \left ( 1 - \frac{1}{\hbar } \langle s_{\rm int}[\phi^*,\phi ] \rangle_0 \right)~. \nonumber\end{aligned}\ ] ] using wick s theorem , we thus have @xmath251 \rangle_0 = \frac{1}{2 } \sum_{\alpha,\alpha ' } \int_0^{\hbar\beta } d\tau \int d{\bf x } \int d{\bf x}'~ g_{0;\alpha,\alpha}({\bf x},\tau;{\bf x},\tau^+ ) \\ & & \hspace*{1.5 in } \times~ \frac{- v({\bf x}-{\bf x}')}{\hbar } g_{0;\alpha',\alpha'}({\bf x}',\tau;{\bf x}',\tau^+ ) \nonumber \\ & & \hspace*{0.52 in } \pm \frac{1}{2 } \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x } \int d{\bf x}'~ g_{0;\alpha,\alpha}({\bf x}',\tau;{\bf x},\tau^+ ) \nonumber \\ & & \hspace*{1.5 in } \times~ \frac{- v({\bf x}-{\bf x}')}{\hbar } g_{0;\alpha,\alpha}({\bf x},\tau;{\bf x}',\tau^+)~. \nonumber\end{aligned}\ ] ] ) up to first order in the interaction . ] to understand the general structure of the perturbation expansion , it is very convenient to represent equations ( [ z1 ] ) and ( [ z2 ] ) in terms of feynman diagrams . the final result is shown in figure 4 , where a wiggly line corresponds to the factor @xmath252 and a thin arrowed line pointing from @xmath253 to @xmath254 represents the noninteracting green s function @xmath255 . in figure 4 we have , for clarity sake , also explicitly indicated the various coordinates and spin degrees of freedom that we have to either integrate or sum over , respectively . it is , however , much more common in practice to suppress in feynman diagrams those degrees of freedom that have to be integrated or summed over , and denote only the degrees of freedom on which the quantity of interest depends explicitly . for example , for the numerator of equation ( [ expec ] ) we obtain @xmath256d[\phi]~ \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) e^{-s_0[\phi^*,\phi]/\hbar } \left ( 1 - \frac{s_{\rm int}[\phi^*,\phi]}{\hbar } \right ) \hspace*{0.4 in } \\ & & = z_0 \left ( - \langle \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) \rangle_0 + \frac{1}{\hbar } \langle \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) s_{\rm int}[\phi^*,\phi ] \rangle_0 \right ) , \nonumber\end{aligned}\ ] ] whose diagrammatic equivalent is then given in figure 5 . up to first order the interacting green s function , which is represented by a thick arrowed line , obeys the equation shown in figure 6 . note that the so - called disconnected diagrams have exactly cancelled . this is a general feature that happens for any expectation value , i.e. , the disconnected diagrams that occur in the numerator are always exactly cancelled by the denominator @xcite . one can also show that the general structure of the perturbation expansion of the green s function is as in figure 7 , in which only one - particle irreducible diagrams contribute to the selfenergy @xmath257 . these one - particle irreducible diagrams can be distinguished from the fact that they do not become disconnected if we cut a single thin arrowed line . this is dyson s equation @xcite . it is particularly insightful for a homogeneous gas or for a trapped gas in the weak - coupling limit , since then the one - particle states @xmath53 are not affected by the interactions and the exact one - particle propagator can be written as @xmath258 note that for realistic trapped atomic gases this weak - coupling limit is essentially always realized in the normal phase of the gas . it is , therefore , also of some experimental interest to consider that limit in more detail first . one then finds that the dyson equation in first instance becomes @xmath259 because @xmath260 and because we have assumed the interactions to be spin independent , we can easily convince ourselves that in perturbation theory it is always true that the selfenergy obeys @xmath261 and as a result also that up to all orders in the interaction @xmath262 . as we will see in section [ hft ] , however , nonperturbative effect can change this result due to the phenomena of spontaneous broken symmetry . ignoring the latter for the moment , we find for each spin state the uncoupled equation @xmath263 or @xmath264 hence @xmath265 which shows that the poles in the green s function are indeed shifted by the interactions . in our lowest order calculation we have found for the selfenergy the diagrammatic result shown in figure 8 . this is the most simple approximation we can think of . in the weak - coupling limit it is mathematically equal to @xmath266 the first term is known as the direct or hartree , and the second term as the exchange or fock contribution to the selfenergy . from equation ( [ self ] ) we conclude that the matrix elements of the interaction enter only in the combination @xmath267 . clearly , this is a refection of the pauli principle , which forces the effective interaction between two fermionic atoms in the same state @xmath37 to vanish . to make the theory selfconsistent we should use in the expression for the selfenergy not the noninteracting propegators , but precisely those that follow from the dyson equation this is the hartree - fock approximation , which is diagrammatically given in figure 9 . it leads in the weak - coupling limit to the new dispersion relation @xmath268 for the ` dressed ' particles or quasiparticles of the gas , where the selfenergy is essentially also found from equation ( [ self ] ) . we only have to replace in the right - hand side @xmath269 by @xmath270 . in this manner we have thus obtained an approximation to the interacting green s function that is nonperturbative in the interaction and effectively sums an infinite number of feynman diagrams . of course , the hartree - fock approximation can also be used in the strong - coupling limit . in that case it diagrammatically still corresponds to the solution of the dyson equation of figure 7 , with a selfenergy as shown in figure 9 . however , we are then no longer allowed to assume that the exact green s function is diagonal in the eigenstates @xmath53 of the external trapping potential . to see more explicitly what this means physically , it is convenient to rederive the hartree - fock theory without making use of perturbation theory . because the hartree - fock theory is very useful in a large number of circumstances , and because we want to illustrate a very useful technique which is nowadays often used in the literature , we are now going to reproduce the above results without making use of a diagramatic expansion . we consider for simplicity a fermionic mixture with an equal number of atoms in two hyperfine states and start by splitting our spin - independent interaction @xmath271 into two spin - dependent parts such that one contributes only to the hartree diagram and the other only to the fock diagram . denoting a spin - dependent interaction by @xmath272 we thus want that @xmath273 with @xmath274 . using operators in spin space , a possible solution to these equations is @xmath275 and @xmath276 where @xmath277 is the exchange operator of the spins of atoms @xmath65 and @xmath278 , and @xmath279 are the usual pauli matrices . clearly , we now have @xmath280 = \frac{1}{2 } \sum_{\alpha,\alpha';\beta,\beta ' } \int_0^{\hbar\beta } d\tau \int d{\bf x } \int d{\bf x}'~ \\ & & \times \left\{\phi^*_{\alpha'}({\bf x},\tau ) \phi_{\alpha}({\bf x},\tau ) v^h_{\alpha',\beta';\alpha,\beta}({\bf x}-{\bf x } ' ) \phi^*_{\beta'}({\bf x}',\tau ) \phi_{\beta}({\bf x}',\tau ) \raisebox{0.15 in } { } \right . \nonumber \\ & & \hspace*{0.5 in } - \left . \raisebox{0.15 in } { } \phi^*_{\alpha'}({\bf x},\tau ) \phi_{\beta}({\bf x}',\tau ) v^f_{\alpha',\beta';\alpha,\beta}({\bf x}-{\bf x } ' ) \phi^*_{\beta'}({\bf x}',\tau ) \phi_{\alpha}({\bf x},\tau ) \right\}~ , \nonumber\end{aligned}\ ] ] which we write as @xmath281 \equiv \frac{1}{2 } ( \phi^*\phi|v^h|\phi^*\phi ) - \frac{1}{2 } ( \phi^*\phi||v^f||\phi^*\phi)~.\ ] ] we next apply a so - called hubbard - stratonovich transformation @xcite to both the hartree and the fock parts of the interaction . first the hartree part . we note that @xmath282}$ ] can be written as a functional integral over the four real fields contained in @xmath283 as @xmath284~ \exp \left\ { \frac{1}{2\hbar } ( \kappa|{v^h}^{-1}|\kappa ) - \frac{1}{2\hbar } ( \kappa|\phi^*\phi ) - \frac{1}{2\hbar } ( \phi^*\phi|\kappa ) \right\}~. \nonumber\end{aligned}\ ] ] if we put @xmath285 = 0 $ ] for a moment , we can substitute this equality in the partition function to obtain @xmath286d[\phi^*]d[\phi ] e^{-s[\kappa,\phi^*,\phi]/\hbar}$ ] with the action @xmath287 = - \frac{1}{2 } ( \kappa|{v^h}^{-1}|\kappa ) + \sum_{\alpha,\alpha ' } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi^*_{\alpha}({\bf x},\tau ) \left\ { \left ( \hbar \frac{\partial}{\partial \tau } \right . \right . \hspace*{0.4 in } \\ & & \hspace*{1.0 cm } \left . \left . - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) + \epsilon_{\alpha } - \mu \right ) \delta_{\alpha,\alpha ' } + \kappa_{\alpha,\alpha'}({\bf x},\tau ) \right\ } \phi_{\alpha'}({\bf x},\tau)~. \nonumber\end{aligned}\ ] ] we see that in this manner the action for the fermions has become quadratic with a selfenergy @xmath288 . hence , we can now integrate out the fermion fields to obtain @xmath286 e^{-s^{\rm eff}[\kappa]/\hbar}$ ] and @xmath289 = - \frac{1}{2 } ( \kappa|{v^h}^{-1}|\kappa ) -\hbar { \rm tr}[\ln ( -g^{-1})]~,\ ] ] with @xmath290 also a functional of @xmath291 . up to now we have not made any approximations and have only performed an exact rewriting of the partition function . however , the action @xmath292 $ ] contains all powers of the fields @xmath291 and is thus quite complicated . it may therefore appear that we have not made any progress . this is not the case , because the partition function will be largest for configurations that minimize the action @xmath292 $ ] . to make use of this observation , we expand the action around its minimum , i.e. , we put @xmath293 and require that @xmath294/\delta\kappa_{\alpha,\alpha'}({\bf x},\tau ) |_{\kappa=\langle\kappa\rangle } = 0 $ ] . if we neglect the fluctuations , we find that @xmath295/\hbar}$ ] . this will turn out to be the hartree approximation , as we will show now . first of all we have @xmath296 substituting this in equation ( [ haction ] ) gives us for the terms linear in the fluctuations @xmath297 - ( \delta\kappa|{v^h}^{-1}|\langle\kappa\rangle)~. \nonumber\end{aligned}\ ] ] if @xmath298 is indeed a minimum of the action @xmath292 $ ] , these linear terms have to vanish , which implies that @xmath299 as promised , this is precisely the most general mathematical expression for the hartree contribution to the selfenergy in figure 9 . taking again @xmath300 , we also have that the hartree approximation to the one - particle propagator obeys @xmath301 , with @xmath302 the average total atomic density , and we obtain the usual hartree selfenergy @xmath303 now we again want to include the fock part of the interaction and treat this also by a hubbard - stratonovich transformation . this requires introducing four real fields that depend on two spatial coordinates . they are denoted by @xmath304 . schematically we then use @xmath305~ \exp \left\ { - \frac{1}{2\hbar } ( \lambda||{v^f}^{-1}||\lambda ) + \frac{1}{2\hbar } ( \lambda||\phi^*\phi ) + \frac{1}{2\hbar } ( \phi^*\phi||\lambda ) \right\}~ , \nonumber\end{aligned}\ ] ] which leads to the total selfenergy @xmath306 \delta(\tau-\tau')\ ] ] and after integration over the fermion fields to the effective action @xmath307 = - \frac{1}{2 } ( \kappa|{v^h}^{-1}|\kappa ) + \frac{1}{2 } ( \lambda||{v^f}^{-1}||\lambda ) - \hbar { \rm tr}[\ln ( -g^{-1})]~.\ ] ] requiring now that also @xmath308/ \delta\lambda_{\alpha,\alpha'}({\bf x},{\bf x}',\tau ) |_{\lambda=\langle\lambda\rangle } = 0 $ ] , then leads to the expected fock selfenergy @xmath309 in the symmetric case it simply becomes @xmath310 with @xmath311 and @xmath312 exactly the hartree - fock approximation to the off - diagonal part of the one - particle density matrix . note that to actually perform the hartree - fock calculation in the latter case , we need to be able to determine the green s function @xmath313 . the easiest way to do so , is by realizing that it is the green s function of the operator in the fermionic piece of the action @xmath314 $ ] obtained after the hubbard - stratonovich transformations . if we diagonalize this operator by solving the eigenvalue problem @xmath315 the desired one - particle propagator aquires the ideal gas form @xmath316 with new one - particle energies @xmath317 and eigenstates @xmath318 that incorporate the average effect of the interactions of an atom with all the other atoms in the gas . clearly , in the cases that the eigenstates are not affected by these so - called mean - field effects , we recover the weak - coupling results of the previous section . although we have thus precisely reproduced our diagrammatic result , there are two important advantages in using the hubbard - stratonovich transformation . first , it is in principle exact , and allows us to also calculate corrections to the hartree - fock approximation . for example , if we expand @xmath319 $ ] up to quadratic order in @xmath320 and @xmath321 , and neglect all higher orders , we find the so - called generalized random phase approximation . the latter approach actually gives us also the opportunity to study the density fluctuations and therefore the collective excitations of the gas . we return to this important topic in section [ modes ] . second , it allows for a beautiful way to describe phase transitions . as mentioned previously , in perturbation theory we always find for a homogeneous gas that @xmath322 due to the same feature of @xmath323 . from a fundamental point of view this is a result of the translational invariance of the gas and of the rotational symmetry in spin space . however , we can imagine that in principle we can also find selfconsistent solutions that do not have this property . we then have a spontaneous breaking of symmetry and therefore a phase transition in our system . for example , if below a certain temperature @xmath324 , we are dealing with a transition to a charge density wave or ( wigner ) crystal . if on the other hand @xmath325 the gas is in a ferromagnetic phase . for a spin - density wave we even have that @xmath326 . in all these cases the hubbard - stratonovich approach used above leads in a natural way to the appropriate landau theory of the phase transition . since the landau theory is also very useful for the understanding of superfluidity in atomic gases , which is clearly the primary goal of the present course , we use the next section to give a short introduction into this subject . in this summary of the landau theory of phase transitions , we restrict ourselves in first instance to the homogeneous case because this makes the discussion more transparent . however , at the end of the section we also briefly mention how the inhomogeneity enters the theory . we have seen , for example in the hartree - fock theory discussed above , that the green s function @xmath327 can have a nonzero value of the average spin density @xmath328 , which is usually called the magnetization @xmath329 in this context . this signals a phase transition to a ferromagnetic phase and the magnetization is called the order parameter of this transition . in the previous section we have also seen how we can , in principle exactly , obtain an expression for the partition function as a functional integral over the field @xmath330 , i.e. , @xmath331~e^{-s^{\rm eff}[{\bf s}]/\hbar}~.\ ] ] we just have to integrate out the fields @xmath332 and @xmath333 . for space and time independent values of the magnetization @xmath5 the action @xmath334 $ ] is @xmath335 , where @xmath336 is the landau free - energy density and @xmath337 is the total volume of system . because of the symmetry of @xmath248 $ ] under rotations of the spin , this free - energy density must also be rotationally invariant and can therefore only depend on the magnitude of @xmath5 , which we denote by @xmath338 . if a phase transition occurs the behaviour of @xmath339 can essentially fall only into two categories . at temperatures very high compared to the critical temperature @xmath340 , the system is fully disorded and the free energy @xmath339 must have a single minimum at @xmath341 to make sure that the order parameter @xmath342 is zero . bringing the temperature closer to @xmath340 , however , the free energy can develop a second local minimum . as long as the free energy in this second local minimum is higher than the minimum at @xmath341 , the equilibrium value of @xmath342 will still be zero and no phase transition has occured . lowering the temperature further , the value of the free energy in the second minimum decrease until , precisely at the critical temperature @xmath340 , it is equal to the free energy at @xmath341 . for temperatures below this critical one the second minimum has actually become the global minimum of the free energy , which implies that @xmath343 and we are in the ordered phase . in this scenario the order parameter has always a discontinuity at the critical temperature . as a result , this corresponds to a discontinuous , or first - order , phase transition . the behaviour of the free energy and the order parameter is illustarted in figure 10 . this should be compared with the behaviour of the free energy and the order parameter for a continuous , i.e. , second or higher order , phase transition , which is quite different and depicted in figure 11 . now the landau free energy @xmath339 has always a single minimum , which for temperatures above @xmath340 is at @xmath341 but at temperatures below the critical temperature shifts to a nonzero value of @xmath338 . in particular , the order parameter thus shows no discontinuity at @xmath340 . in the case of a second - order phase transition , we conclude that near the critical temperature @xmath342 is very small . as a result we can then expand the landau free - energy density in powers of @xmath338 . because the free energy must also be symmetric under @xmath344 , we have @xmath345 with @xmath346 and @xmath347 . thus if @xmath348 we have @xmath349 and @xmath350 . but for @xmath351 , @xmath352 becomes negative and we have @xmath353 note that the free - energy density in this minimum is @xmath354 and has a discontinuity in its second derivative with respect to the temperature . historically , this is the reason why the corresponding phase transition was named to be of second order . in fact , the landau theory of second order phase transitions is slightly more involved , because it also considers slow spatial fluctuations in @xmath5 . since a uniform rotation of @xmath5 costs no energy , we must have that the free energy is @xmath355 = \int d{\bf x}~ \left ( \gamma ( \mbox{\boldmath $ \nabla$ } { \bf s}({\bf x}))^2 + \alpha(t ) |{\bf s}({\bf x})|^2 + \frac{\beta}{2 } |{\bf s}({\bf x})|^4 \right)\ ] ] and @xmath356~e^{-\beta f_l[{\bf s}]}$ ] . landau theory now esentially corresponds to minimizing @xmath357 $ ] and taking only quadratic fluctuations into account . taking also higher order fluctuations into account turns out to be very difficult . it requires renormalization group methods , which we are not going to treat here and that can be found in the literature @xcite . the effects of these fluctuation corrections to landau theory are generally known as critical phenomena . interestingly , they do not occur in first - order phase transitions . more importantly for our purposes is , however , that for atomic gases the fluctuations are only important in a small temperature interval around the critical temperature . for many applications it is , therefore , possible to neglect them . what usually can not be negected is the effect of the inhomogeneity of the gas . in the context of landau theory that implies that the coefficients @xmath352 , @xmath358 and @xmath359 in equation ( [ lfe ] ) become also dependend on the spatial position in the trapping potential . we will see several examples of this feature in the following . we finally want to consider two important second - order phase transitions , that are purely due to quantum effects and can be conveniently treated with the methods that we have developed sofar . moreover , they occur very often in nature , for example in metals , in liquid helium and recently of course also in atomic gases of rubidium @xcite , lithium @xcite , sodium @xcite and hydrogen @xcite . let us first consider the last case , which is associated with a gas of spin - less bosons . for the low temperatures of interest the action is @xmath151 \\ & & = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi^*({\bf x},\tau ) \left\ { \hbar \frac{\partial}{\partial\tau } - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu \right\ } \phi({\bf x},\tau ) \nonumber \\ & & + \frac{1}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x } ~ v_{{\bf 0 } } \phi^*({\bf x},\tau ) \phi^*({\bf x},\tau ) \phi({\bf x},\tau ) \phi({\bf x},\tau)~ , \nonumber\end{aligned}\ ] ] where @xmath271 is taken to be equal to @xmath360 . the justification for this simplificaties is , roughly speaking , that the thermal de broglie wavelength @xmath361 of the atoms is much larger than the range of the interatomic interaction . in this system the phase transition of interest is bose - einstein condensation . the associated order parameter is @xmath362 , since for time independent @xmath363 the above action has presicely the form of a landau theory with a ` free - energy ' of @xmath364 \\ & & = \int d{\bf x}~ \left ( \frac{\hbar^2}{2 m } |\mbox{\boldmath $ \nabla$}\phi({\bf x})|^2 + ( v^{\rm ex}({\bf x})-\mu ) |\phi({\bf x})|^2 + \frac{v_{{\bf 0}}}{2 } |\phi({\bf x})|^4 \right)~. \nonumber\end{aligned}\ ] ] we conclude that in lowest order the critical temperature is determined by @xmath365 , because then the configurations @xmath366 precisely make a vanishing contribution to the quadratic part of the ` free energy ' . this condition makes sense , because it is exactly the condition that we have a bose - einstein condensation in the ideal case : in the ideal bose gas the number of particles in the one - particle ground state is @xmath367 , which indeed diverges for @xmath368 . to determine the corrections to this result we now explicitly substitute @xmath369 into our functional integral . it is for lateron important to realize that to consistently define the fluctuations @xmath370 in this manner , we also have to require that @xmath371 the physical reason behind this condition is that @xmath370 should contain all the configurations that are orthogonal to @xmath372 . in principle , therefore , it should also contain configurations that in effect only multiply @xmath372 by a global phase . such fluctuations lead to the phenomena of phase ` diffusion ' and are discussed in section [ pd ] . in full detail we find after the above substitution that @xmath373 = \hbar\beta f_l[\phi_{\bf 0}^*,\phi_{\bf 0 } ] + s_0[{\phi'}^*,\phi ' ] + s_{\rm int}[{\phi'}^*,\phi']~,\ ] ] where the linear and quadratic terms are given by @xmath374 \\ & & = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ { \phi'}^*({\bf x},\tau ) \left\ { - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu + v_{{\bf 0 } } |\phi_{\bf 0}({\bf x})|^2 \right\ } \phi_{\bf 0}({\bf x } ) \nonumber \\ & & + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi'({\bf x},\tau ) \left\ { - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu + v_{{\bf 0 } } |\phi_{\bf 0}({\bf x})|^2 \right\ } \phi_{\bf 0}^*({\bf x } ) \nonumber \\ & & + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ { \phi'}^*({\bf x},\tau ) \nonumber \\ & & \hspace*{0.85 in } \times \left\ { \hbar \frac{\partial}{\partial\tau } - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu + 2v_{{\bf 0}}|\phi_{\bf 0}({\bf x})|^2 \right\ } \phi'({\bf x},\tau ) \nonumber \\ & & + \frac{1}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } ( \phi_{\bf 0}({\bf x}))^2 { \phi'}^*({\bf x},\tau ) { \phi'}^*({\bf x},\tau ) \nonumber \\ & & + \frac{1}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } ( \phi_{\bf 0}^*({\bf x}))^2 \phi'({\bf x},\tau ) \phi'({\bf x},\tau)~ , \nonumber\end{aligned}\ ] ] and the cubic and quartic terms by @xmath375 = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } \phi_{\bf 0}({\bf x } ) { \phi'}^*({\bf x},\tau ) { \phi'}^*({\bf x},\tau ) \phi'({\bf x},\tau ) \hspace{0.4 in } \\ & & \hspace*{0.1 in } + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } \phi_{\bf 0}^*({\bf x } ) { \phi'}^*({\bf x},\tau ) \phi'({\bf x},\tau ) \phi'({\bf x},\tau ) \nonumber \\ & & \hspace*{0.1 in } + \frac{1}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } { \phi'}^*({\bf x},\tau ) { \phi'}^*({\bf x},\tau ) \phi'({\bf x},\tau ) \phi'({\bf x},\tau)~ , \nonumber \end{aligned}\ ] ] respectively . in the bogoliubov approximation we neglect the last three interaction terms @xcite . furthermore , to make sure that @xmath376 or @xmath377 , we need to require that the terms linear in @xmath378 and @xmath379 drop out of the action @xmath380 $ ] . clearly , this implies that @xmath381 which is the same result as obtained from minimizing the landau ` free - energy ' @xmath382 $ ] . in the context of trapped atomic gases , this equation is known as the gross - pitaevskii equation @xcite . it determines the macroscopic wave function of the condensate . the reason for calling @xmath372 the macroscopic wave function follows from the fact that the total density of the gas now obeys @xmath383 the total number of condensate atoms thus equals @xmath384 . as equation ( [ dens ] ) shows , it is in general always smaller that the total number of atoms in the gas due to the effect of the fluctuations . note that in our present formulation the average @xmath385 physically describes not only the depletion of the condensate due to the usual thermal fluctuations known from the ideal bose gas , but also due to the interactions , i.e. , purely due to quantum fluctuations . assuming that we have solved for the gross - pitaevskii equation , the fluctuation corrections are in the bogoliubov theory determined by a quadratic action of the form @xmath386 \\ & & = - \frac{\hbar}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left [ { \phi'}^*({\bf x},\tau ) , \phi'({\bf x},\tau ) \right ] \cdot { \bf g}^{-1 } \cdot \left [ \begin{array}{c } \phi'({\bf x},\tau ) \\ { \phi'}^*({\bf x},\tau ) \end{array } \right]~ , \nonumber\end{aligned}\ ] ] where the associated green s function @xmath387 has now a matrix structure because not only the normal average @xmath388 but also the anomalous average @xmath389 is now unequal to zero . we thus have that @xmath390 \cdot \left [ { \phi'}^*({\bf x}',\tau ' ) , \phi'({\bf x}',\tau ' ) \right ] \right\rangle~.\ ] ] from equation ( [ s0 ] ) we in fact find that in the bogoliubov approximation @xmath391 \delta({\bf x}-{\bf x } ' ) \delta(\tau-\tau')~. \nonumber\end{aligned}\ ] ] with the noninteracting green s function @xmath392 defined by @xmath393\ ] ] and , of course , @xmath394 this is clearly only the lowest order result for the green s function , because a perturbative treatment of @xmath395 $ ] leads , in the same way as in section [ ifd ] , to higher order corrections . in general , the dyson equation is , however , always of the form @xmath396^{-1 } = \left [ \begin{array}{cc } g_0^{-1 } & 0 \\ 0 & g_0^{-1 } \end{array } \right ] - \left [ \begin{array}{cc } \sigma_{11 } & \sigma_{12 } \\ \sigma_{21 } & \sigma_{22 } \end{array } \right]~.\ ] ] the off - diagonal elements are again called anomalous , since they vanish in the normal phase of the gas . diagrammatically the dyson equations for @xmath397 and @xmath398 are shown in figure 12 . the selfenergy in the bogoliubov approximation is @xmath399 \delta({\bf x}-{\bf x } ' ) \delta(\tau-\tau')~.~~\end{aligned}\ ] ] because we have completely negelected the cubic and quartic terms in the action to obtain this selfenergy , the bogoliubov approach is only valid if the fluctuations are sufficiently small . physically , this impies that the deplection of the condensate must be small . it can , therefore , not be applied to liquid helium , but is expected to be valid for a weakly - interacting atomic gas at such low temperature that it essentially only consists of a condensate . under these conditions the bogoliubov theory not only predicts the condensate density profile @xmath400 but also the collective modes of the condensate . both these predictions have been accurately verified in recent experiments @xcite . theoretically , the eigenfrequencies of the collective modes are again determined by the poles in the one - particle green s function . to understand more clearly how these can be determined , we first consider a homogeneous bose gas . in a box with volume @xmath401 the one - particle states are most easily characterized by the wavevector @xmath402 and equal to @xmath403 . the one - particle energies are thus @xmath404 . moreover , the gross - pitaevskii equation reduces to @xmath405 in that case , because the landau ` free - energy ' is minimized for a macroscopic wave function that is independent of the position in the box . making use of this fact , equation ( [ gb ] ) can immediately be solved by a fourier transformation . the result is @xmath406~. \nonumber\end{aligned}\ ] ] clearly there are poles in @xmath407 if the determinant of the right - hand side is zero or if @xmath408 this is the famous bogoliubov dispersion of the collective excitations . note that to finish the calculation we still have to obtain the condensate density @xmath409 . this is determined by the total density of the gas , which obeys @xmath410 with @xmath411 the density of the ` above ' condensate particles . by inverting the right - hand side of equation ( [ gbf ] ) , we find that the noncondensate density is equal to @xmath412 for a given density and temperature , the last two equations thus fully determine the condensate density . note that equation ( [ ncd ] ) explicitly shows that the condensate is indeed depleted both by thermal as well as quantum effects . the generalization to the inhomogeneous case is straightforward . first we again have to solve the gross - pitaevskii equation at a fixed chemical potential . given the condensate wave function , we can then calculate the collective mode frequencies by finding the poles of @xmath387 , or equivalently but more conveniently , the zero s of @xmath413 . clearly , the latter are located at @xmath414 , where @xmath415 is found from the eigenvalue problem @xmath416 \cdot \left [ \begin{array}{c } u_{\bf n}({\bf x } ) \\ v_{\bf n}({\bf x } ) \end{array } \right ] \hspace*{0.4 in } \\ = \hbar\omega_{\bf n } \left [ \begin{array}{cc } 1 & 0 \\ 0 & - 1 \end{array } \right ] \cdot \left [ \begin{array}{c } u_{\bf n}({\bf x } ) \\ v_{\bf n}({\bf x } ) \end{array } \right]~ \nonumber \end{aligned}\ ] ] and we introduced the operator @xmath417 . this is de bogoliubov - de gennes equation that has recently been applied with great succes to the collective modes of a bose condensed rubidium and sodium gas @xcite . note that a special solution with @xmath418 is given by @xmath419 = [ \phi_{\bf 0}({\bf x}),-\phi^*_{\bf 0}({\bf x})]$ ] . as a result we can , by making use of the fact that the left - hand side of the bogoliubov - de gennes equation involves a hermitian operator , easily proof that all the solutions obey @xmath420 as required by the condition in equation ( [ ort ] ) . moreover , we can similary show that the solutions with @xmath421 can always be normalized as @xcite @xmath422 physically , the zero frequency solution @xmath423 describes the dynamics of the global phase of the condensate @xcite . because of the so - called @xmath424 symmetry of the action , i.e. , its invariance under the simultaneous phase changes @xmath425 and @xmath426 , this solution is essentially of no importance to the thermodynamic properties of a macroscopic gas sample and is therefore usually neglected . nevertheless , it has from a fundamental point of view some interesting consequences , as we will see in section [ pd ] . knowing all the eigenstates of @xmath413 we can then easily perform the inversion and finally again determine the density profile of the noncondensed atoms . keeping the physical significance of the zero frequency mode in mind , we ultimately find , @xmath427 which may be compared with equation ( [ ncd ] ) . at temperatures near absolute zero , we have as a good approximation that @xmath428 and the bogoliubov theory applies . however at nonzero temperatures we thermally excite particles and @xmath429 becomes nonzero . if we treat the effect of the noncondensate part of the gas in the hartree - fock approximation , we find that @xmath395 $ ] on average adds @xmath430 = 2 \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } n'({\bf x } ) { \phi'}^*({\bf x},\tau ) \phi_{\bf 0}({\bf x } ) \\ & & + 2 \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } n'({\bf x } ) \phi'({\bf x},\tau ) \phi_{\bf 0}^*({\bf x } ) \nonumber \\ & & + 2 \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } n'({\bf x } ) { \phi'}^*({\bf x},\tau ) \phi'({\bf x},\tau ) ~ , \nonumber \end{aligned}\ ] ] to the action @xmath380 $ ] . in figure 13 we indicate how this can be understood diagramatically . performing the same analysis as before , we conclude that the gross - pitaevskii equation is modified to @xmath431 and the normal selfenergies are changed into @xmath432 . the bogoliubov - de gennes equation for the elementary excitations is , therefore , now given by @xmath433 \cdot \left [ \begin{array}{c } u_{\bf n}({\bf x } ) \\ v_{\bf n}({\bf x } ) \end{array } \right ] \hspace*{0.4 in } \\ = \hbar\omega_{\bf n } \left [ \begin{array}{cc } 1 & 0 \\ 0 & - 1 \end{array } \right ] \cdot \left [ \begin{array}{c } u_{\bf n}({\bf x } ) \\ v_{\bf n}({\bf x } ) \end{array } \right]~. \nonumber \end{aligned}\ ] ] these last two equations in combination with equation ( [ ncda ] ) are known as the popov theory in the recent literature @xcite . it is much studied at present in the context of bose - einstein condensation in atomic gases , and has been applied with succes to the equilibrium density profile of the gas below the critical temperature @xcite . it has also been used to determine the collective mode frequencies of the gas at nonzero temperatures , however , with much less success @xcite . the reason for the failure of the popov theory in this case is that the bogoliubov - de gennes equation in equation ( [ bdgt ] ) describes physically only the motion of the condensate in the presence of a static noncondensed cloud and not the dynamics of the noncondensed cloud itself . how that can be achieved is discussed lateron in section [ modes ] , when we have obtained a better understanding of the nonequilibrium theory . now we briefly want to make a connection to the interaction parameter @xmath434 used in the bogoliubov and popov theories and the specific two - body scattering properties of the atomic gas of interest in a particular experiment . in the previous section we mentioned that the bogoliubov theory , and to a certain extent also the popov theory , agree very well with experiment . it is clear , however , that to apply these theories to an actual experiment we need to know the interaction parameter @xmath434 . in the case of atomic gases it is indeed possible to perform an _ ab initio _ calculation of this quantity , something which for instance can not be done for liquid helium . the reason why @xmath434 can be determined for atomic gases , is that under the experimental conditions of interest the densities are always so low that we only need to consider all two - body processes taking place in the gas and we can neglect three - body and higher - body processes . this implies physically that we only have to calculate and add the quantum mechanical amplitudes for two atoms to scatter of each other an arbitrary number of times . diagrammatically the procedure can essentially be summarized by the t - matrix equation in figure 14 , because by iteration of this equation we easily see that we are indeed summing all two - body interacion processes . denoting the total momentum of the two incoming particles by @xmath435 and the sum of the two matsubara frequencies by @xmath436 , the t - matrix equation in figure 14 can , with our knowledge of how to perform the matsubara sum in the right - hand side , easily be shown to be mathematically equivalent to @xmath437 it is immediately solved by @xmath438 note that this is a somewhat formal result , because the momentum sum in the right - hand side has an ultra - violet divergence , which from a field - theoretical point of view requires a renormalization procedure . the physical reason for the divergence is that we have used the approximate potential @xmath439 instead of the actual interactomic potential @xmath271 . the argument for using the @xmath440-function approximation was that the thermal de broglie wavelength of the atoms is for the ultra - low temperatures of interest always much larger than the typical range of the interaction . we now see that this argument is not fully correct , because if we calculate corrections in perturbation theory , we have to deal with momentum sums which are not always restricted to momenta of order @xmath441 and are therefore sensitive to the precise details of the interaction potential . to cure this disease we note that if we put in equation ( [ tmatrix ] ) the bose occupation numbers equal to zero , we precisely get the t - matrix or lippmann - schwinger equation @xcite for two atoms interacting with the potential @xmath439 . furthermore , if the atoms interact with the potential @xmath271 the solution of the corresponding lippmann - schwinger equation is known from elementary scattering theory to be equal to @xmath442 for small incoming momenta and energies , where @xmath443 is the @xmath338-wave scattering length . we thus conclude that we must interpret @xmath444 as being equal to @xmath445 and , therefore , that the desired t - matrix in principle obeys @xmath446 to obtain an accurate theory for a trapped atomic bose gas that includes the effect of all two - body processes , we should now use instead of @xmath434 the above t - matrix in the bogoliubov or popov theories . of course , in the normal phase the same is true for the hartree - fock theory . if the temperature is not too close to the critical temperature for bose - einstein condensation , it turns out that @xmath447 and we arrive at the conclusion that for an application to realistic atomic gases we must replace everywhere in section [ sf ] the interaction parameter @xmath434 by @xmath442 . for temperatures close to the critical temperature , this is however no longer true and the bose distribution functions have a nonnegligible effect of the t - matrix @xcite . the theory that , by using ` dressed ' one - particle propagators in figure 14 , selfconsistently includes these effects of the medium on the scattering properties of the atoms is known as the many - body t - matrix theory . it results in an effective interaction that , to avoid confusion , is denoted by @xmath448 in section [ cf ] . as we will see then , it plays a crucial role if we want to arrive at a consistent discription of condensate formation in an atomic bose gas . finally , we want to briefly discuss the bardeen - cooper - schrieffer or bcs theory of superconductivity @xcite , which has received considerable attention recently in connection with ongoing experiments with the fermionic isotope of lithium . the reasons for this will become clear shortly . we are in that case dealing with effective spin @xmath449 fermions and the action is taken to be @xmath151 \\ & & = \sum_{\alpha=\uparrow,\downarrow } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi^*_{\alpha}({\bf x},\tau ) \left\ { \hbar \frac{\partial}{\partial\tau } - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } - \mu \right\ } \phi_{\alpha}({\bf x},\tau ) \nonumber \\ & & + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{\bf 0 } \phi^*_{\uparrow}({\bf x},\tau ) \phi^*_{\downarrow}({\bf x},\tau ) \phi_{\downarrow}({\bf x},\tau ) \phi_{\uparrow}({\bf x},\tau)~. \nonumber\end{aligned}\ ] ] note that the use of a single chemical potential implies that we only consider the optimal case of an equal density in each hyperfine state . this situation is optimal in the sense that , for unequal densities in each hyperfine state , it is no longer possible to pair up all the atoms in the gas . as a result the critical temperature of the gas drops dramatically , and in general becomes experimentally inaccessible . the effect is to a large extent analogous to putting superfluid @xmath450he in an homogeneous magnetic field . furthermore , we in first instance consider the homogeneous case , because we want to illustrate in this section the local - density approximation to include the effect of the external trapping potential . physically , this approximation treats the gas as consisting of a large number of independent gases that are in diffusive equilibrium with each other . such an approach only works if the correlation length of the gas is much smaller than the typical length scale associated with changes of the external potential . fortunately , this is almost always the case for realistic trapped atomic gases and for that reason the local - density approximation is often used in practice . the bcs theory is the theory of bose - einstein condensation of so - called cooper pairs . this means that the order parameter is @xmath451 , in analogy with the order parameter @xmath362 for the bose case just discussed . furthermore , it requires that the interaction parameter @xmath434 is negative , since otherwise the formation of pairs would not be energetically favorable . from now on we assume , therefore , that this is the case . the condensate of cooper pairs can also be nicely treated with a hubbard - stratonovich transformation . we now introduce a complex field @xmath452 and use @xmath453d[\delta]~ \exp \left\ { \frac{1}{\hbar } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left ( \frac{|\delta({\bf x},\tau)|^2}{v_{\bf 0 } } \right . \nonumber \\ & & + \left . \left . \raisebox{0.2 in } { } \delta^*({\bf x},\tau ) \phi_{\downarrow}({\bf x},\tau ) \phi_{\uparrow}({\bf x},\tau ) + \phi^*_{\uparrow}({\bf x},\tau ) \phi^*_{\downarrow}({\bf x},\tau ) \delta({\bf x},\tau ) \right ) \raisebox{0.25 in } { } \right\}~. \nonumber\end{aligned}\ ] ] this leads to a partition function with the action @xmath454 = - \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \frac{|\delta({\bf x},\tau)|^2}{v_{\bf 0 } } \\ & & - \frac{\hbar}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left [ \phi^*_{\downarrow}({\bf x},\tau ) , \phi_{\uparrow}({\bf x},\tau ) \right ] \cdot { \bf g}^{-1 } \cdot \left [ \begin{array}{c } \phi_{\downarrow}({\bf x},\tau ) \\ \phi^*_{\uparrow}({\bf x},\tau ) \end{array } \right]~ , \nonumber\end{aligned}\ ] ] where @xmath455 \delta({\bf x}-{\bf x } ' ) \delta(\tau-\tau')~ , \nonumber\end{aligned}\ ] ] and the noninteracting green s function @xmath392 is defined by @xmath456\ ] ] and @xmath457 we thus see that the fermionic part has exactly the same matrix structure as in the case of a condensed bose gas , only the selfenergy is now @xmath458 = \frac{1}{\hbar } \left [ \begin{array}{cc } 0 & \delta({\bf x},\tau ) \\ \delta^*({\bf x},\tau ) & 0 \end{array } \right ] \delta({\bf x}-{\bf x } ' ) \delta(\tau-\tau')~.~~\ ] ] if we again integrate out the fermion fields , we get the effective action @xmath459 = - \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \frac{|\delta({\bf x},\tau)|^2}{v_{\bf 0 } } - \hbar { \rm tr}[\ln ( -{\bf g}^{-1})]~,\ ] ] which we can expand in powers of @xmath460 by using @xmath461 and therefore @xmath462 = - \hbar { \rm tr}[\ln ( -{\bf g}_0^{-1 } ) ] + \hbar \sum_{m=1}^{\infty } \frac{1}{m } { \rm tr}[({\bf g}_0 \mbox{\boldmath $ \sigma$})^m]~.\ ] ] explicite calculation @xcite shows that for space and time independent @xmath460 we obtain a ` free - energy ' density of the form of the landau - theory of second - order phase transitions , i.e. , @xmath463 with as expected @xmath464 here @xmath465 is the density of states of a single spin state at the fermi energy @xmath466 and the critical temperature is given by @xmath467 if we use the same renormalization procedure as in section [ ap ] to eliminate the interaction parameter @xmath434 in favor of the negative @xmath338-wave scattering length @xmath443 . below @xmath340 we thus have a nonzero average @xmath468 . using this in the green s function for the fermions and neglecting fluctations , we find in momentum space that @xmath469\ ] ] and therefore poles in @xmath470 if @xmath471 this disperion relation has clearly a gap of magnitude @xmath472 at the fermi surface . as a consequence @xmath472 is known as the bcs gap parameter . note that the critical temperature below which the gap becomes nonzero depends exponentally on the parameter @xmath473 . for typical atomic gases this is a quantity which is much larger than one , and the bcs transition is experimentally inaccesible . the only exception at the moment appears to be @xmath474li , with its anomalously large and negative triplet scattering length of @xmath475 @xcite . this explains the present interest in spin - polarized atomic lithium . in view of this exciting possibility , we have recently studied the equilibrium properties of atomic lithium is a harmonic oscillator potential @xmath476 , with a trapping frequency of @xmath477 hz @xcite . to incorporate the effect of the external potential we have , as mentioned above , applied the local - density approximation . the result is shown in figure 15 . note that the use of the local - density approximation implies that we perform the above outlined homogeneous calculation for each point in space with a chemical potential that is equal to @xmath478 the third term in the right - hand side represents the mean - field effect of the hartree contribution to the selfenergy of the fermions . since our hubbard - stratonovich procedure is in principle exact , it is not immediately clear why such a term must be included in the theory . it can , however , be shown that it arises from the fluctuations of the bcs gap parameter . with this remark , we end our development of the equilibrium field theory of trapped atomic gases . we now turn our attention to the nonequilibrium field theory . and energy gap @xmath479 for a @xmath474li atomic gas consisting of @xmath480 atoms in each spin state a ) at @xmath481 nk , b ) at @xmath482 nk , slightly below @xmath340 , and c ) at @xmath483 nk . the left scale of each plot refers to the density and the right scale to the energy gap . the dotted line in c ) shows the density distribution for an ideal fermi gas with the same number of particles and at the same temperature . ] as we have already seen several times , the equlibrium field theory gives us also information on dynamical properties of the gas by means of the substitution @xmath484 or equivalently @xmath485 . however , the popov theory of bose - einstein condensation has shown us that we have to be very careful with this procedure , because we may not always end up with the correct physics needed for a discription of a particular experiment . it is for these cases that a truly nonequilibrium field theory is required and the development of such a theory is the topic of section [ qkt ] . before embarking on that , however , we first consider two important dynamical problems where the equilibrium theory does give us all the answers . in a bose gas with effectively attractive interactions , i.e. , with a negative scattering length @xmath443 , a condensate will always have the tendency to collapse to a high density state due to the gain in energy that can be obtained in this way @xcite . the most important question in this context is , therefore , if a condensate can exist sufficiently long to be experimentally observed . neglecting the variation of the noncondensate density on the size of the condensate @xcite , we know from the popov theory that the dynamics of the collapse is , apart from an unimportant shift in the chemical potential , determined by the gross - pitaevskii equation @xcite @xmath486 more precisely , this determines only the semiclassical dynamics . if we also want to study the quantum fluctuations , which is necessary if we are also interested in how the condensate tunnels through the macroscopic energy barrier , it is most convenient to calculate the grand canonical partition function of the condensate @xcite . quantizing the gross - pitaevskii equation we obtain for this partition function the functional integral @xmath487d[\phi]~ \exp \left\ { - \frac{1}{\hbar } s[\phi^*,\phi ] \right\}~,\ ] ] over the complex field @xmath363 and with the euclidian action @xmath488 = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi^*({\bf x},\tau ) \left ( \hbar \frac{\partial}{\partial \tau } + \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } \right . \\ & & + \left . v^{\rm ex}({\bf x } ) - \mu + \frac{2\pi a\hbar^2}{m } |\phi({\bf x},\tau)|^2 \right ) \phi({\bf x},\tau)~. \nonumber\end{aligned}\ ] ] as always for bose systems , the integration is only over fields that are periodic on the imaginary time axis . although it has recently been shown by freire and arovas that the tunneling process can also be studied in terms of the complex field @xmath363 @xcite , we believe that it leads to somewhat more physical insight if we use instead the fields @xmath489 and @xmath490 that correspond to the density and phase fluctuations of the condensate , respectively . they are introduced by performing the canonical variable transformation @xcite @xmath491 in the functional integral for the partition function . as a result we find @xmath492d[\theta]~ \exp \left\ { - \frac{1}{\hbar } s[\rho,\theta;\mu ] \right\}~,\ ] ] with @xmath493 = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left ( i\hbar \rho({\bf x},\tau ) \frac{\partial \theta({\bf x},\tau)}{\partial \tau } \right . \\ & & + \frac{\hbar^2 \rho({\bf x},\tau)}{2 m } ( \mbox{\boldmath $ \nabla$ } \theta({\bf x},\tau))^2 + \frac{\hbar^2}{8m\rho({\bf x},\tau ) } ( \mbox{\boldmath $ \nabla$ } \rho({\bf x},\tau))^2 \nonumber \\ & & + v^{\rm ex}({\bf x})\rho({\bf x},\tau ) - \mu\rho({\bf x},\tau ) + \left . \frac{2\pi a\hbar^2}{m } \rho^2({\bf x},\tau ) \right)~. \nonumber\end{aligned}\ ] ] next , we notice that this action is only quadratic in the phase fluctuations . the field @xmath490 can therefore be integrated over exactly , because it only involves the evaluation of a gaussian integral . compared to ordinary gaussian integrals there is , however , one slight complication which is associated with the fact that @xmath490 are phase variables . this implies that the periodicity of the original field @xmath363 only constraints the phase field @xmath490 to be periodic up to a multiple of @xmath494 . to evaluate the grand canonical partition function in equation ( [ gr ] ) we must therefore first integrate over all fields @xmath490 that obey the boundary condition @xmath495 and subsequently sum over all possible integers @xmath12 . because these different boundary conditions only affect the zero - momentum part of @xmath490 we first have to evaluate the sum @xmath496~ \exp \left\ { -i \int_0^{\hbar\beta } d\tau~ n_{\bf 0}(\tau ) \frac{\partial \theta_{\bf 0}(\tau)}{\partial \tau } \right\}~ , \nonumber\end{aligned}\ ] ] with @xmath497 the number of condensate particles . after performing a partial integration on the integral in the exponent , we can carry out the path integration over @xmath498 to obtain @xmath499~. \nonumber\end{aligned}\ ] ] as expected , the integration over the global phase of the condensate leads to the constraint of a constant number of condensate particles , i.e. , @xmath500 . moreover , we have @xmath501 , which restricts the number of condensate particles to an integer . putting all these results together , we see that the integration over the zero - momentum part of @xmath489 is only a sum over the number of condensate particles and we have that @xmath502 here we introduced the canonical partition function of the condensate , which is apparently equal to the functional integral @xmath503d[\theta]~ \exp \left\ { - \frac{1}{\hbar } s[\rho,\theta;0 ] \right\}\ ] ] over all the nonzero momentum components of the density and phase fields . the integration over the nonzero momentum components of the phase field @xmath490 is easily performed , because it now involves an ordinary gaussian integral . introducing the green s function for the phase fluctuations @xmath504 by @xmath505 we immediately obtain the desired effective action for the density field @xmath506 = \int_0^{\hbar\beta } d\tau \int d{\bf x } \int d{\bf x}'~ \left(-\frac{\hbar}{2 } \frac{\partial \rho({\bf x},\tau ) } { \partial \tau } g({\bf x},{\bf x}';\rho ) \frac{\partial \rho({\bf x}',\tau ) } { \partial \tau } \right ) \\ & & + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left ( \frac{\hbar^2}{8m\rho({\bf x},\tau ) } ( \mbox{\boldmath $ \nabla$ } \rho({\bf x},\tau))^2 + v^{\rm ex}({\bf x } ) \rho({\bf x},\tau ) \right . \nonumber \\ & & \hspace*{2.4 in } + \left . \frac{2\pi a\hbar^2}{m } \rho^2({\bf x},\tau ) \right)~. \nonumber\end{aligned}\ ] ] being an action for the density fluctuations of the condensate , @xmath507 $ ] also describes all the collisionless modes of the condensate . this is important for our purposes , because the mode which becomes unstable first , determines precisely how the condensate collapses . moreover , it determines the probability with which the collapse is going to take place , both for quantum and thermal fluctuations , since the energy barrier is smallest in that direction of the configuration space . it should be noted that as long as we can neglect the interaction between the condensate and the thermal cloud , the action @xmath507 $ ] describes also the collective modes of a gas with positive scattering length . for various other theoretical approaches that have been applied under these conditions see , for example , refs . the actual measurements of the collective mode frequencies have been performed by jin _ _ @xcite and mewes _ et al . _ @xcite and are at sufficiently low temperatures indeed in good agreement with the theoretical predictions @xcite . we expect the same to be true for a gas with effectively attractive interactions and , therefore , the action @xmath507 $ ] to be a good starting point for the following discussion . to obtain the collisionless modes explicitly we consider first the case of an ideal bose gas by putting @xmath508 . for the ideal bose gas we expect the gaussian profile @xmath509 to describe an exact mode of the condensate . the reason is that in the noninteracting case we can make a density fluctuation by taking one particle from the condensate and putting that in one of the excited states of the external potential . the corresponding density fluctuation obeys @xmath510 for the experimentally relevant case of an isotropic harmonic oscillator @xcite it is more convenient to use instead of the cartesian quantum numbers @xmath164 , the two angular momentum quantum numbers @xmath511 and @xmath4 and the quantum number @xmath512 that counts the number of nodes in the radial wave function @xmath513 . the density fluctuation then becomes @xmath514 with @xmath515 the excitation energy and @xmath516 the size of the condensate wave function . comparing this now with the expansion of the gaussian profile in equation ( [ gprof ] ) around the groundstate density profile , which is obtained by substituting @xmath517 , we find that @xmath518 has precisely the same form as a density fluctuation in which one particle is taken from the condensate and put into the harmonic oscillator state with quantum numbers @xmath519 . the frequency of this so - called ` breathing ' mode described by the gaussian density profile must therefore be equal to @xmath520 . to proof that this is indeed correct , we need to evaluate the effective action @xmath507 $ ] , and hence the green s function @xmath504 , for a gaussian density profile . substituting such a profile in equation ( [ geq ] ) immediately leads to @xmath521 , with @xmath522 the latter equation can be solved , if we can solve the eigenvalue problem @xmath523 this turns out to be an easy task , because substituting @xmath524 gives essentially the radial schrdinger equation for an isotropic harmonic oscillator with frequency @xmath525 , i.e. , @xmath526 the desired eigenfunctions are therefore @xmath527 , with @xmath528 the properly normalized harmonic oscillator states with the energies @xmath529 , and the corresponding eigenvalues are @xmath530 . introducing finally the ` dual ' eigenfunctions @xmath531 , the green s function @xmath532 is given by @xmath533 note that prime on the summation sign indicates that the sum is over all quantum numbers except @xmath534 . the latter is excluded because the associated eigenfunction @xmath535 is just a constant and thus does not contribute to @xmath504 , which is defined as the green s function for all phase fluctuations with nonvanishing momenta . putting all these results together , we see that the dynamics of the collective variable @xmath536 is determined by the action @xmath537 = \int_0^{\hbar\beta } d\tau~ \left\ { \frac{3mn_{\bf 0}}{4 } \left ( \frac{dq}{d\tau } \right)^2 + n_{\bf 0 } \left ( \frac{3\hbar^2}{4mq^2 } + \frac{3}{4}m\omega^2q^2 \right ) \right\ } \\ & & \equiv \int_0^{\hbar\beta } d\tau~ \left\ { \frac{1}{2 } m^ * \left ( \frac{dq}{d\tau } \right)^2 + v(q ) \right\}~ , \nonumber\end{aligned}\ ] ] that is equivalent to the action of a particle with effective mass @xmath538 in a potential @xmath539 . as expected from our previous remarks , this potential has a minimum for @xmath540 and can be expanded near its minimum as @xmath541 it thus comfirms that the gaussian profile describes a breathing mode with frequency @xmath520 around an equilibrium density profile that is given by @xmath542 . our next task is to investigate how interactions affect this result . considering again only gaussian density profiles , the action @xmath543 $ ] is again that of a particle with effective mass @xmath538 but now in the potential @xcite @xmath544 the physically most important feature of this potential is that it is unbounded from below , since @xmath545 if @xmath546 . hence , the condensate indeed always has the tendency to collapse to the high - density state @xmath547 . however , if the number of condensate particles is sufficiently small , or more precisely if @xcite @xmath548 the condensate has to overcome a macroscopic energy barrier before it can collapse . under these conditions the condensate is therefore really metastable and can in principle be observed experimentally . the most important question in this respect is of course : how metastable is the condensate ? within the gaussian approximation this question is easily answered , because then the dynamics of the condensate is equivalent to the dynamics of a particle in an unstable potential , as we have just seen . we therefore only need to evaluate the wkb - expression for the tunnneling rate @xcite and compare this to the rate of decay due to thermal fluctuations by calculating also the height of the energy barrier . the outcome of this comparison for the conditions of the experiment with atomic @xmath549li is presented in reference @xcite and shows that , for the relatively high temperatures @xmath550 that have been obtained thusfar @xcite , the decay by means of thermal fluctuations over the energy barrier is the dominant decay mechanism of the condensate . more important for our purposes , however , is that sufficiently close to the maximum number of condensate particles @xmath551 the collective decay of the condensate discussed above is always much more probable than the decay due to two and three - body collisions that lead to a spin - flip or the formation of @xmath549li molecules , respectively . as a result the collapse of the condensate should be observable within the finite lifetime of the gas . in fact , on the basis of this separation of time scales we expect the condensate to go through a number of growth and collapse cycles @xcite . physically this picture arises as follows . starting from a gas with a number of atoms @xmath552 , the condensate will initially grow as a response to evaporative cooling . however , if the number of condensate atoms starts to come close to @xmath551 , the condensate fluctuates over the energy barrier and collapses in a very short time of @xmath553 @xcite . during the collapse the condensate density increases rapidly and two and three - body inelastic processes quickly remove almost all the atoms from the condensate . after this has occurred the condensate grows again from the noncondensed part of the gas and a new growth and collapse cycle begins . it is only after many of these cycles that enough atoms are removed for the gas to relax to an equilibrium with a number of condensate particles that is less than @xmath554 . this is shows quantitatively in figure 16 for the experimental conditions of interest . in response to evaporative cooling . the long - time decay of the condensate is due to two and three - body inelastic collisions . ] a final issue which needs to be addressed at this point is the actual dynamics of the collapse and , in particular , how we must include the effect of the inelastic growth and decay processes on this dynamics . unfortunately , the inclusion of these effects is rather complicated and we need the results of section [ cf ] to be able to correctly address this problem . it is , however , of some interest because sackett _ et al . _ have recently observed that after a single collapse there remains a remnant of the condensate with about 10% of the initial number of atoms @xcite . at present , it is an important open problem to theoretically understand the magnitude of this remnant . as we have just seen explicitly , a particularly interesting consequence of the finite size of the gas is that quantum fluctuations play a much more important role . although this is especially true for the case of attractive interactions that we considered in section [ nsl ] , it is also true for a bose gas with repulsive interactions . a striking example in this respect is the phenomenon of phase ` diffusion ' , which was recently discussed by lewenstein and you @xcite . we rederive their results for a trapped bose gas in a moment , but first consider also the same phenomenon for a neutral and homogeneous superconductor . in this manner it is possible to bring out the physics involved more clearly . using the approach of section [ sc ] , it can be shown that at zero temperature the dynamics of the superconducting order parameter , i.e. , the bcs gap parameter @xmath555 that is proportional to the wave function of the condensate of cooper pairs , is in a good approximation determined by a time - dependent ginzburg - landau theory @xcite with the action @xmath556 = \frac{n(0)}{4 } \int dt \int d{\bf x}~ \left\ { \frac{\hbar^2}{|\delta_0|^2 } \left| \frac{\partial \delta}{\partial t } \right|^2 \right . \\ & & - \left . \frac{\hbar^2 v_f^2}{3|\delta_0|^2 } \left| \mbox{\boldmath $ \nabla$ } \delta \right|^2 + 2 |\delta|^2 \left ( 1 - \frac{|\delta|^2}{2|\delta_0|^2 } \right ) \right\}~ , \nonumber\end{aligned}\ ] ] where @xmath465 is the density of states for one spin projection at the fermi energy @xmath557 and @xmath558 is the equilibrium value of the order parameter @xcite . writing the complex order parameter in terms of an amplitude and a phase , we immediately observe that the amplitude fluctuations are gapped @xcite and can , therefore , be safely neglected at large length scales . the long - wavelength dynamics of the superconductor is thus dominated by the phase fluctuations , according to the action @xmath559 = \frac{n(0)\hbar^2}{4 } \int dt \int d{\bf x}~ \left\ { \left ( \frac{\partial \theta}{\partial t } \right)^2 - \frac{v_f^2}{3 } \left ( \mbox{\boldmath $ \nabla$ } \theta \right)^2 \right\}~.\ ] ] this also implies that the global phase @xmath560 of the superconductor has a dynamics that is governed by @xmath561 = \frac{n(0)n\hbar^2}{4n } \int dt~ \left ( \frac{d \theta_{\bf 0}}{dt } \right)^2~,\ ] ] using the fact that the total volume @xmath337 of the system is given by @xmath562 . up to now our discussion has again been semiclassical . to consider also the quantum fluctuations , we have to quantize this theory by applying the usual rules of quantum mechanics . doing so , we find that the wave function of the overall phase obeys a schrdinger equation @xmath563 with a ` diffusion ' constant that can easily be shown to be equal to the quantity @xmath564 @xcite and is , most importantly for our purposes , proportional to @xmath565 . in the thermodynamic limit @xmath566 a state with a well defined stationary phase is clearly a solution and we are then dealing with a system having a spontaneously broken @xmath424 symmetry . however , for a finite ( and fixed ) number of particles the global phase can not be well defined at all times and always has to ` diffuse ' in accordance with the above schrdinger equation . note also that in the groundstate the phase is fully undetermined and @xmath567 . maybe surprisingly , the same calculation is somewhat more complicated for a bose gas because the amplitude fluctuations of the order parameter can not be neglected even at the largest length scales . however , taking these amplitude fluctuations into account properly , we nevertheless arrive at an action that is equivalent to equation ( [ phase ] ) and hence again leads to the phenomenon of phase ` diffusion ' . we start again from the action @xmath568 $ ] for the condensate . the difference with the previous subsection is , however , that now we are not so much interested in the dynamics of the density but in the phase dynamics instead . therefore we now want to integrate over the density field @xmath489 . this can not be done exactly and we therefore here consider only the strong - coupling limit , which was also treated by lewenstein and you @xcite . in that limit we are allowed to neglect the gradient of the average density profile @xcite and the action @xmath568 $ ] is for the longest wavelengths well approximated by @xmath569 = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left ( i\hbar \rho({\bf x},\tau ) \frac{\partial \theta({\bf x},\tau)}{\partial \tau } \right . \\ & & + \left . v^{\rm ex}({\bf x})\rho({\bf x},\tau ) - \mu\rho({\bf x},\tau ) + \frac{2\pi a\hbar^2}{m } \rho^2({\bf x},\tau ) \right)~. \nonumber\end{aligned}\ ] ] in equilibrium the average density profile of the condensate thus obeys @xmath570 performing now the shift @xmath571 , we find for the zero - momentum part of the action @xcite @xmath572 = \hbar\beta e_{\bf 0}(\mu ) + \int_0^{\hbar\beta } d\tau \left ( i\hbar \delta n_{\bf 0 } \frac{d \theta_{\bf 0}}{d\tau } + \frac{2\pi a\hbar^2}{mv_{\bf 0}(\mu ) } ( \delta n_{\bf 0})^2 \right)~,\ ] ] where @xmath573 and @xmath574 correspond , respectively , to the energy and the volume of the condensate in the so - called thomas - fermi approximation @xcite . moreover , @xmath575 represents the fluctuations in the total number of condensate particles in that same approximation , implying that the density fluctuations @xmath576 are only nonzero in that region of space where the condensate density does not vanish . performing now the integration over the number fluctuations @xmath577 and the usual wick rotation to real times @xmath485 , we immediately see that the effective action for the global phase of the condensate has precisely the same form as in equation ( [ phase ] ) , i.e. , @xmath578 = \frac{mv_{\bf 0}(\mu)}{8\pi a } \int dt~ \left ( \frac{d\theta_{\bf 0}}{dt } \right)^2~.\ ] ] the appropriate ` diffusion ' constant is therefore equal to @xmath579 , which can easily be shown to be equal to @xmath580 if we make use of the fact that in the thomas - fermi approximation the chemical potential obeys @xmath581 and the radius of the condensate is given by @xmath582 @xcite . hence , the ` diffusion ' constant is proportional to @xmath583 . note that if the condensate where contained in a box the ` diffusion ' constant would be proportional to @xmath584 instead . it is important to note also that , in contrast to the case of a fermionic superfluid , we have to integrate over the amplitude fluctuations of the order parameter to arrive at a quadratic action for the phase fluctuations . this leads to the important conclusion that for a bosonic superfluid it is impossible to be in a state with only phase fluctuations and no density fluctuations , even at the largest length scales . for simplicity we restrict ourselves for the rest of this course to the case of a doubly - polarized atomic bose gas , which for most practical purposes can be seen as a gas of spin - less bosons @xcite . once we have understood the quantum kinetic theory for this particular case , however , we can easily generalize to multi - component bose gases that are receiving considerable attention at present @xcite . furthermore , the same methods can also be used for fermionic gases and have recently already been applied to the problem of cooper - pair formation in a two - component fermion gas @xcite . in textbooks an ideal bose gas is generally discussed in terms of the average occupation numbers of the one - particle states @xmath53 @xcite . given the density matrix @xmath585 of the gas at an initial time @xmath586 , these occupation numbers obey @xmath587~,\ ] ] with @xmath588 and @xmath589 the usual ( heisenberg picture ) creation and annihilation operators of second quantization , respectively . because the hamiltonian of the gas @xmath590 commutes with the number operators @xmath591 , the nonequilibrium dynamics of the system is trivial and the average occupation numbers are at all times equal to their value at the initial time @xmath586 . if we are also interested in fluctuations , it is convenient to introduce the eigenstates of the number operators , i.e. , @xmath592 and to consider the full probability distribution @xmath593~,\ ] ] which is again independent of time for an ideal bose gas . the average occupation numbers are then determined by @xmath594 and the fluctuations can be obtained from similar expressions . as indicated in section [ pd ] , we are also interested in the phase dynamics of the gas . in analogy with the occupation number representation in equation ( [ pn ] ) , we can now obtain a completely different description of the bose gas by making use of coherent states and considering the probability distribution @xmath595 $ ] . although we expect that this probability distribution is again independent of time , let us nevertheless proceed to derive its equation of motion in a way that can be generalized lateron when we consider an interacting bose gas . first , we use that @xmath596 = \int d[\phi^*_0]d[\phi_0]~ p[\phi_0^*,\phi_0;t_0 ] \frac{|\langle\phi;t|\phi_0;t_0\rangle|^2 } { \langle\phi;t|\phi;t\rangle \langle\phi_0;t_0|\phi_0;t_0\rangle}~.\ ] ] this is a particularly useful result , because the time dependence is now completely determined by the matrix element @xmath597 for which the functional integral representation is well known from our discussion in section [ fi ] . it is given by a ` path ' integral over all complex field evolutions @xmath598 from @xmath586 to @xmath30 . more precisely we have @xmath599d[\psi]~ \exp \left\ { \frac{i}{\hbar } s_+[\psi^*,\psi ] \right\}~,\ ] ] with the forward action @xmath600 $ ] given by @xmath601 = \sum_{\bf n } \left\ { -i\hbar \psi^*_{\bf n}(t ) \psi_{\bf n}(t ) \raisebox{0.2 in } { } \right . \\ & & \hspace*{0.6 in } + \left . \int_{t_0}^{t } dt_+~ \psi^*_{\bf n}(t_+ ) \left ( i\hbar \frac{\partial}{\partial t_+ } - \epsilon_{\bf n } \right ) \psi_{\bf n}(t_+ ) \right\}~. \nonumber\end{aligned}\ ] ] in the same manner the matrix element @xmath602 can be written as a ` path ' integral over all possible field configurations @xmath603 evolving backward in time from @xmath30 to @xmath586 , i.e. , @xmath604d[\psi]~ \exp \left\ { \frac{i}{\hbar } s_-[\psi^*,\psi ] \right\}~,\ ] ] with a backward action @xmath605 = \sum_{\bf n } \left\ { -i\hbar \psi^*_{\bf n}(t_0 ) \psi_{\bf n}(t_0 ) \raisebox{0.2 in } { } \right . \\ & & \hspace*{0.6 in } + \left . \int_{t}^{t_0 } dt_-~ \psi^*_{\bf n}(t_- ) \left ( i\hbar \frac{\partial}{\partial t_- } - \epsilon_{\bf n } \right ) \psi_{\bf n}(t_- ) \right\ } \nonumber \\ & & \hspace*{0.1 in } = \sum_{\bf n } \left\ { -i\hbar \psi^*_{\bf n}(t ) \psi_{\bf n}(t ) \raisebox{0.2 in } { } \right . \nonumber \\ & & \hspace*{0.6 in } + \left . \int_{t}^{t_0 } dt_-~ \psi_{\bf n}(t_- ) \left ( -i\hbar \frac{\partial}{\partial t_- } - \epsilon_{\bf n } \right ) \psi^*_{\bf n}(t_- ) \right\}~. \nonumber\end{aligned}\ ] ] putting all these results together , we see that the probability distribution @xmath595 $ ] can in fact be represented by a functional integral over all fields @xmath606 that evolve backwards from @xmath30 to @xmath586 and then forward in time from @xmath586 to @xmath30 . absorbing the factor @xmath607 $ ] into the measure of the functional integral , we thus arrive at the desired result that @xmath608 = \int_{\psi({\bf x},t)=\phi({\bf x})}^{\psi^*({\bf x},t)=\phi^*({\bf x } ) } d[\psi^*]d[\psi]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi ] \right\}~,\ ] ] where the total ( backward - forward ) action in first instance obeys @xmath609 = s_-[\psi^*,\psi ] + s_+[\psi^*,\psi ] = -i\hbar \sum_{\bf n } \left ( \psi^*_{\bf n}(t ) \psi_{\bf n}(t ) - |\phi_{\bf n}|^2 \right ) \hspace{0.4 in } \\ & & + \sum_{\bf n } \int_{{\cal c}^t } dt'~ \left\ { \frac{1}{2 } \left ( \psi^*_{\bf n}(t ' ) i\hbar \frac{\partial}{\partial t ' } \psi_{\bf n}(t ' ) - \psi_{\bf n}(t ' ) i\hbar \frac{\partial}{\partial t ' } \psi^*_{\bf n}(t ' ) \right ) \right . \nonumber \\ & & \hspace*{2.55 in } - \left . \raisebox{0.2 in } { } \epsilon_{\bf n } \psi^*_{\bf n}(t ' ) \psi_{\bf n}(t ' ) \right\}~ , \nonumber\end{aligned}\ ] ] and the integration along the schwinger - keldysh contour @xmath610 is defined by @xmath611 @xcite . note also that in equation ( [ prob2 ] ) we have explicitly specified the boundary conditions on the functional integral . it is interesting to mention that these boundary conditions are essentially dictated by the topological terms in the action @xmath612 $ ] , which is a general feature of the path - integral formulation of quantum mechanics due to the fact that the quantum theory should have the correct ( semi)classical limit @xcite . making use of the periodicity of the field @xmath606 on the schwinger - keldysh contour , the variational principle @xmath613/\delta \psi^*_{\bf n}(t_{\pm } ) = 0 $ ] indeed leads not only to the euler - lagrange equation @xmath614 which agrees with the heisenberg equation of motion @xmath615 $ ] for the annihilation operators , but also with the appropriate boundary condition @xmath616 . in the same way we find the complex conjugate results if we require that @xmath613/\delta \psi_{\bf n}(t_{\pm } ) = 0 $ ] . substituting the boundary conditions in equation ( [ action ] ) and performing a partial integration , we then finally obtain for the action @xmath617 = \sum_{\bf n } \int_{{\cal c}^t } dt'~ \psi^*_{\bf n}(t ' ) \left ( i\hbar \frac{\partial}{\partial t ' } - \epsilon_{\bf n } \right ) \psi_{\bf n}(t')~,\ ] ] which we for completeness sake also rewrite as @xmath618 \\ & & = \int_{{\cal c}^t } dt ' \int d{\bf x}~ \psi^*({\bf x},t ' ) \left ( i\hbar \frac{\partial}{\partial t ' } + \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } - v^{\rm ex}({\bf x } ) \right ) \psi({\bf x},t')~. \nonumber\end{aligned}\ ] ] we are now in a position to derive the equation of motion , i.e. , the fokker - planck equation , for the probability distibution @xmath595 $ ] . this is most easily achieved by performing the variable transformation @xmath619 in equation ( [ prob2 ] ) . in this manner the fields @xmath620 and @xmath621 that live on the backward and forward branch of the schwinger - keldysh contour , respectively , are ` projected ' onto the real time axis . moreover , at the same time we effect a separation between the ( semi)classical dynamics described by @xmath622 and the quantum fluctuations determined by @xmath623 . after the transformation we have @xmath624 \\ & & = \int_{\phi({\bf x},t)=\phi({\bf x})}^{\phi^*({\bf x},t)=\phi^*({\bf x } ) } d[\phi^*]d[\phi ] \int d[\xi^*]d[\xi]~ \exp \left\ { \frac{i}{\hbar } s[\phi^*,\phi;\xi^*,\xi ] \right\}~ , \nonumber\end{aligned}\ ] ] with @xmath625 = \sum_{\bf n } \int_{t_0}^t dt'~ \left\ { \phi^*_{\bf n}(t ' ) \left ( i\hbar \frac{\partial}{\partial t ' } - \epsilon_{\bf n } \right ) \xi_{\bf n}(t ' ) \right . \\ & & \hspace*{1.2 in } + \left . \xi^*_{\bf n}(t ' ) \left ( i\hbar \frac{\partial}{\partial t ' } - \epsilon_{\bf n } \right ) \phi_{\bf n}(t ' ) \right\}~. \nonumber\end{aligned}\ ] ] because this action is linear in @xmath626 and @xmath627 , the integration over the quantum fluctuations leads just to a constraint and we find that @xmath628 = \int_{\phi({\bf x},t)=\phi({\bf x})}^{\phi^*({\bf x},t)=\phi^*({\bf x } ) } d[\phi^*]d[\phi]~ \\ & & \times \prod_{\bf n } \delta \left [ \left ( -i \frac{\partial}{\partial t ' } - \frac{\epsilon_{\bf n}}{\hbar } \right ) \phi^*_{\bf n}(t ' ) \cdot \left ( i \frac{\partial}{\partial t ' } - \frac{\epsilon_{\bf n}}{\hbar } \right ) \phi_{\bf n}(t ' ) \right ] \nonumber\end{aligned}\ ] ] or equivalently that @xmath629 = \int d[\phi^*_0]d[\phi_0]~ p[\phi^*_0,\phi_0;t_0 ] \prod_{\bf n } \delta \left ( |\phi_{\bf n } - \phi^{\rm cl}_{\bf n}(t)|^2 \right)~,\ ] ] where we introduced the quantity @xmath630 obeying the semiclassical equation of motion @xmath631 and the initial condition @xmath632 . the latter equation is thus solved by @xmath633 and we conclude from a simple change of variables in equation ( [ prob3 ] ) that for an ideal bose gas @xmath595 = p[\phi^*,\phi;t_0]$ ] , as expected . we also see from equation ( [ prob3 ] ) that the desired equation of motion for @xmath595 $ ] reads @xmath634 = - \left ( \sum_{\bf n } \frac{\partial}{\partial \phi_{\bf n } } \epsilon_{\bf n } \phi_{\bf n } \right ) p[\phi^*,\phi;t ] \\ & & \hspace*{0.8 in } + \left ( \sum_{\bf n } \frac{\partial}{\partial \phi^*_{\bf n } } \epsilon_{\bf n } \phi^*_{\bf n } \right ) p[\phi^*,\phi;t]~. \nonumber\end{aligned}\ ] ] this is indeed the correct fokker - planck equation for an ideal bose gas @xcite . to see that we first consider @xmath635d[\phi]~ \phi_{\bf n } p[\phi^*,\phi;t]$ ] . multiplying equation ( [ fp1 ] ) with @xmath636 and integrating over @xmath637 , we easily find after a partial integration that @xmath638 which precisely corresponds to the equation of motion of the expectation value @xmath639 $ ] in the operator formalism . similarly , we find that @xmath640 in agreement with the result for @xmath641 $ ] . next we consider the average of @xmath642 , for which we immediately obtain @xmath643 we expect this result to be related to the fact that in the operator formalism the occupation numbers @xmath644 are independent of time . although this turns out to be true , to give the precise relation between @xmath645 and @xmath644 is complicated by the fact that at equal times the operators @xmath589 and @xmath588 do not commute . however , the path - integral formulation of quantum mechanics is only capable of calculating time - ordered operator products @xcite . in our case this implies that @xmath645 is the value at @xmath646 of @xmath647 \\ & & = \theta(t , t ' ) { \rm tr } \left [ \hat{\rho}(t_0 ) \hat{\psi}_{\bf n}(t ) \hat{\psi}^{\dagger}_{\bf n}(t ' ) \right ] + \theta(t',t ) { \rm tr } \left [ \hat{\rho}(t_0 ) \hat{\psi}^{\dagger}_{\bf n}(t ' ) \hat{\psi}_{\bf n}(t ) \right]~ , \nonumber\end{aligned}\ ] ] with @xmath648 the time - ordening operator on the schwinger - keldysh contour and @xmath649 the corresponding heaviside function . since the heaviside function is equal to @xmath449 at equal times @xcite , we conclude that @xmath650 and that equation ( [ phisquare ] ) is thus fully consistent with the operator formalism . an intuitive understanding of the relation between @xmath645 and the occupation numbers can be obtained by noting that , since all our manipulations up to now have been exact , we expect @xmath645 to contain both classical and quantum fluctuations . these correspond precisely to the contributions @xmath644 and @xmath449 , respectively . finally we need to discuss the stationary solutions of the fokker - planck equation . it is not difficult to show that any functional that only depends on the amplitudes @xmath642 is a solution . as it stands the fokker - planck equation , therefore , does not lead to a unique equilibrium distribution . this is not surprising , because for an isolated , ideal bose gas there is no mechanism for redistributing the particles over the various energy levels and thus for relaxation towards equilibrium . however , the situation changes when we allow the bosons in the trap to tunnel back and forth to a reservoir at a temperature @xmath117 . the corrections to the fokker - planck equation that are required to describe the physics in this case are considered next . however , to determine these corrections in the most convenient way , we have to slightly generalize the above theory because with the probability distibution @xmath595 $ ] we are only able to study spatial , but not temporal correlations in the bose gas . to study also those we follow the by now well - known procedure in quantum field theory and construct a generating functional @xmath651 $ ] for all ( time - ordered ) correlation functions . it is obtained by performing two steps . first , we introduce the probability distribution @xmath652 $ ] for a bose gas in the presence of the external currents @xmath653 and @xmath654 by adding to the hamiltonian the terms @xmath655 as a result we have @xmath656 = \int_{\psi({\bf x},t)=\phi({\bf x})}^{\psi^*({\bf x},t)=\phi^*({\bf x } ) } d[\psi^*]d[\psi]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi ] \right\ } \\ & & \hspace*{0.38 in } \times \exp \left\ { i \int_{{\cal c}^t } dt ' \int d{\bf x}~ \left ( \psi({\bf x},t ' ) j^*({\bf x},t ' ) + j({\bf x},t ' ) \psi^*({\bf x},t ' ) \right ) \right\}~. \nonumber\end{aligned}\ ] ] second , we integrate this expression over @xmath637 to obtain the desired generating functional . hence @xmath657 = \int d[\phi^*]d[\phi]~ p_j[\phi^*,\phi;t ] = \int d[\psi^*]d[\psi]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi ] \right\ } \hspace{0.4 in } \\ & & \hspace*{0.32 in } \times \exp \left\ { i \int_{{\cal c}^{\infty } } dt \int d{\bf x}~ \left ( \psi({\bf x},t ) j^*({\bf x},t ) + j({\bf x},t ) \psi^*({\bf x},t ) \right ) \right\}~. \nonumber\end{aligned}\ ] ] it is important to realize that @xmath651 $ ] is indeed independent of the time @xmath30 because of the fact that @xmath652 $ ] is a probability distribution and thus properly normalized . we are therefore allowed to deform the contour @xmath658 to any closed contour that runs through @xmath586 . since we are in principle interested in all times @xmath659 , the most convenient choice is the countour that runs backward from infinity to @xmath586 and then forwards from @xmath586 to infinity . this contour is denoted by @xmath660 and also called the schwinger - keldysh contour in the following because there is in practice never confusion with the more restricted contour @xmath658 that is required when we consider a probability distribution . with this choice it is also clear that @xmath651 $ ] is a generating functional . indeed , equation ( [ z ] ) shows explicitly that all time - ordered correlation functions can be obtained by functional differentiation with respect to the currents @xmath653 and @xmath654 . we have , for instance , that @xmath661 = \left . \frac{1}{i } \frac{\delta}{\delta j^*({\bf x},t ) } z[j , j^ * ] \right|_{j , j^*=0}\ ] ] and similarly that @xmath662 \\ & & = \left . \frac{1}{i^2 } \frac{\delta^2}{\delta j^*({\bf x},t ) \delta j({\bf x}',t ' ) } z[j , j^ * ] \right|_{j , j^*=0}~. \nonumber\end{aligned}\ ] ] note that the times @xmath30 and @xmath663 always have to be larger or equal to @xmath586 for these identities to be valid . inspired by the caldeira - leggett model for a particle experiencing friction @xcite , we take for the reservoir an ideal gas of @xmath3 bosons in a box with volume @xmath337 . the states in this box are labeled by the momentum @xmath664 and equal to @xmath665 . they have an energy @xmath666 , where @xmath667 accounts for a possible bias between the potential energies of a particle in the center of the trap and a particle in the reservoir . the reservoir is also taken to be sufficiently large that it can be treated in the thermodynamic limit and is in an equilibrium with temperature @xmath117 and chemical potential @xmath60 for times @xmath668 . at @xmath586 it is brought into contact with the trap by means of a tunnel hamiltonian @xmath669 with complex tunneling matrix elements @xmath670 that for simplicity are assumed to be almost constant for momenta @xmath671 smaller that a cutoff @xmath672 but to vanish rapidly for momenta larger than this ultraviolet cutoff . moreover , we consider here only the low - temperature regime in which the thermal de broglie wavelength @xmath673 of the particles obeys @xmath674 , since this is the most appropriate limit for realistic atomic gases . to study the evolution of the combined system for times @xmath659 we thus have to deal with the action @xmath675 = \sum_{\bf n } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf n}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \psi_{\bf n}(t ) \\ & & + \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf k}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon({\bf k } ) + \mu \right ) \psi_{\bf k}(t)~ , \nonumber \\ & & - \frac{1}{\sqrt{v } } \sum_{\bf n } \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \left ( t_{\bf n}({\bf k } ) \psi_{\bf n}(t ) \psi^*_{\bf k}(t ) + t^*_{\bf n}({\bf k } ) \psi_{\bf k}(t ) \psi^*_{\bf n}(t ) \right ) \nonumber \end{aligned}\ ] ] if we measure all energies relative to the chemical potential and also introduce the complex field @xmath676 for the degrees of freedom of the reservoir . however , we are only interested in the evolution of the bose gas in the trap and therefore only in the time - ordered correlation functions of this part of the system . the corresponding generating functional @xmath677 = \int d[\psi^*]d[\psi ] \int d[\psi_r^*]d[\psi_r]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi;\psi_r^*,\psi_r ] \right\ } \hspace{0.4 in } \\ & & \times \exp \left\ { i \int_{{\cal c}^{\infty } } dt \int d{\bf x}~ \left ( \psi({\bf x},t ) j^*({\bf x},t ) + j({\bf x},t ) \psi^*({\bf x},t ) \right ) \right\}~. \nonumber\end{aligned}\ ] ] is of the same form as the functional integral in equation ( [ z ] ) , but now with an effective action that is defined by @xmath678 \right\ } \equiv \int d[\psi_r^*]d[\psi_r]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi;\psi_r^*,\psi_r ] \right\}~.\ ] ] hence , our next task is to integrate out the field @xmath679 , which can be done exactly because it only requires the integration of a gaussian . to familiarize ourselves with the schwinger - keldysh formalism , we perform the gaussian integration here in some detail . in principle , this can of course be done explicitly by making use of the fact that the initial density matrix @xmath680 of the reservoir can be expanded in the projection operators on the coherent states by using the coefficients @xmath681 = \prod_{\bf k } \frac{1}{n({\bf k } ) } e^ { - |\phi_{\bf k}|^2/n({\bf k } ) } ~,\ ] ] with @xmath682 the appropriate bose distribution function and @xmath119 . however , in practice it is much more easy to use a different procedure . it is based on the observation that if we introduce the @xmath440 function on the schwinger - keldysh contour defined by@xmath683 and the green s function @xmath684 obeying @xmath685 the action @xmath686 $ ] can be written as a complete square , or more precisely as the sum of two squares @xmath687 $ ] and @xmath688 $ ] that are given by @xmath689 = \sum_{\bf n } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf n}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \psi_{\bf n}(t ) \\ & & \hspace*{0.1 in } - \frac{1}{\hbar v } \sum_{\bf n,\bf n ' } \sum_{\bf k } \int_{{\cal c}^{\infty } } dt \int_{{\cal c}^{\infty } } dt'~ \psi^*_{\bf n}(t ) t^*_{\bf n}({\bf k } ) g({\bf k};t , t ' ) t_{\bf n'}({\bf k } ) \psi_{\bf n'}(t ' ) \nonumber \end{aligned}\ ] ] and @xmath690 \\ & & = \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \left ( \psi_{\bf k}^*(t ) - \frac{1}{\hbar\sqrt{v } } \sum_{\bf n } \int_{{\cal c}^{\infty } } dt'~ t^*_{\bf n}({\bf k } ) \psi^*_{\bf n}(t ' ) g({\bf k};t',t ) \right ) \nonumber \\ & & \hspace*{0.7 in } \times \left ( i\hbar \frac{\partial}{\partial t } - \epsilon({\bf k } ) + \mu \right ) \nonumber \\ & & \hspace*{0.7 in } \times \left ( \psi_{\bf k}(t ) - \frac{1}{\hbar\sqrt{v } } \sum_{\bf n } \int_{{\cal c}^{\infty } } dt'~ g({\bf k};t , t ' ) \psi_{\bf n}(t ' ) t_{\bf n}({\bf k } ) \right)~ , \nonumber \end{aligned}\ ] ] respectively . since the first term is independent of the field @xmath679 , we only need to evaluate @xmath691d[\psi_r]~ \exp \left ( i s_2[\psi^*,\psi;\psi_r^*,\psi_r]/\hbar \right)$ ] . performing a shift in the integration variables , we however see that this functional integral is equal to @xmath692d[\psi_r]~ \exp \left\ { \frac{i}{\hbar } \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf k}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon({\bf k } ) + \mu \right ) \psi_{\bf k}(t ) \right\ } \\ = { \rm tr}[\rho_r(t_0 ) ] = 1~. \nonumber\end{aligned}\ ] ] as a result the effective action @xmath693 $ ] is just equal to @xmath687 $ ] , which can be slightly rewritten to read @xmath694 = \sum_{\bf n,\bf n ' } \int_{{\cal c}^{\infty } } dt \int_{{\cal c}^{\infty } } dt'~ \\ & & \hspace*{0.37 in } \times~ \psi^*_{\bf n}(t ) \left\ { \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \delta_{\bf n,\bf n ' } \delta(t , t ' ) - \hbar \sigma_{{\bf n},{\bf n}'}(t , t ' ) \right\ } \psi_{\bf n'}(t')~ , \nonumber\end{aligned}\ ] ] with a selfenergy @xmath695 obeying @xmath696 this is our first example of an effective action describing the nonequilibrium dynamics of a bose gas . before we can study its consequences we clearly first need to determine the green s function @xmath684 in terms of which the selfenergy is expressed . although we know that this green s function fulfills equation ( [ greena ] ) , we can not directly solve this equation because we do not know the appropriate boundary condition at @xmath697 . to calculate @xmath684 we therefore have to proceed differently . it is however clear from equation ( [ greena ] ) that @xmath684 is a property of the reservoir and we thus expect that it can somehow be related to a time - ordered correlation function of this reservior . to see that explicitly we consider again the generating functional of these correlation functions , i.e. , @xmath698 \\ & & = \int d[\psi_r^*]d[\psi_r]~ \exp \left\ { \frac{i}{\hbar } \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf k}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon({\bf k } ) + \mu \right ) \psi_{\bf k}(t ) \right\ } \nonumber \\ & & \hspace*{0.9 in } \times \exp \left\ { i \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \left ( \psi_{\bf k}(t ) j^*_{\bf k}(t ) + j_{\bf k}(t ) \psi^*_{\bf k}(t ) \right ) \right\}~. \nonumber\end{aligned}\ ] ] it is again a gaussian integral and can thus be evaluated in the same manner as before . the result is now @xmath699 = \exp \left\ { -i \sum_{\bf k } \int_{{\cal c}^{\infty } } dt \int_{{\cal c}^{\infty } } dt'~ j^*_{\bf k}(t ) g({\bf k};t , t ' ) j_{\bf k}(t ' ) \right\}~,\ ] ] which shows by means of equation ( [ corr ] ) that @xmath700~.\ ] ] note first of all that this is indeed consistent with equation ( [ greena ] ) , because the right - hand side obeys @xmath701 \\ & & = i\hbar \delta(t , t ' ) { \rm tr } \left [ \hat{\rho}_r(t_0 ) [ \hat{\psi}_{\bf k}(t),\hat{\psi}^{\dagger}_{\bf k}(t ) ] \right ] \nonumber \\ & & \hspace*{0.97 in } + { \rm tr } \left [ \hat{\rho}_r(t_0 ) t_{{\cal c}^{\infty } } \left ( i\hbar \frac{\partial}{\partial t } \hat{\psi}_{\bf k}(t ) \hat{\psi}^{\dagger}_{\bf k}(t ' ) \right ) \right ] \nonumber \\ & & = i\hbar \delta(t , t ' ) + ( \epsilon({\bf k } ) - \mu ) { \rm tr } \left [ \hat{\rho}_r(t_0 ) t_{{\cal c}^{\infty } } \left ( \hat{\psi}_{\bf k}(t ) \hat{\psi}^{\dagger}_{\bf k}(t ' ) \right ) \right ] \nonumber\end{aligned}\ ] ] in the operator formalism . moreover , from this identification we see that the desired solution fulfilling the appropriate boundary conditions is apparently @xmath702 the specific dependence on the backward and forward branches of the schwinger - keldysh contour is thus solely determined by the heaviside function @xmath649 . as a result it is convenient to decompose the green s function into its analytic pieces @xmath703 and @xmath704 @xcite by means of @xmath705 due to the fact that we are always dealing with time - ordered correlation functions , such a decomposition turns out to be a generic feature of all the functions on the schwinger - keldysh contour that we will encounter in the following @xcite . for a general function @xmath706 , it is however also possible to have @xmath440-function singularities . if that happens the correct decomposition is @xmath707 this more general decomposition is not required here , but will be needed in section [ cf ] when we determine the effective interaction between two atoms in a gas . having obtained the green s function of the reservoir , we can now return to our discussion of the effective action @xmath693 $ ] for the bose gas in the trap . although we have chosen to derive in equation ( [ seff ] ) the effective action for the generating functional @xmath651 $ ] , it is straightforward to show that the effective action for the probability distribution @xmath595 $ ] is obtained by only replacing the contour @xmath708 by @xmath709 . in the following we therefore no longer always specify all the boundary conditions on the time integration , if the precise details of this integration are not important and the discussion applies equally well to both cases . keeping this in mind , we now again perform the transformation @xmath710 to explicitly separate the ( semi)classical dynamics from the effect of fluctuations . it leads in first instance to @xcite @xmath711 = \sum_{\bf n } \int_{t_0 } dt~ \phi^*_{\bf n}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \xi_{\bf n}(t ) \\ & & \hspace*{0.45 in } + \sum_{\bf n } \int_{t_0 } dt~ \xi^*_{\bf n}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \phi_{\bf n}(t ) \nonumber \\ & & \hspace*{0.45 in } - \sum_{\bf n,\bf n ' } \int_{t_0 } dt \int_{t_0 } dt'~ \phi^*_{\bf n}(t ) \hbar \sigma^{(-)}_{{\bf n},{\bf n}'}(t - t ' ) \xi_{\bf n'}(t ' ) \nonumber \\ & & \hspace*{0.45 in } - \sum_{\bf n,\bf n ' } \int_{t_0 } dt \int_{t_0 } dt'~ \xi^*_{\bf n}(t ) \hbar \sigma^{(+)}_{{\bf n},{\bf n}'}(t - t ' ) \phi_{\bf n'}(t ' ) \nonumber \\ & & \hspace*{0.45 in } - \frac{1}{2 } \sum_{{\bf n},{\bf n ' } } \int_{t_0 } dt \int_{t_0 } dt'~ \xi^*_{\bf n}(t ) \hbar \sigma^{k}_{{\bf n},{\bf n'}}(t - t ' ) \xi_{\bf n'}(t')~ , \nonumber\end{aligned}\ ] ] where we introduced the retarded and advanced components of the selfenergy @xmath712 that affect the terms in the action that are linear in @xmath713 and @xmath714 , respectively , and also the keldysh component @xmath715 that is associated with the part quadratic in the fluctuations . the physical content of these various components of the selfenergy is understood most clearly if we now apply again the beautiful functional - integral procedure that is due to hubbard and stratonovich . the basic idea is to write the factor @xmath716 in the integrant of @xmath111d[\phi ] \int d[\xi^*]d[\xi]~ \exp\left ( is^{\rm eff}[\phi^*,\phi;\xi^*,\xi]/\hbar \right)$ ] as a gaussian integral over a complex field @xmath717 . it is equivalent , but in practice more convenient , to just multiply the integrant by a factor @xmath65 that is written as a gaussian integral @xmath718d[\eta]~ \exp\left ( is^{\rm eff}[\eta^*,\eta]/\hbar \right)$ ] with @xmath719 \\ & & = \frac{1}{2 } \sum_{{\bf n},{\bf n } ' } \int_{t_0 } dt \int_{t_0 } dt'~ \left(2 \eta^*_{\bf n}(t ) - \sum_{\bf n '' } \int_{t_0 } dt''~ \xi^*_{\bf n''}(t '' ) \hbar \sigma^k_{{\bf n}'',{\bf n}}(t''-t ) \right ) \nonumber \\ & & \times~ ( \hbar\sigma^k)^{-1}_{{\bf n},{\bf n}'}(t - t ' ) \left(2 \eta_{\bf n'}(t ' ) - \sum_{\bf n '' } \int_{t_0 } dt''~ \hbar \sigma^k_{{\bf n}',{\bf n}''}(t'-t '' ) \xi_{\bf n''}(t '' ) \right ) \nonumber\end{aligned}\ ] ] a complete square . adding this to @xmath720 $ ] the total effective action becomes @xmath721 \\ & & = \sum_{{\bf n},{\bf n } ' } \int_{t_0 } dt \int_{t_0 } dt'~ \phi^*_{\bf n}(t ) \left\ { \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu - \eta^*_{\bf n}(t ) \right ) \delta_{{\bf n},{\bf n } ' } \delta(t - t ' ) \right . \nonumber \\ & & \hspace*{2.57 in } - \left . \raisebox{0.2 in } { } \hbar \sigma^{(-)}_{{\bf n},{\bf n}'}(t - t ' ) \right\ } \xi_{\bf n'}(t ' ) \nonumber \\ & & + \sum_{{\bf n},{\bf n } ' } \int_{t_0 } dt \int_{t_0 } dt'~ \xi^*_{\bf n}(t ) \left\ { \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu - \eta_{\bf n}(t ) \right ) \delta_{{\bf n},{\bf n } ' } \delta(t - t ' ) \right . \nonumber \\ & & \hspace*{2.5 in } - \left . \raisebox{0.2 in } { } \hbar \sigma^{(+)}_{{\bf n},{\bf n}'}(t - t ' ) \right\ } \phi_{\bf n'}(t ' ) \nonumber \\ & & + 2 \sum_{{\bf n},{\bf n } ' } \int_{t_0 } dt \int_{t_0 } dt'~ \eta^*_{\bf n}(t ) ( \hbar\sigma^k)^{-1}_{{\bf n},{\bf n}'}(t - t ' ) \eta_{\bf n'}(t ' ) \nonumber\end{aligned}\ ] ] and is thus linear in @xmath714 and @xmath713 . integrating over these fluctuations we conclude from this action that the field @xmath722 is constraint to obey the langevin equations @xcite @xmath723 and @xmath724 with gaussian noise terms @xmath725 and @xmath726 that from the last term in the right - hand side of equation ( [ snoise ] ) are seen to have the time correlations @xmath727 in the thermodynamic limit . moreover , in that limit the retarded selfenergy is equal to @xmath728 and the advanced selfenergy can be found from the relation @xmath729 . note that the heaviside functions in these retarded and advanced selfenergies are precisely such that eqs . ( [ lan1 ] ) and ( [ lan2 ] ) are causal , i.e. , the evolution of @xmath722 and @xmath730 depends only on their value at previous times . this is clearly a sensible result and explains why these components of the full selfenergy enter into the terms in the action that are linear in the fluctuations . to understand why the keldysh component enters into the terms that are quadratic in the fluctuations is more complicated and is related to the famous fluctuation - dissipation theorem @xcite . physically , the fluctuation - dissipation theorem ensures in our case that , due to the interactions with the reservoir , the gas in the trap now relaxes towards thermal equilibrium in the limit @xmath731 . let us therefore study this limit in somewhat more detail . however , before we do so we would like to mention for completeness that if we apply the above hubbard - stratonovich transformation to the functional integral for the probability distribution @xmath595$],we immediately find that analogous to equation ( [ prob3 ] ) @xmath732 = \int d[\eta^*]d[\eta]~ p[\eta^*,\eta ] \int d[\phi^*_0]d[\phi_0]~ p[|\phi_0|^2;t_0 ] \\ & & \hspace*{1.9 in } \times \prod_{\bf n } \delta \left ( |\phi_{\bf n } - \phi^{\rm cl}_{\bf n}(t)|^2 \right)~ , \nonumber\end{aligned}\ ] ] where @xmath630 now obeys the langevin equation in equation ( [ lan1 ] ) for a particular realization of the noise @xmath725 and with the initial condition @xmath733 . furthermore , the probability distribution for each realization of the noise is given by @xmath734 \\ & & = \exp \left\ { \frac{2i}{\hbar } \sum_{{\bf n},{\bf n } ' } \int_{t_0}^t dt ' \int_{t_0}^t dt''~ \eta^*_{\bf n}(t ' ) ( \hbar\sigma^k)^{-1}_{{\bf n},{\bf n}'}(t'-t '' ) \eta_{\bf n'}(t '' ) \right\}~. \nonumber\end{aligned}\ ] ] this result shows explicitly that , as expected , the fokker - planck equation associated with the langevin equations in eqs . ( [ lan1 ] ) and ( [ lan2 ] ) is in fact the desired fokker - planck equation for the probability distribution @xmath595 $ ] . after this slight digression , we now return to the long - time behaviour of the gas . in the long - time limit two important simplifications occur . looking at the expresions for the selfenergies , we see that their width as a function of the time difference @xmath735 is at most of order @xmath736 in the low - temperature regime of interest where the thermal de broglie wavelength of the particles obeys @xmath674 . if therefore @xmath737 we are allowed to take the limit @xmath738 . taking this limit means physically that we neglect the initial transients that are due to the precise way in which the contact between the trap and the reservoir is made , and focus on the ` universal ' dynamics that is independent of these details . in addition , at long times the dynamics of the gas is expected to be sufficiently slow that we can neglect the memory effects altogether . finally , we consider here first the case of a reservoir that is so weakly coupled to the gas in the trap that we can treat the coupling with second order perturbation theory . as a result we can also neglect the nondiagonal elements of the selfenergies . in total we are then allowed to put ( but see below ) @xmath739 where the fourier transform of the selfenergies is defined by @xmath740 note that this implies in general that @xmath741 , and in particular therefore that @xmath742 . with these simplifications our langevin equations become @xmath743 and just the complex conjugate equation for @xmath744 . the retarded selfenergy in this equation is given by @xmath745 with @xmath746 the usual notation for the limiting procedure @xmath747 . it clearly has real and imaginary parts that we denote by @xmath748 and @xmath749 , respectively . denoting the cauchy principle value part of an integral by @xmath750 , they obey @xmath751 and @xmath752 the interpretation of these results is quite obvious if we consider the average of the langevin equation , i.e. , @xmath753 which is solved by @xmath754 hence the real part of the retarded selfenergy @xmath748 represents the shift in the energy of state @xmath53 due to the coupling with the reservoir . indeed , equation ( [ shift ] ) agrees precisely with the value of this shift in second order perturbation theory . moreover , the fact that @xmath755 shows that the average rate of decay @xmath756 of the state @xmath53 is equal to @xmath757 , which , together with equation ( [ decay ] ) , exactly reproduces fermi s golden rule . from the equation of motion for the average @xmath758 we can also conclude that the right - hand side of the fokker - planck equation contains the ` streaming ' terms @xmath759 \hspace*{1.5 in } \nonumber \\ + \left ( \sum_{\bf n } \frac{\partial}{\partial \phi^*_{\bf n } } ( \epsilon_{\bf n } + \hbar\sigma^{(-)}_{\bf n } - \mu ) \phi^*_{\bf n } \right ) p[\phi^*,\phi;t]~. \nonumber\end{aligned}\ ] ] however , there must now also be a ` diffusion ' term due to the fact that the average @xmath645 is no longer independent of time since the interactions with the reservoir can lead to changes in the occupation numbers @xmath644 . to obtain the ` diffusion ' term we thus need to determine @xmath760 . to do so we first formally solve the langevin equation by @xmath761 multiplying this with the complex conjugate expression and taking the average , we obtain first of all @xmath762 which shows that @xmath763 moreover , the keldysh component of the selfenergy is given by @xmath764 and therefore obeys @xmath765 with @xmath766 the bose distribution function . this is in fact the fluctuation - dissipation theorem , because it relates the strength of the fluctuations determined by @xmath767 , to the amount of dissipation that is given by @xmath768 . as mentioned previously , the fluctuation - dissipation theorem ensures that the gas relaxes to thermal equilibrium . this we can already see from equation ( [ fluc ] ) , because substitution of the fluctuation - dissipation theorem leads to @xmath769 which tells us that in equilibrium @xmath770 . in fact , we have argued that in general @xmath771 . substituting this identity in equation ( [ fluct ] ) and realizing that @xmath772 due to the energy - conserving @xmath440 function in @xmath756 , we indeed obtain the correct rate equation for the average occupation numbers @xmath773 that might justly be called the quantum boltzmann equation for the gas , because the right - hand side contains precisely the rates for scattering into and out of the reservoir . furthermore , equation ( [ fluc ] ) shows that the ` diffusion ' term in the fokker - planck equation is @xmath774~ , \nonumber\end{aligned}\ ] ] because the first term in the right - hand side of equation ( [ fluc ] ) is also present without the noise and is therefore already accounted for in the ` streaming ' terms . in total we have thus arrived at @xmath775 = - \left ( \sum_{\bf n } \frac{\partial}{\partial \phi_{\bf n } } ( \epsilon_{\bf n } + \hbar\sigma^{(+)}_{\bf n } - \mu ) \phi_{\bf n } \right ) p[\phi^*,\phi;t ] \\ & & \hspace*{0.2 in } + \left ( \sum_{\bf n } \frac{\partial}{\partial \phi^*_{\bf n } } ( \epsilon_{\bf n } + \hbar\sigma^{(-)}_{\bf n } - \mu ) \phi^*_{\bf n } \right ) p[\phi^*,\phi;t ] \nonumber \\ & & \hspace*{0.2 in } - \left ( \frac{1}{2 } \sum_{\bf n } \frac{\partial^2}{\partial \phi^*_{\bf n } \partial \phi_{\bf n } } \hbar \sigma^k_{\bf n } \right ) p[\phi^*,\phi;t]~. \nonumber \end{aligned}\ ] ] using again the fluctuation - dissipation theorem , it is not difficult to show that the stationary solution of this fokker - planck equation is @xmath776 = \prod_{\bf n } \frac{1}{n_{\bf n } + 1/2 } \exp \left\ { - \frac{1}{n_{\bf n } + 1/2 } |\phi_{\bf n}|^2 \right\}~.\ ] ] although this appears to be quite reasonable , it is important to note that this result is not fully consistent because the ` streaming ' terms in equation ( [ fp2 ] ) show that the energies of the states @xmath53 are shifted and equal to @xmath777 . we therefore expect that the equilibrium occupation numbers are equal to the bose distribution evaluated at those energies and not at the unshifted values @xmath19 . the reason for this inconsistency is , of course , that we have taken @xmath778 equal to @xmath779 , which was only justified on the basis of second - order perturbation theory . we now , however , see that to obtain a fully consistent theory we must take instead @xmath780 . this implies that to obtain the retarded selfenergy , we first have to solve for the real part @xmath781 and then substitute this into @xmath782 to obtain the imaginary part . the fluctuation - dissipation theorem then again reads @xmath783 but now with the equilibrium distribution @xmath784 . as a consequence the quantum boltzmann equation and the stationary state of the fokker - planck equation will now also contain these correct equilibrium occupation numbers @xcite . having arrived at the fokker - planck equation of our caldeira - leggett - like toy model , we have now demonstrated the use of essentially all the theoretical tools that are required for a discussion of a weakly interacting bose gas . before we start with this discussion , however , we want to make a final remark about the effect of the nondiagonal elements of the selfenergies . physically , including these nondiagonal elements accounts for the change in the wave functions @xmath53 due to the interaction with the reservoir . it can clearly be neglected if the coupling with the reservoir is sufficiently weak , or more precisely if @xmath785 is much smaller than the energy splitting @xmath786 between the states in the trap . for a quantum dot in solid - state phyisics @xcite this is often the case and for that reason we might call our toy model a bosonic quantum dot . however , in magnetically trapped atomic gases the average interaction energy between the atoms is already slightly below the bose - einstein condensation critical temperature larger than the typical energy splitting and the nondiagonal elements of the selfenergies are very important . such a strong - coupling situation can , of course , also be treated in our caldeira - leggett like model . the main difference is that we need to expand our various fields not in the eigenstates of the trapping potential @xmath53 but in the eigenstates @xmath318 of the nonlocal schrdinger equation @xmath787 \chi'_{\bf n}({\bf x } ' ) \nonumber\end{aligned}\ ] ] where @xmath788 are the new eigenvalues and the retarded selfenergie in coordinate space is @xmath789 . in this new basis the nondiagonal elements of the selfenergies can now be neglected and we find essentially the same results as before . we only need to replace @xmath790 by @xmath788 and @xmath670 by @xmath791 . neglecting the nondiagonal elements in this basis only requires that the real part of the retarded selfenergy is much larger that its imaginary part , which is always fulfilled in our case because @xmath792 . summarizing , the coherent and incoherent dynamics of the gas in the trap is for our toy model quite generally solved by @xmath793 and @xmath794 respectively . in the limit @xmath731 , the average of the annihilation operators @xmath758 thus always vanishes but the average occupation numbers @xmath644 relax to the equilibrium distribution @xmath795 . although this appears to be an immediately obvious result , its importance stems from the fact that it is also true if we tune the potential energy bias @xmath667 such that at low temperatures the groundstate @xmath796 acquires a macroscopic occupation , i.e. , @xmath797 . the gas therefore never shows a spontaneous breaking of the @xmath424 symmetry , in agreement with the notion that we are essentially dealing with an ideal bose gas in the grand - canonical ensemble @xcite . the reason of the absense of the spontaneous symmetry breaking can also be understood from our stationary solution of the fokker - planck equation in equation ( [ stat ] ) , which shows that the probability distribution for @xmath798 is proportional to the boltzmann factor @xmath799 in the degenerate regime of interest . hence , the corresponding landau free energy @xmath800 never shows an instability towards the formation of a nonzero average of @xmath798 due to the fact that @xmath801 can never become less then zero . once we introduce interactions between the atoms in the gas , this picture fully changes . we are now in a position to discuss an interacting bose gas , because in an interacting bose gas , the gas is roughly speaking its own thermal reservoir . the fokker - planck equation therefore turns out to be quite similar to the one presented in equation ( [ fp2 ] ) . to be more precise , however , we should mention that we aim in this section to describe the formation of a condensate in an interacting bose gas under the conditions that have recently been realized in experiments with atomic rubidium , lithium , sodium and hydrogen gases . in these experiments the gas is cooled by means of evaporative cooling . numerical solutions of the boltzmann equation have shown that during evaporative cooling the energy distribution function is well described by an equilibrium distribution with time dependent temperature @xmath802 and chemical potential @xmath803 , that is truncated at high energies due to the evaporation of the highest energetic atoms from the trap @xcite . moreover , in the experiments of interest the densities just above the critical temperature are essentially always such that the gas is in the weak - coupling limit , which implies in this context that the average interaction energy per atom is always much less than the energy splitting of the one - particle states in the harmonic trapping potential . for a trapped atomic gas the appropriate one - particle states are the eigenstates @xmath53 of the external potential @xmath6 . as a result the fokker - planck equation for the interacting gas will now quite generally be given by @xmath804 = - \sum_{{\bf n},{\bf n } ' } \frac{\partial}{\partial \phi_{{\bf n } } } \left\ { \left ( ( \epsilon_{{\bf n } } - \mu(t ) ) \delta_{{\bf n},{\bf n } ' } + \hbar\sigma^{(+)}_{{\bf n},{\bf n } ' } \right ) \phi_{{\bf n } ' } \raisebox{0.3 in } { } \right . \\ & & \hspace*{1.4 in } + \left . \sum_{{\bf m},{\bf m } ' } v^{(+)}_{{\bf n},{\bf m};{\bf n}',{\bf m } ' } \phi^*_{{\bf m } } \phi_{{\bf m } ' } \phi_{{\bf n } ' } \right\ } p[\phi^*,\phi;t ] \nonumber \\ & & + \sum_{{\bf n},{\bf n } ' } \frac{\partial}{\partial \phi^*_{{\bf n } } } \left\ { \left ( ( \epsilon_{{\bf n } } - \mu(t ) ) \delta_{{\bf n},{\bf n } ' } + \hbar\sigma^{(-)}_{{\bf n}',{\bf n } } \right ) \phi^*_{{\bf n } ' } \raisebox{0.3 in } { } \right . \nonumber \\ & & \hspace*{1.4 in } + \left . \sum_{{\bf m},{\bf m } ' } v^{(-)}_{{\bf n}',{\bf m}';{\bf n},{\bf m } } \phi^*_{{\bf n } ' } \phi^*_{{\bf m } ' } \phi_{{\bf m } } \right\ } p[\phi^*,\phi;t ] \nonumber \\ & & - \frac{1}{2 } \sum_{{\bf n},{\bf n } ' } \frac{\partial^2}{\partial \phi_{{\bf n } } \partial \phi^*_{{\bf n } ' } } \left\ { \hbar\sigma^k_{{\bf n},{\bf n } ' } + \sum_{{\bf m},{\bf m } ' } v^k_{{\bf n},{\bf m};{\bf n}',{\bf m } ' } \phi^*_{{\bf m}}\phi_{{\bf m } ' } \right\ } p[\phi^*,\phi;t]~. \nonumber\end{aligned}\ ] ] for this equation to be useful , however , we need to determine the various selfenergies and interactions . to do so , let us for simplicity first consider the normal state of the gas which is usually in the weak - coupling limit @xmath805 for the experiments with atomic alkali gases that are of interest here . in this regime we can neglect the nondiagonal elements of the selfenergies and also the nonlinear terms in the right - hand side of equation ( [ fpinhom ] ) . the fokker - planck equation then reduces to @xmath806 = - \sum_{{\bf n } } \frac{\partial}{\partial \phi_{{\bf n } } } ( \epsilon_{{\bf n } } + \hbar\sigma^{(+)}_{{\bf n } } - \mu(t ) ) \phi_{{\bf n } } p[\phi^*,\phi;t ] \\ & & \hspace*{0.23 in } + \sum_{{\bf n } } \frac{\partial}{\partial \phi^*_{{\bf n } } } ( \epsilon_{{\bf n } } + \hbar\sigma^{(-)}_{{\bf n } } - \mu(t ) ) \phi^*_{{\bf n } } p[\phi^*,\phi;t ] \nonumber \\ & & \hspace*{0.23 in } - \frac{1}{2 } \sum_{{\bf n } } \frac{\partial^2}{\partial \phi^*_{{\bf n } } \partial \phi_{{\bf n } } } \hbar\sigma^k_{{\bf n } } p[\phi^*,\phi;t]~ , \nonumber\end{aligned}\ ] ] with @xmath807 and @xmath808 . this equation looks identical to the one we derived for our caldeira - leggett toy model but describes in fact quite different physics because now the selfenergies are not due to an interaction with a reservoir but due to the interactions between the atoms themselves . this is , however , completely hidden in the selfenergies as we show now explicitly . performing the same manipulations as in sections [ ifd ] and [ ap ] , we find in the many - body t - matrix approximation first of all that @xmath809 and similarly that @xmath810 our next task is therefore the evaluation of the various many - body t - matrix elements @xmath811 that occur in these expressions , which is easily carried out with our experience of the homogeneous gas and by noting that the lippmann - schwinger equation is in fact an operator equation that only needs to be expressed in a new basis . in terms of the appropriate matrix elements of the interatomic interaction , i.e. , @xmath812 the retarded and advanced components of the many - body t matrix thus obey the bethe - salpeter equation @xcite @xmath813 and the optical theorem @xmath814 finally , we also need @xmath815 and the keldysh component @xmath816 notice that the appearance of a @xmath440 function in the last three equations in practice never leads to any problems , because in the experiments with trapped atomic gases the number of particles is always so large that the critical temperature of the gas is much larger than the energy splitting of the trap and the discrete sum over states is well approximated by an integral over a continuous spectrum . making use of the above relations we find that for the states with thermal energies the renormalized energy is given by the pseudopotential result @xmath817 with @xmath818 the density profile of the gas cloud . for the states near the one - particle ground state there is , however , a correction given by @xmath819 and @xmath820 . in the weak - coupling limit , the renormalization of the one - particle energies plays by definition not such an important role and we are much more interested in the imaginary part of the retarded and advanced selfenergies . they are given by @xmath821~. \nonumber \end{aligned}\ ] ] furthermore , the keldysh component of the selfenergies equals @xmath822~. \nonumber \end{aligned}\ ] ] as a result the quantum boltzmann equation for the gas becomes @xmath823 with the rate to scatter out of the state @xmath824 given by @xmath825 and the rate to scatter into this state by @xmath826 as expected from fermi s golden rule . the quantum boltzmann equation found above , in combination with the expressions for the shift in the energy levels , fully describes the dynamics of the gas in the normal state . however , close to the critical temperature when the occupation numbers of the one - particle groundstate @xmath827 start to become large , we are no longer allowed to neglect the nonlinear terms in the fokker - planck equation in equation ( [ fpinhom ] ) . applying again the hartree - fock approximation , which is essentially exact in the weak - coupling limit as we have seen in the equilibrium case , and substituting the ansatz @xmath595 = p_0[\phi^*_{\bf 0},\phi_{\bf 0};t ] p_1[\phi'^*,\phi';t]$ ] into equation ( [ fpinhom ] ) , we obtain for the probability distribution of the condensate @xmath828 \\ & & = - \frac{\partial}{\partial \phi_{\bf 0 } } \left ( \epsilon_{\bf 0 } + \hbar\sigma^{(+)}_{\bf 0 } - \mu(t ) + t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0}}(t ) |\phi_{\bf 0}|^2 \right ) \phi_{\bf 0 } p_0[\phi^*_{\bf 0},\phi_{\bf 0};t ] \nonumber \\ & & + \frac{\partial}{\partial \phi^*_{\bf 0 } } \left ( \epsilon_{\bf 0 } + \hbar\sigma^{(-)}_{\bf 0 } - \mu(t ) + t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0}}(t ) |\phi_{\bf 0}|^2 \right ) \phi^*_{\bf 0 } p_0[\phi^*_{\bf 0},\phi_{\bf 0};t ] \nonumber \\ & & - \frac{1}{2 } \frac{\partial^2}{\partial \phi^*_{\bf 0 } \partial \phi_{\bf 0 } } \hbar \sigma^k_{\bf 0 } p_0[\phi^*_{\bf 0},\phi_{\bf 0};t]~ , \nonumber \end{aligned}\ ] ] where it should be remembered that for consistency reasons the selfenergies need to be evaluated at the energies @xmath829 . in equilibrium the fluctuation - dissipation theorem then reads @xmath830 \propto \exp \left\ { -\beta \left ( \epsilon'_{\bf 0 } - \mu + \frac{t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0}}}{2 } |\phi_{\bf 0}|^2 \right ) |\phi_{\bf 0}|^2 \right\}~. \nonumber\end{aligned}\ ] ] the fokker - planck equation , and hence the boltzmann equation , for the noncondensed part of the gas is the same as in the normal state because for a discrete system the contributions from the condenstate to the selfenergies is already included in our expressions for @xmath831 and needs not be separated out explicitly . this implies of course , that if we evaluate the rates @xmath832 and @xmath833 by replacing the sum over states by an integral over a continuum , we first need to separate the contribution from the condensate to obtain the correct result . in summary we have thus found that in the weak - coupling limit the dynamics of bose - einstein condensation in an external trapping potential is described by the coupled equations in equation ( [ beinhom ] ) and equation ( [ fpcinhom ] ) , i.e. , by a boltzmann equation for the noncondensed part of the gas and a fokker - planck equation for the condensate , respectively . if we apply these equations to a bose gas with a positive scattering length , we essentially arrive at the theory recently put forward by gardiner and coworkers @xcite . indeed , neglecting the energy shifts and assuming that the noncondensed cloud is in equilibrium with a chemical potential @xmath834 , our fokker - planck equation for a @xmath424 invariant probability distribution of the condensate becomes @xmath835 = - \frac{\beta}{4 } \hbar \sigma^k_{\bf 0 } \frac{\partial}{\partial \phi_{\bf 0 } } \left ( \epsilon_{\bf 0 } - \mu + t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0 } } |\phi_{\bf 0}|^2 \right ) \phi_{\bf 0 } p_0[|\phi_{\bf 0}|;t ] \\ & & \hspace*{0.2 in } - \frac{\beta}{4 } \hbar \sigma^k_{\bf 0 } \frac{\partial}{\partial \phi^*_{\bf 0 } } \left ( \epsilon_{\bf 0 } - \mu + t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0 } } |\phi_{\bf 0}|^2 \right ) \phi^*_{\bf 0 } p_0[|\phi_{\bf 0}|;t ] \nonumber \\ & & \hspace*{0.2 in } - \frac{1}{2 } \hbar \sigma^k_{\bf 0 } \frac{\partial^2}{\partial \phi^*_{\bf 0 } \partial \phi_{\bf 0 } } p_0[|\phi_{\bf 0}|;t]~. \nonumber \end{aligned}\ ] ] . the inset shows the evolution of the average condensate number and its fluctuations . ] it is equivalent to the fokker - planck equation for a single - mode laser , since introduction of the dimensionless time @xmath836 , the dimensionless condensate number @xmath837 , and the so - called ` pump parameter ' @xmath838 leads to the fokker - planck equation @xmath839 = \frac{\partial}{\partial i } \left\ { 2(i - a)i + 4i \frac{\partial}{\partial i } \right\ } p[i;\tau]\ ] ] well - known from laser theory @xcite . the typical behaviour of @xmath840 $ ] during bose - einstein condensation in shown in figure 17 . calculating @xmath841 from equation ( [ sml ] ) by making use of the mean - field assumption @xmath842 , precisely reproduces the theory of gardiner _ et al . _ a comparison with the experiments of h .- j . shows that the theory of gardiner and coworkers is unfortunately not fully satisfactory @xcite . the reason for this is presumably that these experiments are really in the strong - coupling limit , where the above theory does not apply . as mentioned previously , the theory is however applicable to gases with a negative scattering length . since we know from our experience with homogeneous gases that the chemical potential will never be larger than @xmath843 in this case , we clearly see from the equilibrium probability distribution @xmath844 $ ] that a condensate in a gas with effectively attractive interactions is only metastable and will ultimately collapse to a dense state after it has overcome an energy barrier , either by thermal fluctuations or by quantum mechanical tunneling . this is clearly in full agreement with the discussion in section [ nsl ] . furthermore , we now also understand how the dynamics of the collapse must be treated . it requires the solution of a stochastic , dissipative nonlinear schrdinger equation , in which the dissipative terms and the noise obey an appropriate fluctuation - dissipation theorem . having arrived at this important result , we are finally left with a discussion of the strong - coupling limit . the difficulty with the strong - coupling limit is that the appropriate one - particle states of the gas are no longer the states @xmath824 of the trapping potential because these states are now strongly coupled by the mean - field interaction . for our caldeira - leggett model this problem could be easily resolved by introducing the states @xmath845 that incorporate the average interaction with the reservoir . however , for a trapped bose gas the same procedure would lead to an expansion into states @xmath846 that parametrically depend on time since the mean - field interaction depends on time also . in principle , we thus have to deal with the situation that not only the occupation numbers vary with time but also the states to which these occupation numbers belong . this results in highly nontrivial dynamics since it is _ a priori _ clear that an adiabatic approximation , in which we neglect the nondiagonal matrix elements @xmath847 , is not valid , because of the lack of a separation of time scales . notice that we were able to circumvent the above problems for a bose gas with effectively attractive interactions due to the fact that on the one hand the growth of the condensate essentially takes place in the weak - coupling regime and that on the other hand the collapse , which certainly occurs in the strong - coupling regime , requires in first instance only knowledge of the collisionless dynamics of the condensate . notwithstanding the above remarks , we can make progress in the strong - coupling limit if we realize that in this limit the typical length scale on which the ( noncondensate ) density of the gas varies is always much larger than the trap size . we are therefore justified in performing a gradient expansion . to do so in a systematic fashion , it is convenient to express the various selfenergies and interactions in the coordinate representation , i.e. , by introducing @xmath848 and similarly @xmath849 in this coordinate representation , our fokker - planck equation acquires in all generality the form @xmath850 = - \int d{\bf x } d{\bf x}'~ \frac{\delta}{\delta \phi({\bf x } ) } \\ & & \hspace*{0.2 in } \left\ { \left ( \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu(t ) \right ) \delta({\bf x}-{\bf x } ' ) + \hbar\sigma^{(+)}({\bf x},{\bf x } ' ) \right ) \phi({\bf x } ' ) \right . \nonumber \\ & & \hspace*{0.4 in } + \left . \int d{\bf y } d{\bf y}'~ v^{(+)}({\bf x},{\bf x}';{\bf y},{\bf y } ' ) \phi^*({\bf y } ) \phi({\bf y } ' ) \phi({\bf x } ' ) \raisebox{0.2 in } { } \right\ } p[\phi^*,\phi;t ] \nonumber \\ & & + \int d{\bf x } d{\bf x}'~ \frac{\delta}{\delta \phi^*({\bf x } ) } \nonumber \\ & & \hspace*{0.2 in } \left\ { \left ( \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu(t ) \right ) \delta({\bf x}-{\bf x } ' ) + \hbar\sigma^{(-)}({\bf x}',{\bf x } ) \right ) \phi^*({\bf x } ' ) \right . \nonumber \\ & & \hspace*{0.4 in } + \left . \int d{\bf y } d{\bf y}'~ v^{(-)}({\bf x}',{\bf x};{\bf y}',{\bf y } ) \phi^*({\bf x } ' ) \phi^*({\bf y } ' ) \phi({\bf y } ) \raisebox{0.2 in } { } \right\ } p[\phi^*,\phi;t ] \nonumber \\ & & - \frac{1}{2 } \int d{\bf x } d{\bf x}'~ \frac{\delta^2}{\delta \phi({\bf x } ) \delta \phi^*({\bf x } ' ) } \left\ { \hbar\sigma^k({\bf x},{\bf x } ' ) \raisebox{0.2 in } { } \right . \nonumber \\ & & \hspace{0.4 in } + \left . \int d{\bf y } d{\bf y}'~ v^k({\bf x},{\bf x}';{\bf y},{\bf y } ' ) \phi^*({\bf y } ) \phi({\bf y } ' ) \right\ } p[\phi^*,\phi;t]~. \nonumber\end{aligned}\ ] ] to study its physical content and to familiarize ourselves with how to perform the desired gradient expansion , we first analyze again the normal state of the gas in which we can neglect the nonlinear terms representing the effect of the condensate . in the normal state the condensate is absent and we are only interested in the behavior of the quantity @xmath851 . from the above fokker - planck equation we see that it obeys the equation of motion @xmath852 introducing now the wigner distribution @xcite by means of @xmath853 and similarly the selfenergies @xmath854 by @xmath855 we find after substituting these relations in lowest nonvanishing order in the gradient @xmath856 the expected quantum boltzmann equation @xcite @xmath857 where the energies of the atoms have to be determined selfconsistently from @xmath858~.\ ] ] moreover , the expressions for the retarded energy @xmath859 , and the collision rates @xmath860 and @xmath861 turn out to be identical to the ones for a homogeneous gas @xcite . we only need to replace the one - particle energies @xmath862 and occupation numbers @xmath863 by their local counterparts @xmath864 and @xmath865 , respectively . notice also that the same substitution has to be made in the bethe - salpeter equation for the many - body t matrix , which implies that the effective interaction now implicitly depends both on the position @xmath51 and the time @xmath30 . as an example , let us consider the gas not too close to the critical temperature . then the effective interaction is well approximated by the two - body t matrix and we simply have @xmath866 as a result the equilibrium solution @xmath867 of the quantum boltzmann equation equals a bose - einstein distribution evaluated at the energy @xmath868 . the equilibrium density profile @xmath869 thus corresponds to a local - density approximation but is indeed very accurate for the atomic gases of interest . apart from describing the relaxation of the gas towards equilibrium , the quantum boltzmann equation can also be used to study the collective modes of the trapped gas above the critical temperature . using the same methods that have recently been applied to inhomogeneous bose gases by zaremba , griffin , and nikuni @xcite , we can for example derive the hydrodynamic equations of motion the gas . moreover , the collisionless modes are obtained from @xmath870 solving this equation is equivalent to the generalized random - phase approximation @xcite . more important is that the above collisionless boltzmann equation can be shown to satisfy the kohn theorem @xcite by substituting the shifted equilibrium profile @xmath871 . for an harmonic external potential the center - of - mass of the gas cloud then precisely follows newton s law @xmath872 , which shows that the gas has three collisionless modes with frequencies that are equal to the three frequencies of the external trapping potential . the full quantum boltzmann equation in equation ( [ qbe ] ) also describes the relaxation of the gas towards equilibrium . it can thus be used to study forced evaporative cooling by adding an appropriate loss term to account for the escaping high - energetic atoms . in agreement with our previous picture we then expect that , if the evaporative cooling is performed slowly , the distribution function @xmath865 can be well approximated by a truncated equilibrium distribution with a time - dependent temperature and chemical potential . as a result , we can use the results from the homogeneous gas to show that in the critical regime the gas again becomes unstable towards bose - einstein condensation . this is the case when @xmath873 for positions near the center of the trap where the density of the gas is highest . if that happens we are no longer allowed to neglect the nonlinear terms in the fokker - planck equation and some corrections are neccesary . applying again the hartee - fock approximation , which is correct sufficiently close to the critical temperature where @xmath874 , we find first of all that the quantum boltzmann equation for the noncondensed part of the gas has the same form as in equation ( [ qbe ] ) . we only have to add the by now familiar mean - field interaction @xmath875 due to the condensate to the renormalized energy @xmath864 . finally , the scattering rates @xmath860 and @xmath861 have to be adjusted in precisely the same way as in the homogeneous case to account for the macroscopic occupation of the condensate @xcite . in addition , the resulting quantum boltzmann equation for the noncondensed atoms is coupled to a fokker - planck equation for the condensate . after a gradient expansion the latter equals @xmath876 \\ & & = - \int d{\bf x}~ \frac{\delta}{\delta \phi_{\bf 0}({\bf x } ) } \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu(t ) + \hbar\sigma^{(+)}({\bf x},{\bf 0 } ) \right . \nonumber \\ & & \hspace{1.22 in } + \left . { } ^{{}^{{}^{{}^{{}^{{}^ { } } } } } } t^{(+)}({\bf 0},0 ) |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}({\bf x } ) p_0[\phi_{\bf 0}^*,\phi_{\bf 0};t ] \nonumber \\ & & + \int d{\bf x}~ \frac{\delta}{\delta \phi_{\bf 0}^*({\bf x } ) } \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu(t ) + \hbar\sigma^{(-)}({\bf x},{\bf 0 } ) \right . \nonumber \\ & & \hspace{1.22 in } + \left . { } ^{{}^{{}^{{}^{{}^{{}^ { } } } } } } t^{(+)}({\bf 0},0 ) |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}^*({\bf x } ) p_0[\phi_{\bf 0}^*,\phi_{\bf 0};t ] \nonumber \\ & & - \frac{1}{2 } \int d{\bf x}~ \frac{\delta^2}{\delta \phi_{\bf 0}({\bf x } ) \delta \phi_{\bf 0}^*({\bf x } ) } \hbar\sigma^k({\bf x},{\bf 0 } ) p_0[\phi_{\bf 0}^*,\phi_{\bf 0};t]~ , \nonumber\end{aligned}\ ] ] where @xmath877 denotes @xmath878 . in more detail we have for the real part of the retarded and advanced selfenergies that @xmath879 $ ] and @xmath880 the local energy of a condensate atom . moreover , using our picture of the evaporative cooling of the gas , the imaginary parts follow most conveniently from the fluctuation - dissipation theorem @xmath881 where @xmath882 can in a good approximation be considered as independent of @xmath883 . near equilibrium the fokker - planck equation thus simplifies to @xmath884 \\ & & = - \frac{\beta}{4 } \int d{\bf x}~ \hbar\sigma^k({\bf x},{\bf 0 } ) \frac{\delta}{\delta \phi_{\bf 0}({\bf x } ) } \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + \epsilon'({\bf x},{\bf 0 } ) - \mu \right . \nonumber \\ & & \hspace{1.2 in } + \left . { } ^{{}^{{}^{{}^{{}^{{}^ { } } } } } } t^{(+)}({\bf 0},0 ) |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}({\bf x } ) p_0[|\phi_{\bf 0}|;t ] \nonumber \\ & & - \frac{\beta}{4 } \int d{\bf x}~ \hbar\sigma^k({\bf x},{\bf 0 } ) \frac{\delta}{\delta \phi_{\bf 0}^*({\bf x } ) } \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + \epsilon'({\bf x},{\bf 0 } ) - \mu \right . \nonumber \\ & & \hspace{1.2 in } + \left . { } ^{{}^{{}^{{}^{{}^{{}^ { } } } } } } t^{(+)}({\bf 0},0 ) |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}^*({\bf x } ) p_0[|\phi_{\bf 0}|;t ] \nonumber \\ & & - \frac{1}{2 } \int d{\bf x}~ \hbar\sigma^k({\bf x},{\bf 0 } ) \frac{\delta^2}{\delta \phi_{\bf 0}({\bf x } ) \delta \phi_{\bf 0}^*({\bf x } ) } p_0[|\phi_{\bf 0}|;t ] \nonumber\end{aligned}\ ] ] and the probability distribution of the condensate relaxes to the expected stationary state @xmath885 \propto \exp \left\ { - \beta \int d{\bf x}~ \phi_{\bf 0}^*({\bf x } ) \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + \epsilon'({\bf x},{\bf 0 } ) - \mu \right . \right . \\ & & \hspace*{2.0 in } \left . \left . + \frac{t^{(+)}({\bf 0},0)}{2 } |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}({\bf x } ) \right\}~. \nonumber\end{aligned}\ ] ] from this result we conclude that eqs . ( [ qbe ] ) and ( [ fpsc ] ) give an accurate discription of the dynamics of bose - einstein condensation in the strong - coupling regime . unfortunately , the solution of these coupled equations is not possible analytically and we therefore have to refer to future work for a numerical study and a comparison of this theory with experiment . we here only remark that the semiclassical approximation in this case amounts to the thomas - fermi result for the condensate density @xmath886 such an approximation implicitly assumes , however , that the dominant mechanism for the formation of the condensate is by means of a coherent evolution and not by incoherent collisions . this might be expected not to be very accurate on the basis of the homogeneous result that the second mechanism only diverges logarithmically with the system size @xcite , whereas the first diverges linearly @xcite . nevertheless it seems that only a full solution of equations ( [ qbe ] ) and ( [ fpsc ] ) can settle this important issue . although we have up to now primarily been focussed on the dynamics of bose - einstein condensation itself , it is interesting that we have in fact , similarly as for temperatures above the critical temperature , also arrived at an accurate discription of the collective modes of a condensed gas at nonzero temperatures . since these collective modes are receiving much attention at present , we now briefly want to discuss how such a description emerges from our fokker - planck equation . since most experiments at present are performed in the collisionless limit , we consider only this limit here . it is in principle possible , however , to derive the hydrodynamic equations of motions as well @xcite . let us for simplicitly first assume that the temperature of the gas is sufficiently far below the critical temperature that the many - body t matrix is well approximated by the two - body t matrix , but not so low that the hartree - fock approximation is no longer valid . in this regime we have that @xmath887 and the dynamics of the noncondensed part of the gas is again determined by the collisionless quantum boltzmann equation given in equation ( [ clqbe ] ) , where the total density now consists of two contributions @xmath888 the dynamics of the condensate follows again from the nonlinear schrdinger equation @xmath889 which can be rewritten as the continuity equation @xmath890 together with the josephson relation for the superfluid velocity @xmath891 here , we used that @xmath892 , @xmath893 , and have applied the thomas - fermi approximation , which is appropriate in the strong - coupling limit . it will presumably not come as a surprise that these coupled equations for @xmath865 , @xmath894 , and @xmath895 can be shown to fulfill the kohn theorem exactly . this is an important constraint on any theory for the collective modes of a trapped gas , as stressed recently by zaremba , griffin , and nikuni @xcite . as a first step towards this goal we here consider the time - dependent hartree - fock theory for the collective modes of the bose condensed gas , in which the collisionless boltzmann equation in equation ( [ clqbe ] ) is coupled to the nonlinear schrdinger equation in equation ( [ nlse ] ) . it turns out that a numerical determination of the eigenmodes of the collisionless boltzmann equation is rather difficult . therefore , we have recently proposed a variational approach to solve these coupled equations @xcite . denoting the anisotropic harmonic trapping potential used in the experiments @xcite by @xmath896 , we first of all take for the condensate wave function the gaussian _ ansatz _ @xmath897 with the functions @xmath898 describing the ` breathing ' of the condensate in three independent directions . clearly , this is the anisotropic generalization of the gaussian _ ansatz _ that we used in section [ nsl ] to study the stability of a condensate with negative scattering length . although it is not immediately obvious that such an _ ansatz _ is also appropriate in the strong - coupling regime of interest here , it has been shown by perez - garcia _ et al . _ @xcite that at zero temperature the assumption of a gaussian density profile nevertheless leads to rather accurate results for the frequencies of the collective modes . we therefore anticipate that at nonzero temperatures the same will be the case . next we also need an _ ansatz _ for the distribution function @xmath865 of the thermal cloud . here , we consider the bose distribution @xmath899 where @xmath900 are again three scaling parameters , @xmath901 is the number of noncondensed particles , @xmath902 denotes the equilibrium value of @xmath900 that in general is greater than @xmath65 due to the repulsive interactions between the atoms , and @xmath903 . the functional form of the distribution function is physically motivated by the following reasons . first , by integrating over the momentum @xmath664 we see that the time - dependent density profile of the noncondensed cloud @xmath904 is a time - independent function of @xmath905 , as desired . second , for such a density profile we must have that the local velocity @xmath906 is given by @xmath907 . third , we want the equilibrium distribution to be only anisotropic in coordinate space , and not in momentum space . this explains the appearance of the factor @xmath908 in the exponent . the full lines give the results with a bose distribution _ ansatz _ for the distribution function of the thermal cloud . curves 1 and 2 correspond to the @xmath909 in and out - of - phase modes , respectively . similarly , curves 3 and 4 give the frequencies of the @xmath910 in and out - of - phase modes . the experimental data is also shown . triangles are for a @xmath909 mode and circles for a @xmath910 mode . ] using the ideal gas result for the number of condensate particles and substituting the above _ ansatz _ for the condensate wave function and the wigner distribution function into the nonlinear schrdinger and collisionless boltzmann equations , it is straightforward but somewhat tedious to derive the six coupled equations of motion for the ` breathing ' parameters @xmath898 and @xmath900 . linearizing these equations around the equilibrium solutions @xmath911 and @xmath902 , respectively , we can then also obtain the desired eigenmodes and eigenfrequencies . the results of this calculation for the trap parameters of jin _ et al . _ @xcite are shown in figure 18 together with the experimental data . note that the trap has a disk geometry with @xmath912 and @xmath913 . due to the axial symmetry , the modes of the gas can be labeled by the usual quantum number @xmath4 that describes the angular dependence around the @xmath914-axis . on the whole there appears to be a quite reasonable agreement between this simple theory and experiment . the only exceptions are the two data point almost halfway in between the in and out - of - phase @xmath909 modes . a possible explanation for this discrepancy is that experimentally both modes are excited simultaneously @xcite . to make sure of this , however , we need to determine the oscillator strength of the various modes and it appears that the in - phase mode indeed dominates the response for temperatures larger than about @xmath915 @xcite . with the latter remark we end this course on the use of field - theoretical methods for the study of the equilibrium and nonequilibrium properties of trapped atomic gases . although we have in principle illustrated all the necessary tools for an _ ab initio _ treatment of these new quantum systems , many interesting topics still need to be considered in detail . we already mentioned such topics as fermi - bose mixtures , the dynamics of condensate collapse , the strong - coupling theory for the formation of a condensate in either atomic bose of fermi gases , and the hydrodynamics of single or multicomponent atomic gases . several other important areas of research , which can also be easily addressed within the context of quantum field theory , are atom lasers , the damping of collective modes , the behaviour of spinor condensates , the dynamics of topological excitations such as kinks , vortices and skyrmions , quantum critical phenomena , the optical properties of superfluid gases , and two - dimensional phase transitions . in view of these and many other possibilities that are now experimentally feasible , it appears certain that the physics of degenerate atomic gases will remain very exciting for years to come . 106 e. abers and b.w . lee , phys . rep . * 9c * , 1 ( 1973 ) . k.g . wilson and j. kogut , phys . rep . * 12 * , 75 ( 1974 ) . hohenberg and b.i . halperin , rev . phys . * 49 * , 435 ( 1977 ) . s. sachdev in _ proceedings of the 19th iupap international conference on statistical physics _ , edited by b .- l . hao ( world scientific , singapore , 1996 ) . p.a . lee and t.v . ramakrishnan , rev . phys . * 57 * , 287 ( 1985 ) . anderson , phys . rev . * 109 * , 1492 ( 1958 ) . laughlin in _ the quantum hall effect _ , edited by r.e . prange and s.m . girvin ( springer - verlag , new york , 1990 ) . wen , phys . b * 43 * , 11025 ( 1991 ) mehta , _ random matrices _ ( academic press , new york , 1991 ) . anderson , j.r . ensher , m.r . matthews , c.e . wieman , and e.a . cornell , science * 269 * , 198 ( 1995 ) . bradley , c.a . sackett , j.j . tollett , and r.g . hulet , phys . lett . * 75 * , 1687 ( 1995 ) ; c.c . bradley , c.a . sackett , and r.g . hulet , _ ibid . _ * 78 * , 985 ( 1997 ) . davis , m .- o . mewes , m.r . andrews , n.j . van druten , d.s . durfee , d.m . kurn , and w. ketterle , phys lett . * 75 * , 3969 ( 1995 ) . a. einstein , sitz . ( berlin ) , 3 ( 1925 ) . i.f . silvera and j.t.m . walraven , phys . 44 * , 164 ( 1980 ) . a.l . fetter and j.d . walecka , _ quantum theory of many - particle systems _ ( mcgraw - hill , new york , 1971 ) . note that these authors use the more common notation @xmath916 and @xmath917 for the bosonic creation and annihilation operators . l. mandel and e. wolf , _ optical coherence and quantum optics _ ( cambridge , new york , 1995 ) . negele and h. orland , _ quantum many - particle systems _ ( addison - wesley , new york , 1988 ) . amit , _ field theory , the renormalization group , and critical phenomena _ ( world scientific , singapore , 1984 ) . j. zinn - justin , _ quantum field theory and critical phenomena _ ( oxford , new york , 1989 ) . dyson , phys . rev . * 75 * , 1736 ( 1949 ) . stratonovich , sov . phys * 2 * , 416 ( 1958 ) . j. hubbard , phys . rev . lett , * 3 * , 77 ( 1959 ) . d.g . fried , t.c . killian , l. willmann , d. landhuis , s.c . moss , d. kleppner , and t.j . greytak , phys . * 81 * , 3811 ( 1998 ) . bogoliubov , j. phys . ( u.s.s.r . ) * 11 * , 23 ( 1947 ) pitaevskii , sov . jetp * 13 * , 451 ( 1961 ) and e.p . gross , j. math . phys . * 4 * , 195 ( 1963 ) . jin , j.r . ensher , m.r . matthews , c.e . wieman , and e.a . cornell , phys . lett . * 77 * , 420 ( 1996 ) ; d.s . jin , m.r . matthews , j.r . ensher , c.e . wieman , and e.a . cornell , _ ibid . _ * 78 * , 764 ( 1997 ) . m .- o . mewes , m.r . anderson , n.j . van druten , d.m . kurn , d.s . durfee , c.g . townsend , and w. ketterle , phys . lett . * 77 * , 988 ( 1996 ) . hau , b.d . busch , c. liu , z. dutton , m.m . burns , and j.a . golovchenko , phys . a * 58 * , r54 ( 1998 ) . singh and d.s . rokhsar , phys . lett . * 77 * , 1667 ( 1996 ) . m. edwards , p.a . ruprecht , k. burnett , r.j . dodd , and c.w . clark , phys . 77 * , 1671 ( 1996 ) . fetter , ann . phys . * 70 * , 67 ( 1972 ) . m. lewenstein and l. you , phys . lett . * 77 * , 3489 ( 1996 ) . popov , _ functional integrals in quantum field theory and statistical physics _ ( reidel , dordrecht , 1983 ) and references therein . f. dalfovo , s. giorgini , l. pitaevskii , and s. stringari , rev . mod . * 71 * , 463 ( 1999 ) . hutchinson , e. zaremba , and a. griffin , phys . rev * 78 * , 1842 ( 1997 ) . dodd , m. edwards , c.w . clark , and k. burnett , phys . a * 57 * , r32 ( 1998 ) . b.a . lippmann and j. schwinger , phys . rev . * 79 * , 469 ( 1950 ) . m. bijlsma and h.t.c . stoof , phys . a * 54 * , 5085 ( 1996 ) . h. shi and a. griffin , phys . rep . * 304 * , 1 ( 1998 ) . j. bardeen , l.n . cooper , and j.r . schrieffer , phys . rev . * 108 * , 1175 ( 1957 ) . h. kleinert , forts . phys . * 26 * , 565 ( 1978 ) . abraham , w.i . mcalexander , j.m . gerton , r.g . hulet , r. ct , and a. dalgarno , phys . a * 55 * , r3299 ( 1997 ) . stoof , m. houbiers , c.a . sackett , and r.g . hulet , phys . rev . lett . * 76 * , 10 ( 1996 ) ; m. houbiers , r. ferwerda , h.t.c . stoof , w.i . mcalexander , c.a . sackett , and r.g . hulet , phys . rev . a * 56 * , 4864 ( 1997 ) . for an equilibrium argument at zero temperature see l.d . landau and e.m . lifshitz , _ statistical physics _ ( pergamon , london , 1958 ) and p. nozires , in _ bose - einstein condensation _ , edited by a. griffin , d.w . snoke , and s. stringari , p. 15 ( cambridge , new york , 1995 ) . nonzero temperatures are discussed in h.t.c . stoof , phys . a * 49 * , 4704 ( 1995 ) . m. houbiers and h.t.c . stoof , phys . a * 54 * , 5055 ( 1996 ) . t. bergeman , phys . a * 55 * , 3658 ( 1997 ) . see , for instance , s. coleman in _ aspects of symmetry _ ( cambridge , new york , 1985 ) . freire and d.p . arovas , phys . a * 59 * , 1461 ( 1999 ) . s. stringari , phys . lett . * 77 * , 2360 ( 1996 ) . y. castin and r. dum , phys . 77 * , 5315 ( 1996 ) . perez - garcia , h. michinel , j.i . cirac , m. lewenstein , and p. zoller , phys . lett . * 77 * , 5320 ( 1996 ) . dodd , m. edwards , c.j . williams , c.w . clark , m.j . holland , p.a . ruprecht , and k. burnett , phys . rev . a * 54 * , 661 ( 1996 ) . j. javanainen , phys . rev . a * 54 * , 3722 ( 1996 ) . l. you , w. hoston , and m. lewenstein , phys . a * 55 * , 1581 ( 1997 ) . p. hberg , e.l . surkov , i. tittonen , s. stenholm , m. wilkens , and g.v . shlyapnikov , phys . a * 56 * , r3346 ( 1997 ) . for the anisotropic generalization see , for instance , g. baym and c.j . pethick , phys . lett . * 76 * , 6 ( 1996 ) . fetter , phys . a * 53 * , 4245 ( 1996 ) . the exact result is obtained in p.a . ruprecht , m.j . holland , k. burnett , and m. edwards , phys . a * 51 * , 4704 ( 1995 ) . stoof , j. stat . phys . * 87 * , 1353 ( 1997 ) . for a different calculation of the tunneling rate that however neglects the phase fluctuations of the condensate , see e.v . shuryak , phys . a * 54 * , 3151 ( 1996 ) . sackett , c.c . bradley , m. welling , and r.g . hulet , appl . b * 65 * , 433 ( 1997 ) sackett , h.t.c . stoof , and r.g . hulet , phys . lett . * 80 * , 2031 ( 1998 ) . pitaevskii , phys . a * 221 * , 14 ( 1996 ) . sackett , j.m . gerton , m. welling , and r.g . hulet , phys . * 82 * , 876 ( 1999 ) . anderson , phys . rev . * 112 * , 1900 ( 1958 ) . e. abrahams and t. tsuneto , phys . rev . * 152 * , 416 ( 1966 ) . in principle the effective action for a neutral superconductor contains also a topological term , but this does not affect the final result of the following argument because it only leads to a constant shift in the total number of particles . for more details , see m. stone , int . b * 9 * , 1359 ( 1995 ) and references therein . the magnitude of the frequency gap is actually not accurately described by this time - dependent ginzburg - landau theory , because the action @xmath918 $ ] is implicitly only valid for long wavelengths @xmath919 and low frequencies @xmath920 . this may be compared directly with the results of e.m . wright , d.f . walls , and j.c . garrison , phys . * 77 * , 2158 ( 1996 ) and k. mlmer , phys . a * 58 * , 566 ( 1998 ) . v.v . goldman , i.f . silvera , and a.j . leggett , phys . b * 24 * , 2870 ( 1981 ) . for simplicity we here again neglect a topological term in the action , which essentially plays no role in the following . this is also in agreement with our discussion of the superconductor , where exactly the same topological term was not included in the effective action for the global phase @xcite . thomas , proc . * 23 * , 542 ( 1927 ) and e. fermi , mat . * 6 * , 602 ( 1927 ) . see , however , t .- l . ho and v.b . shenoy , phys . lett * 77 * , 2595 ( 1996 ) . ho , phys . rev . lett . * 81 * , 742 ( 1998 ) . m. houbiers and h.t.c . stoof , phys . a * 59 * , 1556 ( 1999 ) . see , for example , k. huang , _ statistical mechanics _ ( wiley , new york , 1987 ) . j. schwinger , j. math . phys . * 2 * , 407 ( 1961 ) . keldysh , zh . fiz . * 47 * , 1515 ( 1964 ) [ sov . jetp * 20 * , 1018 ( 1965 ) ] . p. danielewicz , ann . ( n.y . ) * 152 * , 239 ( 1984 ) . b. de wit , private communication . p. carruthers and k.s . dy , phys . rev . * 147 * , 214 ( 1966 ) . h. kleinert , _ path integrals in quantum mechanics statistics and polymer physics _ ( world scientific , singapore , 1994 ) . this follows from the requirement that we want to be able to fourier transform the heaviside function . caldeira and a.j . leggett , phys . * 46 * , 211 ( 1981 ) and a.o . caldeira and a.j . leggett , ann . ( n.y . ) * 149 * , 374 ( 1983 ) ; _ ibid . _ * 153 * , 445 ( 1984 ) kadanoff and g. baym , _ quantum statistical mechanics : green s function methods in equilibrium and nonequilibrium problems _ ( addison - wesley , new york , 1962 ) . d.c . langreth and j.w . wilkins , phys . b * 6 * , 3189 ( 1972 ) . in agreement with our previous remarks , the notation @xmath921 implies that the time integration can be either from @xmath586 to @xmath922 or from @xmath586 to @xmath30 . n.g . van kampen , _ stochastic processes in physics and chemistry _ ( north - holland , amsterdam , 1981 ) . d. forster , _ hydrodynamic fluctuations , broken symmetry , and correlation functions _ ( benjamin , reading , 1975 ) . to make a connection with a large body of knowledge in quantum optics , we note that equation ( [ stat ] ) explicitly shows that the probability distribution @xmath595 $ ] does not correspond to a @xmath923 or @xmath924 representation of the density matrix , but to a wigner representation instead . the ` diffusion matrix ' in the fokker - planck equation for @xmath595 $ ] is therefore diagonal and positive , as is explained by c.w . gardiner in _ quantum noise _ ( springer , berlin , 1991 ) , chapter 6 . the same nice feature will also appear in the case of an interacting bose gas . kastner , rev . phys . * 64 * , 849 ( 1992 ) and references therein . see also h.t.c . stoof in _ bose - einstein condensation _ , edited by a. griffin , d.w . snoke , and s. stringari , p. 226 ( cambridge , new york , 1995 ) . luiten , m.w . reynolds , and j.t.m . walraven , phys . rev . a * 53 * , 381 ( 1996 ) . n.p . proukakis and k. burnett , j. res . * 101 * , 457 ( 1996 ) ; n.p . proukakis , k. burnett , and h.t.c . stoof , phys . rev . a * 57 * , 1230 ( 1998 ) . gardiner , p. zoller , r.j . ballagh , and m.j . davis , phys . lett . * 79 * , 1793 ( 1997 ) . h. risken , z. phys . * 186 * , 85 ( 1965 ) ; _ ibid . _ * 191 * , 302 ( 1966 ) . miesner , d.m . stamper - kurn , m.r . andrews , d.s . durfee , s. inouye , and w. ketterle , science * 279 * , 1005 ( 1998 ) . for a recent extension of their theory see c.w . gardiner , m.d . lee , r.j . ballagh , and m.j . davis , and p. zoller , phys . lett . * 81 * , 5266 ( 1998 ) . m. hillery , r.f . oconnell , m.o . scully , and e.p . wigner , phys . rep . * 106 * , 121 ( 1984 ) . d.v . semikoz and i.i . tkachev , phys . lett . * 74 * , 3093 ( 1995 ) . stoof , j. low temp . phys . * 114 * , 11 ( 1999 ) . e. zaremba , a. griffin , and t. nikuni , phys . a * 57 * , 4695 ( 1998 ) . see also a. griffin , w .- c . wu , and s. stringari , phys . 78 * , 1838 ( 1997 ) , g.m . kavoulakis , c.j . pethick , and h. smith , phys . a * 57 * , 2938 ( 1998 ) , t. nikuni and a. griffin , j. low . phys . * 111 * , 793 ( 1998 ) , and v. shenoy and t .- l . ho , phys . rev . lett . * 80 * , 3895 ( 1998 ) . w. kohn , phys . rev . * 123 * , 1242 ( 1961 ) . kirkpatrick and j.r . dorfman , j. low temp . * 58 * , 301 ( 1985 ) ; _ ibid . _ * 58 * , 399 ( 1985 ) . e. zaremba , t. nikuni , and a. griffin , j. low temp . phys . * 116 * , 277 ( 1999 ) . k. damle , s.n . majumdar , and s. sachdev , phys . a * 54 * , 5037 ( 1996 ) . bijlsma and h.t.c . stoof , phys . a * 60 * ( 1999 ) . cornell , private communication . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the final merger of two black holes in a binary system releases more power than the combined light from all the stars in the visible universe . this vast energy comes in the form of gravitational waves , which travel across the universe at the speed of light , bearing the waveform signature of the merger . today , ground - based gravitational - wave detectors stand poised to detect the mergers of stellar black - hole binaries , the corpses of massive stars . in addition , planning is underway for a space - based detector that will observe the mergers of massive black holes , awesome behemoths at the centers of galaxies , with masses of @xmath0 , where @xmath1 is the mass of the sun . since template matching forms the basis of most gravitational - wave data analysis , knowledge of the merger waveforms is crucial . calculating these waveforms requires solving the full einstein equations of general relativity on a computer in three spatial dimensions plus time . as you might imagine , this is a formidable task . in fact , numerical relativists have attempted to solve this problem for many years , only to encounter a host of puzzling instabilities causing the computer codes to crash before they could compute any sizable portion of a binary orbit . remarkably , in the past few years a series of dramatic breakthroughs has occurred in numerical relativity ( nr ) , yielding robust and accurate simulations of black - hole mergers for the first time . in this article , we review these breakthroughs and the wealth of new knowledge about black - hole mergers that is emerging , highlighting key applications to astrophysics and gravitational - wave data analysis . we focus on comparable - mass black - hole binaries , with component mass ratios @xmath2 , where @xmath3 and @xmath4 , @xmath5 are the individual black - hole masses . we will frequently also refer to the _ symmetric mass ratio _ @xmath6 for simplicity , we choose to set @xmath7 and @xmath8 ; with this , we can scale the dynamics and waveforms for black - hole binaries with the total system mass @xmath9 . in particular , we can express both length and time scales in terms of the mass , giving @xmath10 . we begin by setting both the scientific and historical contexts . in sec . [ sec : astro_grav ] we provide a brief overview of astrophysical black - hole binaries as sources for gravitational - wave detectors . we next turn to a historical overview in sec . [ sec : history ] , surveying efforts to evolve black - hole mergers on computers , spanning more than four decades and culminating with the recent triumphs . having thus set the stage , we focus on more in - depth discussions of the key components underlying successful black - hole merger simulations , discussing computational methodologies in sec . [ sec : comp ] , including numerical - relativity techniques and black - hole binary initial data . section [ sec : gws ] is the heart of this review . here we discuss the key results from numerical relativity simulations of black - hole mergers , following a historical development and concentrating on the merger dynamics and the resulting gravitational waveforms . these results have opened up a variety of exciting applications in general relativity , gravitational waves , and astrophysics . we discuss synergistic interactions between numerical relativity and analytic approaches to modeling gravitational dynamics and waveforms in sec . [ sec : nrpn ] , and applications of the results to gravitational - wave data analysis in sec . [ sec : gwda ] . the impact of merger simulations on astrophysics is presented in sec . [ sec : astro ] , which includes discussions of recoiling black holes and potential electromagnetic signatures of the final merger . we conclude with a look at the frontiers and future directions of this field in sec . [ sec : frontiers ] . before we begin , we mention several other resources that may interest our readers . the review articles by @xcite and @xcite provide interesting surveys of numerical relativity several years before the breakthroughs in black - hole merger simulations . the article by @xcite is an early review of the recent successes , covering some of the same topics that we discuss here . @xcite reviewed the status of black - hole simulations producing long waveforms ( including at least ten cycles of the dominant gravitational - wave mode ) and their application to gravitational - wave data analysis . finally , the text books by @xcite and @xcite provide many more details on the mathematical and computational aspects of numerical relativity than we can include here , and serve as useful supplements to our discussions . black holes and gravitational waves are surely among the most exotic and amazing predictions in all of physics . these two offspring of einstein s general relativity are brought together in black - hole binaries , expected to be among the strongest emitters of gravitational radiation . we begin by presenting some basic properties of black holes and gravitational waves . for fuller discussions and more details , see @xcite and @xcite . a black hole forms when matter collapses to infinite density , producing a singularity of infinite curvature in the fabric of spacetime . each black hole is surrounded by an event horizon , at which the escape velocity is the speed of light . the event horizon is a global property of the spacetime , since it is defined by the paths of outgoing " photons that are the boundary between photon trajectories that must fall inward , and those that can escape to infinity . the photons defining the event horizon hover at finite radius at the surface of the black hole . since , in principle , mass ( energy ) can fall into the event horizon at late times which will move the location at which photon paths can hover we must know the entire future development of the system to locate the event horizon . when black holes merge , a single event horizon forms whose area is at least as large as the sum of the individual horizons . since numerical relativists want to know when this occurs during the course of a calculation , they rely on a related concept known as an _ apparent horizon _ , whose location depends only on the properties of the spacetime at any given time @xcite . for quiescent black holes , the apparent and event horizons coincide ; for more general holes , the apparent horizon is always inside the event horizon ( with restrictions on the behavior of the matter involved ) . so , in terms of causality in a numerically - generated spacetime , any physical phenomenon found inside an apparent horizon should not leak out and affect the spacetime outside . the simplest black hole is nonrotating and is described by the spherically symmetric schwarzschild solution to the einstein equations of general relativity in vacuum ( i.e. , with no `` matter '' sources in the spacetime ) . a schwarzschild black hole is fully specified by one quantity , its mass @xmath9 . the horizon is located at coordinate @xmath11 ( in schwarzschild coordinates ) ; its area is @xmath12 . more general black holes can have both charge and spin . since a charged black hole in astrophysics will generally be neutralized rapidly by any surrounding plasma , we can consider only rotating , uncharged black holes . stationary ( i.e. , time independent ) black holes are described by the axisymmetric kerr solution . a kerr black hole is fully specified by two quantities , its mass @xmath9 and its angular momentum per unit mass @xmath13 . the event horizon is located at the boyer - lindquist @xcite radius @xmath14 , where @xmath15 the area of the event horizon is @xmath16 . equation requires @xmath17 ; when @xmath18 the black hole is said to be maximally rotating or `` extremal '' . notice that @xmath19 when @xmath20 , and that @xmath14 decreases as @xmath13 increases , thus bringing the location of the horizon deeper into the potential well as the black - hole spin increases . photons and test particles in the vicinity of a single black hole can experience either stable or unstable orbits . for a schwarzschild black hole of mass @xmath9 , the _ innermost stable circular orbit _ ( isco ) occurs at @xmath21 for massive test particles . in the case of a kerr black hole , the isco is closer in for co - rotating test particles , and farther out for counter - rotating particles . while the concept of an isco is strictly defined only for massive test particles , it has proven useful for studies of the spacetime around two black holes spiralling together on quasicircular orbits . imagine that you put the two black holes on an _ instantaneously _ circular orbit around each other ; at that moment they have neither nonzero radial velocity nor nonzero radial acceleration . at any given separation , the black holes have some angular momentum . the isco is the separation where that angular momentum is a minimum , in analogy to the test particle definition . black holes at closer separations would be expected to fall inward , toward the center , even without radiating angular momentum via gravitational radiation . gravitational waves are ripples in the curvature of spacetime itself . they carry energy and momentum and travel at the speed of light , bearing the message of disturbances in the gravitational field . as with electromagnetic waves , gravitational waves can be decomposed into multipolar contributions that reflect the nature of the source that generates them . recall that electromagnetic radiation has no monopole contribution due to the conservation of total charge . by analogy , conservation of total mass - energy guarantees that there can be no monopole gravitational radiation . since dipolar variations of charge and currents are possible , electromagnetic waves can have a dipole character . however , conservation of linear and angular momenta removes any possibility of dipolar gravitational waves , so the leading - order contribution to gravitational radiation is quadrupolar . gravitational waves are thus generated by systems with time - varying mass quadrupole moments @xcite . in the wave zone , a gravitational wave is described as a perturbation @xmath22 to a smooth underlying spacetime . the wave amplitude is @xmath23 where @xmath24 is the quadrupole moment of the source , @xmath25 is the distance from the source , and @xmath26 is the mass in the source that is undergoing quadrupolar changes . this shows that the strongest gravitational waves will be produced by large masses moving at high velocities , such as binaries of compact stars and black holes . the wave on the left is purely in the @xmath27 polarization state , and the one on the right is purely in the @xmath28 polarization state . the gravitational waves produce tidal forces in planes transverse to the propagation direction . from @xcite . reprinted with permission from aaas.,width=336 ] a gravitational wave is purely transverse , acting tidally in directions perpendicular to its propagation direction . when a gravitational wave impinges on a detector of length scale @xmath29 , it produces a length change in that detector @xmath30 . by substituting in typical values for compact objects in the universe into eq . , one can see that astrophysical sources typically yield wave amplitudes of @xmath31 at the earth . consequently , precision measurements are needed to make detections . gravitational waves have two polarization components , known as @xmath32 and @xmath33 for a linearly polarized wave . figure [ fig : forcelines ] shows the corresponding lines of force for sinusoidal gravitational waves propagating along the @xmath34 axis . in the purely @xmath27 polarization on the left , the wave stretches along one axis and squeezes along the other , alternating sinusoidally as the wave passes . the @xmath28 polarization wave on the right acts similarly , stretching and squeezing along axes rotated by @xmath35 . in general , a gravitational wave is a superposition of these two states , conveniently written as a complex waveform strain @xmath22 , where @xmath36 astronomers have found evidence for black holes throughout the universe on a remarkable range of scales . the smallest of these , stellar black holes , have masses in the range @xmath37 and form as the end - products of massive star evolution . there is good observational evidence for the existence of stellar black holes , based on dynamical measurements of the masses of compact objects in transient systems that undergo x - ray outbursts . since neutron stars can not have masses @xmath38 , compact objects more massive than this must be black holes @xcite . intermediate - mass black holes ( imbhs ) have masses in the range @xmath39 . imbhs may form as the result of multiple mergers of smaller objects in the centers of dense stellar clusters in the present universe @xcite , assuming mass loss from stellar winds is not significant @xcite . they may also arise from the evolution of very massive stars early in the history of the universe , forming black - hole `` seeds '' in the centers of massive halos ( the precursors of the galaxies we see today ) early in the history of the universe , to redshifts @xmath40 @xcite . currently the best observational evidence for imbhs comes from models of ultraluminous x - ray sources @xcite . finally , massive black holes ( mbhs ) have masses in the range @xmath41 and are found at the centers of galaxies , including our own milky way galaxy . the observational case for the existence of mbhs is quite strong , based on dynamical models of stars and gas believed to be moving in the potential well of the central mbh @xcite . black - hole binaries are binary systems in which each component is a black hole . as mentioned above , we focus here on comparable - mass binaries , which are expected to produce the strongest gravitational - wave signals . stellar black - hole binaries may form as the result of binaries composed of two massive stars ; see @xcite and references therein . stellar black - hole binaries may also arise from dynamical processes in which a black hole is captured into an orbit around another black hole in dense stellar environments @xcite . imbh binaries can also form through dynamical processes in stellar clusters @xcite and from mergers of massive halos at high redshifts . since both stellar and imbh binaries are `` dark '' that is , they are generally not surrounded by gas which might produce electromagnetic radiation we have few observational constraints on these types of black - hole binaries . this situation will change dramatically , however , with the detection of gravitational radiation from these systems @xcite , as gravitational waves bring direct information about the dynamical behavior of the orbiting masses and do not rely on electromagnetic emissions from nearby matter . since essentially all galaxies are believed to contain an mbh at the center and to undergo a merger with another galaxy at least once during the history of the universe , mbh binaries can arise when their host galaxies merge @xcite ; see also @xcite and references therein . however , due to the vast cosmic distances involved , and the small angular separations on the sky expected for mbh binaries , only a few candidates are currently known through electromagnetic observations @xcite . when they form , mbh binaries typically have relatively wide separations , and the gravitational radiation they emit is very weak and insufficient to cause the binary to coalesce within the age of the universe . however , various processes such as gaseous dissipation and @xmath42-body interactions with stars can remove orbital energy from the binary and cause the black holes to spiral together @xcite ; see also @xcite and @xcite and references therein . eventually , the black holes reach separations at which gravitational radiation reaction becomes the dominant energy - loss mechanism , leading to the final coalescence of the black holes and the emission of strong gravitational waves @xcite . mergers of comparable - mass black - hole binaries are expected to be among the strongest sources of gravitational waves . this final death spiral of a black - hole binary encompasses three stages : inspiral , merger , and ringdown @xcite . in the early stages of the inspiral , the orbits of most astrophysical black - hole binaries will circularize due to the emission of gravitational radiation @xcite . during the inspiral , the orbital time scale is much shorter than the time scale on which the orbital parameters change ; consequently , the black holes spiral together on quasicircular orbits . since the black holes have wide separations , they can be treated as point particles . the inspiral dynamics and waveforms can be calculated using post - newtonian ( pn ) equations , which result from a systematic expansion of the full einstein equations in powers of @xmath43 , where @xmath44 is the binary separation @xcite . the inspiral phase produces gravitational waves in the characteristic form of a _ chirp _ , which is a sinusoid with both frequency and amplitude increasing with time . as the black holes spiral inward , they eventually reach the strong - field , dynamical regime of general relativity . in this merger stage , the orbital evolution is no longer quasi - adiabatic ; rather , the black holes plunge together and coalesce into a single , highly distorted remnant black hole , surrounded by a common horizon . since the point - particle and pn approximations break down , numerical relativity simulations of the einstein equations in three dimensions are needed to calculate the merger . due to the difficulty of these simulations , the resulting gravitational waveforms were completely unknown until recently . finally , the highly distorted remnant black hole settles down into a quiescent rotating kerr black hole by shedding its nonaxisymmetric modes through gravitational - wave emission . we call this process the `` ringdown , '' in analogy to how a bell that has been struck sheds its distortions as sound waves . various analytic techniques of black - hole perturbation theory @xcite form the basis of ringdown calculations , producing gravitational waveforms in the shape of exponentially damped sinusoids @xcite . the characteristic gravitational - wave frequency of a quasicircular black - hole binary , produced by the dominant ( highest - order ) quadrupole component , is @xmath45 where @xmath46 is the orbital frequency . astrophysical black - hole binaries produce gravitational waves that span three frequency regimes , depending on the black - hole masses @xcite . stellar black - hole binaries and the lower mass end of the imbh binaries radiate in the high frequency band , @xmath47 , which is already being observed by ground - based laser interferometer detectors such as ligo @xcite , and will be observed by the advanced detectors by @xmath48 @xcite . low frequency gravitational waves cover the band @xmath49 and will be observed by the space - based laser interferometer lisa , currently under development @xcite . mbh binaries with masses @xmath50 will be very strong sources for lisa , with the lower mass systems visible out to redshifts @xmath40 @xcite ; the inspirals of imbh binaries will also be detectable @xcite . finally , the very low frequency band @xmath51 will be observed by pulsar timing arrays @xcite . this band is expected to dominated by gravitational waves from a very large population of unresolved mbh binaries @xcite with possibly a few discrete sources @xcite . the quest to calculate the gravitational - wave signals from the merger of two black holes spans more than four decades and encompasses key developments in theoretical and experimental general relativity , astrophysics , and computational science . in this section , we begin by delineating these threads in general terms , and then turn to a more detailed account of select milestones along the path toward successful simulations of black - hole mergers . at the end of the 18th century @xcite and @xcite first speculated , using newtonian gravity , that a star could become so compact that the escape velocity from its surface would exceed the speed of light . in the 20th century , scientists realized that such black holes could form as the final state of total gravitational collapse in general relativity @xcite ; john wheeler would later popularize the term `` black hole '' to describe such an object @xcite . beginning in the 1960s , many highly energetic astrophysical phenomena were discovered with physical properties pointing to extremely strong gravitational fields as underlying mechanisms ; among these are quasars and x - ray binaries such as cygnus x-1 @xcite , the first credible black - hole candidate . as discussed in sec . [ sec : astro_grav : astro_bhs ] , today astrophysical black holes are believed to exist on a vast range of scales throughout the universe , and black - hole binaries are considered to be strong sources of gravitational waves . einstein s equations of general relativity form a coupled set of nonlinear partial differential equations , in which dynamic curved spacetime takes the role of newton s gravitational field and interacts nonlinearly with massive bodies . gravitational waves were first recognized as solutions to the linearized , weak - field einstein equations early in the past century . by mid - century , gravitational waves were recognized as real physical phenomena , carrying energy and being capable of producing a response when impinging on a detector . this development spawned a major branch of experimental general relativity , with concepts for the first gravitational - wave detectors appearing in the 1960s ; see @xcite for a review . the discovery of two neutron stars in a binary system by @xcite provided an astrophysical laboratory for the first _ indirect _ detection of gravitational radiation . decades of observation revealed the binary orbit to be shrinking by precisely the amount expected if the system were emitting gravitational waves according to general relativity @xcite ; hulse and taylor were awarded the nobel prize in 1993 . today , the prognosis for _ direct _ detection of gravitational waves is excellent , with the first events expected from the advanced ground - based interferometers around the middle of this coming decade . as you might expect , einstein s equations pose formidable obstacles to anyone who would dare to probe the physics within . throughout most of the 20th century , relativists uncovered a fairly small number of exact solutions by exploiting symmetries , and made progress toward more general problems using various perturbative expansions . by the 1960s , computers had become powerful enough to encourage attempts at solving einstein s equations numerically , to uncover physics beyond the realm of perturbation theory . the subsequent development of numerical relativity has been made possible in part by continued increases in computer power and advances in algorithms and computational methods . most numerical - relativity simulations start with the idea of decomposing four - dimensional spacetime into a stack of curved three - dimensional spacelike slices threaded by a congruence of timelike curves @xcite . arnowit , deser , and misner ( adm ) pioneered this `` 3 + 1 '' approach as the basis for a canonical formulation of the dynamics of general relativity @xcite . in this hamiltonian formulation , the three - metric @xmath52 on the spatial slices takes the role of the `` configuration variables . '' quantities based on the extrinsic curvature @xmath53 , which is roughly the time derivative of @xmath52 , play the role of `` conjugate momenta . '' variation of an action with respect to @xmath52 produces a set of six first - order evolution equations for the conjugate momenta ; varying the momenta gives six first - order evolution equations for @xmath52 . adm also introduced four lagrange multipliers as freely specifiable gauge or coordinate conditions , representing the four coordinate degrees of freedom in general relativity . variation of these lagrange multipliers yields four equations that must hold on each slice : a hamiltonian constraint , and three momentum constraints . originally intended as a tool for quantizing gravity , the adm formalism later became the basis for most work in numerical relativity . as we discuss in sec . [ sec : comp ] below , key elements in this approach are solving the cauchy problem , beginning with the initial data on a spacelike slice , and then evolving that data forward in time . the constraint equations form the basis for this initial value problem . appropriate choices for the gauge conditions are crucial ingredients for today s successful black - hole merger simulations . @xcite made the first known attempt to simulate the head - on collision of two equal - mass black holes on a computer in 1964 , using a two - dimensional axisymmetric approach . their simulation ran for 50 timesteps to a duration of @xmath54 ; at this point , they decided that the simulation was no longer accurate enough to warrant continued evolution , and stopped the code . smarr and eppley @xcite returned to this problem in the mid-1970s , again employing two - dimensional axisymmetry but now using the adm formalism , specialized coordinates , and improved coordinate conditions . although they encountered problems with instabilities and large numerical errors , they managed to evolve the collision and extract information about the emitted gravitational waves . smarr and eppley had used the most powerful computers of their day . going to the next step , orbiting black holes in three dimensions , was deemed to be not feasible at the time , due to unresolved questions about the instabilities and insufficient computer power . the black - hole merger problem lay dormant for over a decade . in the 1990s , attention returned to black - hole mergers as the ligo project began to move forward , and black - hole mergers were recognized as the strongest sources for this detector . since the signal - to - noise ratio for such ground - based detectors is fairly modest , having a template for the merger waveform is a key part of the data analysis strategy . numerical relativists revisited the problem of colliding two black holes head - on with modern techniques and more powerful computers @xcite . in the mid-1990s , the national science foundation funded a computational grand challenge grant for a large multi - institution collaboration aimed at evolving black - hole mergers in three dimensions and calculating the resulting gravitational waveforms . around the same time , a large and very active numerical relativity group arose at the newly - formed albert - einstein institut ( aei ) in potsdam , germany . during the late 1990s and into the early 2000s , two developments on the experimental side further increased the desire for black - hole merger simulations : the ground - based gravitational - wave detectors started taking data , and interest grew in lisa and its potential for observing gravitational waves from massive black - hole binary mergers . while no one expected the task at hand developing computer codes to solve the full einstein equations in three dimensions for the final few orbits and merger of two black holes to be simple , numerical relativists found that the problem was far more difficult than anticipated . producing a waveform useful for gravitational - wave detection purposes typically would require running a simulation for a duration of several hundred @xmath9 . however , a variety of instabilities plagued the codes , causing them to crash well before any significant portion of an orbit could be achieved . nevertheless , during a period encompassing a little over a decade , much important work was accomplished that laid the foundations for later success . key milestones include these developments : * initial data for binary black holes near the isco @xcite ; * new methods for representing black holes on computational grids such as punctures @xcite and excision @xcite ; * recognition of the importance of hyperbolicity in formulating the einstein equations for numerical solution @xcite ; * improved formulations of the einstein equations @xcite ; * fully three - dimensional evolution codes and their use in evolving distorted black holes @xcite , boosted black holes @xcite , head - on collisions @xcite , and grazing collisions @xcite ; * coordinate conditions that keep the slices from crashing into singularities and the spatial coordinates from falling into the black holes as the evolution proceeds @xcite ; * the cactus computational toolkit , which provided a framework for developing numerical - relativity codes and analysis tools used by many groups ; * modern adaptive mesh refinement finite - difference ( including carpet with cactus , bam @xcite , and paramesh @xcite ) , and multi - domain spectral @xcite infrastructures for numerical relativity . throughout this period , the length of black - hole evolutions gradually increased . improvements in the formalisms allowed simulations of single black holes and , later , two black holes to increase in duration to @xmath55 . the addition of new slicing and shift conditions again increased the evolution times to @xmath56 . the lazarus project took a novel approach to combine these relatively short - duration binary simulations with perturbation techniques for the late - time behavior to produce a hybrid model for a black - hole merger @xcite . by late 2003 , @xcite carried out the first complete orbit of two equal - mass , nonspinning black holes . while this simulation lasted @xmath57 , the code crashed shortly after the orbit was completed and the gravitational waves were not extracted . overall , progress was slow , difficult , and incremental . however , the situation was about to change dramatically . in early 2005 , frans pretorius electrified the relativity community when he achieved the first evolution of an equal - mass black - hole binary through its final orbit , merger , and ringdown using techniques very different from those employed by the rest of the community @xcite . later in 2005 the groups at the university of texas at brownsville ( utb ) and nasa s goddard space flight center independently and simultaneously discovered a new method , called `` moving punctures , '' that also produced successful black - hole mergers @xcite . their presentations were given back - to - back at a workshop on numerical relativity , to the amazement of each other and the assembled participants . since the moving - puncture approach was based on underlying techniques used by several other groups , it was rapidly and readily adopted by most in the community , producing a growing avalanche of results . 2006 was a year of many firsts : the first simulations of unequal - mass black holes and the accompanying recoil of the remnant hole @xcite , the first mergers of spinning black holes @xcite , the first long waveforms ( @xmath58 orbits ) @xcite , the first comparisons with pn results @xcite , and the first systematic parameter study in numerical relativity @xcite . the year 2007 opened with the discovery of `` superkicks '' recoil velocities exceeding @xmath59 @xcite , and in 2008 a black - hole binary merger with mass ratio @xmath60 was accomplished @xcite , while @xcite carried out the first long - term evolution of generic spinning and precessing black - hole binaries . as this article was being written in 2009 the state of the art continues to advance , with the first simulations of black - hole mergers using spectral numerical techniques @xcite , and the first steps towards modeling the flows of gas around the merging black holes @xcite . applications of the merger results in areas such as comparisons with pn expressions for the waveforms , astrophysical computations of black - hole merger rates , and the development of templates for gravitational - wave data analysis have accompanied these technical developments in the simulations . the study of black - hole mergers using numerical relativity is thriving indeed . in this section , we discuss the mathematical and numerical foundations underlying current black - hole merger simulations , highlighting the key issues involved in achieving successful evolutions . for more detailed treatments , we direct the interested reader to @xcite or @xcite . the central task of numerical relativity is solving einstein s field equations @xmath61 where einstein s tensor @xmath62 represents the curvature of spacetime , the energy - momentum tensor @xmath63 contains the matter sources , and @xmath64 . by convention , an index @xmath65 selects a `` time '' component , and @xmath66 selects a `` space '' component . @xmath62 depends on the first and second derivatives of the metric tensor @xmath67 . for vacuum black - hole spacetimes , @xmath68 . note that all of the tensor fields discussed here are symmetric in the indices , e.g. @xmath69 . the metric characterizes the geometry of spacetime by giving the infinitesimal spacetime interval @xmath70 through the following definition : @xmath71 where we use the einstein summation convention which implies summation over all values of a given index that appears twice in an expression . many physical implications of the metric are immediately apparent from eq . ( [ eq : interval ] ) . for example , when @xmath72 , the resulting metric - determined relationship between the time and space coordinates yields the paths that light rays must follow in this spacetime . the dependence of einstein s tensor on the metric can be simply illustrated in coordinates known as harmonic " ; in vacuum eq . takes the form : @xmath73 where @xmath74 is the flat - space wave - operator and @xmath75 represents all terms nonlinear in the metric . if @xmath75 is interpreted as an effective source term , this is a simple wave equation . the familiar form of this equation suggests that its cauchy problem can be solved by specifying the metric and its first derivative on an initial , spatial surface and then integrating in time , as for an ordinary wave equation . einstein s equations admit various formulations and coordinate conditions , which should be tailored to the problem at hand in this case , numerical simulation of black - hole spacetimes . regardless of these choices , current numerical practices universally involve an initial three - dimensional slice of spacetime that is evolved forward in time . here , we review the history and current most common choices of initial data , black - hole representations , formulation , coordinate conditions , and some details of the numerics . the impetus , and most important outcome , of all these developments is the ability to generate gravitational waveforms from black - hole binary sources over many cycles . the cauchy problem concerns solution of the field equations given initial data specified on an initial ( typically spatial ) hypersurface . in practice , the cauchy problem is more conveniently investigated in a `` 3 + 1 '' formulation , explicitly based on a foliation of the spacetime into three - dimensional spatial slices parametrized by a time coordinate . a common 3 + 1 formulation inspired by adm @xcite divides up the components of the metric according to their relationships with space and time , such that the line element takes the form : @xmath76 where @xmath77 is called the lapse function , @xmath78 the shift vector , and @xmath79 is the spatial three - metric . we write the time coordinate @xmath80 and the spatial coordinate indices @xmath81 . note that contraction with @xmath67 or its inverse @xmath82 is used to lower or raise indices of four - dimensional tensors , respectively , while @xmath52 or its inverse @xmath83 is used to lower or raise indices of three - dimensional tensors . the lapse and shift represent coordinate freedom in the metric ; we can choose these quantities arbitrarily . however , since the three - metric @xmath52 ( and its first and second spatial derivatives ) determines the intrinsic curvature of the slice , it carries the information about the gravitational field and thus is constrained by the physics . the meaning of the lapse and shift can be understood by considering two successive spatial slices separated by an infinitesimal time interval @xmath84 ( fig . [ fig : smarr_3plus1 ] ) . an observer along a vector normal to the first slice will measure an elapsed proper time of @xmath85 in evolving to the second slice , and a change in spatial coordinate of @xmath86 . since einstein s equations are second order in time , we must also specify the initial time derivative of the three - metric . rather than specifying this derivative directly , we define a new quantity : @xmath87 where @xmath88 is an ordinary partial derivative and @xmath89 is the spatial covariant derivative . note that the spacetime covariant derivative @xmath90 is a partial derivative with a correction such that it transforms as a vector and satisfies @xmath91 . @xmath89 is the projection of @xmath90 onto the spatial slice and is equivalent to a three - dimensional covariant derivative formed from @xmath52 . for the case of euclidean normal coordinates , @xmath92 and @xmath93 , eq . ( [ eq : kij ] ) reduces to the simple expression @xmath94 . if we define a unit vector @xmath95 normal to the spatial slice , we can show that eq . ( [ eq : kij ] ) is equivalent to @xmath96 . as suggested by this expression , @xmath53 is a measure of the change of the normal vector as it is transported along the slice . in this way @xmath53 gives an extrinsic measure of the curvature of a three - dimensional spatial slice with respect to its embedding in four - dimensional spacetime . it is therefore called the extrinsic curvature . depending on the formulation , the extrinsic curvature might or might not come into the evolution equations ; however , @xmath53 is almost universally utilized when calculating initial data . and @xmath97 are depicted . @xmath77 is the lapse , and @xmath98 represents the proper time lapse between slices . @xmath99 is the shift , and @xmath100 represents the amount by which the spatial coordinates shift between slices . @xmath101 is normal to the slice at @xmath102 . if a ray parallel to @xmath101 intersects the @xmath102 slice at a point @xmath103 , then it will intersect the @xmath97 slice at @xmath104 . @xmath105 is a coordinate time vector . if a ray parallel to @xmath106 intersects the @xmath102 slice at point @xmath103 , then it will also intersect the @xmath97 slice at @xmath103 . , width=336 ] of the ten component equations of eq . , six determine the time - evolution of the metric , while four must be satisfied on a spatial slice at any given time , and are thus constraint equations . with an appropriate choice of time coordinate , and assuming vacuum spacetime , these four constraint equations are equivalent to the condition @xmath107 . the time - time component @xmath108 is called the _ hamiltonian constraint _ , and the time - space components @xmath109 are called the _ momentum constraint_. these take the form of conditions on the extrinsic curvature : @xmath110 where @xmath111 is the three - dimensional ricci scalar associated with the three - metric @xmath52 , and @xmath112 , sometimes called the _ mean curvature_. these constraint equations must be solved in order to obtain an initial spatial slice consistent with einstein s equations . how does one represent an exotic object such as a black hole in a numerical simulation ? in particular , how can one use finite computational methods repeatedly to model an object which , analytically , contains physical and/or coordinate singularities ? fortunately , two successful strategies have emerged to meet this challenge . the unusual topology of black holes offers one way out . as @xcite originally showed , a black hole can be considered a bridge " or wormhole " connecting one universe or worldsheet " , to a second worldsheet ( see fig . [ fig : wormhole_flatspace ] ) . exploiting this topology , a continuous spatial slice which avoids the physical singularity contained within the event horizon of each black hole can be constructed as follows . starting with a spatial slice of schwarzschild spacetime , remove the interior of the event horizon . identify the resulting spherical boundary with the spherical boundary of an identical copy of this space . the two - dimensional analog would be to take two sheets of paper , cut out a disk from each , and then glue the resulting circular edges together . each sheet , or copy of schwarzschild spacetime in this example , is called a worldsheet . the identified spherical boundaries connecting the worldsheets form what is referred to as the throat " of the wormhole . to complete this construction , we require an appropriate coordinate system to continuously cover both worldsheets . brill and lindquist discovered coordinates that will prove convenient , in which the three - metric of a particular spatial slice of schwarzschild spacetime is given by @xcite @xmath113 where @xmath25 is a radial coordinate . in these coordinates , the event horizon is at @xmath114 . we can consider each of the worldsheets described above as being separately labeled by such coordinates . designate one worldsheet as @xmath115 and the other as @xmath116 , and call their radial coordinates @xmath117 and @xmath118 respectively . since the interior of the event horizon has been removed from each worldsheet , assume that @xmath119 and @xmath120 . the metric on each worldsheet has the form of eq . ( [ eq : bls ] ) . as noted by brill and lindquist , this form of the metric is unchanged by the transformation @xmath121 , and @xmath122 when @xmath114 . so if we define a new coordinate @xmath123 by @xmath124 mapping spatial infinity on worldsheet @xmath116 to @xmath125 , we obtain a single continuous coordinate system that applies to worldsheet @xmath115 for @xmath126 and worldsheet @xmath116 for @xmath127 . this avoidance of the physical singularity comes at the expense of a coordinate singularity in the metric at @xmath125 , called a `` puncture '' . however , it turns out this coordinate singularity can be confined to a single scalar variable , as suggested in eq . ( [ eq : bls ] ) . with a suitable change of variables and other means , numerical simulations haven proven capable of handling this irregular scalar field . thus the puncture " method is one way to represent a black hole that is amenable to computation . another strategy is excision , first proposed by unruh @xcite . given that no physical information can escape an event horizon to influence the exterior , the interior of a black hole can in principle be excised from the computational domain . this relies on the fact that all physical information propagates inwards from the event horizon towards the physical singularity , i.e. lightcones tilt inwards , and nothing physical propagates outward from the horizon . extrapolation can be used for the boundary condition of the excised region , and any nonphysical numerical error that escapes the horizon should be negligible . this approach is not constrained to the particular coordinates required by the puncture method , but it can be more difficult due to the need for precise positioning of the excision boundary and accurate extrapolation . excision was used successfully for orbiting binary simulations by @xcite and continues to be used by the caltech - cornell dual - coordinate spectral code @xcite . either of these methods can be useful in representing black holes on the initial data slice . surprisingly , we will also see that either of these representations can be made robust enough to persist as the black holes evolve . the starting point for successful simulations of black - hole mergers is finding initial data for astrophysically realistic inspiralling black holes . if we were simulating the orbits of stars in newtonian gravity , this would be a simple procedure . for example , we could simply specify the masses and spins of the stars , along with their positions and velocities on orbits derived from the dynamics of point particles , and then evolve the system numerically to allow it to `` relax '' into orbits appropriate for bodies of finite size . in general relativity , however , the initial data must satisfy the constraint equations ( [ eq : ham ] ) - ( [ eq : mom ] ) , which are cast in terms of the @xmath128-metric @xmath52 and the extrinsic curvature @xmath53 . since there is no obvious , or natural , connection between these field variables and the astrophysical properties of inspiralling black holes , obtaining suitable initial data is a major challenge . building on the earlier work of @xcite , york developed a general procedure for solving the constraint equations to produce initial data for the cauchy problem in the 1970s ( see @xcite for a review ) . this approach generally requires solving a coupled elliptic system of four nonlinear field equations . we can break this problem down into more manageable pieces with some simplifying assumptions . while these choices do come at the cost of some loss of generality and astrophysical realism in the initial data for two black holes , it has been seen that for sufficiently long evolutions the final orbits and waveform signatures from the black - hole evolution are largely insensitive to this level of detail in the initial data , at least in the case of equal - mass nonspinning black holes . the first simplification is to choose traceless extrinsic curvature , @xmath129 . with this , the hamiltonian constraint is decoupled from the momentum constraints , and can be solved separately . to find solutions corresponding to multiple black holes , we generally further assume that the initial slice is conformally flat . that is , the three - metric is the product of a scalar conformal factor with a flat metric , @xmath130 with this , the problem reduces to solving a ( typically nonlinear ) equation for the scalar field @xmath131 . @xcite found a simple solution , representing @xmath42 black holes momentarily at rest , which gives @xmath132 where @xmath133 is the mass associated with the @xmath134th black hole and @xmath135 represents its coordinate center . each @xmath135 corresponds to the location of a puncture , as described in the last section . this is a valuable solution , but not generally useful , because these black holes lack momentum and spin . an explicit solution by bowen for the momentum constraint was the last crucial step in defining a procedure for calculating initial data for multiple black holes with specified linear and angular momenta @xcite . the bowen - york prescription employed a two - sheeted topology found by misner for the black - hole interiors @xcite . later brandt and brgmann generalized the procedure for the brill - lindquist topology @xcite . the brandt - brgmann puncture data for arbitrary momenta and spin is now widely used because of its ease of implementation . more recently york developed another modeling ansatz , known as the conformal - thin - sandwich ( cts ) approach @xcite , which has certain additional advantages . for example , it turns out the conventional brandt - brgmann puncture can not yield a spin parameter @xmath136 greater than @xmath137 @xcite , while cts data can go higher @xcite . as discussed above , the spatial metric is chosen to be conformally flat , then instead of providing an ansatz for the extrinsic curvature @xmath53 , the initial time - derivative of the conformal metric is specified ( generally to vanish ) . in addition , a condition can be imposed on the slicing ( see below ) . the result is a coupled system of elliptic equations which is solved to enforce the constraints and optionally the choice of slicing condition . boundary conditions may additionally be supplied to enforce rotational motion . either approach provides an ansatz for constructing three - dimensional binary black - hole initial _ field _ data for a specified choice of particle - like parameters , masses , positions , momenta , and spins . inevitably there are differences from the nearly quiescent evolving systems that we seek to represent . generally there is some level of spurious radiation generated from a period of initial transient dynamics through which the system relaxes to become quiescent on sub - orbital time scales . in particular , extraneous radiation content is an unavoidable consequence of conformal flatness , which post - newtonian analysis has shown must deviate from the physically relevant inspiralling binary solution @xcite . this spurious radiation is seen in plots of the gravitational waveforms produced by mergers ; see fig . [ fig : universal_wf ] and other waveform plots in sec . [ sec : gws : universal ] . often the simulations will also undergo a period of initial gauge - evolution which , though physically inconsequential , may affect the quality of the simulation numerically . for puncture initial data @xcite analyzed some of these gauge dynamics , and developed a promising approach ( trumpet " data ) to mitigate it ( see sec . [ sec : movepunc ] for related gauge issues ) . however , for simulations lasting for several orbits these modest transient effects are generally negligible . most black - hole binary simulation studies are designed to represent the astrophysical population of systems , which have circularized before the gravitational radiation becomes observationally significant . these simulations begin with circularly inspiralling initial data configurations . even before reliable numerical relativity simulations were possible , considerable attention had been given to prescribing initial data for these near - circular configurations ( see @xcite for a review ) . within the cts approach , it is particularly natural to impose initially circular motion upon the system , resulting in an initial data prescription for quasicircular orbits @xcite . for either the cts or brandt - brgmann data , quasicircular parameters may be chosen by constraints on either an effective gravitational potential or total energy of the system @xcite . for more realistic inspiralling trajectories , the pn approximation may be used @xcite , or , for higher accuracy , an iterative procedure involving short numerical evolutions @xcite . the einstein evolution equations , which determine the time - development of the initial data , form a set of at least six coupled , nonlinear propagation equations . the exact formulation of these equations depends on the choice of evolved variables as well as how constraint identities are incorporated . although many formulations are possible , not all are equivalent from a numerical point of view . the choice of formulation can affect how numerical errors grow in time , and whether these errors can converge to zero as the resolution of the computational grid is increased . an ideal formulation is well - posed , meaning that : ( i ) for a given set of initial data , a unique solution exists , and ( ii ) solutions depend continuously on perturbations of the initial data . to determine whether a given formulation has the desirable property of well - posedness , we first consider it as a matrix equation : @xmath138 where @xmath139 is a vector , the components of which represent the evolved variables such as components of the metric ; the linear operator @xmath140 is the principal part , containing the highest - order spatial derivatives ; and @xmath141 contains additional terms which may be nonlinear . certain conditions on the eigenvalues and eigenvectors of the operator @xmath140 in eq . are used to characterize the equations , typically as either `` strongly hyperbolic '' or `` weakly hyperbolic '' @xcite . if the eigenvalues are real and the eigenvectors form a complete set ( so any solution can be written as a linear combination of them ) , the equations are strongly hyperbolic . then , assuming adequate boundary conditions , the equations are well - posed and all solutions are bounded by a function that grows ( at most ) exponentially at a rate independent of the initial data . strong hyperbolicity is a key ingredient for successful black - hole merger simulations . on the other hand , if the eigenvalues are real but the eigenvectors are not complete , the equations are weakly hyperbolic . in this case the equations are not well - posed , and they permit solutions which grow at rates that depend on the initial data . weak hyperbolicity implies that small numerical errors in the initial data may grow at a rate which depends on the resolution . it then becomes difficult to show that the numerical solution converges to a well - defined continuum solution , and at high resolutions the simulation may become unstable @xcite . the quest for workable formulations of einstein s equations has proceeded along two parallel lines of development . one originated with consideration of harmonic coordinates " , so called because the coordinates satisfy the wave equation @xmath142 , where @xmath143 is the curved - space wave operator . in these coordinates , einstein s equations can be written such that the principal part resembles a wave equation in terms of the metric as in eq . . in this form , einstein s equations are manifestly hyperbolic @xcite . however , the harmonic coordinate condition is too restrictive for numerical purposes , so generalized harmonic coordinates were eventually developed by introducing a source term into the coordinate condition , i.e. @xmath144 @xcite , a suitable choice for which preserves strong hyperbolicity . the subsequent introduction of _ constraint - damping _ terms , which tend to drive the constraints towards zero , further ensured stability @xcite . this formulation is manifestly second - order in both time and space , and has been implemented numerically as such @xcite , but for more efficient numerical integration a first - order - in - time formulation was also developed @xcite , and is currently being used by some groups . a second line of development originated with the invention of the adm formulation of einstein s equations @xcite . this formulation was refined by @xcite , who suggested evolving the specific quantity of the extrinsic curvature ( eq . ) . the associated evolution equations specify the first - order time - derivatives for the three - metric @xmath52 and the extrinsic curvature @xmath53 , as well as for the gauge variables @xmath77 and @xmath78 . however , the adm formulation is weakly hyperbolic ( see , e.g. , chapter 5 of @xcite ) , and attempts at stable numerical evolutions with it were not successful . there are essentially two ways to modify the adm equations that may affect their hyperbolicity ( while keeping them first - order in time and maintaining their 3 + 1 character ) . one way is to add the constraints , which may change the principal part of the equations without changing the associated physics . the other is to define new independent variables . simple addition of constraints failed to give a strongly hyperbolic formulation however . eventually it was found that strongly hyperbolic versions could be constructed from the adm equations by the promotion of certain derived quantities to the status of independently evolved variables @xcite . although strong hyperbolicity is important for a stable formulation , it is not the only property critical for accurate evolution . of the various strongly hyperbolic 3 + 1 formulations , one eventually emerged as more successful than the rest . its development began with the observation that the numerical accuracy of @xmath145 , the determinant of the three - metric , and @xmath146 , the trace of the extrinsic curvature , could best be preserved by evolving these two quantities independently @xcite . subsequently , to remove the redundancy in evolving the full three - metric and extrinsic curvature , the conformal three - metric @xmath147 and conformal traceless extrinsic curvature @xmath148 were substituted @xcite . to eliminate certain terms with second derivatives ( which contribute to the principal part of this system ) , a new independently evolved variable @xmath149 was also introduced @xcite . further improvements were made shortly afterwards , such as the addition of constraints to eliminate more second - derivative terms @xcite . the resulting formulation , now commonly known as bssn , proved to be robustly stable and accurate for numerical purposes . note that the above choices which comprise its form were empirically motivated , by identifying and eliminating terms which tended to compromise numerical accuracy . some analytic justification for its success was later found by @xcite , who showed that the bssn formulation avoids exponentially growing modes and most zero - speed modes , which accumulate numerical error . finally @xcite and @xcite proved that bssn is also strongly hyperbolic , given an appropriate choice of gauge . a few noteworthy refinements of the bssn formulation followed . @xcite and @xcite suggested adding specific terms proportional to the gamma constraint " ( @xmath150 minus its definition ) and the hamiltonian constraint into the evolution equations . these terms have the effect of damping the constraints and thereby improving stability . later , for evolution of punctures specifically , @xcite evolved the conformal factor in the form @xmath151 , which vanishes at the puncture ( an improvement in accuracy over the previous standard of @xmath152 , which is singular at the puncture ) . this change of variables , however , introduces the potentially singular factor @xmath153 into the evolution equations , necessitating an arbitrary restriction on the minimum value of @xmath154 . an alternative variable @xmath155 was subsequently suggested because resulting appearances of @xmath156 mostly cancel with factors of @xmath157 @xcite . the choice of gauge or coordinate conditions , like the choice of formulation , has important consequences on the numerics , especially the stability of the simulation . important considerations include how to deal with the extreme conditions of black holes such as the physical singularities , the possible coordinate singularities , the strong - field gradients , and the dynamical , surrounding spacetime . the coordinates must accommodate these features in a way that is numerically tractable . an early consideration was how to avoid the physical singularity of a black hole through an appropriate slicing condition . one such choice , _ maximal slicing _ , required solution of a numerically expensive elliptic equation @xcite . later , an interest in constructing a hyperbolic evolution system led to a generalization of harmonic slicing known as the bona - mass family @xcite . one particular member of this family , called 1+log slicing , was found to also be singularity - avoiding , like maximal slicing , but at less numerical cost @xcite . meanwhile , it was recognized that a shift vector was required to counter the large field gradients , or `` slice - stretching '' , incurred in the presence of a black hole @xcite . a hyperbolic shift condition called the `` gamma driver '' was introduced that fulfilled this requirement @xcite : @xmath158 where @xmath159 depends on a conformal three - metric @xmath160 of the evolving spatial slice , @xmath161 is an auxiliary variable , @xmath78 is the shift , and @xmath162 is some scalar field . @xmath163 is a damping parameter that fine - tunes the growth of the shift , which affects the coordinate size of the black - hole horizons , which in turn has bearing on the required numerical resolution @xcite . this shift condition also has the desirable tendency to drive the coordinates to quiescence in synchrony with the physics , for example after a binary merges into a stationary black hole . clearly if @xmath150 vanishes , @xmath161 is damped to zero and @xmath78 approaches a ( typically small ) constant . initially these conditions were intended for use with punctures that remained at fixed coordinate positions ( i.e. comoving coordinates ) , by choosing @xmath162 to vanish at the punctures . however , this led to large gradients between the merging black holes since there the metric had to vanish as the physical distance contracted . pathologies also arose from the twisting of the coordinates as the black holes orbited each other . these issues , which resulted in large numerical errors and instabilities , were eventually resolved with the breakthrough discovery that slight modifications of the 1+log and gamma - driver conditions allowed arbitrary motion of the punctures with robust stability @xcite . in particular , in the shift condition , a critical alteration was unpinning " the puncture by no longer requiring the factor @xmath162 to vanish at the puncture but rather to remain constant . these modifications were subsequently refined to eliminate zero - speed modes , thus preventing the possibility of error build - up @xcite . this was accomplished in part by the addition of advection terms : @xmath164 it is noteworthy that these particular gauge choices , coupled with the bssn equations , also result in strong hyperbolicity @xcite . a further refinement resulted from the observation that , assuming that @xmath165 initially , eqs . and can be integrated to yield @xmath166 ; this substitution allows the removal of the equation for @xmath161 , leaving only a single shift equation @xcite : @xmath167 use of this or similar gauge conditions became known as the moving puncture " method , and proved to be very successful as it became increasingly widespread among the numerical - relativity community . analysis has verified that moving punctures are valid black - hole solutions , although the initial character of the punctures changes significantly during the course of evolution . the second worldsheet shown in fig . [ fig : wormhole_flatspace ] , for example , invariably becomes disconnected in numerical simulations , at a point within the horizon near the throat , due to the action of the shift vector effectively shifting computational grid points onto the first worldsheet . meanwhile the spatial coordinates evolve such that the @xmath168 singularities in the conformal factor @xmath131 become @xmath169 singularities . for a single nonspinning black hole , the numerical result rapidly asymptotes to an exact form of the schwarzschild solution called a trumpet " , recently investigated by @xcite and @xcite , which turns out to be a type of solution first considered by @xcite . development of generalized harmonic coordinates initially proceeded independently of the above 3 + 1-formulated conditions . as mentioned , in harmonic coordinates the dalembertian of each coordinate vanishes . in generalized harmonic coordinates , the wave equation for each coordinate is allowed a source term , i.e. @xmath170 these gauge driving " source terms @xmath171 can be either algebraically specified or evolved such that hyperbolicity is preserved @xcite . the first successful numerical orbit of black holes involved a source term for the time coordinate that effectively kept the lapse close to its minkowski value of unity , while the spatial coordinates remained harmonic @xcite . this was accomplished by evolving the source term itself , according to @xmath172\alpha^{-1}\ ] ] where @xmath173 and @xmath174 are constants . more recently , to dampen extraneous gauge dynamics during the inspiral and merger of generic binaries , @xcite found the following gauge driver to be successful : @xmath175 ^ 3 \\ h_i & = & -\mu_0\left[\log\left(\frac{\sqrt{g}}{\alpha}\right)\right]^2\frac{\beta_i}{\alpha}\end{aligned}\ ] ] where @xmath176 is a specified function of time that starts at zero and eventually increases monotonically to unity . additional noteworthy advances in coordinate conditions were developed for the generalized harmonic formulation to facilitate spectral methods ( see sec . [ sec : comp : computation ] below ) but are in principle generalizable to other frameworks . one of these is the use of multiple coordinate patches , where the coordinates are chosen according to the local physical geometry for optimal accuracy . in the generalized harmonic formulation , characteristic speeds are readily available for use in constructing physical boundary conditions between patches @xcite . the other advance is that of dual coordinates " . because moving punctures involve irregular fields , they are not easily made compatible with spectral methods , which are sensitive to irregularities . an alternative , that of a moving , excised region , can leave a large trail of interpolation error in its wake . the remaining option is to use comoving coordinates @xcite , but these can result in instabilities due to the steep field gradients and other pathologies mentioned in sec . [ sec : comp : gauge ] . dual coordinates were invented to exploit the advantages of both comoving and non - comoving coordinates while avoiding their disadvantages . this is accomplished by computing all tensor field components in the non - comoving coordinate system ( thus avoiding steep gradients ) , yet evolving them as functions of the comoving coordinates ( thus allowing stationary excision boundaries ) @xcite . the initial data , formulation , and gauge are all , in principle , analytic choices applicable to an infinite , continuous manifold . one must finally make choices pertaining explicitly to the finite , discrete mechanics of the computations that will numerically approximate the above analytic specifications . an immediate consideration is the fact that the simulated domain must have finite extent . various conformal compactification schemes which map spatial or null infinity to a finite boundary have been tested @xcite , but currently it is more common to impose artificial boundaries at finite spatial coordinates and apply some form of either radiative boundary condition @xcite or constraint - preserving boundary condition @xcite . to mitigate the effect of inward - propagating errors from the outer boundaries , some form of mesh refinement is typically employed to push the outer boundaries as far away as possible from regions of interest ( e.g. wave sources and extraction regions ) . with a limited number of grid points available , their density is judiciously chosen to be highest near the strong field gradients of the black holes , moderate throughout regions where wave propagation is studied , and coarser beyond that . if the simulated black holes are very dynamic then some algorithm for automatically adapting the mesh refinement is necessary . meanwhile the interfaces between refinement regions require interpolation . on this grid , spatial derivatives are computed in one of two ways . they may be computed with finite differencing stencils across uniformly spaced grid points , derived from taylor expansions , currently up to eighth - order accurate @xcite . alternatively , the spectral approach may be used , in which coefficients of an expansion in basis functions are computed to some order , on a number of collocation points comparable to the number of basis modes , from which the derivatives are then obtained analytically @xcite . in the former case , dissipative terms are often added to the evolution equations to reduce noise @xcite . in the latter case , spectral methods often include a smoothing step for a similar end . last , time must be advanced in discrete steps . various explicit time - integration algorithms have been tested , for example the iterated crank - nicholson scheme , but the fourth - order runge - kutta algorithm has become most widely used , due to its superior accuracy and efficiency . in some codes the timestep size is made to vary with spatial resolution , for even greater efficiency , albeit at the expense of the complication of time - interpolation . implicit timestepping schemes are also being investigated as a means to greatly increase the timestep size @xcite . one of the most important end - results of a simulation of merging black holes is a computation of the emitted gravitational radiation . for this purpose , it is useful to calculate the weyl tensor " , @xmath177 . the weyl tensor , like the einstein tensor , is constructed from derivatives of the metric . unlike the einstein tensor , the weyl tensor has degrees of freedom that do not necessarily depend on a massive source ; it can be nonzero while the einstein tensor vanishes . certain components of the weyl tensor form a complex quantity called @xmath178 , one of five weyl scalars " used to classify spacetimes @xcite . in a special , transverse - traceless " , spherical coordinate system , @xmath178 can be expressed as follows : @xmath179 where the subscripts @xmath180 denote orthonormal tetrad components . in a spacetime with gravitational radiation this quantity typically falls off as @xmath181 @xcite , and can be associated with outgoing radiation at spatial infinity ( @xmath182 ) in asymptotically flat spacetimes @xcite . in terms of the strain introduced in the last section , this can be written @xmath183 , \ ] ] where , in terms of a metric perturbation @xmath184 , @xmath185 what makes @xmath178 a particularly useful measure of the radiation is that to first order in @xmath186 it is coordinate - invariant . although the linearized radiation interpretation for @xmath178 is only strictly valid in the limit as @xmath182 , in numerical simulations it is often extracted on a sphere of large but finite radius . this will approximate the expected radiative behavior if @xmath187 ; if spheres of multiple radii are used then it is also possible to extrapolate the results to infinity . typically @xmath178 is computed on the computational grid points from the metric variables and then interpolated onto the spherical extraction surface . recently , cauchy - characteristic extraction methods have been applied to allow direct evaluation of radiation at future null infinity @xcite . the product of @xmath178 with a spherical harmonic function is then integrated over the sphere , as it is useful to extract specific modes of the field . the structure of the gravitational field is such that it is convenient to expand in spin - weight -2 spherical harmonics @xmath188 @xcite ( just as , analogously , it proves convenient to use the spin - weighted -1 harmonics in expanding the electromagnetic field ) . thus : @xmath189 unlike electrodynamics , there is no dipole moment of the gravitational field in conventional general relativity , so the dominant contribution is the @xmath190 quadrupole moment . other information that is important in analyzing spacetimes relates to the properties of the black holes themselves . this is commonly taken from apparent horizons , surfaces from which all light rays must go inward . it is useful to compute this during a simulation , for example by finding surfaces on which the expansion of light rays are minimized @xcite . when a black hole is sufficiently isolated , it is possible to define a mass and spin magnitude from data evaluated on the horizon @xcite , as well as linear momentum @xcite . these quantities can be used to characterize the physical properties of the black holes , and to label the spacetimes when comparing with pn theory . the final merger of a black - hole binary takes place in the dynamical , strong - field regime of general relativity . for many years , numerical relativists wondered what they would find when they probed this arena using black - hole merger simulations . what would the merger waveforms look like ? how strongly might artifacts in the initial data affect the mergers ? how might the effects of unequal masses and spins change the picture obtained by studying the simplest case of nonspinning , equal - mass mergers ? would they uncover any unexpected phenomena ? recent breakthroughs in black - hole merger simulations are providing important tools for addressing such questions . in this section we explore the dynamics of black - hole mergers and the resulting gravitational waveforms . we start with a look at the lazarus approach , which provides hybrid waveforms by combining analytic methods with brief numerical simulations . we then begin a more comprehensive discussion of black - hole merger physics , starting with mergers of nonspinning , equal - mass black holes . next we consider mergers of unequal - mass holes , followed by mergers with spin . throughout this discussion we aim to provide a historical context while highlighting the key physical results . in the late 1990s , numerical relativity had advanced sufficiently to allow brief simulations of two black holes in three dimensions . by `` brief '' we mean here durations of @xmath191 , which is a small fraction of the estimated binary orbital period of @xmath192 near the isco . while most of the community focused on technical developments aimed at extending the duration of these simulations , a small band of collaborators took a different approach @xcite . their novel idea was to use full numerical - relativity simulations to calculate the strong - field approach to merger during @xmath191 of simulation time , and then to calculate the remaining evolution using perturbative techniques @xcite . they began with traditional puncture black holes ( which remain fixed in the grid ) on quasicircular orbits near the isco , and evolved them towards merger . then , just before the simulation became unstable , they stopped the calculation and extracted the physical data for the merging black holes and the emerging gravitational waves . this data was then interpreted as initial data for a highly distorted single black hole , and evolved using techniques of black - hole perturbation theory . they called this method of reviving a ( nearly ) dead calculation the _ lazarus _ approach . the hybrid waveforms produced by the lazarus collaboration gave the first indications of what might be expected from the final merger of black - hole binaries @xcite . they ran a suite of simulations , varying the initial black - hole separations and the time at which they stopped the calculation and made the transition from the fully numerical evolution to the perturbed black hole evolution . figure [ fig : laz_wf ] shows the dominant quadrupole @xmath193 component of re(@xmath178 ) , which corresponds to the @xmath27 polarization , for the case of equal - mass , nonspinning black holes . note the simple shape of the waveform smoothly tying together what might be the end of an ( inspiral ) chirp with a damped ( ringdown ) sinusoid . by tuning the mass and spin of the background hole for the perturbative evolution , @xcite also determined that the merger remnant was a kerr ( spinning ) hole of mass @xmath194 , and spin @xmath195 . the lazarus method was also applied to mergers of black holes with spins either aligned or anti - aligned with the orbital angular momentum @xcite ; in all cases , a similar waveform shape was seen . how generic was this simple shape and , in particular , would it also arise in situations where the black holes complete one or more orbits before merging ? part of @xmath196 , corresponding to the @xmath27 polarization ( measured on the system s axis ) , is shown for ten simulations : five different initial black - hole separations ( designated qc-0 , etc . ) , each with two transition times to perturbative evolution . ] early in 2005 , frans pretorius stunned the community by achieving the first robust simulation of two equal - mass black holes through a single plunge orbit , merger , and ringdown @xcite . the resulting gravitational waveforms are shown in fig . [ fig : frans_wf ] , where @xmath25re(@xmath178 ) is plotted versus @xmath197 . here @xmath25 is the coordinate distance from the center of the grid to the sphere on which the waveform is extracted ( see sec . [ sec : extract ] ) , and @xmath198 is the mass of a single black hole . ( note that the time axes for all other waveform plots in this paper are scaled by either the total system mass @xmath9 or the mass of the final , remnant black hole @xmath199 ) . , due to lower resolution in the outer regions ] here curves are plotted for waveforms extracted on four spheres with successively larger radii . the curves have been shifted in time so that the waveforms overlap . note that the overall waveform shape is simple . the merger yields a single black hole that is spinning , with spin parameter @xmath200 , where @xmath199 is the mass of the final black hole that forms . to carry out these mergers , pretorius @xcite used techniques that are very different from the more traditional approach based on bssn and punctures , used by nearly all other numerical relativists . as discussed in sec . [ sec : harm ] , he used a generalized harmonic formulation in which the einstein equations are written with second - order time and space derivatives . the spatial domain was compactified , so that the outer boundary of each slice was mapped to spatial infinity . moreover , he excised the black holes ( each formed from the collapse of scalar field `` blobs '' ) and evolved their motion across the grid using adaptive mesh refinement ( see secs . [ sec : bhongrid ] and [ sec : comp : computation ] ) . pretorius had clearly developed a robust method for evolving black - hole binary mergers . in the wake of his remarkable success , many numerical relativists began studying his methods . however , another surprise was soon to emerge . in the autumn of 2005 two research groups , one at utb and the other at goddard , independently developed a powerful technique for evolving mergers of puncture black holes within the traditional bssn approach @xcite . as discussed in sec . [ sec : movepunc ] , the standard puncture method requires the black holes to remain fixed on the numerical grid . in this new `` moving puncture '' approach , simple but novel gauge conditions allow the punctures to move across the grid , producing accurate and stable merger evolutions @xcite . although the utb and goddard codes were both based on the bssn approach , they were independently written and featured somewhat different implementations of moving punctures . in addition , the goddard code used second - order finite differencing ( standard in the community at the time ) . they employed a box - in - box fixed mesh refinement to produce the high resolution in the region around the black holes needed to compute the dynamics accurately , while maintaining adequate resolution in more distant regions and a large enough computational domain to allow accurate extraction of the gravitational waves . the utb code used fourth - order finite differencing , and the `` fish - eye '' coordinate transformation to produce higher resolution around the black - hole orbits and lower resolution in the wave extraction regions . both groups successfully evolved an equal - mass , nonspinning binary through the final plunge orbit , merger , and ringdown , and extracted the gravitational waveforms . in these simulations , the black holes completed @xmath201 orbit before merger , defined as the time at which a single apparent horizon was formed . figure [ fig : utb_tracks ] shows the tracks traced by the black - hole punctures , with the apparent horizons superimposed , for the utb run @xcite . these simulations produced plunge waveforms with a simple shape similar to that found by pretorius ; compare fig . [ fig : frans_wf ] from pretorius s simulations with fig . [ fig : gsfc_wf ] , which shows waveforms from the goddard simulation plus a corresponding lazarus waveform . and @xmath202 . the solid circles denote the centroids of the apparent horizons every @xmath203 during the evolution . the first common horizon , marking the time of merger , forms at @xmath202 , just before the punctures complete @xmath204 orbit . ] , @xmath205 mode of @xmath206 was extracted from the numerical simulation on spheres of radii @xmath207 , and @xmath208 for the medium- and high - resolution runs . the waveforms extracted at different radii have been rescaled by @xmath25 , and shifted in time to account for the wave propagation time between the two extraction spheres . curves are shown for medium ( @xmath209 in the finest grid ) and high ( @xmath210 ) resolutions . a waveform from a lazarus calculation starting with the same initial data is shown for comparison @xcite . the lazarus waveform shown in this figure is scaled differently from those in fig . [ fig : laz_wf ] . ] in most cases , astrophysical black - hole binaries spiral together over many orbits , radiating away any orbital eccentricity in the form of gravitational waves . by the time the black holes reach the final inspiral , their orbits will be quasicircular . all mergers of such equal - mass nonspinning binaries should thus produce the same waveform signature , subject only to rescaling with the total system mass @xmath9 . over the years , concerns had been raised within the relativity community about the effects of deviations from astrophysical initial conditions on the waveforms . for example , spurious eccentricity in the orbits could arise from starting conditions that did not approximate quasicircular inspiral accurately . moreover , the commonly - used conformal flatness prescription ( see sec . [ sec : comp : initialdata ] ) for the initial data is different from the conditions experienced by an astrophysical binary ; these differences would result in spurious gravitational radiation being present in the initial conditions for the binary simulations . how would factors such as these influence the merger waveforms ? having developed robust techniques for evolving black - hole mergers , numerical relativists eagerly pursued longer simulations with the black holes starting at wider separations , completing more orbits , and producing longer wavetrains . the utb group ran a simulation in which the black holes completed nearly 1.5 orbits before the formation of a single apparent horizon @xcite . they observed that the resulting waveform was quite similar to those produced by their earlier simulation that started near the plunge @xcite , with a brief oscillatory signal at the beginning . @xcite anticipated `` that the plunge waveform , when starting from quasicircular orbits , has a generic shape that is essentially independent of the initial separation of the binary . '' the goddard group , simultaneously pursuing this same goal , produced the first `` universal '' waveform for the merger of equal - mass , nonspinning black holes @xcite . using approximately quasicircular initial conditions , they ran a series of four simulations , starting the black holes at increasingly wider separations . in all , the binaries completed @xmath211 , 2.5 , 3.6 , and 4.2 orbits before forming a common apparent horizon ; the systems emitted just under 4% of their energy as gravitational radiation , and the final black holes were spinning with @xmath212 . to compare the results of these models , they chose the fiducial time @xmath102 to be the moment of peak amplitude in the gravitational radiation ; this typically occurs within a few @xmath9 of merger . it is worth noting how close these results are to those obtained from the lazarus approach to the same system ( see sec . [ sec : gws : laz ] ) . orbit before merger ( denoted by * ) . ] the binary orbital dynamics of these runs is clearly seen in fig . [ fig : tracks ] , which shows the tracks of the black - hole punctures . here only one puncture track is shown for each binary , with the trajectories oriented to superpose at the fiducial @xmath102 . the tracks show the effects of eccentricity in the early stages of each run ; plots of the separation @xmath25 versus time show that the initial eccentricity decreases as the initial separation increases . as the holes spiral together into the strong - field regime , the eccentricity diminishes . the puncture tracks lock on to a single universal trajectory , independent of the initial conditions , through the final orbit , plunge , and merger . figure [ fig : universal_wf ] shows the corresponding universal gravitational waveform for equal - mass , nonspinning black holes . specifically , it shows the real part of the @xmath190 , @xmath205 mode of @xmath206 versus time ; for this case , the quadrupole @xmath213 modes strongly dominate all other modes . here the signals produced by each run have been shifted in time so that @xmath102 marks the moment of peak radiation amplitude . starting from @xmath214 , the waveforms for the final burst of radiation show nearly perfect agreement , with differences at the level of @xmath215 . for the preceding few orbits , the waveforms agree in amplitude and phase to @xmath216 , except for the brief initial bursts of spurious radiation ( see sec . [ sec : comp : initialdata ] ) . overall , the merger stage lasts @xmath57 , converting @xmath217 of the initial total mass @xmath9 into gravitational - wave energy . the gravitational wave released during this burst has a luminosity @xmath218 , which is greater than the total luminosity of all the stars in the visible universe . for stellar black - hole binary mergers , this luminosity will last for a few milliseconds , while for mbh binaries it will last for several minutes . , @xmath205 mode ) produced by the four simulations whose puncture tracks are shown in fig . [ fig : tracks ] @xcite . the signals have been shifted in time so that the peak radiation amplitude occurs at @xmath219 . the agreement among the waveforms is excellent , with differences @xmath215 for the final burst of radiation starting at @xmath220 . ] note that this universal waveform has a simple shape . the signal starts with a short inspiral chirp , increases smoothly in amplitude during the plunge and merger , and then transitions to the damped ringdown sinusoid . overall , the frequency increases monotonically , reaching a maximum value that stays constant during the ringdown @xcite . since black - hole - merger waveforms will be used to help find signals in data from gravitational - wave detectors , it was important to verify that various groups consistently achieve the same results . figure [ fig : compare_wf ] provides a comparison of waveforms computed by pretorius and the groups at utb and goddard for equal - mass , nonspinning black - hole mergers @xcite ; here the real part of @xmath206 s dominant @xmath221 , @xmath205 mode is shown , with all three waveforms aligned in time so that the moment of peak gravitational radiation amplitude occurs at @xmath222 . all three simulations evolved through the last @xmath223 orbits before merger ; the goddard run is r1 from @xcite and shown in figs . [ fig : tracks ] and [ fig : universal_wf ] . note that all evolutions produce the same overall waveform shape . the goddard and utb groups evolved black holes with zero spin starting on similar quasicircular initial orbits , and produced nearly identical waveforms . pretorius used co - rotating initial conditions which impart a small spin @xmath224 to each black hole . this difference is consistent with the slightly higher frequencies seen in pretorius results during the ringdown , @xmath225 . . the differences between pretorius waveform and the others during the ringdown , @xmath226 , are consistent with his simulation producing a slightly faster rotating final black hole due to these initial spins . reproduced by permission of the institute of physics . ] how robust is this universal waveform to larger amounts of orbital eccentricity and gravitational radiation in the initial data ? @xcite studied a series of equal - mass , nonspinning binaries starting from roughly the same initial orbital period and having varying amounts of initial orbital eccentricity @xmath227 . for @xmath228 , the initial eccentricity is radiated away , and the binaries circularize and begin a universal plunge at @xmath229 before the time of peak radiation amplitude , producing a final black hole with @xmath230 . for @xmath231 , the black holes do not complete any orbits , but rather plunge together and merge ; the final black hole spin reaches a maximum value @xmath232 around @xmath233 . figure [ fig : psu - ecc - tracks ] shows the puncture tracks , and fig . [ fig : psu - ecc - wfs ] shows the real part of the strain s @xmath221 , @xmath205 mode , for two nonspinning , equal - mass black - hole mergers with nonzero eccentricity from this study ; in both cases , @xcite show quantitatively that the waveform for the final burst of radiation is the same as the universal waveform . in a complementary study , @xcite found that the final spin of the remnant black hole is essentially insensitive to the eccentricity for binaries that do not plunge immediately . @xcite examined the effects of spurious gravitational waves by superposing a tunable packet of gravitational radiation on an equal - mass , nonspinning binary . they found the binary evolution and the spin of the remnant black hole to be robust to modest amounts of added gravitational radiation . ( left ) and @xmath234 ( right ) from @xcite . in these cases , the initial eccentricity is radiated away and the binaries circularize before merger . ] strain mode ( labeled @xmath32 here ) for black - hole mergers shown in fig . [ fig : psu - ecc - tracks ] from @xcite . these runs have initial orbital eccentricity @xmath235 ( top ) and @xmath234 ( bottom ) and enter a universal plunge by @xmath229 before the gravitational radiation reaches its peak value . ] the black - hole simulations discussed so far cover the last @xmath236 four orbits before merger . these achievements marked the triumph of solving a long - standing problem of fundamental importance to general relativity : the two - body problem for the final merger of equal - mass schwarzschild black holes driven by gravitational radiation reaction . these simulations provided the first look at the dynamics and waveforms of black holes evolving and merging in the nonlinear , strong - field regime . they also pointed the way towards applications in detecting the final gravitational - wave burst from black - hole mergers @xcite . however , for more advanced applications to gravitational - wave data analysis and comparisons with pn analytical treatments , longer waveforms that start many cycles before merger are essential . the goddard group @xcite produced the first such long waveforms for equal - mass , nonspinning black holes starting @xmath58 orbits or @xmath237 gravitational - wave cycles before merger . they used improved initial conditions with low eccentricity @xmath238 , and focused on improving accuracy while the black holes traversed the relatively long inspiral . they investigated the observability of black - hole mergers with ground- and space - based gravitational - wave detectors @xcite ( see sec.[sec : gwda : direct ] ) and successfully applied their long waveforms to comparisons with pn results , focusing on the waveform phases @xcite . and applies when the black holes are far enough apart that the black hole speeds remain well below the speed of light . we refer to pn results by the order at which the series is truncated . for example , `` 2 pn '' means that terms of order @xmath239 are retained . see sec . [ sec : nrpn ] for a deeper discussion of the pn approximation in the context of numerical relativity . ] shortly thereafter , the jena group simulated a binary inspiralling for nine orbits ( @xmath240 gravitational - wave cycles ) before merger @xcite . with this , they made the first quantitative comparisons with both the pn phase _ and _ amplitude , and quantified the level of error in the quadrupole approximation . they used higher - order finite differencing @xcite and initial binary parameters calculated using the pn approximation to reduce the initial eccentricity significantly @xcite , enabling a precise measurement of the waveform phase . the caltech - cornell group currently holds the record for the longest and most accurate black - hole binary evolution , starting 16 orbits and 32 gravitational - wave cycles before merger @xcite . using their spectral code ( see sec . [ sec : comp : computation ] ) , they begin with a very small initial orbital eccentricity @xmath241 and evolve with very high accuracy through a relatively long inspiral , then merger and ringdown . the impressive trajectories of their black holes are shown in fig . [ fig : cc_traj ] and the accompanying gravitational waveforms in fig . [ fig : cc_wf ] . showing the @xmath190 , @xmath205 component of re(@xmath206 ) the left panel shows a zoom of the inspiral waveform , and the right panel a zoom of the merger and ringdown . ] comparison of waveforms from different groups remains important in the push for higher accuracy and use in gravitational - wave data analysis . the samurai project @xcite sets the current state of the art for studying the consistency of black - hole binary waveforms . this effort focuses on comparing waveforms from equal - mass , nonspinning binaries , starting with at least six orbits ( or 12 gravitational - wave cycles ) before merger and continuing through the ringdown . they focus on comparing the amplitude @xmath242 and phase @xmath243 for the @xmath190 , @xmath205 mode of @xmath206 , defined as @xmath244 the gravitational - wave frequency of this mode is then @xmath245 . they compare the results from five independent numerical codes : the moving - puncture codes from the aei , goddard , jena , and penn state groups , and the caltech - cornell spectral code . figure [ fig : samurai - phase ] compares the gravitational - wave amplitudes and fig . [ fig : samurai - amps ] the gravitational - wave phases as a function of frequency for the five waveforms . qualitatively , the results appear to be quite consistent . quantitatively , they concluded that these waveforms have sufficient accuracy to be used for detection with all current and planned ground - based detectors . as a function of the dimensionless frequency @xmath246 for the five codes from the samurai comparison @xcite . the scale along the top of the panel labels the frequency in khz scaled with respect to the total binary mass in solar units . ] , @xmath205 component of the gravitational waves produced by the five codes in the samurai comparison @xcite are shown as a function of frequency . ] early in 2006 , numerical relativists took the next step in opening up the parameter space of binary black - hole mergers by simulating nonspinning binaries with unequal masses . this added a new parameter , the mass ratio @xmath247 , to the problem and brought an additional need for adaptive mesh refinement to achieve adequate resolution around the smaller black hole . as shown in fig . [ fig : unequal_tracks ] , the smaller black hole also moves faster , completing an orbit around the center of mass in the same time as the larger hole and thus requiring smaller timesteps for its evolution . these factors combine to make simulations of unequal - mass binaries more technically challenging ; currently , numerical relativists are able to simulate mass ratios up to @xmath60 @xcite . decomposition into spin - weighted spherical harmonic modes ( see sec . [ sec : extract ] ) provides the basis for an in - depth study of black - hole mergers . for the equal - mass case @xmath248 , the @xmath221 , @xmath249 quadrupole mode is dominant and the odd-@xmath250 modes are suppressed by symmetry . as @xmath247 increases , the sub - dominant modes become more important , affecting both the evolution of the source and the emitted radiation . @xcite carried out a set of unequal - mass mergers with mass ratios in the range @xmath251 . they found that , to leading order , the total energy emitted during merger scales @xmath252 while the spin of the final black hole scales @xmath253 , where @xmath254 is the symmetric mass ratio . they also studied the multipolar structure of the gravitational waves . figure [ fig : berti - amps ] shows several @xmath255 modes of the radiation produced in their simulations for the case @xmath256 . they demonstrated that the higher modes carry larger fractions of the total energy as @xmath247 increases ; in particular , the @xmath257 mode generally carries @xmath216 of the emitted energy for @xmath258 . the goddard group @xcite performed a complementary study of nonspinning unequal - mass mergers for mass ratios @xmath259 . they found that the overall simple waveform shape first discovered for equal - mass mergers extends to the cases with @xmath260 ; this is easily seen when the gravitational waves are scaled by @xmath254 . figure [ fig : wfs_eta_scaled ] shows the strain @xmath261 , including the @xmath221 and @xmath262 modes , for an observer located at distance @xmath25 on the system s orbital axis . the waveforms are aligned so that the maximum amplitude of the @xmath190 , @xmath205 mode occurs at @xmath102 . the differences in phase during the final ringdown portion of the waveforms are consistent with the differing spins of the final black holes ; for example , with nonspinning black holes , a larger mass ratio @xmath247 results in a smaller final spin for the remnant black hole . when viewed off - axis , the waveforms show modest amplitude and phase modulations , while still preserving the overall simplicity in shape . for different mass ratios , from @xcite . @xmath32 is calculated for different mass ratios using the @xmath221 and @xmath262 modes . the observer is located at distance @xmath25 along the axis of the system . ] @xcite also found that each of the individual spherical harmonic components is circularly polarized during the inspiral , merger , and ringdown , as seen by distant observers on the system s rotational axis . during the inspiral , the phase and frequency of the different @xmath255 components are the same for each mass ratio ; this is expected since the waveform phase is directly related to the orbital phase . more interestingly , for the @xmath263 modes , this relationship continues to hold during merger and ringdown . based on these observations , they developed a simple conceptual picture in which each @xmath255 mode of the gravitational radiation is produced separately by the @xmath255 mode of some _ implicit rotating source_. the fixed relationship between the phase and frequency of the @xmath263 modes shows that these components of the implicit source rotate rigidly during the entire coalescence from inspiral through ringdown . in comparison , the @xmath264 source components peel away from the main trend of the @xmath263 parts during merger , indicating less rigid rotation . @xcite have carried out the highest mass ratio merger simulation to date , with @xmath60 . this binary radiates @xmath265 of its mass as gravitational radiation as it undergoes @xmath266 orbits before merging to form a black hole with @xmath267 ; these values fit the scaling relations found by @xcite and @xcite . figure [ fig:10to1_wfs1 ] , from @xcite shows that the @xmath268 and @xmath269 modes of re(@xmath270 ) exhibit the simple waveform shape individually ; this trend continues through @xmath271 , the highest mode they studied . unequal - mass black - hole binaries can form as a result of galaxy mergers , dynamical processes in star clusters , and the final stages of binary stellar evolution . mergers of unequal - mass binaries bring a qualitatively new phenomenon of great importance to astrophysical scenarios of black - hole growth and retention : recoils or kicks . as discussed in sec . [ sec : astro_grav : basic ] , the gravitational waves emitted by a merging black - hole binary carry away linear momentum . if the pattern of gravitational - wave emission is not symmetrical ( _ i.e. , _ if there is more radiation emitted in some direction than in others ) , then global conservation of momentum requires the center of mass , and thus the remnant black hole that forms , to recoil in the opposite direction . the situation for an unequal - mass , nonspinning binary is shown schematically in fig . [ fig : schematic_kick ] , from @xcite , who attributed this argument to alan wiseman ; we summarize their discussion here . the two holes are orbiting in a plane about their center of mass . since the smaller black hole moves faster , its wave emission undergoes more `` forward beaming '' than that of the larger hole . instantaneously this gives a net flux of momentum parallel to the smaller hole s velocity , and an opposing recoil or kick at the center of mass . the direction of this kick changes continually as the black holes orbit . if the orbit was circular , the center of mass would move in a circle and suffer no net recoil . however , since the black holes are spiralling together ( due to the energy and angular momentum carried away by the gravitational waves ) , the instantaneous kicks do not cancel exactly but rather accumulate . this causes the final merged black hole to have a non - zero net recoil in the orbital plane . is emitted parallel to the smaller hole s velocity . momentum conservation requires that the system center of mass recoil in the opposite direction , @xmath272 . reproduced with kind permission from springer science+business media : fig . 1 of @xcite . ] this recoil has been studied by several authors , including @xcite . more recent analytic treatments using pn approximations were carried out by @xcite , while some comparable - mass estimates were also made with the lazarus approach @xcite . however , since the dominant part of the kick depends sensitively on the strong - field regime close to merger , where the orbits are less circular , an accurate calculation of the kick requires full numerical relativity simulations of the final stages of binary inspiral , merger and ringdown @xcite . the goddard group @xcite carried out the first accurate calculation of black - hole recoil for the merger of a nonspinning binary with @xmath273 . figure [ fig : gsfc - kick.eps ] shows the kick velocity as a function of time for three simulations in which the black holes start out on orbits with increasingly larger initial separations @xmath274 . the kick velocity grows rapidly during merger , reaching a peak value @xmath191 after the time of peak radiation amplitude in @xmath178 , and then dropping to a lower , final value . it is clear from this figure that the black holes must start far enough apart in order to get a consistent and reliable value for the final recoil velocity . for this @xmath273 case , their results spanned the range @xmath275 , with a best estimate of @xmath276 . following on from this , and other unequal - mass simulations by @xcite , @xcite performed the first systematic parameter study of nonspinning black - hole binary mergers with mass ratios in the range @xmath277 . they found that the maximum recoil velocity @xmath278 occurs for the mass ratio @xmath279 . for the case @xmath60 , @xcite found a recoil velocity @xmath280 . astrophysical black holes are generally expected to be spinning . including the spin vector of both black holes in the binary adds six more dimensions to the parameter space , giving seven in all when the mass ratio @xmath247 is included . exploration of this large parameter space began in 2006 , with the simplest cases of equal masses and spins , and is gradually growing to incorporate more generic binaries . the utb group carried out the first black - hole binary mergers with spin @xcite . they simulated three binaries , each having equal masses and starting on quasicircular orbits with initial orbital frequency @xmath281 , giving an initial orbital period @xmath282 . in the aligned case , both holes have spin parameters @xmath283 and vectors parallel to the orbital angular momentum @xmath284 ; in the anti - aligned case , the black holes have the same spin parameters but the vectors are anti - parallel to @xmath284 . a nonspinning equal - mass binary was run for comparison . the @xmath221 , @xmath205 components of the waveforms are plotted in fig . [ fig : utb - spin ] , which shows all three cases starting out at the same time @xmath102 at the same orbital period and gravitational - wave frequency @xmath285 . notice that the aligned system takes longer to merge , completing more orbits and producing a longer wavetrain . this behavior is caused by a spin - orbit interaction that produces an effective repulsive force between the black holes . in the anti - aligned case , this interaction yields an effective attraction , causing the binary to merge more quickly and with fewer orbits . the nonspinning case is intermediate between the other two . all three cases produce similar waveforms with a clean , simple shape . mode is shown for three cases : aligned spins ( solid / red ) , anti - aligned spins ( dotted / black ) , and no spins ( dashed / blue ) . all three simulations start with the same initial orbital period at @xmath102 . figure kindly provided by m. campanelli . ] it is also interesting to compare the total energy radiated as gravitational waves @xmath286 and the spin parameter of the final black hole @xmath287 for these three cases . the aligned case yields @xmath288 and @xmath289 while the anti - aligned case gives @xmath290 and @xmath291 . the nonspinning case is again between these two cases , with @xmath292 and @xmath212 @xcite . the black - hole mergers discussed so far have resulted from simple binary dynamics with no precession and have produced simple waveforms with a similar shape , at least when examined mode - by - mode in a spin - weighted spherical harmonic decomposition . might more `` generic '' precessing binaries generate waveforms with more complex patterns ? this important question pertains not only to basic orbital dynamics in general relativity , but also to strategies for developing templates to search for signals in data from gravitational - wave detectors ( see sec . [ sec : gwda ] ) . the parameter space for such binaries is very large , and explorations of this space have only recently started . the utb group , now at the rochester institute of technology ( rit ) , pioneered the study of precessing black - hole binary mergers . they evolved a generic black - hole binary with unequal masses and unequal , non - aligned precessing spins that undergoes @xmath293 orbits before merger and produces a relatively long wavetrain @xcite . this system has a mass ratio @xmath294 and arbitrarily oriented spins with magnitudes @xmath295 and @xmath296 . the initial conditions were determined from point - mass evolutions using 2.5 pn and 3.5 pn parameters . figure [ fig : rit-3d - tracks ] shows the difference in the black - hole trajectories @xmath297 . for a non - precessing binary , @xmath297 would have no component in the @xmath34 direction . in this generic case , the precession of the total system spin drives a precession of the orbital plane , producing evolution of the trajectory in the @xmath34 direction . for the generic binary evolution from @xcite . the precession of the system spin induces precession of the orbital plane . ] the resulting waveforms demonstrate the effects of precession on the amplitudes of the sub - dominant modes . figure [ fig : rit - precess ] shows the @xmath221,@xmath298 mode of the strain @xmath32 . the numerical - relativity evolution is shown by the solid curve ; note the amplitude modulations induced by the precession . the dotted curve shows a solution of the pn equations of motion at 3.5 pn order . , @xmath298 mode of the strain ( @xmath32 ) is shown for the numerical - relativity simulation ( solid ) and a 3.5 pn calculation ( dotted ) . the amplitude oscillations are induced by precession . ] the merger of a binary consisting of nonspinning black holes produces a single spinning black hole with spin direction aligned with the orbital angular momentum @xmath284 of the binary . in this case , the spin of the final hole arises purely from @xmath284 . when the individual black holes have spins that are not aligned with @xmath284 , the spin of the final hole will , in general , not be aligned with the initial spin but rather will have a different direction . this phenomenon , in which the spin direction of the final black hole differs significantly from the spin directions of the individual holes prior to merger , is known as a `` spin flip . '' the simplest situation in which a spin flip can occur is a binary with both spins anti - aligned with @xmath284 . the case of generic binaries , with unequal masses and arbitrarily - oriented spins has a much larger parameter space . the rit group @xcite has examined spin flips for several such cases . figure [ fig : rit-3d - flip - tracks ] shows the puncture trajectories for an equal - mass binary with equal spins having magnitude @xmath299 and initially pointing @xmath35 above the initial orbital plane . the black holes execute @xmath300 orbits before merger , during which time the spins , shown as arrows along each trajectory , precess by @xmath301 . the evolution of the spin direction for this case is shown in fig . [ fig : rit - spin - flip ] . the final black hole has @xmath302 with a spin direction flipped by @xmath303 with respect to the component spins just before merger . in a related phenomenon , the direction of the _ total _ angular momentum ( @xmath304 ) may change . this case was studied using general principles and extrapolations from point - particle results by @xcite , who predicted that binaries with @xmath305 will experience a flip in total angular momentum direction provided that the initial spins are equal and anti - aligned with @xmath284 , and that the individual black hole spins obey @xmath306 . in particular , they predicted that , for @xmath307 , a schwarzschild black hole should form if @xmath308 . motivated by this work , @xcite simulated a series of binaries with @xmath309 and having equal anti - aligned spins in the range @xmath310 $ ] . they found that a schwarzschild hole is formed when the initial spin is @xmath311 . above the initial orbital plane . the spin directions are shown as arrows along each trajectory every @xmath312 until merger . ] , from @xcite . the spin direction of the component holes , shown by the red arrows plotted every @xmath312 until merger , varies continuously due to precession . the blue arrow shows the final black - hole spin , with direction flipped discontinuously from that of the component holes just prior to merger . ] the mergers of asymmetrical spinning binaries will produce recoiling remnant black holes . in the simplest case , the binary components have equal masses and spin magnitudes , and the asymmetry arises from the spin vectors pointing in opposite directions . in early 2007 , several new results appeared in quick succession , generating considerable excitement among numerical relativists as well as astrophysicists eager to apply these results . in this section we discuss the basic results and defer to sec . [ sec : astro ] their impact on astrophysics . @xcite simulated the merger of equal - mass black holes with equal spin magnitudes @xmath313 and having one spin vector aligned and the other anti - aligned with the orbital angular momentum . these mergers produced kicks directed purely in the orbital plane with magnitudes @xmath314 . shortly thereafter @xcite showed that , for the specific case @xmath315 , @xmath316 ; linear extrapolation of results to the maximally spinning case @xmath317 yields @xmath318 . @xcite carried out a systematic study of equal - mass mergers with spin magnitude @xmath319 and direction aligned with the orbital angular momentum . the other black hole has spin magnitude @xmath320 , and direction both aligned and anti - aligned . they found a maximum kick velocity @xmath321 . these results all demonstrated that mergers of spinning black holes can produce significantly larger kick velocities than nonspinning mergers however , new results would soon show that kicks from the mergers of spinning black holes could get much larger indeed . the idea of `` superkicks '' was first discussed by the rit group @xcite . they observed that a pn treatment @xcite predicts that the recoil due to spin is maximized when @xmath322 , @xmath323 , and the spin directions are anti - aligned with each other and lying in the orbital plane . the jena group first simulated binaries in this configuration and demonstrated the phenomenon of superkicks . using @xmath324 , they found a resulting kick velocity @xmath325 @xcite . figure [ fig : jena - kick ] shows the coordinate positions of the black - hole punctures from one of their simulations . notice that the trajectories move out of the initial orbital plane and that the final black hole is kicked in the @xmath326 direction . the rit group @xcite carried out similar simulations with @xmath327 . their results show kicks perpendicular to the orbital plane with magnitudes up to @xmath328 ; using the expression from @xcite they predict a maximum recoil @xmath329 for the case of maximally spinning black holes , @xmath317 . @xcite performed a multipolar analysis of recoil from black - hole mergers , for both unequal masses and non - zero , non - precessing spins . they found that multipole moments up to and including @xmath330 are sufficient to accurately reproduce the final recoil velocity ( within @xmath331 ) , and that only a few dominant modes contribute significantly to it ( within @xmath332 ) . @xcite studied the role of spin in producing superkicks . they showed that the recoil velocity is almost entirely due to the asymmetry between the @xmath333 and @xmath334 modes of @xmath178 . in addition , the major contribution to the recoil occurs within a period @xmath335 before and after merger , after the time at which a standard pn treatment of the evolution breaks down . , initially oppositely directed in the orbital plane . during the evolution , the black holes move out of the original plane , and the final black hole recoils with velocity @xmath336 in the negative @xmath34-direction . ] einstein s theory of general relativity implies predictions for the dynamics and gravitational - wave generation for generic black - hole binaries at all stages of the merger process . as we have seen , numerical relativity now provides an excellent tool for concretely deriving these predictions beginning in the last orbits and continuing through the conclusion of the coalescence in the ringdown of the remnant black hole . there is no hard limit to how early numerical relativity can be applied in these simulations , but the considerable costs of these simulations tends to limit the duration to tens of orbital periods . though the pn expansion provides only an approximation to general relativity , it succeeds in making accurate and efficient predictions over very long time - scales , while the black holes remain well separated . late in the merger , however , as the black - hole velocities increase , these errors grow , and pn theory is no longer reliable a full understanding of general relativity s predictions for the merger process must draw on both approaches . the maturation of numerical relativity has fueled efforts toward synthesizing a complete understanding of black - hole mergers built from the results and techniques of both these fields . even a coarse general review of the pn formalism and results is outside the scope of this presentation . decades of work in this area have already been covered @xcite . here we focus exclusively on key synergistic areas where fruitful research interactions drawing from both numerical relativity and pn results and formalisms are yielding a more complete general - relativistic understanding of black - hole binary systems . the broad range of research that interfaces these two approaches can be grouped into four categories , based on the different ways that pn theory impacts numerical - relativity research . pn theory : 1 ) provides independent results for cross - checking and comparing with nr ; 2 ) provides models for the initial values needed to begin astrophysically realistic nr simulations ; 3 ) provides insight for interpreting nr results ; and 4 ) may provide the basis for empirical models representing the combined knowledge drawn from pn and nr investigations . as mentioned , pn theory is based on an approximate expansion of einstein s equations in powers of velocity @xmath337 providing general relativistic corrections to the newtonian , small - velocity limit . this approach has a long history of success as the primary framework for deriving explicit predictions from general relativity for physical and astrophysical observations . the relatively small velocities involved in most observations have made low - order pn theory an ideal tool for testing general relativity as the standard model for gravitational physics in many contexts including earth - orbit experiments , solar - system dynamical observations , and in even precision pulsar timing observations of binary neutron star systems @xcite . gravitational - wave researchers have applied pn theory to represent general relativity - based signal expectations for the vast majority of gravitational - wave searches for anticipated observations of black - hole and neutron - star binary inspirals , and likewise in the development and planning for current and future gravitational - wave instruments . the early epoch of black - hole binary inspiral is also well - described by low - order pn theory . for near - circular inspirals , @xmath338 , where the binary separation @xmath44 is scaled by the total system mass @xmath9 . at large @xmath44 , the pn expansion provides excellent predictions , even at low order . as the black holes lose energy and sink closer together , the velocity grows , requiring higher - order pn terms for sufficiently accurate predictions . currently , pn predictions are available up to 3.5 pn order ( 2.5 pn for spin - terms ) . these higher - order expansions seem to provide pn predictions sufficiently accurate for the analysis of data from current instruments for all but the last several orbits . once the separation approaches , say , @xmath339 , the accuracy of the pn expansion diminishes . even at ( hypothetical ) arbitrarily high orders , the expansion may fail , if the expansion parameter @xmath340 exceeds the series radius of convergence . precisely when the pn approximation effectively fails depends on the details of the system being studied and the requisite accuracy , but generally , for the last orbits and merger pn theory is no longer reliable . the strongest gravitational radiation is generated in the late stages of inspiral or merger where the internal consistency of the pn approach has been , at best , difficult to assess @xcite . numerical simulation can treat the late portions of the mergers , with practical resource limitations on the duration of the simulations , and consequently how far apart the black holes can be at the start of the simulation . would it be practical , however , to run numerical simulations long enough to `` overlap '' with the part of the problem treated successfully by pn methods , or would there be an intermediate region requiring yet another approach @xcite ? the first numerical inspiral results were quickly compared against pn calculations @xcite . these first comparisons yielded promising indications that the gap between nr and pn could be bridged , but also made clear that pn results were not uniquely determined for the final part of the inspiral . numerical results can be verified by internal consistency studies , examining , for instance , the convergence of the results toward consistency with einstein s equations as the resolution is increased . external verification , however , would strengthen the case that these new results are indeed correct . even more importantly , these comparisons would allow an independent check of the late inspiral pn predictions which are not easily confirmed by self - consistency studies . roughly one year after the first robust numerical - relativity results , simulations lasting @xmath341 were conducted , and quantitatively compared with various pn approximations . quantitatively cross - checking waveform comparisons requires some care . how do the differences between pn and numerical results compare with intrinsic numerical error estimate how do different variants of pn predictions compare with the numerical results ? because each waveform comes with no meaningful absolute reference in time and phase , how can the freedom to offset these parameters be controlled for the comparison ? effective comparisons also require longer waveforms than those produced in the earliest simulations . the first comparison addressing these issues came at the end of 2006 @xcite . the study considered the case of equal - mass non - spinning black - hole mergers , comparing a 3.5 pn waveform with numerical results covering the last seven orbits . the compared waveforms are shown in fig . [ fig : pnprlwf ] . in the comparison the waveform phases agreed within 1 rad of phase drift , for a little over ten gravitational - wave cycles preceding the last orbit before merger , comparable to numerical error estimates . this result gave clear indication that pn waveforms could be accurate in the last orbits approaching merger , and that pn and nr , combined , could treat the complete waveform signal . not all equally valid variants ( approximants ) of the approximate pn waveforms agree this well with numerical results . to understand this , it is worthwhile to consider pn results in a little more detail . the pn approximation is formally understood as an expansion in powers of the speed of the merging black holes , @xmath340 . a concrete result of the theory will typically be a taylor expansion for a specific dependent variable , in terms of a chosen independent variable . the obvious choice , to express the waveform itself ( gravitational - wave strain ) as a function of time , would give a poor result since the sinusoidal shape of the waveform is difficult to approximate by a polynomial . pn waveform results are typically expressed as separate expansions for the orbital phase , and polarization component amplitudes . these are given as expansions in the time to merger @xmath342 or orbital frequency @xmath343 . the orbital phase information may also be expressed by an expansion for @xmath344 , referred to as the chirp - rate ; see @xcite for a review of the approach and a particular explicit waveform expansion . the pn information may also be encoded in an hamiltonian formulation description of the dynamics , which may be integrated numerically . researchers also choose between taylor series and other `` resummed '' expressions of the results based , for instance , on pad approximants @xcite . in the comparisons described above , the pn waveform phasing was found to agree with numerical results derived from a taylor series expansion for chirprate @xmath345 expanded in powers of frequency . this is known as the taylort4 approximant in the language of the ground - based gravitational - wave community . later studies with longer and more accurate numerical simulations provided more precise tests of phases and amplitudes for a variety of pn approximants @xcite . @xcite found even closer agreement ( than found in @xcite ) with the taylort4 approximant , to within 0.05 rad over nearly 30 cycles before the orbital frequency reaches @xmath346 at roughly an orbit before merger . up to @xmath347 , while the comparison in right panel extends this range to @xmath348 . ] figure [ fig : boylepncomp ] shows the results of four pn approximants with terms kept to various pn orders . the excellent agreement of the taylort4 approximant with numerical results is not matched by other equally consistent pn waveform approximants in the late portions of the waveforms . however , all four approximants agree to within about a radian if the comparison is cut off at waveform frequency @xmath349 , about five orbits before merger [ note that this is the frequency of the dominant @xmath350 mode , twice the orbital frequency @xmath351 . while most attention has so far been given to the nonspinning cases , some results exist for other regions of parameter space . for equal - mass configurations of black holes with spins aligned with the orbital angular momentum , the phase agreement of the taylort4 approximant is not generally best , and phase differences ( over ten cycles preceding @xmath352 ) are much larger than those seen in the nonspinning case , with differences near 1 rad or more for spins @xmath306 @xcite . pn spin effects are not yet derived beyond 2.5 pn order , possibly limiting the accuracy of the pn waveforms in these comparisons . some pn comparisons have also been examined for other configurations , including a precessing system @xcite and for eccentric configurations of equal - mass , non - spinning mergers @xcite . in comparing fourier - domain pn approximants with numerical results for a set of non - spinning mergers with various mass ratios , @xcite discovered that the abrupt truncation of the fourier - domain waveforms , inducing gibbs oscillations in the time - domain , can approximately emulate the physical quasinormal ringdown that terminates the numerical - simulation waveforms . even better agreement is possible if @xmath254 , the symmetric mass ratio , is allowed to extend beyond its physical range ( to @xmath353 ) or if an extra pseudo-4pn order term is added . more recent studies have confirmed that these simple models are directly useful in some gravitational - wave observation applications @xcite . these results have motivated further development of phenomenological full - waveform models . for the early part of the waveforms various pn approximants agree at 3.5 pn order , the best currently available , providing excellent approximations . at late times , the results of different approximants diverge , and the numerical - simulation results are the only way to accurately derive the predictions of general relativity . can an analytic waveform description be developed that simultaneously encodes the results of both complementary theoretical approaches ? the generally simple features of the numerically simulated merger waveforms raise hopes that fairly accurate approximate analytic waveform descriptions may be produced with simple dependence on the binary system parameters , providing an efficient means also of interpolating from a sparse sampling of the full parameter space for which numerical - simulation studies are completed . these waveform models would be computationally efficient ( compared to numerical simulations ) and may have applications in a broad class of observational data analysis applications . ajith and collaborators have developed a fourier - domain full - waveform model phenomenological approach resembling the treatment of @xcite , but with greater emphasis on matching the numerical waveforms @xcite . they begin combining information from the pn and nr - based waveforms by first stitching together time - series data for the dominant ( @xmath190,@xmath205 ) component waveforms , including a long pn precursor joined to a numerical merger waveform . fourier transforms of waveforms are then fit to a parametrized model of the waveform similar to waveform families previously applied in phenomenological treatments of purely pn waveforms @xcite . their waveform model for mergers of non - spinning binaries has the form @xmath354 where @xmath355 is a constant related to the masses , and @xmath356 is the distance to the source . the effective fourier amplitude @xmath357 is modeled piecewise , @xmath358 with distinct power - law segments before and after a merger frequency @xmath359 , and a lorentzian @xmath360 decay beyond the transition frequency @xmath361 demarking the ringdown of the final black hole . the fourier phase is represented by a single power - law expansion @xmath362 for each waveform , the free amplitude and phase coefficients are determined by a fit to the stitched nr - pn hybrid waveforms . it was found that the mass - ratio parameter dependence could be modeled by simple fits to quadratic functions of the symmetric mass ratio . these waveforms have been applied in gravitational - wave data analysis studies ( see sec . [ sec : gwda : direct];@xcite ) . recently @xcite have developed a generalization of this model for non - precessing system of spinning black holes . stitching together pn and nr waveforms can be avoided , and closer contact with the basic physics maintained , by requiring direct pn consistency in the full waveform model . some pn approximants can resemble the numerical late - merger results . if such approximants can be tuned to agree with numerical results right up to merger by careful choice of adjustable parameters which only affect pn consistency beyond known pn order , the result would simultaneously encode pn and numerical - relativity results . the effective - one - body ( eob ) family of pn hamiltonian models has promise as a tunable model for encoding both pn and nr results @xcite . in the time domain , techniques have also been developed for extending these waveforms into the ringdown of the final black hole . analysis of the early numerical results in comparison with an untuned eob model gave promising indications that the waveforms could be closely approximated this way @xcite . with tuning , it appears that this construction can provide an analytic but potentially very accurate , approximation to the complete coalescence waveform . in the eob model , the binary motion is recast as the motion of a single _ effective _ body of mass @xmath363 moving about a central potential , as is familiar from newtonian mechanics . in the general relativistic version of this framework , the effective body s motion follows a geodesic ( to 2 pn order ) around a modified version of a schwarzschild metric . the motion of the effective body is described by an effective hamiltonian , which , for systems of nonspinning black holes , may take the form @xmath364 } \,.\end{aligned}\ ] ] the expressions for @xmath365 and @xmath366 are chosen so that the pn - expansion of the hamiltonian is consistent with the results from pn theory to known order ( 3 pn for nonspinning black holes ) @xcite . the nonconservative contribution to the motion , arising from the loss of angular momentum to gravitational radiation , is encoded in an additional flux term entering as an external force in hamilton s equations , which is also constrained to be consistent with pn theory ( typically to 3.5pn order ) . hamilton s equations are integrated to derive the effective body s motion , and hence the black - hole trajectories . waveforms are constructed using pn extension of the quadrupole formula relating the motion of the black holes to the amplitude and phase of the multipolar radiation components . the last part of the radiation , arising after merger , is completed by continuously matching a sum of quasinormal ringdown modes to the waveforms truncated near the point of merger . while consistent with pn theory , the formalism can be adjusted to also match the wave forms derived from numerical - relativity simulations . the flux function , as well as @xmath367 and @xmath368 , can be adjusted by the addition of higher - order pn terms to modify the strong - field dynamics to agree with nr without violating pn consistency . an early implementation of this approach , encoding the combined results pn and nr for nonspinning black - hole mergers was presented by @xcite . figure [ fig : faithful ] shows their result comparing numerical and adjusted eob - model waveforms for a 4:1 mass - ratio merger . component of the gravitational - wave strain from the merger of binary with mass ratio 4:1 . the waveform is for an observer located an at inclination angle @xmath369 from the axis of rotation . [ fig : faithful],scaledwidth=45.0% ] subsequent work involving more accurate numerical simulations , and more careful tuning of the eob model has improved eob model waveforms @xcite . recent comparisons of improved eob models with high - accuracy numerical results from @xcite yield differences comparable to numerical errors showing phase agreement within about @xmath370 rad through @xmath371 30 gravitational - wave cycles @xcite . recently the eob approach has been extended to match cases of non - precessing spinning black - hole mergers@xcite . there remain open questions in extending these waveform models to generic spin configurations , more extreme mass ratios , and full multipolar content . before a numerical simulation can commence , a numerical relativist requires a model for the initial configuration of the two black holes and the geometric fields which represent them . as discussed in sec . [ sec : comp : initialdata ] this involves not only solving the general relativistic constraint equations , but also providing some ansatz for the free data . in some cases , some information from pn theory may be applied to produce an initial data model that more precisely represents the desired physical configuration . to understand how pn information can be applied , it will be useful to revisit the discussion of numerical initial data . typical simulations begin with a modeling ansatz , the brandt - brgmann model , for instance , which produces initial _ field _ data from a given set of particle - like parameters related to the black holes masses , spins , initial positions , and momenta . usually , the initial data models make no attempt to represent the gravitational radiation fields which would have been previously generated by the motion of the black holes , and do not utilize pn information in this reduction of the field degrees of freedom to particle parameters . information from pn theory is frequently applied in choosing the specific particle parameters which correspond to the sought - after simulation , particularly for circular inspiral configurations . work has also begun toward making richer use of pn information for improving the field ansatz . most of the simulations discussed have been designed to represent black holes in circularized orbits . before long - lasting numerical - simulation results were available , comparative studies of numerical initial data sets with pn - derived information provided a gauge of the results [ see for instance @xcite ] . without evolutions , these studies focused on theoretical properties of the black - hole configuration space such as the isco ( see sec . [ sssec : astro_grav : basic : bh ] ) . still subtle imbalances in the initial data , such as excess angular momentum for the chosen separation , can lead to eccentricity in the simulation which impacts the simulated radiation waveforms @xcite . the residual eccentricity can be reduced via a straightforward iterative procedure @xcite , but this requires repeated simulations lasting several orbits ; expensive in both time and computational resources . to minimize this eccentricity without resorting to iteration , several groups use trajectory information from pn theory in setting up the parameters for the numerical initial data . this approach enables simulations with eccentricities @xmath372 for equal - mass non - spinning mergers with initial separations @xmath373 @xcite . the technique is helpful in simulations with unequal masses and nonvanishing spins as well , though the residual eccentricity is generally larger @xcite . work is underway on techniques allowing the richer application of pn information from numerical initial data modeling . the common assumption of a conformally flat spatial metric in numerical initial data , disagrees with pn results at the 2 pn order @xcite , making this a likely leading source of modeling error which produces spurious initial transients in the simulations . though not yet as well - developed as the widely applied models , an alternative approach applies metric information from pn theory for a non - trivial initial conformal metric @xcite . with these techniques it is also possible to encode in the initial data information about the prior radiation generated by the system before the `` initial '' time of the simulation @xcite . we have noted several specific areas where research in numerical simulations makes contact with pn theory . these alone do not provide a full picture of the interplay between the two theoretical approaches . most broadly , pn theory provides a foundation for interpreting numerical results . numerical relativists draw widely from pn - based background knowledge of the black - hole binary . many important phenomena , such as the isco , spin - orbit coupling , spin precession , and ` gravitational rocket ' kicks were first studied in the pn approximation , providing a foundation for subsequent numerical studies . small examples can be noted throughout this review . in particular we point to the value of pn results in interpreting mergers of spinning black holes ( sec . [ sec : gws : spinning ] ) and kicks ( sec . [ sec : gws : uneq ] ) . analytic formulas for approximately expressing the final state of the black hole , mass , spin , and momentum have also drawn heavily from insights based on the pn treatment ( see sec . [ sec : astro : recoil_formulae ] . ) several detectors are active or in their planning stages to detect the gravitational - wave signals from astrophysical processes . prominent among these are the ground - based interferometric detectors ligo ( u.s.a . ) @xcite , geo ( germany ) , virgo ( italy ) , tama ( japan ) , and aigo ( australia ) sensitive to frequencies in the range @xmath374 , as well as next - generation instruments such as the einstein telescope @xcite . also planned to launch at the end of the next decade is the laser interferometer space antenna ( lisa ) , with complementary frequency sensitivities between @xmath375 . all of these instruments are subject to a variety of noise sources ; in the ground - based detectors , these sources will completely overwhelm the signal , unless filtered intelligently . figure [ fig : ligo_virgo_noise ] shows the design sensitivity curve resulting from these disparate noise sources , for the ligo and virgo detectors . , while virgo does a little better ( note that achieved sensitivities may not perfectly mirror the design curves ) . taken from @xcite . reproduced by permission of the institute of physics.,width=288 ] burst analysis can extract many signals , but optimal results in gravitational - wave data analysis require _ matched filtering _ of the noisy input signal @xcite . fundamental to this approach is some measure of overlap between a physical signal @xmath376 and filtering template waveform @xmath377 . for this it is convenient to use a frequency - space inner product @xmath378 between signals , defined as @xcite @xmath379 , \label{eq : inner_product_da}\end{aligned}\ ] ] where @xmath380 and @xmath381 are the fourier transforms of the signals , and @xmath382 is the ( one - sided ) noise power spectral density of the detector we are interested in . this can be normalized to produce the _ match _ ( or _ overlap _ ) @xcite between two waveforms . to use requires the availability of a set of waveform _ templates _ simple , few - parameter analytic model waveforms to filter against . even for a relatively simple system such as a black - hole binary , the combination of possible source and detector configurations leads to a 17-dimensional parameter space . some of these parameters are intrinsic to the source : total binary mass @xmath383 , symmetric mass ratio @xmath254 , spins @xmath384 , time to coalescence @xmath385 , eccentricity @xmath227 , and eccentric phase @xmath386 . the remaining parameters have to do with the relation between source and detector : ( luminosity ) distance to source @xmath387 , inclination @xmath388 , orbital phase @xmath389 , waveform polarization @xmath131 , and position of the binary on the detector s sky @xmath390 . restricting consideration to binaries that have circularized by the time they enter the detector s window @xcite allows us to neglect the eccentricity parameters @xmath391 . crucially also all observables have a simple dependence on the total ( redshifted ) mass , @xmath392 . similarly , the observed waveforms have a trivial dependence on the time to coalescence @xmath385 . as this means only one theoretical waveform has to be generated to cover all astrophysical masses and coalescence times , @xmath392 and @xmath385 are sometimes treated as extrinsic parameters instead . nevertheless , numerical relativists are left with a seven - dimensional parameter space to cover ( @xmath254 , @xmath393 , @xmath394 ) , with no obvious shortcuts to lighten the workload further . moreover , numerical methods will always struggle with finite accuracy . equipped with a parametrized set of template waveforms @xmath395 , we can test an incoming detector data stream @xmath396 , consisting of a ( possible ) `` exact '' gravitational wave @xmath397 and detector noise @xmath398 by calculating the _ signal - to - noise ratio _ ( snr ) @xcite : @xmath399 in the case where the signal @xmath397 in the data stream corresponds perfectly ( up to overall scaling ) with the model waveform @xmath400 for some value of the model parameters @xmath401 , we obtain the optimal snr : @xmath402 . we refer to this simply as the `` snr '' @xmath403 for the remainder of this section . expected snrs depend strongly on the detector in question . generally speaking , the ground - based detectors in operation ( ligo , geo , virgo , etc ) are expected to observe with only modest optimal snrs , while lisa is expected to observe mbh mergers with snrs of hundreds @xcite . of course , real model waveforms will be inaccurate , due to incomplete knowledge of the underlying physics , or perhaps a desire for simplicity . in this case , the achieved snr can not be optimal . @xcite defined the _ fitting factor _ ( ff ) as the reduction in signal from such imperfect templates : @xmath404 a fitting factor of 1 means that the exact signal lies somewhere in the model space . however , for a particular detector , it is impossible to generate templates arbitrarily close together in any one parameter dimension . we can develop a sense of how closely spaced a set of templates must be so that any physical signal will be detected by at least one of them with high likelihood . @xcite developed a geometric picture of a template - space metric defining the distance between neighboring closely - spaced templates . with these tools , he was able to link the number of templates @xmath405 to the desired _ minimal match _ ( mm ) the worst value of the match between the signal and any template . the connection between these concepts is nicely laid out by @xcite . figure [ fig : lindblom_fig3 ] , from @xcite , demonstrates how @xmath406 and @xmath407 combine to produce an effective mismatch @xmath408 . , model waveform with the same physical parameters @xmath409 , actual `` best fit '' model waveform @xmath410 , and template bank waveforms @xmath411 , @xmath412 . [ figure taken from @xcite],width=288 ] the percentage of detected signals scales as the cube of the fitting factor . for this reason , detector analysts require rather stringent fitting factors from their template banks , e.g. , 97% ( implying detection of 90% of all signals present ) . for initial ligo , for instance , @xcite estimate that @xmath371 100 zero - spin templates ( produced from @xmath371 10 different numerical simulations ) would suffice to detect all nonspinning merging binaries . obviously we expect that the addition of spins of significant magnitude and arbitrary direction will greatly increase this number . detection is only a first step , although an important one . when signal strengths are high , as is expected with lisa for massive binaries , we might expect to be able to read off some of the system s intrinsic and extrinsic parameters . achieving this requires templates that accurately cover the parameter space of the binary , with a one - to - one relationship with the parameters of the binary that is , that the best - fit model waveform @xmath410 and closest - parameters model waveform @xmath409 in fig . [ fig : lindblom_fig3 ] coincide closely ( within statistical errors ) . it also requires that we have an understanding of the probabilistic correlations between parameters . for high-@xmath403 signals , the simplest way to estimate parameter uncertainties is using the _ fisher information matrix _ @xcite , which uses simple derivatives of the template waveform shape with respect to the physical parameters : @xmath413 in the high - snr limit , the parameters are assumed to have a multivariate gaussian distribution , and the fisher matrix is the inverse of the covariance matrix between parameters . more sophisticated approaches that do not assume high snr include markov - chain monte carlo methods @xcite . different requirements can be made on the quality of the developed templates , as described by @xcite . the first is that they be _ effectual_. roughly speaking , this means that the templates must be capable in the bulk of detecting gravitational - wave signals . in the language of matched filtering above , we demand that the fitting factor @xmath414 be extremely close to unity for any signal . more stringently , we might demand that the templates be _ faithful _ : that is , that the template waveform corresponds to underlying source parameters that correspond closely to the parameters of the detected waveform source . in the language of @xcite , the model waveforms @xmath409 and @xmath410 coincide closely ( see fig . [ fig : lindblom_fig3 ] ) . this requirement will be crucial for identifying the astrophysical source of the radiation . traditionally , templates used in the search for comparable - mass black - hole binaries have depended wholly on pn theory . as detection strategies are most sensitive to phase discrepancies between template and observed signal ( see below ) , the determination of waveform phase to high accuracy has been crucial . modern pn templates include phase terms up to 3.5pn order for nonspinning binaries , with 2.5 pn order spin corrections . many templates use this high - accuracy phase with the leading - order quadrupole amplitude to produce _ restricted _ templates ; however , amplitude corrections are known to 2.5 pn order ( higher for certain low-@xmath415 modes ) @xcite . while such pn - driven templates offer excellent phase accuracy during the long , slow inspiral of the binary , their usefulness becomes questionable as we approach the last pre - merger orbits of the binary . we may ask , then , what does our new numerical insight into the final moments of merger bring us ? once full numerical waveforms became available , several groups used the results to test older predictions of the effect of the merger segment on observability . for the equal - mass nonspinning case , the numerical merger gave results consistent with @xcite for initial ligo , though with a smaller merger snr @xcite . the boost in snr from merger is significantly greater for advanced ligo and lisa , as shown in fig . [ fig : ligosnr ] , from @xcite . before merger ) is used ; the dotted curve uses the signal between this time and when final plunge commences ( @xmath416 to @xmath417 before merger ; the thin solid curve uses just the merger waveform ( starting @xmath417 before merger ) . the thick solid curve is the combined result of using the entire waveform . the snr , and hence distance reach , of the detector is clearly greatly enhanced by the final merger portion.,title="fig:",width=336 ] before merger ) is used ; the dotted curve uses the signal between this time and when final plunge commences ( @xmath416 to @xmath417 before merger ; the thin solid curve uses just the merger waveform ( starting @xmath417 before merger ) . the thick solid curve is the combined result of using the entire waveform . the snr , and hence distance reach , of the detector is clearly greatly enhanced by the final merger portion.,title="fig:",width=336 ] the relevance of the final plunge and merger to the overall snr of the binary depends on the total mass of the binary and where it falls in the window of the detector . figure [ fig : lisahchar ] , adapted from @xcite , shows this for lisa , plotting the `` characteristic signal strain '' @xmath418 , the late - merger and ringdown signal will fall outside lisa s sensitivity band . for such low masses , inspiral - only waveforms should be adequate for data - analysis purposes . curve , we mark times before the peak amplitude ( circle ) one hour ( square ) , one day ( diamond ) and one month ( triangle).,width=336 ] the ligo and virgo detectors are sensitive to frequencies above @xmath419 ( see fig . [ fig : ligo_virgo_noise ] ) . the merger of binary systems is marked by a chirp gravitational - wave signal , whose monotonically increasing frequency saturates at the dominant quasinormal mode , which depends on the post - merger hole s mass and spin . for a nonspinning binary of total mass @xmath9 , this @xmath420 . this means that a merging binary of mass greater than @xmath421 will never be seen by ligo . in contrast , most post - newtonian templates stop at @xmath422 , the frequency at the innermost stable circular orbit ( well defined only for test particles ; see sec . [ sssec : astro_grav : basic : bh ] ) . this is @xmath423 for an incomplete pn waveform to be useful , we want the missing merger - ringdown section ( i.e. the part of the signal with @xmath424 ) to contribute as little as possible to the total snr . for instance , if we take the high end of ligo s sensitivity window to be @xmath425 , this is @xmath422 for a binary of total mass @xmath426 . then , for binaries with @xmath427 , @xmath428 will fall within the visible window ; that is , ligo will see the final stages of the binary merger , where no acceptable pn waveform is available . recent work by @xcite supplied a more precise answer : inspiral - only pn templates can be trusted for all ligo configurations for systems with @xmath429 . above this critical mass , there is a gap in reliable information , which must be filled by numerical results . figure [ fig : panetal_snrdist ] , adapted from @xcite , shows the importance of the last stages of merger in initial ligo as a function of mass range . the thin blue curves are hybrid nr / eob waveforms , while the thick red waveforms are generated from these by `` whitening '' : that is , they have been fourier - transformed to the frequency domain , then rescaled by @xmath430 , and finally re - transformed to the time domain . the whitened amplitude in a segment is proportional to the contribution of that segment to the total snr ; each marked segment accounts for 10% of the total . as the total system mass @xmath9 increases , the whitened amplitude becomes more bulked toward the merger time . for the largest total mass shown , @xmath431 , more than 90% of the signal power comes from the last cycle + merger + ringdown . increases , the final merger and ringdown account for a greater fraction of the total power . adapted from @xcite.,width=336 ] in sec . [ sec : gws : longer_wfs ] , we referenced the recent `` samurai '' project @xcite , which establishes the consistency of all of the `` long '' ( pre - merger duration @xmath432 ) equal - mass nonspinning waveforms within their stated numerical accuracy . as well as direct comparisons of phase and amplitude errors over time , they conducted mismatch tests in the regime of the ligo and virgo ground - based detectors . they demonstrated that quadrupole waveforms ( @xmath433 ) from the five numerical codes have tiny mismatches ( @xmath434 ) for binary masses above @xmath435 with `` enhanced '' ligo , and masses above @xmath436 with advanced ligo , virgo and advanced virgo . all waveforms would be indistinguishable for snrs below 14 , entirely reasonable for the current generation of ground - based detectors ; this is shown for the advanced ligo detector in fig . [ fig : samurai_measuredaligo ] . below which different `` long '' numerical waveforms are indistinguishable for the advanced ligo detector . even the earliest , lowest - accuracy published long waveforms are indistinguishable for snr levels below 14 , as demonstrated in this figure from the samurai paper @xcite.,width=336 ] studies using incomplete ( non - merger ) pn waveforms have shown that introducing fuller harmonics of spinning and precessing binary systems greatly enhances parameter estimation , removing some parameter correlations and hence reducing certain errors by several orders of magnitude ( though other parameters are significantly less improved ) @xcite . it is expected that the introduction of full numerical waveforms for the merger portion will lead to further improvements , at least in regions of the detector s window where the binary merger is visible . a recent study by @xcite used phenomenological merger - inspiral - ringdown templates ( discussed in sec . [ sec : gwda : templates ] ) to assess these improvements for advanced ligo and advanced virgo . they showed that parameter - estimate accuracy is improved substantially compared to inspiral - only waveforms for system masses @xmath437 ; in particular , the average sky - position error for an @xmath438 equal - mass , nonspinning binary at a luminosity distance of 1 gpc drops to about one - tenth of a square degree . since they only uses the dominant mode for this investigation , there will be systematic errors as well . numerical verification of the extent of parameter estimation improvements for lisa has led to differing estimates @xcite . in particular , @xcite find that 50% of injected signals could be located on the detector sky within an error box of 3 arc min , while @xmath385 could be measured to less than a second and @xmath387 within 1.5% ( ignoring issues of weak lensing ) . they attributed these impressive results to a combination of very high snr ( between 900 and 9000 ) and the use of higher multipoles of the radiation . in contrast , @xcite found smaller ( by about one order of magnitude ) improvements over inspiral - only waveforms . this may in part be attributable to the use of different mass ratios and different - length waveforms by the two groups , but has not been fully resolved . one of the eventual aims of numerical - relativity simulations of binary mergers is the production of a set of gravitational - wave `` templates '' that will cover the many - dimensional space of astrophysical parameters . numerical merger simulations are still orders of magnitude too slow to run on the fly to filter incoming detector data streams ; thus there is a need to develop analytic waveform expressions that encode the numerics . it has been demonstrated in the physical regime of nonspinning binary mergers @xcite that inspiral - only pn - based templates become inconsistent with each other ( and with full waveforms ) at total binary masses @xmath439 for initial and advanced ligo configurations ; above this mass , full templates including the numerical - simulation - based understanding of merger / ringdown become necessary . one such set of templates incorporating numerical - relativity data is due to @xcite . these were extensions of the stationary - phase approximation ( spa ) templates used in ligo / virgo data analysis , with input from equal - mass numerical waveforms supplied by frans pretorius @xcite , as well as equal- and unequal - mass numerical waveforms from the goddard group . these templates are effectual they achieved fitting factors of @xmath440 but they did this by extending source parameters into unphysical space . to date a handful of useful faithful full - waveform template banks has been produced , aimed at covering all mass ratios for nonspinning binaries . the `` phenomenological '' templates of @xcite are simple three - segment curves in frequency space , with matching parameters tuned by numerical data generated by the ` bam ` @xcite and ` ccatie ` @xcite codes . the templates of @xcite , coded in the lsc algorithm library as `` eobnr '' , are an extension of effective - one - body waveforms , with a single `` pseudo-4pn '' parameter tuned by numerical data ( initially from the goddard hahndol code , and later using waveforms from the caltech - cornell group . both the phenomenological and eobnr template banks are discussed in sec . [ sec : nrpn : analytic ] . both template banks are faithful in the sense described above , in the restricted @xmath441 parameter space , and both have been used in data - analysis injection tests @xcite . currently , eobnr templates are being used both for injection and filtering in the high - mass region of ligo test analysis , and also for injection for the ringdown band . they were also both used in the matched - filter analyses of the ninja project ( see sec . [ sec : gwda : inject ] ) , performing as well as the inspiral - only templates in detection , and considerably better in the limited parameter estimation attempted . figure [ fig : ninja_eobnr_parest ] , from @xcite , shows the accuracy with which the total mass @xmath9 and time of merger were extracted from all detected injections in ninja , when the eobnr templates were used . results for the phenomenological templates were similar for @xmath9 , but performed slightly less well for time of merger , as it is not an explicit part of this model ( given the relatively small number of samples in the ninja tests , such minor differences may not be significant ) . , for all detected signals , while the lower shows the error in merger time . adapted from @xcite . reproduced by permission of the institute of physics.,width=288 ] , for all detected signals , while the lower shows the error in merger time . adapted from @xcite . reproduced by permission of the institute of physics.,width=288 ] as significant spin is expected in astrophysical black holes at all scales , it is important to develop templates that encode spin also . @xcite demonstrated that matched filtering with nonspinning merger waveforms does a good job in detecting waveforms from binaries with significant spins . as their spinning waveforms come from aligned - spin mergers where the total spin is zero ( one hole s spin is aligned with the orbital angular momentum , while the other s is anti - aligned ) , their configurations will lack some prominent spin - related effects , including spin - orbit effects ( pulling in or pushing out the isco of the orbit ) and precession of the orbital plane . their conclusions , then , are likely to be optimistic , and more generic configurations can be expected to do much worse with nonspinning templates . the phenomenological templates of ajith _ have recently been extended to include non - precessing spinning systems @xcite . these have been shown to be both effectual and faithful in searches over non - precessing binary hybrid waveforms with ligo , where earlier zero - spin templates failed badly . for maximum simplicity , these new phenomenological template families are ultimately parametrized by three numbers : the total mass @xmath9 , the symmetric mass ratio @xmath254 , and a single effective spin @xmath442 where @xmath443 . an obvious application of numerical waveforms for data analysis is in directly testing analysis algorithms . as current templates are overwhelmingly based on pn information either extended heuristically to include merger and ringdown , or terminating before merger so current detection and parameter estimation algorithms are also based on incomplete information . recently , a multi - group collaboration ( the ninja project ) has performed the first direct insertion of explicit numerical waveforms into a simulated ligo datastream , and investigated detection efficiency and systematic errors @xcite . essentially all active numerical - relativity groups contributed waveforms ; these were scaled randomly to represent different - mass sources , resulting in 126 different injections . nine data - analysis groups analyzed the post - injection data stream , with a variety of methods . these fell into three main groups : matched filtering , burst analysis , and bayesian techniques . for the matched filtering , some groups used standard `` taylorf2 '' inspiral - only templates generated by pn theory , some used ringdown - only templates , and some used full inspiral - merger - ringdown templates , such as the eobnr and phenomenological waveforms discussed below and in sec . [ sec : nrpn : analytic ] . both the taylorf2 and the full template waveforms yielded quite high detection rates around 80 of the 126 injected signals in triple coincidence between the ligo detectors while the ringdown templates performed more poorly , detecting only 45 signals in triple coincidence . however for parameter estimation of the total mass @xmath9 , most of the taylorf2 estimates have an error of 40% or more , significantly more than for the eobnr and phenomenological estimates . figure [ fig : ninja_eobnr_parest ] shows errors in detected mass ( upper panel ) and coalescence time ( lower panel ) when using the eobnr templates . while mass errors are still quite large , between @xmath444 and @xmath445 for the bulk of detected signals , they are significantly less biased than those from the taylorf2 templates ( compare with fig . 8 of @xcite ) . estimation of the coalescence time @xmath385 was better for the eobnr and phenomenological template banks as well , as might be expected ( since taylorf2 templates must necessarily cut off before coalescence ) . however , the eobnr and phenomenological templates typically covered a larger mass range as well ( up to @xmath446 compared to @xmath447 for taylorf2 ) , which will bias the results when the injected signals had a large mass . burst analysis is a more abstract approach to the problem , using partial information about signals , such as dominant frequency and approximate duration . in ninja , two burst techniques were used the `` q pipeline '' and the hilbert - huang transform ( hht ) . the q pipeline fits signals to a sine - gaussian model parametrized by a central frequency and a @xmath24 factor ( @xmath448 , the number of oscillations under the gaussian ) . at the single - detector level , the detection rate is comparable to that of the eobnr and phenomenological matched - filter searches . parameter estimation is limited to estimates of the peak power time ( close to @xmath385 ) and peak frequency ; for these parameters , performance was comparable to the ringdown matched - filter search . the hilbert - huang transform decomposes input data into `` intrinsic mode functions '' , each characterized by a single frequency scale . applied to the signal , the hht generates a high - resolution time - frequency map of the data @xcite . the hht applied to the ninja data found @xmath449 coincident events , competitive with the best matched - filter results ; however , no parameter estimation is currently possible with this method . bayesian analysis attempts to reconstruct the posterior probability density function of a parametrized signal based on the observed signal . two variant bayesian approaches were taken in ninja analysis , involving different monte carlo methods markov chain monte carlo ( mcmc ) and nested sampling . in both cases , a parametrized waveform model is needed . for the mcmc study , a restricted pn waveform was used , only 1.5 pn in phase , but including leading spin effects . detection of injected signals performed well , but the high mass of the injections meant that the bulk of the snr occurred in the merger - ringdown phase ; as a result the inspiral - only model masses merger time was significantly biased . the nested sampling effort instead uses the taylorf2 templates ( restricted amplitude , 2.0 pn in phase ) as well as phenomenological templates . in detection , the phenomenological templates significantly out - perform the taylorf2 model ; in parameter estimation , it achieves results consistent with those of the matched - filter analysis . overall , current results are encouraging , with high detection rates of injected waveforms , and good , if limited , accuracy in estimated parameters when attempted with faithful template banks . however , the project was hampered by the limited length of the contributed waveforms and unrealistically stationary ligo and virgo noise profiles . follow - on work in this area will demand longer numerical waveforms that span the ligo sensitivity window , or more likely , a systematic blending of long pn inspiral waveforms to late - inspiral / merger numerical waveforms , but also may be extended to other gravitational - wave sources . the recent successes in modeling of binary black - hole mergers have captured the attention of astronomers and astrophysicists . while black - hole binaries have been studied both observationally and theoretically for many years , most efforts have been primarily within the framework of newtonian gravity . the new developments in numerical relativity now supply the missing piece : the effects of the final merger in the strong - field general relativistic regime . in this section we highlight three key areas of current interest : recoiling black holes , the spin of the merged black hole , and electromagnetic signatures from the final merger . one of the more dramatic implications of gravitational - wave computations for astrophysical observations is the possibility that gravitational radiation - induced recoil will eject black holes from their host galaxies . the largest recoil velocities predicted by numerical relativity exceed the escape velocities of many galaxies ( @xmath450 ) . calculating the probability of this kind of event requires a detailed understanding of the recoil on mass ratios , spin magnitudes , and spin orientations , together with some expectations about the distributions of these parameters . it would be useful to have in hand a simple , analytic formula expressing the relevant dependencies of the recoil velocity . the highly nonlinear , strong - field interactions of merger dynamics determine the final recoil velocity of the merged remnant . numerical simulations are required to accurately compute such results . however , in attempting to construct an ansatz for a phenomenological formula , one might initially assume a form consistent with pn predictions . this will suggest the functional dependence on mass and spin , and their leading powers . for example , pn analysis @xcite suggests that the recoil due to unequal masses may be proportional to @xmath451 ( the fitchett formula " ) , where @xmath254 is the symmetric mass ratio . meanwhile , to leading pn orders , the spins should contribute through components of @xmath452 and @xmath453 @xcite . such an ansatz can be further supported and constrained by symmetry arguments . an economical approach is to consider the recoil velocity as a taylor expansion in all six spin components ( three for each black hole ) , where the coefficients are functions of mass and initial separation . rotation , parity , and exchange symmetries of the binary system then impose relationships between these coefficients @xcite . by such arguments , for example , the component of the recoil velocity parallel to the orbital angular momentum can be shown to be proportional , to leading order in spin , to the dot product of the coordinate separation vector between the black holes and some linear combination of the spins @xcite . unknown coefficients in the ansatz can then be fit to numerical results , and correction terms can be added as needed for numerical agreement . with these considerations , a tentative formula has been taking shape through the combined efforts of the numerical - relativity community , of the following form @xcite : @xmath454 \nonumber\\ & + & \frac{k_s(q-1)\eta^2}{(q+1)^3 } \nonumber\\ & \times&\left[q^2\alpha^{\perp}_1 \cos(\phi_1-\phi_1)+\alpha^{\perp}_2\cos(\phi_2-\phi_2)\right ] \label{eq : v_parallel}\end{aligned}\ ] ] where @xmath455 is the contribution due to mass asymmetry , @xmath456 is the contribution due to spin that yields a kick perpendicular to the orbital angular momentum , @xmath457 is the contribution due to spin that yields a kick parallel to the orbital angular momentum , @xmath458 is the projection of the dimensionless spin vector @xmath459 of black hole @xmath134 along the orbital angular momentum , @xmath460 is the magnitude of its projection @xmath461 into the orbital plane , @xmath462 refers to the angle made by @xmath461 with respect to some reference angle in the orbital plane , and @xmath463 and @xmath464 are constants for a given mass ratio and initial separation . here the spins are considered to have been measured at some point before merger or ideally , arbitrarily close to merger . @xmath463 and @xmath464 encode the amount of precession of each spin before merger . one can then evaluate the formula over ranges of expected mass ratios and spin magnitudes , and for all spin orientations , to compute the probability of a given recoil speed . for example , for mass ratios between 1 and 10 , and spin magnitudes of 0.9 , the probability of exceeding @xmath59 is predicted to be @xmath216 . studies of various model - dependent speed probabilities are given by @xcite and @xcite . among the significant astrophysical consequences of gravitational - radiation recoil , growth rates of black holes can be affected . for example , the recoil of massive black holes imparting less than the escape velocity of their host galaxies tends to have regulatory effects on the amount of mass accreted @xcite . recoil velocities that exceed galactic escape velocities can impact massive black - hole growth in a different way , by reducing the chances of subsequent mergers @xcite . this in turn may modestly reduce the rate of coalescence events observable by lisa @xcite . the recoil might also have more directly observable consequences . some quasars are thought to originate from the coalescence of massive black holes during galactic mergers . quasars ejected from their host galaxies might therefore be expected . a study by @xcite found no evidence for such events , indicating that they are rare . however , recent observations of two rapidly moving extragalactic quasars @xcite are strong candidates for such recoiled black holes @xcite . the final spin of the merged remnant of a binary is also of astrophysical interest . it would be useful to know the probability of various spin magnitudes in constructing matched filtering templates and estimating gravitational signal detectability for ligo and lisa @xcite . the ability to predict a final spin given initial binary parameters also implies that observation of the spin of a black hole may help understand its origin . in particular , an understanding of the relationship between the final spin of a black hole and the orbital angular momentum of its binary precursor may help explain the formation of x - shaped jets @xcite . as with the kick , it would be advantageous to construct a simple analytic formula with which to estimate the spin . observations from numerical simulations suggest this might be possible . for example , in the case of nonspinning , equal - mass binaries undergoing circular inspiral , the final spin is insensitive to the initial separation , being determined by the terminal dynamics of the merger . such evolution results in a universal final spin of @xmath465 @xcite . additionally , for unequal - mass , nonspinning binaries , the final spin scales roughly linearly with symmetric mass ratio @xcite . various approaches to constructing an ansatz for the final spin have been proposed . for aligned - spin , equal - mass mergers , @xcite produced a simple formula quadratic in the ( common ) initial spin ; this satisfied the `` cosmic censorship '' hypothesis : @xmath466 . for nonspinning mergers , @xcite simply assumed @xmath467 , for fitting parameters @xmath13 and @xmath468 , which agreed with data from nonspinning binaries reasonably well . @xcite arrived at a more generic ansatz , for initial black holes of arbitrary spins , using the pn approximation . @xcite proposed a different expression for generic binaries , using a set of assumptions about the approximate conservation of the magnitudes and relative angles of certain angular momentum vectors . several other formulas have been suggested @xcite , each in good agreement with some subset of available numerical data . however , broad agreement on an analytic formula for the final spin from generically precessing binaries has yet to be achieved . we have seen above that black - hole mergers are expected to be `` loud , '' since they produce strong gravitational - wave signals . but will they also be `` bright ? '' that is , will there be an accompanying display of photons , detectable by telescopes in any frequency range of the full electromagnetic spectrum ? the answer to this question depends critically on the amount and distribution of gas and magnetic fields surrounding the merging binary . for stellar black - hole binaries , and intermediate - mass black - hole binaries in stellar clusters , any matter in an accretion disk would be consumed by one or both black holes and disappear relatively quickly , making an electromagnetic counterpart highly unlikely . massive black - hole binaries formed from galaxy mergers , however , present a very different situation . in a gas - rich or `` wet '' merger , there is likely enough gas available to feed accretion disks around each black hole that eventually evolve into a circumbinary disk ; this provides a source of gas and magnetic fields that could generate detectable electromagnetic emission . even in the case of gas - poor or `` dry '' mergers , the thin hot gas present is such ( elliptical ) galaxies might be sufficient to produce an electromagnetic signature . such electromagnetic signatures would be valuable to astrophysics . identification of the source on the sky would help confirm and characterize the merger , and probe accretion physics . a measurement of its redshift using electromagnetic radiation , taken together with the determination of luminosity distance using gravitational waves observed by lisa @xcite , would provide an independent calibration of the distance scale and a precise probe of cosmology including the nature of the mysterious dark energy @xcite . differences in arrival times of the electromagnetic and gravitational - wave signatures could also test fundamental principles such as the relative propagation speed of photons and gravitational waves . with the recent successes in numerical - relativity modeling of black - hole mergers , we find considerable interest in understanding possible electromagnetic signals from these events . most work to date has focused on mechanisms that could produce emission from a surrounding accretion disk , including signals induced by a recoiling merged black hole encountering the disk @xcite . moreover , explorations of possible electromagnetic signals that could arise in the dynamic spacetime near the binary during its last few orbits and merger are just now beginning . @xcite recently studied the effects of a merging black - hole binary on a surrounding magnetic field . the equal - mass , nonspinning holes start out with separation @xmath469 on quasicircular orbits somewhat outside the isco . the magnetic field is initially poloidal and assumed to be generated by currents in a distant circumbinary disk located at @xmath470 . the electric field around the binary is initially zero . there is no matter near the black holes , as predicted when the circumbinary disk is thin @xcite . they solved the coupled einstein - maxwell equations for a binary evolving in presence of externally sourced electromagnetic fields . as the system evolves , the inspiralling black holes stir up the fields . before the merger , from @xcite . the electric field lines are twisted around the black holes , while the magnetic field lines are mainly aligned with the @xmath34 axis . ] , except at a later time , @xmath471 before merger . ] figures [ fig : em_early ] and [ fig : em_later ] show the electric and magnetic field lines at @xmath472 and @xmath471 before the merger , respectively . the magnetic fields are mostly aligned with the @xmath34 axis , while the electric fields are twisted around the black holes . the binary dynamics induce oscillations in the electromagnetic energy flux , with a period approximately half of the dominant ( quadrupole ) gravitational - wave signal . the energy in the electromagnetic field is also enhanced gradually . @xcite have certainly made an interesting start on this important problem . can this scenario generate detectable electromagnetic emission ? the answer awaits more realistic simulations , including spinning black holes . other astrophysical scenarios allow both matter and magnetic fields in the vicinity of the binary close to merger . in this case , calculation of possible electromagnetic signatures requires solving the equations of general relativistic magnetohydrodynamics in a dynamically evolving spacetime governed by the einstein equations . @xcite took a step towards solving this problem by mapping the flow of pressureless matter , modeled as non - interacting point particles , in the dynamical spacetime around the merging black holes . they started with a distribution of particles around the black - hole binary , and then evolved the binary using numerical relativity while tracing the motion of the particles along geodesics as the binary evolves . to estimate the energetics of the flow , they detected `` collisions '' by looking for particles within a small distance ( typically , @xmath473 ) of each other and then calculating the lorentz factors @xmath474 in the center - of - mass frame of each collision . they began with equal - mass inspiralling black - hole binaries @xmath475 orbits before merger , and considered both black holes to have either zero spin , or @xmath476 with spin vectors aligned with the binary orbital angular momentum . a single black hole of mass @xmath9 is also evolved as a control case . approximately 75,000 geodesic particles are then initially distributed uniformly throughout a solid annulus centered on the binary and having inner radius @xmath477 , outer radius @xmath478 , and vertical full thickness @xmath479 . particles within the inner radius are excluded , to avoid transient signatures from particles initially near the horizons . note that these circumbinary disks are geometrically thick , and thus potentially have high enough inward radial speeds to keep up with the shrinking separation of the inspiralling binary , providing a source of gas near the black holes during their merger @xcite . @xcite studied two initial velocity configurations for the particles . in the `` orbital '' configuration , the initial velocities are randomly distributed around a tangential velocity @xmath480 that would give a circular orbit in a schwarzschild spacetime of mass @xmath9 , resulting in a scale height of @xmath481 . in contrast , the `` isotropic '' configuration is an extreme ( that is , very hot ) case in which the particles only have random velocities , with each component sampled from a gaussian distribution of standard deviation @xmath482 . for the orbital initial velocities , the particles remain mostly in a rotational configuration ; some particles do enter the region around the black holes , but are soon ejected by a gravitational slingshot . the collisions typically occur at relatively low velocities with lorentz factors @xmath483 . by the time of merger , there are essentially no particles near the black holes , and thus there is no energetic signature of the merger . the situation is quite different with the isotropic initial conditions , which provide a continual influx of particles throughout the evolution . moreover , there is a clear signature of the merger visible in the maximum lorentz factor , which takes on values @xmath484 during the inspiral and spikes up to @xmath485 just before merger for the nonspinning case . for rotating black holes , @xmath486 during the inspiral , spiking to @xmath487 just before merger . in addition , gravitational torques from the merging binary effectively stir the particle distribution , leading to high - velocity outflows with particles reaching @xmath488 or more . @xcite point out that , in realistic astrophysical disks , viscosity would cause angular momentum transport , bringing material inward towards the merging black holes . this could produce a scenario between the two extremes they studied , with a clear merger signal . more recent work by @xcite , using hydrodynamical simulations of gas around black - hole binaries , provided strong support in this regard . further confirmation of this interesting possibility , along with other mechanisms involving general relativistic magnetohydrodynamics , awaits more detailed simulations with gas and magnetic fields , currently in development . five years ago , we did not know what obstacles might conspire to prevent the effective application of numerical - relativity simulations in deriving predictions of general relativity to address questions of astrophysical black - hole binaries . up to that time , the numerical relativity researchers addressed their research primarily to questions of theoretical and computational physics . then , quite suddenly , the last obstacles vanished . as reported , numerical relativity can now be applied to understand general relativity s predictions of strong - field gravitational physics and to begin addressing questions of astrophysics and gravitational - wave data analysis . we expect future work in this field to be guided not by internal problems in gravitational theory , but by questions of astrophysics , and other areas where strong - field gravitational theory applies . these questions will motivate both continued , more detailed , investigation of some of the phenomena which have already been revealed , and the development of new capabilities to bring simulation results to bear on new questions . though the work has begun in applying numerical - relativity results to gravitational - wave observations , much more work is still needed . it is now clear that the signal - to - noise ( snr ) ratio from the last moments of the merger will dominate over the inspiral signal in many potential observations . though numerical - relativity simulations have shed considerable light on the basic features of the burst of gravitational radiation that completes the predicted signals , applications in data analysis will require _ comprehensive _ quantitative knowledge of the merger signals generated over the full parameter space of mergers . while we discussed examples of mergers that sample the binary black - hole parameter space along what may be its principal axes , a full quantification of the signal space will require a long systematic program of investigation . even with the aid of empirical models for encoding numerical results , many simulations must be conducted to qualify possible models . initial models suitable for detection and ( perhaps ) parameter estimation of low - snr signals , will require much further development for application to the high - snr observations expected with future instrument upgrades , such as advanced ligo , and future instruments such as lisa . considerably more simulations of high quality will be required to achieve these goals . in addition to understanding the signals generated by mergers through the bulk of parameter space in greater depth , relativists must also extend the region of parameter space covered by simulations . typical current simulations are limited in the range of parameters that can be practically covered . currently popular initial data models , particularly those assuming conformal flatness , are limited to @xmath489 @xcite . alternatives more suited to rapidly spinning kerr black holes are less well developed in other ways . while astrophysical limits on spin magnitudes remain unclear , it is likely that larger spins will have to be treated with possibly novel initial data models . other , related challenges in the simulations may also exist . current simulations are also restricted to comparable mass ratios . simulations beyond @xmath490 require somewhat heroic investments of computational resources with present techniques . astrophysical mergers may occur , however , over a broad range of mass ratios . in the limit of extremely large mass ratios , approximation techniques are possible that treat the motion of the smaller object as a perturbed geodesic in the spacetime of the larger black hole . such analytic methods look promising in this limit , where numerical simulations are less practical . however , there is a rather large middle ground , say @xmath491 , where new techniques may be required . one problem is that the small spatial scale of the smaller black hole requires extremely small timesteps for stable explicit time integrations , even though there is nothing interesting happening on these timescales . alternatives such as implicit schemes @xcite may open the door to a broader range of applications . most of the simulations discussed also focus on circular , or nearly circular orbits . while this is clearly an important portion of the black - hole - binary population , scenarios have been proposed in which one black hole captures another from nearly parabolic initial encounters . understanding the potential signals from such systems may require new techniques appropriate for long periods of slow , effectively newtonian evolution , punctuated by brief periods of strong gravitational interaction . as discussed , the numerical discovery of strong gravitational - wave recoils in asymmetric mergers of spinning black holes has had significant impact in areas of astrophysics beyond direct gravitational - wave observations . these results have spurred excitement in the black - hole astronomy community about other possible applications of numerical relativity . since other areas of astronomy are based on ( primarily ) electromagnetic observations , these interactions naturally lead to applications involving the interactions of black holes and electromagnetically visible matter . there are many relevant phenomena of interest : accretion disks around black holes , possibly disturbed by gravitational recoil , black - hole neutron - star binaries , neutron - star neutron - star binaries , jets from active galactic nuclei , quasars , and the mysterious origins of gamma ray bursts . a key direction for future work has been adding physics beyond the purely gravitational simulations we have reviewed here . as discussed , efforts are currently underway to use and develop magnetohydrodynamics in a dynamical spacetime . as such simulations are applied with increasing realism to more astrophysical scenarios , refinement of the techniques will be required . adequately resolving shocks and turbulence interacting with magnetic fields , achieving robust accuracy in the presence of very strong magnetic fields , and enforcing the vanishing divergence of the magnetic fields , all in curved spacetime , are among the challenges that need to be met . astrophysical mergers typically involve nearly circular orbits , or in extreme scenarios , initially parabolic encounters . numerical simulations can be applied to other sorts of black - hole interactions as well . for high - velocity black - hole encounters , one expects a phase transition dividing the spacetimes in which the black holes merge from those in which they pass each other hyperbolically , never to approach each other again : see fig . [ fig : crit_traj ] . numerical simulation studies are beginning to elucidate critical behavior near this phase transition @xcite . these fascinating phenomena at the extreme limit of strong gravitational dynamics have been discussed in the context of upcoming experimental measurements . in the trans - planckian energy limit , some particle collisions may be describable by classical black - hole dynamics @xcite . in highly speculative theories involving large extra spatial dimensions , it is conceivable that tev scale physics may be sufficient to produce such black - hole - like collisions @xcite . these possibilities motivate the first numerical - simulation studies of ultra - relativistic black - hole collisions @xcite . einstein gave us general relativity nearly 100 years ago . over the past century this theory , our standard model of gravitational physics , has passed experimental and observational tests over ranging from laboratory - scale physics experiments and solar - system tests to observations of compact astronomical objects and cosmology @xcite . these measurements have , so far , provided no need for a refinement of einstein s theory of gravity [ though some models to explain cosmic acceleration do involve alternative gravity @xcite ] . numerical simulations are now revealing the detailed predictions of einstein s theory for the final merger dynamics of a black - hole binary , and its record in the emitted gravitational radiation . gravitational - wave observations of these events will expose such gravitational phenomena to measurements in a strong - gravity regime far beyond anything which has previously been tested . these include the first tests of higher - order terms in the pn expansion and , indeed , of the physical predictions which numerical relativity is just now revealing . the possibility that general relativity will find its limits in these observations motivates a better understanding of what waveforms might be predicted by alternative theories of gravity @xcite . alternative - gravity models of black - hole mergers would also likely require numerical simulation to derive waveform predictions , but little work has been done in this direction so far [ see one related example from @xcite ] . if einstein s theory proves correct in predicting the detailed physics revealed by numerical relativity , it would stand as a truly incredible achievement of scientific induction , divining the details of phenomena vastly removed from physical observations on which the theory was founded . if not , then these observations , together with a confident understanding of einstein s predictions founded in numerical relativity simulations , may indeed lay the foundation for the next theory of gravity . this review draws on the work of a broad research community and would not have been possible without the many contributions and support of our colleagues . we especially want to thank a few individuals who have made particularly valuable contributions . william d. boggs , bernd brgmann , alessandra buonanno , mark hannam , richard matzner , cole miller , and harald pfeiffer gave insightful and helpful comments on our article . manuela campanelli and harald pfeiffer supplied us with figures from their simulations that were not in the published literature . we also benefited from many useful discussions with sean mcwilliams . we acknowledge support from nasa grant no . 06-befs06 - 19 . b.j.k . was supported in part by an appointment to the nasa postdoctoral program at the goddard space flight center , administered by oak ridge associated universities through a contract with nasa . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: in the hot field of computer simulation of biological macromolecules , available potential energy functions are often not accurate enough to properly describe complex processes such as the folding of proteins @xcite . in order to improve the situation , it is convenient to extract ab initio information from quantum mechanical calculations with the hope of being able to devise less computationally demanding methods that can be used to tackle large systems . in this spirit , the effective potential for the nuclei calculated in the non - relativistic born - oppenheimer approximation is typically considered as a good reference to assess the accuracy of cheaper potentials @xcite . the study of molecules at this level of theoretical detail and the design of computationally efficient approximations for solving the demanding equations that appear constitute the major part of the field called _ quantum chemistry _ @xcite . in this work , we voluntarily circumscribe ourselves to the basic formalism needed for the ground - state quantum chemical calculations that are typically performed in this context . for more general expositions , we refer the reader to any of the thorough accounts in refs . @xcite . in sec . [ sec : qc_molecular_h ] , we introduce the molecular hamiltonian and a special set of units ( the atomic ones ) that are convenient to simplify the equations . in sec . [ sec : qc_born - oppenheimer ] , we present in an axiomatic way the concepts and expressions related to the separation of the electronic and nuclear problems in the born - oppenheimer scheme . in sec . [ sec : qc_variational ] , we introduce the variational method that underlies the derivation of the basic equations of the hartree and hartree - fock approximations , discussed in sec . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] respectively . the computational implementation of the hartree - fock approximation is tackled in sec . [ sec : qc_roothaan ] , where the celebrated roothaan - hall equations are derived . in sec . [ sec : qc_basis_sets ] , the main issues related to the construction and selection of gaussian basis sets are discussed , and , finally , in sec . [ sec : qc_modern_developments ] , the hottest areas of modern research are briefly reviewed and linked to the issues in the rest of the work , with a special emphasis in the development of linear - scaling methods . since 1960 , the international scientific community has agreed on an ` official ' set of basic units for measurements : _ le ystme nternational dunits _ , or si for short ( see ` http://www.bipm.org/en/si/ ` and ref . @xcite ) . the meter ( m ) , the kilogram ( kg ) , the second ( s ) , the ampere ( a ) , the kelvin ( k ) , the mole ( mol ) , the joule ( j ) and the pascal ( pa ) are examples of si units . sticking to the si scheme , the non - relativistic quantum mechanical hamiltonian operator of a molecule consisting of @xmath0 nuclei ( with atomic numbers @xmath1 and masses @xmath2 , @xmath3 ) and @xmath4 electrons ( i.e. , the _ molecular hamiltonian _ ) is expressed as note that the non - relativistic molecular hamiltonian does not depend on spin - like variables . ] : @xmath5 where @xmath6 stands for @xmath7 , being @xmath8 planck s constant , @xmath9 denotes the electron mass , @xmath10 the proton charge , @xmath11 the position of the @xmath12-th electron , @xmath13 that of the @xmath14-th nucleus , @xmath15 the vacuum permittivity and @xmath16 the laplacian operator with respect to the coordinates of the @xmath12-th particle . although using a common set of units presents obvious communicative advantages , when circumscribed to a particular field of science , it is common to appeal to non - si units in order to simplify the most frequently used equations by getting rid of some constant factors that always appear grouped in the same ways and , thus , make the numerical values in any calculation of the order of unity . in the field of quantum chemistry , _ atomic units _ ( see table [ tab : atomic_units ] ) , proposed in ref . @xcite and named in ref . @xcite , are typically used . in these units , ( [ eq : chqc_ham_tot_si ] ) is substantially simplified to @xmath17 [ tab : atomic_units ] since all the relevant expressions in quantum chemistry are derived in one way or another from the molecular hamiltonian , the simplification brought up by the use of atomic units propagates to the whole formalism . consequently , they shall be the choice all throughout this work . [ tab : unit_conversion ] apart from the atomic units and the si ones , there are some other miscellaneous units that are often used in the literature : the _ ngstrm _ , which is a unit of length defined as , and the units of energy _ _ cm__@xmath18 ( which reminds about the spectroscopic origins of quantum chemistry and , even , quantum mechanics ) , _ electronvolt _ ( ev ) , _ kilocalorie per mole _ ( kcal / mol ) and _ kilojoule per mole _ ( kj / mol ) . the last two are specially used in the field of macromolecular simulations and quantify the energy of a mole of entities ; for example , if one asserts that the torsional barrier height for @xmath19 is @xmath20 kcal / mol , one is really saying that , in order to make a mole of @xmath19 ( i.e. , @xmath21 molecules ) rotate @xmath22 around the o o bond , one must spend @xmath20 kcal . for the conversion factors between the different energy units , see table [ tab : unit_conversion ] . finally , to close this section , we rewrite eq . ( [ eq : chqc_ham_tot_au ] ) introducing some self - explanatory notation that will be used in the subsequent discussion : [ eq : chqc_ham_tot_op ] @xmath23 to think of a macromolecule as a set of quantum objects described by a _ wavefunction _ @xmath24 dependent on the spatial and [ page : foot : spin_convenient ] spin one convenient way of thinking about functions that depend on spin - like variables is as an @xmath25-tuple of ordinary @xmath26 functions , where @xmath25 is the finite number of possible values of the spin . in the case of a one - particular wavefunction describing an electron , for example , @xmath27 can take two values ( say , @xmath28 and @xmath29 ) in such a way that one may picture any general _ spin - orbital _ @xmath30 as a 2-tuple @xmath31 . of course , another valid way of imagining @xmath30 is simply as a function of four variables , three real and one discrete . ] degrees of freedom , , of the electrons and on those of the nuclei , @xmath32 , would be too much for the imagination of physicists and chemists . all the language of chemistry would have to be remade and simple sentences in textbooks , such as `` rotation about this single bond allows the molecule to avoid steric clashes between atoms '' or even `` a polymer is a long chain - like molecule composed of repeating monomer units '' , would have to be translated into long and counter - intuitive statements involving probability and ` quantum jargon ' . conscious or not , we think of molecules as classical objects . more precisely , we are ready to accept that electrons are quantum ( we know of the interference experiments , electrons are light , we are accustomed to draw atomic ` orbitals ' , etc . ) , however , we are reluctant to concede the same status to nuclei . nuclei are heavier than electrons ( at least @xmath33 times heavier , in the case of the single proton nucleus of hydrogen ) and we picture them in our imagination as ` classical things ' that move , bond to each other , rotate around bonds and are at precise points at precise times . we imagine nuclei ` slowly moving ' in the field of the electrons , which , for each position of the first , immediately ` adjust their quantum state ' . the formalization of these ideas is called _ orn - ppenheimer _ ( bo ) _ approximation _ @xcite and the confirmation of its being good for many relevant problems is a fact that supports our intuitions about the topic and that lies at the foundations of the vast majority of the images , the concepts and the research in quantum chemistry there are many phenomena , however , in which the born - oppenheimer approximation is broken . for example , in striking a flint to create a spark , mechanical motion of the nuclei excites electrons into a plasma that then emits light @xcite . ] . like any approximation , the born - oppenheimer one may be either _ derived _ from the exact problem ( in this case , the entangled behaviour of electrons and nuclei as the same quantum object ) or simply _ proposed _ on the basis of physical intuition , and later confirmed to be good enough ( or not ) by comparison with the exact theory or with the experiment . of course , if it is possible , the first way should be preferred , since it allows to develop a deeper insight about the terms we are neglecting and the specific details that we will miss . however , although in virtually every quantum chemistry book @xcite hand - waving derivations up to different levels of detail are performed and the born - oppenheimer approximation is typically presented as unproblematic , it seems that the fine mathematical details on which these ` standard ' approaches are based are far from clear @xcite . this state of affairs does not imply that the final equations that will need to be solved are ill - defined or that the numerical methods based on the theory are unstable ; in fact , it is just the contrary ( see the discussion below ) , because the problems are related only to the precise relation between the concepts in the whole theory and those in its simplified version . nevertheless , the many subtleties involved in a _ derivation _ of the born - oppenheimer approximation scheme from the exact equations suggest that the second way , that of _ proposing _ the approximation , be taken . hence , in the following paragraphs , an _ axiomatic _ presentation of the main expressions , aimed mostly to fix the notation and to introduce the language , will be performed . first of all , if we examine the hamiltonian operator in eq . ( [ eq : chqc_ham_tot_au ] ) , we see that the term @xmath34 prevents the problem from being separable in the nuclear and electronic coordinates , i.e. , if we define @xmath35 as the set of all electronic coordinates ( spatial and spin - like ) and do likewise with the nuclear coordinates @xmath36 , the term @xmath34 prevents any wavefunction @xmath37 solution of the _ time - independent schrdinger equation _ , @xmath38 from being written as a product , @xmath39 , of an electronic wavefunction and a nuclear one . if this were the case , the problem would still be difficult ( because of the coulomb terms @xmath40 and @xmath41 ) , but we would be able to focus on the electrons and on the nuclei separately . the starting point for the born - oppenheimer approximation consists in assuming that a less strict separability is achieved , in such a way that , for a pair of suitably chosen @xmath42 and @xmath43 , any wavefunction solution of eq . ( [ eq : chqc_schrodinger ] ) ( or at least those in which we are interested ; for example , the eigenstates corresponding to the lowest lying eigenvalues ) can be expressed as @xmath44 where we have used a ` ; ' to separate the two sets of variables in the electronic part of the wavefunction in order to indicate that , in what follows , it is convenient to use the image that ` from the point of view of the electrons , the nuclear degrees of freedom are fixed ' , so that the electronic wavefunction depends ` parametrically ' on them . in other words , that the @xmath36 are not quantum variables in eq . ( [ eq : chqc_clamped ] ) below . of course , it is just a ` semantic ' semicolon ; if anyone feels uncomfortable about it , she may drop it and write a normal comma . notably , in ref . @xcite , hunter showed that any solution of the schrdinger equation can in fact be written exactly in the form of eq . ( [ eq : chqc_bo_wf ] ) , and that the two functions , @xmath42 and @xmath43 , into which @xmath37 is split may be interpreted as marginal and conditional probability amplitudes respectively . however , despite the insight that is gained from this treatment , it is of no practical value , since the knowledge of the exact solution @xmath37 is required in order to compute @xmath42 and @xmath43 in hunter s approach . in the born - oppenheimer scheme , an additional assumption is made in order to avoid this drawback : _ the equations obeyed by the electronic and nuclear parts of the wavefunction are supposed to be known_. hence , @xmath43 is assumed to be a solution of the time - independent _ clamped nuclei schrdinger equation _ , @xmath45 where the _ electronic hamiltonian operator _ @xmath46 and the _ electronic energy _ @xmath47 ( both dependent on the nuclei positions ) have been defined , and , since the nuclear spins do not enter the expression , we have explicitly indicated that @xmath48 depends parametrically on @xmath49 and not on @xmath36 . the common interpretation of the clamped nuclei equation is , as we have advanced at the beginning of the section , that the nuclei are much ` slower ' than the electrons and , therefore , the latter can automatically adjust their quantum state to the instantaneous positions of the former . physically , eq . ( [ eq : chqc_clamped ] ) is just the time - independent schrdinger equation of @xmath4 particles ( the electrons ) of mass @xmath9 and charge @xmath50 in the _ external electric field _ of @xmath0 _ point charges _ ( the nuclei ) of size @xmath51 at locations @xmath13 . mathematically , it is an eigenvalue problem that has been thoroughly studied in the literature and whose properties are well - known @xcite . in particular , it can be shown that , in the case of neutral or positively charged molecules ( i.e. , with @xmath52 ) , the clamped nuclei equation has an infinite number of normalizable solutions in the discrete spectrum of @xmath46 ( _ bound - states _ ) for every value of @xmath49 @xcite . these solutions must be regarded as the different _ electronic energy levels _ , and a further approximation that is typically made consists in , not only accepting that the electrons immediately ` follow ' nuclear motion , but also that , for each value of the nuclear positions @xmath49 , they are in the electronic ground - state this is customarily assumed in the literature and it is supported by the general fact that electronic degrees of freedom are typically more difficult to excite than nuclear ones . hence , in the vast majority of the numerical implementations of the theory , only the electronic ground - state is sought . we will see this in the forecoming sections . ] , i.e. , the one with the lower @xmath47 . consequently , we define @xmath53 to be the _ effective electronic field _ in which the nuclei move , in such a way that , once we have solved the problem in eq . ( [ eq : chqc_clamped ] ) and know @xmath54 , the time - independent _ nuclear schrdinger equation _ obeyed by @xmath42 is : @xmath55 where the _ effective nuclear hamiltonian _ @xmath56 has been implicitly defined . now , to close the section , we put together the main expressions of the born - oppenheimer approximation for quick reference and we discuss them in some more detail : [ eq : chqc_bo_summary ] @xmath57 to start , note that the above equations are written in the logical order in which they are imagined and used in any numerical calculation . first , we assume the nuclei fixed at @xmath49 and we ( hopefully ) solve the clamped nuclei electronic schrdinger equation ( eq . ( [ eq : chqc_bo_summary_a ] ) ) , obtaining the electronic ground - state @xmath58 with its corresponding energy @xmath59 . next , we repeat this procedure for all possible values of course , this can not be done in practice . due to the finite character of available computational resources , what is customarily done is to define a ` grid ' in @xmath49-space and compute @xmath54 in a finite number of points . ] of @xmath49 and end up with an hyper - surface @xmath54 in @xmath49-space . finally , we add this function to the analytical and easily computable @xmath60 and find the _ effective potential _ that determines the nuclear motion : @xmath61 it is , precisely , this effective potential that is called _ otential nergy urface _ ( pes ) ( or , more generally , _ otential nergy yper - urface _ ( pehs ) ) in quantum chemistry and that is the central object through which scientists picture chemical reactions or conformational changes of macromolecules @xcite . in fact , the concept is so appealing and the classical image so strong that , after ` going quantum ' , we can ` go classical ' back again and think of nuclei as perfectly classical particles that move in the classical potential @xmath62 . in such a case , we would not have to solve eq . ( [ eq : chqc_bo_summary_c ] ) but , instead , integrate the newtonian equations of motion . this is the basic assumption of every typical force field used for classical ground - state _ molecular dynamics _ , such as the ones in the popular charmm @xcite , amber @xcite or opls @xcite packages . finally , we would like to remind the reader that , despite the hand - waving character of the arguments presented , up to this point , every computational step has a clear description and eqs . ( [ eq : chqc_bo_summary_a ] ) through ( [ eq : chqc_bo_summary_c ] ) could be considered as _ definitions _ involving a certain degree of notational abuse . to assume that the quantities obtained through this process are close to those that proceed from a rigorous solution of the time independent schrdinger equation ( eq . ( [ eq : chqc_schrodinger ] ) ) is where the approximation really lies . hence , the more accurate eqs . ( [ eq : chqc_bo_summary_d ] ) are , the better the born - oppenheimer guess is , and , like any other one , if one does not trust in the heuristic grounds on which the final equations stand , they may be taken as axiomatic and judged a posteriori according to their results in particular cases until now , two approximations have been done : the non - relativistic character of the objects studied and the born - oppenheimer approximation . in the forecoming , many more will be done . the a priori quantification of their goodness in large molecules is a formidable task and , despite the efforts in this direction , in the end , the comparison with experimental data is the only sound method for validation . ] . in quantum chemistry , the born - oppenheimer approximation is assumed in a great fraction of the studies and it allows the central concept of potential energy surface to be well - defined , apart from considerably simplifying the calculations . the same decision is taken in this work . there exists a mathematically appealing way of deriving the time independent schrdinger equation ( eq . ( [ eq : chqc_schrodinger ] ) ) from an extremal principle . to this end , we first define the functional ( see appendix a ) that corresponds to the expected value of the energy , @xmath63 : = \langle\psi|\,\hat{h}\,|\psi\rangle \ , \ ] ] where the traditional _ bra _ and _ ket _ notation is read as @xmath64 @xmath65 being any operator in the space of wavefunctions and @xmath66 a dummy variable representing all possible coordinates on which @xmath67 depends . the norm of @xmath67 , in this notation is expressed as @xmath68 , and we shall say that @xmath67 is _ normalized _ if @xmath69 . if we want to optimize the energy functional above restricting the search space to the normalized wavefunctions , the constrained - extremals problem that results can be solved via the _ lagrange multipliers method _ ( see appendix b ) by constructing the _ associated functional _ @xmath70 $ ] , where we introduce a _ lagrange multiplier _ @xmath71 to force normalization : @xmath72 : = \mathcal{f}[\psi ] + \lambda \,\big ( \langle \psi|\psi\rangle - 1 \big ) \ .\ ] ] if we now ask that the functional derivative of @xmath70 $ ] with respect to the complex conjugate @xmath73 of the wave [ page : foot : deriving_conjugate ] function a function of a complex variable @xmath74 ( or , analogously , a functional on a space of complex functions ) may be regarded as depending on two different sets of independent variables : either @xmath75 and @xmath76 or @xmath74 and @xmath77 . the choice frequently depending on technical issues . ] be zero , i.e. , we look for the stationary points of @xmath78 $ ] conditioned by , we obtain the eigenvalues equation for @xmath79 , i.e. , the time - independent schrdinger equation . additionally , it can be shown , first , that , due to the self - adjointedness of @xmath79 , the equation obtained from the stationarity condition with respect to @xmath67 ( not @xmath73 ) is just the complex conjugate and adds no new information . moreover , one can see that the reverse implication is also true @xcite , so that , if a given normalized wavefunction @xmath67 is a solution of the eigenvalue problem and belongs to the discrete spectrum of @xmath79 , then the functional in eq . ( [ eq : chqc_functional_ftilde ] ) is stationary with respect to @xmath73 : @xmath80}{\delta\psi^ { * } } = 0 \quad \longleftrightarrow \quad \hat{h}\,\psi=-\lambda\,\psi:=e\,\psi \quad \mathrm{and } \quad \langle\psi|\psi\rangle = 1 \ .\ ] ] this result , despite its conceptual interest , is of little practical use , because it does not indicate an operative way to solve the schrdinger equation different from the ones that we already knew . the equivalence above simply illustrates that mathematical variational principles are over - arching theoretical statements from which the differential equations that actually contain the details of physical systems can be extracted . nevertheless , using similar ideas , we will derive another simple theorem which is indeed powerfully practical : the _ variational theorem_. let @xmath81 be a basis of eigenstates of the hamiltonian operator @xmath79 and @xmath82 their corresponding eigenvalues . since @xmath79 is self - adjoint , the eigenstates @xmath83 can be chosen to be orthonormal ( i.e. , @xmath84 ) and any normalized wavefunction @xmath67 in the hilbert space can be written as a linear combination of them we assume here , for the sake of simplicity and in order to highlight the relevant concepts , that @xmath79 has only discrete spectrum . the ideas involved in a general derivation are the same , but the technical details and the notation are more complicated @xcite . ] : @xmath85 if we now denote by @xmath86 the lowest @xmath87 ( i.e. , the energy of the ground - state ) its existence is not guaranteed : it depends on the particular potential in @xmath79 . however , for the physically relevant cases , there is indeed a minimum energy in the set @xmath82 . ] and calculate the expected value of the energy on an arbitrary state @xmath67 such as the one in eq . ( [ eq : chqc_decomposition ] ) , we obtain @xmath88 this simple relation is the _ variational theorem _ and it states that any wavefunction of the hilbert space has an energy larger than the one of the ground - state ( the equality can only be achieved if @xmath89 ) . however trivial this fact may appear , it allows a very fruitful ` everything - goes ' strategy when trying to approximate the ground - state in a difficult problem . if one has a procedure for finding a promising guess wavefunction ( called _ variational ansatz _ ) , no matter how heuristic , semi - empirical or intuitive it may be , one may expect that the lower the corresponding energy , the closer to the ground - state it is of course , this not necessarily so ( and , in any case , it depends on the definition of ` closer ' ) , since it could happen that the @xmath90 landscape in the constrained subset of the hilbert space in which the search is performed be ` rugged ' . in such a case , we may have very different wavefunctions ( say , in the sense of the @xmath91-norm ) with similar energies @xmath90 . the only ` direction ' in which one can be sure that the situation improves when using the variational procedure is the ( very important ) energetic one . that one is also moving towards better values of any other observable is , in general , no more than a _ bona fide _ assumption . ] this provides a systematic strategy for improving the test wavefunction which may take a number of particular forms . one example of the application of the variational theorem is to propose a family of normalized wavefunctions @xmath92 parametrically dependent on a number @xmath93 and calculate the @xmath93-dependent expected value of the energy note that , if the functions @xmath92 were not normalized , then we should deal with the constrained problem as in ( [ eq : chqc_functional_ftilde ] ) , or , equivalently , we could include a dividing _ overlap term _ @xmath94 in ( [ eq : chqc_e_theta ] ) . ] : @xmath95 then , one may use the typical tools of one - variable calculus to find the minimum of @xmath96 and thus make the best guess of the energy @xmath86 constrained to the family @xmath92 . if the ansatz is cleverly chosen , this estimate could be rather accurate , however , for large systems that lack symmetry , it is very difficult to write a good enough form for @xmath92 . when dealing with a large number of particles , there exists another protocol based on the variational theorem that will permit us to derive the hartree and hartree - fock equations for the electronic wavefunction @xmath48 ( see secs . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] , respectively ) . the first step is to devise a restricted way ( a function @xmath97 with no free parameters ) to express @xmath48 in terms of _ one - electron wavefunctions _ , also called _ orbitals _ and denoted in principle , there could be more orbitals than electrons , however , in both the hartree and hartree - fock applications of this formalism , the index @xmath98 runs , just like @xmath12 , from 1 to @xmath4 . ] by @xmath99 , thus reducing the search space to a ( typically small ) subset of the whole hilbert space : @xmath100 the second step consists in establishing a ( possibly infinite ) number of constraints on the one - electron functions actually , both restrictions ( the one at the level of the total wavefunction in eq . ( [ eq : chqc_constraint_total ] ) and the one involving the one - particle ones in eq . ( [ eq : chqc_constraint_mono ] ) ) are simply constraints ( see appendix b ) . the distinction is not fundamental but operative , and it also helps us to devise variational ansatzs separating the two conceptual playgrounds . ] , @xmath101 with these two ingredients , we can now write the lagrange functional that describes the constrained problem in terms of the orbitals @xmath102 ( see eq . ( [ eq : chqc_functional_ftilde ] ) ) : @xmath103 = \big\langle f\big(\{\psi_{a}(x_{i})\}\big ) \,\big|\ , \hat{h}\ , \big|\ , f\big(\{\psi_{a}(x_{i})\}\big ) \big\rangle + \sum_{k } \lambda_{k } l_{k}\big(\{\psi_{a}(x_{i})\}\big ) \ .\ ] ] finally , we take the derivatives of @xmath104 $ ] with respect to every @xmath105 ( normally , with respect to the complex conjugate @xmath106 , see footnote [ foot : deriving_conjugate ] in page ) and we ask each one to be zero ( see appendix a ) . this produces the final equations that must be solved in order to find the stationary one - electron orbitals . of course , these final equations may have multiple solutions . in the cases discussed in this work , there exist procedures to check that a particular solution ( found computationally ) is , not only stationary , but also minimal @xcite . however , to assure that it is , not only locally minimal , but also globally ( i.e. , that is _ optimal _ ) , could be , in general , as difficult as for any other multi - dimensional optimization problem @xcite . in the hartree and hartree - fock cases , discussed in secs . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] respectively , the aufbau principle and a clever choice of the starting guess constitute particular techniques intended to alleviate this problem . assuming the born - oppenheimer approximation ( see sec . [ sec : qc_born - oppenheimer ] and eqs . ( [ eq : chqc_bo_summary ] ) ) , the central problem that one must solve in quantum chemistry is _ to find the ground - state of the electronic hamiltonian for a fixed position _ @xmath49 _ _ of the nuclei _ _ since , from now on , we will only be dealing with the ` electronic problem ' , the notation has been made simpler by dropping superfluous subindices @xmath10 where there is no possible ambiguity . as a consequence , for example , the electronic hamiltonian is now denoted by @xmath79 , the electronic kinetic energy by @xmath107 and the electronic wavefunction by @xmath108 ( dropping the parametric dependence on @xmath49 in the same spirit ) . ] : @xmath109 as already remarked in sec . [ sec : qc_born - oppenheimer ] , this problem is well posed for neutral and positively charged molecules , and , in the same way in which the term @xmath34 prevented the total wavefunction to be a product of an electronic and a nuclear part , the term @xmath41 in the expression above breaks the separability in the one - electron variables @xmath110 of the electronic time - independent schrdinger equation associated to @xmath79 . hence , a general solution @xmath108 can not be a product of orbitals and the search must be a priori performed in the whole hilbert space . however , this is a much too big place to look for @xmath108 , since the computational requirements to solve the schrdinger equation grow exponentially on the number of electrons . partially recognizing this situation , in the first days of quantum mechanics , dirac wrote that , _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the underlying physical laws necessary for the mathematical theory of a large part of physics and the whole of chemistry are thus completely known , and the difficulty is only that the exact application of these equations leads to equations much too complicated to be soluble . it therefore becomes desirable that approximate practical methods of applying quantum mechanics should be developed , which can lead to an explanation of the main features of complex atomic systems without too much computation @xcite . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the description of the most popular approximate methods , which the great physicist envisaged to be necessary , will be the objective of the following sections . two basic points responsible of the relative success of such an enterprise are the severe reduction of the space in where the ground - state is sought ( which , of course , leads to only an approximation of it ) and the availability of computers unimaginably faster than anything that could be foreseen in times of dirac . one of the first and most simple approximations aimed to solve the problem posed in the previous section is due to hartree in 1927 @xcite ( although the way in which the hartree equations will be derived here , using the variational theorem , is due to slater @xcite ) . in this approximation , the total wavefunction is constrained to be a product ( typically referred to as _ hartree product _ ) of @xmath4 one - electron orbitals ( see eq . ( [ eq : chqc_constraint_total ] ) ) , where the spin of the electrons and the antisymmetry ( i.e. , the pauli exclusion principle ) are not taken into account we shall denote with capital greek letters the wavefunctions depending on all the electronic variables , and with lowercase greek letters the one - electron orbitals . in addition , by @xmath67 ( or @xmath111 ) , we shall indicate wavefunctions containing spin part ( called _ spin - orbitals _ ) and , by @xmath112 ( or @xmath113 ) , those that depend only on spatial variables . ] : @xmath114 where the @xmath98 index in the orbitals has been substituted by @xmath12 due to the fact that each function is paired to a specific set of electron coordinates , consequently being the same number of both of them . also , the additional requirement that the one - particle wavefunctions be normalized is imposed ( see eq . ( [ eq : chqc_constraint_mono ] ) ) : @xmath115 with these two ingredients , we can construct the auxiliary functional whose zero - derivative condition produces the solution of the constrained stationary points problem ( see eq . ( [ eq : chqc_constrained_functional ] ) ) . to this effect , we introduce @xmath4 lagrange multipliers @xmath116 that force the normalization constraints note that the normalization of the total wavefunction is a consequence of the normalization of the one - electron ones and needs not to be explicitly asked . ] : @xmath117 = \left \langle \prod_{i=1}^{n}\phi_{i}({\bm r}_{i } ) \right | \hat{h } \left | \prod_{i=1}^{n}\phi_{i}({\bm r}_{i } ) \right \rangle - \sum_{i=1}^{n } \varepsilon_{i } \big ( \langle\phi_{i}|\phi_{i}\rangle-1 \big ) \ , \ ] ] where the minus sign in the lagrange multipliers term is chosen in order to get to the most common form of the final equations without having to define new quantities this functional may be considered to depend on @xmath118 independent functions : the @xmath4 one - electron @xmath119 and their @xmath4 complex conjugates ( see footnote [ foot : deriving_conjugate ] in page ) . the _ hartree equations _ are then obtained by imposing that the functional derivative of @xmath120 with respect to @xmath121 be zero for @xmath122 . in order to obtain them and as an appetizer for the slightly more complicated process in the more used hartree - fock approximation , the functional derivative will be here computed in detail following the steps indicated in appendix a. first , we write out the limits in sums and products are dropped if there is no possible ambiguity . ] the first term in the right - hand side of eq . ( [ eq : chqc_constrained_functional_h ] ) : @xmath123 where @xmath124 denotes the euclidean @xmath125 volume element @xmath126 . now , we realize that the products outside the integrals can be dropped using the constraints in eq . ( [ eq : chqc_constraint_mono_h ] ) ( see the last paragraphs of appendix b for a justification that this can be done before taking the derivative ) . then , using the previous expression and conveniently rearranging the order of the integrals and sums , we write out the first term in the numerator of the left - hand side of eq . ( [ eq : def_funct_der ] ) that corresponds to an infinitesimal variation of the function @xmath121 : @xmath127:= } \nonumber \\ \lefteqn { \widetilde{\mathcal{f}}\,\big[\phi_{1},\phi_{1}^{*},\ldots , \phi_{k},\phi_{k}^{*}+ \epsilon \delta \phi_{k}^{*},\ldots , \phi_{n},\phi_{n}^{*}\big ] = } \nonumber \\ & & \mbox { } -\frac{1}{2}\sum_{i } \int\phi_{i}^{*}({\bm r})\nabla^{2}\phi_{i}({\bm r})\,\mathrm{d}{\bm r } - \sum_{i } \int\big|\phi_{i}({\bm r})\big|^{2 } \left ( \sum_{a=1}^{n_{n}}\frac{z_{a}}{|{\bm r}-{\bm r}_{a}| } \right ) \mathrm{d}{\bm r } \nonumber \\ & & \mbox { } + \frac{1}{2}\sum_{i}\int\big|\phi_{i}({\bm r})\big|^{2 } \left ( \int\frac{\sum_{j \ne i}\big|\phi_{j}({\bm r}\ , ' ) \big|^{2 } } { |{\bm r}-{\bm r}\ , ' |}\,\mathrm{d}{\bm r}\ , ' \right ) \mathrm{d}{\bm r } - \sum_{i}\varepsilon_{i } \left ( \int\big|\phi_{i}({\bm r})\big|^{2 } \mathrm{d}{\bm r } - 1 \right ) \nonumber \\ & & \mbox { } -\frac{1}{2}\epsilon\int\delta\phi_{k}^{*}({\bm r})\nabla^{2 } \phi_{k}({\bm r})\,\mathrm{d}{\bm r } - \epsilon\int\delta\phi_{k}^{*}({\bm r})\,\phi_{k}({\bm r } ) \left ( \sum_{a=1}^{n_{n}}\frac{z_{a}}{|{\bm r}-{\bm r}_{a}| } \right ) \mathrm{d}{\bm r } \nonumber \\ & & \mbox { } + \epsilon\int\delta\phi_{k}^{*}({\bm r})\,\phi_{k}({\bm r } ) \left ( \int\frac{\sum_{i \ne k}\big|\phi_{i}({\bm r}\ , ' ) \big|^{2 } } { |{\bm r}-{\bm r}\ , ' |}\mathrm{d}{\bm r}\ , ' \right ) \mathrm{d}{\bm r } -\epsilon\,\varepsilon_{k}\int\delta\phi_{k}^{*}({\bm r})\,\phi_{k}({\bm r } ) \,\mathrm{d}{\bm r } \ .\end{aligned}\ ] ] we subtract from this expression the quantity @xmath128 $ ] , so that the first four terms cancel , and we can write @xmath129 - \widetilde{\mathcal{f}}\,\big[\phi_{k}^{*}\big]}{\epsilon } = } \nonumber \\ & & \int \bigg { [ } -\frac{1}{2 } \nabla^{2 } \phi_{k}({\bm r } ) - \left ( \sum_{a=1}^{n_{n}}\frac{z_{a } } { |{\bm r}-{\bm r}_{a}| } \right ) \phi_{k}({\bm r } ) \nonumber \\ & & \mbox { } + \left ( \int\frac{\sum_{i \ne k}|\phi_{i}({\bm r}\ , ' ) |^{2 } } { |{\bm r}-{\bm r}\ , ' |}\,\mathrm{d}{\bm r}\ , ' \right ) \phi_{k}({\bm r } ) - \varepsilon_{k}\phi_{k}({\bm r } ) \bigg { ] } \,\delta\phi_{k}^{*}({\bm r } ) \,\mathrm{d}{\bm r } \ .\end{aligned}\ ] ] now , by simple inspection of the right - hand side , we see that the functional derivative ( see eq . ( [ eq : def_funct_der ] ) ) is the part enclosed by square brackets : @xmath130 } { \delta \phi_{k}^{*}}= \left ( - \frac{1}{2}\nabla^{2 } + \hat{v}_{e}({\bm r } ) + \hat{v}_{e}^{k}({\bm r } ) - \varepsilon_{k } \right ) \phi_{k}({\bm r } ) \ , \ ] ] where the _ nuclear potential energy _ and the _ electronic potential energy _ have been respectively defined as compare the notation with the one in eqs . ( [ eq : chqc_ham_tot_op ] ) , here a subindex @xmath10 has been dropped to distinguish the new objects defined . ] [ eq : chqc_potentials_h ] @xmath131 finally , if we ask the functional derivative to be zero for @xmath122 , we arrive to the equations that the stationary points must satisfy , the _ hartree equations _ : @xmath132 \ , \phi_{k}({\bm r } ) : = \left ( - \frac{1}{2}\nabla^{2 } + \hat{v}_{n}({\bm r } ) + \hat{v}_{e}^{k}({\bm r } ) \right ) \phi_{k}({\bm r } ) = \varepsilon_{k}\,\phi_{k}({\bm r})\ , \ ] ] for all @xmath122 . let us note that , despite the fact that the object @xmath133 $ ] defined above is not a operator strictly speaking , since , as the notation emphasizes , it depends on the orbitals @xmath134 , we will stick to the name _ hartree operator _ for it , in order to be consistent with most of the literature . now , some remarks related to the hartree equations are worth making . first , it can be shown that , if the variational ansatz in eq . ( [ eq : chqc_constraint_total_h ] ) included the spin degrees of freedom of the electrons , all the expressions above would be kept , simply changing the orbitals @xmath135 by the spin - orbitals @xmath136 . secondly , and moving into more conceptual playgrounds , we note that the special structure of @xmath137 in eq . ( [ eq : chqc_potentials_h_b ] ) makes it mandatory to interpret the hartree scheme as one in which each electron ` feels ' only the average effect of the rest . in fact , if the _ quantum charge density _ @xmath138 is regarded for a moment as a classical continuum distribution , then the potential produced by all the electrons but the @xmath139-th is precisely the one in eq . ( [ eq : chqc_potentials_h_b ] ) . supporting this image , note also the fact that , if we write the joint probability density of electron 1 being at the point @xmath140 , electron @xmath141 being at the point @xmath142 and so on ( simply squaring eq . ( [ eq : chqc_constraint_total_h ] ) ) , @xmath143 we see that , _ in a probabilistic sense _ , the electrons are independent ( they could not be independent in a physical , complete sense , since we have already said that they ` see ' each other in an average way ) . anyway , despite these appealing images and also despite the fact that , disguised under the misleading ( albeit common ) notation , these equations seem ` one - particle ' , they are rather complicated from a mathematical point of view . on the one hand , it is true that , whereas the original electronic schrdinger equation in ( [ eq : chqc_bo_summary_a ] ) depended on @xmath144 spatial variables , the expressions above only depend on 3 . this is what we have gained from drastically reducing the search space to the set of hartree products in eq . ( [ eq : chqc_constraint_total_h ] ) and what renders the approximation tractable . on the other hand , however , we have paid the price of greatly increasing the mathematical complexity of the expressions , so that , while the electronic schrdinger equation was one linear differential equation , the hartree ones in ( [ eq : chqc_hartree_eqs ] ) are @xmath4 coupled non - linear integro - differential equations @xcite . this complexity precludes any analytical approach to the problem and forces us to look for the solutions using less reliable iterative methods . typically , in computational studies , one proposes a _ starting guess _ for the set of @xmath4 orbitals @xmath145 ; with them , the hartree operator @xmath146 $ ] in the left - hand side of eq . ( [ eq : chqc_hartree_eqs ] ) is constructed for every @xmath139 and the @xmath4 equations are solved as simple eigenvalue problems . for each @xmath139 , the @xmath147 that corresponds to the lowest @xmath148 is selected and a new hartree operator @xmath149 $ ] is constructed with the @xmath150 . the process is iterated until ( hopefully ) the @xmath151-th set of solutions @xmath152 differ from the one @xmath153 less than a reasonably small amount . many technical issues exist that raise doubts about the possible success of such an approach . the most important ones being related to the fact that a proper definition of the _ hartree problem _ should be : _ find the global minimum of the energy functional @xmath154 under the constraint that the wavefunction @xmath112 be a hartree product _ and not : _ solve the hartree equations _ ( [ eq : chqc_hartree_eqs ] ) , whose solutions indeed include the global minimum sought but also all the rest of stationary points . while the possibility that a found solution be a maximum or a saddle point can be typically ruled out @xcite , as we remarked in sec . [ sec : qc_variational ] and due to the fact that there are an infinite number of solutions to the hartree equations @xcite , to be sure that any found minimum is the global one is impossible in a general case . there exists , however , one way , related to a theorem by simon and lieb @xcite , of hopefully biasing a particular found solution of the hartree equations to be the global minimum that we are looking for . they showed that , first , for neutral or positively charged molecules ( @xmath155 ) , the hartree global minimization problem has a solution ( its uniqueness is not established yet @xcite ) and , second , that the minimizing orbitals @xmath156 correspond to the lowest eigenvalues of the @xmath133 $ ] operators self - consistently constructed with them in quantum chemistry , where the number of electrons considered is typically small , the version of the hartree equations that is used is the one derived here , with the hartree operators depending on the index @xmath139 in a non - trivial way . however , if the number of electrons is large enough ( such as in condensed matter applications ) , is customary to add to the effective electronic repulsion in eq . ( [ eq : chqc_potentials_h_b ] ) the self - interaction of electron @xmath139 with himself . in such a case , the hartree operator is independent of @xmath139 so that , after having achieved self - consistency , the orbitals @xmath157 turn out to be eigenstates corresponding to different eigenvalues of the same hermitian operator , @xmath158 $ ] , and , therefore , mutually orthogonal . ] . now , although _ the reverse of the second part of the theorem is not true in general _ ( i.e. , from the fact that a particular set of orbitals are the eigenstates corresponding to the lowest eigenvalues of the associated hartree operators , does not necessarily follow that they are the ones that minimize the energy ) @xcite , in practice , the insight provided by lieb and simon s result is invoked to build each successive state in the iterative procedure described above , choosing the lowest lying eigenstates each time . in this way , although one can not be sure that the global minimum has been reached , the fact that the found one has a property that the former also presents is regarded as a strong hint that it must be so ( see also the discussion for the hartree - fock case in the next section ) . this drawback and all the problems arising from the fact that an iterative procedure such as the one described above could converge to a fixed point , oscillate eternally or even diverge , are circumvented in practice by a clever choice of the starting guess orbitals @xmath145 . if they are extracted , for example , from a slightly less accurate theory , one may expect that they could be ` in the basin of attraction ' of the true hartree minimum ( so that the stationary point found will be the correct one ) and close to it ( so that the iterative procedure will converge ) . this kind of wishful thinking combined with large amounts of heuristic protocols born from many decades of trial - and - error - derived knowledge pervade and make possible the whole quantum chemistry discipline . the hartree theory discussed in the previous section is not much used in quantum chemistry and many textbooks on the subject do not even mention it . although it contains the seed of almost every concept underlying the _ hartree - fock approximation _ discussed in this section , it lacks an ingredient that turns out to be essential to correctly describe the behaviour of molecular species : _ the indistinguishability of the electrons_. this was noticed independently by fock @xcite and slater @xcite in 1930 , and it was corrected by proposing a variational ansatz for the total wavefunction that takes the form of a so - called _ slater determinant _ ( see eq . ( [ eq : chqc_constraint_total_sd_ghf ] ) below ) . the most important mathematical consequence of the indistinguishability among a set of @xmath4 quantum objects of the same type is the requirement that the total @xmath4-particle wavefunction must either remain unchanged ( _ symmetric _ ) or change sign ( _ antisymmetric _ ) when any pair of coordinates , @xmath110 and @xmath159 , are swapped . in the first case , the particles are called _ bosons _ and must have integer spin , while in the second case , they are called _ fermions _ and have semi - integer spin . electrons are fermions , so the total wavefunction must be antisymmetric under the exchange of any pair of one - electron coordinates . this is a property that is certainly not met by the single hartree product in eq . ( [ eq : chqc_constraint_total_h ] ) but that can be easily implemented by forming linear combinations of many of them . the trick is to add all the possible hartree products that are obtained from eq . ( [ eq : chqc_constraint_total_h ] ) changing the order of the orbitals labels while keeping the order of the coordinates ones it is immaterial whether the orbitals labels are kept and the coordinates ones changed or vice versa . ] , and assigning to each term the sign of the permutation @xmath160 needed to go from the natural order @xmath161 to the corresponding one @xmath162 . the sign of a permutation @xmath160 is 1 if @xmath160 can be written as a composition of an even number of two - element transpositions , and it is @xmath163 if the number of transpositions needed is odd . therefore , we define @xmath164 as the minimum number it can be shown that the parity of all decompositions of @xmath160 into products of elementary transpositions is the same . we have chosen the minimum only for @xmath164 to be well defined . ] of transpositions needed to perform the permutation @xmath160 , and we write the sign of @xmath160 as @xmath165 . using this , an antisymmetric wavefunction constructed from hartree products of @xmath4 different orbitals may be written as @xmath166 where the factor @xmath167 enforces normalization of the total wavefunction @xmath67 ( if we use the constraints in eq . ( [ eq : chqc_constraint_mono_ghf ] ) ) and @xmath168 denotes the _ symmetric group _ of order @xmath4 , i.e. , the set of all permutations of @xmath4 elements ( with a certain multiplication rule ) . the above expression is more convenient to perform the calculations that lead to the hartree - fock equations , however , there is also a compact way of rewriting eq . ( [ eq : chqc_constraint_total_ghf ] ) which is commonly found in the literature and that is useful to illustrate some particular properties of the problem . it is the _ slater determinant _ : @xmath169 now , having established the constraints on the form of the total wavefunction , we ask the hartree - fock one - electron orbitals to be , not only normalized , like we did in the hartree case , but also mutually orthogonal : @xmath170 additionally note that , contrarily to what we did in the previous section , we have now used one - electron wavefunctions @xmath171 dependent also on the spin @xmath27 ( i.e. , spin - orbitals ) to construct the variational ansatz . general _ spin - orbital note that , if we had not included the spin degrees of freedom , the search space would have been half as large , since , where we now have @xmath118 functions of @xmath172 ( i.e. , @xmath173 and @xmath174 , with @xmath175 ) , we would have had just @xmath4 ( the @xmath176 ) . ] may be written as ( see also footnote [ foot : spin_convenient ] in page ) @xmath177 where the functions @xmath14 and @xmath178 correspond to the _ spin - up _ and _ spin - down _ eigenstates of the operator associated to the @xmath74-component of the one - electron spin . they are defined as @xmath179 the formalism obtained when these general spin - orbitals are used is accordingly called _ eneral artree - ock _ ( ghf ) . the first part of the mathematical treatment in the following paragraphs shall be performed assuming this situation . the advantage of such a choice is that , later on , by imposing additional constraints to the spin part of the one - electron orbitals , we will be able to derive the basic equations for some other flavours of the hartree - fock theory , such as uhf , rhf and rohf , in a very direct way . now , to calculate the expected value of the energy in a state such as the one in eqs . ( [ eq : chqc_constraint_total_ghf ] ) and ( [ eq : chqc_constraint_total_sd_ghf ] ) , let us denote the one - particle part ( that operates on the @xmath12-th coordinates ) of the total electronic hamiltonian @xmath79 in eq . ( [ eq : chqc_el_ham_1 ] ) by @xmath180 in such a way that , @xmath181 where @xmath182 . [ page : one_e_calculation ] we shall compute separately each one of the sums in the expression above . let us start now with the first one : for a given @xmath12 in the sum , and due to the structure of the electronic wavefunction in ( [ eq : chqc_constraint_total_ghf ] ) , the expected value @xmath183 is a sum of @xmath184 terms of the form @xmath185 but , since @xmath186 operates only on @xmath187 and due to the orthogonality of the spin - orbitals with different indices , we have that the only non - zero terms are those with @xmath188 . taking this into account , all permutations @xmath160 appear still as terms of the sum , and we see that every orbital @xmath189 occurs depending on every coordinate @xmath187 . given a particular pair @xmath12 and @xmath190 , this happens in the terms for which @xmath191 and one of such terms may be expressed as @xmath192 where we have used that @xmath193 , and we have dropped the index @xmath12 from @xmath186 noticing that the integration variables in @xmath194 are actually dummy . next , we use again the one - electron wavefunctions constraints in eq . ( [ eq : chqc_constraint_mono_ghf ] ) to remove the product of norms in brackets , and we realize that , for each @xmath190 , there are as many terms like the one in the expression above as permutations of the remaining @xmath195 orbital indices ( i.e. , @xmath196 ) . in addition , we recall that every @xmath190 must appear and perform the first sum in eq . ( [ eq : chqc_expected_h_ghf ] ) , yielding @xmath197 where all factorial terms have canceled out . the next step is to calculate the second sum in eq . ( [ eq : chqc_expected_h_ghf ] ) . again , we have that , for each pair @xmath198 , @xmath199 is a sum of @xmath184 terms like @xmath200 [ page : two_e_calculation ] for this expected value , contrarily to the case of @xmath186 and due to the two - body nature of the operator @xmath201 , not only do the terms with @xmath188 survive , but also those in which @xmath160 and @xmath202 differ over only a pair of values @xmath12 and @xmath190 , i.e. , those for which @xmath203 , @xmath204 and @xmath205 . the reason for this is that , even if @xmath206 and @xmath207 , the integral @xmath208 does not vanish . now , using that @xmath201 operates only on @xmath187 and @xmath209 , the orthonormality conditions in eq . ( [ eq : chqc_constraint_mono_ghf ] ) and the fact that @xmath193 , we have that , when @xmath210 depends on @xmath187 and @xmath211 depends on @xmath209 , the @xmath188 part of the corresponding terms in eq . ( [ eq : chqc_calculation_rij_1 ] ) reads @xmath212 where we have defined @xmath213 next , we see that , for each pair @xmath214 , and keeping @xmath215 , we can make @xmath216 permutations among the @xmath217 indices of the orbitals on which @xmath201 does not operate and still find the same expression ( [ eq : chqc_calculation_rij_2 ] ) . therefore , for each pair @xmath214 , we have a sum of @xmath216 identical terms . in addition , if we perform the sum on @xmath12 and @xmath190 in eq . ( [ eq : chqc_expected_h_ghf ] ) and remark that the term in eq . ( [ eq : chqc_calculation_rij_2 ] ) does not depend on the pair @xmath198 ( which is obvious from the suggestive notation above ) , we have that the @xmath188 part of the second sum in eq . ( [ eq : chqc_expected_h_ghf ] ) , which is typically called _ coulomb energy _ , reads @xmath218 where we have used that the sum @xmath219 is performed on @xmath220 identical terms which do not depend neither on @xmath12 nor on @xmath190 . on the other hand , in the case in which @xmath160 and @xmath202 only differ in that the indices of the orbitals that depend on @xmath110 and @xmath159 are swapped , all the derivation above applies except for the facts that , first , @xmath221 and , second , the indices @xmath139 and @xmath222 must be _ exchanged _ in eq . ( [ eq : chqc_calculation_rij_3 ] ) ( it is immaterial if they are exchanged in the bra or in the ket , since the indices are summed over and are dummy ) . henceforth , the remaining part of the second sum in eq . ( [ eq : chqc_expected_h_ghf ] ) , typically termed _ exchange energy _ , may be written as @xmath223 finally , the expected value of the energy in the ghf variational state @xmath67 turns out to be @xmath224 where the _ one - electron integrals _ @xmath225 have been defined together with the _ two - electron integrals _ , @xmath226 and @xmath227 , and the fact that @xmath228 has been used to include the diagonal terms in the second sum . now , the energy functional above is the quantity that we want to minimize under the orthonormality constraints in eq . ( [ eq : chqc_constraint_mono_ghf ] ) . so we are prepared to write the auxiliary functional @xmath120 , introducing @xmath229 lagrange multipliers @xmath230 ( see eq . ( [ eq : chqc_constrained_functional ] ) and compare with the hartree example in the previous section ) : @xmath231 = \sum_i h_i + \frac{1}{2 } \sum_{i , j } ( j_{ij}-k_{ij } ) - \sum_{i , j } \lambda_{ij}\ , \big ( \langle\psi_{i}|\psi_{j}\rangle-\delta_{ij } \big ) \ .\ ] ] in order to get to the _ hartree - fock equations _ that the stationary orbitals @xmath210 must satisfy , we impose that the functional derivative of @xmath232 $ ] with respect to @xmath233 be zero . to calculate @xmath234 , we follow the procedure described in appendix a , using the same notation as in eq . ( [ eq : chqc_h_step_2 ] ) . the variation with respect to each @xmath233 shall yield the hartree - fock equations for the unconjugated @xmath171 . the equations for the @xmath235 are obtained either by differentiating @xmath232 $ ] with respect to each @xmath210 or , if the @xmath230 matrix is hermitian ( which will turn out to be the case ) , by simply taking the complex conjugate of both sides of the final equations in ( [ eq : chqc_hf_eqs_v1_ghf ] ) . now , @xmath236 - \widetilde{\mathcal{f}}\,\big[\psi_{k}^{*}\big]}{\epsilon } = } \nonumber \\ & & \langle\delta\psi_{k}|\,\hat{h}\,|\psi_{k}\rangle + \sum_j\bigg(\langle\delta\psi_k\psi_j|\,\frac{1}{r}\,|\psi_k\psi_j\rangle - \langle\delta\psi_k\psi_j|\,\frac{1}{r}\,|\psi_j\psi_k\rangle\bigg ) - \sum_j \lambda_{kj } \langle\delta\psi_{k}|\psi_{j}\rangle = \nonumber \\ & & \int \bigg { [ } \hat{h}\,\psi_{k}(x ) + \sum_j \left ( \psi_k(x ) \!\ ! \int \frac{|\psi_{j}(x^{\prime})|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } - \psi_j(x ) \!\ ! \int \frac{\psi^{*}_{j}(x^{\prime})\,\psi_{k}(x^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \right ) \nonumber \\ & & \mbox { } - \sum_j \lambda_{kj}\,\psi_{j}(x ) \bigg { ] } \,\delta\psi_{k}^{*}(x ) \,\mathrm{d}x \ , \end{aligned}\ ] ] where we have used the more compact notation @xmath237 instead of @xmath238 . then , like in the previous section , by simple inspection of the right - hand side , we see that the functional derivative is the part enclosed by square brackets ( see eq . ( [ eq : def_funct_der ] ) ) : @xmath239 } { \delta \psi_{k}^{*}}= \left[\hat{h } + \sum_j \big(\hat{j}_{j}[\psi]- \hat{k}_{j}[\psi]\big)\right]\psi_{k}(x)- \sum_j\lambda_{kj}\psi_{j}(x ) \ , \ ] ] where the _ coulomb _ and _ exchange operators _ are respectively defined by their action on an arbitrary function @xmath240 as follows like in the hartree case in the previous section , the word _ operator _ is a common notational abuse if they act upon the very @xmath171 on which they depend . this is again made explicit in the notation . ] : [ eq : chqc_operators_ghf ] @xmath241\,\varphi(x):=\left(\int\frac{|\psi_{j}(x^{\prime})|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \right ) \varphi(x)\ , \label{eq : chqc_operators_ghf_a } \\ & \hat{k}_{j}[\psi]\,\varphi(x):= \left ( \int\frac{\psi^{*}_{j}(x^{\prime})\,\varphi(x^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \right ) \psi_{j}(x)\ . \label{eq : chqc_operators_ghf_b } \end{aligned}\ ] ] therefore , if we define the ghf _ fock operator _ as @xmath242:= \hat{h}+\sum_j \big(\hat{j}_{j}[\psi]-\hat{k}_{j}[\psi]\big ) \ , \ ] ] we arrive to a first version of the _ hartree - fock equations _ by asking that the functional derivative in eq . ( [ eq : chqc_funct_der_ghf ] ) be zero : @xmath243\,\psi_{i}(x)=\sum_j\lambda_{ij}\psi_{j}(x)\ , \qquad i=1,\ldots , n\ .\ ] ] now , in order to obtain a simpler version of them , we shall take profit from the fact that the whole problem is invariant under a unitary transformation among the one - electron orbitals . if we repeat the calculation in eq . ( [ eq : chqc_limit_ghf ] ) but varying @xmath210 this time , instead of @xmath244 , and use the following relation : @xmath245\psi_j(x)\mathrm{d}x \ , \ ] ] we arrive to the ghf equations for the conjugated spin - orbitals : @xmath246\,\psi^*_{i}(x)=\sum_j\lambda_{ji}\psi^*_{j}(x)\ . \qquad i=1,\ldots , n\ .\ ] ] then , we may subtract the complex conjugate of eq . ( [ eq : chqc_hf_eqs_v1_ghf_conj ] ) from eq . ( [ eq : chqc_hf_eqs_v1_ghf ] ) yielding @xmath247 therefore , since the set of the @xmath189 is orthogonal and hence linearly independent , we have that the @xmath248 matrix @xmath249 of lagrange multipliers is hermitian : @xmath250 this actually means that we have a set of three equations that the stationary spin - orbitals satisfy , but only two of them are independent . these equations are the ghf equations for the @xmath171 and @xmath235 , in ( [ eq : chqc_hf_eqs_v1_ghf ] ) and ( [ eq : chqc_hf_eqs_v1_ghf_conj ] ) , respectively , and ( [ eq : chqc_lambdakjjk_ghf_herm ] ) . any pair of them could be in principle be chosen as the basic equations , however , in common practice the first and the last one of them are typically picked . in any case , due to ( [ eq : chqc_lambdakjjk_ghf_herm ] ) , a unitary matrix @xmath251 exists that _ diagonalizes _ @xmath252 ; in the sense that @xmath253 is a diagonal matrix , i.e. , @xmath254 . using this unitary matrix @xmath251 , we can transform the set of orbitals @xmath255 into a new one @xmath256 : @xmath257 this transformation is physically legitimate since it only changes the @xmath4-electron wavefunction @xmath67 in an unmeasurable phase @xmath258 . to see this , let us denote by @xmath259 the @xmath260-element of the matrix inside the slater determinant in eq . ( [ eq : chqc_constraint_total_sd_ghf ] ) , i.e. , @xmath261 . then , after using the expression above , the @xmath260-element of the new matrix @xmath262 can be related to the old ones via @xmath263 , in such a way that @xmath264 and the desired result follows : @xmath265 now , we insert eq . ( [ eq : chqc_transformation_phi_ghf ] ) into the first version of the hartree - fock equations in ( [ eq : chqc_hf_eqs_v1_ghf ] ) : @xmath266\,\left(\sum_j u_{ij}\psi^{\prime}_{j}(x)\right)= \sum_{j , k}\lambda_{ij}u_{jk}\psi^{\prime}_{k}(x)\ , \qquad i=1,\ldots , n\ .\ ] ] next , we multiply by @xmath267 each one of the @xmath4 expressions and sum in @xmath12 : @xmath268\,\left(\sum_{i , j } \underbrace{u^{-1}_{li}u_{ij}}_{\displaystyle \delta_{lj } } \psi^{\prime}_{j}(x)\right)= \sum_{i , j , k}\underbrace{u^{-1}_{li}\lambda_{ij}u_{jk } } _ { \displaystyle \varepsilon_{lk}=\delta_{lk}\varepsilon_{l } } \psi^{\prime}_{k}(x ) } \nonumber \\[6pt ] & & \longrightarrow\quad \hat{f}^\mathrm{ghf}[u\psi^{\prime}]\,\psi^{\prime}_{l}(x)= \varepsilon_{l}\psi^{\prime}_{l}(x)\ , \qquad l=1,\ldots , n\ .\end{aligned}\ ] ] although this new version of the hartree - fock equations can be readily seen as a _ pseudo - eigenvalue problem _ and solved by the customary iterative methods , we can go a step further and show that , like the @xmath4-particle wavefunction @xmath67 ( see eq . ( [ eq : chqc_phi_invariant_ghf ] ) ) , the fock operator @xmath269 $ ] , as a function of the one - electron orbitals , is invariant under a unitary transformation such as the one in eq . ( [ eq : chqc_transformation_phi_ghf ] ) . in fact , this is true for each one of the sums of coulomb and exchange operators in eq . ( [ eq : chqc_fock_op_ghf ] ) separately : @xmath270\,\varphi(x)= \sum_j \bigg(\int\frac { \left |\sum_k u_{jk } \,\psi^{\prime}_{k}(x^{\prime } ) \right|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \bigg)\,\varphi(x ) = } \nonumber \\ & & \sum_j \bigg(\int\frac { \sum_{k , l } u^{*}_{jk}u_{jl } \,\psi^{\prime * } _ { k}(x^{\prime } ) \,\psi^{\prime}_{l}(x^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \bigg ) \,\varphi(x)= \nonumber \\ & & \bigg(\int\frac { \sum_{j , k , l } u^{-1}_{kj}u_{jl } \,\psi^{\prime * } _ { k}(x^{\prime } ) \,\psi^{\prime}_{l}(x^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \bigg)\,\varphi(x)= \nonumber \\ & & \sum_k \bigg(\int\frac { |\psi^{\prime}_{k}(x^{\prime})|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}x^{\prime } \bigg)\,\varphi(x)= \sum_j \hat{j}_{j}[\psi^{\prime}]\,\varphi(x ) \ , \qquad \forall \varphi(x ) \ , \end{aligned}\ ] ] where , in the step before the last , we have summed on @xmath190 and @xmath222 , using that . performing very similar calculations , one can show that @xmath271\,\varphi(x ) = \sum_j \hat{k}_{j}[\psi^{\prime}]\,\varphi(x)\ , \qquad \forall \varphi(x ) \ , \ ] ] and therefore , that @xmath272=f^\mathrm{ghf}[\psi^{\prime}]$ ] . in such a way that any unitary transformation on a set of orbitals that constitute a solution of the hartree - fock equations in ( [ eq : chqc_hf_eqs_v1_ghf ] ) yields a different set that is also a solution of _ the same _ equations . for computational and conceptual reasons ( see , for example , koopmans theorem below ) , it turns out to be convenient to use this freedom and choose the matrix @xmath251 in such a way that the lagrange multipliers matrix is diagonalized ( see eq . ( [ eq : chqc_lambdakjjk_ghf ] ) and the paragraph below it ) . the particular set of one - electron orbitals @xmath256 obtained with this @xmath251 are called _ canonical orbitals _ and their use is so prevalent that we will circumscribe the forecoming discussion to them and drop the prime from the notation . using the canonical orbitals , the _ hartree - fock equations _ can be written as @xmath273\,\psi_{i}(x)=\varepsilon_{i}\psi_{i}(x)\ , \qquad i=1,\ldots , n\ .\ ] ] many of the remarks related to these equations are similar to those made about the hartree ones in ( [ eq : chqc_hartree_eqs ] ) , although there exist important differences due to the inclusion of the indistinguishability of the electrons in the variational ansatz . this is clearly illustrated if we calculate the joint probability density , associated to a wavefunction like the one in eq . ( [ eq : chqc_constraint_total_ghf ] ) , of the coordinates with label 1 taking the value @xmath274 , the coordinates with label @xmath141 taking the value @xmath275 , and so on : @xmath276 if we compare this expression with eq . ( [ eq : chqc_hartree_rho ] ) , we see that the antisymmetry of @xmath67 has completely spoiled the statistical independence among the one - electron coordinates . however , there is a weaker quasi - independence that may be recovered : if , using the same reasoning about permutations that took us to the one - electron part @xmath277 of the energy functional in page , we calculate the marginal probability density of the @xmath12-th coordinates taking the value @xmath187 , we find @xmath278 now , since the coordinates indices are just immaterial labels , the actual probability density of finding _ any _ electron with coordinates @xmath66 is given by @xmath279 which can be interpreted as a _ charge density _ ( except for the sign ) , as , in atomic units , the charge of the electron is @xmath280 . the picture being consistent with the fact that @xmath281 is normalized to the number of electrons @xmath4 : @xmath282 additionally , if we perform the same type of calculations that allowed to calculate the two - electron part of the energy functional in page , we have that the two - body probability density of the @xmath12-th coordinates taking the value @xmath187 and of the @xmath190-th coordinates taking the value @xmath209 reads @xmath283 and , if we reason in the same way as in the case of @xmath284 , in order to get to the probability density of finding _ any _ electron with coordinates @xmath66 at the same time that _ any other _ electron has coordinates @xmath285 , we must multiply the function above by @xmath286 , which is the number of immaterial @xmath198-labelings , taking into account that the distinction between @xmath66 and @xmath285 is also irrelevant : @xmath287 finally , taking eq . ( [ eq : chqc_qrhoi_ghf ] ) to this one , we have @xmath288 where the first term corresponds to independent electrons and the second one , called the _ interference term _ , could be interpreted as an exchange correction . although , in general , this is the furthest one may go , when additional constraints are imposed on the spin part of the one - electron wavefunctions ( see the discussion about restricted hartree - fock in the following pages , for example ) , the exchange correction in eq . ( [ eq : chqc_qrhoij_qrhoi_ghf ] ) above vanishes for electrons of opposite spin , i.e. , electrons of opposite spin turn out to be pairwise independent . however , whereas it is true that more correlation could be added to the hartree - fock results by going to higher levels of the theory and , in this sense , hartree - fock could be considered the first step in the ` correlation ladder ' , one should not regard it as an ` uncorrelated ' approximation , since , even in the simplest case of rhf ( see below ) , hartree - fock electrons ( of the same spin ) are statistically correlated . all of this has its roots in the _ pauli principle _ , which states that no pair of electrons can share all the quantum numbers . let us now point out that , like in the hartree case , the left - hand side of the hartree - fock equations in ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) is a complicated , non - linear function of the orbitals @xmath255 and the notation chosen is intended only to emphasize the nature of the iterative protocol that is typically used to solve the problem . however , note that , while the hartree operator @xmath133 $ ] depended on the index of the orbital @xmath157 on which it acted , the fock operator in eq . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) is the same for all the spin - orbitals @xmath171 . this is due to the inclusion of the @xmath289 terms in the sum of the coulomb and exchange two - electron integrals in eq . ( [ eq : chqc_expected_h_2_ghf ] ) and it allows to perform the iterative procedure solving only one eigenvalue problem at each step , instead of @xmath4 of them like in the hartree case ( see however the uhf and rohf versions of the hartree - fock problem in what follows ) . the one - particle appearance of eqs . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) is again strong and , whereas the ` eigenvalues ' @xmath116 are not the energies of the individual electrons , they are called _ orbital energies _ due to the physical meaning they receive via the well - known _ koopmans theorem _ @xcite . to get to this result , let us multiply eq . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) from the left by @xmath290 , for a given @xmath12 , and then integrate over @xmath66 . using the definition of the fock operator in eq . ( [ eq : chqc_fock_op_ghf ] ) together with the coulomb and exchange ones in eqs . ( [ eq : chqc_operators_ghf ] ) , we obtain @xmath291 where we have used the same notation as in eq . ( [ eq : chqc_expected_h_2_ghf ] ) and the fact that the one - electron orbitals are normalized . if we next sum on @xmath12 and compare the result with the expression in eq . ( [ eq : chqc_expected_h_2_ghf ] ) , we found that the relation of the eigenvalues @xmath116 with the actual hartree - fock energy is given by @xmath292 [ page : koopman ] finally , if we assume that upon ` removal of an electron from the @xmath139-th orbital ' the rest of the orbitals will remain unmodified , we can calculate the _ ionization energy _ using the expression in ( [ eq : chqc_expected_h_2_ghf ] ) together with the equations above : @xmath293 and this is koopmans theorem , namely , that _ the @xmath139-th ionization energy in the frozen - orbitals approximation is @xmath294_. moving now to the issue about the solution of the hartree - fock equations in ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) , we must remark that the necessity of using the relatively unreliable iterative approach to tackle them stems again from their complicated mathematical form . like in the hartree case , we have managed to largely reduce the dimension of the space on which the basic equations are defined : from @xmath144 in the electronic schrdinger equation in ( [ eq : chqc_bo_summary_a ] ) to 3 in the hartree - fock ones . however , to have this , we have payed the price of dramatically increasing their complexity @xcite , since , while the electronic schrdinger equation was one linear differential equation , the hartree - fock ones in ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) are @xmath4 coupled non - linear integro - differential equations , thus precluding any analytical approach to their solution . a typical iterative procedure the process described in this paragraph must be taken only as an outline of the one that is performed in practice . it is impossible to deal in a computer with a general function as it is ( a non - countable infinite set of numbers ) , and the problem must be discretized in some way . the truncation of the one - electron hilbert space using a finite basis set , described in secs . [ sec : qc_roothaan ] and [ sec : qc_basis_sets ] , is the most common way of doing this . moreover , the iterative procedure is normally performed using not the spin - orbitals but the spatial ones . in this sense , the restricted versions of the hartree - fock problem , discussed below , are closer to the actual implementation of the theory in computer applications . ] begins by proposing a _ starting guess _ for the set of @xmath4 spin - orbitals @xmath295 . with them , the fock operator @xmath296 $ ] in the left - hand side of eq . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) is constructed and the set of @xmath4 equations is solved as one simple eigenvalue problem . then , the @xmath297 that correspond to the @xmath4 lowest eigenvalues @xmath298 are selected ( see the discussion of the aufbau principle below ) and a new fock operator @xmath299 $ ] is constructed with them . the process is iterated until ( hopefully ) the @xmath151-th set of solutions @xmath300 differs from the @xmath301-th one @xmath302 less than a reasonably small amount ( defining the distance among solutions in some suitable way typically combined with a convergence criterium related to the associated energy change ) . when this occurs , the procedure is said to have converged and the solution orbitals are called _ self - consistent _ ; also , a calculation of this kind is commonly termed _ elf - onsistent ield _ ( scf ) . again , like in the hartree case , many issues exist that raise doubts about the possible success of such an approach . the most important ones are related to the fact that a proper definition of the _ hartree - fock problem _ should be : _ find the global minimum of the energy functional @xmath303 under the constraint that the wavefunction @xmath67 be a slater determinant of one - electron spin - orbitals _ , and not : _ solve the hartree - fock equations _ ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) . the solutions of the latter are all the stationary points of the constrained energy functional , while we are interested only in the particular one that is the global minimum . even ruling out the possibility that a found solution may be a maximum or a saddle point ( which can be done @xcite ) , one can never be sure that it is the global minimum and not a local one . there exists , however , one way , related to the hartree - fock version of the theorem by simon and lieb @xcite mentioned in the previous section , of hopefully biasing a particular found solution of eqs . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) to be the global minimum that we are looking for . they showed , first , that for neutral or positively charged molecules ( @xmath304 ) , the hartree - fock global minimization problem has a solution ( its uniqueness is not established yet @xcite ) and , second , that the minimizing orbitals @xmath255 correspond to the @xmath4 lowest eigenvalues of the fock operator @xmath269 $ ] that is self - consistently constructed with them . therefore , although _ the reverse is not true in general _ @xcite ( i.e. , from the fact that a particular set of orbitals are the eigenstates corresponding to the lowest eigenvalues of the associated fock operator , does not necessarily follow that they are the optimal ones ) , the information contained in simon and lieb s result is typically invoked to build each successive state in the iterative procedure described above by keeping only the @xmath4 orbitals that correspond to the @xmath4 lowest eigenvalues @xmath116 . indeed , by doing that , one is effectively constraining the solutions to have a property that the true solution does have , so that , in the worst case , the space in which one is searching is of the same size as the original one , and , in the best case ( even playing with the possibility that the reverse of simon and lieb s theorem be true , though not proved ) , the space of solutions is reduced to the correct global minimum alone . this wishful - thinking way of proceeding is termed the _ aufbau principle _ @xcite , and , together with a clever choice of the starting - guess set of orbitals @xcite ( typically extracted from a slightly less accurate theory , so that one may expect that it could be ` in the basin of attraction ' of the true hartree - fock minimum ) , constitute one of the many heuristic strategies that make possible that the aforementioned drawbacks ( and also those related to the convergence of iterative procedures ) be circumvented in real cases , so that , in practice , most of scf calculations performed in the field of quantum chemistry do converge to the true solution of eqs . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) in spite of the theoretical notes of caution . now , to close this ghf part , let us discuss some points regarding the imposition of constraints as a justification for subsequently introducing three commonly used forms of the hartree - fock theory that involve additional restrictions on the variational ansatz ( apart from those in eqs . ( [ eq : chqc_constraint_total_ghf ] ) and ( [ eq : chqc_constraint_mono_ghf ] ) ) . in principle , the target systems in which we are interested in our group and to which the theory developed in this work is meant to be applied are rather complex ( short peptides , small ligands , etc . ) . they have many degrees of freedom and the different interactions that drive their behaviour typically compete with one another , thus producing complicated , ` frustrated ' energy landscapes ( see refs . @xcite , but note , however , that we do not need to think about macromolecules ; a small molecule like co@xmath305 already has 22 electrons ) . this state of affairs renders the a priori assessment of the accuracy of any approximation to the exact equations an impossible task . as researchers calculate more and more properties of molecular species using quantum chemistry and the results are compared to higher - level theories or to experimental data , much empirical knowledge about ` how good is theory a for calculating property x ' is being gathered . however , if the characterization of a completely new molecule that is not closely related to any one that has been previously studied is tackled with , say , the hartree - fock approximation , it would be very unwise not to ` ask for a second opinion ' . all of this also applies , word by word , to the choice of the constraints on the wavefunction in variational approaches like the one discussed in this section : for example , it is impossible to know a priori what will be the loss of accuracy due to the requirement that the @xmath4-particle wavefunction @xmath67 be a slater determinant as in eq . ( [ eq : chqc_constraint_total_ghf ] ) . however , in the context of the hartree - fock approximation , there exists a way of proceeding , again , partly based on wishful thinking and partly confirmed by actual calculations in particular cases , that is almost unanimously used to choose additional constraints which are expected to yield more _ efficient _ theories . it consists of imposing constraints to the variational wavefunction that are _ properties that the exact solution to the problem does have_. in such a way that the obvious loss of accuracy due to the reduction of the search space is expected to be minimized , while the decrease in computational cost could be considerable . this way of thinking is clearly illustrated by the question of whether or not one should allow that the one - electron spin - orbitals @xmath171 ( and therefore the total wavefunction @xmath67 ) be complex valued . indeed , due to the fact that the electronic hamiltonian in eq ( [ eq : chqc_el_ham_1 ] ) is self - adjoint , the real and imaginary parts of any complex eigenfunction solution of the time independent schrdinger equation in ( [ eq : chqc_bo_summary_a ] ) are also solutions of it @xcite . therefore , the ground - tate , which is the exact solution of the problem that we are trying to solve , may be chosen to be real valued . nevertheless , the exact minimum will not be achieved , in general , in the smaller space defined by the hartree - fock constraints in eqs . ( [ eq : chqc_constraint_total_ghf ] ) and ( [ eq : chqc_constraint_mono_ghf ] ) , so that there is no a priori reason to believe that allowing the hartree - fock wavefunction to take complex values would not improve the results by finding a lower minimum . in fact , in some cases , this happens @xcite . nevertheless , if one constrains the search to real orbitals , the computational cost is reduced by a factor two , and , after all , the whole formalism discussed in this work profits from the imposition of constraints ( starting by the consideration of only one slater determinant ) , all of which save some computational effort at the expense of a reduction in the accuracy . the search for the most efficient of these approximations constitutes the main part of the quantum chemistry field . apart from these ` complex vs. real ' considerations , there exist three further restrictions that are commonly found in the literature and that affect the spin part of the one - electron orbitals @xmath171 . the @xmath4-electron wavefunction @xmath67 of the ghf approximation ( which is the one discussed up to now ) is not an eigenstate of the total - spin operator , @xmath306 , nor of the @xmath74-component of it , @xmath307 @xcite . however , since both of them commute with the electronic hamiltonian in eq . ( [ eq : chqc_el_ham_1 ] ) , the true ground - state of the exact problem can be chosen to be an eigenstate of both operators simultaneously . therefore three additional types of constraints on the spin part of the ghf wavefunction in ( [ eq : chqc_constraint_total_ghf ] ) are typically made that force the variational ansatz to satisfy these ground - state properties and that should be seen in the light of the above discussion , i.e. , as reducing the search space , thus yielding an intrinsically less accurate theory , but also as being good candidates to hope that the computational savings will pay for this . the first approximation to ghf ( in a logical sense ) is called _ nrestricted artree - ock _ ( uhf ) and it consists of asking the orbitals @xmath171 to be a product of a part @xmath308 depending on the positions @xmath172 times a spin eigenstate of the one - electron @xmath309 operator , i.e. , either @xmath310 or @xmath311 ( see eq . ( [ eq : chqc_alpha_beta ] ) ) . this is denoted by @xmath312 , where @xmath313 is either the @xmath14 or the @xmath178 function . now , if we call @xmath314 and @xmath315 the number of spin - orbitals of each type , we have that , differently from the ghf one , the uhf wavefunction @xmath67 is an eigenstate of the @xmath307 operator with eigenvalue @xmath316 ( in atomic units , see sec . [ sec : qc_molecular_h ] ) . however , it is not an eigenstate of @xmath317 and , when this deviation results into a poor description of the observables in which we are interested , we talk about _ spin contamination _ although the uhf wavefunction can be projected into pure @xmath317-states , the result is multideterminantal @xcite and will not be considered here since it spoils many of the properties that render hartree - fock methods a low - cost choice . regarding the computational cost of the uhf approximation , it is certainly lower than that of ghf , since the search space is half as large : in the latter case , we had to consider @xmath118 ( complex or real ) functions of @xmath125 ( the @xmath318 and the @xmath319 , see eq . ( [ eq : chqc_general_so ] ) ) , while in uhf we only have to deal with @xmath4 of them : the @xmath320 . now , if we introduce into the expression for the ghf constrained functional in ( [ eq : chqc_constrained_functional_ghf ] ) the following relations that hold for the uhf spin- and spatial orbitals of course , the average values at both sides of the expressions are taken over different variables : over @xmath66 and @xmath321 on the left - hand side , and over @xmath172 and @xmath322 on the right - hand side . also , let us remark that , although placing _ functions _ as arguments of the kronecker s delta @xmath323 is a bit unorthodox mathematically , it constitutes an intuitive ( and common ) notation . ] [ eq : chqc_help_relations_uhf ] @xmath324 we may perform a derivation analogous to the one performed for the ghf case , and get to a first version of the _ uhf equations _ [ eq : chqc_hf_eqs_vfinal_uhf ] @xmath325\,\phi_{i}({\bm r}):= \left(\hat{h}+\sum^n_j \hat{j}_j[\phi ] - \sum^n_{j}\delta_{\gamma_i \gamma_j } \hat{k}_j[\phi ] \right)\,\phi_{i}({\bm r})= \sum_j \lambda_{ij}\phi_{j}({\bm r})\ , \label{eq : chqc_hf_eqs_vfinal_uhf_a}\\ & \hat{f}^{\mathrm{uhf}}_{i}[\phi]\,\phi^*_{i}({\bm r})= \sum_j \lambda_{ji}\phi^*_{j}({\bm r})\ , \label{eq : chqc_hf_eqs_vfinal_uhf_b}\end{aligned}\ ] ] for all @xmath175 and where the coulomb and exchange operators dependent on the spatial orbitals @xmath326 are defined by their action on an arbitrary function @xmath327 as follows : [ eq : chqc_operators_uhf ] @xmath328\,\varphi({\bm r}):= \left ( \int\frac{|\phi_{j}({\bm r}^{\prime})|^{2 } } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}{\bm r}^{\prime } \right ) \varphi({\bm r})\ , \label{eq : chqc_operators_uhf_a } \\ & \hat{k}_{j}[\phi]\,\varphi({\bm r}):= \left ( \int\frac{\phi^{*}_{j}({\bm r}^{\prime})\,\varphi({\bm r}^{\prime } ) } { |{\bm r}-{\bm r}^{\prime}|}\,\mathrm{d}{\bm r}^{\prime } \right ) \phi_{j}({\bm r})\ . \label{eq : chqc_operators_uhf_b } \end{aligned}\ ] ] now , one must note that , differently from the ghf case , due to the fact that the exchange interaction only takes place between orbitals ` of the same spin ' , the _ uhf fock operator _ $ ] depends on the index @xmath12 . this precludes the solution of uhf as a _ pseudoeigenvalue problem ( c.f . ( [ eq : chqc_hf_eqs_vfinal_ghf ] ) ) and makes necessary some further considerations in order to arrive to a more directly applicable form of the expressions : first , although the equations for @xmath326 and @xmath330 in ( [ eq : chqc_hf_eqs_vfinal_uhf ] ) can be combined in the same way as in the ghf case to yield the hermiticity conditions @xmath331 , there are fewer lagrange multipliers in the uhf scheme than in the previous derivation . to see this , one must notice that the orthogonality constraints must be imposed _ on the spin - orbitals , not on the spatial orbitals _ ( see eq . ( [ eq : chqc_constraint_mono_ghf ] ) ) . therefore , since two uhf spin - orbitals @xmath332 and @xmath333 are orthogonal no matter the value of @xmath334 due to the different spin parts , the corresponding lagrange multiplier @xmath230 needs not to be included in the constrained functional from which the uhf equations come . this may be incorporated into the formalism by simply using that the matrix @xmath249 in eqs . ( [ eq : chqc_hf_eqs_vfinal_uhf ] ) presents the following block - diagonal form : @xmath335 0 & \lambda^\beta \end{array } \right ) \ , \ ] ] where we have assumed ( without loss of generality ) that the uhf spin - orbitals are ordered in such a way that the @xmath14 ones occur first , and @xmath336 indicates a block of zeros of the appropriate size . ( of course , redundant constraints may be imposed on the orbitals by , for example in this case , including matrix terms in @xmath252 that connect the @xmath14 and @xmath178 spaces . however , in order to know the exact freedom we have in the choice of the constraints , it is convenient to use the minimal number of lagrange multipliers . if this approach were not followed , for example , the discussion below about the diagonalization of @xmath337 and @xmath338 would become much less direct . ) the next step consists in noticing that , despite the dependence of @xmath339 on the orbital index @xmath12 in eqs . ( [ eq : chqc_hf_eqs_vfinal_uhf ] ) , there are actually only _ two _ different fock operators : one for the @xmath14 orbitals and one for the @xmath178 ones . defining the sets of indices @xmath340 and @xmath341 , we can write these two @xmath14 and @xmath178 operators : [ eq : chqc_alpha_beta_f_uhf ] @xmath342=\hat{h}+\sum^{n}_{j=1 } \hat{j}_j[\phi ] - \sum_{j \in a } \hat{k}_j[\phi ] \ , \label{eq : chqc_alpha_beta_f_uhf_a } \\ & \hat{f}^{\mathrm{uhf}}_\beta[\phi]=\hat{h}+\sum^{n}_{j=1 } \hat{j}_j[\phi ] - \sum_{j \in b } \hat{k}_j[\phi ] \ . \label{eq : chqc_alpha_beta_f_uhf_b } \end{aligned}\ ] ] with them , and using the particular structure of the matrix @xmath343 in ( [ eq : chqc_lambda_uhf ] ) , the original uhf equations in ( [ eq : chqc_hf_eqs_vfinal_uhf ] ) are split into two disjoint sets of expressions that are only coupled through the fock operators on the left hand sides , plus the hermiticity condition : [ eq : chqc_hf_eqs_vfinal_1_uhf ] @xmath342\phi_i({\bm r } ) = \sum_{j \in a } \lambda_{ij}^\alpha \phi_j({\bm r } ) \ , \quad \mathrm{if } \ i \in a \ , \label{eq : chqc_hf_eqs_vfinal_1_uhf_a } \\ & \hat{f}^{\mathrm{uhf}}_\beta[\phi]\phi_i({\bm r } ) = \sum_{j \in b } \lambda_{ij}^\beta \phi_j({\bm r } ) \ , \quad \mathrm{if } \ i \in b \ , \label{eq : chqc_hf_eqs_vfinal_1_uhf_b } \\ & \lambda_{ij } = \lambda_{ji}^ * \ . \label{eq : chqc_hf_eqs_vfinal_1_uhf_c}\end{aligned}\ ] ] the last step needed to arrive to the final form of the uhf equations ( found by pople and nesbet @xcite and named after them ) is the diagonalization of both the @xmath344 and @xmath338 hermitian matrices above . in order to achieve this , the orbitals @xmath326 must be transformed similarly to the ghf case . however , this is trickier than it was then , since not only the @xmath4-electron wavefunction @xmath67 and the fock operators must remain invariant under the sought transformation , but also the uhf constraints must be kept . as we saw before , any unitary transformation @xmath251 in the set of spin - orbitals @xmath171 like the one in eq . ( [ eq : chqc_transformation_phi_ghf ] ) is physically legitimate , since it changes the slater determinant by only an unmeasurable phase and leave the fock operators invariant . then , if we write each spin - orbital as in ( [ eq : chqc_general_so ] ) : @xmath345 we can make use of ( [ eq : chqc_transformation_phi_ghf ] ) , to obtain @xmath346 \ .\ ] ] by setting @xmath347 and @xmath348 in this expression , we see that any transformation @xmath251 in the spin - orbitals @xmath171 induces exactly the same transformation in their spatial components @xmath349 and @xmath350 , @xmath351 now , if we order the sets of spin and spatial orbitals : @xmath352 and @xmath353 , with @xmath354 , we can express the _ uhf constraints _ by saying that the @xmath355 must have the form [ eq : chqc_constraints_uhf ] @xmath356 then , since the fact that the @xmath14 orbitals appear first constitutes no loss of generality , we must ask the transformed @xmath357 , with @xmath354 , to have also the structure in ( [ eq : chqc_constraints_uhf ] ) if we want to remain inside the uhf scheme . as a consequence , and due to the linear independence among the orbitals , not every unitary matrix @xmath251 is allowed , but only those of the form @xmath358 0 & u^\beta \end{array } \right ) \ , \ ] ] using the same notation as in eq . ( [ eq : chqc_lambda_uhf ] ) . [ eq : qc_rohf_argument ] this can be easily proved by focusing on a particular value for @xmath12 and @xmath359 in eq . ( [ eq : chqc_transformation_phi_i ] ) , say , @xmath360 and @xmath361 . due to the uhf constraints in ( [ eq : chqc_constraints_uhf ] ) , we know that the left - hand side of such an expression is zero and that only spatial orbitals with @xmath362 appear in the sum on the right - hand side , yielding the relation @xmath363 . but the @xmath364 , with @xmath365 , form an orthonormal , and therefore linearly independent set , so that the only possibility that such a relation can hold is that all coefficients @xmath366 be zero . by repeating this for all @xmath360 and , then , for @xmath367 , the result follows . fortunately , this limited freedom in the choice of @xmath251 is still enough to independently diagonalize @xmath344 and @xmath338 in eq . ( [ eq : chqc_hf_eqs_vfinal_1_uhf ] ) ( which are both hermitian ) by suitably choosing the unitary submatrices @xmath368 and @xmath369 respectively . this takes us to the final , diagonal form of the uhf equations , the _ pople - nesbet equations _ @xcite : [ eq : chqc_hf_eqs_vfinal_2_uhf ] @xmath342\phi_i({\bm r } ) = \varepsilon_{i}^\alpha \phi_i({\bm r } ) \ , \quad \mathrm{if } \ i \in a \ , \label{eq : chqc_hf_eqs_vfinal_2_uhf_a } \\ & \hat{f}^{\mathrm{uhf}}_\beta[\phi]\phi_i({\bm r } ) = \varepsilon_{i}^\beta \phi_i({\bm r } ) \ , \quad \mathrm{if } \ i \in b \ . \label{eq : chqc_hf_eqs_vfinal_2_uhf_b}\end{aligned}\ ] ] although these equations are coupled through the coulomb term in the fock operators on the left - hand side , at each step of the iterative scf procedure , they can be solved as two independent eigenvalue problems . this has allowed to implement them in most quantum chemical packages and it has made uhf calculations now routine . to close the uhf discussion , we shall now study the statistical properties of the probability densities appearing in this model . if we introduce the special form of the uhf orbitals in ( [ eq : chqc_constraints_uhf ] ) into the general expression in ( [ eq : chqc_qrhoij_ghf ] ) , we can calculate the two - body probability density of finding _ any _ electron with coordinates @xmath66 at the same time that _ any other _ electron has coordinates @xmath285 : @xmath370 if we compute this probability density for ` electrons of the same spin ' , i.e. , for @xmath371 , we obtain placing _ a function and a coordinate _ as arguments of the kronecker s delta @xmath372 is even more unorthodox mathematically than placing two functions ( in fact , @xmath372 is exactly the same as @xmath373 ) , however , the intuitive character of the notation compensates again for this . ] @xmath374 & & \frac{1}{2 } \sum_{k , l \in i_\sigma } \left ( \big|\phi_k({\bm r})\big|^{2}\big|\phi_l({\bm r}^{\prime})\big|^{2 } - \phi^{*}_k({\bm r})\,\phi^{*}_l({\bm r}^{\prime})\,\phi_l({\bm r})\ , \phi_k({\bm r}^{\prime } ) \right ) = \nonumber \\ & & \frac{1}{2}\left(\rho^{\mathrm{uhf}}({\bm r},\sigma)\,\rho^{\mathrm{uhf}}({\bm r}^{\prime},\sigma ) - \sum_{k , l } \delta_{\gamma_k \sigma}\delta_{\gamma_l \sigma } \phi^{*}_k({\bm r})\ , \phi^{*}_l({\bm r}^{\prime})\,\phi_l({\bm r})\ , \phi_k({\bm r}^{\prime})\right ) \ , \end{aligned}\ ] ] where the following expression for the one - electron charge density has been used : @xmath375 at this point , note that eq . ( [ eq : chqc_qrhoij_samespins_uhf ] ) contains , like in the ghf case , the exchange correction to the first ( independent electrons ) term . nevertheless , as we advanced , if we calculate the two - body @xmath376 for ` electrons of opposite spin ' , i.e. , for @xmath377 , we have that @xmath378 i.e. , that _ uhf electrons of opposite spin are statistically pairwise independent_. there is another common approximation to ghf that is more restrictive than uhf and is accordingly called _ estricted artree - ock _ ( rhf ) . apart from asking the orbitals @xmath171 to be a product of a spatial part times a spin eigenstate of the one - electron @xmath309 operator ( like in the uhf case ) , in rhf , the number of ` spin - up ' and ` spin - down ' orbitals is the same , @xmath379 ( note that this means that rhf may only be used with molecules containing an even number of electrons ) , and each spatial wavefunction occurs twice : once multiplied by @xmath310 and the other time by @xmath311 . this is typically referred to as a _ closed - shell _ situation and we shall denote it by writing @xmath380 if @xmath381 , and @xmath382 if @xmath383 ; in such a way that there are @xmath384 different spatial orbitals denoted by @xmath385 , with , @xmath386 . using the same notation as in ( [ eq : chqc_constraints_uhf ] ) , the _ rhf constraints _ on the spin - orbitals are [ eq : chqc_constraints_rhf ] @xmath387 due to these additional restrictions , we have that , differently from the ghf and uhf ones , the rhf wavefunction @xmath67 is an eigenstate of both the @xmath306 and the @xmath307 operators , with zero eigenvalue in both cases @xcite , just like the ground - state of the exact problem . i.e. , _ there is no spin - contamination in rhf_. regarding the computational cost of the rhf approximation , it is even lower than that of uhf , since the size of the search space has been reduced to one quarter that of ghf : in the latter case , we had to consider @xmath118 ( complex or real ) functions of @xmath125 ( the @xmath318 and the @xmath319 , see eq . ( [ eq : chqc_general_so ] ) ) , while in rhf we only have to deal with @xmath384 of them : the @xmath388 ( see fig . [ fig : hf_computational_cost ] ) . now , using again the relations in ( [ eq : chqc_help_relations_uhf ] ) , we can derive a first version of the _ rhf equations _ : @xmath389\,\phi_{i}({\bm r}):= \left [ \hat{h}+\sum^{n/2}_j \bigg ( 2\hat{j}_j[\phi ] - \hat{k}_j[\phi ] \bigg ) \right]\phi_{i}({\bm r})= \sum_{j}^{n/2}\lambda_{ij}\phi_{j}({\bm r})\ , \ ] ] for all @xmath386 , where we have used that the _ rhf fock operator _ @xmath390 $ ] , differently from the uhf case , does not depend on the index @xmath391 of the orbital on which it operates , and following the same steps as before , the minimal lagrange multipliers matrix needed to enforce the orthogonality constraints among the _ _ spin__orbitals in the rhf case is @xmath392 0 & \lambda^{(n/2)}/2 \end{array } \right ) \ , \ ] ] where , this time , @xmath393 is an arbitrary @xmath394 hermitian matrix , and the @xmath29 has been included in order to get to the classical rhf equations in ( [ eq : chqc_hf_eqs_vfinal_1_rhf ] ) without irrelevant numerical factors . using the same arguments as for uhf , it is clear that , in order to ` remain inside rhf ' upon an unitary transformation of the spin - orbitals @xmath171 , not every unitary matrix @xmath251 is allowed , but only those of the form @xmath395 0 & u^{(n/2 ) } \end{array } \right ) \ .\ ] ] finally , by suitable choosing the @xmath394 unitary block @xmath396 , the hermitian matrix @xmath397 can be diagonalized and the final , diagonal form of the _ rhf equations _ can be written : @xmath398\,\phi_{i}({\bm r}):= \left [ \hat{h}+\sum^{n/2}_j \bigg ( 2\hat{j}_j[\phi ] - \hat{k}_j[\phi ] \bigg ) \right]\phi_{i}({\bm r})= \varepsilon_{i}\phi_{i}({\bm r})\ , \ ] ] with @xmath386 . as we can see , this version of the hartree - fock theory can be numerically solved as a _ single _ pseudoeigenvalue problem . this , together with the aforementioned small size of the rhf space , has made the rhf approximation ( in its real - valued version ) the first one cast into a computationally manageable form @xcite and the most used one in recent literature @xcite ( for molecules with an even number of electrons ) . now , in order to investigate the statistical features of rhf , if we introduce the special form of the orbitals in ( [ eq : chqc_constraints_rhf ] ) into the general expression in ( [ eq : chqc_qrhoij_ghf ] ) , we can calculate the rhf two - body probability density of finding _ any _ electron with coordinates @xmath66 at the same time that _ any other _ electron has coordinates @xmath285 : @xmath399 where the following expression for the rhf one - electron charge density has been used : @xmath400 we notice that the situation is the same as in the uhf case : for rhf electrons with equal spin , there exists an exchange term in @xmath401 that corrects the ` independent ' part , whereas _ rhf electrons of opposite spin are statistically pairwise independent_. finally , if we follow the same steps as for ghf , in page , we can relate the _ rhf energy _ to the eigenvalues @xmath402 and the two - electron spatial integrals : @xmath403 to close this section , we shall discuss a fourth flavour of hartree - fock which is called _ estricted pen - shell artree - ock _ ( rohf ) . compared to the rest of variants , rohf is the most difficult to derive theoretically and it shall be described here only in an introductory manner . the rohf wavefunction is somewhat between the rhf and the uhf ones , and both of them can be obtained as particular cases of the rohf scheme ( the rhf one provided that the molecule has an even number of electrons ) . in the ( monodeterminantal ) rohf case , the one - electron spin orbitals @xmath171 are constrained to be of two different types : @xmath404 of them are _ doubly occupied _ , like in the rhf case , in such a way that they are formed by only @xmath405 spatial orbitals @xmath406 , each one of them appearing once multiplied by @xmath310 and once by @xmath311 . the associated @xmath404 spin - orbitals are said to belong to the _ closed shell _ part of the wavefunction . the remaining @xmath407 ones are _ singly occupied _ , like in the uhf case , and are said to belong to the _ open shell_. among them , @xmath314 are multiplied by an @xmath310 spin part , and @xmath315 by @xmath311 . if we number the whole set of rohf spatial orbitals @xmath406 , with @xmath408 , in such a way that the doubly occupied ones occur first , with @xmath409 , then the alpha ones , with @xmath410 , and finally the beta ones , with @xmath411 , we can express the _ rohf constraints _ on the spin - orbitals using the same notation as in ( [ eq : chqc_constraints_uhf ] ) and ( [ eq : chqc_constraints_rhf ] ) : [ eq : chqc_constraints_rohf ] @xmath412 where the particular ordering has been chosen in order to facilitate the forecoming calculations . the general monodeterminantal rohf wavefunction considered here and constructed using the above constraints has the same spin properties as the uhf one , i.e. , it is an eigenstate of the the @xmath307 operator with eigenvalue @xmath316 , but it is not an eigenstate of @xmath317 . however , in the particular ( and common ) case in which all the @xmath413 open shell orbitals are constrained to present parallel spin parts ( either all @xmath14 or all @xmath178 ) , the rohf wavefunction becomes an eigenstate of the @xmath317 operator too , with eigenvalue @xmath414 , thus avoiding the problem of spin contamination @xcite . in order to construct wavefunctions with the same @xmath415 but lower @xmath416 , several rohf slater determinants must be linearly combined . the subtleties arising from such a procedure are beyond the scope of this review ; the interested reader may want to check references @xcite , which discuss this topic . regarding the size of the variational space in rohf , it is somewhere between uhf and rhf , depending on the @xmath417 ratio . now , if the rohf constraints in ( [ eq : chqc_constraints_rohf ] ) are imposed on the ghf energy in eq . ( [ eq : chqc_expected_h_2_ghf ] ) , the rohf analogue in terms of the spatial orbitals @xmath406 can be calculated : @xmath418= \sum_{a}f_a\langle\phi_{a}|\hat{h}|\phi_{a}\rangle } \nonumber \\ & & \mbox { } + \frac{1}{2 } \sum_{a , b } \bigg(f_a f_b \langle\phi_a\phi_b|\frac{1}{r}|\phi_a\phi_b\rangle - g_{ab}\langle\phi_a\phi_b|\frac{1}{r}|\phi_b\phi_a\rangle \bigg ) \ , \end{aligned}\ ] ] where the @xmath419 are a sort of ` occupation numbers ' that take the value @xmath420 when @xmath421 ( i.e. , when it corresponds to a closed shell orbital ) and @xmath422 otherwise . the matrix @xmath423 is defined as @xmath424 \hline\\[-8pt ] 1 & \begin{array}{c@{\hspace{5pt}}c } 1^\alpha & 0 \\[5pt ] 0 & 1^\beta \end{array } \end{array } \right ) \ , \ ] ] being @xmath425 a @xmath426 box of 2 s , @xmath427 and @xmath428 , @xmath429 and @xmath430 boxes of 1 s respectively . the off - diagonal blocks contain in all elements the number indicated and are of the appropriate size . then , as we did for ghf , in order to derive the rohf equations , we construct the functional for the conditioned stationary values problem , adding to the energy in ( [ eq : chqc_energy_rohf ] ) the lagrange multipliers terms needed to enforce the orthonormality constraints : @xmath431 = e^{\mathrm{rohf}}\big[\{\phi_a\}\big ] - \sum_{a , b } \lambda_{ab}\ , \big ( \langle\phi_a|\phi_b\rangle-\delta_{ab } \big ) \ .\ ] ] now , like in all restricted hf cases , special attention must be payed to the structure of the matrix @xmath432 , since the requirement is that all _ spin - orbitals _ be orthogonal , not the spatial orbitals . in the rohf case , this leads to explicitly impose the orthonormality conditions ( of course ) inside the three sets of @xmath406 : the doubly occupied , the alpha and the beta ones ; but also between the doubly occupied and the alpha ones , and between the doubly occupied and the beta ones . the orthormality between the alpha and beta sets , however , needs not to be enforced , since the associated spin - orbitals are already orthogonal due to the different spin parts . these considerations lead to the following form for the minimal lagrange multipliers matrix : @xmath433 ( \lambda^{d\alpha})^+ & \lambda^\alpha & 0 \\[5pt ] ( \lambda^{d\beta})^+ & 0 & \lambda^\beta \end{array } \right ) \ , \ ] ] where the notation used for the different blocks connecting the doubly occupied , alpha and beta shells is self - explanatory , and the fact that @xmath434 is hermitian after reaching stationarity has been advanced . next , we impose the condition of zero functional derivative on the functional in ( [ eq : chqc_constrained_functional_rohf ] ) , and obtain a first version of the _ rohf equations _ : [ eq : chqc_hf_eqs_v1_rohf ] @xmath435\phi_{a}({\bm r})= \sum_{b}\lambda_{ab}\phi_{b}({\bm r})\ , \qquad a = 1,\ldots , n_d+n_\alpha+n_\beta \ , \label{eq : chqc_hf_eqs_v1_rohf_a } \\ & \hat{f}_a^{\mathrm{rohf}}[\phi]\,\phi^*_{a}({\bm r})= \sum_{b}\lambda_{ba}\phi^*_{b}({\bm r})\ , \qquad a = 1,\ldots , n_d+n_\alpha+n_\beta \ , \label{eq : chqc_hf_eqs_v1_rohf_b}\end{aligned}\ ] ] where the _ rohf operator _ @xmath436 $ ] is defined as @xmath437:= f_a \hat{h}+ \sum_b \bigg ( f_a f_b \hat{j}_b[\phi ] - g_{ab}\hat{k}_b[\phi ] \bigg ) \ , \ ] ] and the hermiticity property of the lagrange multipliers matrix @xmath438 follows from conjugation and subtraction in eqs . ( [ eq : chqc_hf_eqs_v1_rohf ] ) . again , although the fock operator depends on the index @xmath98 of the orbital upon which it operates , this dependence presents a very particular structure , yielding _ only three _ different types of operators : [ eq : chqc_rohf_ops ] @xmath439:= 2 \hat{h}+ \sum_b \bigg ( 2 f_b \hat{j}_b[\phi ] - f_{b}\hat{k}_b[\phi ] \bigg ) \ , \qquad a \in d \ , \label{eq : chqc_hf_rohf_ops_a } \\ & \hat{f}_\alpha^{\mathrm{rohf}}[\phi]:= \hat{h}+ \sum_{b \in ( d \cup a ) } \bigg(\hat{j}_b[\phi ] - \hat{k}_b[\phi ] \bigg ) \ , \qquad a \in a \ , \label{eq : chqc_hf_rohf_ops_b } \\ & \hat{f}_\beta^{\mathrm{rohf}}[\phi]:= \hat{h}+ \sum_{b \in ( d \cup b ) } \bigg ( \hat{j}_b[\phi ] - \hat{k}_b[\phi ] \bigg ) \ , \qquad a \in b \ . \label{eq : chqc_hf_rohf_ops_c}\end{aligned}\ ] ] note that , in the particular case that we had no open - shell orbitals , the operator for the doubly occupied ones does not reduce ( as it should ) to the rhf one in ( [ eq : chqc_hf_eqs_vfinal_1_rhf ] ) . this is because we hid a factor 2 in the definition of the lagrange multipliers in the rhf derivation . at this point , it would be desirable to continue with the same program that we followed in the uhf and rhf cases and diagonalize the matrix @xmath438 in order to arrive to a system of three pseudoeigenvalue equations , coupled only via the fock operators . nevertheless , this is not possible in rohf , as we shall show in the following lines , and it is the root of rohf being the most tricky flavour in the hartree - fock family . the obstruction to achieve this diagonalization comes from the fact , already used in the uhf case , that not every unitary transformation of the _ spin - orbitals _ is allowed if we want to remain inside the rohf scheme , i.e. , if we want that the transformed spin - orbitals satisfy the same rohf constraints in ( [ eq : chqc_constraints_rohf ] ) that the untransformed ones did . to begin with , if we recall that the doubly occupied spatial orbitals must be orthogonal to both the alpha and the beta ones , we can use the same reasoning as in page to show that the alpha - beta connecting parts of the allowed unitary matrix in this case must be zero . hence , we can write @xmath440 u^{\prime d\alpha } & u^\alpha \end{array } & 0 \\[5pt ] \hline\\[-8pt ] 0 & \begin{array}{c@{\hspace{5pt}}c } u^{\prime d } & u^{d\beta } \\[5pt ] u^{\prime d\beta } & u^\beta \end{array } \end{array } \right ) \ , \ ] ] where the size of each block may be easily found from ( [ eq : chqc_constraints_rohf ] ) and the notation is again self - explanatory . now , in order to obtain further restrictions to the form of @xmath441 , we write the transformation of the _ two _ spin - orbitals in the closed shell that correspond to _ the same _ spatial orbital @xmath406 , with @xmath442 . to this end , we use ( [ eq : chqc_u_rohf ] ) , ( [ eq : chqc_constraints_rohf ] ) , and the appropriate lines in the first and third lines of blocks in ( [ eq : chqc_u_rohf ] ) : @xmath443 using the orthogonality relations inside and among the three sets of spatial orbitals , we multiply the above expression by any @xmath444 , with @xmath445 and integrate on @xmath172 . from this , the equality of @xmath446 and @xmath447 follows , and the corresponding sums in ( [ eq : chqc_rohf_proof1 ] ) subtract to zero , yielding @xmath448 which equates one vector in the linear space spanned by the alpha orbitals to another in the linear space spanned by the beta ones . however , only the zero vector can belong to both spaces if we want the rohf assumptions regarding the spin - orbitals to hold . to see this , we can use a _ reductio ad absurdum _ type of argument : assume that both sides of ( [ eq : chqc_rohf_proof2 ] ) are different from zero . then , we may perform a unitary transformation changing only the alpha and beta spaces , and with no elements connecting the two . this is allowed , since it does not change neither the @xmath4-electron wavefunction , nor the fock operators , nor the rohf constraints . now , if we select the partial unitary transformations in the alpha and beta sets so that they turn the vectors at both sides of ( [ eq : chqc_rohf_proof2 ] ) into single elements in the bases of their respective spaces , we have that a single alpha orbital _ equals _ a beta one . although this can happen in particular cases , we can not ask it or we would be changing the fundamental assumptions made in ( [ eq : chqc_constraints_rohf ] ) . therefore , both sides of ( [ eq : chqc_rohf_proof2 ] ) must be zero , and , since the alpha and beta sets are linearly independent , all coefficients must be zero too . the proof that @xmath449 and @xmath450 are also zero is performed using similar arguments , and the final form of @xmath451 satisfying all the restrictions reads @xmath452 0 & u^\alpha \end{array } & 0 \\[5pt ] \hline\\[-8pt ] 0 & \begin{array}{c@{\hspace{5pt}}c } u^d & 0 \\[5pt ] 0 & u^\beta \end{array } \end{array } \right ) \ .\ ] ] finally , if we write the associated matrix using the @xmath98 indices , i.e. , operating on the set of spatial orbitals with the doubly occupied ones _ unrepeated _ : @xmath453 & u^\alpha & \\[5pt ] 0 & & u^\beta \end{array } \right ) \ , \ ] ] then , the transformed @xmath454 is related to the original one in ( [ eq : chqc_lambda_rohf ] ) through simple matrix multiplication : @xmath455 ( dropping the rohf superindices ) . it is clear that such a restricted @xmath456 does not operate on the off - diagonal blocks of @xmath438 in ( [ eq : chqc_lambda_rohf ] ) and , therefore , _ the sought diagonalization is not possible_. using the fact that , however , the @xmath457 , @xmath344 and @xmath338 do allow to be diagonalized , we can write the final , simplest possible form of the _ rohf equations _ ( forgetting their complex conjugate counterparts ) : [ eq : chqc_hf_eqs_vfinal_rohf ] @xmath439\phi_{a}({\bm r})= \varepsilon^d_a \phi_{a}({\bm r } ) + \sum_{b \in a}\lambda^{d\alpha}_{ab}\phi_{b}({\bm r } ) + \sum_{b \in b}\lambda^{d\beta}_{ab}\phi_{b}({\bm r})\ , \quad a \in d \ , \label{eq : chqc_hf_eqs_vfinal_rohf_a } \\ & \hat{f}_\alpha^{\mathrm{rohf}}[\phi]\phi_{a}({\bm r})= \varepsilon^\alpha_a \phi_{a}({\bm r } ) + \sum_{b \in d}(\lambda^{d\alpha}_{ba})^*\phi_{b}({\bm r})\ , \quad a \in a \ , \label{eq : chqc_hf_eqs_vfinal_rohf_b } \\ & \hat{f}_\beta^{\mathrm{rohf}}[\phi]\phi_{a}({\bm r})= \varepsilon^\beta_a \phi_{a}({\bm r } ) + \sum_{b \in d}(\lambda^{d\beta}_{ba})^*\phi_{b}({\bm r})\ , \quad a \in b \ , \label{eq : chqc_hf_eqs_vfinal_rohf_c}\end{aligned}\ ] ] where the superindices in the matrix elements are only written for visual convenience when comparing with ( [ eq : chqc_lambda_rohf ] ) . in this final form , it is evident that the off - diagonal elements of the lagrange multipliers matrix , i.e. , those related to the orthogonality constraints between the closed and open shells , introduce a coupling in the right - hand side of the rohf equations that completely spoils the possibility of casting them into pseudoeigenvalue ones . in the previous uhf and rhf versions , _ all _ orthogonality constraints were handled by simply choosing a special basis in the space spanned by the spatial orbitals in which the lagrange multipliers matrix was diagonal . however , in the rohf scheme , there is no such basis and we must deal with the problem in a different way , resulting into higher computational and theoretical difficulty . since the pioneering work by roothaan @xcite , the solution of the rohf problem has been attempted by distinct means , ranging from directly tackling the rohf equations in ( [ eq : chqc_hf_eqs_vfinal_rohf ] ) explicitly forcing the orthogonality constraints @xcite , to the construction of a so - called _ unified coupling operator _ @xcite , which allows to turn the rohf scheme into a single pseudoeigenvalue problem at the price of introducing certain ambiguities in the one - electron orbital energies @xcite . the details and subtleties involved in these issues are beyond the scope of a review of the fundamental topics such as this one . the interested reader may want to check the more specialized accounts in @xcite . the hartree - fock equations in the rhf form in expression ( [ eq : chqc_hf_eqs_vfinal_rhf ] ) are a set of @xmath384 coupled integro - differential equations . as such , they can be tackled by finite - differences methods and solved on a discrete grid ; this is known as _ numerical hartree - fock _ @xcite , and , given the present power of computers , it is only applicable to very small molecules . in order to deal with larger systems , such as biological macromolecules , independently proposed by roothaan @xcite and hall @xcite in 1951 , a different kind of discretization must be performed , not in @xmath458 but in the hilbert space @xmath459 of the one - electron orbitals . hence , although the actual dimension of @xmath459 is infinite , we shall approximate any function in it by a finite linear combination of @xmath460 different functions @xmath461 in all this section and the next one , the indices belonging to the first letters of the alphabet , @xmath462 , etc . , run from 1 to @xmath460 ( the number of functions in the finite basis set ) ; whereas those named with capital letters from @xmath391 towards the end of the alphabet , @xmath463 , etc . , run from 1 to @xmath384 ( the number of spatial wavefunctions @xmath464 , also termed the number of _ occupied orbitals _ ) . ] . in particular , the one - electron orbitals that make up the rhf wavefunction , shall be approximated by @xmath465 in both cases , numerical hartree - fock and discretization of the function space , the correct result can be only be reached asymptotically ; when the grid is very fine , for the former , and when @xmath466 , for the latter . this exact result , which , in the case of small systems , can be calculated up to several significant digits , is known as the _ hartree - fock limit _ @xcite . in practical cases , however , @xmath460 is finite ( often , only about an order of magnitude larger than @xmath384 ) and the set @xmath467 in the expression above is called the _ basis set_. we shall devote the next section to discuss its special characteristics , but , for now , it suffices to say that , in typical applications , the functions @xmath461 are _ atom - centred _ , i.e. , each one of them has non - negligible value only in the vicinity of a particular nucleus . therefore , like all the electronic wavefunctions we have dealt with in the last sections , they parametrically depend on the positions @xmath49 of the nuclei ( see sec . [ sec : qc_born - oppenheimer ] ) . this is why , sometimes , the functions @xmath461 are called _ _ tomic rbitals _ _ some authors @xcite suggest that , being strict , the term _ atomic orbitals _ should be reserved for the one - electron wavefunctions @xmath464 that are the solution of the hartree - fock problem ( or even to the exact schrdinger equation of the isolated atom ) , and that the elements @xmath461 in the basis set should be termed simply _ localized functions_. however , it is very common in the literature not to follow this recommendation and choose the designation that appear in the text @xcite . we shall do the same for simplicity . ] ( ao ) ( since they are localized at individual atoms ) , the @xmath464 are referred to as _ olecular rbitals _ ( mo ) ( since they typically have non - negligible value in the whole space occupied by the molecule ) , and the approximation in eq . ( [ eq : chqc_lcao ] ) is called _ inear ombination of tomic rbitals _ ( lcao ) . in addition , since we voluntarily circumscribe to real - rhf , we assume that both the coefficients @xmath468 and the functions @xmath461 in the above expression are real . now , if we introduce the linear combination in eq . ( [ eq : chqc_lcao ] ) into the hartree - fock equations in ( [ eq : chqc_hf_eqs_vfinal_rhf ] ) , multiply the result from the left by @xmath469 ( for a general value of @xmath470 ) and integrate on @xmath471 , we obtain @xmath472 where we denote by @xmath473 the @xmath474-element of the _ _ fock matrix _ _ note that the rhf superindex has been dropped from @xmath475 . ] , and by @xmath476 the one of the _ overlap matrix _ , defined as @xmath477 \ , | \chi_a \rangle \qquad \mathrm{and } \qquad s_{ba}:=\langle \chi_b | \chi_a \rangle \ , \ ] ] respectively . note that we do not ask the @xmath461 in the basis set to be mutually orthogonal , so that the overlap matrix is not diagonal in general . next , if we define the @xmath478 matrices @xmath479:=(f_{ab})$ ] and @xmath480 , together with the ( column ) @xmath460-vector @xmath481 , we can write eq . ( [ eq : chqc_rh_1 ] ) in matricial form : @xmath482 c_i=\varepsilon_i\,sc_{i}\ .\ ] ] hence , using the lcao approximation , we have traded the @xmath384 coupled integro - differential hartree - fock equations in ( [ eq : chqc_hf_eqs_vfinal_rhf ] ) for this system of @xmath384 algebraic equations for the @xmath384 orbital energies @xmath402 and the @xmath483 coefficients @xmath468 , which are called _ roothaan - hall equations _ @xcite and which are manageable in a computer . now , if we forget for a moment that the fock matrix depends on the coefficients @xmath468 ( as stressed by the notation @xmath479 $ ] ) and also that we are only looking for @xmath384 vectors @xmath484 while the matrices @xmath475 and @xmath485 are @xmath486 , we may regard the above expression as a @xmath460-dimensional _ generalized eigenvalue problem_. many properties are shared between this kind of problem and a classical eigenvalue problem ( i.e. , one in which @xmath487 ) @xcite , being the most important one that , due to the hermiticity of @xmath479 $ ] , one can find an orthonormal set of @xmath460 vectors @xmath488 corresponding to real eigenvalues @xmath489 ( where , of course , some eigenvalue could be repeated ) . in fact , it is using this formalism how most of actual hartree - fock computations are performed , although the reader must also note that other approaches , in which the orthonormality constraints are automatically satisfied due to the choice of variables also exist in the literature @xcite . the general outline of the iterative procedure is essentially the same as the one discussed in sec . [ sec : qc_hartree - fock ] : choose a _ starting - guess _ for the coefficients @xmath468 ( let us denote it by @xmath490 ) , construct the corresponding fock matrix @xmath491 $ ] note ( in eq . ( [ eq : chqc_matrices_2 ] ) , for example ) that the fock matrix only depends on the vectors @xmath488 with @xmath492 . ] and solve the generalized eigenvalue problem in eq . ( [ eq : chqc_rh_2 ] ) . by virtue of the aufbau principle discussed in the previous section , from the @xmath460 eigenvectors @xmath488 , keep the @xmath384 ones @xmath493 that correspond to the @xmath384 lowest eigenvalues @xmath494 , construct the new fock matrix @xmath495 $ ] and iterate ( by convention , the eigenvalues @xmath496 , for all @xmath151 , are ordered from the lowest to the largest as @xmath98 runs from 1 to @xmath460 ) . this procedure ends when the @xmath151-th solution is close enough ( in a suitable defined way ) to the @xmath301-th one . also , note that , after convergence has been achieved , we end up with @xmath460 orthogonal vectors @xmath497 . only the @xmath384 ones that correspond to the lowest eigenvalues represent real one - electron solutions and they are called _ occupied orbitals _ ; the @xmath498 remaining ones do not enter in the @xmath4-electron wavefunction ( although they are relevant for calculating corrections to the hartree - fock results ) and they are called _ virtual orbitals_. regarding the mathematical foundations of this procedure , let us stress , however , that , whereas in the finite - dimensional ghf and uhf cases it has been proved that the analogous of lieb and simon s theorem ( see the previous section ) is satisfied , i.e. , that the global minimum of the original optimization problem corresponds to the lowest eigenvalues of the self - consistent fock operator , in the rhf and rohf cases , contrarily , no proof seems to exist @xcite . of course , in practical applications , the positive result is assumed to hold . finally , if we expand @xmath499 in eq . ( [ eq : chqc_matrices_1 ] ) , using the shorthand @xmath500 for @xmath501 , we have @xmath502 } \underbrace{\left ( 2 \langle ac|\,\frac{1}{r}\,|bd\rangle - \langle ac|\,\frac{1}{r}\,|db\rangle\right ) } _ { \displaystyle g^{cd}_{ab } } \ , \ ] ] where we have introduced the _ density matrix _ @xmath503 $ ] , and also the matrix @xmath504 , made up by the _ two - electron four - centre integrals _ @xmath505 ( also called _ lectron epulsion ntegrals _ ( eris ) ) defined by @xmath506 it is also convenient to introduce the _ coulomb _ ( @xmath507 $ ] ) and _ exchange _ ( @xmath508 $ ] ) _ matrices _ [ eq : chqc_coulomb_exchange ] @xmath509 : = \sum_{c , d}d_{cd}[c]\langle ac|\,\frac{1}{r}\,|bd\rangle \ , \label{eq : chqc_coulomb_exchange_a } \\ & k_{ab}[c ] : = \sum_{c , d}d_{cd}[c]\langle ac|\,\frac{1}{r}\,|db\rangle \ , \label{eq : chqc_coulomb_exchange_b } \end{aligned}\ ] ] in terms of which , the fock operator in eq . ( [ eq : chqc_matrices_2 ] ) may be expressed as @xmath510 - k_{ab}[c ] \ .\ ] ] after scf convergence has been achieved , the _ rhf energy _ in the finite - dimensional case can be computed using the discretized version of eq . ( [ eq : chqc_energy_rhf ] ) : @xmath511d_{bd}[c]\langle ab|\,\frac{1}{r}\,|cd\rangle\ , \end{aligned}\ ] ] where a convenient rearrangement of the indices in the two sums has been performed from the first to the second line . in principle , arbitrary functions may be chosen as the @xmath461 to solve the roothaan - hall equations in the previous section , however , in eq . ( [ eq : chqc_matrices_2 ] ) , we see that one of the main numerical bottlenecks in scf calculations arises from the necessity of calculating the @xmath512 four - centre integrals if the symmetry properties of the integrals are used , the precise number of eris is found to be @xmath513 @xcite . ] @xmath514 , since the solution of the generalized eigenvalue problem in eq . ( [ eq : chqc_rh_2 ] ) typically scales only like @xmath515 , and there are two - centre @xmath516 integrals ( see however the next section ) . either if these integrals are calculated at each iterative step and directly taken from ram memory ( _ direct scf _ ) or if they are calculated at the first step , written to disk , and then read from there when needed ( _ conventional scf _ ) , an appropriate choice of the functions @xmath461 in the finite basis set is essential if accurate results are sought , @xmath460 is intended to be kept as small as possible and the integrals are wanted to be computed rapidly . when one moves into higher - level theoretical descriptions and the numerical complexity scales with @xmath460 even more unpleasantly , the importance of this choice greatly increases . in this section , in order to support that study , we shall introduce some of the concepts involved in the interesting field of basis - set design . for further details not covered here , the reader may want to check refs . @xcite . the only analytically solvable molecular problem in non - relativistic quantum mechanics is the _ hydrogen - like atom _ , i.e. , the system formed by a nucleus of charge @xmath517 and only one electron ( h , he@xmath518 , li@xmath519 , etc . ) . therefore , it is not strange that all the thinking about atomic - centred basis sets in quantum chemistry is much influenced by the particular solution to this problem . the spatial eigenfunctions of the hamiltonian operator of an hydrogen - like atom , in atomic units and spherical coordinates , read for consistency with the rest of the text , the born - oppenheimer approximation has been also assumed here . so that the _ reduced mass _ @xmath520 that should enter the expression is considered to be the mass of the electron @xmath521 ( recall that , in atomic units , @xmath522 and @xmath523 ) . ] @xmath524 ^ 3 } } \left ( \frac{2z}{n } r \right)^l l^{2l+1}_{n - l-1}\left ( \frac{2z}{n } r \right ) e^{-zr / n } y_{lm}(\theta,\varphi ) \ , \ ] ] where @xmath151 , @xmath222 and @xmath25 are the energy , total angular momentum and @xmath74-angular momentum quantum numbers , respectively . their ranges of variation are coupled : all being integers , @xmath151 runs from 1 to @xmath525 , @xmath222 from 0 to @xmath526 and @xmath25 from @xmath527 to @xmath222 . the function @xmath528 is a _ generalized laguerre polynomial _ @xcite , for which it suffices to say here that it is of order @xmath529 ( thus having , in general , @xmath529 zeros ) , and the function @xmath530 is a _ spherical harmonic _ , which is a simultaneous eigenfunction of the total angular momentum operator @xmath531 ( with eigenvalue @xmath532 ) and of its @xmath74-component @xmath533 ( with eigenvalue @xmath25 ) . the hope that the one - electron orbitals that are the solutions of the hartree - fock problem in many - electron atoms could not be very different from the @xmath534 [ page : foot : orbitals_only_in_hf ] above note that the @xmath4-electron wavefunction of the exact ground - state of a non - hydrogen - like atom depends on 3@xmath4 spatial variables in a way that can not be written , in general , as a slater determinant of one - electron functions . the image of single electrons occupying definite orbitals , together with the possibility of comparing them with the one - particle eigenfunctions of the hamiltonian of hydrogen - like atoms , vanishes completely outside the hartree - fock formalism . ] , together with the powerful chemical intuition that states that ` atoms - in - molecules are not very different from atoms - alone ' , is what mainly drives the choice of the functions @xmath461 in the basis set , and , in the end , the variational procedure that will be followed is expected to fix the largest failures coming from these too - simplistic assumptions . hence , it is customary to choose functions that are centred at atomic nuclei and that partially resemble the exact solutions for hydrogen - like atoms . in this spirit , the first type of aos to be tried @xcite were the _ later - ype rbitals _ ( stos ) , proposed by slater @xcite and zener @xcite in 1930 : @xmath535 where @xmath536 is a normalization constant and @xmath537 is an adjustable parameter . the index @xmath538 is that of the nucleus at which the function is centred , and , of course , in the majority of cases , there will be several @xmath539 corresponding to different values of @xmath98 centred at the same nucleus . the integers @xmath540 and @xmath541 can be considered quantum numbers , since , due to the fact that the only angular dependence is in @xmath542 ( see below for a definition ) , the sto defined above is still a simultaneous eigenstate of the one - electron angular momentum operators @xmath531 and @xmath533 ( with the origin placed at @xmath543 ) . the parameter @xmath544 , however , should be regarded as a ` principal ( or energy ) quantum number ' only by analogy , since , on the one hand , it does not exist a ` one - atom hamiltonian ' whose exact eigenfunctions it could label and , on the other hand , only the leading term of the laguerre polynomial in eq . ( [ eq : chqc_hydrogen - like ] ) has been kept in the sto if we notice that , within the set of all possible stos ( as defined in eq . ( [ eq : chqc_sto ] ) ) , every hydrogen - like energy eigenfunction ( see eq . ( [ eq : chqc_hydrogen - like ] ) ) can be formed as a linear combination , we easily see that the stos constitute a complete basis set . this is important to ensure that the hartree - fock limit could be actually approached by increasing @xmath460 . ] . additionally , in the above notation , the fact that @xmath539 parametrically depends on the position of a certain @xmath538-th nucleus has been stressed , and the functions @xmath542 , which are called _ real spherical harmonics _ @xcite ( remember that we want to do real rhf ) , are defined in terms of the classical _ spherical harmonics _ @xmath545 by [ eq : chqc_real_spherical_harmonics ] @xmath546 where @xmath547 stands for _ cosine _ , @xmath548 for _ sine _ , the functions @xmath549 are the _ associated legendre polynomials _ @xcite , and the spherical coordinates @xmath550 and @xmath551 also carry the @xmath538-label to remind that the origin of coordinates in terms of which they are defined is located at @xmath543 . also note that , using that @xmath552 , there is the same number of real spherical harmonics as of classical ones . these @xmath539 have some good physical properties . among them , we shall mention that , for @xmath553 , they present a _ cusp _ ( a discontinuity in the radial derivative ) , as required by kato s theorem @xcite ; and also that they decay at an exponential rate when @xmath554 , which is consistent with the image that , an electron that is taken apart from the vicinity of the nucleus must ` see ' , at large distances , an unstructured point - like charge ( see , for example , the sto in fig . [ fig : contraction ] ) . finally , the fact that they do not present radial nodes ( due to the aforementioned absence of the non - leading terms of the laguerre polynomial in eq . ( [ eq : chqc_hydrogen - like ] ) ) can be solved by making linear combinations of functions with different values [ page : foot : r_exponent_idontcare ] of @xmath537 this way of proceeding renders the choice of the exponent carried by the @xmath555 part ( @xmath556 in the case of the sto in eq . ( [ eq : chqc_sto ] ) ) a rather arbitrary one . as a consequence , different definitions may be found in the literature and the particular exponent chosen in actual calculations turns out to be mostly a matter of computational convenience . ] . now , despite their being good theoretical candidates to expand the mo @xmath464 that make up the @xmath4-particle solution of the hartree - fock problem , these stos have serious computational drawbacks : whereas the two - centre integrals ( such as @xmath516 in eq . ( [ eq : chqc_matrices_2 ] ) ) can be calculated analytically , the four - centre eris @xmath505 can not @xcite if functions like the ones in eq . ( [ eq : chqc_sto ] ) are used . this fact , which was known as `` the nightmare of the integrals '' in the first days of computational quantum chemistry @xcite , precludes the use of stos in practical ab initio calculations of large molecules . a major step to overcome these difficulties that has revolutioned the whole field of quantum chemistry @xcite was the introduction of _ artesian aussian - ype rbitals _ ( cgto ) : @xmath557 where the @xmath558 and the @xmath559 , with @xmath560 , are the euclidean coordinates of the electron and the @xmath538-th nucleus respectively , and the integers @xmath561 , @xmath562 and @xmath563 , which take values from 0 to @xmath525 , are called _ _ orbital quantum numbers _ _ since the harmonic - oscillator energy eigenfunctions can be constructed as linear combinations of cartesian gtos , we have that the latter constitute a complete basis set and , like in the case of the stos , we may expect that the hartree - fock limit is approached as @xmath460 is increased . ] . although these gtos do not have the good physical properties of the stos ( compare , for example , the sto and the gto in fig . [ fig : contraction ] ) , in 1950 , boys @xcite showed that all the integrals appearing in scf theory could be calculated analytically if the @xmath461 had the form in eq . ( [ eq : chqc_gto ] ) . the enormous computational advantage that this entails makes possible to use a much larger number of functions to expand the one - electron orbitals @xmath326 if gtos are used , partially overcoming their bad short- and long - range behaviour and making the gaussian - type orbitals the universally preferred choice in scf calculations @xcite . to remedy the fact that the angular behaviour of the cartesian gtos in eq . ( [ eq : chqc_gto ] ) is somewhat hidden , they may be linearly combined to form _ pherical aussian - ype rbitals _ ( sgto ) : @xmath564 which are proportional to the real spherical harmonic @xmath565 , and to which the same remarks made in footnote [ foot : r_exponent_idontcare ] in page for the stos , regarding the exponent in the @xmath555 part , may be applied . the fine mathematical details about the linear combination that relates the cartesian gtos to the spherical ones are beyond the scope of this introduction . we refer the reader to refs . @xcite and @xcite for further information and remark here some points that will have interest in the subsequent discussion . first , the cgtos that are combined to make up a sgto must have all the same value of @xmath566 and , consequently , this sum of the three orbital quantum numbers @xmath561 , @xmath562 and @xmath563 in a particular cartesian gto is typically ( albeit dangerously ) referred to as the _ angular momentum _ of the function . in addition , apart from the numerical value of @xmath540 , the spectroscopic notation is commonly used in the literature , so that cgtos with @xmath567 are called _ s _ , _ p _ , _ d _ , _ f _ , _ g _ , _ h _ , , respectively . where the first four come from the archaic words _ harp _ , _ rincipal _ , _ iffuse _ and _ undamental _ , while the subsequent ones proceed in alphabetical order . second , for a given @xmath568 , there are more cartesian gtos ( @xmath569 ) than spherical ones ( @xmath570 ) , in such a way that , from the @xmath569 functionally independent linear combinations that can be formed using the cgtos of angular momentum @xmath540 , only the angular part of @xmath570 of them turns out to be proportional to a real spherical harmonic @xmath571 ; the rest of them are proportional to real spherical harmonic functions with a different value of the angular momentum quantum number . for example , from the six different d - cartesian gtos , whose polynomial parts are @xmath572 , @xmath573 , @xmath574 , @xmath575 , @xmath576 and @xmath577 ( using an evident , compact notation ) , only five different spherical gtos can be constructed : the ones with polynomial parts proportional to @xmath578 , @xmath576 , @xmath577 , @xmath579 and @xmath575 @xcite . among these new sgtos , which , in turn , are proportional ( neglecting also powers of @xmath580 , see footnote [ foot : r_exponent_idontcare ] in page ) to the real spherical harmonics @xmath581 , @xmath582 , @xmath583 , @xmath584 and @xmath585 , the linear combination @xmath586 is missing , since it presents the angular behaviour of an s - orbital ( proportional to @xmath587 ) . finally , let us remark that , whereas cartesian gtos in eq . ( [ eq : chqc_gto ] ) are easier to be coded in computer applications than sgtos@xcite , it is commonly accepted that these spurious spherical orbitals of lower angular momentum that appear when cgtos are used do not constitute efficient choices to be included in a basis set @xcite ( after all , if we wanted an additional s - function , why include it in such an indirect and clumsy way instead of just designing a specific one that suits our particular needs ? ) . consequently , the most common practice in the field is to use cartesian gtos removing from the basis sets the linear combinations such as the @xmath586 above . now , even if the integrals involving cgtos can be computed analytically , there are still @xmath512 of them in a scf calculation . for example , in the model dipeptide hco - l - ala - nh@xmath305 , which is commonly used to mimic an alanine residue in a protein @xcite , there are 62 electrons and henceforth 31 rhf spatial orbitals @xmath464 . if a basis set with only 31 functions is used ( this is a lower bound that will be rarely reached in practical calculations due to symmetry issues , see below ) , near a million of four - centre @xmath588 integrals must be computed . this is why , one must use the freedom that remains once the decision of sticking to cgtos has been taken ( namely , the choice of the _ exponents _ @xmath537 and the _ angular momentum _ @xmath540 ) to design basis sets that account for the relevant behaviour of the systems studied while keeping @xmath460 below the ` pain threshold ' . the work by nobel prize john pople s group has been a major reference in this discipline , and their sto-@xmath151 g family @xcite , together with the split - valence gaussian basis sets , 3 - 21 g , 4 - 31 g , 6 - 31 g , etc . @xcite , shall be used here to exemplify some relevant issues . however , note that most of the concepts introduced are also applicable to more modern basis sets , such as those by dunning @xcite . to begin with , let us recall that the short- and long - range behaviour of the slater - type orbitals in eq . ( [ eq : chqc_sto ] ) is better than that of the more computationally efficient gtos . in order to improve the physical properties of the latter , it is customary to linearly combine @xmath589 cartesian gtos , denoted now by @xmath590 ( @xmath591 ) , and termed _ rimitive aussian - ype rbitals _ ( pgto ) , having the same atomic centre @xmath543 , the same set of orbital quantum numbers , @xmath561 , @xmath562 and @xmath563 , but different exponents @xmath592 , to make up a _ ontracted aussian - ype rbitals _ ( cgto ) , defined by @xmath593 where the normalization constants @xmath594 have been kept inside the sum because they typically depend on @xmath595 . also , we denote now by @xmath596 the number of contracted gtos and , by @xmath597 , the number of primitive ones . in the sto-@xmath151 g family of basis sets , for example , @xmath151 primitive gtos are used for each contracted one , fitting the coefficients @xmath598 and the exponents @xmath592 to resemble the radial behavior of slater - type orbitals @xcite . in fig . [ fig : contraction ] , the 1s - contracted gto ( see the discussion below ) of the hydrogen atom in the sto-3 g basis set is depicted , together with the three primitive gtos that form it and the sto that is meant to [ page : foot : basisform ] be approximated basis sets were obtained from the extensible computational chemistry environment basis set database at ` http://www.emsl.pnl.gov/forms/basisform.html ` , version 02/25/04 , as developed and distributed by the molecular science computing facility , environmental and molecular sciences laboratory which is part of the pacific northwest laboratory , p.o . box 999 , richland , washington 99352 , usa , and funded by the u.s . department of energy . the pacific northwest laboratory is a multi - program laboratory operated by battelle memorial institute for the u.s . department of energy under contract de - ac06 - 76rlo 1830 . contact karen schuchardt for further information . ] . we can see that the contracted gto has a very similar behavior to the sto in a wide range of distances , while the single gto that is also shown in the figure ( with the same norm and exponent as the sto ) has not . typically , the fitting procedure that leads to contracted gtos is performed on isolated atoms and , then , the already mentioned chemical intuition that states that ` atoms - in - molecules are not very different from atoms - alone ' is invoked to keep the linear combinations fixed from there on . obviously , better results would be obtained if the contraction coefficients were allowed to vary . moreover , the number of four - centre integrals that need to be calculated depends on the number of _ primitive _ gtos ( like @xmath599 ) , so that we have not gained anything on this point by contracting . however , the size of the variational space is @xmath596 ( i.e. , the number of _ contracted _ gtos ) , in such a way that , once the integrals @xmath588 are calculated ( for non - direct scf ) , all subsequent steps in the iterative self - consistent procedure scale as powers of @xmath596 . also , the disk storage ( again , for non - direct schemes ) depends on the number of _ contracted _ gtos and , frequently , it is the disk storage and not the cpu time the limiting factor of a calculation . an additional chemical concept that is usually defined in this context and that is needed to continue with the discussion is that of _ shell _ : atomic shells in quantum chemistry are defined analogously to those of the hydrogen atom , so that each electron is regarded as ` filling ' the multi - electron atom ` orbitals ' according to hund s rules @xcite . hence , the _ occupied shells _ of carbon , for example , are defined to be 1s , 2s and 2p , whereas those of , say , silicon , would be 1s , 2s , 2p , 3s an 3p . each shell may contain @xmath600 electrons if complete , where @xmath601 accounts for the orbital angular momentum multiplicity and the 2 factor for that of electron spin . [ tab : contraction ] thus , using these definitions , all the basis sets in the aforementioned sto-@xmath151 g family are _ minimal _ ; in the sense that they are made up of only @xmath601 contracted gtos for each completely or partially occupied shell , so that the sto-@xmath151 g basis sets for carbon , for example , contain two s - type contracted gtos ( one for the 1s- and the other for the 2s - shell ) and three p - type ones ( belonging to the 2p - shell ) . moreover , due to rotational - symmetry arguments in the isolated atoms , all the @xmath601 functions in a given shell are chosen to have the same exponents and the same contraction coefficients , differing only on the polynomial that multiplies the gaussian part . such @xmath601 cgtos shall be said to constitute a _ aussian hell _ ( gs ) , and we shall also distinguish between the _ rimitive _ ( pgs ) and _ ontracted _ ( cgs ) versions . in table [ tab : contraction ] , the exponents @xmath595 and the contraction coefficients @xmath598 of the primitive gtos that make up the three different shells in the sto-3 g basis set for carbon are presented ( see ref . @xcite and footnote [ foot : basisform ] in page ) . the fact that the exponents @xmath595 in the 2s- and 2p - shells are constrained to be the same is a particularity of some basis sets ( like this one ) which saves some computational effort and deserves no further attention . next , let us introduce a common notation that is used to describe the contraction scheme : it reads ( _ primitive shells _ ) / [ _ contracted shells _ ] , or alternatively ( _ primitive shells _ ) @xmath602 [ _ contracted shells _ ] . according to it , the sto-3 g basis set for carbon , for example , is denoted as ( 6s,3p ) @xmath602 [ 2s,1p ] , or ( 6,3 ) @xmath602 [ 2,1 ] . moreover , since for organic molecules it is frequent to have only hydrogens and the 1st - row atoms c , n and o ( whose occupied shells are identical ) in proteins , one may also have sulphur in cysteine and methionine residues . ] , the notation is typically extended and the two groups of shells are separated by a slash ; as in ( 6s,3p/3s ) @xmath602 [ 2s,1p/1s ] for sto-3 g . [ tab : contraction2 ] the first improvement that can be implemented on a minimal basis set such as the ones in the sto-@xmath151 g is the _ splitting _ , which consists in including more than one gaussian shell for each occupied one . if the splitting is evenly performed , i.e. , each shell has the same number of gss , then the basis set is called _ ouble eta _ ( dz ) , _ riple eta _ ( tz ) , _ uadruple eta _ ( qz ) , _ uintuple eta _ ( 5z ) , _ extuple eta _ ( 6z ) , and so on ; where the word _ zeta _ comes from the greek letter @xmath603 used for the exponents . a hypothetical tz basis set in which each cgto is made by , say , four primitive gtos , would read ( 24s,12p/12s ) @xmath602 [ 6s,3p/3s ] in the aforementioned notation . at this point , the already familiar intuition that says that ` atoms - in - molecules are not very different from atoms - alone ' must be refined with another bit of chemical experience and qualified by noticing that ` core electrons are less affected by the molecular environment and the formation of bonds than valence electrons ' recall that , for the very concept of ` core ' or ` valence electrons ' ( actually for any label applied to a single electron ) to have any sense , we must be in the hartree - fock formalism ( see footnote [ foot : orbitals_only_in_hf ] in page ) . ] . in this spirit , the above evenness among different shells is typically broken , and distinct basis elements are used for the energetically lowest lying ( _ core _ ) shells than for the highest lying ( _ valence _ ) ones . on one side , the contraction scheme may be different . in which case , the notation used up to now becomes ambiguous , since , for example , the designation ( 6s,3p/3s ) @xmath602 [ 2s,1p/1s ] , that was said to correspond to sto-3 g , would be identical for a _ different _ basis set in which the 1s - gaussian shell of heavy atoms be formed by 4 pgss and the 2s - gaussian shell by 2 pgss ( in 1st - row atoms , the 2s- and 2p - shells are defined as valence and the 1s - one as core , while in hydrogen atoms , the 1s - shell is a valence one ) . this problem can be solved by explicitly indicating how many primitive gss form each contracted one , so that , for example , the sto-3 g basis set is denoted by ( 33,3/3 ) @xmath602 [ 2,1/1 ] , while the other one mentioned would be ( 42,3/3 ) @xmath602 [ 2,1/1 ] ( we have chosen to omit the angular momentum labels this time ) . the other point at which the core and valence gaussian shells may differ is in their respective ` zeta quality ' , i.e. , the basis set may contain a different number of contracted gaussian shells in each case . for example , it is very common to use a single cgs for the core shells and a multiple splitting for the valence ones . these type of basis sets are called _ split - valence _ and the way of naming their quality is the same as before , except for the fact that a capital v , standing for _ alence _ , is added either at the beginning or at the end of the acronyms dz , tz , qz , etc . , thus becoming vdz , vtz , vqz , etc . or dzv , tzv , qzv , etc . pople s 3 - 21 g @xcite , 4 - 31 g @xcite , 6 - 31 g @xcite and 6 - 311 g @xcite are well - known examples of split - valence basis sets that are commonly used for scf calculations in organic molecules and that present the two characteristics discussed above . their names indicate the contraction scheme , in such a way that the number before the dash represents how many primitive gss form the single contracted gss that is used for core shells , and the numbers after the dash how the valence shells are contracted , in much the same way as the notation in the previous paragraphs . for example , the 6 - 31 g basis set ( see table [ tab : contraction2 ] ) , contains one cgs made up of six primitive gss in the 1s - core shell of heavy atoms ( the 6 before the dash ) and two cgs , formed by three and one pgss respectively , in the 2s- and 2p - valence shells of heavy atoms and in the 1s - shell of hydrogens ( the 31 after the dash ) . the 6 - 311 g basis set , in turn , is just the same but with an additional single - primitive gaussian shell of functions in the valence region . finally , to fix the concepts discussed , let us mention that , using the notation introduced above , these two basis sets may be written as ( 631,31/31 ) @xmath602 [ 3,2/2 ] and ( 6311,311/311 ) @xmath602 [ 4,3/3 ] , respectively . two further improvements that are typically used and that may also be incorporated to pople s split - valence basis sets are the addition of _ polarization _ @xcite or _ diffuse functions _ @xcite . we shall discuss them both to close both this section and the work . up to now , neither the contraction nor the splitting involved gtos of larger angular momentum than the largest one among the occupied shells . however , the molecular environment is highly anisotropic and , for most practical applications , it turns out to be convenient to add these _ polarization _ ( large angular momentum ) gaussian shells to the basis set , since they present lower symmetry than the gss discussed in the preceding paragraphs . typically , the polarization shells are single - primitive gss and they are denoted by adding a capital p to the end of the previous acronyms , resulting into , for example , dzp , tzp , vqzp , etc . , or , say , dz2p , tz3p , vqz4p if more than one polarization shell is added . in the case of pople s basis sets @xcite , these improvements are denoted by specifying , in brackets and after the letter g , the number and type of the polarization shells separating heavy atoms and hydrogens by a comma there also exists an old notation for the addition of a single polarization shell per atom that reads 6 - 31 g * * and that is equivalent to 6 - 31g(d , p ) . ] . for example , the basis set 6 - 31g(2df , p ) contains the same gaussian shells as the original 6 - 31 g plus two d - type shells and one f - type shell centred at the heavy atoms , as well as one p - type shell centred at the hydrogens . finally , for calculations in charged species ( specially anions ) , where the charge density extends further in space and the tails of the distribution are more important to account for the relevant behaviour of the system , it is common to augment the basis sets with _ diffuse functions _ , i.e. , single - primitive gaussian shells of the same angular momentum as some preexisting one but with a smaller exponent @xmath603 than the smallest one in the shell . in general , this improvement is commonly denoted by adding the prefix _ aug_- to the name of the basis set . in the case of pople s basis sets , on the other hand , the insertion of a plus sign ` + ' between the contraction scheme and the letter g denotes that the set contains one diffuse function in the 2s- and 2p - valence shells of heavy atoms . a second + indicates that there is another one in the 1s - shell of hydrogens . for example , one may have the doubly augmented ( and doubly polarized ) 6 - 31++g(2d,2p ) basis set . the ground - breaking advances reviewed in the previous sections allow to routinely calculate , using hartree - fock scf methods , physical properties of molecules of tens of atoms in present day computers . however , the simplest algorithms that can be devised to perform the limiting steps in such calculations are far from optimal . this large room for improvement , which may be enough to accommodate the exciting possibility of linearly scaling approaches that could open the door to thousands atoms computations , has been steering many innovative lines of research in the last years to close this review , we shall briefly outline here some of the hottest areas of modern development related to the topics discussed , specially those aimed to the reduction of the in principle @xmath512 complexity associated to the calculation of the @xmath588 eris in eq . ( [ eq : chqc_twoe_integrals ] ) , as well as the @xmath515 cost of the diagonalization of the fock operator in eq . ( [ eq : chqc_rh_2 ] ) in principle , careful distinction must be made between the number of primitive gtos @xmath604 and the number of contracted ones @xmath596 ( see the previous section ) . however , since in this section we will be dealing only with approximate scalings without worrying much about the prefactor , the ` neutral ' notation @xmath460 has been chosen to denote a quantity which is linear on both @xmath604 and @xmath596 . ] . for wider reviews on these topics , we recommend to the interested reader the accounts in refs . @xcite . the first class of attempts to reduce the cost associated to the calculation of the @xmath512 eris @xmath505 are those aimed to the improvement of the algorithms for analytically calculating them _ without approximations_. there are basically two issues that render the construction of these algorithms a non trivial task : first , the fact that the only four - center eris that can be straightforwardly computed are the ones corresponding to a product of four s - type gtos , while higher angular momentum eris may obtained from them in a non unique way . after using the _ gaussian product rule _ @xcite ( see the previous section for the notation used ) @xmath605 which allows the four - centre integral to be turned into a two - center one , and whose absence for the case of stos is the essential reason for their being non practical , we can turn the 6-dimensional eri into a simple one - dimensional integral that can be readily calculated by different means @xcite : @xmath606 with [ eq : chqc_fund_expl ] @xmath607 such an integral is called a _ fundamental _ eri and , as we said , eris involving gtos with higher angular momentum than @xmath608 are obtained from this fundamental one through iterative differentiations of eq . ( [ eq : chqc_fund ] ) with respect to the nuclear positions , leading to recurrence relations expressing eris of a given angular momentum as a function of the lower angular momentum ones . the particular flavour of these recurrence relations that is used is one of the matters in which the algorithms for calculating eris differ . the second issue that renders the construction of algorithms for computing eris non - trivial is related to how the contraction of primitive gtos is handled . in the geminal paper by boys @xcite , the most naive procedure was suggested , namely , the conversion of each eri of contracted gtos into a quadruple sum of eris of primitive gtos . however , this does not take profit , for example , from the fact , mentioned in the previous section , that all cgtos in the same contracted gaussian shell are formed by pgtos with the same set of exponents @xmath609 . much profit can be taken from this and other constraints and , in fact , the cost - scaling profile of each algorithm with the contraction degree @xmath589 of the cgtos is strongly correlated to the moment at which the transformation between cgtos and pgtos is performed @xcite . among the most used of these analytical algorithms , we can mention the pople - hehre ( ph ) one @xcite , which additionally exploits the fact that for each four - center eri of low angular momentum there is a privileged cartesian axis system in which many primitive integrals vanish by symmetry ; the _ cmurchie - avidson _ ( md ) approach @xcite , which avoids the rotation in ph thus being more efficient for high angular momentum eris ; the _ bara - aika - chlegel _ ( oss ) algorithm @xcite ; and a better defined and improved version of it : the _ ead - ordon - ople _ ( hgp ) algorithm @xcite . finally , if the moment at which the contraction is handled is chosen dynamically depending on the type of gto appearing in the eri , we have the prism modifications of md and hgp : the md - prism @xcite and hgp - prism @xcite algorithms , as well as a generalization of all the previous methods , called cold prism @xcite . now , even if we implement any of these efficient methods for calculating the eris , there is still @xmath512 of them . this scaling is simply too harsh for a too large class of applications . therefore , the next natural step is to try to devise approximate methods that minimize the necessary decrease in accuracy at the same time that maximize the savings in computer time . thanks to the particular characteristics of the eris , the gaussian basis functions and the concrete physical problem intended to solve the locality of many - electron quantum mechanics @xcite , which is related to the _ nearsightedness _ concept introduced by kohn @xcite , is the main property that allows to finally achieve linearity . ] , this endeavour has been successfully pursued by many researchers and the ` holy grail ' @xcite of linear scaling with @xmath460 is asymptotically approaching current calculations in large systems . here , we shall discuss the basic issues that make this possible . for more in depth reviews , we point the readers to the accounts in references @xcite . the first point to consider in order to reduce the @xmath512 scaling attains the so - called _ radial overlap_. if we take a look at eq . ( [ eq : chqc_contracted_gto ] ) , we can see that , irrespective of the polynomial prefactor which contains all the angular dependence of the gto , every function @xmath610 contains a radial exponential part ( actually , a sum of exponentials ) . therefore , if we consider any product @xmath611 of two gtos , we will always find a multiplying sum of terms such as those depicted in the expression for the gaussian product rule in ( [ eq : chqc_gpr ] ) . the exponential decay of all quantities @xmath612 in those terms with the distance @xmath613 between the nuclei on which the gtos are centred indicates that , among the @xmath614 possible pair products @xmath615 , only @xmath616 of them will be non - negligible . to see this , note that if we fix ( say ) @xmath98 , the only values of @xmath470 that will yield a non - negligible product @xmath611 are those for which @xmath613 is ` small ' . since atoms do not interpenetrate in most of the conformations that shall be studied , this can only happen for a number of different @xmath470 s which is not @xmath616 but a constant independent of the size of the molecule . there are @xmath460 different possible values of @xmath98 for which the above reasoning can be repeated , and the result follows . as a consequence , if only @xmath616 pairs @xmath617 are non - negligible , then only @xmath618 eris @xmath588 may in principle contribute in a significant way to the fock operator in eq . ( [ eq : chqc_matrices_2 ] ) and not @xmath512 . ( for similar estimations based on slightly different hypotheses , see @xcite . ) one must also note that the discussion is complicated by the fact that the eris do not appear alone , but contracted with the density matrix elements @xmath619 . this allows for further improvements of the scaling beyond @xmath618 which are discussed later in this section . now , knowing that most of the eris are too small to be relevant , _ we do not know which ones _ and , if we calculated the @xmath512 of them in order to spot the little ones , then we would have not gained anything . this simple argument shows the necessity of finding a set of _ estimators _ that allow us to selectively drop eris _ without calculating them_. of course , the number of estimators must also scale at worst like @xmath618 in order for the scheme to be useful . one of the first and simplest such estimators , was introduced by almlf et al . at the same time that they proposed the direct scf method @xcite . in their scheme , each eri @xmath588 was approximated by the corresponding radial overlap factor @xmath620 ( note that there are only @xmath618 numbers @xmath612 ) . although this estimator was relatively successful , it presented the important drawback of not being an upper - bound for the eris , thus rendering the control of errors an a priori impossible task . in order to overcome this problem , hser and ahlrichs @xcite later proposed a different estimator which can be assured to be always greater than the associated eri . following them , after the proof of positive definiteness by roothaan @xcite , the electrostatic interaction energy between two continuous charge distributions , @xmath621 can be easily shown to satisfy the properties of an inner product . hence , if we choose @xmath622 , and @xmath623 , we can use the well - known schwarz inequality to show that @xmath624 in such a way that , by calculating only the @xmath614 different two - index eris in the last term , we can safely bound from above the whole set containing @xmath512 of them . after these seminal works , more sophisticated and tighter bounds have been developed through the years @xcite , their application being nowadays routine in quantum chemistry packages . in a second generation of methods , the @xmath618 formal scaling achieved in practical calculations @xcite by using the above ideas has been recently attacked . to this end , more specifically physical properties of the problem are exploited ( see footnote [ locality ] in page ) , and different strategies are used to deal with the coulomb and exchange parts of the fock operator in ( [ eq : chqc_matrices_2 ] ) . the main difference between the behaviours of these two contributions lies in the way in which the eris are contracted with the density matrix @xmath619 : in the ` classical ' coulomb part , the relative sizes of the eris @xmath588 are largely correlated with the relative sizes of the associated elements @xmath619 @xcite , so that no decrease in the @xmath625 scaling is expected from density matrix considerations . differently , in the exchange terms , the fact that the elements @xmath619 couple ` exchanged ' indices in the @xmath626 eris produces cancellations which make this type of ` quantum ' contributions rather short - range ( for non - metallic species ) @xcite . hence , as for every short - range interaction , a number of terms scaling _ linearly _ with the system size is expected . apart from the different treatment that this difference suggests , note that the separation of the coulomb and exchange tasks allows for improved parallelization the computer codes @xcite . for the exchange part , the aforementioned short - range behaviour allows to devise @xmath616 algorithms just by intelligently ordering the loops in which the eris are calculated . the fine details of these methods are rather technical and they are beyond the scope of this review . we point the reader to @xcite and references therein for further information . in the coulomb case , on the other hand , @xmath618 terms still enter the sum in ( [ eq : chqc_matrices_2 ] ) , and more physically - based approximations must be used . the _ ontinuous ast ultipole ethod _ ( cfmm ) by white et al . @xcite , for example , is probably one of the most celebrated algorithms for calculating the coulomb part of the fock operator . it is a generalization for continuous charge distributions of the _ ast ultipole ethod _ ( fmm ) , introduced by greengard and rokhlin @xcite in a ground - breaking paper and aimed for point - like charges . in both fmm and cfmm , a clever hierarchical tree - like division of the space into cells is performed , and the far away regions are approximated via truncated multipole expansions . these two ingredients , which allow to calculate the coulomb contribution in @xmath616 steps for large systems , are common to most of the fast coulomb algorithms . despite this similarity , the room for improvement seems still large enough to accommodate a vigorous field with many publications appearing each year . let us mention here , for example , the _ eneralized ell ultipole ethod _ ( gcmm ) by kutteh et al . , which can use moments higher than monopole @xcite ; the _ uantum hemical ree ode _ ( qctc ) by challacombe et al . @xcite , which independently thresholds ` bra ' and ` ket ' distributions ; and the _ aussian ery ast ultipole ethod _ ( gvfmm ) by strain et al . @xcite , which benefits from the idea , introduced in @xcite for point - like charges , of using a dynamical maximum angular momentum to further speed up the calculations note however , that the near - field contributions in these methods are still calculated without approximations and represent a great portion of the computer time . in this line , some modern algorithms are appearing to alleviate this part of the work , such as , for example , the _ j matrix engine _ by white and head - gordon @xcite , which uses and improves the ideas discussed in the first part of this section about analytically calculating the eris ; or the method by izmaylov et al . @xcite , which implements a hierarchy of screening levels to eliminate negligible integrals . according to recent reports @xcite , the combination of cfmm , with the j matrix engine technique and with the _ ourier ransform oulomb _ ( ftc ) method by fsti - molnr and pulay @xcite is nowadays probably the fastest way for assembling the coulomb matrix . now , once the construction process of the whole fock matrix @xmath499 ( via its coulomb and exchange parts ) has been cast into the form of an @xmath616 algorithm , the importance of the second rate - limiting step in scf procedures , the diagonalization of @xmath499 , comes into focus . although the prefactor of the ( in principle ) @xmath515 diagonalization step is very small and , for systems of less than a few thousands of atoms , the absolute time spent on it is smaller than the one needed for the formation of the fock matrix @xcite , it is clear that , in the long run , it will dominate the calculations in larger systems and will become the relevant bottleneck @xcite . we shall point out the essentials regarding the methods aimed to reduce the scaling the diagonalization step . for more in depth reviews on the topic , we suggest to the reader the accounts in @xcite . the @xmath515 complexity of classical diagonalization methods ( such as the givens - householder one @xcite ) can be easily understood if we think that the core of the algorithm is just the multiplication of @xmath486 matrices . nevertheless , if the matrices multiplied are _ sparse _ , i.e. , they have a number of non - negligible elements that scale not as @xmath618 but as @xmath616 , then the product can be obtained in @xmath616 steps . as a result of the already mentioned locality properties of many - body quantum mechanics ( see footnote [ locality ] in page ) , some matrices appearing in the scf methods discussed in this review , namely , the density matrix @xmath627 and the fock operator @xmath499 , are indeed sparse if the system is non - metallic , i.e. , if it has a non - vanishing homo - lumo gap @xcite . the idea behind most of the modern algorithms for achieving ( or avoiding ) diagonalization with @xmath616 effort consists essentially in performing all operations using only these ( local ) sparse matrices , and avoiding ( non - local ) dense ones , such as the mo coefficients matrix @xmath628 . the attempts to improve the scaling of the diagonalization steps fall basically in two groups @xcite . in the first one , profit is taken from the use of mos which , instead of being extended over the whole molecule , such as the canonical orbitals used in the previous sections , are _ localized _ in a small region of space . in these class of methods , diagonalization @xcite or pseudo - diagonalization ( annihilation of the occupied - virtual blocks of the fock matrix ) @xcite is still performed , and the @xmath616 scaling is achieved because the representation of all operators in the basis of localized mos is sparse . the second group of algorithms do not use the mos as variables but the density matrix itself . among them , two subfamilies of methods may be found : in the first one , the search for the optimal density matrix is simply treated as a standard optimization problem , being the score function the hf energy , and the variables the density matrix elements or a set of parameters of some suitable truncated expansion of it @xcite . see , for example , the approaches by ochsenfeld and head - gordon @xcite , by salek et al . @xcite , by millam and scuseria @xcite , or by ordejn et al . the other subfamily of density matrix - based methods use iterative procedures , in such a way that , at each step , the fock operator is considered fixed and the equations are solved for the density matrix ( much in the spirit of the mos - based algorithms discussed in the previous sections ) . in this group , we can find , for example , the approach in @xcite using the lanczos algorithm @xcite , or the method by helgaker et al . @xcite . these algorithms , combined with new strategies that also avoid diagonalization and improve scf convergence properties , such as the one described in @xcite , represent the final step towards linear hartree - fock methods in quantum chemistry . to close this section , although we have been concerned , up to now , with the calculation of the electronic ground - state given a fixed position of the nuclei , let us stress that it is also very common to use quantum chemical methods for finding the local energy minima of molecules . to this end , geometry optimizations must be performed and not only must we be able to compute the energy of the molecule , but also their derivatives with respect to the nuclear coordinates . additionally , these derivatives are also needed to do ab initio molecular dynamics in the ground - state born - oppenheimer pes @xcite . the most naive approach , namely , the computation of the gradient of the energy @xmath629 ( using a simpler notation for it than @xmath630 in ( [ eq : chqc_v_eff ] ) ) by finite differences , is very inefficient for anything but the smallest molecules . to see this , one just need to notice that the gradient has as many components as the system degrees of freedom @xmath151 hence , in order to obtain it , say , at a point @xmath631 , we would have to compute @xmath632 times a single point energy , in order to know @xmath633 and @xmath634 , being @xmath635 , with @xmath636 , a small displacement in each of the nuclear degrees of freedom . this drawback was overcome in the late 60s by pulay and others ( see @xcite and references therein ) with the introduction of the so - called _ analytical derivatives _ , in which the gradient ( and higher - order derivatives ) are expressed , like the energy itself , just as a function of eris involving the wavefunction at the point @xmath631 . this marked an inflexion point in the development of optimization and molecular dynamics algorithms that continues nowadays , as analytical derivatives are routinely introduced together with almost any new method for calculating the energy . in relation to the improvements reviewed in this work , for example , let us note that , in @xcite , the extension of the j matrix engine method to calculate the derivatives of the coulomb part with respect of the nuclei coordinates is introduced ; in @xcite , linear scaling exchange gradients are developed ; analytic derivatives for the gvfmm are provided in @xcite ; the hgp algorithm for calculating the eris is extended to the computation of derivatives of the eris as well in @xcite ; and we may find similar developments for the ftc method @xcite or for algorithms that achieve diagonalization with linear effort @xcite . this work has been supported by the research projects e24/3 and pm048 ( aragn government ) , mec ( spain ) and mcyt ( spain ) . p. echenique has been supported by a bifi research contract and by a mec ( spain ) postdoctoral grant . a _ functional _ @xmath78 $ ] is a mapping that takes functions to numbers ( in this work , only functionals in the real numbers are considered ) : @xmath637 \end{array}\ ] ] for example , if the function space @xmath638 is the hilbert space of square - integrable functions @xmath91 ( the space of states of quantum mechanics ) , the objects in the domain of @xmath639 ( i.e. , the functions in @xmath91 ) can be described by infinite - tuples @xmath640 of complex numbers and @xmath639 may be pictured as a function of infinite variables . when dealing with function spaces @xmath638 that meet certain requirements we will not discuss the issue further but let it suffice to say that @xmath91 does satisfy these requirements . ] , the limit on the left - hand side of the following equation can be written as the integral on the right - hand side : @xmath641 - \mathcal{f}[\psi_{0}]}{\varepsilon } : = \int \frac{\delta \mathcal{f}[\psi_{0}]}{\delta\psi}(x ) \delta\psi(x ) \mathrm{d}x\ , \ ] ] where @xmath66 denotes a point in the domain of the functions in @xmath638 , and the the object @xmath642 / \delta\psi)(x)$ ] ( which is a function of @xmath66 not necessarily belonging to @xmath638 ) is called the _ functional derivative _ of @xmath78 $ ] in the in the _ point _ @xmath643 . one common use of this functional derivative is to find stationary points of functionals . a function @xmath643 is said to be an _ stationary point _ of @xmath78 $ ] if : @xmath644}{\delta\psi}(x ) = 0 \ .\ ] ] in order to render this definition operative , one must have a method for computing @xmath645 / \delta\psi)(x)$ ] . interestingly , it is possible , in many useful cases ( and in all the applications of the formalism in this work ) , to calculate the sought derivative directly from the left - hand side of eq . ( [ eq : def_funct_der ] ) . the procedure , in such a situation , begins by writing out @xmath646 $ ] and clearly separating the different orders in @xmath647 . secondly , one drops the terms of zero order ( by virtue of the subtraction of the quantity @xmath648 $ ] ) and those of second order or higher ( because they vanish when divided by @xmath647 and the limit @xmath649 is taken ) . the remaining terms , all of order one , are divided by @xmath647 and , finally , @xmath645 / \delta\psi)(x)$ ] is identified out of the resulting expression ( which must written in the form of the right - hand side of eq . ( [ eq : def_funct_der ] ) . for a practical example of this process , see secs . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] . very often , when looking for the stationary points of a function ( or a functional ) , the search space is not the whole one , in which the derivatives are taken , but a certain subset of it defined by a number of constraints . an elegant and useful method for solving the constrained problem is that of the _ lagrange multipliers_. although it can be formally generalized to infinite dimensions ( i.e. , to functionals , see appendix a ) , here we will introduce the method in @xmath650 in order to gain some geometrical insight and intuition . the general framework may be described as follows : we have a differentiable function @xmath651 that takes points in @xmath650 to real numbers and we want to find the stationary points of @xmath97 restricted to a certain subspace @xmath652 of @xmath650 , which is defined by @xmath653 _ constraints _ if the constraints are functionally independent , one must also ask that @xmath654 . if not , @xmath652 will be either a point ( if @xmath655 ) or empty ( if @xmath656 ) . ] : the points that are the solution of the constrained problem are those @xmath658 belonging to @xmath652 where the first order variation of @xmath97 would be zero if the derivatives were taken ` along ' @xmath652 . in other words , the points @xmath658 where the gradient @xmath659 has only components ( if any ) in directions that ` leave ' @xmath652 ( see below for a rigorous formalization of these intuitive ideas ) . thus , when comparing the solutions of the unconstrained problem to the ones of the constrained problem , three distinct situations arise ( see fig . [ fig : lagrange ] ) : 1 . a point @xmath658 is a solution of the unconstrained problem ( i.e. it satisfies @xmath660 ) but it does not belong to @xmath652 . hence , it is not a solution of the constrained problem . this type of point is depicted as a white - filled circle in fig . [ fig : lagrange ] . a point @xmath658 is a solution of the unconstrained problem ( i.e. it satisfies @xmath660 ) and it belongs to @xmath652 . hence , it is also a solution of the constrained problem , since , in particular , the components of the gradient in directions that do not leave @xmath652 are zero . this type of point is depicted as a gray - filled circle inside a white - filled circle in fig . [ fig : lagrange ] . 3 . a point @xmath658 is not a solution of the unconstrained problem ( i.e. , one has ) but it belongs to @xmath652 and the only non - zero components of the gradient are in directions that leave @xmath652 . hence , it is a solution of the constrained problem . this type of point is depicted as a gray - filled circle in fig . [ fig : lagrange ] . from this discussion , it can be seen that , in principle , no conclusions about the number ( or existence ) of solutions of the constrained problem may be drawn only from the number of solutions of the unconstrained one . this must be investigated for each particular situation . in fig . [ fig : lagrange ] , an schematic example in @xmath125 is depicted . the constrained search space @xmath652 is a 2-dimensional surface and the direction note that , only if @xmath661 , i.e. , if the dimension of @xmath652 is @xmath195 , there will be a vector perpendicular to the constrained space . for @xmath662 , the dimensionality of the vector space of the directions in which one ` leaves ' @xmath652 will be also larger than 1 . ] in which one leaves @xmath652 is shown at several points as a perpendicular vector @xmath663 . in such a case , the criterium that @xmath659 has only components in the direction of leaving @xmath652 may be rephrased by asking @xmath659 to be parallel to @xmath663 , i.e. , by requiring that there exists a number @xmath71 such that @xmath664 . the case @xmath665 is also admitted and the explanation of the minus sign will be given in the following . it is clear that , requiring the gradient of @xmath667 to be zero , one recovers the condition @xmath669 , which is satisfied by the points solution of the constrained stationary points problem . if one also asks that the derivative of @xmath667 with respect to @xmath71 be zero , the constraint @xmath670 that defines @xmath652 is obtained as well . this process illustrates the _ lagrange multipliers _ method in this particular example . in the general case , described by eq . ( [ eq : constraints ] ) and the paragraph above it , it can be proved that the points @xmath658 which are stationary subject to the constraints imposed satisfy also , it is worth remarking here that any two pair of functions , @xmath674 and @xmath675 , of @xmath650 whose restrictions to @xmath652 are equal ( i.e. , that satisfy @xmath676 ) obviously represent the same constrained problem and they may be used indistinctly to construct the auxiliary function @xmath667 . this fact allows us , after having constructed @xmath667 from a particular @xmath97 , to use the equations of the constraints to change @xmath97 by another simpler function which is equal to @xmath97 when restricted to @xmath652 . this freedom is used to derive the hartree and hartree - fock equations , in secs . [ sec : qc_hartree ] and [ sec : qc_hartree - fock ] , respectively . the formal generalization of these ideas to functionals ( see appendix a ) is straightforward if the space @xmath650 is substituted by a functions space @xmath639 , the points @xmath658 by functions , the functions @xmath97 , @xmath677 and @xmath667 by functionals and the requirement that the gradient of a function be zero by the requirement that the functional derivative of the analogous functional be zero . finally , let us stress something that is rarely mentioned in the literature : _ there is another ( older ) method , apart from the lagrange multipliers one , for solving a constrained optimization problem : simple substitution . i.e. , if we can find a set of @xmath678 independent _ adapted coordinates _ that parameterize @xmath652 and we can write the score function @xmath97 in terms of them , we would be automatically satisfying the constraints . actually , in practical cases , the method chosen is a suitable combination of the two ; in such a way that , if substituting the constraints in @xmath97 is difficult , the necessary lagrange multipliers are introduced to force them , and vice versa . as a good example of this , the reader may want to check the derivation of the hartree equations in sec . [ sec : qc_hartree ] ( or the hartree - fock ones in sec . [ sec : qc_hartree - fock ] ) . there , we start by proposing a particular form for the total wavefunction @xmath112 in terms of the one - electron orbitals @xmath326 ( see eq . ( [ eq : chqc_constraint_total_h ] ) ) and we write the functional @xmath639 ( which is the expected value of the energy ) in terms of that special @xmath112 ( see eq . ( [ eq : chqc_constrained_functional_h ] ) ) . in a second step , we impose the constraints that the one - particle orbitals be normalized ( @xmath679 ) and force them by means of @xmath4 lagrange multipliers @xmath680 . despite the different treatments , both conditions are constraints standing on the same footing . the only difference is not conceptual , but operative : for the first condition , it would be difficult to write it as a constraint ; while , for the second one , it would be difficult to define adapted coordinates in the subspace of normalized orbitals . so , in both cases , the easiest way for dealing with them is chosen . mackerell jr . , a. d. , brooks , b. , brooks iii , c. l. , nilsson , l. , roux , b. , won , y. , and karplus , m. in _ the encyclopedia of computational chemistry _ , p. v. r. schleyer , p. r. schreiner , n. l. allinger , t. clark , j. gasteiger , p. kollman and h. f. schaefer iii ( eds ) , pp . 217277 , john wiley & sons , chichester ( 1998 ) . cancs , e. , defranceschi , m. , kutzelnigg , w. , le bris , c. , and maday , y. in _ handbook of numerical analysis . volume x : special volume : computational chemistry _ , p. ciarlet and c. le bris ( eds ) , pp . 3270 , elsevier ( 2003 ) . schlegel , h. b. and mcdouall , j. j. w. in _ computational advances in organic chemistry : molecular structure and reactivity _ , c. gretir and i. g. csizmadia ( eds ) , pp . 167185 , kluwer academic , the netherlands ( 1991 ) . rodrguez , a. m. , baldoni , h. a. , suvire , f. , nieto vzquez , r. , zamarbide , g. , enriz , r. d. , farkas , . , perczel , a. , mcallister , m. a. , torday , l. l. , papp , j. g. , and csizmadia , i. g. , * 455 * , 275301 ( 1998 ) . garca de la vega , j. m. and miguel , b. in _ introduction to advanced topics of computational chemistry _ , l. a. montero , l. a. daz and r. bader ( eds ) , chapt . 3 , pp . 4180 , editorial de la universidad de la habana ( 2003 ) . shao , y. , molnar l. f. , jung , y. , kussmann , j. , ochsenfeld , c. , brown , s. t. , gilbert , a. t. b. , slipchenko , l. v. , levchenko , s. v. , oneill , d. p. , distasio , r. a. , lochan , r. c. , wang , t. , beran , g. j. o. , besley , n. a. , herbert , j. m. , lin , c. y. , van voorhis , t. , chien , s. h. , sodt , a , steele , r. p. , rassolov , v. a. , maslen , p. e. , korambath , p. p. , adamson , r. d. , austin , b. , baker , j. , byrd , e. f. c. , dachsel , h. , doerksen , r. j. , dreuw , a , dunietz , b. d. , dutoi , a. d. , furlani , t. r. , gwaltney , s. r. , heyden , a. , hirata , s. , hsu , c. p. , kedziora , g. , khalliulin , r. z. , klunzinger , p. , lee , a. m. , lee , m. s. , liang , w. , lotan , i. , nair , n. , peters , b. , proynov , e. i. , pieniazek , p. a. , rhee , y. m. , ritchie , j. , rosta , e. , sherrill , c. d. , simmonett , a. c. , subotnik , j. e. , woodcock , h. l. , zhang , w. , bell , a. t. , chakraborty , a. k. , chipman , d. m. , keil , f. j. , warshel , a. , hehre , w. j. , schaefer , h. f. , kong , j. , krylov , a. i. , gill , p. m. w. , head - gordon , m. , * 8 * , 31723191 ( 2006 ) . Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: after iras @xcite showed that most galaxies in the nearby universe contain at least some dust , the main question moved from `` whether '' to `` when , where , and how much '' . `` when '' did galaxies started to be enriched in metals and dust and how did their dust content evolve as a function of redshift ? `` where '' is the dust located within a galaxy and how is it distributed among hubble types ? `` how much '' dust there is in galaxies , how much stellar light is absorbed by dust and how is it re - emitted ? more than one third of the bolometric luminosity of local galaxies is processed by dust into the infrared . the fraction of stellar radiation reprocessed into the iras 8120 @xmath0 m window is @xmath125%30% @xcite ; this implies an infrared energy fraction around 35%40% of the total bolometric energy within the local @xmath1100 mpc , once bolometric corrections from the iras window to the full infrared range are included @xcite . this average value comes with a large variation between galaxies , depending on luminosity , morphological type , and activity level ( see section 3 ) . the cosmic infrared background ( cib ) observed by cobe @xcite sets equally stringent limits on the total amount of stellar energy re - processed by dust into the infrared at _ all _ redshifts . about half , and possibly more , of the background radiation between 0.1 @xmath0 m and 1000 @xmath0 m is emerging at wavelengths @xmath240 @xmath0 m @xcite . this implies that at some epoch(s ) in the past galaxies have been infrared luminous at least as much as , and possibly more than , today to produce the observed balance between uv / optical and infrared energy in the cosmic background . bright sources are also found to be about two orders of magnitude more numerous in the past than today ( section 5.2 and * ? ? ? the infrared luminosity of a galaxy is related to both its dust content and nuclear / star - formation activity . the lower metal content of distant galaxies does not necessarily make them more transparent systems than galaxies today . the dust content of galaxies is , in fact , proportional to both metallicity and gas content ; as stellar populations evolve in time and more gas is locked into stars , the metallicity of a galaxy increases while its gas content decreases . the two trends are opposite to each other , and the dust content reaches a peak sometimes during the evolution of the galaxy . models predict that galaxies at z@xmath112 have about twice the dust opacity of local galaxies @xcite . high - redshift , star - forming galaxies are indirectly observed to have large gas column densities ; the average _ sfr per unit galactic area _ increases by more than an order of magnitude between redshift 0 and 3 @xcite , which is indication of a similar increase in the average gas column density via the global schmidt law @xcite . the presence of dust in galaxies limits our ability to interpret the local and distant universe , because dust extinction dims and redden the galaxy light in the traditional uv to near - ir windows , where the vast majority of the data have been obtained . dust effects hamper the interpretation of galaxies seds in terms of their fundamental parameters , such as age , stellar population mix , star formation rates ( sfrs ) , and stellar initial mass function ( imf ) . since young stellar populations are on average more deeply embedded in dust clouds than older stars , dust opacity variations within a galaxy impact the derivation of a galaxy s star formation history , by altering the census of the populations in the different age bins . heavy dust extinction in nearby starburst galaxies still prevents establishing whether the stellar imf in these systems is different from that of more quiescent environments @xcite . the increase of dust content with luminosity in spirals may be responsible for the non - linear behavior of the tully - fisher relation , and this translates into biased distance estimates @xcite . determinations of luminosity functions , mass - to - light ratios , and number counts , especially at short wavelengths , suffer from both the presence of dust and the variations of dust content from galaxy to galaxy . light dimming by dust biases measurements of galaxies sfrs , both locally @xcite and at high redshift @xcite , and is the cause of uncertain cross - calibrations between different indicators . the evolution with redshift of the sfr of galaxies per unit comoving volume is still debated : unclear is whether it has a peak around z@xmath113 @xcite or is mostly flat between z@xmath11 and z@xmath14 @xcite . answering this question , which requires understanding the dust opacity characteristics and the nature of high redshift galaxies , can discriminate between different flavors of galaxy formation scenarios @xcite . finally , measuring the fraction of metals locked into dust at different epochs is a crucial ingredient for tracing the chemical evolution of galaxies . the quantification of the dust opacity of galaxies of all types and at all redshifts as a function of wavelength has required the collective effort of many investigators over the last @xmath11015 years and is still , in many aspects , an on - going process . this review attempts to capture the main results obtained so far ; it concentrates on star - forming galaxies , because of their key role for intepreting the star formation history and evolution of galaxies . the topic has acquired with time large breadth and numerous publications have appeared in the field ; despite the effort to include as many relevant results as possible , some important contributions may have been inadvertenly overlooked . after iras , the infrared space observatory ( iso ) continued to shed light on the dust emission properties of galaxies , and comprehensive reviews are given in @xcite and in the proceedings of the 1998 paris conference ` the universe as seen by iso ' @xcite . the status of the topic before iso is reported in the proceedings of the 1994 cardiff ( u.k . ) conference ` the opacity of spiral disks ' @xcite and of the 1996 johannesburg ( south africa ) conference ` new extragalactic perspectives in the new south africa ' @xcite . throughout the following , the adopted cosmology is a flat universe with @xmath3=0 and hubble constant h@xmath4=65 km s@xmath5 mpc@xmath5 . ` local universe ' refers to a sphere of 100200 mpc radius centered on the milky way . the transfer of radiation through dust is governed by the well - known equation ( e.g. , * ? ? ? * ; * ? ? ? * ) : @xmath6 where i@xmath7(s,@xmath8 ) is the radiation intensity , ds is the element of linear distance along the direction of propagation , @xmath9 is the element of solid angle as seen by the dust grain , @xmath10 is the opacity ( absorption plus scattering ) of the dust per unit separation s , @xmath11 is the emissivity of the sources along the direction of propagation , @xmath12 is the albedo , i.e. the fraction of the opacity due to scattering , and @xmath13 is the phase function of the dust grains expressed as a function of @xmath14 , the angle between the directions determined by the incident radiation ( @xmath15 ) and the scattered radiation ( @xmath16 ) . one commonly used expression of the scattering phase function , the henyey - greenstein function @xcite , is characterized by a single parameter , the asymmetry parameter @xmath17=@xmath18cos(@xmath14)@xmath19 , which is the average of the cosine of the scattering angle . determining the dust opacity and recovering the intrinsic physical properties of complex systems like galaxies typically requires the full treatment of the transfer of radiation through dust ( equation 1 ) . the efforts to solve the integro - differential equation 1 in the case of galactic environments have generated increasingly sophisticated radiative transfer models and codes to deal with the complexity of both the dust physical properties and the dust / emitter geometry . a representative , although non - exhaustive , list of models and their salient characteristics is given in table [ tbl-1 ] . a detailed description and comparison between these models and codes is beyond the scope of the present review and the interested reader is referred to the papers listed in the table . although the simplifying assumptions implemented in many cases often appear to yield as good a solution as more complex treatments ( see , for instance , * ? ? ? * ) , the large number of free parameters is a fundamental shortcoming of radiative transfer models . workarounds to this shortcoming go from assembling large , coherent sets of independent observational data on just one or a small number of objects to making over - simplistic assumption on the age and metallicity mix of the galaxy stellar populations , and often both . this clearly bears deeply on the generality and applicability of any result and may account to an extent for the wide variety of conclusions on the opacity of galaxies reached by different authors . a point - like source ( a star or an agn ) behind a screen of dust is the simplest source / dust geometry , and equation 1 has solution : @xmath20 where i@xmath21 and i@xmath22 are the attenuated ( observed ) and intrinsic radiation intensity , and @xmath23 is the optical depth of the dust screen and a@xmath7 that will be used later . ] . the total - to - selective extinction , i.e. the part of the optical depth which only depends on the physics of the dust grains and , therefore , only on the wavelength , is often expressed as k(@xmath24)=a@xmath25/e(b@xmath26v ) . the color excess e(b@xmath26v ) is defined as a@xmath27(b)@xmath26a@xmath27(v ) , hence k(b)@xmath26k(v)=1 . although many studies traditionally employ a ` mean galactic extinction curve ' to correct galactic and extragalactic sources for the effects of dust obscuration , the well - studied milky way interstellar extinction has been shown to be spatially variable and highly dependent on the dust grain environment along the line of sight ( see reviews of @xcite and @xcite ) . @xcite demonstrated that the variations can be effectively parametrized by the total - to - selective extinction at v , r(v)=a@xmath28/e(b@xmath26v ) . both the steepness of the far - uv ( @xmath290.3 @xmath0 m ) rise of the extinction curve and the strength of the 0.2175 @xmath0 m absorption feature decrease for increasing r(v ) , as schematically shown in figure [ fig1 ] @xcite . most sightlines have r(v)@xmath3026 , with the low values corresponding to the diffuse ism ( for which r(v)=3.1 is a reasonable average value , ( * ? ? ? * e.g. , ) ) and the high values pertaining to dense clouds . the large values of r(v ) in dense clouds , usually accompanied by a gray extinction curve , are consistent with dust grains having systematically larger sizes than in the diffuse medium @xcite . the only galaxies external to the milky way for which some measurements of the uv extinction curves exist are the magellanic clouds ( figure [ fig1 ] ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , although tentative measurements in m31 have been made @xcite . despite the large variation of extinction properties found within each of the magellanic clouds ( * ; * ? ? ? * ; * ? ? ? * e.g. , ) , the _ average _ curves of the clouds appear to have some intrinsic differences from the average milky way curve , including a smaller r(v ) ( r(v)@xmath312.42.9 , @xcite ) . this has been interpreted as an effect of the lower metallicity in the clouds , which causes the molecular clouds to be more diffuse @xcite and , possibly , to produce on average smaller dust grains @xcite . the weak or absent uv absorption ` bump ' at 0.2175 @xmath0 m in the extinction curves of the lmc 30 doradus region and of the smc bar , respectively , ( figure [ fig1 ] ) could be related to local processes in these regions , possibly to the ongoing star formation activity @xcite . although the bump carriers have not been positively identified yet ( e.g. , * ? ? ? * ) , mid and far infrared studies of star forming regions in the milky way have provided evidence that the environmental uv energy density level affects the dust properties , such as the grain size distribution , via destruction or coagulation , and/or the dust physical characteristics , via ionization state changes @xcite . thus , the environmental uv energy density appears to play as important a role as the metallicity in shaping the extinction curve . if confirmed by further observations , this hypothesis may explain the general absence of a 0.2175 @xmath0 m absorption ` bump ' in the extinction curve of starburst galaxies ( section 4.1 ) and may partially account for its weakness in qsos ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? in addition to absorbing stellar light , dust manifests itself through emission and absorption features and continuum emission at @xmath21 @xmath0 m . features in the near / mid - ir include , among others , the silicate absorption at 9.7 and 18 @xmath0 m , h@xmath32o and co@xmath32 ice absorption at 3.0 and 4.3 @xmath0 m ( see @xcite and @xcite for a list of other absorption features ) , and the emission bands at 3.3 , 6.2 , 7.7 , 8.6 , and 11.3 @xmath0 m @xcite . the continuum emission beyond a few @xmath0 m and up to @xmath14070 @xmath0 m is considered to be mainly due to discrete photon heating of very small dust grains , the smallest among which are transiently heated up to @xmath301,000 k ( e.g. , * ? * ; * ? ? ? * ) ; dust heated by massive stars to temperatures @xmath1970 k also contributes to the flux in the wavelength region below @xmath150 @xmath0 m , albeit with decreasing filling factor for increasing temperature @xcite . beyond 4070 @xmath0 m , the emission is mainly due to dust grains in nearly steady balance with the average heating by starlight ; the transition wavelength between predominance of discrete photon heating versus that of nearly - steady state depends on the average color temperature of the dust and , thus , on the mean uv energy density level of the galaxy @xcite . the original dust model of bare graphite and silicate grains @xcite has evolved over time by including more complex grain compositions and extended grain size distributions , to accomodate all the details of the absorption and emission , and the variability of the environment , while at the same time accounting for the constraints imposed by the heavy elements overall abundance ( * ? ? ? * e.g. , ) . polycyclic aromatic hydrocarbons @xcite , hydrogenated amorphous carbons @xcite , icy and organic refractory mantles @xcite , composite and/or porous grains @xcite , as well as grain size distributions that significantly depart from the original power law model @xcite , have all been suggested as possible alternatives / complements to the graphite / silicate model . in the late 80s , @xcite and @xcite revived the decade - long debate on whether galaxy disks are mainly transparent or opaque to dust . using multi - wavelength information and surface brightness - versus - inclination tests , respectively , the two groups of authors reached the similar conclusion that the disks of galaxies are fairly opaque , with total face - on extinctions in the b band a@xmath331 mag . ] . these results countered earlier findings , also based on the surface brigthness - versus - inclination tests , that galaxy disks are mostly transparent with face - on extinctions at b less than 1 mag @xcite . the reason why similar approaches can produce very different results can be found in the complexity of the absolute and relative distribution of stars and dust inside galaxies . when there is a significant departure from the simplest geometry of a point source ( a star ) behind a homogeneous screen of dust , the dust distribution , rather than the optical depth of the dust layer , becomes the dominant factor in determining the obscuration suffered by the stellar radiation @xcite . in particular , differential optical depth effects come into play . if , for instance , the stars are homogeneously mixed with the dust , the shorter the wavelength of the radiation the smaller the optical depth of the dust layer from which it can emerge ; thus , an observer will detect blue light only from the closest outer ` skin ' of the mixed distribution , while redder light will come from deeper layers . an extreme case is that of infinitely optically thick clumps in front of an extended light distribution : the stellar emission will come only from those lines of sight that do not intersect clumps , and the net result will be an unreddened , albeit dimmer than the original , spectral energy distribution ( sed ) . the scattering component of the extinction curve will also modify the general characteristics of the dust obscuration , as light will be scattered _ into the line of sight _ as well as out of it . the net result of all these effects is a general blueing of the emerging sed relative to the foreground homogeneous dust screen case , or , in other words , an often much greyer ` effective extinction ' , even in the presence of large dust optical depths @xcite . for illustrative purposes , figures [ fig2 ] and [ fig3 ] show how photometric and spectroscopic quantities are affected by the dust geometry for five ` toy models ' . the models represent plane - parallel geometries of dust and stars ( details are given in * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) for : ( 1 ) the simple case of a homogeneous , non - scattering dust screen foreground to the light source ; ( 2 ) a homogeneous , scattering dust screen foreground to the light source ( e.g. , a screen in close proximity of the source , so that scattering into the line of sight becomes an important effect ) ; ( 3 ) a poissonian distribution of clumps in front of the light source , with average number @xmath34=10 ; ( 4 ) a homogeneous mixture of dust and emitters ; ( 5 ) a homogeneous dust / emitter distribution averaged over all inclination angles , with the uv and ionized gas emission having the same scaleheight of the dust while the scaleheight of the optical light gradually increases to twice that of the dust at v and beyond . models 25 include scattering in the calculationsm feature is due to pure absorption ( no scattering ) @xcite . these @xmath35 and @xmath36 values are assumed here to be representative of dust properties , although the following caveats need to be kept in mind : some values , especially for the albedo , are still unsettled @xcite ; the generalization of the reflection nebulae results ( r(v)@xmath193.1 ) to the diffuse dust is not immediate @xcite ; the uv @xmath35 and @xmath36 values could be lower in the magellanic clouds than in the milky way @xcite . ] . in all cases , the light source is assumed to be extended and to be represented by a 300 myr old stellar population undergoing constant star formation , with solar metallicity and a salpeter imf in the mass range 0.1100 m@xmath37 @xcite . the ` blueing ' effect of geometries more complex than a foreground dust screen is evident from the comparison of the five models . at fixed optical depth @xmath38 ( fixed amount of dust ) , the reddening of the v@xmath26i color of a stellar population induced by a generic geometry of stars and dust will be less , and the effective extinction a@xmath39 lower , than that induced by a foreground screen ( figure [ fig2 ] , panels a and b ) . the same will be true for colors at other wavelengths ( figure [ fig2 ] ) , where , again , the foreground non - scattering screen will provide the maximum reddening . for mixtures of dust and stars ( models 4 and 5 ) , reddening of colors and effective extinction reach asymptotic values ; beyond certain values of the optical depth , an increase in @xmath38 corresponds to small or negligible changes in colors and a@xmath39 . this is due to the fact that the shortest wavelength emission is contributed only by the dust layers closest to the observer . one common consequence of the greyer effective extinction caused by complex dust / emitter geometries is that a limited number of diagnostics over a small wavelength range will lead to underestimates of the true galaxy opacity , especially if the actual dust geometry is not known and is ` a priori ' assumed to be foreground @xcite . to aid the comparison with data on local starbursts and high - redshift star - foming galaxies ( sections 4 and 5 ) , the models shown in figure [ fig3 ] are a function of the ratio of the attenuated - to - intrinsic balmer line ratio : @xmath40 and of the uv spectral slope . ] . the luminosity ratios of nebular hydrogen lines , like h@xmath41 , h@xmath42 , pa@xmath41 , br@xmath43 , etc . , are rather insensitive to the details of the underlying stellar population and of the imf , being affected at the @xmath15%10% level by variations of the gas physical conditions ( e.g. , * ? ? ? * ) ; they represent , therefore , accurate probes of dust reddening and geometry . table [ tbl-2 ] lists the commonly used hydrogen emission line ratios , and the differential extinction between each pair . when multiple line ratios widely separated in wavelength are available , they can constrain the dust geometry in the sampled wavelength range . the two drawbacks of using nebular lines are : ( 1 ) they are produced by ionizing photons , therefore an ionizing stellar population needs to be present in the galaxy ; and ( 2 ) they only probe the attenuation of the _ ionized gas _ , and such attenuation may or may not be directly related to that of the stellar population ( section 4 ) . the explicit expression of r@xmath44 depends on the dust geometry ; in the simple case of a foreground , non - scattering screen : @xmath45},\ ] ] where the values of k(h@xmath42 ) and k(h@xmath46 are given in table [ tbl-2 ] . degeneracies between the reddening induced by dust and that induced by variations in the age , metallicity , and imf of the stellar populations play an non - negligible role in opacity determinations of galaxies . these effects are especially important when only broad band colors or low resolution spectroscopy are available , and age - sensitive or imf - sensitive stellar features are not identifiable ( e.g. , discussion in * ? ? ? * ) . for the age - dust degeneracy , an example is given in figure [ fig4 ] ; here , the broad band colors and magnitudes of a dust - free , ageing stellar population produced by an instantaneous burst of star formation are compared with the analogous quantities of an increasingly extincted 6 myr old stellar population . a clumpy dust distribution with @xmath472 in front of the young population reddens and attenuates it enough that its broad - band uv , optical , and near - ir colors closely resemble those of a 20 myr old , dust - free population ( corresponding to the color v@xmath26i@xmath10.6 in figure [ fig4 ] ) . thus , the dusty , young population can be mistaken for a dust - free population a factor @xmath13 older . this example is just one out of a large range of possibilities in the multi - dimensional dust / age / imf / metallicity parameter space . in summary , the grey effective extinction produced by dust geometry and scattering , the age - imf - metallicity - dust degeneracy , together with uncertainties on the appropriate extinction curve to be applied to different galaxies ( section 2.1 ) and the absence of strong emission and absorption features below @xmath11 @xmath0 m ( apart from the @xmath10.22 absorption bump which is not universal ) , all contribute to make dust obscuration measurements very difficult in galaxies . efforts to determine the dust opacity of local galaxies have spawned the development of a number of independent techniques that have been applied to a large array of samples . the most widely used methods in the case of galaxy disks are tests of the dependence of the surface brightness on inclination , multi - wavelength comparisons , and statistical analysis of the color and number count variations induced by a foreground galaxy onto background sources . each method has its strengths and weaknesses , as briefly detailed below . as @xcite , @xcite , and @xcite pointed out , probing galaxy dust opacity by measuring the surface brightness dependence of galaxy disks on inclination can suffer from drastic selection biases . for increasing inclination , an apparent - magnitude - limited catalog will mimick the condition for transparent galaxies , with mildly increasing surface brightnesses , increasing diameters , and constant total luminosities . conversely , an apparent - diameter - limited catalog will mimick the conditions for opaque disks , with roughly constant surface brightnesses , increasing magnitudes and constant diameters as a function of increasing inclination @xcite . selection effects can , however , be controlled by combining different samples and by using galaxy redshifts / distances to discriminate among the different biases , and analysis of large samples of late - type ( mainly sbc and sc ) spirals @xcite has shown that the central regions of disk galaxies are opaque while the external regions beyond 23 scale lengths are almost completely transparent ; there is also a trend for luminous galaxies to be more opaque than less luminous galaxies @xcite . in bright , m@xmath4821 disk galaxies , the observed magnitude correction from edge - on to face - on , @xmath49m@xmath501 mag , corresponds to a face - on central optical depth is the optical depth of the dust at the geometrical center of the tri - dimensional dust distribution for a face - on disk galaxy . ] @xmath38(0)@xmath515 ; this implies effective face - on extinctions a@xmath520.40.5 mag , a@xmath530.2 mag , and a@xmath540.1 mag @xcite , for a double exponential distribution of dust and stars , and for practically transparent external galaxy regions @xcite . less luminous galaxies are progressively less opaque in their central regions , with a edge - on to face - on correction of @xmath49m@xmath500.5 mag at m@xmath5518 , corresponding to effective face - on extinctions a@xmath560.1 mag @xcite . qualitative results similar to those above have been obtained by @xcite and @xcite for a small number of edge - on late - type spirals using multi - wavelength kinematic information . the method , proposed by @xcite , consists of comparing rotation curves of an edge - on galaxy at widely different wavelengths . if the edge - on galaxy is optically thick at a given wavelength , say , h@xmath42 , the rotation curve measured at this wavelength shows apparent solid body rotation throughout the entire disk , because the h@xmath42 emission is coming only from the external regions of the galaxy . a comparison with a mostly unextincted rotation curve measured at a longer wavelength , such as hi , can then establish whether optical thickness at h@xmath42 exists . this method , though fairly direct in principle , may suffer from two limitations : first , dust in galaxies tends to be patchy and the h@xmath42 emission may be detected along the least extincted sight lines even in the presence of considerable opacity ; second , the method implicitly assumes that h@xmath42-emitting regions are uniformly distributed across the disk . the multi - wavelength comparison method provides an independent approach to the problem of determining the dust opacity of galaxies . in broad terms , the technique consists of using multiwavelength data to solve simultaneously for the intrinsic colors of the stellar populations and for the reddening / attenuation induced by the dust . generally , broad band images covering as large as possible wavelength baseline , often from b to k band , are used to determine the opacity of the disks , by exploiting the differential extinction property of dust @xcite . indeed , in face - on quiescent galaxies k - band emission appears negligibly affected by dust obscuration , with @xmath5710% of the light lost to dust absorption and scattering @xcite . the multiwavelength technique is the most widely adopted for measuring dust opacity in galaxies , because it employes easily accessible information on diverse galaxy types , and enables detailed modelling of optical depths as a function of galactic region . there are , however , three main weaknesses to the method : first , this technique needs as wide a wavelength coverage as possible , often difficult to obtain , to yield unambiguous results ; second , if the dust optical depth exceeds unity at some wavelength , different volumes of the galaxy will be sampled at different wavelengths ; third , the need for large datasets and detailed modelling makes obtaining results on large samples relatively cumbersome . results vary from author to author , due to the different sample selections , galaxy and dust distribution models , and measurement techniques applied . optical and near - ir image analysis gives typical central optical depths @xmath38(0 ) in the range 0.32.5 for types sab - sc @xcite , although higher values have been obtained by @xcite . for exponential stellar and dust profiles , these values of @xmath38(0 ) result in optical depths @xmath581 averaged over the inner 23 scale - lengths @xcite , and a@xmath520.30.4 mag and a@xmath590.15 mag over the same area . much lower values of the optical depth are found for late - type , low surface brightness galaxies , with face - on , central values @xmath38(0)@xmath10.15 @xcite . in galaxy disks , arm regions tend to be opaque , with @xmath471.45.5 ( a@xmath530.20.7 mag ) and peak values at @xmath471012 , while interarm regions are generally transparent , with @xmath470.31.1 ( a@xmath530.050.15 mag , @xcite , but see the higher interarm extinction values found by @xcite ) . energy balance estimates between dust absorption and emission can also be classified as a ` multiwavelength technique ' , but it is conceptually different from the one above . this technique employes direct measurements of the dust emission in the infrared to derive an estimate of the total stellar energy absorbed by dust , and then derive opacities at shorter wavelengths ; whenever possible , the wavelength baseline is extended to the uv to probe the direct light from massive stellar populations and constrain bolometric quantities @xcite . by combining uv and infrared ( iras ) data , @xcite , @xcite , and @xcite find typically lower @xmath60 values than those derived from optical / near - ir images . this is not incompatible with those results , since @xcite and @xcite s models assume plane - parallel dust and stars , instead of exponential profiles , and thus the opaque inner regions are averaged with the nearly transparent outer regions @xcite . still , the uv selection of the samples may bias the conclusions towards lower optical depths and the use of iras data leaves some uncertainty as to the bolometric correction of the dust emission . the use of iso data , that extend the infrared wavelength coverage out to 200 @xmath0 m , appear indeed to favor higher values of the effective extinction @xcite . @xcite find that opacity in disk galaxies is a function of luminosity ( figure [ fig5 ] ) , in agreeement with @xcite . these authors parametrize the opacity as : @xmath60=@xmath61 , with @xmath62=0.6 and l@xmath63 the visible luminosity of a l@xmath64 galaxy in the local universe . similarly , @xcite note that sb scd galaxies have @xmath650.8 and are on average 1.5 times more opaque than sa sab , and @xmath13 times more opaque than sd irr galaxies , with effective face - on extinctions a@xmath520.3 mag , a@xmath520.15 mag , and a@xmath660.1 mag for sb scd , sa sab , and sd - irr , respectively . all methods described so far use the stellar radiation of the galaxy itself as light source for the dust , with the complication that intrinsic colors , age and metallicity gradients of the stellar populations and the relative distributions of dust and stars need to be disentangled from the extinction proper . a technique that overcomes these problems is the one that uses background sources as ` light bulbs ' for the dust in foreground galaxies . in this case , the dust opacity is measured across the full thickness of the foreground galaxy . there are basically two approaches to the technique . in the first approach the foreground galaxy is partially projected onto a nearby background galaxy @xcite . in general , the background galaxy needs to present a smooth light profile , thus early types and preferably ellipticals are required ( see , however , * ? ? ? * ) ; the dimming and color changes induced by the foreground galaxy dust onto the isophotes of the background source provides a measurement of dust opacity . the drawbacks of this approach are twofold : ( 1 ) there are very few known pairs of galaxies that fullfil all the requirements and ( 2 ) the method is limited by the degree of symmetry of both galaxies @xcite . the second approach uses distant galaxies as background light sources . statistical variations in number counts and colors of the background galaxies relative to control fields provide a handle on the total dust optical depth of the foreground galaxy @xcite . this method overcomes most of the limitations of the first one , but it presents some of its own : ( 1 ) high angular resolution images ( e.g. , from hst ) are needed in order to identify background galaxies ; ( 2 ) even with such images it is often difficult to discriminate background sources from extended regions within the foreground galaxy ( e.g. , hii regions ) , especially in crowded areas . because of these limitations , the second method may also tend to avoid crowded and/or heavily extincted areas , thus biasing the results towards less dusty galactic regions . results from various authors indicate that the spiral arms of galaxies tend to be opaque with a@xmath670.52 mag and a@xmath680.31.5 mag at any radius , while the opacity of interarm regions decreases with increasing distance from the galaxy center , starting at a@xmath690.7 mag within 0.3 r@xmath70 and decreasing to practically zero at r@xmath70 @xcite . the dust masses inferred from these extinction measures agree within a factor @xmath12 with masses derived from the infrared / submm ( iso and scuba ) dust emission @xcite . one common trend generally found is the greyer - than - galactic reddening in the arm regions ( see , however , * ? ? ? the clumpiness of the dust within each resolution bin has been suggested to be responsible for the grey reddening @xcite . less agreement exists on the shape of the reddening curve in the interarm regions , due to the larger uncertainties resulting from the lower extinction values . the measured extinctions are front - to - back , i.e. represent the optical depth of the intervening dust between the observer and the background sources . to relate these numbers to those derived from the inclination and multi - wavelength tests , one has to make assumptions as to the distribution of stars relative to the dust in the foreground galaxy ; for a mixed geometry of stars and small dusty clumps , the measured optical depths correspond to effective i extinctions of a@xmath530.30.4 mag and a@xmath710.2 mag for the arms and the interarm regions , respectively . multi - wavelength optical and near - ir imaging of edge - on or nearly edge - on systems indicate that the scalelength of the dust is about 40% larger than the scalelength of the stars @xcite , while the dust / stars scaleheight ratio is in the range 0.250.75 , with a mean value of 0.5 , @xcite for late - type spirals ( sb and later ) and appears to have ratio @xmath11 in early - type spirals @xcite . iras 100 @xmath0 m @xcite and iso long wavelength maps @xcite confirm that cold dust emission extends beyond the limits of the optical disks along the radial direction , with scales that are @xmath140% larger than those of the b - band emitting stars , but still well within the hi disks ( e.g. , * ? ? ? using statistical measurements of color variations between background galaxies and control fields , @xcite give a @xmath512 @xmath72 detection of a@xmath690.04 in galaxy haloes , at a distance of about 60 kpc from the galaxy centers , along the optical major axis . spiral and irregular galaxies follow a global metallicity - luminosity ( and metallicity - mass ) relation , where the oxygen abundance increases by a factor @xmath1100 for an increase of @xmath110@xmath73 in absolute blue luminosity @xcite . irregulars occupy the low metallicity locus of the relation , with @xmath3010 times on average lower oxygen abundance than spirals . since the hi masses per unit l@xmath74 in irregulars are a factor @xmath303 larger than in spirals @xcite , the mass in interstellar metals per unit l@xmath74 is @xmath303 times smaller . adding the h@xmath32 gas into the balance does not change this conclusion , since the h@xmath32/hi mass ratio in irregulars appear to be no larger than that in spirals , even accounting for the uncertainties in the co - to - h@xmath32 conversion factor @xcite . measurements give the h@xmath32 mass as 25%50% of the hi mass in late spirals and as @xmath3020% of the hi mass in irregulars @xcite . smaller metal masses for the irregulars translate into smaller dust masses per unit l@xmath74 , for standard values of the metal - to - dust ratio @xcite , and into the expectation that irregular galaxies are at least 3 times more transparent at b than spiral galaxies , for similar dust / star geometries . the analysis of the blue , h@xmath42 , and iras - detected infrared emission confirms that irregular galaxies are relatively transparent systems @xcite . in a study of the magellanic clouds using background galaxies , @xcite find that the central regions have a@xmath750.25 in the lmc and a@xmath750.16 in the smc . in their uv selected samples , @xcite find that a 0.1 l@xmath64 galaxy has a mean optical depth about 1/3 of the optical depth of an l@xmath64 galaxy , and @xcite reach a similar conclusion by comparing sb - scd galaxies with sd - irr . thus , if the typical face - on effective extinction for an sc galaxy is a@xmath520.40.5 mag , the analogous quantities for an irregular are a@xmath520.15 and a@xmath530.050.10 . although early hubble type galaxies ( ellipticals and s0s ) do not generally fall under the category of ` star - forming ' galaxies , they occupy a niche in terms of dust opacity properties that is at the same time opposite and complementary to that of other galaxies . ellipticals and s0s are at the opposite end of the hubble classification relative to irregulars , but they appear to be no more opaque than those late systems . early type galaxies cover a metallicity range similar to that of late type spirals @xcite , but have total ( hi@xmath76h@xmath32 ) gas fractions per unit optical luminosity that are @xmath13 times or more smaller @xcite . thus , for standard interstellar metal - to - dust ratios , early type galaxies are expected to be more transparent than disk galaxies . the iras survey has shown that between 10% and 50% of all nearby early type systems ( 45% of ellipticals and 68% of s0s ) contain dust @xcite , although the involved masses are generally modest , typically in the range 10@xmath7710@xmath78 m@xmath37 or m@xmath79/l@xmath8010@xmath8110@xmath82 m@xmath37 l@xmath83 @xcite . the exact fraction of dusty systems appears to depend upon the adopted detection threshold @xcite , but it is likely a lower bound to the actual value , given the limits in the iras sensitivity . the 50% figure seems in agreement with the results of optical surveys : at least 40% of the local ellipticals and s0s show presence of disks , lanes and/or patchy distributions of dust in visual images @xcite . about 4050% , and possibly as many as 80% , of ellipticals also show evidence for circumnuclear dust on scales @xmath18100 pc , as seen in hst images ; the small - scale dust is mostly associated with nuclear radio activity @xcite . despite the abundance of evidence for dust , optical extinction determinations of dust features in ellipticals generally account for only @xmath11/10 to @xmath11/5 of the dust mass measured from the infrared ( iras ) emission ; the remaining 80%90% of the dust mass is probably associated with a component smoothly distributed across the galaxy , and thus difficult to detect with standard techniques @xcite . limited data from iso in the wavelength range 120200 @xmath0 m confirm these findings @xcite . using the elliptical galaxy model of @xcite , @xcite determine that the observed infrared luminosities are compatible with central optical depths of the diffuse component @xmath38(0)@xmath570.7 , with a typical value @xmath38(0)@xmath10.20.3 ( see , however , the somewhat larger values predicted by the models of @xcite ) . the corresponding effective extinction is a@xmath840.050.10 mag and a@xmath850.020.04 mag . the giant ellipticals found at the centers of cooling flow clusters are often surrounded by extended , @xmath3010100 kpc , filamentary , and dusty emission - line nebulae , with a higher frequency in those clusters with cooling times shorter than a hubble time @xcite . dust lanes in the central galaxies have , indeed , been directly observed @xcite . optical and uv emission line studies give extinction values in the range a@xmath860.32 mag for the dust associated with the nebula @xcite ; infrared and sub mm measurements give dust masses of the order of 10@xmath8710@xmath88 m@xmath37 , or m@xmath79/l@xmath8010@xmath8210@xmath89 m@xmath37 l@xmath83 , for these systems @xcite . the existing controversy on the origin of the dust associated with the central galaxies of cooling flow clusters will be briefly reviewed in section 6 . such galaxies are , however , not a statistically significant component of ellipticals in general , thus their contribution to the dust mass budget of early type galaxies is small . the galaxies discussed so far tend to be mostly ` quiescent ' , i.e. their bolometric output is _ not dominated _ by a single region of active star formation or by nuclear non - thermal activity , and have modest infrared outputs . their infrared luminosities in the 31000 @xmath0 m band are typically @xmath3010@xmath90 l@xmath37 and less than @xmath146@xmath9110@xmath92 l@xmath37 , with infrared - to - blue luminosity ratios l@xmath93/l@xmath940.40.5 @xcite , and @xmath301/2 or less of the stellar bolometric luminosity converted into infrared emission via dust reprocessing of the dust emission is derived from the infrared luminosity measured in the 8120 @xmath0 m or 40120 @xmath0 m iras window @xcite multiplied by @xmath11.4 or @xmath12.2 , respectively , to account for bolometric correction factors as modeled by @xcite from iso data . the blue luminosity is l@xmath95(0.44 @xmath0 m ) . ] ( see table [ tbl-3 ] ) . galaxies with l@xmath966@xmath9110@xmath92 l@xmath37 and l@xmath93/l@xmath971 are progressively associated with dustier and more active systems for increasing values of the infrared luminosity @xcite . the brightest infrared sources are not necessarily associated with the most massive galaxies , despite the existence of a relation between a galaxy luminosity ( mass ) and its dust content @xcite . infrared luminosity is a combination of dust content and activity : brighter sources have larger amounts of star formation and/or non - thermal activity embedded in dust . most of the luminosity from galaxies with l@xmath98l@xmath9910@xmath100 l@xmath10134 l@xmath64 appears associated with sources heavily obscured by dust @xcite . at the bright end of the infrared luminosity function @xcite are the ultraluminous infrared galaxies ( ulirgs ) with luminosities from l@xmath10210@xmath103 l@xmath37 up to @xmath18@xmath9110@xmath103 l@xmath37 @xcite , infrared - to - blue ratios around 30400 @xcite , and optical extinctions in excess of 1040 magnitudes @xcite . these sources show a continumm of properties as a function of increasing luminosity , with the fraction of starburst - dominated ulirgs going from @xmath180%85% at the low luminosity end to @xmath150% for l@xmath962@xmath9110@xmath103 l@xmath37 , and the remaining fraction being agn - dominated @xcite . comprehensive reviews of the properties of ulirgs are given in @xcite and @xcite . despite their large luminosities , bright infrared sources do not represent a major contributor to the energy budget of galaxies in the present - day universe . such objects are relatively rare within the local 100200 mpc : galaxies with l@xmath10410@xmath100 l@xmath37 and with l@xmath10410@xmath103 l@xmath37 represent @xmath16% and @xmath181% , respectively , of the total infrared emission in the local universe @xcite . however , the contribution of infrared bright sources to the galaxy energy budget increases with redshift ( see section 5.2 ) , as does their relevance in the framework of galaxy evolution . although the issue of dust opacity of galaxies is far from settled , an overall picture is apparent : galaxies in the local universe are only moderately opaque , and extreme values of the opacity are only found in the statistically non - dominant , more active systems . table [ tbl-3 ] summarizes the main conclusions of this section , by listing the effective extinction values for local galaxies according to morphological type . for disk and irregular galaxies , both the face - on and the inclination - averaged extinctions are given . for instance , in sb - scd galaxies , the typical inclination - averaged extinctions are 0.60.65 mag larger than face - on at 0.15 @xmath0 m , @xmath10.4 mag at b , @xmath10.25 mag at i , and 0.1 mag at k , for a stellar to dust scaleheight ratio that varies between 1 in the uv and 2 in the i and k bands ( model 5 of section 2.2 ) . among the non - active galaxies , luminous sb scd galaxies contain the largest amounts of dust , which absorb @xmath301/2 of their bolometric light . for comparison , dust emission represents between 35% and 40% of the bolometric luminosity of our milky way galaxy @xcite . the dust content generally decreases in less luminous galaxies and for earlier and later type galaxies ; dust absorbs less than @xmath115% of the bolometric light in e / s0 galaxies , and the fraction rises to @xmath3030% in irregulars ( table [ tbl-3 ] ) . interpolating from table [ tbl-3 ] , the inclination - averaged effective extinction of the stellar emission in late spiral galaxies at the restframe wavelength of h@xmath42 is a@xmath1050.60 mag . this value should be compared with the effective extinction measured for the h@xmath42 line emission , a@xmath1060.81.1 mag , from the ratio of the line to the reddening - free thermal radio emission @xcite . althought there are relatively large uncertainties associated with both numbers , the two extinction values suggest that the nebular emission is more reddened than the stellar continuum , with a@xmath1071.5 a@xmath108 . studies of controlled samples indicate a value of 2 as more typical for the ratio between the dust attenuation of the ionized gas and that of the stellar continuum , both in the case of spirals and irregulars @xcite and in the case of moderate - intensity starbursts ( section 4 ) . among local galaxies , starbursts occupy a special niche . the average uv and infrared luminosities ( both proxies for sfr ) per unit comoving volume increase with redshift by roughly an order of magnitude up to z@xmath112 @xcite . active star formation played a larger role in the past than today , and may have been a fundamental ingredient in the shaping of galaxy populations . nuclear and circumnuclear starbursts occur in galaxies spanning a wide range of properties , from almost dust - free , infrared - faint , low - mass irregulars ( e.g. , tol1924@xmath26416 and ngc1569 ) to dust - rich , massive spirals ( e.g , ngc5236 ) , to infrared - bright interacting galaxies ( e.g. , ngc7714 and ngc6090 ) , to mergers and ulirgs ( e.g. , ngc4194 and ngc1614 ; arp220 , * ? ? ? nuclear and circumnuclear bursts of star formation can be triggered by a variety of mechanisms , including secular evolution of bars @xcite and galaxy interactions and mergers @xcite . the latter produce the most luminous starbursts ( ulirgs ) , which rapidly deplete the molecular gas content of the galaxy as merging progresses @xcite . less extreme starbursts , hosted in galaxies with l@xmath1095@xmath9110@xmath100 l@xmath37 , obey a positive correlation between the luminosity of the starburst and the luminosity and mass of the host galaxy @xcite : @xmath110 where sfr@xmath111 is in m@xmath37 yr@xmath5 for a salpeter imf in the mass range 0.1100 m@xmath37 ( from the formulae of * ? ? ? equation 5 is probably a consequence of the fact that large sfrs are substained by the massive gas inflows that only a deep potential well can support @xcite . indeed , from causality arguments , the maximum luminosity of a starburst scales as @xmath112v@xmath113 , with @xmath72 the dispersion velocity of the gas and v the rotation velocity of the host galaxy @xcite . there are notable exceptions to this average trend : m82 is an example of a dwarf , m@xmath110@xmath92 m@xmath37 @xcite and l@xmath1145@xmath9110@xmath90 l@xmath37 , galaxy hosting a relatively powerful starburst , with sfr@xmath16 m@xmath37 yr@xmath5 and l@xmath1023@xmath9110@xmath92 l@xmath37 . a secondary effect of both the sfr@xmath111host galaxy mass correlation and the mass - metallicity correlation is that the more powerful starbursts are hosted in the more metal - rich and dustier host galaxies ( * ? ? ? * also figure [ fig7 ] , panel c ) : @xmath115 ( see , also , * ? ? ? although equation 6 is not applicable to ulirgs , still in the local universe there are no examples of ` naked ' arp220-like galaxies , emitting most of the luminosity produced by their sfr@xmath1100200 m@xmath37 yr@xmath5 directly in the uv . objects like arp220 tend to be heavily extincted , with a@xmath861545 mag for arp220 itself @xcite , and to be bright infrared sources ( section 3.5 ) . the rest of section 4 discusses the dust reddening and obscuration properties of the local starburst galaxies with l@xmath1095@xmath9110@xmath100 l@xmath37 . the specific sample contains starbursts that are uv luminous enough to have been detected by iue @xcite ; it will be termed ` uv - selected ' in what follows . the sample is not statistically complete or rigorously defined ; however , it is representative of the broad range of properties of local uv bright starbursts @xcite . these galaxies are not as extreme in dust extinction and activity level as ulirgs . they have sfrs per unit area in the range @xmath10.320 m@xmath37 yr@xmath5 kpc@xmath116 and cover a limited range in the parameter space of optical attenuation , a@xmath10602.5 mag . the attenuation refers to the integrated value across the starburst site ; individual regions or clumps can have effective extinctions as high as a@xmath10630 mag @xcite . the host galaxies cover a wide variety of late type morphologies , from grand design spirals to irregular and amorphous , to distuberd morphologies that result from interactions or merging processes . the starbursts mostly occupy the inner region of solid body rotation of the host galaxy , with sizes @xmath10.54 kpc @xcite . the uv selection @xcite aids the discrimination from their more opaque and infrared - brighter counterparts . although the uv selection results in some limitations to the applicability of the obscuration properties discussed below ( see end of section 4.2 ) , it does not entirely exclude dusty systems ; the sample contains objects in which up to 90%95% of the bolometric energy is re - processed by dust . table [ tbl-4 ] gives three representative examples each for both the high - luminosity , dust - rich end and the low - luminosity , dust - poor end of the uv - selected starbursts , together with the average of the sample . although the uv - selected starbursts form an heterogeneous set , their dust reddening and obcuration properties are remarkably uniform . higher dust optical depths produce redder uv to near ir seds and nebular emission line ratios ( figure [ fig6 ] ) , almost independent of the details of the dust distribution and of the intrinsic stellar population @xcite . the geometry of the dust that best describes the reddening of the ionized gas emission in these systems , in the wavelength range 0.48 @xmath0m2.2 @xmath0 m , is that of a foreground like distribution . figure [ fig6 ] , panel b , shows the comparison between observations and the models of section 2.2 for the attenuated - to - intrinsic hydrogen line ratios h@xmath42/h@xmath41 and h@xmath41/br@xmath43 . internal dust does not appear to be a major component in the starburst region(s ) , and the little present is likely to be in compact clumps @xcite . the main source of opacity appears to be given by dust that is external , or mostly external , to the starburst region ( although still internal to the host galaxy ) , similar to a ( clumpy ) dust shell surrounding a central starburst . this is , of course , verified only in the b to k wavelength range where the reddening of the nebular emission lines has been measured ; measurements at longer wavelengths could easily reveal more complex geometries . from a practical point of view , the data in figure [ fig6 ] , panel b , are well described by a simple foreground , non - scattering dust distribution , with mw extinction curve for the diffuse ism ( r(v)=3.1 ) ; the reddening of the ionized gas is thus parametrized by the color excess e(b@xmath26v)@xmath117 ( equation 4 ) . the stellar continuum colors show the trend to become redder for increasing reddening of the hydrogen emission line ratios ( figure [ fig6 ] , panels a the implications are twofold : ( 1 ) variations of the intrinsic sed from galaxy to galaxy are secondary relative to the trend induced by dust obscuration , and ( 2 ) within each starburst , the same foreground like dust geometry holds for both the ionized gas and the non - ionizing stellar continuum . the first implication is not surprising , at least at uv wavelengths . since we are observing the rayleigh - jeans part of the massive stars spectrum , the intrinsic uv spectral slope and @xmath118 refer to the intrinsic and observed , respectively , uv spectral slope measured between 0.125 @xmath0 m and 0.26@xmath0 m as defined in @xcite , see appendix b for more details . ] @xmath119 is relatively constant over a fairly large range of ages , with values @xmath262.7@xmath1202.2 for constant star formation over 10@xmath78 yr@xmath57age@xmath5710@xmath90 yr ( see * ? ? ? * and the table in appendix b ) . in the same conditions , the production of ionizing photons is also constant . in an instantaneous burst of star formation , the ionizing photons are more age - sensitive than the non - ionizing photons that produce the stellar continuum and disappear before appreciable changes in the uv spectral shape can be observed . the observed uv spectral slope @xmath118 of the uv - selected starbursts span the range @xmath262.5@xmath1210.4 , much larger than what expected from stellar population variations alone , and increases for increasing values of the attenuated - to - intrinsic hydrogen line ratios ( i.e. , color excess , figure [ fig6 ] ) . in addition to correlate with the reddening of the ionized gas , the reddening of the uv stellar continuum , as measured by @xmath118 , correlates with the the infrared - to - blue and the infrared - to - uv luminosity ratios , l@xmath93/l@xmath74 and l@xmath93/l@xmath122 ( ( figure [ fig7 ] and * ? ? ? * ; * ? ? ? these ratios are a measure of the total dust opacity in the starburst region , because the bulk of the starburst s energy is coming out in the uv b . in addition , dust _ absorption _ is the main process that removes light from the line of sight @xcite ; when large galactic regions are observed light scattered by dust out of the line of sight compensates , on average , that scattered into the line of sight . the opacity reddening correlation means that measurements of reddening can be used to infer the total dust absorption in uv - selected starbursts @xcite with an uncertainty , @xmath49a@xmath1230.6 mag for individual objects and @xmath120% when averaged over large samples @xcite , that is dominated by variations in the starburst populations and in the details of the dust geometry from galaxy to galaxy . such a correlation is expected for a foreground - like dust geometry ( figure [ fig3 ] , panel d ; see discussion in * ? ? ? the slopes of the correlations in figure [ fig6 ] identify the reddening suffered by the stellar continuum , a@xmath124a@xmath125=[k@xmath126(@xmath127)-k@xmath126(@xmath128)]e(b@xmath26v)@xmath117 , with k@xmath126(@xmath24 ) the obscuration curve , while the total obscuration measurements of figure [ fig7 ] provide the zeropoint for k@xmath126(@xmath24 ) ( see , also , * ? ? ? the intrinsic starburst flux density f@xmath129 is thus recovered from the observed flux density f@xmath22 via : @xmath130 where the obscuration curve for the stellar continuum , k@xmath126(@xmath24 ) , is given by @xcite : @xmath131 the above expression folds into a single functional form a variety of effects : extinction proper , scattering , and the geometrical distribution of the dust relative to the emitters . it is derived from the spatially integrated colors of the entire stellar population in the starburst , and represents the ` net ' obscuration of the population itself . as mentioned above , the obscuration curve has mainly an absorption component , as the effects of scattering are averaged out . thus , it should not be confused with the extinction curve k(@xmath24 ) defined in section 2.1 . equation 8 implies that 1 mag of reddening between b and v for the ionized gas , a@xmath132a@xmath133=1 mag , corresponds to a@xmath134a@xmath135=0.44 mag of reddening for the stellar continuum . this fact can be alternatively expressed as : @xmath136 in other words , the stellar continuum suffers roughly half of the reddening suffered by the ionized gas ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * see , also , figure [ fig3 ] , panels e and f ) . although e(b@xmath26v)@xmath137 parametrizes the amount of reddening of the stellar continuum , it is not a color excess proper as in the case of individual stars , because of the dust geometry effects folded in the expression of k@xmath126(@xmath24 ) . one obvious characteristic of the obscuration curve is the absence of the 0.2175 @xmath0 m bump , which is a prominent feature of the mw extinction curve ( figure [ fig1 ] ) . dust / emitter geometries that can dilute the bump include mixed distributions with large optical depths @xcite . however , mixed geometries do not account for the full range of observational data ( figure [ fig3 ] ) , because the sources that contribute to the emerging sed are located at progressively lower optical depths @xmath38 for decreasing wavelength . in particular , the reddest uv slope produced by mixed geometries is @xmath138 @xcite , to be compared with @xmath139 observed in starburst galaxies ( figure [ fig6 ] , panel a ) . thus , the lack of the feature is probably intrinsic to the _ extinction curve _ of the starbursts @xcite , and may be due to the high uv energy densities that characterize these regions ( sections 2.1 and 4.2 ) . this section describes a dust model that attempts to reconcile two apparently contradictory facts : ( 1 ) stars in starbursts are on average a factor @xmath302 less reddened than the ionized gas ( equation 9 ) , and ( 2 ) both stars and ionized gas are affected by the same foreground - like dust distribution ( figure [ fig6 ] ) . a schematic representation of a dust / star / gas configuration that can at the same time account for both effects is shown in figure [ fig8 ] . stars are born within optically thick molecular clouds , and the short - lived , most massive stars remain closely associated with their parental cloud and the gas they ionize for their entire lifetime ( e.g. , * ? ? ? these stars and their surrounding gas will be generally highly attenuated . as the starburst population evolves , the inside of the region of star formation becomes depleted of dust @xcite . with energy densities @xmath140100 times higher than in the local ism , starburst environments are rather inhospitable to dust . shocks from supernovae destroy dust grains , via grain - grain collisions and sputtering @xcite . the average lifetime of a refractory grain is @xmath18@xmath9110@xmath78 yr , for a supernova rate of 0.05 yr@xmath5 and an ism mass of 5@xmath9110@xmath88 m@xmath37 ( e.g. , * ? ? ? * ) , about one order of magnitude or more shorter than the typical lifetimes of starbursts @xcite . in addition , hot - star - winds- and supernova - driven outflows evacuate both interstellar gas and dust from the region @xcite . r136 , the central cluster in 30 doradus , is an example of a few million years old cluster surrounded by an evacuating region @xcite . ionized gas separated , in projection , from the hot stars has also been observed in nearby starbursts @xcite . the evacuated ( clumpy ) dust will act as a foreground - like distribution for both the gas ( also at the edges of the region ) and for the central stars ( figure [ fig8 ] , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ; however , the ionized gas will be more obscured than the stars , because of its spatial location and closer association with the dust @xcite . this gas is more likely to contribute to the nebular emission observed at uv optical wavelengths than the gas in molecular clouds , because of its milder obscuration . in the presence of multiple generations of starburst populations , the long - lived , non - ionizing stars have time to ` diffuse ' into regions of lower dust density @xcite , as their native clusters are disrupted by evaporation or by the host galaxy s gravitational field @xcite . cool stellar populations in the lmc have , for instance , been observed to be less embedded in dust than hotter stellar populations @xcite . a similar trend is observed in the young ( @xmath1820 myr ) open clusters of the milky way @xcite . massive , non - ionizing stars can still produce significant uv emission . hence , the integrated uv optical stellar continuum from the combination of the ageing , diffusing populations and the starburst - embedded young populations will be less obscured than the emission from the nebular gas @xcite . for the stellar continuum and the ionized gas to have correlated obscuration values ( equation 9 ) , multiple starburst generations or long - lasting star formation events ( rather than ` instantaneous ' events ) are needed to attain an equilibrium between the number of stars drifting out of molecular clouds / starburst regions and the number of newly - born , dusty stars . starbursts hosted in dwarf galaxies are observed to have on - going star formation over @xmath51200 myr @xcite , and durations could be as long in more massive starburst galaxies @xcite . although the obscuration curve can be widely applied to uv - selected starbursts , it is not necessarily applicable to systems with different stellar population or ism characteristics . in the case of non - starburst galaxies there is no obvious mechanism for creating a foreground - dust - like geometry . in extreme starbursts as the ulirgs , the non - applicability of the obscuration curve has been demonstrated observationally ( * ? ? ? * and figure [ fig7 ] , panel a ) . in these galaxies , the high central concentration of gas and dust is likely to induce more extreme geometries than in uv - selected galaxies , and to produce combinations of clumpy mixed and foreground distributions . galaxies whose infrared properties are intermediate between the ulirgs and the uv - selected galaxies will progressively deviate from the dust geometry of the latter @xcite . dust in subsonic hii regions is likely to be mixed with the stars , rather than having been pushed away or destroyed as in starbursts , and scattering of stellar light into the line of sight may be an important component in the attenuation budget @xcite . because of the differences in the physical environment between the two types of objects , the foreground - dust - like approximation no longer holds for subsonic hii regions and equation 8 does not apply . differences in the star formation history can also contribute to the difference between subsonic hii regions and starbursts , with instantaneous bursts being a better description for the former and extended star formation for the latter ; thus , differential attenuation between ionized gas and stars is not necessarily expected in the hii regions . supersonic hii regions could be at the boundary between the two extremes just discussed ; one example is 30 doradus , that has evolved to the point of resembling a mini - starburst @xcite . to conclude , the starburst obscuration curve is a purely empirical result , and its derivation is independent of any assumption about the starburst s population , the dust geometry , or the details of the extinction curve . it provides statistical estimates of the reddening and obscuration of local uv - selected starbursts , and deviations as large as @xmath10.6 mag from the mean obscuration at 0.16 @xmath0 m are to be expected on an case - by - case basis . the physics of the starburst environment provides an explanation of the ` foreground - dust - like ' behavior of the curve , of the correlation between the reddening of the ionized gas and that of the stellar continuum , and of the correlation between reddening and total obscuration . as a sample , local uv - selected starbursts are characterized by modest amounts of reddening and obscuration , that are correlated with the star formation activity ( equation 6 ) . the median color excess is e(b@xmath26v)@xmath1410.35 @xcite , corresponding to uv continuum attenuations of a@xmath1421.6,1.4,1.1 mag at 0.15,0.20,0.30 @xmath0 m , respectively ( table [ tbl-4 ] and * ? ? ? * ; * ? ? ? observables like the uv slope , the colors , and the luminosities and equivalent widths of the hydrogen lines can be readily related to the dust - induced color excess of the ionized gas and , through equation 9 , of the stellar continuum . the uv slope is correlated with e(b@xmath26v)@xmath117 ( figure [ fig6 ] panel a ) , and equation 8 gives : @xmath143 where a list of @xmath144 values is given in the table of appendix b. the slope of the correlation between @xmath118 and e(b - v)@xmath117 is somewhat steeper than that measured by @xcite directly from the data ( 1.88@xmath1450.28 ) , but the two are within 1 @xmath72 of each other . it should also be noted that equation 10 becomes a lower envelope to the data points in figure [ fig6 ] , panel a , for @xmath1462.25 , a value appropriate for dust - free , star - forming populations . this indicates that variations of the dust / stars / gas geometry from the schematic model of figure [ fig8 ] and/or of the stellar population from the theoretical extreme @xmath1472.25 has the effect of reddening the uv slope even for e(b - v)@xmath1410 . the vertical scatter in figure [ fig6 ] , panel a should thus be interpreted as due to such variations from galaxy to galaxy . reddening expressions for optical colors can be directly derived from equation 8 ( see , also , figure [ fig6 ] , panels c and d ) . a mixed uv optical color , 0.28@xmath26v , could be useful to infer the reddening of intermediate redshift ( @xmath18z@xmath1480.6 ) starburst galaxies observed in b and i , and the reddening of one relative to the other is given by : @xmath149 under the assumption that the restframe uv emission and the nebular lines are both due to massive stars , the reddening of the uv continuum relative to the nebular hydrogen emission is given by : @xmath150 figure [ fig6 ] , panel e shows the comparison between data of local galaxies and equation ( 13 ) @xcite . while the agreement between the data in the first four panels of figure [ fig6 ] and equation 8 is a direct consequence of the fact that those data were used to derive the starburst obscuration curve , the match between data and expected trend in the bottom two panels of the figure is a consistency check , since it was not imposed a priori . one immediate consequence of the differential reddening between gas and stars is that the equivalent widths of the nebular lines depend on the amount of gas obscuration ; for the h@xmath42 and h@xmath41 emission lines the relations are @xcite : @xmath151= -0.64 e(b - v)_{gas},\\ \log [ ew(h\alpha)_{a}/ew(h\alpha)_{o}]= -0.40 e(b - v)_{gas},\end{aligned}\ ] ] where the attenuated and intrinsic ews are directly compared . comparison of these trends with the data points is shown in panel f of figure [ fig6 ] ; the trend given by the obscuration curve marks an upper envelope to the ensemble of observed ews , because the presence of older stellar populations underlying the starbursts dilutes the contrast between the emission lines and the stellar continuum in the data . statistically speaking , the uv slope is a good estimator of the total obscuration affecting the starburst ( * ? ? ? * and figure [ fig7 ] ) ; in particular , the ratio of the dust infrared luminosity , l@xmath93 , to the stellar uv or blue luminosity , l@xmath122 or l@xmath74 , is related to @xmath118 via : @xmath152 \simeq { a_{0.16 , star}\over 2.5 } = 0.84 ( \beta_{26}-\beta_{26 , 0}),\\ \log \biggl[{1\over 3.85 } \biggl({l_{ir}\over l_b}\biggr)_{star } + 1\biggr]\simeq { a_{b , star}\over 2.5 } = 0.43 ( \beta_{26}-\beta_{26 , 0}).\end{aligned}\ ] ] the two constant values on the left - hand - side of equations 17 and 18 , 1.68 and 3.85 , represent the bolometric corrections of the stellar emission relative to the uv or b. the correction for the b - band is larger and subject to more uncertainties than that for the uv , hence the larger scatter in the data point of figure [ fig7 ] , panel b. equation 17 can be rewritten as : @xmath153 = 1.76 e(b - v)_{gas},\ ] ] where the equation above represents a lower envelope to the data ( figure ( 7 ) , panel d and * ? ? ? * ) , and has a scatter comparable to that of figure [ fig6 ] , panel a. the nature of the scatter in the @xmath118 vs. e(b - v)@xmath117 and in the l@xmath93/l@xmath122 vs. e(b - v)@xmath117 relations is the same : it is due to variations from galaxy to galaxy of the starburst populations and of the dust / star / gas geometry . as discussed in the introduction , the energy balance between uv optical and infrared cosmic background suggests that there have been infrared luminous galaxies at all epochs . the number density of quasars was @xmath19100 times higher at z@xmath12.5 than today ( e.g. , * ? ? ? * ) and star formation activity was @xmath13 to 10 times higher at z@xmath511 than at z@xmath10 @xcite , flattening or decreasing only beyond z@xmath112 @xcite . if the association between activity and dust opacity observed in the local universe ( equation 6 ) holds at high redshift as well , galaxies in the past may have been as dusty as or dustier than today ( section 5.2 ) . among the methods described in section 3 , the multiwavelength comparison is the only practical approach for measuring the dust opacity of galaxies beyond the local few hundred mpc . however , because of observational difficulties , the wavelength coverage of the typical sample is limited to a few windows , often with little overlap from sample to sample . measures of the content , nature , and opacity effects of the dust in distant galaxies are therefore still at a very early stage , and the picture is vastly incomplete and controversial . the very first evidence for presence of dust in high redshift systems came from studies of damped ly@xmath42 systems ( dlas , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? dlas are the largest hi column density absorption systems seen in the spectra of background quasars , with n(hi)@xmath15410@xmath155 cm@xmath116 . their metallicities cover the range @xmath11/1001/10 solar for z@xmath10.54 , with a mean value @xmath11/13 solar at z@xmath10.53 and little evidence for evolution with redshift @xcite . selection biases against the dustiest , and most metal - rich dlas and problems with dust depletion corrections have been advocated as possible reasons for the observed weak metallicity evolution ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * see , however , @xcite ) . the dust fraction in dlas is measured by comparing the abundance of dust - depleted metals such as cr ii , fe ii , and si ii against that of a metallicity tracer such as zn ii , all of them producing rest - frame uv absorption lines @xcite . the dust - to - metals ratio in dlas is in the range 50%60% of the milky - way value ; this , combined with the generally low metallicities , indicates that dlas are relatively transparent systems , with dust - to - gas ratios between 2% and 25% of the milky - way value @xcite . the absence of the 0.2175 @xmath0 m dust absorption associated with dlas led @xcite to conclude that the extinction curve in these systems is probably similar to the one observed in the magellanic clouds . more recently , though , @xcite reported a @xmath13 @xmath72 detection of the dust absorption feature in the composite spectra of mgii absorbers . although presence of modest quantities of dust is well established in dlas , the weak or absent evolution of the metal content and of the cosmological mass density with redshift @xcite questions their relationship to luminous galaxies , and favors the interpretation that dlas preferentially trace galaxies with extended gas distributions and low metal and dust contents @xcite and the unevolved outskirts of galaxies @xcite . in recent years , facilities / instruments like iso and the sub - millimeter common user bolometer array ( scuba ) at the james clerk maxwell telescope ( jcmt ) have achieved high enough sensitivities in the mid and far infrared to detect _ dust emission _ beyond a few @xmath0 m from statistically significant samples of galaxies at cosmological distances . studies on the iso and scuba surveys , especially the identification of the optical / near ir counterparts to the infrared sources and the redshift measurements , are still ongoing and the results described below should be considered in many instances preliminary . iso surveys have yielded @xmath301,500 sources in the redshift range 0@xmath57z@xmath571.5 , down to a sensitivity limit of @xmath300.1 mjy in the mid ir ( 6.75 @xmath0 m , 12 @xmath0 m , and 15 @xmath0 m ) and @xmath300.1 jy in the far ir ( 90 @xmath0 m and 175 @xmath0 m , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? full lists of all surveys are given in @xcite and @xcite . the number counts at 175 @xmath0 m account for @xmath110% of the cib @xcite , showing that the bulk of the background has not been resolved yet at these wavelengths . neverthless , the number counts at mid and far ir wavelengths are about 310 times above the no - evolution extrapolation of the local iras counts @xcite . the implication is a strong luminosity evolution , l@xmath157(1@xmath76z)@xmath158 , or a combination of luminosity and density evolution of the infrared sources out to z@xmath11 @xcite , that is comparable to or stronger than the luminosity evolution observed at restframe uv wavelengths out to the same redshift @xcite . the iso counts are thus dominated by luminous infrared galaxies at z@xmath571 , with l@xmath159l@xmath16010@xmath100 h@xmath161 l@xmath37 @xcite . although this conclusion is based on the highly uncertain conversion of the 15 @xmath0 m emission1 the flux density at 15 @xmath0 m is centered on the restframe emission of the aromatic features that dominate the region between 6 @xmath0 m and 12 @xmath0 m ; the relation between the intensity of the features and the total infrared emission is still unsettled @xcite . ] to a total infrared luminosity , it is qualitatively supported by the optical counterparts of the iso sources . close to 100 mid ir sources , with median redshift @xmath18z@xmath1480.70.8 have been so far identified at optical / near ir wavelengths @xcite [ @xmath18z@xmath1480.10.2 for the 12@xmath0 m sample @xcite ] . many of the restframe optical spectra show characteristics of dusty luminous starbursts , with current bursts of star formation superimposed on a prominent population of a stars , and with attenuation values a@xmath1620.51.8 mag @xcite . at the time of writing , scuba surveys at 850 @xmath0 m of ` blank ' fields and targeted lensed objects have detected a total of @xmath30100 sources with flux densities in the range @xmath10.5 mjy to @xmath5110 mjy and redshifts between z@xmath11 and 3 , although sources as far away as z@xmath305 may be present @xcite . the number counts down to 2 mjy and to 0.5 mjy account for @xmath120%30% and more than 90% , respectively , of the cib @xcite . however , the latter fraction is potentially affected by small number statistics : so far only a few sources have flux densities below 2 mjy , the scuba confusion limit in blank fields , all of them from the survey of lensed galaxies @xcite . observations at 850 @xmath0 m probe the raileigh - jeans tail of the thermal dust emission of galaxies at least to redshift @xmath3057 . a source at z@xmath1541 and with an 850 @xmath0 m flux density of 2 mjy corresponds to a luminosity l@xmath1022@xmath9110@xmath103 h@xmath161 l@xmath37 , for typical seds of local infrared selected galaxies @xcite ; this is roughly the luminosity of a ulirg like arp220 . the few optical / near ir counterparts which have been identified are faint , often with i@xmath1924 and k@xmath512021 @xcite ; in most cases the ratio l@xmath93/l@xmath1630.05 , where l@xmath164 is the luminosity in the observer frame r band , and the infrared emission accounts for more than 95%99% of the bolometric luminosity . it is indeed widely accepted , also on the basis of the few available multi - wavelength seds , that the scuba sources are ` cosmological ulirgs ' @xcite , although this generalization has been recently questioned @xcite . the median redshift of the surveys is @xmath18z@xmath1652.5 @xcite ; however , @xmath1810% of all the sources have reliable redshift determinations from optical / near ir counterparts , and a total of @xmath3050% of the sample have photometric redshifts , most often derived from the radio / sub - mm spectral - index relation of @xcite . the comoving number density of scuba sources in the redshift range z=13 brighter than 6 mjy is @xmath135@xmath9110@xmath81 h@xmath166 mpc@xmath89 in our cosmology @xcite , about two orders of magnitude larger than the corresponding number density of local ulirgs brighter than 10@xmath103 l@xmath37 @xcite . a strong luminosity evolution , possibly as strong as l@xmath157(1@xmath76z)@xmath167 to z@xmath12 , is also implied by the number counts ( e.g. , * ? ? ? although the statistical properties of the sources are often inferred using the star - forming arp220 as a template , a fraction of agns is present in the sample @xcite . current best estimates , based on the x - ray luminosity of the sources favor an agn fraction @xmath120% @xcite ; this value will generally depend on the source luminosity range , as it does in the local universe @xcite . multi - wavelength observations of optically - selected samples of intermediate and high redshift galaxies are generally limited to the rest - frame uv and optical windows , with few exceptions ( e.g. , * ? ? ? * ; * ? ? ? such surveys suffer the same limitations of local uv optical samples , with the added complication of the often incomplete wavelength coverage . by probing only the _ differential attenuation _ , the uv optical data provide , in principle , lower limits to the actual dust opacity of the galaxy ( section 2.2 , but see also section 4 ) . in order to differentiate an ageing stellar population from dust as the cause of the sed reddening , a variety of approaches have been adopted by different authors . common diagnostics include the balmer line decrement h@xmath42/h@xmath41 , line - to - continuum ratios , e.g. h@xmath42/uv and h@xmath41/uv , and multi - color sed fitting . limitations to each of these diagnostics are discussed in section 2.2 and in the sections below , as appropriate . for a uv - selected sample of star - forming galaxies at z@xmath300.2 , @xcite and @xcite derive an average attenuation a@xmath1060.8 mag and find tentative evidence that those systems follow an obscuration sfr correlation similar to that observed in more local galaxies @xcite . the authors use a combination of uv ( 0.20 @xmath0 m ) , optical ( h@xmath42 and h@xmath41 ) and radio ( 1.4 ghz ) data to analyze their sample , although their reddening estimates are mostly based on the subsample for which they can measure h@xmath42/h@xmath41 ratios . if the starburst obscuration curve ( section 4 ) applies to z@xmath300.2 star - forming galaxies , an attenuation a@xmath1060.8 mag for the nebular line corresponds to a@xmath1681.3 and a@xmath1231.4 for the uv stellar continuum . although these figures agree with local measurements of the extinction at uv and h@xmath42 in star - forming galaxies ( section 3.6 and table [ tbl-3 ] ) , the uv selection of the sample potentially biases the reddening measures towards low values by excluding more heavily extincted galaxies . the z@xmath11 balmer break sample of @xcite is also a restframe - uv / blue selected one . exploiting the presence of the balmer discontinuity at @xmath10.365 @xmath0 m in galaxy seds to select z@xmath11 candidates ( a derivative of the lyman - break technique , see next section ) , these authors have obtained spectroscopic confirmation for @xmath30700 galaxies . the sample contains a large variety of galaxies , from active starbursts to quiescent star - forming to post - starbursts . if the z@xmath11 galaxies with blue ( @xmath1690.3 ) seds are analogs of local starbursts and follow the same ir / uv - versus-@xmath41 relation , they also follow an obscuration sfr relation similar to the local one , with a comparable range of attenuations ( figure [ fig9 ] , panel a ) . however , for fixed opacity , the z@xmath11 galaxies tend to have on average brighter uv@xmath76ir luminosities ( larger sfrs ) than the z@xmath10 galaxies . thus , even taking selection effects into account , the distant galaxies are , on average , more actively star - forming for the same amount of obscuration than the local starbursts @xcite . @xcite , @xcite , and @xcite use restframe h@xmath42 observations and h@xmath42/uv comparisons to infer dust reddening in star - forming galaxies at redshift z@xmath112.2 . the i - band selected sample of @xcite corresponds to restframe @xmath1v for a z=1 galaxy : the longer wavelength selection of the samples mitigates , although does not remove completely , the bias against dust obscured galaxies introduced by uv selections [ for the canada - france redshift survey employed by @xcite , see @xcite , and for the sample used by @xcite , see @xcite ] . the presence of dust reddening in the z@xmath112.2 samples is derived from the discrepancy between the measured l@xmath170/l@xmath171 luminosity ratio and the ratio predicted by dust - free models / l@xmath172 ratio ; the small extrapolation to l@xmath170/l@xmath171 is done here assuming a constant star formation population . ] . figure [ fig10 ] compares the range of attenuated to intrinsic values ( l@xmath170/l@xmath171)@xmath173/(l@xmath170/l@xmath171)@xmath174 derived from data by the three groups of authors / l@xmath171)@xmath174 is an average of h@xmath42 and uv luminosities for a variety of imfs , star formation histories , and metallicities of stellar populations , see table 3 in @xcite . ] with predictions from the dust models of section 2.2 . geometries where dust and stars are homogeneously mixed , either in spheroidal or flattened distributions , appear less effective in reproducing the observed h@xmath42/uv ratios than foreground geometries ( figure [ fig10 ] ) . if the foreground and starburst distributions bracket the possible range of dust / emitter morphologies for the z@xmath112 h@xmath42 galaxies , the front - to - back optical depths are @xmath470.64.9 mag ( figure [ fig10 ] , bottom panel ) , and the effective attenuations are a@xmath1060.64.3 mag , a@xmath1751.55 mag , a@xmath12337 mag , for smc extinction . mw - like and smc - like extinction curves give results that are less than 0.1 mag apart at @xmath1760.28 @xmath0 m , but @xmath11.2 mag different at @xmath24=0.16 @xmath0 m , with larger obscurations given by the smc extinction curve . uncertainties in the metallicity , imf , and star formation history of the galaxy stellar populations introduce a factor @xmath512 additional scatter on the inferred opacities ( table 3 in * ? ? ? the major limitation is , however , that dust opacities measured from the current h@xmath42 surveys of galaxies are probably not representative of the z@xmath112 population as a whole . by number , @xcite s and @xcite s samples are small ( 13 galaxies and 5 galaxies , respectively ) ; @xcite s sample , albeit 2.5 times larger than @xcite s , does not measure the h@xmath42 emission on the same galaxies for which uv is available . for such small samples sensitivity limits induce a strong selection effect : the observed galaxies will be among the brightest in h@xmath42 and , in addition , contamination from agn / seyfert nuclei is not excluded @xcite . therefore , such surveys sample the high end of the activity and dust obscuration ranges for optically - detectable systems at z@xmath112 ( * ? ? ? * ; * ? ? ? * and equation 6 ) . in summary , the limited information available on galaxies at z@xmath301 suggests that dust opacities cover a range similar to that of local star - forming and starburst galaxies ; they may follow a similar relation a@xmath177(sfr)@xmath178 ( see equation 6 ) , with the proportionality constant revised downward relative to the local sample . to date , the surveys that have most efficiently secured large populations of galaxies beyond z@xmath12 are those that exploit the lyman discontinuity at restframe 0.0912 @xmath0 m to identify high - redshift candidates ; these are the so called lyman break galaxies ( lbgs , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? so far , there are @xmath11,000 spectroscopically confirmed lbgs at z@xmath13 , and @xmath170 at z@xmath14 . the lbg population is a substantial component of the high redshift `` zoo '' : their comoving number density at z=3 is n(z=3)@xmath13@xmath9110@xmath116 h@xmath166 mpc@xmath89 @xcite , comparable to the density of local galaxies @xcite . by selection , i.e. due to the need of presenting a relatively blue continuum longward of ly@xmath42 , this high redshift population tends to be biased against systems with predominantly old populations , or systems with large amounts of dust extinction . indeed , most lbgs show evidence for ongoing or recent star formation activity @xcite . from a purely phenomenological point of view , lbgs resemble local uv - selected starburst galaxies ( section 4 ) . the obscuration corrected sfrs per unit galactic area are @xmath10.225 m@xmath37 yr@xmath5 kpc@xmath116 , for a 0.1100 m@xmath37 salpeter imf @xcite , corresponding to @xmath10.5%50% the maximum sfr per unit galactic area measured in the local universe @xcite ; this is similar to the observed range in nearby uv - selected starbursts . restframe uv spectra show a wealth of absorption features , and often p - cygni profiles in the civ(1550 ) line @xcite , which indicates the predominance of young , massive stars . nebular emission lines observed in the brightest lbgs @xcite have intensities comparable to those of local starbursts @xcite . the observed blueshifts in the uv interstellar absorption lines relative to the systemic redshifts of lbgs have been interpreted as bulk gas outflows with velocities v@xmath1200500 km s@xmath5 @xcite , again very similar to what is observed in nearby starbursts @xcite . finally , more luminous lbgs tend to be dustier ( figure [ fig9 ] , panel b and equation 6 ; * ? ? ? * ; * ? ? ? however , dissimilarities also exist between the two classes of galaxies . star formation extends over a large fraction of the galactic body in lbgs , typically 530 h@xmath161 kpc@xmath113 @xcite , while it is confined within the inner kpc@xmath113 in local starbursts . as a result , lbgs are , on average , more rapidly star - forming than either z@xmath10 or z@xmath11 galaxies ( figure [ fig9 ] ; * ? ? ? * ; * ? ? ? the bulk outflows in lbgs have velocities that are similar to those of local infrared bright starbursts like ngc6240 and a few times larger than those of hii galaxies @xcite . the implications of such differences are not fully understood yet . establishing how much dust lbgs contain is the subject of ongoing debate . metals are definitely present in these systems . although the exact value of the metallicity has not been constrained , it most likely lies between 0.1 z@xmath37 and 0.7 z@xmath37 @xcite . the uv spectral slopes of lbgs have median value @xmath1791.5 @xcite , significantly redder than the value @xmath1802.2@xmath262.7 expected for a dust - free star - forming population ( appendix b ) . if interpreted in the same fashion as the uv slope of nearby uv - selected starbursts , lbgs suffer from dust obscuration at the level of a@xmath12312 mag @xcite , and , possibly , more @xcite . however , a number of uncertainties and systematics limit the usefulness of uv data alone to derive obscuration estimates . first and foremost , the high redshift galaxy for which reddening corrections are sought needs to possess active , on - going star - formation ( i.e. , not to be a post - starburst ) ; this appears to be true for the lbgs as a population , but may not be true on a individual basis @xcite . second , for most lbgs the uv slope is derived from photometry alone ; the conversion of g@xmath181 colors filters have central wavelengths 0.483 @xmath0 m and 0.693 @xmath0 m , respectively @xcite . ] to @xmath118 suffers from uncertainties due to the presence of both intrinsic ( stellar lines and lyman break ) and intervening ( lyman@xmath42 forest ) absorption in the photometric filters @xcite . finally , because of the age dust degeneracy , the intrinsic uv slope is a free parameter that can lead to @xmath191 mag difference in the derived obscuration value at 0.16 @xmath0 m ( compare * ? ? ? * ; * ? ? ? supplementing the uv with restframe optical data provides further leverage for determining the level of dust reddening in lbgs . the negligible correlation between l@xmath182/l@xmath122 and the uv slope found for a sample of @xmath5715 luminous lbgs(0.5007 @xmath0 m ) emission line @xcite are not discussed because of the strong dependence of the nebular line intensity , up to one order of magnitude variation , on the chemical and physical conditions of the gas @xcite . ] supports the notion that lbgs are subject to modest reddening @xcite . although this result is still dominated by the large scatter induced by both observational and model uncertainties @xcite , it is consistent with what is observed in local uv - selected starbursts1.5 corresponds to obscurations a@xmath1231.2 mag and a@xmath1830.50 mag , and a differential obscuration between the two of @xmath10.7 mag ; the latter is probably too large to be reconciled with observational data . highly inclined disks are not considered in the obscuration budget , as they would probably be too dim in the uv to be selected as lbgs . ] , for which an obscuration of 12 mag at 0.16 @xmath0 m corresponds to the modest reddening a@xmath184a@xmath1830.150.3 ( section 4.3 ) . although the age dust degeneracy can not be easily broken even with the addition of long wavelength data , extending the wavelength coverage of lbgs to the restframe optical has helped delimiting the locus of allowed combinations in the age dust plane @xcite . using complementary datasets of lbgs in terms of luminosity range , @xcite and @xcite have established that there is an inverse correlation between the age of the stellar populations in a galaxy and the amount of dust reddening it suffers : the older the galaxy , the lower its reddening . in agreement with previous results @xcite , lbgs whose seds are compatible with young ( @xmath5710@xmath88 yr ) stellar populations are also dusty . whatever dust is present , it is not preferentially obscuring selected regions of the galaxies , because of the similarity between the restframe uv and optical morphologies @xcite . a typical lbg obscuration of a@xmath12312 mag , with a median of @xmath11.5 mag and a high - end tail around 4.55 mag @xcite , is currently the most broadly accepted value . despite all the uncertainties in estimating dust obscuration in lbgs , the inferred opacity values are modest enough that the high redshift galaxies are expected to be faint sub - mm emitters , with 850 @xmath0 m flux densities at or below the scuba detection limit of @xmath112 mjy @xcite . out of @xmath5110 lbgs targeted with scuba , only a couple have been detected @xcite . table [ tbl-5 ] gives three examples ( two lbgs and one scuba source ) of highly reddened z@xmath13 galaxies , plus the average values of a typical lbg . the three examples are meant to show some of the most attenuated cases that are also detected at restframe uv . table [ tbl-5 ] demonstrates that the exact prediction of the flux density at restframe @xmath30200 @xmath0 m is heavily dependent on the adopted dust temperature(s ) and emissivity , because long infrared wavelengths only probe the raileigh - jeans tail of the thermal dust emission in galaxies ( figure [ fig11 ] ) . local galaxies , although extensively used as analogs of high redshift galaxies , are of limited help in this respect , given the wide range of properties they show . if the flux density of a local galaxy beyond 6070 @xmath0 m is parametrized by a modified blackbody with dust emissivity @xmath185 : @xmath186 ulirgs and bright ir - selected galaxies tend to be well described by either a single - temperature component with t@xmath140 k and @xmath1871.5 @xcite or by a warm / cool two - temperature model with @xmath188=2 @xcite ; however , there is tentative indication that uv bright , metal - poor starbursts have hotter dust seds , with t@xmath150 k and @xmath188=2 ( tol1924416 , * ? ? ? equation 20 , which is used for purely illustrative purposes , is clearly an over - simplification of the dust sed in local galaxies , which is characterized by a range of temperatures and , at restframe @xmath294070 @xmath0 m , by the emission from non - equilibrium dust grains ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * see , also , section 2.1 ) . in starbursts , the latter component represents @xmath125% of the total infrared emission . given the uncertainty of the sub - mm measures and the current poor understanding of the detailed physics of dust emission in both local and high - redshift galaxies , the opacity of lbgs is still relatively unconstrained by sub - mm observations ( * ? ? ? * ; * ? ? ? * and table [ tbl-5 ] ) . further complications ( and uncertainties ) arise when the comparison between predictions and observations of infrared flux densities is made for individual objects @xcite ; reddening / obscuration corrections at uv - optical wavelengths are accurate only in a statistical sense and , generally , do not apply on a one - to - one basis to individual objects . for instance , if the dust geometry in a particular system is such that the scattering out of the line of sight is larger than that into the line of sight , the infrared energy , which is due to the absorption part of the extinction only , will be overestimated when derived from uv data . there is so far limited overlap between infrared and optical surveys at intermediate and high redshift ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? consequently multi - wavelength coverage from the restframe @xmath1uv optical to the infrared is available only for a small number of galaxies . to make matters worse , observational limitations lead to vastly different selection criteria for the optical surveys in different redshift bins @xcite . thus , conclusions about the dust opacity of galaxy populations as a whole and as a function of redshift are at the moment based on ` reasonable ' extrapolations of existing data , rather than hard evidence . available infrared and optical data support the scenario that z@xmath301 star forming galaxies cover at least the same range of dust opacities as local galaxies . the luminosity ( or combined luminosity and density ) evolution implied by the iso counts suggests that luminous infrared galaxies , with l@xmath159l@xmath16010@xmath100 h@xmath161 l@xmath37 , were common at z@xmath571 . the range of dust attenuation values at uv wavelengths inferred for optically - selected z@xmath11 galaxies is comparable to those of local disk galaxies and uv selected starbursts ( figure [ fig9 ] ) , although the distant galaxies tend to be more luminous ( more active ) on average . larger samples with more extended wavelength coverage are needed before a more complete picture can be drawn in this redshift range . the ism of z@xmath13 galaxies has been observed to be metal - polluted to at least @xmath300.1 z@xmath37 , implying that non - negligible amounts of dust must be present in the galaxy population at large by z=3 and , in many cases , earlier @xcite . the observed properties of the z@xmath191 scuba sources and of the lbgs suggest that both objects are the high redshift analogs of local actively star - forming galaxies , where dust content and level of activity determines a continuum of characteristics . the scuba sources are at the high end of such continuum , with powerful activity , both thermal and non - thermal , and large dust contents that make them similar to local ulirgs ; more than 95% of their bolometric light is absorbed by dust . lbgs appear to span a larger range of the continuum of dust content and activity level , although they tend to have , on average , lower values than the scuba sources . typical uv attenuations are a@xmath12312 mag , corresponding to 35%95% of the bolometric light being absorbed by dust . assuming the local infrared luminosity function can be applied to high redshift sources as well , the comoving number density of z@xmath303 scuba galaxies with 850 @xmath0 m flux densities brighter than 2 mjy and 0.5 mjy ( corresponding to @xmath12@xmath9110@xmath103 l@xmath37 and @xmath15@xmath9110@xmath100 l@xmath37 for an arp220-like sed ) would be @xmath304@xmath9110@xmath82 h@xmath166 mpc@xmath89 and @xmath3010@xmath116 h@xmath166 mpc@xmath89 , respectively . high - redshift infrared galaxies as bright as arp220 are almost 100 times less numerous than lbgs ; for comparison , in the local universe ulirgs are more than 4 orders of magnitude less numerous than optically selected galaxies . at lower 850 @xmath0 m luminosities , z@xmath13 scuba sources are about as numerous as lbgs at the same redshift , but there could be large overlap between the two samples at the mjy level . lbgs are , in fact , predicted to occupy the faint end of the scuba 850 @xmath0 m source counts , with flux densities of a few mjy or less @xcite . thus , although scuba sources down to 0.5 mjy at all redshifts already account for @xmath1990% of the cib , a non - negligible fraction of this , at least 25% , could be due to lbgs @xcite . intracluster and intergalactic dust are briefly discussed in view of their potential importance in the interpretation of the intermediate and high redshift universe . the amount , nature , and sources of the dust in the intracluster ( icm ) and intergalactic ( igm ) media are still a controversial matter , despite the large number of independent researches tackling the issue . the metallicity of the present - day icm is fairly high , about 1/31/2 solar , implying that a non - negligible fraction of the gas in the icm has been ejected from the cluster s galaxies @xcite . the high metallicity is not automatically accompanied , however , by a high dust content of the icm . the icm hot gas effectively destroys dust through sputtering . the typical timescale for sputtering in a @xmath5110@xmath87 k hot ambient gas is t@xmath18910@xmath78a / n , where t@xmath190 is measured in yrs , a is the size of the dust grain in @xmath0 m , and n is the density of the icm in cm@xmath89 @xcite . in the high - density cores of clusters ( n@xmath5110@xmath89 cm@xmath89 ) , a 0.1 @xmath0 m dust grain survives a few times 10@xmath88 yr , with smaller grains being destroyed on shorter timescales than the large grains @xcite . small grains can survive only outside the cores of clusters , i.e. , beyond the central @xmath30mpc . dust needs to be constantly replenished in the icm , and the likely sources are enriched winds from the cluster s galaxies , intracluster stars , and , in dynamically young clusters , winds from infalling spiral galaxies @xcite . the timescales for the dust injection in the icm are of order 10@xmath90 yr or longer , implying that the grain size distribution , and therefore the shape of the extinction curve , depends mainly on the sputtering efficiency at the selected cluster location @xcite . various direct and indirect methods have been employed to detect dust in the icm . deficits in background counts of galaxies , quasars , and clusters behind galaxy clusters of various richness have been interpreted as extinction in the range a@xmath670.20.5 , with patches that can exceed a@xmath1921 mag @xcite . conversely , reddening measurements of cluster members @xcite or background galaxies @xcite give upper limits of a@xmath670.2 mag . the two methods thus give contradictory results and neither is completely bias - free : deficits in background counts can be due not only to dust , but also to confusion ; low reddening values can be attributed to a gray extinction curve , greyer than a milky - way - like curve , because of the selective destruction of small grains in the icm . tentative infrared detections of icm dust emission , using iras and iso data @xcite , are either at low significance level , or have been argued to be due to emission from cluster members @xcite . presence of dust is well established in the centers of cooling flow clusters ( see discussion in section 3.4 ) , but the issue of its origin is still controversial . in particular , it is not clear whether this is an extended component , such as condensates from the cooled cluster gas or a by - product of star formation triggered by the cooling flow , @xcite , or is directly associated with the central galaxy , possibly from a merging event that the galaxy underwent @xcite . the milky - way - like form of the extinction curve in the central galaxies has been extensively used as an argument to support the galactic origin of the dust in cooling flow clusters ; as remarked above , icm - processed dust would tend to be depleted in small grains and therefore produce a gray extinction @xcite . however , if soft x - ray absorption is present in the centers of the clusters , dust absorption in the cooling flow would appear to be the direct physical explanation @xcite . the tentative detection of the neutral oxygen edge , and thus of dust , in the perseus cluster @xcite and the widespread presence of co emission in low - radio - emitting cooling flow clusters @xcite are countered by xmm data that do not support the presence of large amounts of cool gas in the center of some of the clusters and , therefore , do not require a large amount of dust to explain the soft x - ray spectral shape @xcite . at present times , the presence and nature of dust in the igm is even more uncertain than that in the icm , though it is widely accepted that the igm is metal - enriched . hydrodynamical cdm simulations identify the ly@xmath42 forest as the igm , i.e. , modest baryon density fluctuations in a ionizing background @xcite . the ly@xmath42 forest accounts for a significant fraction , @xmath1950% , of the baryons at z@xmath303 @xcite ; its metallicity at z@xmath12.53 is @xmath10.21% z@xmath37 down to column densities n(hi)@xmath110@xmath193 cm@xmath116 @xcite , while metallicity values at lower column densities remain uncertain @xcite . these values could be lower limits to the igm metallicity , if there is a ` missing metals ' problem as pointed out by @xcite and @xcite on the basis of the metals produced by star formation by z@xmath12.5 . the metallicity of the ly@xmath42 forest increases for decreasing redshift , being @xmath110 times more metal rich at z@xmath10.5 than at z@xmath12.53 @xcite . in the local universe , the ly@xmath42 forest accounts for only @xmath3020% of the baryons @xcite and matter associated with galaxies and clusters for another @xmath3025%30% @xcite ; thus , @xmath150% of the baryons at z=0 , and possibly of the metals , are contained in a yet undetected phase of the igm , proposed to be hot gas @xcite . various mechanisms have been considered for the metal , and possibly dust , pollution of the igm ; they include supernova ejecta from galaxies and/or subgalactic structures , dynamical encounters between galaxies , ram - pressure stripping , and population iii stars @xcite . the different mechanisms are non - exclusive , although each one could be dominant relative to the other at specific cosmological epochs . early attempts to detect diffuse dust in the igm investigated the evolution of the mean quasar spectral index with redshift @xcite , with the basic result that presence of igm dust is negligible for milky - way - like extinction , up to at least z@xmath302 @xcite . milky - way - like igm dust should also leave an imprint on the cosmic microwave background , as it would re - emit in the sub - mm / mm regime the absorbed background radiation @xcite . the tight cobe upper limit to deviations from blackbody emission @xcite limits the igm dust to a@xmath1940.3 mag to z=5 , with negligible value to z=3 @xcite . however , if sputtering is efficient in the hot igm , dust has a greyer extinction than milky - way - like ( * ? ? ? * for a challenge to the sputtering argument see @xcite ) . gray igm dust can be present in larger quantities than milky - way - like dust without violating the cobe limits , but its emission would represent @xmath5175% of the cib at 850 @xmath0 m @xcite . source counts with scuba account already for @xmath1990% of the cib ( section 5.2 ) , which challenges the case for large amounts of gray dust . so far , the strongest limits have come from observations of cosmological supernovae . for gray dust with minimum grain size @xmath10.1 @xmath0 m , @xcite s model gives a@xmath74-a@xmath1940.01 and a@xmath74-a@xmath500.2 for a@xmath860.20.3 to z=0.5 ; at z=2 , a@xmath860.60.7 mag . using observations of type 1a supernovae at z@xmath10.5 and z@xmath11.7 , @xcite and @xcite have placed a limit of a@xmath860.10.15 to the amount of diffuse , gray igm dust to z=0.5 . in summary , although the actual numbers are still not settled , igm dust is likely to produce small attenuations at most redshifts . current observational data indicate that the population of local galaxies is characterized by modest values of the dust opacity . typically , 50% or less of the total stellar energy is absorbed by dust . the same evidence points out that the definition of ` modest opacity ' is wavelength - dependent ; for example , in the uv , at 0.15 @xmath0 m , a non active galaxy can lose up to 80% of its energy to dust . luminous , intermediate to late type spirals are dustier than ellipticals , by @xmath11.01.4 mag in the uv , and than irregulars , by 0.50.8 mag . luminous galaxies are on average more metal rich and dust rich than their faint counterparts . dust opacity also increases with thermal and non - thermal activity , with the high - end of the distribution given by the ulirgs . although the wealth of data and wavelength coverage for local galaxies has started to uncover the dust opacity of these systems , we are still missing a few crucial pieces of the puzzle . the largest among these is the one - to - one relation between absorbed stellar seds and dust emission seds . at first order , hotter stars are expected to heat the dust to higher temperatures than cooler stars @xcite , but we do nt have yet the direct link between individual dust emission and heating components ; we do nt know the dependence of the dust sed on parameters like metallicity , radiation field , gas density , galactic environment , etc . ; and questions still surround the relation between radiation intensity and pah emission , despite the progresses made with iso @xcite . the main reason for these limitations to our comprehension of the dust absorption emission relationship is the large mismatch in angular resolution between uv - to - near ir observations and current mid / far infrared data . while observations probing the direct stellar light can easily reach sub - arcsecond resolution , existing data on dust emission are at the arcminute resolution level , especially at wavelengths above @xmath140 @xmath0 m where the bulk of the energy from dust emerges . upcoming infrared missions like sirtf ( the space infrared telescope facility ) , sofia ( the stratospheric observatory for infrared astronomy ) , and first ( now the herschel space observatory ) promise to bridge this gap . they will also push the wavelength coverage all the way to 600700 @xmath0 m , thus probing the cold dust emission in local galaxies , which is the repository of most of the dust mass . uv and optical selected , intermediate and high redshift , galaxy surveys carry the inherent limitation that physical parameters directly derived from the observations are affected by dust in a non - easily quantifiable way . this is further complicated by the fact that instrumental sensitivity limits favor detection of the luminous representatives of any galaxy class , that may tend to have relatively high dust content . estimates of the sfrs in intermediate and high redshift galaxies using balmer lines @xcite and infrared emission @xcite are generally a few times higher than the analogous estimates using the restframe uv emission @xcite . as discussed in section 5 , presence of dust in galaxies at cosmological distances is reasonably established , but the quantification of its effects as a function of redshift is still an ongoing process . linked to this is the puzzle of the nature of the uv selected and ir selected sources , and of their relationship : whether they represent overlapping populations , or the two ends of a continuum of properties possibly parametrized by activity and dust content is still an open question . the absence of full characterization of the dust emission and absorption in distant galaxies is mostly due to the scant overlap between optically selected and infrared selected surveys . this is due to sensitivity limitations at both ends of the spectrum , to angular resolution limitations at long wavelengths , to the small number of objects in some of the samples , and to the sparse wavelength coverage of the dust emission sed by current sub - mm / radio instrumentation . sirtf and first initially , and then alma ( the atacama large millimeter array ) and ngst ( the next generation space telescope ) will be able to fill the gaps . in particular the last two instruments , with their sub - arcsecond resolution and high sensitivities , should mark a break - through in our understanding of the high redshift universe and of its stellar , gas , and dust content up to z@xmath120 . the author is grateful to mark dickinson , megan donahue , paul goudfrooij , claus leitherer , and mark voit for insightful comments on specific sections of this review . she thanks kurt adelberger for data on the lyman break and balmer break galaxies , and max pettini and chuck steidel for the uv spectrum of ms1512-cb58 . she is also indebted with veronique buat , bruce draine , tim heckman , nino panagia , marcia rieke , and max pettini for a critical reading of the manuscript , which has greatly improved its content and presentation . the dimming effects of light are generally termed ` extinction ' only in the simple geometrical case of a point source behind a dust screen ( section 2.1 ) , where the attenuation suffered by the source can be directly related to the optical depth of the dust screen . in this review , the extinction at wavelength @xmath24 is indicated as : @xmath195 ( equation 2 ) . the dimming of background galaxies due to the dust in a foreground galaxy can generally be considered extinction , because the foreground galaxy can be modeled in first approximation as a homogeneous screen in front of point sources ( section 3.1 ) . for more general distributions of dust and emitters , the dimming and reddening of the light is no longer directly related to the extinction or the optical depth of the dust layer , and is called indiscriminately ` effective extinction ' , ` obscuration ' , or ` attenuation ' . the effective extinction a@xmath7 is defined as : @xmath196 where m@xmath197 is the magnitude of the dust free source and m@xmath198 is the measured ( attenuated ) magnitude ; a@xmath7 measures the magnitudes of attenuation suffered by an extended source at the specified wavelength / band . by definition , a@xmath199a@xmath25 for a foreground , homogeneous , non - scattering dust screen . for dust / emitter distributions that do not have spherical symmetry , such as spiral galaxies , the face - on effective extinction is termed a@xmath200 , while a@xmath7 indicates the effective extinction averaged over all inclinations . throughout this review , two different definitions of the uv spectral slope , @xmath41 and @xmath118 have been used . in this context , the spectral slope is derived from the fit f(@xmath24)@xmath201 within a specified wavelength range , with f(@xmath24 ) the flux density of the source . the various definitions reflect historical reasons driven mostly by observational constraints . the default @xmath41 adopted in this review ( section 2.2 ) is measured between 0.125 @xmath0 m and 0.30@xmath0 m , using a set of wavelength windows that avoid the strong stellar and interstellar absorption features . an example of such windows is given in @xcite , although the ones used to derive @xmath41 here are better tuned for use with model spectra rather than with observed spectra , and extend all the way to 0.3 @xmath0 m . this definition of @xmath41 has been adopted to ensure an appropriate comparison between different extinction curves and dust models , as in figures [ fig2 ] and [ fig3 ] , because the fitting windows are selected to avoid the strong and wide dust feature at 0.2175 @xmath0 m . the uv slope @xmath118 used in section 4 is a fit through the 10 windows of @xcite , and corresponds to deriving the slope in the wavelength range 0.1250.26 @xmath0 m , using spectral regions mostly devoid of strong absorption lines . this definition of the uv slope is observation - driven , as it was used to obtain measurements from iue spectra of a local uv - selected starburst sample . the wavelength range is bluer than that of @xmath41 above , and was chosen by @xcite to avoid including in the fits the older stellar populations underlying the starbursts . these older populations can contribute to the spectra starting around 0.270.28 @xmath0 m , when the fluxes of f - stars become non - negligible . for reference , a list of @xmath118 values for dust - free stellar populations , both for instantanous burst and for constant star formation , is given in table [ tbl-6 ] . an convenient definition of the uv - slope for studies of high - redshift galaxies is @xmath202 , measured roughly between @xmath10.12 @xmath0 m and @xmath10.18@xmath0 m . at z@xmath13 , the optical window give information shortward of @xmath10.2 @xmath0 m , thus forcing the uv slope fitting to a much shorter range than possible with local galaxies . the details of the actual wavelength range vary with redshift and with the observing technique ( whether spectroscopy or imaging , and depending on the filters used ) . the relation between the three quantities , as given by models of stellar populations reddened by the starburst obscuration curve of section 4 , is : @xmath203 @xmath204 the error bar following each coefficient refers to the maximum variance observed by varying a number of parameters in the models and fits : ( a ) stellar populations from both @xcite and @xcite models ; ( b ) stellar populations with constant star formation in the range 51000 myr , and instantaneous bursts in the age range 512 myr ; ( c ) both 0.2 z@xmath37 and z@xmath37 metallicities ; ( d ) fitting windows that progressively include or exclude weak absorption lines in the spectra . the latter turns out to be the main driver of the variance , while changes in the characteristics of the stellar populations have only a small impact . the relations between the three slopes would , however , change from equations ( b1)(b2 ) if more extreme cases of stellar population mix , such as the addition of an old ( age@xmath191 gyr ) component , were considered ; in this case , @xmath41 and @xmath118 would be strongly affected , because of the long wavelength baseline . of interest for high - redshift studies is the consideration that while @xmath118 and @xmath41 follow each other closely , @xmath202 is always redder ( more positive ) than the other two , and the discrepancy increases for increasing value of the slopes . the reason for this counter - intuitive behavior is the presence of the well - known ` iron curtain ' at long uv wavelengths : this is a wide dip in the 0.230.28 @xmath0 m range of the uv spectra due to the presence of a large number of closely spaced fe absorption lines @xcite . the uv slopes fitted with the longer wavelength baseline have windows in this continuum - depressed region , hence their more negative values . increasing dust obscuration affects the short - wavelength slope more than the long - wavelength ones ; hence the trend for @xmath202 to become progressively redder than @xmath118 as their values increase ( i.e. , as the uv spectra become more dust reddened ) . @xcite derive an empirical relation between @xmath118 and a short - wavelength uv slope @xmath205 ( close , but not identical , to @xmath202 ) from iue spectra of the uv - selected starbursts . their findings , that @xmath118=@xmath2060.16 , are consistent with the long wavelength slope being bluer than the short wavelength one . however , they do not measure an increasingly redder @xmath205 for increasing values of both slopes . a possible reason is that noise in the iue data may have prevented the detection of such effect . in this review , it is assumed that equations ( b1)(b2 ) are valid for uv - selected starbursts . adelberger , k.l . , & steidel , c.c . , 2000 , , 544 , 218 adelberger , k.l . , steidel , c.c . , giavalisco , m. , dickinson , m.e . , pettini , m. , & kellogg , m. 1998 , , 505 , 18 aguirre , a. 1999 , , 525 , 583 aguirre , a. , & haman , z. 2000 , , 532 , 28 aguirre , a. , hernquist , l. , schaye , j. , katz , n. , weinberg , d.h . , & gardner , j. 2001 , , accepted ( astroph/0105065 ) allen , s.w . 1995 , , 276 , 947 aloisi , a. , tosi , m. , & greggio , l. 1999 , , 118 , 302 alonso - herrero , a. , engelbracht , c.w . , rieke , m.j . , rieke , g.h . , & quillen , a.c . 2001 , , 546 , 952 altieri , b , metcalfe , l. , kneib , j.p . , mcbreen , b. , aussel , h. et al . 1999 , , 343 , l65 alton , p.b . , trewhella , m. , davies , j.i . , evans , r. , bianchi , s. , gear , w. , thronson , h.a . , valentijn , e. , & witt , a.n . , 1998 , , 335 , 807 arimoto , n. , sofue , y. , & tsojimoto , t. 1996 , pasj , 48 , 275 arnaud , k.a . , & mushotzky , r.f . 1998 , , 501 , 119 aussel , h. , cesarsky , c.j . , elbaz , d. , & starck , j.l . 1999 , , 342 , 313 baker , a.j . , lutz , d. , genzel , r. , tacconi , l.j . , & lehnert , m.d . 2001 , , in press ( astroph/0104345 ) barger , a.j . , cowie , l.l . , mushotzky , r.f . , & richards , e.a . 2001 , , 121 , 662 barger , a.j . , cowie , l.l . , & richards , e.a . , 2000 , , 119 , 2092 barger , a.j . , cowie , l.l . , & sanders , d.b . 1999a , , 518 , l5 barger , a.j . , cowie , l.l . , sanders , d.b . , fulton , e. , taniguchi , y. , sato , y. , kawara , k. , & okuda , h. 1998 , nature , 294 , 248 barger , a.j . , cowie , l.l . , smail , i. , ivison , r.j . , blain , a.w . , & kneib , j .- 1999b , , 117 , 2656 barlow , t.a . , & tytler , d. 1998 , , 115 , 1725 baugh , c.m . , cole , s. , frenk , c.s . , & lacey , c.g . 1998 , , 498 , 504 beck , s.c . , turner , j.l . , ho , p.t . , lacy , j.h . , & kelly , d.m . 1996 , , 457 , 610 beckman , j.e . , peletier , r.f . , knapen , j.h . , corradi , r.l.m . , & gentet , l.j . 1996 , , 467 , 175 bell , e.f . 2001 , private communication bell , e.f . , gordon , k.d . , kennicutt , r.c . , & zaritsky d. 2001 , in prep . bell , e.f . , & kennicutt , r.c . 2001 , , in press ( astroph/0010340 ) berlind , a.a . , quillen , a.c . , pogge , r.w . , & sellgren , k. 1997 , , 114 , 107 bianchi , l. , clayton , g.c . , bohlin , r.c . , hutchings , j.b . , & massey , p. 1996 , , 471 , 203 bianchi , s. , ferrara , a. , davies , j.i . , & alton , p.b . 2000 , , 311 , 601 blain , a.w . , ivison , r.j . , kneib , j .- & smail , i. 1999a , in the hy - redshift universe , ed . bunker & w.j.m . van breugel , asp conf . 193 ( asp : san francisco ) , 246 blain , a.w . , jameson , a. , smail , i. , longair , m.s . , kneib , j .- p . , & ivison , r.j . 1999b , , 309 , 715 blain , a.w . , kneib , j .- p . , ivison , r.j . , & smail , i. 1999c , , 512 , l87 blain , a.w . , smail , i. , ivison , r.j . , & kneib , j .- 1999d , , 302 , 632 block , d.l . , & greenberg , j.m . 1996 , new extragalactic perspectives in the new south africa ( dordrecht : kap ) block , d.l . , witt , a.n . , grosbol , p. , stockton , a. , & moneti , a. 1994 , , 288 , 383 bogart , r. , & wagoner , r. 1973 , , 181 , 609 boiss , p. , le brun , v. , bergeron , j. , & deharveng , j .- m . 1998 , , 333 , 841 bond , j.r . , carr , b.j . , & hogan , c.j . 1991 , , 367 , 420 boselli , a. , lequeux , j. , sauvage , m. , boulade , o. , boulanger , f. , et al . 1998 , , 335 , 53 bosma , a. , byun , y. , freeman , k.c . , & athanassoula , e. 1992 , , 400 , l21 bouchet , p. , lequeux , j. , maurice , e. , prevot , l. , & prevot - burnichon , m.l . 1985 , , 149 , 330 boulanger , f. , beichman , c. , dsert , f.x . , helou , g. , prault , m. , & ryter , c. 1988 , , 332 , 328 boyle , b.j . , fong , r. , & shanks , t. 1988 , , 231 , 897 bregman , j.n . , snider , b.a . , grego , l. , & cox , c.v . 1998 , , 499 , 670 brodie , j.p . , & huchra , j.p.1991 , , 379 , 157 bruzual , g. , & charlot , s. 2001 , in preparation bruzual , g.a . , magris , g.c . , & calvet , n. 1988 , , 333 , 673 buat , v. , & burgarella , d. 1998 , , 334 , 772 buat , v. , & xu , c. 1996 , , 306 , 61 bunker , a.j . 1999 , in photometric redshifts and high redshift galaxies , conf . ser . 191 , eds . weymann , l.j . storrie - lombardi , m. sawicki , & r.j . brunner ( san francisco : asp ) , 317 burgh , e.b . 2001 , phd thesis ( the johns hopkins university , u.s.a . ) burstein , d. , haynes , m.p . , & faber , s.m . 1991 , nature , 353 , 515 byun , y.i . 1993 , , 105 , 993 byun , y.i . , freeman , k.c . , & kylafis , n.d . 1994 , , 432 , 114 calzetti , d. , 1997a , , 113 , 162 calzetti , d. 1997b , in the ultaviolet universe at low and high redshift , eds . waller , m.n . fanelli , j.e . hollis , & a.c . danks , aip conf . 408 ( aip : woodbury ) , 403 calzetti , d. , armus , l. , bohlin , r.c . , kinney , a.l . , koornneef , j. & storchi - bergmann , t. 2000 , , 533 , 682 calzetti , d. , bohlin , r.c . , gordon , k.d . , witt , a.n . , & bianchi , l. 1995a , , 446 , l97 calzetti , d. , bohlin , r.c . , kinney , a.l . , storchi - bergmann , t. , & heckman , t.m . 1995b , , 443 , 136 calzetti , d. , conselice , c.j . , gallagher , j.s . , & kinney , a.l . 1999 , , 118 , 797 calzetti , d. , & heckman , t.m . 1999 , , 519 , 27 calzetti , d. , kinney , a.l . , & storchi - bergmann , t. 1994 , apj , 429 , 582 calzetti , d. , kinney , a.l . , & storchi - bergmann , t. 1996 , , 458 , 132 calzetti , d. , meurer , g.r . , bohlin , r.c . , garnett , d.r . , kinney , a.l . , leitherer , c. , & storchi - bergmann , t. 1997 , , 114 , 1834 caplan , j. , & deharveng , l. 1986 , , 155 , 297 cardelli , j.a . , clayton , g.c . , & mathis , j.s . 1989 , , 345 , 245 carilli , c.l . , & yun , m.s . 1999 , , 513 , l13 carollo , c.m . , franx , m. , illingworth , g.d . , & forbes , d.a . 1997 , , 481 , 710 cen , r. , & bryan , g.l . 2001 , , 546 , l81 cen , r. , miralda - escud , j. , ostriker , j.p . , & rauch , m. 1994 , , 437 , l9 cen , r. , & ostriker , j.p . 1999 , , 514 , 1 cesarsky , d. , lequeux , j. , abergel , a. , perault , m. , palazzi , e. , & tran , d. 1996 , , 315 , l305 chandrasekhar , s. 1960 , radiative trnasfer , new york : dover chapman , s.c . , richards , e.a . , lewis , g.f . , wilson , g. , & barger , a.j . 2001 , , 548 , l147 chapman , s.c . , scott , d. , steidel , c.c . , borys , c. , halpern , m. , et al . 2000 , , 319 , 318 charlot , s. , & fall , s.m . 2000 , , 539 , 718 charlot , s. , & longhetti , m. 2001 , , in press ( astroph/0101097 ) chary , r. , & elbaz , d. 2001 , , in press ( astroph/0103067 ) cheng , f.h . , gaskell , c.m . , & koratkar , a.p . 1991 , , 370 , 487 chini , r. , krgel , e. , & kreysa , e. 1986 , , 167 , 315 choloniewski , j. 1991 , , 250 , 486 clements , d.l . , dsert , f .- x . , & franceschini , a. 2001 , , in press ( astroph/0103242 ) clements , d.l . , dsert , f .- x . , franceschini , a. , reach , w.t . , baker , a.c . , davies , j.k . , & cesarsky , c. 1999 , , 346 , 383 clements , d.l . , sutherland , w.j . , saunders , w. , efstathiou , g.p . , mcmahon , r.g . 1996 , , 279 , 459 corradi , r.l.m . , beckman , j.e . , & simonneau , e. 1996 , , 282 , 1005 cowie , l.l , songaila , a. , & barger , a.j . 1999 , , 118 , 603 cox , c.v . , bregman , j.n . , & schombert , j.m . 1995 , , 99 , 405 cox , p. , & kessler , m.f . 1999 , the universe as seen by iso ( noordwijk : esa publications division ) cox , p. , krgel , e. , & mezger , p.g . 1986 , , 155 , 380 crawford , c.s . , allen , s.w . , ebeling , h. , edge , a.c . , & fabian , a.c . 1999 , , 306 , 857 dale , d.a . , helou , g. , contursi , a. , silbermann , n.a . , & kolhatkar , s. 2001 , , 549 , 215 davies , j.i . , alton , p.b . , trewhella , m. , evans , r. , & bianchi , s. 1999 , , 304 , 495 davies , j.l . , & burstein , d. 1995 , the opacity of spiral disks ( dordrecht : kap ) davies , j.i . , phillips , s. , booyce , p.j . , & disney , m.j . 1993 , , 260 , 491 dsert , f .- x . , boulanger , f. , & puget , j.l . 1990 , , 237 , 215 de jong , t. , clegg , p.e . , soifer , b.t . , rowan - robinson , m. , habing , h.j . , houck , j.r . , aumann , h.h . , & raimond , e. 1984 , , 278 , l67 de vaucouleurs , g. 1959a , , 64 , 397 de vaucouleurs , g. 1959b , handb der physik , 53 , 311 devereux , n.a . , & young , j.s . 1991 , , 371 , 515 di bartolomeo , a. , barbaro , g. , & perinotto , m. 1995 , , 277 , 1279 de young , d.s . , & heckman , t.m . 1994 , , 431 , 598 dickinson , m.e . 1998 , in the hubble deep field , stsci may symposium , eds . m. livio , s.m . fall & p. madau ( cambridge : cup ) , 219 dickinson , m.e . 2000 , phil . royal soc . lond . a , 358 , 2001 disney , m. , davies , j. , & phillips , s. 1989 , , 239 , 939 dole , h. , gispert , r. , lagache , g. , puget , j .- l . , bouchet , f.r . , et al . 2001 , , in press ( astroph/0103434 ) domingue , d.l . , keel , w.c . , ryder , s.d . , white , r.e . 1999 , , 118 , 1542 donahue , m. , mack , j. , voit , g.m . , sparks , w. , elston , r. , & maloney , p.r . 2000 , , 545 , 670 draine , b.t . , & anderson , n. 1985 , , 292 , 494 draine , b.t . , & lee , h.m . 1984 , , 285 , 89 draine , b.t . , & salpeter , e.e . 1979 , , 231 , 77 duley , w.w . , jones , a.p . , & williams , d.a . 1989 , , 236 , 709 dunne , l. , eales , s. , edmunds , m. , ivison , r. , alexander , p. , & clements , d.l . , 2000 , , 315 , 115 dunne , l. , & eales , s.a . 2001 , , in press ( astroph/0106362 ) dutra , c.m . , bica , e. , clari , j.j . , piatti , a.e . , & ahumada , a.v . 2001 , , in press ( astroph/0103518 ) dwek , e. , rephaeli , y. , & mather , j.c . 1990 , , 350 , 104 eales , s.a . , lilly , s. , gear , w. , dunne , l. , bond , j.r . , hammer , f. , le fevre , o. , & crampton , d. 1999 , , 515 , 518 eales , s.a . , lilly , s. , webb , t. , dunne , l. , gear , w. , clements , d. , & yun , m. 2000 , , 120 , 2244 ebneter , k. , & balick , b. 1985 , , 90 , 183 ebneter , k. , djorgovski , s. , & davis , m. 1988 , , 95 , 422 edge , a.c . , ivison , r.j . , smail , i. , blain , a.w . , & kneib , j .- 1999 , , 306 , 599 edge , a.c . , 2001 , , in press ( astroph/0106225 ) edmunds , m.g . , & phillips , s. , 1997 , , 292 , 733 efstathiou , a. , oliver , s. , rowan - robinson , m. , surace , c. , summer , t. , et al . 2000 , , 319 , 1169 efstathiou , a. , rowan - robinson , m. , & siebenmorgen , r. 2000 , , 313 , 734 elbaz , d. 1999 , in building galaxies : from the primordial universe to the present , eds . f. hammer , thuan , t.x . , cayatte , v. , guiderdoni , b. , & j. tran thanh van ( singapore : world scientific publishing ) , 195 elbaz , d. , cesarsky , c.j . , fadda , d. , aussel , h. , dsert , f.x . , et al . 1999 , , 351 , l37 ellingson , e. , yee , h.k.c . , bechtold , j. , & elston , r. 1996 , , 466 , l71 ellison , s.l . , lewis , g.f . , pettini , m. , chaffee , f.h . , & irwin , m.j . 1999 , , 520 , 456 ellison , s.l . , songaila , a. , schaye , j. , & pettini , m. 2000 , , 120 , 1175 ellison , s.l . , yan , l. , hook , i.m . , pettini , m. , wall , j.v . , & schaver , p. 2001 , , submitted elmegreen , d.m . 1980 , , 43 , 37 emsellem , e. 1995 , , 303 , 673 fabian , a.c . , johnstone , r.m . , & daines , s.j . 1994 , , 271 , 737 fabian , a.c . , smail , i. , iwasaka , k. , allen , s.w . , blain , a.w . , crawford , c.s . , ettori , s. , ivison , r.j . , johnstone , r.m . , kneib , j .- p . , & wilman , r.j . 2000 , , 315 , l8 fanelli , m.n . , oconnell , r.w . , & thuan , t.x . 1988 , , 334 , 665 fall , s.m . , & pei , y.c . 1993 , , 402 , 479 fall , s.m . , pei , y.c . , & mcmahon , r.g . 1989 , , 341 , l5 ferguson , h.c . 1993 , , 263 , 343 ferguson , a.m.n . , gallagher , j.s . , & wyse , r.f.g . 1998 , , 116 , 673 ferrara , a. , bianchi , s. , cimatti , a. , & giovanardi , c. 1999a , , 123 , 437 ferrara , a. , nath , b. , sethi , s.k . , shchekinov , y. 1999b , , 303 , 301 ferrara , a. , pettini , m. , & shchekinov , y. 2000 , , 319 , 539 ferrara , a. , & tolstoy , e. 2000 , , 313 , 291 fitzpatrick , e.l . 1986 , , 92 , 1068 fitzpatrick , e.l . 1999 , , 111 , 63 fixsen , d.j . , cheng , e.s . , gales , j.m . , mather , j.c . , shafer , r.a . , & wright , e.l . 1996 , , 473 , 576 fixsen , d.j . , dwek , e. , mather , j.c . , bennett , c.l . , & shafer , r.a . 1998 , , 508 , 123 flores , h. , hammer , f. , dsert , f.x . , cesarsky , c.j . , thuan , t. , et al . 1999a , , 343 , 389 flores , h. , hammer , f. , thuan , t.x . , cesarsky , c.j . , dsert , f.x . , et al . 1999b , , 517 , 148 forbes , d.a . , franx , m. , & illingworth , g.d . 1995 , , 109 , 1988 frster schreiber , n.m . , genzel , r. , lutz , d. , kunze , d. , & sternberg , a. 2001 , , 552 , 544 fox , m.j . , efstathiou , a. , rowan - robinson , m. , dunlop , j.s . , scott , s. et al . 2001 , , submitted ( astroph/0107585 ) franceschini , a. , mazzei , p. , dezotti , g. , & danese , l. 1994 , , 427 , 140 friedli , d. , & benz , w. 1995 , , 301 , 649 fukugita , m. , hogan , c.j . , & peebles , p.j.e . 1998 , , 503 , 518 furton , d.g . , laiho , j.w . , & witt , a.n . 1999 , , 526 , 752 gasda , s. , & leitherer , c. 1996 , in from stars to galaxies : the impact of stellar physics on galaxy evolution , eds . c. leitherer , u. fritze - von alvensleben & j. huchra , asp conf . 98 ( asp : san francisco ) , 451 ge , j. , bechtold , j. , & kulkarni , v.p . 2001 , , 547 , l1 genzel , r. , & cesarsky , c.j . 2000 , , in press genzel , r. , lutz , d. , sturm , e. , egami , e. , kunze , d. , moorwood , a.f.m . , rigopoulou , d. , spoon , h.w.w . , sternberg , a. , tacconi - garman , l.e . , tacconi , l. , & thatte , n. 1998 , , 498 , 579 giavalisco , m. , livio , m. , bohlin , r.c . , macchetto , f.d . , & stecher , t.p . 1996a , , 112 , 369 giavalisco , m. , steidel , c.c . , adelberger , k.l . , dickinson , m.e . , pettini , m. , & kellogg , m. 1998 , , 503 , 543 giavalisco , m. , steidel , c.c . , & macchetto , f.d . , 1996b , , 470 , 189 giovanelli , r. , haynes , m.p . , salzer , j.j . , wegner , g. , da costa , l.n . , & freudling , w. 1994 , , 107 , 2036 giovanelli , r. , haynes , m.p . , salzer , j.j . , wegner , g. , da costa , l.n . , & freudling , w. 1995 , , 110 , 1059 glazebrook , k. , blake , c. , economou , f. , lilly , s. , & colless , m. 1999 , , 306 , 843 gnedin , n.y . 1998 , , 294 , 407 goad , j.w . , & roberts , m.s . 1981 , , 250 , 79 goldader , j.d . , meurer , g.r . , heckman , t.m . , seibert , m. , sanders , d. , calzetti , d. , & steidel , c.c . 2001 , in prep . gonzlez , r.a . , allen , r.j . , dirsch , b. , ferguson , h.c . , calzetti , d. , & panagia , n. 1998 , , 506 , 152 gozalez delgado , r.m . , leitherer , c. , heckman , t.m . , lowenthal , j.d . , ferguson , h.c . , & robert , c. 1998 , , 495 , 698 gordon , k.d . , calzetti , d. , & witt , a.n . 1997 , , 487 , 625 gordon , k.d . , & clayton , g.c . 1998 , , 500 , 816 gordon , k.d . , clayton , g.c . , witt , a.n . , & misselt , k.a . 2000 , , 533 , 236 gordon , k.d . , misselt , k.a . , witt , a.n . , & clayton , g.c . 2001 , , in press ( astroph/0011575 ) goudfrooij , p. , hansen , l. , jorgensen , h.e . , & norgaard - nielsen , h.u . 1994 , , 105 , 341 goudfrooij , p. , & de jong , t. 1995 , , 298,784 granato , g.l . , lacey , c.g . , silva , l. , bressan , a. , baugh , c.m . , cole , s. , & frenk , c.s . 2000 , , 542 , 710 greenberg , j.m . 1984 , sci . , 250 , no . 6 , 124 greenberg , j.m . 1989 , in interstellar dust , iau symp . 135 , eds . allamandola & a.g.g.m . tielens ( dordrecht : kap ) , 239 greenberg , j.m . , gillette , j.s . , munz caro , g.m . , mahajan , t.b . , zare , r.n . , 2000 , , 531 , l71 greggio , l. , tosi , m. , clampin , m. , de marchi , g. , leitherer , c. , nota , a. , & sirianni , m. 1998 , , 504 , 725 guhathakurta , p. , tyson , j.a . , & majewski , s.r . 1990 , , 357 , l9 guiderdoni , b. , hivon , e. , bouchet , f.r . , & maffei , b. 1998 , , 295 , 877 haas , m. 1998 , , 337 , l1 hansen , l. , jrgensen , h.e . , & nrgaard - nielsen , h.u . 1995 , , 297 , 13 hauser , m.g . , arendt , r.g . , kelsall , t. , dwek , e. , odegard , n. , et al . 1998 , , 508 , 25 hauser , m.g . , & dwek , e. 2001 , , in press . heckman , t.m . 1994 , in mass - transfer - induced activity in galaxies , ed . i. shloshman ( cambridge : cup ) , 234 heckman , t.m . , in origins , eds . woodward , j.m . shull , & h.a . thronson ( san francisco : asp conf . series ) 148 , 127 heckman , t.m . , armus , l. , & miley , g.k . 1990 , , 74 , 833 heckman , t.m . , baum , s.a . , van breugel , w.j.m . , & mccarthy , p. 1989 , , 338 , 48 heckman , t.m . , & leitherer , c. 1997 , , 114 , 69 heckman , t.m . , lehnert , m.d . , strickland , d.k . , & armus , l. 2000 , , 129 , 493 heckman , t.m . , robert , c. , leitherer , c. , garnett , d.r . , & van der rydt , f. 1998 , , 503 , 646 helou , g. 1986 , , 311 , l33 helou , g. , khan , i.r . , malek , l. , & boehmer , l. 1988 , , 68 , 151 henyey , l.g . , & greenstein , j.l . 1941 , , 93 , 70 hernquist , l. , katz , n. , weinberg , d.h . , & miralda - escud , j. 1996 , , 457 , l51 hibbard , j.e . 1997 , in star formation near and far , the 7th annual astrophysics conference in maryland , eds . holt & l.g . mundy ( woodbury : aip ) 393 , 259 holmberg , e. 1958 , medd . lunds astr . ii , no . 136 hornschemeier , a.e . , brandt , w.n . , garmire , g.p . , schneider , d.p . , broos , p.s . , townsley , l.k . , bautz , m.w . , burrows , d.n . , chartas , g. , feigelson , e.d . , griffith , r. , lumb , d. , nousek , j.a . , & sargent , w.l.w . 2000 , , 541 , 49 hopkins , a.m. , connolly , a.j . , haarsma , d.b . , & cram , l.e . 2001 , , in press ( astroph/0103253 ) hu , e.m . 1992 , , 391 , 608 hu , e.m . , cowie , l.l . , kaaret , p. , jenkins , e.b . , york , d.g . , & roesler , f.l . 1983 , , 275 , l27 hughes , d.h . , serjeant , s. , dunlop , j. , rowan - robinson , m. , blain , a. , ivison , r. , et al . 1998 , nature , 394 , 241 huizinga , e. , & van albada , t.s . 1992 , , 254 , 677 hunter , d.a . 1997 , , 109 , 937 hunter , d.a . , & gallagher , j.s . 1997 , , 475 , 65 hunter , d.a . , gallagher , j.s . , rice , w.l . , & gillett , f.c . 1989 , , 336 , 152 hunter , d.a . , gillett , f.c . , gallagher , j.s . , rice , w.l . , & low , f.j . , 1984 , , 303 , 171 hunter , d.a . , kaufman , m. , hollenbach , d.j . , rubin , r.h , malhotra , s. , dale , d.a . , brahuer , j.r . , silbermann , n.a . , helou , g. , contursi , a. & lord , s.d . 2001 , , in press ( astroph/0012354 ) hunter , d.a . , & sage , l. 1993 , , 105 , 374 hunter , d.a . , & thronson , h.a . 1996 , , 461 , 202 ivison , r.j . , smail , i. , barger , a.j . , kneib , j .- blain , a.w . , owen , f.n . , kerr , t.h . , & cowie , l.l . 2000 , , 315 , 209 ivison , r.j . , smail , i. , le borgne , j .- f . , blain , a.w . , kneib , j .- p . , bezecourt , j. , kerr , t.h . , & davies , j.k . 1998 , , 298 , 583 jansen , r.a . 2000 , phd thesis jenniskens , p. 1993 , a&a , 274 , 653 jones , h. , davies , j.i . , & trewhella , m. 1996 , , 283 , 316 jones , a.p . , tielens , a.g.g.m . , hollenbach , d.j . , & mckee , c.f . 1994 , , 433 , 797 jura , m. 1977 , nature , 266 , 702 jura , m. , kim , d.w . , knapp , g.r . , & guhathakurta , p. 1987 , , 312 , l11 kaastra , j.s . , ferrigno , c. , tamura , t. , paerels , f.b.s . , peterson , j.r . , & mittaz , j.p.d . 2001 , , 365 , l99 karachentsev , i.d . , & lipovetskii , v.a . 1969 , sov . , 12 , 909 kawara , k. , sato , y. , matsuhara , h. taniguchi , y. , okuda , h. et al . 1998 , , 336 , l9 keel , w.c . 1983 , , 88 , 1579 keel , w.c . , kennicutt , r.c . , hummel , e. , & van der hulst , j.m . 1985 , , 90 , 708 keel , w.c . , & white , r.e . 2001a , , in press ( astroph/0011370 ) keel , w.c . , & white , r.e . 2001b , in press ( astroph/0106056 ) kennicutt jr . , r.c . 1983 , , 272 , 54 kennicutt jr . 1998 , , 36 , 189 kim , s .- h . , martin , p.g . , & hendry , p.d . 1994 , , 422 , 164 kim , s.s . , morris , m. , & lee , h.m . 1999 , , 525 , 228 kim , d.c . , & sanders , d.b . 1998 , , 119 , 41 kinney , a.l . , bohlin , r.c . , calzetti , d. , panagia , n. , & wyse , r.f.g . , , 86 , 5 knapp , g.r . , guhathakurta , p. , kim , d.w . , & jura , m. , 70 , 329 koornneef , j. , & code , a.d . 1981 , , 247 , 860 kuchinski , l.e . , terndrup , d.m . , gordon , k.d . , & witt , a.n . 1998 , , 115 , 1438 kunth , d. , mas - hesse , j.m . , terlevich , e. , terlevich , r. , lequeux , j. , & fall , s.m . 1998 , , 334 , 11 kylafis , n.d . , & bahcall , j.n . 1987 , , 317 , 637 lanzetta , k.m . , chen , h .- w . , fernndez - soto , a. , pascarelle , s. , yahata , n. , & yahil , a. 1999 , in the hy - redshift universe : galaxy formation and evolution at high redshift , eds . bunker & w.j.m . van breugel ( asp : san francisco ) , 544 larson , r.b . , & tinsley , b.m . 1978 , , 219 , 46 lauer , t.r . , ajhar , e.a . , byun , y .- i . , dressler , a. , faber , s.m . , grillmair , c.,kormendy , j. , richstone , d. , & tremaine , s. 1995 , , 110 , 2622 le brun , v. , bergeron , j. , boiss , p. , & deharveng , j .- 1997 , , 321 , 733 lger , a. , & puget , j.l . 1984 , , 137 , l5 lehnert , m.d . , & heckman , t.m . 1996 , , 472 , 546 leisawitz , d. , & hauser , m.g . 1988 , , 332 , 954 leitherer , c. , & heckman , t.m . 1995 , , 96 , 9 leitherer , c. , leao , j. , heckman , t.m . , lennon , d. , pettini , m. , & robert , c. 2001 , , 550 , in press leitherer , c. , schaerer , d. , goldader , g.d . , gonzalez delgado , r.m . , robert , c. , kune , d.f . , de mello , d.f . , devost , d. , & heckman , t.m . 1999 , , 123 , 3 lequeux , j. , maurice , e. , prevot - burnichon , m.l . , prevot , l. , & rocca - volmerange , b. 1982 , , 113 , l15 lequeux , j. , maurice , e. , prevot , l. , prevot - burnichon , m .- l . , & rocca - volmerange , b. 1984 , structure and evolution of the magellanic clouds , i.a.u . symposium no . 108 , s. van den bergh & k.s . de boer eds . ( dordrecht : reidel ) , 405 li , a. , & draine , b.t . 2001 , , 554 , 778 lilly , s.j . , eales , s.a . , gear , w.k.p . , hammer , f. , le fevre , o. , crampton , d. , bond , j.r . , & dunne , l. 1999 , , 518 , 641 lilly , s.j . , lefevre , o. , crampton , d. , hammer , f. , & tresse , l. 1995 , , 455 , 50 lilly , s.j . , lefevre , o. , hammer , f. , & crampton , d. 1996 , , 460 , l1 lisenfeld , u. , isaak , k.g . , & hills , r. 2000 , , 312 , 433 loeb , a. , & haiman , z. 1997 , , 490 , 571 londale - persson , c.j . , & helou , g. 1987 , , 314 , 513 lowenthal , j.d . , koo , d.c . , guzman , r. , gallego , j. , phillips , a.c . , et al . 1997 , , 481 , 673 lu , l. , sargent , w.l.w . , barlow , t.a . , churchill , c.w . , & vogt , s.s . 1996 , , 107 , 475 lutz , d. , feuchtgruber , h. , genzel , r. , kunze , d. , rigopoulou , d. , et al . 1996 , , 315 , l269 lutz , d. , spoon , h.w.w . , rigopoulou , d. , moorwood , a.f.m . , & genzel , r. 1998 , , 505 , l103 mac low , m .- m . , & ferrara , a. 1999 , , 513 , 142 madau , p. ferguson , h.c . , dickinson , m.e . , giavalisco , m. , steidel , c.c . , & fruchter , a. 1996 , , 283 , 1388 madau , p. , ferrar , a. , & rees , m.j . 2001 , , in press maz - apellniz , j. , & walborn , n.r . 2000 , poster presented at the iau xxivth general assembly , manchester malhotra , s. 1997 , , 488 , l101 maoz , d. 1995 , , 455 , l115 martin , m.c . 1998 , , 131 , 77 marzke , r.o . , huchra , j.p . , & geller , m.j . 1994 , , 428 , 43 mas - hesse , j.m . , arnault , p. , & kunth , d. 1989 , ap&ss , 157 , 131 mathis , j.s . 1990 , araa , 28 , 37 mathis , j.s . 1996 , , 472 , 643 mathis , j.s . , mezger , p.g . , & panagia , n. 1983 , , 128 , 212 mathis , j.s . , rumpl , w. , & nordsieck , k.h . 1977 , , 217 , 425 mathis , j.s . , & whiffen , g. 1989 , , 341 , 808 matthews , l.d . , & wood , k. 2001 , , 548 , 150 mccarthy , p.j . , yan , l. , freudling , w. , teplitz , h.i . , malumuth , e.m . , et al 1999 , , 520 , 548 mckee , c.f . 1989 , in interstellar dust , iau symposium 135 , eds . allamandola & a.g.g.m . tielens ( dordracht : kap ) , 431 mcquade , k. , calzetti , d. , & kinney , a.l . 1995 , , 97 , 331 meier , d.s . , turner , j.l . , crosthwaite , l.p . , & beck , s.c . 2001 , , in press ( astroph/0011582 ) merluzzi , p. 1998 , , 338 , 807 meurer , g.r . , heckman , t.m . , & calzetti , d. 1999 , , 521 , 64 meurer , g.r . , heckman , t.m . , lehnert , m.d . , leitherer , c. , & lowenthal , j. 1997 , , 114 , 54 meyer , d.m . , & york , d.g . 1987 , , 319 , l45 meyer , d.m . , & roth , k.c . 1990 , , 363 , 57 mezger , p.g . , mathis , j.s . , & panagia , n. 1982 , , 105 , 372 mihos , j.c . , & hernquist , l. 1996 , , 464 , 641 misiriotis , a. , kylafis , n.d . , papamastorakis , j. , xilouris , e.m . 2000 , , 353 , 117 misselt , k.a . , clayton , g.c . , & gordon , k.d . 1999 , , 515 , 128 misselt , k.a . , gordon , k.d . , clayton , g.c . , & wolff , m.j . 2001 , , in press ( astroph/0011576 ) moorwood , a.f.m . , van der werf , p.p . , cuby , j.c . , & oliva , e. 2000 , , 362 , 9 moriondo , g. , giovanelli , r. , & haynes , m.p . 1998 , , 338 , 795 murphy , t.w . , soifer , b.t . , matthews , k. , armus , l. , & kiger , j.r . 2001 , , 121 , 97 mushotzky , r.f . , loewenstein , m. , arnaud , k.a . , tamura , t. , fukazawa , y. , matsushita , k. , kikuchi , k. , & hatsukade , i. 1996 , , 466 , 686 mushotzky , r.f . , & loewenstein , m. , 1997 , , 481 , l63 nandy , k. , morgan , d.h . , willis , a.j . , wilson , r. , gondhalekar , p.m. 1981 , , 196 , 955 nath , b.b . , & trentham , n. 1997 , , 291 , 505 natta , a. , & panagia , n. 1976 , , 50 , 191 natta , a. , & panagia , n. 1984 , , 287 , 228 nelson , a.e . , zaritsky , d. , & cutri , r.m . 1998 , , 115 , 2273 niklas , s. , klein , u. , & wielebinski , r. 1997 , , 322 , 19 norman , c.a , sellwood , j.a . , & hasan , h. 1996 , , 462 , 114 oliver , s.j . , goldschmidt , p. , franceschini , a. , serjeant , s.b.g . , efstathiou , a. , et al . 1997 , , 289 , 471 ortolani , s. , renzini , a. , gilmozzi , r. , marconi , g. , babuy , b. , bica , e. , & rich , m.r . 1995 , nature , 377 , 701 osterbrock , d.e . 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( mill valley : university science books ) ostriker , j.p . , & gnedin , n.y . 1996 , , 472 , l63 ouchi , m. , yamada , t. , kawai , h. , & ohta , k. 1999 , , 517 , l19 pak , s. , jaffe , d.t . , van dishoeck , e.f . , johansson , l.e.b . , & booth , r.s . , 1998 , , 498 , 735 papovich , c. , dickinson , m. , & ferguson , h.c . 2001 , , in press peacock , j.a . , rowan - robinson , m. , blain , a.w . , dunlop , j.s . , efstathiou , a. , et al . 2000 , , 318 , 535 pei , y.c . , fall , s.m . , & bechtold , j. 1991 , , 378 , 6 pei , y.c , fall , s.m . , & hauser , m.g . 1999 , , 522 , 604 peletier , r.f . , valentijn , e.a . , moorwood , a.f.m . , freudling , w. , knapen , j.h . , & beckman , j.e . 1995 , , 300 , l1 peletier , r.f . , & willner , s.p . 1992 , , 103 , 1761 penton , s.v . , shull , j.m . , & stocke , j.t . 2000 , , 544 , 150 peterson , j.r . , paerels , f.b.s . , kaastra , j.s . , arnaud , m. , reiprich , t.h . , fabian , a.c . , mushotzky , r.f . , jernigan , j.g . , & sakelliou , i. 2001 , , 365 , l104 petitjean , p. , muecket , j.p . , & kates , r.e . 1995 , , 295 , l9 pettini , m. 1999 , in chemical evolution from zero to high redshift , ed . j. walsh & m. rosa ( berlin : springer ) , 233 pettini , m. , boksenberg , a. , & hunstead , r.w . , 1990 , , 348 , 48 pettini , m. , ellison , s.l . , steidel , c.c . , shapley , a.e . , & bowen , d.v . 2000a , , 532 , 65 pettini , m. , ellison , s.l . , steidel , c.c . , & bowen , d.v . 1999 , , 510 , 576 pettini , m. , kellogg , m. , steidel , c.c . , dickinson , m.e . , adelberger , k.l . , & giavalisco , m. 1998 , , 508 , 539 pettini , m. , king , d.l . , smith , l.j . , & hunstead , r.w . 1997a , , 478 , 536 pettini , m. , shapley , a.e . , steidel , c.c . , cuby , j .- g . , dickinson , m. , moorwood , a.f.m . , adelberger , k.l . , & giavalisco , m. 2001 , , 554 , 981 pettini , m. , smith , l.j . , hunstead , r.w . , & king , d.l . 1994 , , 426 , 79 pettini , m. , smith , l.j . , king , d.l . , & hunstead , r.w . , & king , d.l . 1997b , , 486 , 665 pettini , m. , steidel , c.c . , adelberger , k.l . , dickinson , m.e . , & giavalisco , m. 2000b , , 528 , 96 pinkney , j. , holtzman , j. , garasi , c. , et al . 1996 , , 468 , l13 pizagno , j. , & rix , h .- w . 1998 , , 116 , 2191 poggianti , b.m . , & wu , h. 2000 , , 529 , 157 popescu , c.c . , tuffs , r.j . , fischera , j. , & vlk , h. 2000 , , 354 , 480 prochaska , j.x . , & wolfe , a.m. 2000 , , 533 , l5 puget , j .- l . , abergel , a. , bernard , j .- p . , boulanger , f. , burton , w.b . , desert , f .- x . , & hartmann , d. 1996 , , 308 , l5 puget , j.l . , lagache , g. , clements , d.l , reach , w.t . , aussel , h. , et al . 1999 , , 345 , 29 quillen , a.c . , rieke , g.h . , rieke , m.j . , caldwell , n. , & engelbracht , c. 1999 , , 518 , 632 radovich , m. , kahanp , j. , & lemke , d. 2001 , , in press ( astroph/0108075 ) rao , s.m . , & turnshek , d.a . 2000 , , 130 , 1 rauch , m. , miralda - escud , j. , sargent , w.l.w . , barlow , t.a . , weinberg , d.h . , hernquist , l. , katz , n. , cen , r. , & ostriker , j.p . 1997 , , 489 , 7 reach , w.t . , boulanger , f. , contursi , a. , & lequeux , j. , 2000 , , 361 , 895 renzini , a. 1999 , in chemical evolution from zero to high redshift , ed . j. walsh & m. rosa ( berlin : springer ) , 185 rieke , g.h . , & lebofsky , m.j . 1985 , , 288 , 618 rieke , g.h . , & lebofsky , m.j . 1986 , , 304 , 326 rieke , g.h . , lebofsky , m.j . , thompson , r. i. , low , f.j . , & tokunaga , a.t . 1980 , , 238 , 24 rieke , g.h . , loken , k. , rieke , m.j . , & tamblyn , p. 1993 , , 412 , 99 riess , a. , filippenko , a.v . , liu , m.c . , et al . 2000 , , 536 , 62 riess , a. , nugent , p. , smith , b. , tonry , j. , gilliland , r. , et al . 2001 , , in press rigopoulou , d. , franceschini , a. , aussel , h. , genzel , r. , van der werf , p. , et al . 2000 , , 537 , l85 rix , h .- w . , & rieke , m.j . 1993 , , 418 , 123 roberts , m.s . , & haynes , m.p . 1994 , , 32 , 115 roberts , m.s . , hogg , d.e . , bregman , j.n . , forman , w.r . , & jones , c. 1991 , , 75 , 751 romani , r.w . , & maoz , d. 1992 , , 386 , 36 rowan - robinson , m. , & crawford , j. 1989 , , 238 , 523 rowan - robinson , m. , & efstathiou , a. 1993 , , 263 , 675 rowan - robinson , m. , mann , r.g . , oliver , s.g . , efstathiou , a. , eaton , n. et al . 1997 , , 289 , 490 rowan - robinson , m. , oliver , s. , efstathiou , a. , gruppioni , c. , serjeant , s. , et al . 1999 , in the universe as seen by iso , eds . p. cox & m.f . kessler ( noordwijk : esa publications division ) 1011 rowan - robinson , m. , negroponte , j. , & silk , j. 1979 , nature , 281 , 635 sadler , e.m . , & gerhard , o.e . 1985 , , 214 , 177 sanders , d.b . , & mirabel , i.f . 1996 , , 34 , 749 sanders , d.b . , soifer , b.t . , elias , j.h , madore , b.f . , matthews , k. , neugebauer , g. , & scoville , n.z . 1988 , , 325 , 74 savaglio , s. , panagia , n. , & stiavelli , m. 2000 , in cosmic evolution and galaxy formation : structure , interactions , and feedback , asp conf . 215 , eds . j. franco , e. terlevich , o. lpez cruz , & i. aretxaga ( asp : chelsea ) , 65 sawicki , m. 2001 , , in press ( astroph/0102208 ) sawicki , m. , & yee , h.k.c . 1998 , , 115 , 1329 schutte , w.a . , van der hucht , k.a . , whittet , d.c.b . , boogert , a.c.a . , tielens , a.g.g.m . , 1998 , , 337 , 261 scott , d. , lagache , g. , borys , c. , chapman , s.c . , halpern , m. , et al . 2000 , , 357 , l5 scott , s.e . , fox , m.j . , dunlop , j.s . , serjeant , s. , peacock , j.a . , et al . 2001 , , submitted ( astroph/0107446 ) scoville , n.z . , evans , a.s . , dinshaw , n. , thompson , r. , rieke , m.,schneider , g. , low , f.j . , hines , d. , stobie , e. , becklin , e. , & epps , h. 1998 , , 492 , l107 scowen , p.a . , hester , j.j . , sankrit , r. , gallagher , j.s . , ballester , g.e . , et al . 1998 , , 116 , 163 sellgren , k. 1984 , , 277 , 623 sellgren , k. 1994 , the first symposium on the infrared cirrus and diffuse insterstellar clouds , cutri & latter , san francisco : asp , 58 , 243 sellgren , k. 2000 , spectrochimica acta a , feb . 2001 , in press ( astroph/0010138 ) serjeant , s. , oliver , s. , rowan - robinson , m. , crockett , h. , missoulis , v. , et al . 2000 , , 316 , 768 severgnini , p. , maiolino , r. , salvati , m. , axon , d. , cimatti , a. , fiore , f. , gilli , r. , la franca , f. , marconi , a. , matt , g. , risaliti , g. , & vignali , c. 2000 , , 360 , 457 shapley , a.e . , steidel , c.c . , adelberger , k.l . , dickinson , m. , giavalisco , m. , & pettini , m. 2001 , in prep . shaver , p.a . , hook , i.m . , jackson , c.a . , wall , j.v . , & kellermann , k.i . 1999 , in highly redshifted radio lines , asp conf . carilli , s.j.e . radford , k.m . menten & g.i . langston ( asp : san francisco ) , 163 shioya , y. , trentham , n. , & taniguchi , y. 2001 , , in press ( atroph/0011467 ) silva , l. , granato , g.l . , bressan , a. , & danese , l. 1998 , , 509 , 103 simonsen , j.t . , & hannestad , s. 1999 , , 351 , 1 skillman , e.d . , kennicutt , r .. , & hodge , p.w . 1989 , , 347 , 875 smail , i. , ivison , r.j . , & blain , a.w . 1997 , , 490 , l5 smail , i. , ivison , r.j . , blain , a.w . , & kneib , j .- p . 1998 , , 507 , l21 smail , i. , ivison , r.j . , kneib , j .- blain , a.w . , barger , a.j . , owen , f.n . , & morrison , g. 1999 , , 308 , 1061 smail , i. , ivison , r.j . , owen , f.n . , blain , a.w . , & kneib , j .- 2000 , , 528 , 612 sobolev , v.v . 1963 , a treatise on radiative transfer , princeton : van nostrand sodroski , t.j . , bennett , c. , boggess , n. , dwek , e. , franz , b.a . , hauser , m.g . , kelsall , t. , moseley , s.h . , odegard , n. , silverberg , r.f . , & weiland , j.l . 1994 , , 428 , 638 sofue , y. , reuter , h .- p . , krause , m. , wielebinski , r. , & nakai , n. 1992 , , 395 , 126 soifer , b.t . , houck , j.r . , & neugebauer , g. 1987 , , 25 , 187 soifer , b.t . , & neugebauer , g. 1991 , , 101 , 354 songaila , a. & cowie , l.l . 1996 , , 112 , 335 sparks , w.b . , ford , h.c . , & kinney , a.l . 1993 , , 413 , 531 sparks , w.b . , macchetto , f. , & golombek , d. 1989 , , 345 , 153 sprayberry , d. , & foltz , c.b . 1992 , , 390 , 39 steidel , c.c . , adelberger , k.l . , giavalisco , m. , dickinson , m.e . , & pettini , m. 1999 , , 519 , 1 steidel , c.c . , giavalisco , m. , dickinson , m.e . , & adelberger , k.l . 1996a , , 112 , 352 steidel , c.c . , giavalisco , m. , pettini , m. , dickinson , m. , & adelberger , k. 1996b , , 462 , l17 steidel , c.c . , & hamilton , d. 1992 , , 104 , 941 stickel , m. , lemke , d. , mattila , k. , haikala , l.k . , & haas , m. 1998 , , 329 , 55 storchi - bergmann , t. , kinney , a.l . , & challis , p. 1995 , , 98 , 103 sullivan , m. , treyer , m.a . , ellis , r.s . , bridges , t.j . , milliard , b. , & donas , j. 2000 , , 312 , 442 sullivan , m. , mobasher , b. , chan , b. , cram , l. , ellis , r. , treyer , m. , & hopkins , a. 2001 , , in press ( astroph/0104425 ) szalay , a.s . , hollsi , j. , & tth , g. 1989 , , 339 , l5 tamura , t. , kaastra , j.s . , peterson , j.r . , paerels , f.b.s . , mittaz , j.p.d . , trudolyubov , s.p . , stewart , g. , fabian , a.c . , mushotzky , r.f . , lumb , d.h . , & ikebe , y. 2001 , , 365 , l87 taniguchi , y. , cowie , l.l . , sato , y. , sanders , d.b . , kawara , k. et al . 1998 , , 328 , l9 teplitz , h.i . , malkan , m.a . , steidel , c.c . , mclean , i.s . , becklin , e.e . , et al . 2000a , , 542 , 18 teplitz , h.i . , mclean , i.s . , becklin , e.e . , figer , d.f . , gilbert , a.m. , et al . 2000b , , 533 , l65 theuns , t. , leonard , a. , efstathiou , g. , pearce , f.r . , & thomas , p.a . 1998 , , 301 , 478 tielens , a.g.g.m . 1989 , in interstellar dust , iau symp . 135 , eds . allamandola & a.g.g.m . tielens ( dordrecht : kap ) , 239 tran , h. , tsvetanov , z. , ford , h.c . , davies , j. , jaffe , w. , van den bosch , f.c . , & rest , a. 2001 , , accepted ( astroph/0102292 ) tremonti , c.a . , calzetti , d. , leitherer , c. , & heckman , t.m . 2001 , , in press ( astroph/0103432 ) trentham , n. , kormendy , j.k . , & sanders , d.b . 1999 , , 117 , 2152 trewhella , m. 1998 , , 297 , 807 trewhella , m. , madore , b. , & kuchinski , l. 1999 , observational cosmology : the development of galaxy systems , giuricin , mezzetti & salucci , san francisco : asp , 176 , 454 trewhella , m. , davies , j.i . , alton , p.b . , bianchi , s. , & madore , b.f . 2000 , , 543 , 153 trewhella , m. , davies , j.i . , disney , m.j . , & jones , h.g.w . 1997 , , 288 , 397 turnshek , d.a . , rao , s.m . , & nestor , d.b . 2001 , in extragalactic gas at low redshift , asp conf . , j. mulchaey and j. stocke eds . ( asp : ... ) , in press ( astroph/0108142 ) tytler , d. , fan , x.m . , burles , s. , cottrell , l. , davis , c. , kirkman , d. , & zuo , l. 1995 , in qso absorption lines , ed . g. meylan ( garching : eso ) , 289 valentijn , e.a . 1990 , nature , 346 , 153 van den bergh , s. , 1974 , , 193 , 63 van der werf , p.p . , kraiberg knudsen , k. , labb , i. , & franx , marijn 2001 , in the far - infrared and submillimeterspectral energy distributions of active and starburst galaxies , eds . p. barthel , b. wilkes , & i. van bemmel ( elsevier : the netherlands ) , ... van dokkum , p.g . & franx , m. 1995 , , 110 , 2027 vrosi , f. , & dwek , e. 1999 , , 523 , 265 veilleux , s. , kim , d .- , & sanders , d.b . 1999 , , 522 , 113 verter , f. , & hodge , p. 1995 , , 446 , 616 vladilo , g. 1998 , , 493 , 583 voit , g.m . , & donahue , m. 1992 , , 452 , 164 voit , g.m . , & donahue , m. 1997 , , 486 , 242 voshchinnikov , n.v . & mathis , j.s . 1999 , , 526 , 257 wainscoat , r.j , hyland , a.r . , & freeman , k.c . 1990 , , 348 , 85 walborn , n.r . 1991 , in massive stars in starbursts , eds . c. leitherer , n.r . walborn , t.m . heckman , & c.a . norman ( cambridge : cup ) , 145 walborn , n.r . , barb , r.h . , brandner , w. , rubio , m. , grebel , e.k . , & probst , r.g . 1999 , , 117 , 225 wang , b. 1991 , , 383 , l37 wang , b. , & heckman , t.m . 1996 , , 457 , 645 wang , j. , heckman , t.m . , & lehnert , m.d . 1998 , , 515 , 97 weingartner , j.c . , & draine , b.t . 2001 , , 548 , 296 white , s.d.m . , & frenk , c.s . 1991 , , 379 , 52 white , r.e . , & keel , w.c . 1992 , nature , 359 , 129 white , r.e . , keel , w.c . , & conselice , c.j . 2000 , , 542 , 761 whittle , m. 1992 , , 79 , 49 wilson , c.d . , walker , c.e . , & thornely , m.d . 1997 , , 483 , 210 wise , m.w . , oconnell , r.w . , bregman , j.n . , & roberts , m.s . 1993 , , 405 , 94 wise , m.w . , & silva , d.r . 1996 , , 461 , 155 witt , a.n . , friedmann , b.c . , & sasseen , t.p . 1997 , , 481 , 809 witt , a.n , & gordon , k.d . 1996 , , 463 , 681 witt , a.n . , & gordon , k.d . 2000 , , 528 , 799 witt , a.n . , thronson , h.a . , & capuano , j.m . 1992 , , 393 , 611 wolff , m.j . , clayton , g.c . , & gibson , s.j . 1998 , , 503 , 815 wright , e.l . 1981 , , 250 , 1 xilouris , e.m . , kylafis , n.d . , papamastorakis , j. , paleologou , e.v . , & haerendel , g. 1997 , , 325 , 135 xilouris , e.m . , byun , y.i . , kylafis , n.d . , paleologou , e.v . , & papamastorakis , j. , 1999 , , 344 , 868 xu , c. 2000 , , 541 , 134 xu , c. , & buat , v. 1995 , , 293 , l65 xu , c. , & helou , g. 1996 , , 456 , 163 yadat , r.k.s . , & sagar , r. 2001 , , in press yamamoto , t.m . , & vansevicius , v. 1999 , pasj , 51 , 405 yan , l. , mccarthy , p.j . , freudling , w. , teplitz , h.j . , malumuth , e.m . , weymann , r.j . , & malkan , m.a . 1999 , , 519 , l47 young , j.s . , & scoville , n.z . 1991 , , 29 , 581 yu , g. & solomon , p.m. 1999 , , 512 , l99 yun , m.s . , reddy , n.a . , & condon , j.j . 2001 , , accepted ( astroph/0102154 ) zaritsky , d. 1994 , , 108 , 1619 zaritsky , d. 1999 , , 118 , 2824 zaritsky , d. , kennicutt , r.c . , & huchra , j.p . 1994 , , 420 , 87 zhang , y. , anninos , p. , & norman , m.l . 1995 , , 453 , l57 zuo , l. , beaver , e.a . , burbidge , e.m . , cohen , r.d . , junkkarinen , v.t . , & lyons , r.w . 1997 , , 477 , 568 zwicky , f. , 1957 , in morphological astronomy ( berlin : springer - verlag ) , 101 llllll spherical & homog . & central/ & & & + & & spherical & @xcite&mw & multiple density profiles , + & & & & & multiple scale - lengths + & & & @xcite & & starburst / ir modelling , + & & & & & central source only + & & & @xcite & & multiple scale - lengths + & & & @xcite & & also external source + & & & @xcite&mw , smc & multiple dust / star density + & & & & & profiles + & & & @xcite & & starburst / ir modelling , + & & & & & multiple spheres , central + & & & & & source only + & & & @xcite&mw , smc&updated albedo / phase + & & & & & function values + & & & & & + spherical & clumpy & central/ & & & + & & spherical & @xcite & mw & central source only + & & & @xcite & & also external source + & & & @xcite&mw , smc & updated albedo / phase + & & & & & function values + & & & & & + plane - paral.&homog.&plane - paral.&@xcite&mw , lmc & multiple scale - heights + & & & @xcite & & no scattering + & & & @xcite&mw & multiple scale - heights + & & & @xcite & & multiple scale - heights + & & & & & + expon . & homog . & expon . & @xcite&mw & edge - on only , isothermal + & & & & & vertical distribution + & & & @xcite & & no scattering + & & & @xcite&mw & stellar bulge , multiple + & & & & & scale lengths / heights + & & & @xcite&mw & face - on only + & & & @xcite&mw & stellar bulge + & & & @xcite&mw , smc & stellar bulge , multiple + & & & & & scale lengths / heights + & & & & & + expon . & clumpy & expon . & @xcite&mw & stellar bulge + & & & @xcite&mw & stellar bulge , ir + & & & & & modelling + & & & @xcite&mw & clumping of stars as well + & & & & & + arbitrary&homog./ & arbitrary & & & + & clumpy & & @xcite & & multi - phase clumpy medium + & & & & & and emitters + & & & @xcite&all&multi - phase clumpy medium + & & & & & and emitters , polarization , + & & & & & ir modelling + lrrr line & @xmath127,@xmath128 & l@xmath207/l@xmath208 & k(@xmath128)@xmath26k(@xmath127 ) + ratio & ( @xmath0m,@xmath0 m ) & & + h@xmath42 , h@xmath41 & 0.6563,0.4861 & 2.87 & 1.163 + h@xmath42 , h@xmath43 & 0.6563,0.4340 & 6.16 & 1.628 + br@xmath43 , h@xmath41 & 2.166,0.4861 & 0.0302 & 3.352 + br@xmath43 , pa@xmath41 & 2.166,1.282 & 0.175 & 0.457 + pa@xmath42 , h@xmath42 & 1.876,0.6563 & 0.123 & 2.104 + lccccccccc e / s0&0.050.15 & 0.100.20 & & 0.050.10 & & 0.020.04 & & @xmath180.02 & + sa sab&&0.400.65 & 0.901.25&0.200.40&0.500.75 & 0.100.15 & 0.300.40&@xmath570.05 & @xmath570.15 + sb scd&0.450.65 & 0.600.80 & 1.201.45 & 0.300.50&0.650.95 & 0.150.20&0.400.45 & 0.050.10 & 0.150.20 + irr & 0.250.40 & 0.300.45 & 0.600.75 & 0.100.15 & 0.300.40 & @xmath10.05 & @xmath10.15 & @xmath570.02 & @xmath570.06 + lcrrrrrrrr name & d & morphology & l@xmath209 & sfr & l@xmath79/l@xmath209 & a@xmath210 & a@xmath74 & a@xmath211 & a@xmath170 + & ( mpc ) & & ( l@xmath37 ) & ( m@xmath37 yr@xmath5 ) & & ( mag ) & ( mag ) & ( mag ) & ( mag ) + + ngc1614 & 74 . & sbc , merger . & 4.8e+11 & 54.9 & 0.91 & 5.67 & 2.78 & 1.20 & 3.08 + ngc4194 & 40 . & ibm , merger & 1.1e+11 & 9.9 & 0.86 & 3.20 & 1.57 & 0.68 & 1.74 + ngc6090 & 135 . & sd , merger & 4.1e+11 & 52.4 & 0.88 & 3.41 & 1.67 & 0.72 & 1.85 + + ngc1140 & 23 . & ibm & 9.1e+09 & 1.1 & 0.59 & 0.83 & 0.40 & 0.17 & 0.45 + ngc1510 & 12 . & sa0 & 8.1e+08 & 0.06 & 0.42 & 0.52 & 0.25 & 0.11 & 0.28 + tol1924@xmath26416 & 44 . & irr & 1.9e+10 & 2.8 & 0.42 & 0.52 & 0.25 & 0.11 & 0.28 + + & & & & & & 1.60 & 0.78 & 0.34 & 0.86 + rlrrrr & parameter & smmj14011@xmath760252 & ms 1512@xmath26cb58 & west mmd11 & l@xmath64 lbg + 1 & z & 2.565 & 2.729 & 2.982 & @xmath13 + 2 & ab@xmath212 & 21.25 & 20.41 & 24.05 & @xmath124.5 + 3 & g@xmath181 & 0.69 & 0.72 & 0.80 & 0.6 + 4 & @xmath118 & @xmath260.74 & @xmath261.34 & @xmath260.90 & @xmath261.5 + 5 & a@xmath213 & 3.1 & 1.8 & 2.7 & 1.5 + 6 & e(b@xmath26v)@xmath137(uv ) & 0.31 & 0.18 & 0.27 & 0.15 + 7 & e(b@xmath26v)@xmath137(h@xmath41 ) & & 0.1@xmath1450.1 & 0@xmath1450.2 & + 8 & f@xmath214(ir ) & 3.7e@xmath2613 & 2.3e@xmath2613 & 1.9e@xmath2614 & 3.2e@xmath2615 + 9 & f@xmath215(450 ) & 42@xmath1457 & & 22@xmath14523 & @xmath1810 + 10 & f@xmath214(450 ) & 25.9 , 71.6 & 19.5 , 51.9 & 2.1 , 5.3 & 0.36 , 0.89 + 11 & f@xmath215(850 ) & 15@xmath1452 & @xmath184 & 5.5@xmath1451.4 & @xmath181 + 12 & f@xmath214(850 ) & 4.0 , 19.0 & 3.2 , 14.6 & 0.4 , 1.6 & 0.06 , 0.27 + lrr age ( myr ) & @xmath216 & @xmath217 + 2 & @xmath262.70 & @xmath262.68 + 5 & @xmath262.50 & @xmath262.61 + 10 & @xmath262.45 & @xmath262.57 + 15 & @xmath262.30 & @xmath262.55 + 30 & @xmath262.09 & @xmath262.50 + 50 & @xmath261.93 & @xmath262.45 + 100 & @xmath261.61&@xmath262.38 + 300 & @xmath260.20&@xmath262.28 + 500 & & @xmath262.26 + 900 & & @xmath262.23 + Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: there is a general lack of information about the nature of uv sources fainter than the completeness limit of the td-1 photometric catalog ( thompson 1978 ) . this is , primarily , the result of a lack of surveys of the uv sky . only partial surveys , in limited regions of the sky , have been conducted since the td-1 mission . one of the more extended , in terms of general sky coverage , was by the faust experiment . faust is a @xmath1 instantaneous field - of - view ( fov ) telescope , covering the spectral range between 1400 and 1800 with an angular resolution of 3.5 arcminutes . it has a 20 cm entrance aperture and it utilizes a microchannel plate detector with wedge and strip anode , which records the position of each detected photon . faust operated on board the space shuttle ( sts ) atlantis in march 1992 as part of the atlas1 mission . a description of faust and its operation aboard atlas1 was given by lampton ( 1993 ) . during this flight 19 pointed exposures were obtained , of which three were short exposures for pointing checks . the other fixed - pointing images were exposed for 12 to 18 minutes . in addition , two @xmath030@xmath2 long scans of the sky were obtained by rolling the sts during the exposure . details of the image reconstruction from the time - tagged photon stream and subsequent reductions are given in bowyer ( 1993 ) . a catalog of 4660 faust sources , originating from the fixed - pointings as well as from the sky scans , was produced by bowyer ( 1995 ; fsc ) . this used a uniform thresholding algorithm for source detection , and the identifications were obtained from correlations against catalogs ( mainly from the simbad data base ) . the faust images are being studied systematically at tel aviv in order to identify sources and perform , as much as possible , ground - based follow - up studies . in this context , we already presented results from five images at high galactic latitude sky regions visible from the northern hemisphere : the north galactic pole ( brosch 1995 ) , the virgo cluster region ( brosch 1997 ) , and in the direction of the coma cluster ( brosch 1998 ) . this paper presents our analysis of four faust fields in the southern sky . it uses , for the first time , data from an objective - prism survey ( the hamburg - eso survey = hes , wisotzki 1996 ; reimers 1997 ) to provide additional ground - based spectroscopic information about the possible optical counterparts . the uv sources in this part of the milky way ( mw ) may help the understanding of the morphology of the disk and halo of the mw . this is because stellar uv sources represent special stellar populations ; the majority are rather massive main - sequence objects with spectral types a & f , as our previous studies have shown . their study allows us to test the mw structure in four spatial windows . in this region there are indications of bends , corrugations , valleys , and other morphological peculiarities of the galactic disk ( alfaro & efremov 1996 ) . the paper is arranged as follows . we first describe the faust observations and the source detection on the uv images . we then describe the process of source identification based on correlations with published data , mainly from electronic data banks . we describe our use of the he s objective - prism plates to confirm or modify our proposed identification . the discussion concentrates on ( a ) a comparison of the fields which have he s data with those fields lacking this information , and ( b ) the general properties of the uv sources we detected , in the context of the fourth galactic quadrant in which they are located . one field is located at the border between the third and the fourth quadrant ; it is analyzed here as though it is one of the 4th quadrant fields . the regions observed by faust , and in which the source identification process has been performed , are listed in table 1 . the faust images are shown in figures 1 to 4 , for the dorado , centaurus , m83 , and telescopium ( hereafter dor , cen , m83 , and tel ) regions . all four regions are located in , or very close to , the fourth galactic quadrant , the part of the mw galaxy which ( as seen from the solar neighbourhood ) lies between the galactic center ( gc ) and l@xmath3270@xmath2 . they are all located at intermediate galactic latitudes , two north of the mw plane and two south of it . the lines of sight sample , therefore , the disk and halo of the mw at similar `` elevation '' angles from the galactic equator and at approximately the same aspect angles with respect to the galactic bulge . therefore , apart from the general interest in source detection and identification , the present observations allow , in principle , an inter - comparison of regions which sample approximately the same mw environments . each faust image has location - specific exposure levels within the image . the reason is that the space shuttle attitude control system permits some platform motion , and the information was collected as a time - tagged photon stream . the images were reconstructed on the ground by tracking relatively bright stars . this implies that during a pointed exposure a larger area of the sky was imaged than the instantaneous fov of faust . table 1 lists , for each region , the actual sky area covered by each image and the lowest and highest exposure levels , generally collected near the edge and in the central part of the image , respectively . the sky area covered by each image was calculated from the total number of non - zero exposure pixels , and by assuming that each pixel is a square of 1.1 arcmin sides . this is an approximation , because electrostatic distortions of the image caused by the detector causes pixels to have slightly different areas . an impartial source detector algorithm , described in brosch ( 1995 ) , operated on the faust images and produced lists of uv source candidates with their faust uv fluxes and equatorial coordinates . because of the physical nature of the faust frames , the exposure times , and therefore the image depths , differ for different regions in each image . variations range from a few hundred seconds in the central regions to a few tens of seconds at the frame periphery . this results in regions with reasonably high signal , whereas other regions ( with low exposure ) have low signal levels . the detector used in faust has as only sources of noise the poisson statistics of photon arrival and detection , and the sky background . as a result , the detection method must be signal - to - noise limited rather than flux - limited . the source detector algorithm used here required an acceptance threshold ( at ) level of 5.5 times the standard deviation of the local background ( @xmath4 ) to decide that a source was probably real . this was determined by examining the behavior of the number of source candidates with threshold level ; the number of possible sources increased exponentially with at for thresholds lower than 5 , but only linearly for thresholds higher than this value . we infer that , if the at is too low , large numbers of `` false '' sources are introduced by the detection method . by adopting a high threshold level one is in danger of losing valuable information . we selected the specific 5.5@xmath5 level used here at the inflexion point in the plot of the number of detected sources ( n@xmath6 ) _ vs. _ at , where the linear regime and the exponential regime cross over . the total number of sources detected by this method in each field is also listed in table 1 . the astrometry of the sources positions is based on a best - fit to the @xmath040 brightest uv sources in each field . these were identified in predicted uv images , which were created with data from the sao catalog and the hipparcos input catalog ( hic ) , together with the algorithm for predicting the uv magnitude using the visual magnitude and spectral type of stars ( brosch 1991 , and further modifications ) . the fits were usually good to @xmath030 " and were checked using the derived coordinates of the remaining @xmath020 visually identified sources . the photon flux from each detected source was measured by integrating all count rates within a simulated round aperture centered on the location of the object , and subtracting the local background . a round aperture of diameter 12 pixels , equivalent to 13.2 arcmin , was used for the majority of sources except for the very strong sources where a diameter of 24 pixels ( 26.4 arcmin ) was used . this aperture is much larger than the resolution of faust and it allows the collections of virtually all the uv photons from a source . the photometric error is composed of the aperture photometry error , that originating from the background estimation , and a systematic error in flux resulting from the laboratory calibration of faust and the intrinsic errors of the iue photometric scale ( 15.8% of the observed flux ) . monochromatic faust magnitudes ( [ uv]=2.5 log(flux)21.175 ) are used here to represent the detected brightness , where the flux is in erg/@xmath7/sec / and was derived from the count rate of faust , as explained in brosch ( 1995 ) . the photometric error translates to a minimum error of 0.16 mag in the uv magnitude used here . the lists of uv sources , separate for each faust field , are presented in table 2 . each source is identified by a running number and two leading letters , the second that identifies the faust field in which this source was identified . for instance , source fd1 is the first detected uv source in the dorado ( dor ) field , and fm83 in the 83rd source in the m83 field . we give the j2000 coordinates of the sources in columns 2 and 3 , the faust magnitude and error in column 4 , and the corresponding uv source number from the faust source catalog ( fsc : bowyer 1995 ) , for the cases where the fsc lists a source in column 5 . it is important to emphasize that the source detection algorithm used here is different from the one used in creating the fsc ( bowyer 1995 ) . because of this , a comparison of the number of detected sources listed here and in the fsc is in order . we searched the fsc for all entries within 3 of each of the sources detected here . the corresponding fsc number is listed in table 2 for all entries where a match was found . in a few cases more than one match was found . this was usually due to the proximity of another source and the `` ownership '' was resolved by visually inspecting the uv image . for three sources , all in the m83 field , bowyer ( 1995 ) detected two components , while we detected only one . the vast majority of the fsc detections have matches in our lists . in the three other fields ( dor , cen , and tel ) most fsc sources undetected by us are all very faint and remained unidentified in the fsc . they could be artefacts , but there is one exception : fsc4623 = sao193381 in the tel field , apparently a bright uv source , whose presence we could not confirm with our detection technique . a comparison for the m83 field is problematic due to the overlap of one of the faust scans with the pointed m83 exposure . nevertheless , the number of fsc entries for the m83 region not detected here is small ; ten such entries were found within 3@xmath2.4 of the m83 image center . this number does not include the three cases where the fsc showed two adjacent entries for only one source detected here . in these cases we included both fsc numbers in the lists , but caution that this may be due to a single object having slightly different fsc astrometric solutions , from the scan and from the fixed pointing . since almost all of the ten other `` undetected here '' m83 fsc entries have either nearby detected sources or are very close the edge of the image , we conclude that they originate from the faust scan , not from the pointed observation . we estimate the number of fsc sources we did not detect and which originate from the m83 fixed exposure to be at most five . table 3 summarizes this comparison of results and emphasizes the significant addition of uv sources to those in the fsc by using the present detection algorithm . on average , fsc sources make up only 2/3 of the uv sources detected by us . therefore , with our detection scheme we increased by @xmath050% the number of faust uv sources in comparison with the fsc . figure 5 shows histograms of the uv magnitude distribution in each of the four fields . in general , all fields show a steady increase in the number of detected sources with increasing apparent magnitude , but there are some subtle differences . the cen field , which is at the lowest galactic latitude , has many more sources than the other fields . on the other hand , the source distribution with uv magnitude in the cen field starts to become incomplete already at m@xmath811 , approximately 1 - 2 magnitudes brighter than in the three other fields . one important feature is noticeable by even a cursory inspection of table 1 , which gives the total number of detections along with the location of each image and its depth . the m83 image has almost the same number of detections as has the tel field , yet the two images differ by a factor of @xmath9 in exposure depth , with the tel field being shallower . the discrepancy can not be due to a higher stellar concentration in tel , because there are no ( listed ) star clusters in this direction and the galactic latitudes of the two fields are almost identical . this will be discussed further , below . the primary identification used on - line data bases , such as the sao catalog ( sao 1966 ) , the hipparcos input catalogue ( turon 1993 ) , the tycho catalog ( esa 1997 ) , and the ned . the identification criterion was the angular distance : the nearer an object is to the calculated uv source position , the more likely it is to be the actual counterpart . additional constraints were imposed on a physical basis . the bluer an optical candidate , the more likely it is to be the real counterpart . the spectral type of the object , if it was listed in the catalog source , was an important input to the determination of the accepted counterpart . sources for which no physical properties could be found were correlated with guide star catalog ( lasker , jenkner , & russell 1988 ) objects , if such could be found in the neighbourhood of a source . the remaining identifications were extracted mainly from the first edition of the us naval observatory two - colour catalog ( usno - a1.0 , hereafter usno : monet 1996 ) . this catalog is based on scans of the palomar and eso / serc sky survey blue and red schmidt telescope plates . the absolute scale of the blue and red magnitudes is not accurate , but a `` local '' comparison of colours between a number of candidates for the same uv source is useful for identifying the bluest and brightest object , which is most likely to be the counterpart of the uv source . the second version of this catalog ( usno - a2.0 : monet 1998 ) has improved astrometric positions and photometric calibration and was used to check a few problematic sources . the photometric improvement is based on the b and v magnitudes of the tycho catalog for the brighter objects and on ccd observations of standard stars for some sky regions . however , the final reported photometric accuracy ( 0.15 mag internal error and up to 0.5 mag due to systematic effects ) is not much better that that of the a1.0 version of the catalog . the usno catalog , whose candidates are designated by a `` usno '' prefix in the lists , contains both stellar and non - stellar sources . on completing the identification process and analyzing the results , a few ( @xmath1010 for all four fields combined ) fairly strong sources , all very near the edge of the images , remained unidentified . as these are relatively strong uv sources they are very probably not artefacts , and a special effort was exerted to identify them . an additional search in the tycho catalog , using a search radius of 5 arcmin , produced good identifications for most sources with optical counterparts located within 3 - 4 arcmin of the uv source . analysis of other identified sources in the same vicinity showed that a systematic positional error appears sometimes at the edge of the images . this is presumably due to the electrostatic distortions being the greatest at the edges of the detector , and not having enough pre - identified stars in these locations as input for the astrometric solution used to compute the celestial coordinates . these additional identifications have been included in the lists shown in tables 2 and 4 . the entries of table 4 , which reports the identifications , repeat the source identifier used here in table 2 column 1 , give a leading catalog identifier for the proposed counterpart ( column 2 ) , its v - band magnitude ( or b - band , in some cases ) in column 3 , a note on its spectral type if it is a star or on its morphological classification if it is a galaxy ( column 4 ) , its listed b - v colour ( or b - r , in some cases ) in column 5 , and the candidate s celestial position in columns 6 and 7 . all positions reported here are for j2000 . the he s ( wisotzki 1996 ) is an objective - prism survey aimed at detecting bright qsos . it was performed with the eso schmidt telescope using iiia - j plates and a 4@xmath2 objective prism . the plates have a resolution of 15 at h@xmath11 ( seeing - dependent ) and cover the range 3200 to 5400 . the celestial area covered by the he s is mainly in the extragalactic sky , roughly above @xmath12b@xmath12=30@xmath2 . because of this sky coverage limitation , two of the faust regions analysed here could not be compared with the he s data sets . in particular , we limit the discussion of the objective - prism spectra to the fields of m83 and of dorado , as the other two regions have a galactic latitude too low to be included in the he s . the he s plates were scanned on the hamburg pds with a 30@xmath1330@xmath14m@xmath15 aperture , and spectra were automatically extracted using object detection on the digitized sky survey . each spectrally observed candidate has , therefore , an astrometric solution good to @xmath02 arcsec and a linearized spectrum ( using the plate sensitometric relation ) covering the range of the he s objective - prism plates , which is @xmath0300 pixels long . on average , @xmath075% of the spectra do not show overlap with other neighbouring spectra and can be used to verify or revise the classification and assignation of an optical candidate to a faust uv source . the search for optical counterparts on the he s data set was done by lw in early - november 1998 . a search in a region 3 arcmin wide around each faust source was performed and the candidates were extracted and examined by nb . in a number of instances , the candidate optical object was so bright as to cause partial or even full saturation of the objective - prism plate . for several cases , the spectral information allowing a rough classification could be recovered using additional scan lines from the two - dimensional scan data . we could thus use the balmer absorption lines and the 4000 ca ii break as classifier criteria in objects as bright as b@xmath169.5 . an optical candidate was accepted as the probable counterpart of a uv source if it : 1 . was the brightest source , 2 . was the nearest source , 3 . showed strong balmer absorption lines , 4 . showed a significantly strong blue - uv `` tail '' of the spectrum not all these conditions had to be fulfilled simultaneously for a source to be accepted as a counterpart . we are aware that these may seem to be subjective criteria , but adopt them here in absence of a systematic spectral classification of all he s spectra . an example of the output produced from the he s plates for this search of optical counterparts , one page of the many produced for the various candidates and inspected by us , is shown in figure 6 . three objects are displayed in this figure ; the bottom one is source fm 31 identified as a new sub - dwarf star . the various entries for each candidate include the number of the eso - schmidt plate on which the object appears , various classification parameters , the image of the field from the digitized sky survey with the candidate object centered , and two representations of the objective - prism spectrum , one at full resolution ( centered ) and another binned to low resolution ( to the right ) . the strong blue tail , indicative of a very hot object , is evident for candidate he1342 - 2728=fm 31 . the he s plates yield also blue ( b@xmath17 ) magnitudes , from the blue part of the spectrum and a calibration against standard stars ( vanelle 1996 ) . the calibration is good to @xmath00.2 mag provided that the spectrum is not overexposed ( b@xmath1811.5 ) . in principle , it is also possible to determine the u b and b v colour indices of the candidates from the spectra , using a calibration of spectral slope against colour indices ; however , here one has to guard even more carefully against chance partial spectrum overlap . we emphasize that the assignation of optical counterparts in the dor and m83 fields was done in a completely independent way from the assignments based on catalog correlations . in most cases , the he s spectra confirmed the original catalog selection of a counterpart . the counterparts selected from the he s spectra are listed in table 4 if they were `` better '' than those originally selected using catalog correlations , or if they offered the choice of a single counterpart , when the catalog searches yielded a number of optical sources in the same error ellipse . the he s sources can be recognized by the he prefix identifier and by the approximate spectral type , derived from the visual inspection of the spectrum tracing . note that in many cases the location of the uv source on the he s contains no object with a spectrum which could be associated with the faust source . in the dor field 37 uv sources had no such he s counterpart . in a number of cases , we could not solve the question of the proper counterpart , even using the he s data . in such cases we reverted to the usno catalog ( version a2.0 ) and selected the best counterpart to be the nearest and bluest listed object , and this is the entry adopted for table 4 . in this section we summarize results from the detection and identification process , including the confirmatory or enhancing results with the he s spectra using the data from table 4 . the summary , which splits the source identification into three broad classes ( extragalactic , stars , and unknown ) is given in table 5 . one item stands out immediately , that is the low number of `` unknown '' sources in the dorado and m83 fields . in these two faust fields the percentage of such sources is 19% and 5% , respectively , while it is @xmath030% in the other two fields . the only difference between the first pair of fields and the second is that for dor and m83 we used the additional information from the he s survey . this indicates that using the he s resource we could classify ( in broad spectral types ) an additional @xmath015% of the faust uv source , some of which belong to the interesting `` hot , evolved '' stellar variety . in this context , an important statistic is the number of such sources ( sub - dwarfs and white dwarfs ) counted in each field . whereas the images without he s data have extremely few such sources ( one in cen and none in tel ) , the other two fields contain significantly more sources ( nine in dor and five in m83 ) . this emphasizes the importance of additional spectral information in untangling the nature of the ( optically faint ) hot uv sources . a comparison of the detected and identified stars with the predicted number of stars , using the updated model of the uv galaxy of brosch ( 1991 ) . the original model uses the bahcall & soneira ( 1980 ) predictor routines , where the milky way is approximated as an exponential disk and a bulge , each with its own luminosity function . the extinction by dust is treated by assuming a slab - like distribution of dust along the galactic equator with an exponential scale height above the disk . the bahcall & soneira model yields predictions for optical bands , and it has also been extended to the near - ir . the uv application of this model uses an optical - to - uv transformation for stars based on their colours . this was derived from iue observations and was extended to late - type stars using kurucz ( 1991 ) model atmospheres . in addition , the uv predictor model includes the gould belt , and a thick disk of white dwarfs with the scale height determined by boyle ( 1989 ) . the comparison of the observations with the model predictions is shown in figures 7 and 8 . we show there the distribution of the differential star counts ( number of stars per uv magnitude ) vs. the model predictions in the left panels , where the actual star counts are represented by squares with error bars derived from poisson statistics . the right - hand plot for each region corresponds to the distribution of colours . to produce this , we deleted from the star list those objects classified as `` hot evolved stars '' , such as hot wds and sd s , based on the catalog information or on data extracted from the he s spectra , because they are not predicted well by our model . we also deleted all the extragalactic objects , which are not included in the predictor model . our model predicts the star numbers and magnitude distributions per square degree , and the predicted colour distribution depends on the faint magnitude cutoff . the fainter this cutoff , the redder is the peak and the median colour of the stellar distribution . allowing fainter stars in the calculation of the colour distribution brings in more red objects than blue ones , driving the total distribution to the red . the actual cutoff in a specific faust exposure is location - dependent . this is because the exposure depth is shallower the closer an object is to the edge of the faust image , the part of the area with the low exposure time . for this reason , we adopted the following strategy when scaling the predictions to the `` effective area '' surveyed : 1 . we selected a magnitude cutoff for the colour prediction at 70% of the peak in the magnitude distribution . that is , we set the cutoff to a fainter magnitude than the peak of the magnitude distribution of the specific field , at the ( interpolated ) magnitude level where the stellar magnitude distribution reached 70% of the peak value . we decided on a scaling constant from the `` per square degree '' prediction of the model and the real star counts in * both * the magnitude and colour distributions using the total number of stars detected to be brighter than or equal to m@xmath19=11 [ n(m@xmath20 ) ] and the similar number of stars , brighter than 11 mag , predicted by the model [ n(m@xmath21 ) ] . the ratio of these values was used to scale both the magnitude distribution prediction and the colour distribution prediction . the actual detected distributions , and those predicted by the model and scaled as explained above , are shown in figures 7 and 8 . in deriving the uv - v colour of an object we used the magitude of the counterpart as listed in table 4 . in many cases , this is the usno catalog `` blue '' magnitude , which may be substantially different from the actual v magnitude of the objects . however , we estimate that this is a random error , which will only increase the scatter of the results but will not introduce a systematic bias , as most of the stars that have spectral types tend to be a and f , which have small b v values . the results , shown in figures 7 and 8 for the four fields , indicate that the procedure is valid and that the predicted stellar populations , as well as the general colour distributions , correspond to those actually observed . we emphasize that , while the choice of the scaling factor seems arbitrary by using the total number of stars to 11 mag . , the use of the same constant for scaling the colour distribution is a constraint , and the subsequent reasonable fit represents an independent confirmation of this procedure . the possible sources of discrepancy between the model predictions and the actual measurements are presumably stars that are too blue in the uv than the model is able to predict . such stars are , for example , hot evolved objects that are not included in the model . the discrepancy is emphasized by the trend of the plots in the right panels of figs . 7 and 8 to under - predict the observed colour distribution at the blue end . this indicates that probably there are more hot evolved stars among those observed by faust than revealed by the spectroscopic classification . the assumption is supported by the size of the under - prediction : the discrepancy is higher for the two fields where we do not have he s spectroscopic data than in the two fields with such data . this finding indicates that collections of uv observations have the potential of enlarging the samples of such objects , with obvious influences on our understanding of the structure and evolution of the milky way . in this section we discuss those stellar sources classified as either hot evolved stars ( sd s , wds , etc . ) , or as red stars ( k0 and later ) , which represent the hot and the cold extremes of the stellar distribution . it is noticeable that of 24 such objects in the four faust fields treated here , only two are in a image for which we have no he s information . this demonstrates , again , the value of ancilliary spectral information from which we extract the special - interest objects . we identified 15 stars as hot evolved objects , mostly in the dor and m83 fields . these are relatively faint uv sources , with a median magnitude of 11.44 , and with very blue colours ( median uv - v , or uv - b=3.65 ) that imply rather faint optical magnitudes . the optical faintness explains the lack of spectral information ; only wide - field surveys using multi - band colours or an objective prism could have picked these objects out . most are sdo , sdb , or sda , but one ( fc 89 ) is classified sdf2 and is indeed very red , with uv - v=3.94 . this is a high proper motion star , as expected of a subdwarf . the source with the most negative colour index is fd 46 , with uv - v=5.03 . we searched the lists of objects for those with extremely blue uv - v colours ( @xmath224 ) and found very few of these . specifically , we detected three such stars in the dor field , three in cen , seven in m83 , and two in tel . six of these are classified in simbad or by us as sub - dwarfs , one is an emission - line galaxy , and for the rest we do not have information beyond the identification with a star - like object and a rough optical magnitude . it is possible that most are also hot sub - dwarfs or binary systems containing hot , evolved stars , and spectroscopic follow - ups are necessary to clarify this point . there are 10 objects classified k0 or later in the four faust fields . as expected of red stars , their uv - v colours are in general large and positive . the colour indices are , though , much bluer than could be expected from the effective temperature which corresponds to their spectral type . the variations of the colour index within the same allocated spectral type are considerable ; among four k2 stars the spread of uv - v is more than 5.5 mag ! normally , k stars are expected to have uv - v@xmath163.2 - 7.4 , but some stars in this class are more uv - bright than this , presumably because of some form of coronal activity ( see jordan 1987 ) , or because of the presence of a hot secondary . one object , fm 81 , is classified as a m0 star with an expected uv - v@xmath169.45 colour , yet it shows an almost nil colour index . checking with simbad , we find this star to be rw hya , a well - known symbiotic system consisting of an m2 giant and a @xmath080,000k wd ( kenyon & mikolajewska 1995 ) . this demonstrates the advantage of a wide wavelength base , from the uv to the optical , with which to identify potentially peculiar objects . another one , fc 168 , is an m3iii star with uv - v=+5.4 , while we expect for such an object uv - v=+9.67 . although there are no indications in simbad of any peculiarity concerning this star , it is likely that this is yet another binary system , with a hot component responsible for the uv excess . we remarked in earlier papers that metallic - line a stars ( am ) appear fainter in the uv than their regular counterparts . the present study is particularly useful to check this behavior because it yielded a relatively large stellar sample . table 4 lists six am stars ( earlier than a5 ) and one late - bm star . these are plotted as filled squares in the colour - colour diagram shown in figure 9 . the comparison is done with regular a stars from the same faust images . the a0 stars are plotted as circles and the late - a stars ( a7 , a8 , and a9 ) are plotted as filled triangles . we added the reddening vector assuming the standard savage & mathis ( 1979 ) relation as a vector for e(b - v)=0.2 to demonstrate that the discrepancy could not be due to reddening by a standard milky way extinction law . it is clear that the am stars are uv - deficient in comparison to most of the a - type regular stars , including the latest a stars ( a9 ) . this deficiency has been noted before ( by van dijk 1978 ) and is presumably due to line blanketing of their spectral energy distributions . note though that the study of van dijk , and others dealing with the issue of uv deficiency of a stars , concentrated mostly on ap objects rather than on am stars . in principle , this points a way to use the property of these a stars to be less uv - bright than expected from their rough optical spectral classification , to identify possible candidate am ( or ap ) stars from among those objects in our four fields for which we found spectral classifications . the requirement is for good uv and optical photometry of the candidate objects , which should have been previously classified as a stars . fifty - four objects in table 4 have been identified as extragalactic objects . in two instances two optical counterparts could correspond to specific uv sources ; the extragalactic ones are listed in table 6 . as seen in previous cases , faust detected more uv flux than iue in the few instances where such observations are available . in the case of ngc 5253 , the iue spectra yield monochromatic uv magnitudes of [ 1482]=11.03 and [ 1913]=11.39 , whereas faust measured 9.89@xmath230.19 mag . the iue measurements refer to `` standard '' bands in the sw spectral range , centered at the listed wavelengths . this discrepancy between faust and iue is undoubtedly the result of measuring a large object with a small aperture . for ngc 5236=fm 83 the faust magnitude is 7.92@xmath230.17 whereas iue measured [ 1482]=10.48 and [ 1913]=10.82 . ngc 5135 has [ faust]=13.84@xmath231.01 and [ 1482]=13.94 and [ 1913]=13.87 , presumably the result of most of the uv emission being produced in the `` composite '' seyfert 2+starburst compact nucleus ( phillips 1983 ) . the amount of uv flux external to the iue aperture may be quite significant in some cases . among the more interesting extragalactic objects detected by faust in these four fields we note m83 ( source fm 70 ) , a starburst galaxy in which no less than six supernovae were discovered in a period of 60 years . a similar case is tol 34 ( source fm 181 ) , which contains a starburst and a sy 2 nucleus . some interacting galaxies appear as uv sources : the compact group of ngc 6845 with four peculiar spirals is such an example . unfortunately , the angular resolution of faust is not sufficient to show which of the objects is the uv source . this group would make an interesting target for a future uv imaging observation . we detected nine uv - emitting galactic nuclei , of which six were identified from the he s spectroscopic data . one of the more interesting such sources is the optically - violent - variable bl lac object euve j2009 - 48.8 ( source ft 34 ) . ours is the first listed uv measurement of this object , which has been detected in @xmath11-rays and xrays , as well as in the euv range . note that the uv - v colour index we measure is based on the optical magnitude listed in the ned . because of the intrinsic variability of a bl lac object , the real colour index may be different from the `` instantaneous '' one . note also the detection of the cd galaxy eso 444-g046 , the largest in the cluster abell 3558 , as a faust uv source . this object has a neutral colour index , uv - v@xmath16 + 0.06 . one other object detected by faust could belong to the same cluster : this is source fm 185 , a star - like object listed only with its gsc number but which showed emission lines in he s spectra and appears almost star - like on the blue plates of the he s . the object was classified by us as an `` emission - line galaxy '' ( elg ) . the colour index exhibited by this galaxy is one magnitude bluer than that of the cd galaxy . the elg is apparently part of a chain of larger galaxies in the cluster , extending from north - west to south - east . lacking a redshift , the assumption that this galaxy is a member of the a3558 cluster remains to be proven . we remarked above on the puzzling discrepancy in the source count between the m83 and the tel areas . briefly , the discrepancy is that these two sky regions , at almost identical @xmath24b@xmath24 , have approximately the same number of uv sources but one ( m83 ) has three times more exposure than the other . the puzzle only deepens when we consider the normalized source count , that is , the projected source density . we first define a representative projected source density for a field @xmath25 as @xmath26 , where @xmath27 is the number of sources brighter than m@xmath19=11 detected in the specific field , and @xmath28 is the angular coverage obtained by faust and listed in table 1 . this magnitude cutoff makes it likely that we are not missing sources that bright even in the image with the shortest exposure time ( 280 sec for the central region of the tel image ) , as indeed the source detection histograms in fig . 5 show . comparing the four values of @xmath29 , where @xmath30 is the exposure time for the image of area @xmath25 , we find that the tel regions has a @xmath31 higher than the three other regions , which have almost identical @xmath32 values . the tel region appears to have many more uv sources that it is `` entitled '' to have , given its projected sky area and depth of exposure . the discrepancy can * not * be explained by the galactic model . the modified bahcall - soneira predictions appear to fit reasonably well both the colour and the magnitude distributions of uv sources . one possibility is that the tel region contains even more of the hot , evolved stellar types than do the other three regions . this could , in a way , be understood when considering the pointing direction : the line - of - sight to the tel region grazes the bulge of the galaxy whereas the three other images have wider separations from the galactic center . if the excess of uv sources in the direction of telescopium would be due to hot , horizontal - branch stars in the bulge , these would appear as the sources near the faint completeness limit of the faust exposure . we estimate this from the colours and magnitudes of zero - age horizontal - branch models listed in table 2 of mould ( 1996 ) and from a galactocentric distance of 8 kpc . another possibility could be for the m83 region to have , for some reason , more extinction than the tel area . we checked this using the stars identified as a0 in table 4 . there are 15 such stars in the m83 frame and 11 in the tel image . the mean b v values are 0.09@xmath230.10 and 0.06@xmath230.07 , respectively . the mean uv v values , on the other hand , are 0.68@xmath231.06 and 0.16@xmath2377 , respectively . the difference between the mean b v values of the two regions could imply an expected difference in the mean uv v values of @xmath00.15 , assuming the wavelength dependence of the extinction as in the `` typical '' galactic law ( savage & mathis 1979 ) , whereas we measure @xmath33(uv v)@xmath160.84 . this could be the case if the slope of the extinction law would be steeper in the faust - observed regions than the savage & mathis relation , and is supported also by the slope of the uv v _ vs. _ b v relation for all the a0 stars in table 4 . the 43 objects have a regression slope of 6.36@xmath231.29 , whereas the savage & mathis relation indicates a slope of 4.9 . a steeper dependence of the uv extinction may be an indicator for the presence of smaller dust grains than encountered on average in the milky way . we note this as a possibility , but do not explore it further in this paper . we presented faust uv observations in four sky area , mostly in the fourth galactic quadrant . our detection algorithm identified 777 uv sources in these regions . we identified the sources with optical counterparts mainly through correlations with catalogued object , and succeeded in increasing the number and quality of identifications by including objective - prism spectra . the stellar objects detected and identified in this program fit reasonably well our model of the uv galaxy . we discussed the more interesting stellar and extragalactic sources , and pointed out that the tel area may contain horizontal branch stars on the outskirts of the galactic bulge . this paper made use of the simbad data bank , established and maintained by the centre du donnes stellaire in strasbourg . the paper is dedicated to the memory of dr . barry lasker , one of the `` founding fathers '' of the guide star catalog and of other catalogs derived from scanning of photographic plates . uv research at tel aviv university is supported by grants from the ministry of science and arts through the israel space agency , from the austrian friends of tel aviv university , and from a center of excellence award from the israel science foundation . nb acknowledges support from a us - israel binational award to study uv sources measured by the faust experiment . we are grateful to mr . benny bilenko for updating the model of uv stellar distributions in the milky way . constuctive remarks by an anonymous referee are gratefully acknowledged . alfaro , e. & efremov , yu . n. 1996 , rev.mex.a.a . , 4 , 1 bahcall , j.n . & soneira , r.m . 1980 astrophys . j. suppl 44 , 73 . boyle , b.j . 1989 mon . not . 240 , 533 . bowyer , s. , sasseen , t.p . , lampton , m. & wu , x. 1993 , apj , 415 , 875 bowyer , s. , sasseen , p.t . , xiauyi , w. & lampton , m. 1995 , apjs , 96 , 461 ( fsc ) brosch , n. 1991 , mnras , 250 , 780 brosch , n. , almoznino , e. , leibowitz , e.m . , netzer , h. , sasseen , t. , bowyer , s. , lampton , m. & wu , x. 1995 , apj , 450 , 137 brosch , n. , formiggini , l. , almoznino , e. , sasseen , t. , lampton , m. & bowyer , s. 1997 , apjs , 111 , 143 brosch , n. , ofek , e. , almoznino , e. , sasseen , t. , lampton , m. , & bowyer , s. 1998 , mnras , 295 , 959 esa `` the hipparcos and tycho catalogues '' 1997 , esa sp-1200 hagen , h .- j . , groote , d. , engels , d. & reimers , d. 1995 , a&as , 111 , 195 jordan , c. , ayres , t.r . , brown , a. , linsky , j.l . & simon , t. 1987 , mnras , 225 , 903 kenyon , s.j . & mikolajewska , j. 1995 , aj , 110 , 391 kurucz , r.l . 1991 nato asi ser . c. , math . , 341 , 441 . lampton , m. , sasseen , t.p . , wu , x. & bowyer , s. 1993 , geophys . res . lett . * 20 * , 539 lasker , b. , jenkner , h. & russell , j.l . 1988 in `` mapping the sky : past heritage and future directions '' ( s. debarat , ed . ) , dordrecht : kluwer , p. 229 . monet , d. 1996 , `` usno - a1.0 catralog '' , washigton : u.s . naval observatory monet , d. 1998 , `` usno - a2.0 catalog '' , washigton : u.s . naval observatory mould , j.r . , watson , a.m. , gallagher , j.s . iii , ballester , g.e . , burrows , c.j . , casertano , s. , clarke , j.t . , crisp , d. , griffiths , r.e . , hester , j.j . , hoessel , j.g . , holtzman , j.a . , scowen , p.a . , stapelfeldt , k.r . , trauger , j.t . & westphal , j.a . 1996 , apj , 461 , 762 paturel , g. , vauglin , i. , garnier , r. , marthinet , m.c . , petit , c. , di nella , h. , bottinelli , l. , gouguenheim , l. & durand , n. 1992 _ leda : the lyon - meudon extragalactic database _ , cd - rom observatorie de lyon phillips , m.m . , charles , p.a . & baldwin , j.a . 1983 , apj , 266 , 485 reimers , d. & wisotzki , l. 1997 , the messenger , 88 , 14 sao 1966 , `` smithsonian astrophysical observatory star catalog '' , washington : smithsonian institution , publication no . 4652 savage , b.d . & mathis , j.s . 1979 , araa , 17 , 73 thompson , g.i . , nandy , k. , jamar , c. , monfils , a. , houziaux , l. , carnochan , d.j . & wilson , r. 1978 , _ catalogue of stellar ultraviolet fluxes _ , the science research council ( td-1 ) turon , c. 1993 bull . cds 43 . van dijk , w. , kerssiers , a. , hammerschlag - hensberg , g. & wesselius , p.r . 1978 , a&a , 66 , 187 vanelle , c. 1996 , msc thesis , hamburger sternwarte wisotzki , l. , kohler , t. , groote , d. & reimers , d. 1996 , a&as , 115 , 227 cccccccccc dor & 04 15 36 & -56 04 12 & 266.0 & -43.7 & 200 - 610 & 60.5 & 142 & 58 & 12.7 cen & 12 47 24 & -40 46 12 & 302.1 & 22.1 & 300 - 1100 & 62.0 & 249 & 129 & 12.5 m83 & 13 34 48 & -28 45 00 & 314.3 & 33.2 & 250 - 850 & 60.1 & 197 & 81 & 13.0 tel & 20 11 38 & -47 06 00 & 352.5 & -32.8 & 100 - 280 & 52.6 & 189 & 105 & 12.35 fc 1 & 12 47 36 & -37 23 22 & 11.85 & 0.40 & 3466 fc 2 & 12 55 45 & -37 17 35 & 11.26 & 0.32 & 3574 fc 3 & 12 40 22 & -37 51 52 & 7.99 & 0.17 & 3350 fc 4 & 12 56 32 & -37 39 28 & 11.75 & 0.29 & 3580 fc 5 & 12 41 54 & -37 53 32 & 10.07 & 0.21 & 3378 fc 6 & 12 42 37 & -37 54 09 & 11.30 & 0.23 & 3392 fc 7 & 13 03 21 & -37 44 39 & 10.73 & 0.29 & 3654 fc 8 & 12 56 50 & -37 57 31 & 12.19 & 0.36 & 3584 fc 9 & 12 52 21 & -38 02 20 & 13.23 & 0.76 & 3538 fc 10 & 12 53 32 & -38 03 43 & 12.56 & 0.45 & 3548 fc 11 & 12 38 03 & -38 15 09 & 11.50 & 0.33 & 3304 fc 12 & 12 49 58 & -38 18 24 & 12.36 & 0.39 & 3510 fc 13 & 12 43 48 & -38 20 52 & 11.85 & 0.34 & 3410 fc 14 & 12 48 21 & -38 22 30 & 11.40 & 0.25 & 3481 fc 15 & 12 39 42 & -38 24 46 & 11.98 & 0.37 & 3333 fc 16 & 13 04 38 & -38 23 54 & 9.39 & 0.19 & 3665 fc 17 & 12 39 09 & -38 44 46 & 10.08 & 0.20 & 3317 fc 18 & 12 34 51 & -38 49 43 & 10.39 & 0.23 & 3253 fc 19 & 12 59 41 & -38 42 32 & 8.81 & 0.17 & 3614 fc 20 & 12 47 45 & -38 51 16 & 10.49 & 0.20 & 3467 fc 21 & 12 45 27 & -38 51 25 & 11.19 & 0.24 & 3434 fc 22 & 13 00 35 & -38 44 38 & 12.24 & 0.28 & 3631 fc 23 & 12 58 01 & -38 55 04 & 7.48 & 0.16 & 3599 fc 24 & 12 40 19 & -39 02 42 & 13.07 & 0.83 & 3346 fc 25 & 12 51 54 & -38 58 24 & 13.58 & 0.92 & 3526 fc 26 & 12 31 48 & -39 13 14 & 10.70 & 0.30 & 3208 fc 27 & 12 52 43 & -39 03 40 & 12.16 & 0.40 & 3542 fc 28 & 12 58 53 & -39 06 28 & 6.73 & 0.16 & 3608 fc 29 & 13 03 37 & -38 58 24 & 10.58 & 0.20 & 3657 fc 30 & 12 57 11 & -39 06 51 & 10.80 & 0.22 & 3587 fc 31 & 13 01 04 & -39 03 05 & 11.79 & 0.30 & 3633 fc 32 & 12 36 10 & -39 22 40 & 12.51 & 0.58 & 3276 fc 33 & 12 58 01 & -39 16 06 & 9.85 & 0.18 & 3600 fc 34 & 12 50 26 & -39 26 33 & 10.94 & 0.25 & ? fc 35 & 13 00 40 & -39 17 38 & 11.67 & 0.27 & 3629 fc 36 & 12 58 12 & -39 22 23 & 11.24 & 0.20 & 3601 fc 37 & 12 53 46 & -39 29 10 & 10.85 & 0.22 & 3553 fc 38 & 12 39 18 & -39 39 35 & 9.89 & 0.22 & 3320 fc 39 & 12 52 01 & -39 41 41 & 5.23 & 0.16 & 3529 fc 40 & 12 49 09 & -39 43 25 & 5.82 & 0.16 & 3498 fc 41 & 12 35 58 & -39 52 00 & 5.22 & 0.16 & 3271 fc 42 & 12 44 12 & -39 46 12 & 9.60 & 0.18 & 3420 fc 43 & 12 39 53 & -39 58 08 & 5.42 & 0.16 & 3338 fc 44 & 12 42 08 & -40 04 42 & 10.66 & 0.26 & 3386 fc 45 & 12 43 26 & -40 10 42 & 8.47 & 0.17 & 3407 fc 46 & 12 53 25 & -40 11 10 & 6.04 & 0.16 & 3545 fc 47 & 12 31 29 & -40 16 34 & 10.63 & 0.24 & 3198 fc 48 & 12 49 41 & -40 13 26 & 12.30 & 0.46 & 3507 fc 49 & 13 03 33 & -40 14 59 & 12.64 & 0.40 & 3655 fc 50 & 12 40 21 & -40 37 31 & 10.91 & 0.24 & 3348 fc 51 & 12 49 12 & -40 37 58 & 10.12 & 0.19 & 3501 fc 52 & 12 34 13 & -40 49 54 & 11.17 & 0.24 & 3245 fc 53 & 12 48 40 & -40 42 36 & 12.77 & 0.32 & 3486 fc 54 & 12 59 07 & -40 37 53 & 11.85 & 0.33 & 3610 fc 55 & 12 51 35 & -40 44 18 & 13.81 & 1.04 & 3524 fc 56 & 12 35 46 & -41 01 15 & 6.94 & 0.16 & 3270 fc 57 & 12 46 54 & -40 53 27 & 11.60 & 0.27 & 3458 fc 58 & 13 04 47 & -40 42 34 & 11.85 & 0.30 & 3666 fc 59 & 12 52 31 & -40 56 22 & 10.71 & 0.21 & 3539 fc 60 & 12 59 55 & -40 52 40 & 9.57 & 0.18 & 3619 fc 61 & 12 45 15 & -41 02 35 & 13.06 & 0.57 & 3431 fc 62 & 13 07 39 & -40 45 59 & 12.50 & 0.49 & 3680 fc 63 & 12 38 19 & -41 12 22 & 9.76 & 0.19 & 3308 fc 64 & 12 44 40 & -41 13 17 & 12.85 & 0.52 & 3429 fc 65 & 12 51 01 & -41 15 06 & 10.22 & 0.18 & 3519 fc 66 & 12 33 35 & -41 25 24 & 8.87 & 0.17 & 3232 fc 67 & 12 52 16 & -41 19 40 & 8.01 & 0.16 & 3536 fc 68 & 13 10 37 & -40 56 50 & 11.09 & 0.28 & 3702 fc 69 & 12 31 32 & -41 31 47 & 10.15 & 0.21 & 3199 fc 70 & 13 02 40 & -41 13 42 & 11.58 & 0.26 & 3648 fc 71 & 12 47 50 & -41 28 01 & 12.58 & 0.43 & 3470 fc 72 & 12 59 55 & -41 23 11 & 11.04 & 0.22 & 3620 fc 73 & 12 43 14 & -41 36 15 & 11.97 & 0.35 & 3403 fc 74 & 13 08 15 & -41 23 05 & 9.00 & 0.17 & 3684 fc 75 & 13 04 01 & -41 24 09 & 12.65 & 0.44 & 3662 fc 76 & 12 50 25 & -41 42 59 & 11.24 & 0.26 & 3516 fc 77 & 12 59 14 & -41 36 27 & 8.94 & 0.17 & 3611 fc 78 & 12 39 26 & -41 50 48 & 9.57 & 0.18 & 3328 fc 79 & 13 05 00 & -41 30 41 & 14.48 & 2.02 & 3667 fc 80 & 12 33 50 & -41 57 49 & 11.24 & 0.28 & 3239 fc 81 & 12 30 29 & -41 57 11 & 11.05 & 0.27 & 3185 fc 82 & 12 50 25 & -41 54 09 & 10.20 & 0.20 & 3515 fc 83 & 12 37 24 & -42 01 19 & 12.22 & 0.36 & 3292 fc 84 & 13 03 01 & -41 46 57 & 10.48 & 0.20 & 3651 fc 85 & 12 56 46 & -41 55 04 & 12.09 & 0.38 & 3582 fc 86 & 12 44 19 & -42 05 35 & 12.70 & 0.50 & 3422 fc 87 & 13 01 06 & -41 58 55 & 11.48 & 0.26 & 3634 fc 88 & 12 56 09 & -42 09 10 & 7.87 & 0.17 & 3576 fc 89 & 12 36 14 & -42 18 47 & 11.92 & 0.35 & 3275 fc 90 & 12 58 23 & -42 05 43 & 13.31 & 0.80 & 3603 fc 91 & 13 07 33 & -42 01 59 & 9.37 & 0.18 & 3678 fc 92 & 12 55 10 & -42 18 26 & 4.98 & 0.16 & 3569 fc 93 & 12 39 45 & -42 21 41 & 12.09 & 0.39 & 3335 fc 94 & 13 10 06 & -42 12 23 & 9.21 & 0.18 & 3697 fc 95 & 12 37 30 & -42 35 45 & 11.20 & 0.25 & 3293 fc 96 & 13 11 08 & -42 12 60 & 11.43 & 0.25 & 3703 fc 97 & 13 00 16 & -42 29 48 & 11.74 & 0.29 & 3624 fc 98 & 13 01 28 & -42 29 59 & 12.20 & 0.34 & 3636 fc 99 & 13 02 12 & -42 28 60 & 13.25 & 0.40 & 3646 fc 100 & 12 39 33 & -42 58 09 & 11.60 & 0.33 & 3329 fc 101 & 12 52 05 & -42 55 28 & 10.53 & 0.21 & 3530 fc 102 & 12 38 01 & -43 01 17 & 12.49 & 0.33 & 3305 fc 103 & 12 45 59 & -43 05 06 & 7.91 & 0.16 & 3444 fc 104 & 12 38 03 & -43 03 35 & 12.51 & 0.33 & ? fc 105 & 12 53 29 & -43 04 17 & 9.41 & 0.18 & 3546 fc 106 & 12 51 40 & -43 10 34 & 11.46 & 0.30 & 3525 fc 107 & 12 49 28 & -43 12 44 & 14.67 & 2.49 & 3503 fc 108 & 12 48 22 & -43 19 37 & 11.89 & 0.35 & 3482 fc 109 & 12 41 31 & -43 37 55 & 9.46 & 0.18 & 3372 fc 110 & 13 01 30 & -43 22 37 & 10.42 & 0.20 & 3639 fc 111 & 13 02 39 & -43 20 59 & 10.36 & 0.21 & 3647 fc 112 & 12 48 45 & -43 42 36 & 7.55 & 0.16 & 3489 fc 113 & 13 08 57 & -43 27 50 & 8.38 & 0.17 & 3689 fc 114 & 12 35 49 & -43 49 27 & 11.35 & 0.31 & 3269 fc 115 & 12 42 02 & -43 53 10 & 11.96 & 0.35 & 3379 fc 116 & 12 56 14 & -43 46 40 & 11.70 & 0.32 & 3577 fc 117 & 12 49 02 & -43 56 26 & 9.51 & 0.18 & 3494 fc 118 & 13 00 55 & -43 47 50 & 10.83 & 0.24 & 3632 fc 119 & 12 43 22 & -44 00 13 & 12.09 & 0.28 & 3404 fc 120 & 12 42 49 & -44 05 49 & 11.72 & 0.26 & 3395 fc 121 & 12 59 20 & -43 59 42 & 11.01 & 0.24 & 3613 fc 122 & 13 04 58 & -43 54 50 & 10.49 & 0.23 & 3668 fc 123 & 12 55 10 & -44 10 27 & 12.27 & 0.35 & 3567 fc 124 & 12 41 25 & -44 17 34 & 13.25 & 1.03 & 3368 fc 125 & 13 06 11 & -44 06 08 & 11.80 & 0.45 & 3674 fc 126 & 12 54 15 & -44 23 15 & 12.16 & 0.35 & 3559 fc 127 & 12 43 06 & -44 34 02 & 10.56 & 0.23 & 3401 fc 128 & 12 48 18 & -44 43 22 & 5.92 & 0.16 & 3479 fc 129 & 12 41 14 & -44 41 51 & 11.06 & 0.30 & 3363 fc 130 & 13 06 15 & -44 24 36 & 9.92 & 0.22 & 3673 fc 131 & 12 45 30 & -44 43 49 & 11.01 & 0.25 & 3437 fc 132 & 13 03 25 & -44 34 04 & 8.75 & 0.18 & 3653 fc 133 & 13 01 20 & -44 45 55 & 5.69 & 0.16 & 3635 fc 134 & 12 56 24 & -44 51 50 & 11.93 & 0.40 & 3578 fc 135 & 13 04 04 & -44 55 37 & 10.34 & 0.27 & 3663 fc 136 & 12 53 52 & -45 05 47 & 11.64 & 0.36 & 3555 fc 137 & 13 01 29 & -45 11 02 & 9.40 & 0.21 & 3637 fc 138 & 12 48 39 & -45 25 42 & 10.85 & 0.27 & 3488 fc 139 & 12 46 05 & -45 29 59 & 9.98 & 0.23 & 3448 fc 140 & 12 52 24 & -45 30 01 & 11.16 & 0.37 & 3537 fc 141 & 12 44 33 & -37 18 20 & 11.95 & 0.46 & ? fc 142 & 12 53 14 & -37 34 32 & 11.62 & 0.16 & ? fc 143 & 12 45 01 & -37 43 04 & 12.41 & 0.51 & ? fc 144 & 12 50 55 & -37 50 52 & 12.81 & 0.64 & ? fc 145 & 13 00 32 & -37 49 28 & 12.16 & 0.44 & ? fc 146 & 12 32 32 & -38 31 40 & 11.72 & 0.37 & ? fc 147 & 12 32 38 & -38 34 39 & 11.77 & 0.37 & ? fc 148 & 12 50 20 & -38 29 37 & 13.22 & 0.80 & 3514 fc 149 & 12 36 00 & -38 38 02 & 11.62 & 0.16 & ? fc 150 & 12 56 00 & -38 39 57 & 15.18 & 3.90 & ? fc 151 & 12 53 32 & -38 47 20 & 13.23 & 0.76 & 3544 fc 152 & 12 44 09 & -38 58 16 & 13.96 & 1.38 & ? fc 153 & 12 49 39 & -39 01 08 & 11.62 & 0.16 & ? fc 154 & 12 44 42 & -39 06 17 & 13.55 & 0.86 & ? fc 155 & 12 40 10 & -39 22 23 & 11.77 & 0.33 & 3344 fc 156 & 12 41 23 & -39 25 15 & 12.36 & 0.45 & ? fc 157 & 13 00 02 & -39 14 40 & 12.64 & 0.28 & 3621 fc 158 & 12 54 19 & -39 19 14 & 12.19 & 0.36 & 3560 fc 159 & 12 43 12 & -39 31 42 & 12.25 & 0.40 & 3402 fc 160 & 12 47 27 & -39 30 35 & 13.94 & 0.68 & ? fc 161 & 12 47 27 & -39 32 58 & 13.32 & 0.45 & ? fc 162 & 12 30 26 & -39 40 36 & 12.49 & 0.71 & ? fc 163 & 12 45 13 & -39 37 46 & 14.01 & 1.42 & ? fc 164 & 12 33 23 & -39 48 36 & 11.62 & 0.16 & ? fc 165 & 12 47 33 & -39 47 24 & 11.51 & 0.27 & 3465 fc 166 & 12 49 37 & -39 54 43 & 12.85 & 0.48 & ? fc 167 & 12 34 38 & -40 01 57 & 11.21 & 0.32 & ? fc 168 & 12 47 21 & -40 06 59 & 13.81 & 1.32 & ? fc 169 & 13 09 18 & -39 51 04 & 16.14 & 0.16 & ? fc 170 & 12 30 34 & -40 16 59 & 12.41 & 0.37 & ? fc 171 & 12 45 27 & -40 18 07 & 13.81 & 1.09 & ? fc 172 & 12 47 53 & -40 22 24 & 11.62 & 0.73 & ? fc 173 & 12 38 12 & -40 29 26 & 11.30 & 0.30 & ? fc 174 & 12 56 36 & -40 21 02 & 12.97 & 0.58 & ? fc 175 & 12 59 20 & -40 25 47 & 10.04 & 0.19 & 3612 fc 176 & 12 58 18 & -40 36 17 & 14.97 & 1.65 & ? fc 177 & 12 36 57 & -40 50 55 & 13.29 & 0.82 & ? fc 178 & 13 01 46 & -40 36 22 & 12.02 & 0.32 & 3642 fc 179 & 13 04 42 & -40 32 08 & 13.64 & 0.54 & ? fc 180 & 13 07 47 & -40 30 09 & 11.62 & 0.16 & ? fc 181 & 13 11 23 & -40 28 46 & 12.16 & 0.55 & ? fc 182 & 12 38 51 & -41 01 41 & 13.65 & 1.08 & ? fc 183 & 12 33 56 & -41 02 06 & 11.62 & 0.16 & ? fc 184 & 12 32 04 & -41 04 27 & 15.87 & 9.24 & 3213 fc 185 & 12 50 00 & -41 05 35 & 11.62 & 0.16 & ? fc 186 & 12 59 17 & -40 59 13 & 13.68 & 0.61 & ? fc 187 & 12 49 05 & -41 11 57 & 12.92 & 0.57 & ? fc 188 & 13 05 57 & -40 56 55 & 13.97 & 1.38 & ? fc 189 & 12 41 45 & -41 17 18 & 12.78 & 0.59 & ? fc 190 & 12 38 17 & -41 22 22 & 14.59 & 1.11 & ? fc 191 & 12 30 06 & -41 24 33 & 12.02 & 0.44 & ? fc 192 & 12 58 50 & -41 10 32 & 13.02 & 0.59 & ? fc 193 & 12 53 41 & -41 18 59 & 15.09 & 3.51 & 3550 fc 194 & 12 33 02 & -41 38 48 & 17.55 & 0.16 & ? fc 195 & 13 10 44 & -41 19 13 & 12.03 & 0.38 & ? fc 196 & 12 45 48 & -41 49 08 & 14.74 & 2.19 & ? fc 197 & 12 52 02 & -41 52 05 & 13.06 & 0.39 & ? fc 198 & 12 52 36 & -41 56 33 & 12.85 & 0.38 & ? fc 199 & 12 52 46 & -41 58 17 & 13.02 & 0.42 & ? fc 200 & 12 53 18 & -42 00 55 & 12.61 & 0.32 & ? fc 201 & 12 42 43 & -42 10 01 & 13.14 & 0.70 & ? fc 202 & 13 08 17 & -41 57 20 & 13.00 & 0.49 & ? fc 203 & 12 42 01 & -42 17 44 & 14.13 & 1.69 & 3383 fc 204 & 13 01 25 & -42 07 19 & 13.80 & 0.62 & ? fc 205 & 13 04 28 & -42 06 35 & 11.62 & 0.16 & ? fc 206 & 12 49 14 & -42 27 49 & 16.28 & 9.22 & ? fc 207 & 12 46 01 & -42 31 38 & 15.17 & 3.01 & ? fc 208 & 12 39 02 & -42 34 49 & 11.62 & 0.16 & ? fc 209 & 12 52 11 & -42 31 26 & 12.81 & 0.64 & ? fc 210 & 12 35 17 & -42 38 58 & 12.48 & 0.53 & 3264 fc 211 & 12 49 07 & -42 38 11 & 13.50 & 0.83 & 3495 fc 212 & 12 51 22 & -42 43 03 & 12.66 & 0.48 & ? fc 213 & 12 59 42 & -42 35 02 & 13.16 & 0.44 & ? fc 214 & 12 44 16 & -42 46 34 & 13.98 & 1.30 & 3418 fc 215 & 13 08 51 & -42 27 57 & 13.66 & 1.17 & 3688 fc 216 & 12 47 52 & -42 48 46 & 14.45 & 2.25 & ? fc 217 & 12 44 22 & -43 02 09 & 12.64 & 0.50 & ? fc 218 & 13 10 02 & -42 41 48 & 12.14 & 0.46 & ? fc 219 & 12 41 21 & -43 15 57 & 13.56 & 1.15 & ? fc 220 & 13 04 17 & -43 02 40 & 14.07 & 0.91 & ? fc 221 & 13 01 51 & -43 05 11 & 14.34 & 1.69 & ? fc 222 & 12 39 00 & -43 25 44 & 12.54 & 0.49 & 3312 fc 223 & 13 04 29 & -43 08 36 & 13.27 & 0.54 & ? fc 224 & 12 44 43 & -43 25 14 & 11.62 & 0.16 & ? fc 225 & 12 57 47 & -43 17 47 & 12.78 & 0.35 & 3591 fc 226 & 12 57 04 & -43 19 39 & 12.98 & 0.43 & ? fc 227 & 12 58 53 & -43 27 14 & 11.62 & 1.01 & ? fc 228 & 13 00 32 & -43 25 24 & 13.44 & 0.49 & ? fc 229 & 13 10 05 & -43 16 00 & 11.92 & 0.46 & ? fc 230 & 13 04 32 & -43 23 45 & 12.77 & 0.59 & ? fc 231 & 13 02 55 & -43 31 37 & 13.48 & 0.63 & ? fc 232 & 13 07 20 & -43 35 40 & 13.25 & 1.03 & ? fc 233 & 13 02 19 & -43 40 23 & 14.42 & 2.37 & ? fc 234 & 13 08 22 & -43 40 12 & 14.24 & 3.01 & ? fc 235 & 12 59 06 & -43 51 13 & 12.74 & 0.34 & 3609 fc 236 & 12 51 07 & -44 12 28 & 13.62 & 1.21 & ? fc 237 & 12 58 52 & -44 18 45 & 15.73 & 7.63 & ? fc 238 & 13 00 44 & -44 15 45 & 12.58 & 0.61 & ? fc 239 & 12 48 57 & -44 28 34 & 12.98 & 0.46 & ? fc 240 & 13 03 53 & -44 20 17 & 12.57 & 0.42 & ? fc 241 & 13 03 58 & -44 22 34 & 12.86 & 0.52 & ? fc 242 & 12 59 55 & -44 27 30 & 11.71 & 0.40 & ? fc 243 & 12 51 09 & -44 51 50 & 11.84 & 0.38 & ? fc 244 & 12 57 28 & -45 07 36 & 12.00 & 0.48 & ? fc 245 & 12 57 23 & -45 26 17 & 12.69 & 1.26 & ? fc 246 & 12 58 50 & -40 27 18 & 11.23 & 0.21 & ? fc 247 & 12 51 41 & -41 18 49 & 9.52 & 0.17 & 3522 fc 248 & 12 52 43 & -41 11 23 & 12.18 & 0.27 & 3541 fc 249 & 12 52 10 & -43 02 07 & 12.10 & 0.27 & 3531 fd 1 & 3 48 31 & -54 16 44 & 9.22 & 0.20 & 497 fd 2 & 3 52 58 & -54 05 47 & 9.36 & 0.19 & 499 fd 3 & 3 53 27 & -54 50 39 & 6.13 & 0.16 & 500 fd 4 & 4 03 38 & -52 55 38 & 10.87 & 0.24 & 516 fd 5 & 3 48 06 & -56 41 44 & 12.74 & 0.93 & 496 fd 6 & 3 59 59 & -54 10 23 & 12.25 & 0.43 & 509 fd 7 & 3 57 09 & -55 05 33 & 12.32 & 0.49 & 505 fd 8 & 4 00 59 & -54 23 16 & 8.82 & 0.17 & 510 fd 9 & 3 51 37 & -56 39 25 & 9.56 & 0.19 & 498 fd 10 & 4 03 16 & -54 39 51 & 11.66 & 0.30 & 514 fd 11 & 3 56 30 & -56 19 37 & 11.72 & 0.33 & 504 fd 12 & 3 57 42 & -56 15 16 & 11.17 & 0.26 & 506 fd 13 & 4 03 49 & -55 24 23 & 9.14 & 0.18 & 518 fd 14 & 4 15 23 & -52 09 32 & 9.33 & 0.19 & 534 fd 15 & 3 59 35 & -56 28 21 & 9.46 & 0.18 & 508 fd 16 & 4 10 25 & -53 34 56 & 12.27 & 0.36 & 525 fd 17 & 4 01 12 & -56 19 48 & 8.91 & 0.17 & 512 fd 18 & 3 58 37 & -57 06 20 & 10.56 & 0.22 & 507 fd 19 & 4 11 07 & -54 03 12 & 12.73 & 0.56 & 528 fd 20 & 4 11 43 & -53 51 17 & 12.93 & 0.68 & ? fd 21 & 4 13 41 & -53 26 35 & 11.26 & 0.25 & 531 fd 22 & 3 55 52 & -58 19 19 & 11.99 & 0.46 & 503 fd 23 & 4 18 36 & -52 52 09 & 11.27 & 0.26 & 542 fd 24 & 4 09 24 & -55 44 52 & 12.53 & 0.53 & 524 fd 25 & 4 02 39 & -57 27 48 & 11.38 & 0.30 & 513 fd 26 & 4 06 34 & -56 34 50 & 10.94 & 0.25 & 522 fd 27 & 4 16 05 & -54 01 25 & 11.93 & 0.36 & 537 fd 28 & 4 04 25 & -57 09 57 & 11.89 & 0.33 & 519 fd 29 & 4 15 26 & -54 22 43 & 10.60 & 0.21 & 536 fd 30 & 4 19 22 & -53 19 25 & 12.26 & 0.39 & 545 fd 31 & 4 00 53 & -58 39 30 & 7.76 & 0.17 & 511 fd 32 & 4 18 46 & -54 08 43 & 7.49 & 0.16 & 543 fd 33 & 4 19 05 & -53 51 01 & 12.45 & 0.45 & 544 fd 34 & 4 19 59 & -54 04 45 & 10.53 & 0.21 & 548 fd 35 & 4 17 09 & -54 58 14 & 11.30 & 0.24 & 539 fd 36 & 4 04 58 & -58 09 44 & 12.59 & 0.55 & 520 fd 37 & 4 22 54 & -53 32 18 & 12.57 & 0.43 & 552 fd 38 & 4 05 35 & -58 28 31 & 9.36 & 0.18 & 521 fd 39 & 4 14 07 & -56 11 16 & 12.14 & 0.42 & 533 fd 40 & 4 25 31 & -53 06 01 & 10.40 & 0.23 & 555 fd 41 & 4 19 57 & -54 56 46 & 9.68 & 0.19 & 547 fd 42 & 4 22 34 & -54 09 18 & 10.09 & 0.19 & 551 fd 43 & 4 08 49 & -58 29 04 & 9.56 & 0.18 & 523 fd 44 & 4 23 32 & -54 17 51 & 13.01 & 0.70 & 553 fd 45 & 4 27 24 & -53 23 44 & 11.06 & 0.28 & 562 fd 46 & 4 28 30 & -53 27 15 & 10.10 & 0.20 & 565 fd 47 & 4 17 30 & -56 44 23 & 12.74 & 0.52 & 540 fd 48 & 4 13 40 & -58 01 10 & 9.14 & 0.18 & 532 fd 49 & 4 29 39 & -53 13 03 & 11.62 & 0.16 & ? fd 50 & 4 11 39 & -58 41 36 & 12.30 & 0.44 & 529 fd 51 & 4 25 48 & -55 07 15 & 11.90 & 0.35 & 556 fd 52 & 4 31 12 & -53 35 18 & 11.14 & 0.31 & 567 fd 53 & 4 27 55 & -55 04 09 & 12.29 & 0.48 & 564 fd 54 & 4 22 06 & -56 59 34 & 11.42 & 0.27 & 550 fd 55 & 4 21 19 & -57 32 32 & 12.69 & 0.59 & 549 fd 56 & 4 27 03 & -56 05 19 & 10.12 & 0.20 & 561 fd 57 & 4 24 06 & -57 15 42 & 7.71 & 0.16 & 554 fd 58 & 4 16 39 & -59 17 34 & 9.74 & 0.19 & 538 fd 59 & 4 26 35 & -56 27 53 & 12.15 & 0.37 & 559 fd 60 & 4 19 13 & -58 44 49 & 8.75 & 0.17 & 546 fd 61 & 4 36 18 & -53 42 58 & 8.15 & 0.17 & 572 fd 62 & 4 25 56 & -57 12 13 & 12.03 & 0.36 & 557 fd 63 & 4 34 20 & -55 00 41 & 5.03 & 0.16 & 570 fd 64 & 4 31 08 & -55 55 31 & 11.35 & 0.32 & 566 fd 65 & 4 32 43 & -55 35 23 & 10.28 & 0.21 & 569 fd 66 & 4 36 33 & -54 37 25 & 9.92 & 0.24 & 576 fd 67 & 4 26 33 & -59 06 07 & 10.51 & 0.23 & 560 fd 68 & 4 34 48 & -56 36 07 & 11.51 & 0.29 & 571 fd 69 & 4 36 06 & -56 17 32 & 12.22 & 0.49 & 573 fd 70 & 4 40 50 & -56 02 32 & 10.58 & 0.24 & 584 fd 71 & 4 40 20 & -56 11 39 & 12.29 & 0.32 & 583 fd 72 & 4 37 03 & -57 28 29 & 12.27 & 0.50 & 578 fd 73 & 4 39 21 & -57 19 34 & 11.03 & 0.25 & 582 fd 74 & 4 36 10 & -58 36 40 & 12.21 & 0.56 & 575 fd 75 & 4 42 39 & -56 41 59 & 11.77 & 0.33 & 587 fd 76 & 4 38 51 & -58 58 05 & 8.83 & 0.18 & 580 fd 77 & 4 40 14 & -58 44 35 & 11.64 & 0.39 & ? fd 78 & 4 03 33 & -52 28 53 & 11.07 & 0.29 & 515 fd 79 & 3 55 31 & -53 29 43 & 13.68 & 1.56 & ? fd 80 & 3 57 34 & -54 20 33 & 13.39 & 0.94 & ? fd 81 & 3 48 33 & -56 32 11 & 14.59 & 3.76 & ? fd 82 & 4 06 00 & -52 39 49 & 12.62 & 0.57 & ? fd 83 & 3 58 19 & -54 51 39 & 12.38 & 0.39 & ? fd 84 & 3 51 34 & -56 23 23 & 12.24 & 0.56 & ? fd 85 & 4 09 29 & -52 28 06 & 12.69 & 0.63 & ? fd 86 & 4 04 01 & -54 04 22 & 12.90 & 0.67 & ? fd 87 & 4 13 30 & -52 24 19 & 13.56 & 1.12 & ? fd 88 & 4 00 14 & -56 05 14 & 11.62 & 0.16 & ? fd 89 & 4 06 09 & -54 35 51 & 11.62 & 0.16 & ? fd 90 & 4 13 23 & -52 53 27 & 13.71 & 1.18 & 530 fd 91 & 3 55 11 & -57 45 16 & 13.22 & 1.01 & ? fd 92 & 4 03 47 & -56 01 00 & 13.17 & 0.89 & ? fd 93 & 4 06 59 & -55 19 29 & 12.89 & 0.68 & ? fd 94 & 4 12 55 & -54 01 46 & 13.64 & 0.97 & ? fd 95 & 4 08 50 & -55 37 09 & 11.62 & 0.16 & ? fd 96 & 4 03 03 & -57 41 48 & 13.50 & 0.98 & ? fd 97 & 4 10 56 & -56 28 58 & 12.92 & 0.58 & 527 fd 98 & 4 00 56 & -58 55 36 & 11.44 & 0.32 & ? fd 99 & 4 03 41 & -58 57 02 & 12.94 & 0.83 & ? fd 100 & 4 26 31 & -53 09 01 & 11.73 & 0.37 & 558 fd 101 & 4 10 21 & -57 47 14 & 12.89 & 0.69 & 526 fd 102 & 4 14 18 & -57 02 59 & 14.31 & 1.89 & ? fd 103 & 4 29 47 & -52 58 52 & 13.13 & 0.95 & ? fd 104 & 4 17 12 & -57 01 21 & 12.93 & 0.64 & ? fd 105 & 4 26 43 & -54 16 53 & 13.17 & 0.81 & ? fd 106 & 4 23 34 & -55 53 51 & 13.36 & 0.90 & ? fd 107 & 4 11 47 & -59 08 38 & 12.61 & 0.52 & ? fd 108 & 4 31 54 & -53 29 58 & 13.93 & 1.01 & ? fd 109 & 4 17 15 & -58 11 30 & 12.45 & 0.52 & 541 fd 110 & 4 33 51 & -53 09 36 & 14.89 & 4.19 & ? fd 111 & 4 32 00 & -53 46 44 & 14.16 & 2.76 & ? fd 112 & 4 26 03 & -56 20 39 & 13.21 & 0.41 & ? fd 113 & 4 29 38 & -57 10 21 & 13.23 & 0.81 & ? fd 114 & 4 28 51 & -57 36 39 & 13.02 & 0.70 & ? fd 115 & 4 40 48 & -54 04 44 & 13.03 & 1.01 & ? fd 116 & 4 37 31 & -55 41 40 & 11.28 & 0.34 & ? fd 117 & 4 30 36 & -58 10 39 & 12.07 & 0.42 & ? fd 118 & 4 37 41 & -55 56 25 & 12.19 & 0.60 & ? fd 119 & 4 39 14 & -55 38 19 & 14.35 & 3.66 & ? fd 120 & 4 33 05 & -57 32 35 & 13.22 & 0.70 & ? fd 121 & 4 37 34 & -56 20 24 & 11.62 & 0.16 & ? fd 122 & 4 36 53 & -56 38 57 & 11.84 & 0.39 & 577 fd 123 & 4 40 20 & -55 34 29 & 11.62 & 0.16 & ? fd 124 & 4 43 14 & -54 41 49 & 12.18 & 0.58 & ? fd 125 & 4 30 60 & -58 52 52 & 13.05 & 1.05 & ? fd 126 & 4 43 58 & -54 47 09 & 12.73 & 0.86 & ? fd 127 & 4 36 11 & -57 52 50 & 15.23 & 5.16 & ? fd 128 & 4 43 22 & -55 46 25 & 13.01 & 0.75 & ? fd 129 & 4 33 51 & -58 51 57 & 14.51 & 3.81 & ? fd 130 & 4 42 19 & -56 07 32 & 12.83 & 0.73 & 586 fd 131 & 4 45 20 & -55 06 26 & 13.33 & 1.40 & ? fd 132 & 4 35 31 & -58 28 25 & 14.37 & 2.75 & ? fd 133 & 4 40 38 & -56 54 25 & 14.25 & 1.51 & ? fd 134 & 4 43 47 & -56 46 10 & 12.69 & 0.59 & ? fd 135 & 4 44 51 & -56 42 02 & 12.23 & 0.45 & ? fd 136 & 4 40 46 & -58 29 47 & 12.19 & 0.53 & ? fd 137 & 4 46 02 & -56 54 28 & 14.87 & 4.03 & ? fd 138 & 4 03 41 & -52 11 25 & 9.11 & 0.18 & 517 fd 139 & 3 56 01 & -52 51 56 & 11.07 & 0.35 & 502 fd 140 & 3 47 06 & -58 06 32 & 7.79 & 0.17 & 495 fd 141 & 4 31 20 & -59 45 28 & 8.92 & 0.19 & 568 fd 142 & 3 55 17 & -54 52 36 & 11.55 & 0.25 & 501 fm 1 & 13 55 36 & -28 59 15 & 12.40 & 0.50 & 3942/3 fm 2 & 13 53 60 & -28 09 49 & 7.70 & 0.17 & 3938 fm 3 & 13 53 60 & -27 56 29 & 10.24 & 0.21 & 3937 fm 4 & 13 53 42 & -27 07 16 & 11.35 & 0.33 & 3935 fm 5 & 13 53 14 & -27 20 60 & 11.67 & 0.35 & 3934 fm 6 & 13 52 56 & -27 41 09 & 10.87 & 0.24 & 3930 fm 7 & 13 53 54 & -30 01 20 & 9.29 & 0.18 & 3936 fm 8 & 13 52 37 & -27 31 43 & 9.37 & 0.18 & 3929 fm 9 & 13 52 18 & -26 48 08 & 11.35 & 0.30 & 3928 fm 10 & 13 52 17 & -30 47 45 & 9.17 & 0.18 & 3927 fm 11 & 13 51 23 & -29 07 27 & 11.35 & 0.25 & 3923 fm 12 & 13 51 52 & -30 48 28 & 9.16 & 0.18 & 3924 fm 13 & 13 50 03 & -29 05 48 & 5.92 & 0.16 & 3915 fm 14 & 13 51 22 & -30 48 10 & 12.36 & 0.34 & ? fm 15 & 13 49 53 & -28 11 48 & 10.30 & 0.20 & 3914 fm 16 & 13 49 11 & -26 55 26 & 11.13 & 0.25 & 3911 fm 17 & 13 50 39 & -30 15 24 & 12.21 & 0.38 & 3917 fm 18 & 13 49 46 & -28 38 55 & 11.77 & 0.31 & 3913 fm 19 & 13 49 12 & -27 25 49 & 11.61 & 0.29 & 3912 fm 20 & 13 50 40 & -31 12 27 & 9.02 & 0.18 & 3918 fm 21 & 13 50 21 & -31 03 57 & 9.58 & 0.18 & 3916 fm 22 & 13 49 11 & -28 35 27 & 12.67 & 0.32 & 3910 fm 23 & 13 48 27 & -27 56 22 & 13.13 & 0.70 & 3907 fm 24 & 13 48 06 & -29 19 50 & 12.59 & 0.46 & 3905 fm 25 & 13 45 38 & -26 06 26 & 5.51 & 0.16 & 3895 fm 26 & 13 47 51 & -31 32 40 & 11.40 & 0.30 & 3903 fm 27 & 13 46 27 & -28 02 03 & 12.25 & 0.35 & 3898 fm 28 & 13 47 23 & -30 27 02 & 12.57 & 0.47 & 3902 fm 29 & 13 46 57 & -29 41 34 & 13.35 & 0.70 & 3900 fm 30 & 13 45 33 & -27 51 33 & 12.17 & 0.32 & 3894 fm 31 & 13 45 11 & -27 43 07 & 11.44 & 0.25 & 3892 fm 32 & 13 46 27 & -30 26 01 & 13.52 & 0.85 & ? fm 33 & 13 46 06 & -30 12 11 & 11.89 & 0.30 & 3896 fm 34 & 13 44 38 & -26 46 33 & 10.63 & 0.20 & 3891 fm 35 & 13 46 24 & -31 45 34 & 12.60 & 0.54 & 3897 fm 36 & 13 45 16 & -29 01 15 & 12.80 & 0.52 & 3893 fm 37 & 13 43 52 & -26 47 37 & 12.22 & 0.37 & 3880 fm 38 & 13 44 10 & -28 16 50 & 12.86 & 0.53 & 3886 fm 39 & 13 43 53 & -27 46 10 & 11.85 & 0.29 & 3882 fm 40 & 13 43 54 & -28 42 03 & 12.75 & 0.46 & 3881/3 fm 41 & 13 44 25 & -30 07 07 & 10.94 & 0.21 & 3888 fm 42 & 13 44 10 & -31 49 07 & 11.81 & 0.32 & 3885 fm 43 & 13 43 00 & -30 10 02 & 9.71 & 0.18 & 3877 fm 44 & 13 43 28 & -31 26 27 & 12.11 & 0.35 & 3879 fm 45 & 13 41 31 & -27 16 16 & 9.93 & 0.19 & 3871 fm 46 & 13 42 15 & -29 45 20 & 10.74 & 0.21 & 3874 fm 47 & 13 41 42 & -28 45 09 & 12.31 & 0.26 & 3872 fm 48 & 13 42 32 & -31 02 37 & 12.32 & 0.35 & 3875 fm 49 & 13 41 42 & -29 53 43 & 11.14 & 0.23 & 3873 fm 50 & 13 40 47 & -28 14 29 & 12.01 & 0.34 & 3866 fm 51 & 13 40 53 & -28 57 18 & 12.40 & 0.42 & 3867 fm 52 & 13 39 54 & -27 08 47 & 10.00 & 0.19 & 3863 fm 53 & 13 41 13 & -31 34 28 & 12.43 & 0.27 & 3868 fm 54 & 13 39 42 & -28 08 13 & 10.71 & 0.21 & 3860 fm 55 & 13 40 14 & -29 33 34 & 12.02 & 0.32 & 3865 fm 56 & 13 38 44 & -27 15 36 & 9.64 & 0.18 & 3853 fm 57 & 13 38 31 & -26 54 19 & 9.63 & 0.18 & 3850 fm 58 & 13 39 57 & -31 38 05 & 9.89 & 0.19 & 3862 fm 59 & 13 39 10 & -30 14 09 & 12.44 & 0.37 & 3856 fm 60 & 13 38 42 & -29 32 48 & 9.95 & 0.18 & 3851 fm 61 & 13 39 05 & -30 44 04 & 12.34 & 0.34 & 3855 fm 62 & 13 39 44 & -32 29 31 & 12.38 & 0.46 & 3861 fm 63 & 13 37 35 & -27 03 44 & 10.86 & 0.22 & 3848 fm 64 & 13 38 23 & -29 49 40 & 11.02 & 0.23 & 3849 fm 65 & 13 36 48 & -26 30 13 & 7.37 & 0.16 & 3838 fm 66 & 13 37 14 & -26 42 39 & 12.18 & 0.35 & 3845 fm 67 & 13 38 46 & -31 50 34 & 11.74 & 0.31 & 3852 fm 68 & 13 37 04 & -27 39 32 & 12.87 & 0.57 & 3843 fm 69 & 13 37 03 & -27 54 14 & 12.59 & 0.46 & 3842 fm 70 & 13 37 00 & -29 50 41 & 7.92 & 0.17 & 3840 fm 71 & 13 35 53 & -25 56 39 & 12.21 & 0.38 & 3833 fm 72 & 13 37 36 & -32 08 53 & 11.28 & 0.24 & 3847 fm 73 & 13 36 29 & -29 07 03 & 12.25 & 0.35 & 3835 fm 74 & 13 36 06 & -28 41 47 & 11.86 & 0.29 & 3834 fm 75 & 13 36 41 & -30 17 51 & 13.03 & 0.56 & 3836 fm 76 & 13 36 59 & -32 18 02 & 12.28 & 0.28 & 3839 fm 77 & 13 36 49 & -32 24 45 & 10.66 & 0.22 & 3837 fm 78 & 13 35 01 & -26 07 13 & 12.67 & 0.51 & 3824 fm 79 & 13 35 31 & -29 08 32 & 8.42 & 0.17 & 3827 fm 80 & 13 35 42 & -29 21 49 & 11.48 & 0.25 & 3830 fm 81 & 13 34 12 & -25 25 52 & 8.86 & 0.17 & 3818 fm 82 & 13 35 39 & -31 25 04 & 12.52 & 0.26 & 3828 fm 83 & 13 35 40 & -32 01 15 & 10.73 & 0.21 & 3829 fm 84 & 13 34 13 & -27 55 12 & 11.99 & 0.34 & 3819 fm 85 & 13 35 15 & -32 13 16 & 9.82 & 0.19 & 3826 fm 86 & 13 34 47 & -31 11 39 & 12.80 & 0.48 & 3821 fm 87 & 13 32 54 & -26 36 00 & 9.77 & 0.18 & 3811 fm 88 & 13 33 10 & -29 17 04 & 10.54 & 0.18 & 3815 fm 89 & 13 32 34 & -28 41 37 & 5.47 & 0.16 & 3806 fm 90 & 13 33 21 & -31 39 08 & 10.08 & 0.18 & 3816 fm 91 & 13 32 36 & -29 17 04 & 9.91 & 0.17 & 3805 fm 92 & 13 31 49 & -26 09 20 & 12.10 & 0.37 & 3803 fm 93 & 13 32 37 & -29 33 07 & 11.41 & 0.25 & 3808 fm 94 & 13 33 09 & -31 47 29 & 10.83 & 0.21 & 3812 fm 95 & 13 31 33 & -28 06 44 & 6.69 & 0.16 & 3801 fm 96 & 13 31 01 & -25 42 36 & 9.06 & 0.18 & 3796 fm 97 & 13 31 22 & -27 32 14 & 11.69 & 0.29 & 3798 fm 98 & 13 32 45 & -32 46 54 & 12.04 & 0.39 & 3810 fm 99 & 13 31 48 & -29 42 25 & 12.26 & 0.39 & 3802 fm 100 & 13 32 12 & -32 06 22 & 12.16 & 0.25 & 3804 fm 101 & 13 31 36 & -32 08 57 & 10.86 & 0.22 & 3799 fm 102 & 13 30 02 & -27 02 52 & 10.71 & 0.22 & 3792 fm 103 & 13 29 48 & -26 35 11 & 8.56 & 0.17 & 3790 fm 104 & 13 30 57 & -31 44 42 & 11.77 & 0.29 & 3794 fm 105 & 13 29 33 & -27 23 39 & 10.23 & 0.19 & 3789 fm 106 & 13 29 54 & -28 56 11 & 10.00 & 0.18 & 3791 fm 107 & 13 29 34 & -29 04 58 & 11.08 & 0.19 & 3788 fm 108 & 13 28 17 & -26 34 47 & 11.17 & 0.24 & 3780 fm 109 & 13 28 42 & -31 59 43 & 11.69 & 0.31 & 3783 fm 110 & 13 28 01 & -29 46 48 & 12.83 & 0.49 & 3779 fm 111 & 13 27 12 & -29 03 28 & 11.15 & 0.22 & 3777 fm 112 & 13 26 51 & -29 51 53 & 12.87 & 0.50 & 3775 fm 113 & 13 26 03 & -27 28 15 & 11.62 & 0.27 & 3771 fm 114 & 13 25 60 & -30 23 11 & 11.71 & 0.20 & 3769 fm 115 & 13 25 14 & -30 22 12 & 8.68 & 0.17 & 3765 fm 116 & 13 24 58 & -31 20 15 & 11.17 & 0.25 & 3764 fm 117 & 13 23 51 & -26 53 60 & 11.05 & 0.24 & 3761 fm 118 & 13 22 43 & -28 47 03 & 11.08 & 0.23 & 3758 fm 119 & 13 22 45 & -30 11 08 & 11.43 & 0.24 & 3757 fm 120 & 13 21 58 & -28 26 52 & 12.02 & 0.34 & 3754 fm 121 & 13 20 46 & -29 29 29 & 11.11 & 0.24 & 3745 fm 122 & 13 19 39 & -30 20 40 & 6.21 & 0.16 & 3737 fm 123 & 13 18 02 & -27 36 19 & 11.28 & 0.33 & 3730 fm 124 & 13 55 02 & -29 07 28 & 12.83 & 0.69 & ? fm 125 & 13 51 24 & -27 24 05 & 14.75 & 2.96 & 3922 fm 126 & 13 51 18 & -28 06 27 & 12.22 & 0.43 & 3921/0 fm 127 & 13 51 10 & -29 20 03 & 11.67 & 0.28 & 3919 fm 128 & 13 49 02 & -28 17 46 & 11.73 & 0.29 & 3909 fm 129 & 13 49 35 & -30 02 06 & 14.11 & 1.11 & ? fm 130 & 13 48 05 & -27 16 56 & 12.45 & 0.46 & 3904 fm 131 & 13 48 25 & -29 06 48 & 12.98 & 0.55 & ? fm 132 & 13 48 49 & -30 12 30 & 13.62 & 0.85 & ? fm 133 & 13 47 38 & -28 55 02 & 14.30 & 1.64 & ? fm 134 & 13 47 11 & -30 46 12 & 12.87 & 0.53 & 3901 fm 135 & 13 45 53 & -28 34 06 & 14.57 & 1.65 & ? fm 136 & 13 44 34 & -26 18 22 & 11.46 & 0.29 & 3889 fm 137 & 13 43 39 & -26 29 55 & 13.11 & 0.71 & ? fm 138 & 13 44 40 & -29 39 34 & 13.69 & 0.88 & ? fm 139 & 13 44 14 & -28 48 53 & 13.39 & 0.42 & 3887 fm 140 & 13 45 17 & -31 22 33 & 13.48 & 0.83 & ? fm 141 & 13 43 21 & -28 55 36 & 13.50 & 0.46 & 3878 fm 142 & 13 43 18 & -29 00 16 & 14.07 & 0.71 & ? fm 143 & 13 42 59 & -29 07 15 & 13.80 & 0.56 & ? fm 144 & 13 42 11 & -28 28 59 & 16.06 & 3.81 & ? fm 145 & 13 41 13 & -27 29 26 & 14.53 & 1.73 & 3869 fm 146 & 13 41 56 & -29 28 49 & 14.33 & 0.80 & ? fm 147 & 13 41 25 & -29 06 22 & 14.45 & 0.89 & ? fm 148 & 13 40 27 & -27 43 12 & 13.69 & 1.00 & ? fm 149 & 13 41 59 & -32 09 29 & 12.59 & 0.51 & ? fm 150 & 13 41 23 & -30 52 59 & 12.78 & 0.44 & ? fm 151 & 13 41 43 & -31 37 13 & 13.47 & 0.45 & ? fm 152 & 13 39 30 & -26 40 11 & 13.04 & 0.60 & 3859 fm 153 & 13 40 59 & -31 14 31 & 13.44 & 0.76 & ? fm 154 & 13 40 15 & -29 53 12 & 14.01 & 1.10 & 3864 fm 155 & 13 40 34 & -32 17 11 & 13.79 & 1.22 & ? fm 156 & 13 39 13 & -29 05 45 & 14.07 & 1.18 & ? fm 157 & 13 39 28 & -29 54 45 & 11.62 & 0.16 & ? fm 158 & 13 36 57 & -25 54 28 & 12.26 & 0.38 & 3841 fm 159 & 13 38 00 & -31 06 14 & 13.22 & 0.43 & ? fm 160 & 13 38 01 & -31 10 40 & 13.15 & 0.36 & ? fm 161 & 13 37 40 & -32 21 50 & 12.50 & 0.47 & 3846 fm 162 & 13 37 11 & -31 29 23 & 13.78 & 0.87 & ? fm 163 & 13 35 50 & -30 51 28 & 13.74 & 0.58 & 3831 fm 164 & 13 33 21 & -27 42 04 & 13.02 & 0.51 & 3817 fm 165 & 13 32 10 & -30 44 25 & 13.86 & 0.84 & ? fm 166 & 13 30 34 & -25 29 15 & 14.32 & 1.11 & ? fm 167 & 13 30 56 & -30 18 29 & 13.11 & 0.50 & 3795 fm 168 & 13 30 41 & -30 47 08 & 12.93 & 0.53 & ? fm 169 & 13 29 00 & -25 36 03 & 12.49 & 0.65 & 3785 fm 170 & 13 29 20 & -28 27 46 & 12.74 & 0.50 & 3787 fm 171 & 13 29 40 & -29 36 24 & 16.34 & 7.80 & ? fm 172 & 13 29 06 & -31 02 01 & 13.73 & 0.90 & ? fm 173 & 13 28 46 & -29 16 29 & 13.41 & 0.73 & 3784 fm 174 & 13 27 48 & -30 46 28 & 13.40 & 0.66 & 3778 fm 175 & 13 27 41 & -31 31 45 & 13.91 & 0.56 & ? fm 176 & 13 27 07 & -30 04 33 & 13.63 & 0.79 & 3776 fm 177 & 13 26 20 & -26 51 42 & 13.01 & 0.56 & ? fm 178 & 13 27 06 & -31 30 15 & 14.23 & 0.77 & ? fm 179 & 13 26 48 & -32 28 51 & 13.27 & 0.59 & ? fm 180 & 13 25 36 & -27 39 21 & 13.28 & 0.66 & ? fm 181 & 13 25 47 & -29 49 49 & 13.84 & 1.01 & 3767 fm 182 & 13 24 57 & -28 15 51 & 14.52 & 1.88 & ? fm 183 & 13 24 02 & -28 24 56 & 13.94 & 1.19 & 3762 fm 184 & 13 23 17 & -27 05 24 & 12.12 & 0.38 & ? fm 185 & 13 20 29 & -29 52 49 & 13.64 & 1.23 & ? fm 186 & 13 19 53 & -27 48 56 & 13.03 & 0.81 & 3738 fm 187 & 13 17 42 & -28 48 03 & 12.37 & 0.54 & 3728 fm 188 & 13 31 21 & -29 33 53 & 13.31 & 0.40 & ? fm 189 & 13 31 36 & -29 30 21 & 12.77 & 0.29 & 3800 fm 190 & 13 26 07 & -32 33 26 & 10.36 & 0.23 & 3770 fm 191 & 13 48 09 & -32 13 29 & 11.17 & 0.32 & 3906 fm 192 & 13 32 05 & -32 11 40 & 12.50 & 0.30 & ? fm 193 & 13 34 47 & -25 18 49 & 11.91 & 0.28 & 3822 fm 194 & 13 41 19 & -28 45 37 & 12.89 & 0.35 & ? fm 195 & 13 35 20 & -31 24 27 & 13.07 & 0.32 & ? fm 196 & 13 40 10 & -31 31 06 & 13.06 & 0.36 & ? fm 197 & 13 36 21 & -29 41 57 & 13.86 & 0.64 & ? ft 1 & 19 55 30 & -49 27 45 & 10.27 & 0.24 & 4536 ft 2 & 19 53 16 & -48 27 36 & 11.25 & 0.37 & 4532 ft 3 & 20 00 09 & -50 09 49 & 10.09 & 0.24 & 4544 ft 4 & 19 54 51 & -48 31 15 & 11.27 & 0.34 & 4534 ft 5 & 19 55 18 & -48 16 52 & 10.60 & 0.27 & 4535 ft 6 & 20 01 44 & -49 58 38 & 10.16 & 0.23 & 4551 ft 7 & 20 04 50 & -49 56 01 & 11.11 & 0.30 & 4560 ft 8 & 20 00 10 & -48 43 38 & 12.18 & 0.54 & 4546 ft 9 & 19 56 08 & -47 23 08 & 9.34 & 0.19 & 4538 ft 10 & 20 04 46 & -49 44 03 & 11.19 & 0.30 & 4559 ft 11 & 20 03 31 & -49 22 26 & 11.28 & 0.30 & 4555 ft 12 & 19 53 15 & -46 22 23 & 9.35 & 0.20 & 4533 ft 13 & 20 00 01 & -48 11 02 & 11.80 & 0.39 & 4545 ft 14 & 20 05 07 & -49 29 09 & 10.78 & 0.27 & 4562 ft 15 & 20 09 35 & -50 25 49 & 10.43 & 0.25 & 4584 ft 16 & 20 05 59 & -49 30 24 & 11.84 & 0.29 & 4567 ft 17 & 20 07 50 & -49 58 27 & 12.45 & 0.58 & 4572 ft 18 & 19 59 41 & -47 33 41 & 12.26 & 0.54 & ? ft 19 & 20 03 54 & -48 37 00 & 13.09 & 0.91 & ? ft 20 & 19 57 02 & -46 35 36 & 10.35 & 0.22 & 4540 ft 21 & 20 06 16 & -48 59 27 & 9.60 & 0.20 & 4568 ft 22 & 20 04 59 & -48 36 25 & 11.36 & 0.32 & 4561 ft 23 & 20 02 55 & -47 58 16 & 12.14 & 0.44 & 4554 ft 24 & 20 01 26 & -47 23 38 & 10.15 & 0.18 & 4550 ft 25 & 20 12 44 & -50 18 27 & 10.99 & 0.28 & 4597 ft 26 & 20 02 07 & -47 18 50 & 8.69 & 0.17 & 4552 ft 27 & 20 13 54 & -50 19 21 & 8.65 & 0.18 & 4602 ft 28 & 20 05 56 & -48 16 17 & 8.92 & 0.18 & 4566 ft 29 & 20 08 57 & -49 3 54 & 12.30 & 0.53 & 4581 ft 30 & 20 12 53 & -49 58 40 & 11.18 & 0.29 & 4599 ft 31 & 20 07 39 & -48 39 25 & 12.06 & 0.46 & 4570 ft 32 & 20 05 35 & -47 56 03 & 10.08 & 0.20 & 4564 ft 33 & 19 58 60 & -46 06 29 & 11.91 & 0.38 & 4542 ft 34 & 20 09 26 & -48 49 48 & 11.56 & 0.34 & 4583 ft 35 & 20 05 13 & -47 29 42 & 9.47 & 0.19 & 4563 ft 36 & 19 58 41 & -45 36 50 & 9.24 & 0.18 & 4541 ft 37 & 20 18 21 & -50 39 31 & 10.59 & 0.25 & 4609 ft 38 & 20 00 53 & -45 42 34 & 8.04 & 0.17 & 4548 ft 39 & 20 11 25 & -48 22 45 & 10.75 & 0.23 & 4593 ft 40 & 20 08 32 & -47 33 54 & 10.96 & 0.25 & 4578 ft 41 & 20 19 11 & -50 09 55 & 12.04 & 0.50 & 4615 ft 42 & 20 08 09 & -47 11 30 & 9.28 & 0.18 & 4575 ft 43 & 20 00 46 & -45 06 44 & 8.33 & 0.17 & 4547 ft 44 & 20 15 37 & -48 58 12 & 11.91 & 0.38 & 4606 ft 45 & 20 07 35 & -46 51 45 & 12.59 & 0.59 & 4571 ft 46 & 20 21 01 & -50 00 27 & 7.51 & 0.17 & 4619 ft 47 & 20 15 11 & -48 44 17 & 12.28 & 0.48 & ? ft 48 & 20 06 25 & -46 05 07 & 9.26 & 0.18 & 4569 ft 49 & 20 18 09 & -48 57 25 & 12.53 & 0.61 & 4608 ft 50 & 20 02 04 & -44 26 49 & 7.11 & 0.16 & 4553 ft 51 & 20 08 37 & -46 19 58 & 11.59 & 0.31 & 4579 ft 52 & 20 05 43 & -45 34 38 & 11.78 & 0.37 & 4565 ft 53 & 20 22 22 & -49 29 13 & 9.17 & 0.18 & 4625 ft 54 & 20 04 29 & -44 57 08 & 10.44 & 0.22 & 4558 ft 55 & 20 23 35 & -49 46 36 & 12.57 & 0.76 & 4631 ft 56 & 20 03 25 & -44 35 11 & 9.59 & 0.19 & 4556 ft 57 & 20 18 34 & -48 32 33 & 11.51 & 0.33 & 4613 ft 58 & 20 10 30 & -46 25 39 & 12.32 & 0.42 & 4590 ft 59 & 20 23 01 & -49 19 55 & 11.28 & 0.24 & 4628 ft 60 & 20 20 06 & -48 28 29 & 9.02 & 0.18 & 4617 ft 61 & 20 03 46 & -43 49 14 & 9.01 & 0.18 & 4557 ft 62 & 20 13 56 & -46 33 51 & 11.56 & 0.30 & 4603 ft 63 & 20 12 25 & -46 10 46 & 10.62 & 0.21 & 4598 ft 64 & 20 07 47 & -44 55 47 & 12.58 & 0.62 & 4574 ft 65 & 20 16 42 & -47 05 24 & 12.12 & 0.40 & ? ft 66 & 20 19 18 & -47 35 05 & 11.32 & 0.31 & 4616 ft 67 & 20 12 23 & -45 51 35 & 11.14 & 0.26 & 4596 ft 68 & 20 09 40 & -44 46 27 & 12.70 & 0.70 & 4586 ft 69 & 20 08 55 & -44 31 16 & 11.83 & 0.43 & 4582 ft 70 & 20 13 37 & -45 36 15 & 8.02 & 0.17 & 4601 ft 71 & 20 21 57 & -47 43 25 & 10.33 & 0.21 & 4622 ft 72 & 20 23 58 & -48 13 09 & 11.38 & 0.31 & 4632 ft 73 & 20 30 13 & -49 21 49 & 8.83 & 0.18 & 4650 ft 74 & 20 09 38 & -44 19 20 & 12.87 & 0.82 & 4587 ft 75 & 20 18 19 & -46 33 12 & 12.04 & 0.38 & 4611 ft 76 & 20 27 12 & -48 34 42 & 11.38 & 0.34 & 4641 ft 77 & 20 10 27 & -44 08 21 & 11.39 & 0.34 & 4589 ft 78 & 20 26 36 & -47 56 53 & 10.29 & 0.23 & 4639 ft 79 & 20 12 03 & -44 16 19 & 9.84 & 0.20 & 4595 ft 80 & 20 10 46 & -43 36 02 & 10.34 & 0.20 & 4591 ft 81 & 20 26 22 & -47 30 36 & 11.94 & 0.43 & 4638 ft 82 & 20 10 60 & -43 28 51 & 9.81 & 0.19 & 4592 ft 83 & 20 22 29 & -46 27 09 & 12.16 & 0.45 & 4626 ft 84 & 20 12 40 & -43 50 20 & 10.53 & 0.23 & 4600 ft 85 & 20 14 11 & -44 10 37 & 12.69 & 0.74 & 4604 ft 86 & 20 27 00 & -47 15 39 & 10.41 & 0.22 & 4640 ft 87 & 20 29 33 & -47 47 07 & 9.56 & 0.19 & 4645 ft 88 & 20 15 34 & -44 28 42 & 11.57 & 0.33 & ? ft 89 & 20 11 59 & -43 24 30 & 11.60 & 0.43 & 4594 ft 90 & 20 20 45 & -45 41 25 & 11.80 & 0.32 & 4620 ft 91 & 20 26 06 & -46 39 46 & 9.10 & 0.18 & 4636 ft 92 & 20 22 00 & -45 39 05 & 8.86 & 0.17 & 4624 ft 93 & 20 14 58 & -43 39 54 & 9.85 & 0.21 & 4605 ft 94 & 20 28 24 & -46 56 53 & 10.68 & 0.24 & 4643 ft 95 & 20 33 45 & -47 56 28 & 11.65 & 0.39 & 4653 ft 96 & 20 26 05 & -46 10 09 & 11.74 & 0.40 & 4637 ft 97 & 20 29 39 & -46 09 44 & 7.00 & 0.16 & 4647 ft 98 & 20 22 40 & -44 35 38 & 11.14 & 0.31 & 4627 ft 99 & 20 33 58 & -46 49 56 & 10.07 & 0.21 & 4654 ft 100 & 20 29 51 & -45 55 25 & 11.55 & 0.40 & 4649 ft 101 & 20 29 33 & -45 23 16 & 10.17 & 0.21 & 4646 ft 102 & 20 27 16 & -44 45 25 & 11.20 & 0.29 & 4642 ft 103 & 20 23 00 & -43 39 26 & 10.14 & 0.20 & 4630 ft 104 & 20 22 46 & -43 29 04 & 9.18 & 0.20 & 4629 ft 105 & 20 35 48 & -46 37 18 & 10.91 & 0.32 & 4655 ft 106 & 20 23 53 & -43 36 56 & 11.09 & 0.27 & 4633 ft 107 & 20 29 25 & -44 57 48 & 12.42 & 0.42 & ? ft 108 & 20 24 20 & -43 40 09 & 11.75 & 0.33 & ? ft 109 & 20 36 52 & -46 26 44 & 9.32 & 0.20 & 4657 ft 110 & 20 32 19 & -45 21 27 & 10.72 & 0.28 & 4652 ft 111 & 20 36 33 & -45 32 42 & 9.22 & 0.20 & 4656 ft 112 & 20 30 33 & -43 58 04 & 10.59 & 0.34 & 4651 ft 113 & 20 01 32 & -50 15 50 & 11.88 & 0.53 & ? ft 114 & 20 02 55 & -50 17 38 & 12.32 & 0.72 & ? ft 115 & 19 56 53 & -48 19 13 & 12.58 & 0.42 & ? ft 116 & 20 01 22 & -49 33 41 & 12.20 & 0.56 & ? ft 117 & 19 54 21 & -47 36 48 & 12.62 & 0.83 & ? ft 118 & 19 55 12 & -47 47 48 & 12.39 & 0.73 & ? ft 119 & 19 57 16 & -48 20 54 & 12.59 & 0.46 & ? ft 120 & 20 03 60 & -49 47 48 & 12.68 & 0.46 & ? ft 121 & 19 56 56 & -47 25 48 & 10.75 & 0.21 & 4539 ft 122 & 19 57 07 & -47 02 54 & 12.84 & 0.90 & ? ft 123 & 19 55 18 & -46 48 39 & 12.23 & 0.61 & ? ft 124 & 20 01 20 & -48 27 16 & 12.70 & 0.76 & ? ft 125 & 19 54 34 & -46 27 26 & 12.33 & 0.58 & ? ft 126 & 20 00 13 & -47 55 55 & 12.78 & 0.81 & ? ft 127 & 19 54 45 & -46 13 16 & 12.49 & 0.69 & ? ft 128 & 20 07 01 & -49 31 30 & 12.57 & 0.44 & ? ft 129 & 20 12 03 & -50 39 49 & 12.48 & 0.81 & ? ft 130 & 19 59 28 & -47 05 44 & 12.37 & 0.55 & 4543 ft 131 & 20 01 07 & -47 04 17 & 12.50 & 0.54 & 4549 ft 132 & 19 58 28 & -46 08 04 & 13.46 & 0.73 & ? ft 133 & 19 57 45 & -45 50 31 & 12.82 & 0.71 & ? ft 134 & 20 13 13 & -49 38 29 & 12.71 & 0.71 & ? ft 135 & 20 09 55 & -48 35 38 & 12.79 & 0.66 & ? ft 136 & 19 58 27 & -45 06 34 & 12.77 & 0.88 & ? ft 137 & 20 02 44 & -45 59 33 & 12.46 & 0.57 & ? ft 138 & 20 16 08 & -48 49 27 & 12.86 & 0.43 & 4607 ft 139 & 20 08 01 & -46 44 14 & 12.99 & 0.48 & ? ft 140 & 20 20 44 & -49 48 58 & 12.47 & 0.43 & ? ft 141 & 20 18 32 & -49 15 10 & 12.36 & 0.50 & ? ft 142 & 20 14 38 & -48 12 26 & 12.78 & 0.68 & ? ft 143 & 20 19 56 & -49 21 59 & 12.71 & 0.72 & ? ft 144 & 20 24 37 & -50 23 17 & 11.27 & 0.34 & ? ft 145 & 20 06 18 & -45 51 36 & 12.34 & 0.50 & ? ft 146 & 20 20 30 & -49 06 44 & 12.77 & 0.66 & 4618 ft 147 & 20 09 01 & -46 07 16 & 14.47 & 1.20 & ? ft 148 & 20 09 38 & -46 09 04 & 12.33 & 0.45 & 4585 ft 149 & 20 09 50 & -46 02 37 & 13.59 & 0.63 & ? ft 150 & 20 05 11 & -44 40 02 & 12.81 & 0.78 & ? ft 151 & 20 04 53 & -44 30 40 & 12.48 & 0.59 & ? ft 152 & 20 10 57 & -45 54 08 & 13.54 & 1.11 & ? ft 153 & 20 10 01 & -45 40 43 & 13.70 & 1.22 & 4588 ft 154 & 20 08 36 & -45 06 48 & 12.34 & 0.53 & ? ft 155 & 20 23 31 & -48 49 47 & 13.26 & 0.90 & ? ft 156 & 20 07 49 & -44 37 54 & 14.62 & 3.42 & ? ft 157 & 20 06 22 & -44 02 51 & 12.27 & 0.56 & ? ft 158 & 20 15 33 & -46 21 51 & 14.23 & 1.98 & ? ft 159 & 20 10 12 & -44 53 29 & 12.29 & 0.53 & ? ft 160 & 20 07 59 & -44 12 04 & 14.41 & 2.52 & 4577 ft 161 & 20 16 55 & -46 31 54 & 13.05 & 0.74 & ? ft 162 & 20 15 15 & -46 04 49 & 12.87 & 0.68 & ? ft 163 & 20 07 59 & -43 33 44 & 12.10 & 0.61 & 4576 ft 164 & 20 27 07 & -48 21 14 & 13.07 & 0.92 & ? ft 165 & 20 25 28 & -47 50 18 & 12.34 & 0.57 & 4635 ft 166 & 20 14 04 & -44 50 33 & 11.69 & 0.34 & ? ft 167 & 20 34 15 & -49 00 19 & 11.84 & 0.49 & ? ft 168 & 20 18 43 & -45 20 15 & 12.63 & 0.56 & 4614 ft 169 & 20 16 43 & -44 42 22 & 12.44 & 0.52 & ? ft 170 & 20 26 44 & -46 27 28 & 13.36 & 1.21 & ? ft 171 & 20 35 41 & -48 22 03 & 11.26 & 0.35 & ? ft 172 & 20 24 51 & -45 43 39 & 12.90 & 0.70 & ? ft 173 & 20 29 19 & -46 28 36 & 12.04 & 0.45 & ? ft 174 & 20 30 17 & -46 28 32 & 11.80 & 0.44 & ? ft 175 & 20 27 48 & -45 41 12 & 11.62 & 0.16 & ? ft 176 & 20 19 11 & -43 32 24 & 12.46 & 0.72 & ? ft 177 & 20 24 44 & -44 38 13 & 14.14 & 2.42 & ? ft 178 & 20 21 37 & -43 47 53 & 12.43 & 0.62 & 4621 ft 179 & 20 35 22 & -46 56 34 & 12.40 & 0.61 & ? ft 180 & 20 25 13 & -44 26 49 & 11.91 & 0.45 & ? ft 181 & 20 27 59 & -44 59 28 & 12.93 & 0.77 & ? ft 182 & 20 29 19 & -45 06 47 & 11.92 & 0.29 & 4644 ft 183 & 20 31 20 & -44 47 35 & 11.89 & 0.53 & ? ft 184 & 20 29 26 & -49 43 37 & 8.74 & 0.17 & ? ft 185 & 20 29 52 & -49 47 23 & 8.52 & 0.17 & 4648 ft 186 & 19 52 07 & -46 50 40 & 10.33 & 0.25 & 4531 ft 187 & 20 15 08 & -43 34 26 & 10.15 & 0.22 & ? ft 188 & 20 05 12 & -47 20 23 & 11.50 & 0.29 & ? ft 189 & 20 24 09 & -43 40 39 & 11.98 & 0.40 & ? fc 1 & usno0525 - 12907188 & b11.1 & u & ? & ? & 12 47 28.6 & -37 22 34 fc 2 & usno0525 - 13025279 & b12.6 & u & ? & ? & 12 55 26.4 & -37 19 18 fc 3 & sao203688 & 7.7 & s & a1v & 0.09 & 12 40 16.0 & -37 51 25 fc 4 & tyc7772 - 851 - 1 & 10.93 & s & ? & 0.39 & 12 56 30.5 & -37 40 10 fc 5 & sao203718 & 9.3 & s & a0 & 0.17 & 12 41 47.0 & -37 53 12 fc 6 & sao203731 & 8.1 & s & f0v & 0.38 & 12 42 32.2 & -37 54 10 fc 7 & tyc777 - 835 - 1 & 9.93 & s & ? & 0.47 & 13 03 39.7 & -37 46 52 fc 8 & usno0450 - 15027418 & b13.7 & u & ? & ? & 12 56 43.5 & -37 58 06 c 9 & usno - a2 - 0450 - 14831165 & b14.7 & u & ? & * -0.7 & 12 52 18.2 & -38 02 35 fc 10 & sao203908 & 9.7 & s & a & 0.31 & 12 53 25.7 & -38 03 32 c 11 & usno - a2 - 0450 - 14517802 & b12.0 & u & ? & * 0.4 & 12 38 01.9 & -38 15 11 fc 12 & usno0450 - 14861545 & b11.2 & u & ? & * 0.5 & 12 49 54.7 & -38 18 16 fc 13 & usno0450 - 14722885 & b14.0 & u & ? & ? & 12 43 53.0 & -38 21 25 fc 14 & sao203822 & 9.2 & s & a3 & 0.23 & 12 48 18.4 & -38 22 04 fc 15 & tyc7758 - 812 - 1 & 9.68 & s & ? & 0.32 & 12 39 35.9 & -38 24 39 fc 16 & sao204099 & 9.0 & s & a0 & 0.06 & 13 04 38.5 & -38 23 40 fc 17 & usno0450 - 14623199 & b11.9 & u & ? & * 0.0 & 12 39 09.8 & -38 44 51 fc 18 & sao203609 & 9.0 & s & a3ii & 0.23 & 12 34 55.5 & -38 49 55 fc 19 & usno0450 - 15096908 & b12.9 & u & ? & * -1.0 & 12 59 39.5 & -38 42 35 fc 20 & sao203816 & 8.9 & s & a3 & 0.24 & 12 47 48.3 & -38 51 45 fc 21 & usno0450 - 14756756 & b12.6 & u & ? & * -0.1 & 12 45 27.0 & -38 51 15 fc 22 & tyc7773 - 1670 - 1 & 10.09 & s & ? & 0.29 & 13 00 33.8 & -38 44 58 fc 23 & sao203979 & 6.8 & s & a3v & 0.13 & 12 58 01.0 & -38 55 02 fc 24 & sao203690 & 9.8 & s & f2 & 0.39 & 12 40 21.9 & -39 02 10 fc 25 & usno - a2 - 0450 - 14821964 & b11.8 & u & ? & * 0.2 & 12 51 54.7 & -38 58 10 fc 26 & sao203562 & 9.6 & s & a0 & 0.17 & 12 31 52.82 & -39 13 57 fc 27 & eso323-g025 & 13.39 & ig & sa(rs)b & ? & 12 52 38.7 & -39 01 42 fc 28 & sao204001 & 6.94 & s & a0v & 0.03 & 12 58 52.8 & -39 06 20 fc 29 & sao204080 & 7.88 & s & f0v & 0.36 & 13 03 36.7 & -38 58 07 fc 30 & usno0450 - 15038765 & b13.7 & u & ? & * -0.3 & 12 57 12.7 & -39 06 54 fc 31 & usno - a2 - 0450 - 15040174 & b14.9 & u & ? & * -0.8 & 13 01 04.0 & -39 03 19 fc 32 & ngc4553 & 13.5 & gp & sa0 & 0.93 & 12 36 08.0 & -39 26 19 fc 33 & sao203981 & 7.53 & s & a9iv & 0.34 & 12 58 03.1 & -39 16 02 fc 34 & usno0450 - 14874256 & b11.9 & u & ? & * 1.1 & 12 50 24.5 & -39 25 37 fc 35 & sao204034 & 8.2 & s & f2 & 0.41 & 13 00 44.4 & -39 17 43 fc 36 & sao20398 & 10.0 & s & a0 & 0.26 & 12 58 15.4 & -39 22 59 fc 37 & tyc7776 - 1116 - 1 & 10.95 & s & ? & 0.32 & 12 53 53.2 & -39 28 38 fc 38 & usno0450 - 14627245 & b13.6 & u & ? & * 0.5 & 12 39 21.4 & -39 41 04 fc 39 & sao203881 & 5.97 & s & b8v & -0.1 & 12 51 57.0 & -39 40 50 fc 40 & sao203835 & 8.1 & s & b5iv & -0.09 & 12 49 10.1 & -39 43 02 fc 41 & sao203621 & 5.78 & s & a0v & 0.00 & 12 36 01.1 & -39 52 11 fc 42 & tyc7775 - 798 - 1 & 10.73 & s & ? & 0.03 & 12 44 13.6 & -39 46 17 fc 43 & sao203681 & 4.63 & s & b8ii & -0.08 & 12 39 52.6 & -39 59 15 fc 44 & usno - a2 - 0450 - 14607200 & 9.9 & g & sabcd & * -1.3 & 12 42 16.2 & -40 06 25 fc 45 & sao203746 & 6.44 & s & a7iii & 0.24 & 12 43 26.3 & -40 10 41 fc 46 & sao203907 & 4.28 & s & a4iv & 0.22 & 12 53 26.2 & -40 10 44 fc 47 & sao203554 & 8.6 & s & a6iii & 0.29 & 12 31 29.1 & -40 16 28 fc 48 & usno - a2 - 0450 - 14768318 & b14.8 & u & ? & * -0.1 & 12 49 36.4 & -40 13 56 fc 49 & tyc7777 - 1416 - 1 & 10.37 & s & ? & 0.35 & 13 03 36.3 & -40 15 14 fc 50 & sao223587 & 8.9 & s & a5 & 0.31 & 12 40 22.0 & -40 37 52 fc 51 & sao223687 & 9.8 & s & a2 & 0.07 & 12 49 15.1 & -40 37 59 fc 52 & sao223519 & 8.0 & s & a5 & 0.17 & 12 34 13.8 & -40 49 44 fc 53 & usno0450 - 14828057 & b12.2 & u & ? & ? & 12 48 35.9 & -40 44 36 fc 54 & tyc7776 - 1480 - 1 & 10.09 & s & ? & 0.25 & 12 59 13.0 & -40 38 03 fc 55 & usno0450 - 14907321 & b13.6 & u & ? & * -0.4 & 12 51 42.7 & -40 44 09 fc 56 & sao223542 & 5.12 & s & a7iii & 0.22 & 12 35 45.5 & -41 01 19 fc 57 & tyc7775 - 757 - 1 & 10.34 & s & ? & 0.15 & 12 46 55.7 & -40 53 24 fc 58 & sao223879 & 7.93 & s & f6v & 0.45 & 13 04 45.2 & -40 40 58 fc 59 & sao223727 & 8.3 & s & a9iv & 0.28 & 12 52 34.3 & -40 56 28 fc 60 & sao223824 & 8.0 & s & a0 & 0.19 & 13 00 01.1 & -40 53 01 fc 61 & pgc042966 & 15.17 & ig & e & ? & 12 45 03.8 & -41 00 47 fc 62 & sao223916 & 10.26 & s & a3 & 0.28 & 13 07 38.1 & -40 46 30 fc 63 & tyc7762 - 215 - 1 & 11.11 & s & ? & 0.21 & 12 38 19.8 & -41 12 41 fc 64 & tyc7775 - 1763 - 1 & 10.63 & s & ? & 0.16 & 12 44 45.2 & -41 13 38 fc 65 & sao223706 & 8.47 & s & a2 & 0.21 & 12 51 03.9 & -41 15 05 fc 66 & sao223511 & 6.7 & s & a7v & 0.24 & 12 33 28.8 & -41 24 51 fc 67 & tyc7780 - 107 - 1 & 9.95 & s & ? & -0.07 & 12 52 19.4 & -41 19 38 fc 68 & tyc7778 - 1705 - 1 & 10.98 & s & ? & 0.18 & 13 10 27.0 & -40 57 24 fc 69 & sao223484 & 7.7 & s & a5iv & 0.32 & 12 31 18.8 & -41 30 17 fc 70 & sao223859 & 9.0 & s & a3 & 0.29 & 13 02 43.2 & -41 14 01 fc 71 & ngc4683 & 13.8 & ig & sb(s)0 & 0.93 & 12 47 42.6 & -41 31 44 fc 72 & sao223823 & 7.4 & s & f2 & 0.37 & 12 59 59.0 & -41 23 27 fc 73 & usno0450 - 14710416 & b14.1 & u & ? & * 0.5 & 12 43 18.5 & -41 37 04 fc 74 & sao223924 & 8.1 & s & a3v & 0.09 & 13 08 12.4 & -41 23 41 fc 75 & ngc4930 & 12.0 & g & sb(rs ) & 0.76 & 13 04 04.7 & -41 24 44 fc 76 & sao223701 & 8.8 & s & f3v & 0.43 & 12 50 31.7 & -41 41 47 fc 77 & sao223808 & 9.5 & s & a0 & 0.01 & 12 59 16.9 & -41 36 34 fc 78 & sao223577 & 9.11 & s & b9 & 0.11 & 12 39 25.2 & -41 51 03 fc 79 & tyc7781 - 616 - 1 & 10.04 & s & ? & 0.27 & 13 04 59.3 & -41 31 54 fc 80 & sao223513 & 7.95 & s & f2 & 0.37 & 12 33 40.8 & -41 57 04 fc 81 & sao223475 & 8.53 & s & ? & 0.29 & 12 30 11.6 & -41 55 28 fc 82 & sao223700 & 8.93 & s & a3 & 0.22 & 12 50 30.4 & -41 54 24 fc 83 & sao223558 & 8.72 & s & a5 & 0.38 & 12 37 21.7 & -42 01 42 fc 84 & sao223864 & 9.2 & s & a2 & 0.19 & 13 03 03.1 & -41 47 21 fc 85 & sao223778 & 8.2 & s & f3v & 0.44 & 12 56 50.2 & -41 54 50 fc 86 & sao223635 & 8.04 & s & f5v & 0.49 & 12 44 20.1 & -42 04 48 fc 87 & usno - a2 - 0450 - 15041646 & b11.3 & u & ? & * -0.1 & 13 01 07.7 & -41 58 55 fc 88 & tyc7780 - 1433 - 1 & 10.38 & s & ? & -0.03 & 12 56 14.2 & -42 09 11 fc 89 & sao223545 & 7.98 & s & sdf2 & 0.45 & 12 36 05.3 & -42 18 35 fc 90 & usno0450 - 15063833 & b11.6 & u & ? & * 0.7 & 12 58 16.0 & -42 03 15 fc 91 & tyc7781 - 83 - 1 & 9.85 & s & ? & 0.10 & 13 07 30.9 & -42 02 40 fc 92 & sao223756 & 6.82 & s & b5vn & -0.13 & 12 55 07.0 & -42 17 28 fc 93 & sao223580 & 9.51 & s & a3 m & 0.32 & 12 39 40.6 & -42 22 29 fc 94 & sao223946 & 8.95 & s & a0 & 0.10 & 13 10 03.7 & -42 13 16 fc 95 & tyc7766 - 1334 - 1 & 10.03 & s & ? & 0.14 & 12 37 25.3 & -42 36 07 fc 96 & sao223960 & 5.79 & s & f7iv & 0.51 & 13 11 08.9 & -42 13 59 fc 97 & sao223828 & 7.83 & s & f3iv & 0.43 & 13 00 17.8 & -42 30 04 fc 98 & tyc7781 - 907 - 1 & 10.52 & s & ? & 0.37 & 13 01 29.9 & -42 29 58 fc 99 & tyc7781 - 802 - 1 & 10.71 & s & ? & 0.35 & 13 02 11.6 & -42 29 13 fc 100 & sao223578 & 9.0 & s & f0v & 0.4 & 12 39 31.9 & -42 59 04 fc 101 & tyc7780 - 609 - 1 & 10.42 & s & ? & 0.15 & 12 52 08.7 & -42 55 26 fc 102 & sao223561 & 9.19 & s & f2 & 0.41 & 12 38 00.2 & -43 02 02 fc 103 & tyc7779 - 1941 - 1 & 10.56 & s & ? & -0.13 & 12 45 59.1 & -43 05 22 fc 104 & tyc7766 - 345 - 1 & 10.75 & s & ? & 0.57 & 12 37 52.3 & -43 05 24 fc 105 & sao223737 & 7.29 & s & a5v & 0.25 & 12 53 33.5 & -43 04 30 fc 106 & usno0450 - 14908873 & b12.8 & u & ? & * -1.2 & 12 51 46.9 & -43 10 44 fc 107 & sao223691 & 9.27 & s & f2 & 0.40 & 12 49 31.4 & -43 12 44 fc 108 & sao223678 & 8.2 & s & f0iii & 0.37 & 12 48 22.4 & -43 19 59 fc 109 & sao223605 & 9.38 & s & a2 & 0.08 & 12 41 30.5 & -43 38 43 fc 110 & tyc7785 - 239 - 1 & 10.89 & s & ? & 0.05 & 13 01 30.2 & -43 22 53 fc 111 & sao223857 & 7.9 & s & a7v & 0.26 & 13 02 38.0 & -43 21 18 fc 112 & sao223681 & 8.82 & s & b9iii & -0.03 & 12 48 47.2 & -43 42 56 fc 113 & tyc7785 - 2089 - 1 & 10.5 & s & ? & -0.10 & 13 08 46.2 & -43 27 52 fc 114 & usno0450 - 14550141 & b13.0 & u & ? & * -0.3 & 12 35 42.8 & -43 49 17 fc 115 & sao223609 & 10.01 & s & a2 & 0.32 & 12 41 58.9 & -43 53 52 fc 116 & tyc7784 - 810 - 1 & 10.93 & s & ? & 0.29 & 12 56 15.1 & -43 46 45 fc 117 & sao223683 & 7.8 & s & a3iii & 0.17 & 12 49 04.0 & -43 56 58 fc 118 & sao223836 & 9.67 & s & a0 & 0.16 & 13 00 53.3 & -43 47 45 fc 119 & usno0450 - 14711408 & b12.4 & u & ? & * 0.3 & 12 43 21.2 & -43 58 34 fc 120 & usno0450 - 14699795 & b12.0 & u & ? & * 0.2 & 12 42 47.8 & -44 06 09 fc 121 & sao223809 & 9.61 & s & a0 & 0.23 & 12 59 19.5 & -43 59 32 fc 122 & sao223880 & 8.79 & s & a2 & 0.25 & 13 04 48.3 & -43 55 29 fc 123 & sao223753 & 5.88 & s & g2iv & 0.63 & 12 54 58.6 & -44 09 07 fc 124 & tyc7783 - 1498 - 1 & 10.49 & s & ? & 0.40 & 12 41 17.9 & -44 18 10 fc 125 & sao223896 & 9.6 & s & a2 & 0.07 & 13 05 58.2 & -44 05 44 fc 126 & sao223742 & 9.1 & s & f0v & 0.38 & 12 54 15.6 & -44 24 06 fc 127 & tyc7783 - 1223 - 1 & 11.36 & s & ? & -0.07 & 12 43 02.4 & -44 34 25 fc 128 & sao223677 & 6.88 & s & b9v & -0.02 & 12 48 16.5 & -44 43 08 fc 129 & sao223598 & 10.28 & s & a2 & 0.21 & 12 41 11.8 & -44 42 26 fc 130 & sao223899 & 9.38 & s & a2iv & 0.18 & 13 06 08.8 & -44 25 14 fc 131 & sao223646 & 9.42 & s & a2 & 0.26 & 12 45 26.3 & -44 44 06 fc 132 & sao223866 & 7.8 & s & a2iv & 0.13 & 13 03 17.9 & -44 33 46 fc 133 & sao223842 & 7.66 & s & b6v & -0.08 & 13 01 15.2 & -44 45 07 fc 134 & sao223769 & 9.44 & s & a0 & 0.31 & 12 56 24.0 & -44 52 07 fc 135 & tyc7785 - 1568 - 1 & 10.88 & s & ? & 0.02 & 13 04 20.5 & -44 59 37 fc 136 & sao223740 & 9.0 & s & a9v & 0.34 & 12 53 54.3 & -45 06 33 fc 137 & sao223846 & 9.74 & s & b9 & 0.11 & 13 01 30.1 & -45 11 23 fc 138 & sao223679 & 8.4 & s & f0 & 0.33 & 12 48 33.8 & -45 25 11 fc 139 & sao223655 & 8.31 & s & f0 & 0.27 & 12 46 03.3 & -45 30 05 fc 140 & sao223722 & 9.33 & s & a2 & 0.03 & 12 52 18.7 & -45 29 29 fc 141 & tyc7259 - 115 - 1 & 9.94 & s & ? & 0.311 & 12 44 52.9 & -37 18 48 fc 142 & usno0450 - 14941695 & b12.3 & u & ? & * 0.4 & 12 53 07.4 & -37 31 45 fc 143 & usno0450 - 14747828 & b12.3 & u & ? & * 1.4 & 12 45 02.2 & -37 43 30 fc 144 & usno0450 - 14885379 & b11.3 & u & ? & * 0.7 & 12 50 49.6 & -37 49 39 fc 145 & usno0450 - 15121279 & b12.4 & u & ? & * 0.3 & 13 00 40.9 & -37 49 05 fc 146 & usno0450 - 14481710 & b13.7 & u & ? & * 0.4 & 12 32 26.5 & -38 32 45 fc 147 & usno0450 - 14485189 & b10.7 & u & ? & * -0.1 & 12 32 36.5 & -38 39 11 fc 148 & usno0450 - 14870154 & b13.2 & u & ? & * 0.0 & 12 50 14.8 & -38 29 34 fc 149 & usno0450 - 14551298 & b12.6 & u & ? & * 1.0 & 12 35 46.2 & -38 39 54 fc 150 & usno0450 - 15008287 & b12.8 & u & ? & * -0.5 & 12 55 55.5 & -38 39 59 fc 151 & usno0450 - 14951320 & b11.6 & u & ? & * 0.9 & 12 53 31.5 & -38 47 16 fc 152 & usno - a2 - 0450 - 14652712 & b15.9 & u & ? & * -0.2 & 12 44 24.5 & -38 56 15 fc 153 & usno0450 - 14854637 & b15.7 & u & ? & * -0.1 & 12 49 38.7 & -39 02 20 fc 154 & tyc7771 - 925 - 1 & 10.43 & s & ? & 0.42 & 12 44 37.7 & -39 05 51 fc 155 & usno0450 - 14643933 & b11.6 & u & ? & * 1.4 & 12 40 08.3 & -39 23 40 fc 156 & tyc7775 - 674 - 1 & 9.99 & s & ? & 0.46 & 12 41 33.8 & -39 27 38 fc 157 & usno0450 - 15104855 & b12.6 & u & ? & * 0.5 & 12 59 59.8 & -39 14 56 fc 158 & tyc7772 - 1576 - 1 & 9.81 & s & ? & 0.36 & 12 54 20.1 & -39 19 32 fc 159 & usno0450 - 14709560 & b11.5 & u & ? & * 1.0 & 12 43 16.0 & -39 32 22 fc 160 & sao203807 & 8.81 & s & g3v & 0.63 & 12 47 23.2 & -39 28 54 fc 161 & ngc4679 & 13.15 & g & sa(s)bc & ? & 12 47 30.2 & -39 34 21 fc 162 & tyc7762 - 1018 - 1 & 11.20 & s & ? & 0.30 & 12 30 31.9 & -39 39 53 fc 163 & tyc7775 - 54 - 1 & 10.99 & s & ? & 0.53 & 12 45 11.9 & -39 38 34 fc 164 & tyc7762 - 206 - 1 & 10.03 & s & ? & 1.03 & 12 33 31.5 & -39 49 01 fc 165 & usno - a2 - 0450 - 14720843 & b16.9 & u & ? & * 1.7 & 12 47 33.0 & -39 46 46 fc 166 & usno0450 - 14850792 & b13.1 & u & ? & * 0.4 & 12 49 30.1 & -39 53 31 fc 167 & usno0450 - 14528258 & b12.2 & u & ? & * 1.6 & 12 34 40.1 & -40 03 17 fc 168 & sao203806 & 8.4 & s & m3iii & 1.58 & 12 47 19.4 & -40 07 377 fc 169 & sao204171 & 8.06 & s & f5v & 0.48 & 13 09 10.5 & -39 52 24 fc 170 & sao203541 & 8.8 & s & a2 & 0.41 & 12 30 27.9 & -40 16 06 fc 171 & sao223652 & 9.1 & s & f8 & 0.56 & 12 45 42.5 & -40 16 41 fc 172 & tyc7775 - 1831 - 1 & 10.14 & s & ? & 0.44 & 12 47 49.7 & -40 22 53 fc 173 & tyc7762 - 1543 - 1 & 10.56 & s & ? & 0.35 & 12 38 21.2 & -40 30 58 fc 174 & usno0450 - 15031167 & b14.0 & u & ? & * 0.0 & 12 56 53.2 & -40 21 00 fc 175 & tyc7776 - 987 - 1 & 10.96 & s & ? & 0.14 & 12 59 25.0 & -40 25 53 fc 176 & usno0450 - 15063491 & b13.3 & u & ? & * 0.4 & 12 58 15.1 & -40 36 25 fc 177 & usno0450 - 14575914 & b13.7 & u & ? & * 0.5 & 12 36 55.9 & -40 52 01 fc 178 & usno - a2 - 0450 - 15059539 & b12.2 & u & ? & * 0.0 & 13 01 52.8 & -40 37 45 fc 179 & usno0450 - 15215568 & b13.5 & u & ? & * -0.7 & 13 04 37.6 & -40 32 15 fc 180 & tyc7777 - 392 - 1 & 10.5 & s & ? & 0.81 & 13 07 40.9 & -40 30 02 fc 181 & tyc7778 - 1639 - 1 & 9.36 & s & ? & 0.33 & 13 11 12.2 & -40 28 26 fc 182 & tyc7762 - 1945 - 1 & 10.18 & s & ? & 0.56 & 12 38 52.6 & -41 00 55 fc 183 & tyc7762 - 1926 - 1 & 10.57 & s & ? & 0.24 & 12 33 49.9 & -40 59 52 fc 184 & sao223494 & 7.3 & s & k2iii & 1.24 & 12 31 59.2 & -41 03 19 fc 185 & usno0450 - 14860017 & b11.4 & u & ? & * 0.8 & 12 49 50.9 & -41 04 23 fc 186 & eso323-g054 & 14.86 & ig & s & ? & 12 59 18.2 & -40 57 41 fc 187 & usno - a2 - 0450 - 14757081 & b12.4 & u & ? & * 0.0 & 12 49 08.7 & -41 12 26 fc 188 & usno0450 - 15248516 & b13.2 & u & ? & * -0.4 & 13 05 59.3 & -40 57 58 fc 189 & usno0450 - 14676419 & b14.7 & u & ? & * 1.4 & 12 41 41.4 & -41 16 59 fc 190 & tyc7766 - 717 - 1 & 10.64 & s & ? & 0.37 & 12 38 18.1 & -41 24 34 fc 191 & sao223465 & 8.03 & s & f3iv & 0.39 & 12 29 48.4 & -41 24 09 fc 192 & tyc7776 - 1098 - 1 & 10.55 & s & ? & 0.39 & 12 58 49.4 & -41 09 07 fc 193 & usno - a2 - 0450 - 14865062 & 16.4 & u & ? & * -0.4 & 12 53 42.8 & -41 18 29 fc 194 & sao223501 & 9.38 & s & f0 & 0.26 & 12 32 49.5 & -41 37 58 fc 195 & usno - a2 - 0450 - 15278266 & b11.8 & u & ? & * -0.2 & 13 10 52.4 & -41 18 04 fc 196 & tyc7779 - 1219 - 1 & 11.05 & s & ? & 0.43 & 12 45 45.7 & -41 49 32 fc 197 & usno - a2 - 0450 - 14825935 & b14.5 & u & ? & * 1.0 & 12 52 04.9 & -41 52 03 fc 198 & usno0450 - 14926401 & b14.6 & u & ? & * 0.7 & 12 52 30.3 & -41 55 28 fc 199 & usno0450 - 14931221 & b14.7 & u & ? & * 0.4 & 12 52 41.9 & -41 57 41 fc 200 & usno0450 - 14945366 & b15.0 & u & ? & * 1.0 & 12 53 16.5 & -42 00 40 fc 201 & tyc7779 - 733 - 1 & 10.93 & s & ? & 0.32 & 12 42 37.3 & -42 10 13 fc 202 & tyc7781 - 114 - 1 & 11.19 & s & ? & 0.08 & 13 08 12.6 & -41 57 12 fc 203 & usno0450 - 14682772 & b14.3 & u & ? & ? & 12 41 59.7 & -42 18 16 fc 204 & usno0450 - 15142256 & b14.3 & u & ? & * 0.5 & 13 01 34.0 & -42 07 48 fc 205 & usno0450 - 15213947 & b12.1 & u & ? & ? & 13 04 33.6 & -42 08 21 fc 206 & usno0450 - 14842856 & b13.2 & u & ? & * 0.6 & 12 49 11.6 & -42 26 10 fc 207 & tyc7779 - 1570 - 1 & 10.28 & s & ? & 0.52 & 12 45 58.5 & -42 31 52 fc 208 & tyc7766 - 1132 - 1 & 10.51 & s & ? & 0.37 & 12 38 58.0 & -42 34 53 fc 209 & usno - a2 - 0450 - 14919623 & b13.1 & u & ? & * 0.5 & 12 52 13.9 & -42 31 01 fc 210 & sao223535 & 8.95 & s & f2 & 0.36 & 12 35 13.1 & -42 40 21 fc 211 & sao223684 & 7.85 & s & f5iv & 0.54 & 12 49 07.0 & -42 38 03 fc 212 & tyc7780 - 747 - 1 & 10.7 & s & ? & 0.45 & 12 51 24.6 & -42 44 48 fc 213 & sao223819 & 8.8 & s & g5v & 0.71 & 12 59 52.0 & -42 33 41 fc 214 & usno - a2 - 0450 - 14647502 & 9.5 & u & ? & * 0.4 & 12 44 18.5 & -42 47 25 fc 215 & usno0450 - 15318910 & b11.7 & u & ? & ? & 13 08 52.4 & -42 27 49 fc 216 & tyc7779 - 2014 - 1 & 9.82 & s & ? & 0.37 & 12 47 55.1 & -42 50 35 fc 217 & tyc7779 - 2027 - 1 & 9.89 & s & ? & 0.31 & 12 44 10.1 & -43 04 30 fc 218 & sao223942 & 9.56 & s & ? & 0.37 & 13 09 44.9 & -42 43 14 fc 219 & sao223596 & 9.32 & s & f5 & 0.43 & 12 41 10.5 & -43 16 59 fc 220 & usno - a2 - 0450 - 15116369 & b14.5 & u & ? & * 1.5 & 13 04 15.1 & -43 02 04 fc 221 & usno0450 - 15150211 & b12.4 & u & ? & * 0.1 & 13 01 54.1 & -43 03 39 fc 222 & sao223571 & 9.35 & s & a0 m & 0.37 & 12 39 01.6 & -43 25 26 fc 223 & usno0450 - 15209781 & b11.6 & u & ? & ? & 13 04 23.1 & -43 08 30 fc 224 & tyc7783 - 60 - 1 & 9.81 & s & ? & 0.73 & 12 44 25.1 & -43 22 45 fc 225 & tyc7784 - 1348 - 1 & 10.77 & s & ? & 0.24 & 12 57 53.9 & -43 18 20 fc 226 & usno0450 - 15036393 & b12.2 & u & ? & * 0.4 & 12 57 06.7 & -43 20 59 fc 227 & usno0450 - 15075910 & b11.6 & u & ? & * 0.8 & 12 58 46.7 & -43 27 41 fc 228 & tyc7785 - 188 - 1 & 10.26 & s & ? & 0.27 & 13 00 36.8 & -43 27 17 fc 229 & eso269-ig056 & 14.01 & ig & sb(s)mp & ? & 13 10 01.2 & -43 12 46 fc 230 & sao223878 & 9.89 & s & a0 & 0.33 & 13 04 28.6 & -43 23 56 fc 231 & usno0450 - 15171172 & b12.6 & u & ? & * 1.3 & 13 02 46.7 & -43 32 10 fc 232 & usno0450 - 15282695 & b12.6 & u & ? & * 0.5 & 13 07 23.3 & -43 36 50 fc 233 & usno0450 - 15158550 & b12.9 & u & ? & * -1.2 & 13 02 14.6 & -43 39 19 fc 234 & tyc7785 - 367 - 1 & 10.54 & s & ? & 0.40 & 13 08 18.1 & -43 35 18 fc 235 & eso269-ig022/3 & 14.7 & ip & s0 & * -1.3 & 12 58 50.6 & -43 52 35 fc 236 & tyc7784 - 485 - 1 & 9.71 & s & ? & 0.54 & 12 51 01.0 & -44 11 32 fc 237 & usno0450 - 15075030 & b14.7 & u & ? & * 0.1 & 12 58 44.4 & -44 19 32 fc 238 & pgc044774 & 16.0 & gp & sab(s)p & ? & 13 00 31.1 & -44 14 39 fc 239 & usno0450 - 14840957 & b12.0 & u & ? & * 1.4 & 12 48 46.7 & -44 27 22 fc 240 & usno0450 - 15201324 & b13.5 & u & ? & * 0.1 & 13 04 01.7 & -44 19 42 fc 241 & usno0450 - 15199021 & b13.3 & u & ? & * -0.3 & 13 03 56.1 & -44 22 31 fc 242 & usno0450 - 15100880 & b12.7 & u & ? & * 1.1 & 12 59 49.7 & -44 27 49 fc 243 & sao223704 & 9.61 & s & f2 & 0.36 & 12 51 02.6 & -44 53 04 fc 244 & usno0375 - 17118731 & b12.9 & u & ? & * 0.8 & 12 57 28.5 & -45 05 36 fc 245 & tyc8246 - 51 - 1 & 10.82 & s & ? & 0.34 & 12 57 29.5 & -45 25 06 fc 246 & tyc7776 - 1164 - 1 & 9.57 & s & ? & 0.28 & 12 58 58.9 & -40 27 40 fc 247 & sao223716 & 9.52 & s & a0v & 0.11 & 12 51 45.8 & -41 18 27 fc 248 & sao223728 & 8.5 & s & f3v & 0.43 & 12 52 44.7 & -41 11 04 fc 249 & sao223721 & 9.45 & s & a5 & 0.32 & 12 52 16.0 & -43 02 14 fd 1 & sao233280 & 6.92 & s & a4v & 0.59 & 3 48 24.9 & -54 17 37 fd 2 & sao233313 & 8.69 & s & a0 & 0.08 & 3 53 01.3 & -54 06 231 fd 3 & sao233314 & 7.75 & s & b6v&-0.13 & 3 53 22.8 & -54 51 08 fd 4 & sao233384 & 7.52 & s & f0iv & 0.34 & 4 03 40.6 & -52 55 44 fd 5 & tyc8503 - 860 - 1 & 11.15 & s & a & 0.17 & 3 48 14.0 & -56 41 13 fd 6 & sao233351 & 7.81 & s & f5v & 0.42 & 3 59 54.3 & -54 09 41 fd 7 & tyc8504 - 1517 - 1 & 10.42 & s & ? & 0.11 & 3 57 13.7 & -55 05 25 fd 8 & sao233363 & 7.68 & s & a3v & 0.14 & 4 00 59.1 & -54 23 31 fd 9 & sao233306 & 9.25 & s & a0 & 0.07 & 3 51 44.2 & -56 39 06 fd 10 & usno0300 - 01218129 & 13.19 & s & b a & 0.45 & 4 03 21.8 & -54 39 54 fd 11 & sao233334 & 8.66 & s & f0iii & 0.33 & 3 56 34.5 & -56 19 30 fd 12 & sao233341 & 8.36 & s & a9iii & 0.28 & 3 57 45.5 & -56 15 01 fd 13 & hic18969 & 7.66 & s & a3v & 0.17 & 4 03 53.0 & -55 24 00 fd 14 & sao233451 & 9.51 & s & b9&-0.03 & 4 15 18.3 & -52 09 48 fd 15 & hic18656 & 6.84 & s & a2 m & 0.24 & 3 59 37.5 & -56 28 22 fd 16 & sao233423 & 9.49 & s & f8 & 0.53 & 4 10 31.0 & -53 35 06 fd 17 & sao233368 & 9.20 & s & a0 & 0.06 & 4 01 15.4 & -56 19 34 fd 18 & hic18592 & 6.05 & s & f2iv & 0.45 & 3 58 42.9 & -57 06 09 fd 19 & hic19543 & 8.55 & s & f3v & 0.42 & 4 11 08.9 & -54 02 22 fd 20 & gsc0850201085 & 14.78 & s & f & ? & 4 11 31.6 & -53 49 57 fd 21 & usno0300 - 01267605 & b13.7 & s & a & * 0.6 & 4 13 44.9 & -53 26 19 fd 22 & he0365 - 5827 & b13.54 ? & s & sdo & * 1.8 & 3 55 53.6 & -58 18 23 fd 23 & hic20109 & 6.09 & s & a8v & 0.48 & 4 18 40.0 & -52 51 36 fd 24 & usno0300 - 01253020 & 11.87 & s & a & * 0.9 & 4 09 25.4 & -55 44 27 fd 25 & hic18878 & 8.45 & s & f0iv & 0.28 & 4 02 49.7 & -57 27 34 fd 26 & sao233400 & 8.78 & s & a5 & 0.31 & 4 06 37.0 & -56 34 32 fd 27 & sao233460 & 9.63 & s & f2 & 0.28 & 4 16 10.1 & -54 00 53 fd 28 & sao233392 & 9.64 & s & a2 & 0.29 & 4 04 32.8 & -57 09 35 fd 29 & gsc0850200572 & 14.80 & s & sdob & ? & 4 15 30.3 & -54 21 58 fd 30 & he0418 - 5326 & 15.76 & s&wfd a & ? & 4 19 24.8 & -53 19 18 fd 31 & hic18731 & 7.04 & s & a2v & 0.11 & 4 00 44.9 & -58 39 42 fd 32 & hic20117 & 7.17 & s & a1v & 0.11 & 4 18 48.2 & -54 08 08 fd 33 & iue - bpfm 1 7731 & 15.54 & s & wfd a & * -1.2 & 4 19 10.1 & -53 50 43 fd 34 & usno0300 - 01307552 & 14.28 & s & sdo+cool & * 0.3 & 4 20 02.7&-54 04 14 fd 35 & iue - wd0416 - 55 & 15.49 & s & wd & * -1.4 & 4 17 11.4 & -54 57 47 fd 36 & tyc8507 - 189 - 1 + 2 & 9.40 & s & f5iv & 0.47 & 4 04 50.3 & -58 10 40 fd 37 & sao233500 & 9.85 & s & a2 & 0.24 & 4 22 59.4 & -53 32 06 fd 38 & hic19086 & 7.97 & s & a4v & 0.16 & 4 05 31.3 & -58 28 42 fd 39 & hic19759 & 7.55 & s & f5v & 0.45 & 4 14 11.3 & -56 10 36 fd 40 & hic20650 & 6.95 & s & f0v & 0.35 & 4 25 29.8 & -53 06 43 fd 41 & ngc1566 & 10.33 & g & sab(rs ) & 0.56 & 4 20 00.6 & -54 56 17 fd 42 & usno0300 - 01320497 & 13.87 & s & sdo & * -0.8 & 4 22 37.3 & -54 08 49 fd 43 & hic19353 & 9.40 & s & a1v & 0.07 & 4 08 45.9 & -58 29 14 fd 44 & usno0300 - 01325652 & 15.06 & s & a & * -0.4 & 4 23 36.7 & -54 17 17 fd 45 & hic20799 & 8.06 & s & b ? & 0.36 & 4 27 24.2 & -53 24 16 fd 46 & usno0300 - 01351412 & 15.13 & s & sdb & * -0.9 & 4 28 31.0 & -53 27 53 fd 47 & usno0300 - 01295274 & 12.51 & s & a & * -0.4 & 4 17 35.3 & -56 43 36 fd 48 & hic19726 & 7.32 & s & a5iv & 0.22 & 4 13 39.9 & -58 01 19 fd 49 & usno0300 - 01357213 & 13.04 & s & a & * 0.0 & 4 29 35.4 & -53 13 13 fd 50 & usno0300 - 01265933 & b15.7 & u & ? & * -0.8 & 4 11 45.7 & -58 41 06 fd 51 & tyc8505 - 1427 - 1 & 10.62 & s & b - a & 0.19 & 4 25 53.9 & -55 06 11 fd 52 & usno0300 - 01365856 & b12.8 & s & b(l)-a & * -0.5 & 4 31 11.0 & -53 35 27 fd 54 & sao233498 & 9.71 & s & a2 & 0.25 & 4 22 06.0 & -56 58 59 fd 55 & sao233493 & 9.40 & s & f2 & 0.37 & 4 21 23.9 & -57 32 30 fd 56 & sao233526 & 9.80 & s & a0 & 0.01 & 4 27 04.8 & -56 04 24 fd 57 & hic20543 & 7.53 & s & a2v & 0.01 & 4 24 06.9 & -57 15 11 fd 58 & hic19921 & 4.44 & s & k2iv & 1.08 & 4 16 28.9 & -59 18 07 fd 59 & sao233524 & 9.10 & s & f0 & 0.31 & 4 26 33.0 & -56 26 41 fd 63 & hic21281 & 3.30 & s & a0v & -0.08 & 4 33 59.8 & -55 02 42 fd 64 & sao233547 & 9.70 & s & a0 & 0.01 & 4 31 10.2 & -55 54 43 fd 65 & hic21200 & 8.17 & s & f2v & 0.38 & 4 32 43.2 & -55 35 49 fd 66 & hic21460 & 8.50 & s & ap & 0.28 & 4 36 30.8 & -54 37 16 fd 67 & usno0300 - 01340927 & b13.6 & u & ? & * -0.4 & 4 26 29.3 & -59 06 10 fd 68 & tyc8512 - 2014 - 1 & 9.55 & s & f & 0.29 & 4 34 48.1 & -56 35 30 fd 69 & usno0300 - 01391935 & b14.4 & u & ? & * 0.4 & 4 35 59.8 & -56 16 50 fd 70 & hic21787 & 7.75 & s & a8iv & 0.34 & 4 40 48.9 & -56 01 52 fd 71 & sao233619 & 9.35 & s & a & 0.28 & 4 40 18.5 & -56 11 34 fd 75 & tyc8512 - 2052 - 1 & 10.17 & s & ? & 0.16 & 4 42 36.9 & -56 41 35 fd 76 & hic21645 & 7.20 & s & a4v & 0.19 & 4 38 54.1 & -58 58 50 fd 77 & eso118-g34 & 13.52 & g & s0 pec & * 0.4 & 4 40 17.4&-58 44 43 fd 78 & hic18929 & 8.39 & s & f0iii & 0.34 & 4 03 28.1 & -52 30 21 fd 127 & usno - a2 - 0300 - 01385078 & b14.7 & u & ? & * 1.3 & 4 36 17.0 & -57 53 32 fd 128 & tyc8512 - 1176 - 1 & 11.33 & s & ? & 0.84 & 4 43 34.8 & -55 49 34 fd 129 & usno - a2 - 0300 - 01371953 & b12.2 & u & ? & * 0.9 & 4 33 53.7 & -58 50 21 fd 130 & hic21874 & 8.04 & s & f5v & 0.46 & 4 42 10.5 & -56 07 08 fm 2 & hfd 121081 & 8.9 & s & b9v & 0.08&13 53 58.4 & -28 8 34 fm 3 & usno0600 - 09386883 & 14.01 & s & b ? & * -1.1&13 54 00.2 & -27 55 16 fm 4 & sao182054 & 9.34 & s & a2 & 0.28&13 53 44.8 & -27 06 02 fm 5 & hfd 120970 & 10.3 & s & a2 & 0.22&13 53 15.0 & -27 18 37 fm 6 & tyc6724 - 799 - 1 & 11.68 & s & a - f ? & 0.42&13 52 54.8 & -27 40 08 fm 7 & sao182056 & 8.1 & s & a6iv & 0.24&13 53 53.1 & -30 00 39 fm 8 & hfd 120867 & 10.2 & s & a0 & 0.09&13 52 36.0 & -27 30 00 fm 9 & sao182034 & 10.24 & s & a0 & 0.11&13 52 18.7 & -26 47 16 fm 10 & sao204927 & 7.9 & s & a3 & 0.10&13 52 17.6 & -30 47 06 fm 11 & hfd 120657 & 10.5 & s & a2 & 0.24&13 51 23.1 & -29 08 44 fm 12 & hfd 120733 & 9.69 & s & a0 & 0.10 & 13 51 48.7 & -30 48 12 fm 13 & sao181999 & 6.19 & s & a0v & -0.03&13 50 06.4 & -29 04 51 fm 14 & tyc7279 - 1224 - 1 & 12.99 & s & ? & -0.04 & 13 51 28.9 & -30 48 27 fm 15 & sao181996 & 7.70 & s & a6 & 0.45&13 49 54.2 & -28 11 11 fm 16 & usno0600 - 09347896 & 15.37 & u & ? & * -0.1&13 49 10.2 & -26 55 30 fm 17 & sao204899 & 8.6 & s & f6v & 0.41&13 50 38.2 & -30 15 41 fm 18 & sao181993 & 8.7 & s & a2 & 0.36&13 49 47.3 & -28 38 24 fm 19 & usno0600 - 09348084 & b15.2 & s & f ? & * 0.4&13 49 11.6&-27 25 30 fm 20 & sao204900 & 7.1 & s & a5 & 0.17&13 50 39.6 & -31 12 20 fm 21 & sao204892 & 7.2 & s & f0 & 0.29 & 13 50 20.9 & -31 04 03 fm 22 & tyc6728 - 518 - 1 & 11.16 & s & a ? & 0.27&13 49 10.9 & -28 34 31 fm 23 & usno0600 - 09343173 & 13.52 & s & a ? & * -0.2&13 48 35.8 & -27 56 33 fm 24 & tyc6727 - 804 - 1 & 11.41 & s & b - a ? & 0.19&13 48 05.8 & -29 20 22 fm 25 & sao181931 & 5.8 & s & a0v & 0.02&13 45 36.3 & -26 06 57 fm 26 & tyc7266 - 692 - 1 & 10.69 & s & a0 & 0.10 & 13 47 50.9 & -31 32 49 fm 27 & usno0600 - 09325326 & 11.83 & u & ? & * 0.5&13 46 27.3 & -28 02 45 fm 28 & n5291/b & 14.00 & g & s pec & ? & 13 47 24.1 & -30 24 27 fm 29 & sao181949 & 8.6 & s & f8 & 0.51&13 46 57.5 & -29 40 18 fm 30 & hfd 119753 & 8.2 & s & f3v & 0.60&13 45 35.0 & -27 51 15 fm 32 & sao204825 & 8.3 & s & g0 & 0.59&13 46 17.0 & -30 28 28 fm 33 & tyc7266 - 1276 - 1 & 11.17 & s & a - f ? & 0.13 & 13 46 07.6 & -30 12 29 fm 34 & sao181919 & 8.6 & s & a3 & 0.30 & 13 44 36.7 & -26 46 48 fm 35 & tyc7266 - 1014 - 1 & 11.47 & s & a - f ? & 0.17 & 13 46 25.2 & -31 45 48 fm 36 & sao181928 & 8.8 & s & f0 & 0.31&13 45 17.8 & -29 01 00 fm 37 & hfd 119463 & 10.5 & s & a3 & 0.24 & 13 43 52.4 & -26 47 38 fm 38 & usno0600 - 09305363 & 15.35 & s & a - f(e ) ? & * -0.2&13 44 10.0 & -28 17 21 fm 39 & sao181910 & 8.8 & s & a2 & 0.26 & 13 43 53.9 & -27 45 59 fm 41 & sao181914 & 9.3 & s & a2 & 0.07&13 44 25.4 & -30 07 50 fm 42 & sao204787 & 8.5 & s & f0 & 0.37&13 44 09.9 & -31 48 59 fm 43 & sao181891 & 7.0 & s & a2v & 0.90&13 43 00.9 & -30 10 58 fm 44 & sao204771 & 9.0 & s & f0iv & 0.38&13 43 27.1 & -31 26 39 fm 45 & sao181862 & 8.7 & s & a2v & 0.19&13 41 31.2 & -27 16 21 fm 46 & sao181876 & 8.5 & s & f0 & 0.31&13 42 15.5 & -29 46 12 fm 47 & sao181868 & 8.0 & s & f2 & 0.37&13 41 43.6 & -28 45 44 fm 48 & usno0525 - 13768846 & 12.17 & s & a ? & * -0.1&13 42 29.8 & -31 3 21 fm 49 & n5264 & 12.6 & g & ib(s)m & 0.55&13 41 31.9 & -29 55 08 fm 50 & hic66742 & 8.01 & s & f3 & 0.44 & 13 40 47.0 & -28 14 32 fm 51 & usno - a2 - 0600 - 15938674 & b13.7 & u & ? & * -2.0 & 13 40 46.1 & -28 56 54 fm 52 & sao181838 & 8.4 & s & a3 & 0.10&13 39 52.9 & -27 09 05 fm 53 & usno0525 - 13747141 & b13.8 & s & a ? & * 0.1&13 41 13.3 & -31 34 54 fm 54 & gsc06726001897 & 14.58 & s & ? & ? & 13 39 42.5 & -28 8 37 fm 56 & sao181827 & 8.1 & s & a3iv & 0.16&13 38 44.5 & -27 15 35 fm 57 & sao181823 & 8.4 & s & a1v & 0.09&13 38 31.2 & -26 54 21 fm 58 & ngc5253 & 10.87 & g&im pec & 0.38&13 39 56.8 & -31 38 11 fm 59 & sao204696 & 9.59 & s & a3 & 0.35&13 39 07.3 & -30 15 30 fm 60 & sao181825 & 5.8 & s & f3v & 0.42 & 13 38 42.1 & -29 33 39 fm 61 & sao204695 & 9.37 & s & f5 & 0.38&13 39 03.6 & -30 44 53 fm 62 & sao204711 & 9.8 & s & a0 & 0.18 & 13 39 41.8 & -32 29 21 fm 63 & sao181807 & 8.6 & s & a5 & 0.17&13 37 34.8 & -27 03 31 fm 64 & sao181821 & 7.0 & s & f2iii & 0.40&13 38 23.3 & -29 50 20 fm 69 & he1334 - 2739 & 16.41 & ag & qso cand . & & 13 37 04.7 & -27 54 20 fm 70 & n5236 & 8.2 & g & sab(s)c & 0.61&13 36 58.8 & -29 51 52 fm 71 & i4293 & 13.50 & g & sa0 & * 1.5&13 36 01.7 & -25 52 53 fm 72 & sao204668 & 9.08 & s & a3 & 0.35&13 37 33.4 & -32 08 24 fm 73 & he1333 - 2853 & 15.92 & s & sda or qso cand . & & 13 36 29.5 & -29 08 53 fm 74 & hfd 118250 & 10.7 & s & a2 & 0.25&13 36 06.8 & -28 42 20 fm 75 & usno0525 - 13671585 & b13.4 & s & a ? & * 0.5&13 36 40.5 & -30 18 36 fm 76 & hfd 118350 & 9.8 & s & a0 & 0.31 & 13 36 58.8 & -32 17 38 fm 77 & sao204651 & 7.6 & s & a9v & 0.42&13 36 46.7 & -32 24 13 fm 78 & usno0600 - 09235666 & 13.7 & us & ? & * 1.1&13 35 00.6 & -26 05 32 fm 79 & sao181775 & 8.4 & s & a0 & 0.01&13 35 30.8 & -29 09 18 fm 80 & sao181779 & 8.1 & s & f2v & 0.41&13 35 41.1 & -29 22 40 fm 81 & sao181760 & 8.9 & s & m0 & 1.27&13 34 18.0 & -25 22 49 fm 82 & hfd 118147 & 10.03 & s & a3 & 0.27 & 13 35 34.3 & -31 25 26 fm 83 & hfd 118162 & 9.7 & s & a0 & 0.17 & 13 35 37.8 & -32 01 04 fm 85 & sao204623 & 6.8 & s & f0 & 0.37&13 35 12.2 & -32 12 45 fm 86 & sao204613 & 9.7 & s & f2 & 0.37&13 34 42.4 & -31 12 13 fm 87 & sao181742 & 6.9 & s & a9iv & 0.33&13 32 54.0 & -26 35 15 fm 88 & sao181749 & 8.7 & s & a2 & 0.07&13 33 09.2 & -29 17 58 fm 89 & sao181737 & 5.7 & s & a1v & 0.04&13 32 35.9 & -28 41 34 fm 90 & sao204584 & 10.2 & s & a0 & 0.01&13 33 19.7 & -31 39 13 fm 91 & sao181736 & 8.2 & s & a4v & 0.18&13 32 35.1 & -29 17 45 fm 93 & sao181735 & 6.4 & s & f5iv & 0.44&13 32 34.5&-29 33 55 fm 94 & ppm291518 & 10.49 & s & a ? & 0.20 & 13 33 06.6 & -31 47 36 fm 95 & sao181723 & 6.4 & s & a1v & 0.09&13 31 33.2 & -28 06 46 fm 96 & sao181716 & 7.5 & s & a4iv & 0.18&13 31 07.4 & -25 39 25 fm 97 & hfd 117531 & 9.70 & s & a2 & 0.27&13 31 20.8 & -27 31 46 fm 98 & tyc7269 - 1285 - 1 & 11.21 & s & ? & 0.04 & 13 32 42.5 & -32 47 00 fm 105 & sao181690 & 7.6 & s & a5 & 0.17&13 29 30.4 & -27 23 22 fm 106 & sao181698 & 9.2 & s & a3v & 0.16&13 29 53.0 & -28 56 41 fm 107 & sao181691 & 9.3 & s & a0 & 0.01&13 29 32.5 & -29 05 28 fm 108 & gsc06721 - 01689 & 13.4 & s & hot star ? & * -0.8&13 28 18.2 & -26 33 51 fm 119 & sao181601 & 8.8 & s & a2 & 0.07&13 22 42.7 & -30 11 51 fm 120 & gsc06712 - 01800 & 13.4 & s & a ? & * -0.3&13 21 56.2 & -28 26 31 fm 121 & usno - a2 - 0600 - 15616353 & b12.1 & s & a ? & * 0.3 & 13 20 45.5 & -29 29 33 fm 122 & sao204359 & 8.48 & s & b5iii & -0.11&13 19 43.0 & -30 19 56 fm 136 & sao181916 & 9.2 & s & k0 & 0.86&13 44 26.8 & -26 19 27 fm 137 & usno0600 - 09301026 & b14.6 & u & ? & * 1.3&13 43 36.1 & -26 29 45 fm 138 & gsc06727 - 01025 & 13.0 & s & f ? & * 1.2&13 44 43.6 & -29 37 49 fm 139 & hfd 119523 & 10.54 & s & f5 & 0.31 & 13 44 14.6 & -28 48 56 fm 174 & sao204493 & 8.3 & s & f3v & 0.48&13 27 45.0 & -30 47 03 fm 175 & eso444 - 46 & 15 & g & e & ? & 13 27 56.8 & -31 29 43 fm 176 & eso444 - 37 & 14.82 & g & sb(s)dm & * 0.6&13 26 59.8 & -30 04 27 fm 177 & ppm261656 & 9.7 & s & f8 & 0.38&13 26 21.8 & -26 50 54 fm 190 & sao204469 & 6.84 & s & f0v & 0.40&13 26 08.1 & -32 32 51 fm 191 & sao204853 & 9.1 & s & a2 & 0.07&13 48 10.1 & -32 14 18 fm 192 & sao204562 & 9.1 & s & f0 & 0.31&13 32 03.0 & -32 12 44 fm 193 & sao181770 & 9.00 & s & f8 & 0.54&13 35 05.1 & -25 18 15 fm 194 & hfd 119027 & 10.5 & s & a3 & 0.48&13 41 19.8 & -28 47 00 fm 195 & sao20462 & 8.0 & s & f3 & 0.46&13 35 15.9 & -31 24 31 fm 196 & gsc07266 - 00123 & 12.97 & s & ? & * 1.8&13 40 06.6 & -31 31 11 fm 197 & sao181785 & 9.6 & s & g5 & 0.64 & 13 36 13.4 & -29 39 40 ft 1 & ppm325630 & 10.57 & s & ? & 0.03 & 19 55 29.2 & -49 31 21 ft 2 & sao229911 & 8.97 & s & ? & 0.36 & 19 53 14.7 & -48 28 10 ft 3 & hic98444 & 6.67 & s & f2iv & 0.40 & 20 00 08.4 & -50 13 00 ft 4 & usno0375 - 38494969 & b12.8 & u & ? & * -0.5 & 19 54 53.6 & -48 32 33 ft 5 & hic98040 & 8.14 & s & f0v & 0.31 & 19 55 20.4 & -48 16 56 ft 6 & tyc8399 - 146 - 1 & 9.05 & s & ? & 0.49 & 20 02 07.3 & -49 58 54 ft 7 & usno0375 - 38670988 & b13.3 & u & ? & * -0.3 & 20 04 52.6 & -49 57 14 ft 8 & hic98449 & 7.67 & s & f6v & 0.47 & 20 00 11.8 & -48 44 54 ft 9 & sao229930 & 9.87 & s & a0 & 0.07 & 19 56 10.6 & -47 23 15 ft 10 & hic98880 & 8.55 & s & f0v & 0.37 & 20 04 47.8 & -49 45 18 ft 11 & usno0375 - 38647724 & b14.1 & u & ? & * 0.3 & 20 03 31.2 & -49 21 06 ft 12 & sao229910 & 8.19 & s & a2 & 0.14 & 19 53 09.0 & -46 22 12 ft 13 & sao229967 & 8.92 & s & a5 & 0.29 & 19 59 57.9 & -48 11 25 ft 14 & sao230028 & 9.6 & s & a2 & 0.03 & 20 05 07.1 & -49 30 17 ft 15 & sao246465 & 9.2 & s & a3 & 0.28 & 20 09 33.0 & -50 26 25 ft 16 & tyc8400 - 1697 - 1 & 9.56 & s & ? & 0.22 & 20 05 59.3 & -49 31 10 ft 17 & tyc8400 - 1353 - 1 & 11.1 & s & ? & 0.68 & 20 07 46.2 & -49 56 35 ft 18 & tyc8395 - 622 - 1 & 9.08 & s & f2 & 0.36 & 19 59 45.4 & -47 32 59 ft 19 & tyc8396 - 100 - 1 & 10.18 & s & ? & 0.36 & 20 03 51.2 & -48 37 04 ft 20 & sao229945 & 9.23 & s & a2 & 0.17 & 19 57 03.4 & -46 35 26 ft 21 & sao230036 & 10.42 & s & a0 & -0.05 & 20 06 17.2 & -48 59 39 ft 22 & tyc8396 - 580 - 1 & 9.66 & s & ? & 0.30 & 20 04 59.6 & -48 36 49 ft 23 & sao230006 & 10.12 & s & a3 & 0.16 & 20 02 52.5 & -47 58 52 ft 24 & hic98577 & 6.99 & s & f2iv & 0.35 & 20 01 25.3 & -47 23 51 ft 25 & hic99595 & 7.92 & s & f3v & 0.39 & 20 12 43.2 & -50 18 41 ft 26 & tyc8395 - 1024 - 1 & 10.5 & s & ? & 0.16 & 20 02 06.0 & -47 18 51 ft 27 & hic99703 & 8.42 & s & a2v & 0.12 & 20 13 54.9 & -50 19 16 ft 28 & usno0375 - 38688655 & b13.3 & u & ? & * -0.5 & 20 05 55.8 & -48 16 10 ft 29 & hic99260 & 8.83 & s & f2v & 0.38 & 20 08 58 & -49 04 42 ft 30 & usno0375 - 38800440 & b12.3 & u & ? & * -0.7 & 20 12 52.0 & -49 58 28 ft 31 & sao230049 & 7.95 & s & f2 & 0.43 & 20 07 39.5 & -48 39 08 ft 32 & ngc6851a / b & 14.81 & gp & sb & 0.84 & 20 05 39.7 & -47 58 40 ft 33 & hic98355 & 7.49 & s & f6v & 0.47 & 19 58 58.6 & -46 05 17 ft 34 & euvej2009 - 48.8 & 15.3 & qs & bl lac & * 0.6 & 20 09 25.4 & -48 49 54 ft 35 & sao230030 & 9.06 & s & a0 & 0.10 & 20 05 10.8 & -47 29 17 ft 36 & usno0375 - 38564499 & b11.6 & u & ? & * 1.0 & 19 58 43.0 & -45 36 31 ft 37 & tyc8417 - 667 - 1 & 10.02 & s & ? & 1.10 & 20 18 17.8 & -50 39 33 ft 38 & sao229984 & 8.78 & s & a0 & 0.00 & 20 00 53.4 & -45 41 58 ft 39 & sao230088 & 9.02 & s & a3 & 0.24 & 20 11 23.6 & -48 22 46 ft 40 & sao230060 & 8.09 & s & f0 & 0.29 & 20 08 30.1 & -47 33 34 ft 41 & tyc8413 - 1194 - 1 & 10.96 & s & ? & 0.58 & 20 19 24.6 & -50 12 41 ft 42 & sao230054 & 8.77 & s & a2 & 0.02 & 20 08 07.1 & -47 10 56 ft 43 & hic98512 & 5.8 & s & a8iii & 0.00 & 20 00 48.3 & -45 06 47 ft 44 & sao230121 & 9.75 & s & ? & 0.22 & 20 15 36.5 & -48 58 21 ft 45 & hic99130 & 8.72 & s & f0v & 0.39 & 20 07 31.0 & -46 50 50 ft 46 & hic100349 & 7.31 & s & a3v & 0.13 & 20 21 02.2 & -50 01 02 ft 47 & tyc8409 - 5 - 1 & 10.71 & s & ? & 1.2 & 20 15 15.6 & -48 43 43 ft 48 & tyc8392 - 1957 - 1 & 11.21 & s & ? & -0.05 & 20 06 24.1 & -46 04 05 ft 49 & usno0375 - 38880708 & b14.1 & u & ? & * -2.2 & 20 18 07.4 & -48 57 32 ft 50 & sao230000 & 7.91 & s & a0 & -0.03 & 20 02 10.2 & -44 27 58 ft 51 & sao230062 & 8.28 & s & f5v & 0.43 & 20 08 37.9 & -46 18 05 ft 52 & usno0375 - 38684700 & b15.7 & u & ? & * -0.3 & 20 05 41.5 & -45 33 25 ft 53 & tyc8413 - 818 - 1 & 8.03 & s & ? & 1.08 & 20 22 17.1 & -49 26 51 ft 54 & hic98859 & 7.83 & s & a9iv & 0.30 & 20 04 29.8 & -44 56 49 ft 55 & sao230181 & 9.54 & s & f5 & 0.35 & 20 23 36.6 & -49 47 44 ft 56 & sao230012 & 8.34 & s & a2 & 0.17 & 20 03 29.4 & -44 35 36 ft 57 & tyc08409 - 887 - 1 & b11.9 & u & ? & 0.58 & 20 18 29.1 & -48 32 11 ft 58 & usno0375 - 38762110 & b11.2 & u & ? & * 1.2 & 20 10 25.9 & -46 25 34 ft 59 & tyc8413 - 1203 - 1 & 10.68 & s & ? & 0.90 & 20 23 18.8 & -49 20 47 ft 60 & tyc8409 - 365 - 1 & 9.09 & s & ? & 0.05 & 20 20 07.0 & -48 28 41 ft 61 & sao230017 & 9.25 & s & a0 & 0.06 & 20 03 51.4 & -43 50 01 ft 62 & sao230111 & 9.68 & s & ? & 0.21 & 20 13 54.5 & -46 33 23 ft 63 & sao230099 & 9.07 & s & a0 & 0.19 & 20 12 22.3 & -46 10 00 ft 64 & usno - a2 - 0450 - 38582716 & b15.8 & u & ? & * 1.5 & 20 07 43.4 & -44 56 45 ft 65 & usno0375 - 38856050 & b12.8 & u & ? & * 1.1 & 20 16 32.2 & -47 06 59 ft 66 & hic100184 & 6.13 & s & f5v & 0.48 & 20 19 17.7 & -47 34 47 ft 67 & tyc8392 - 1398 - 1 & 10.42 & s & ? & 0.46 & 20 12 25.6 & -45 49 29 ft 68 & tyc7959 - 412 - 1 & 10.74 & s & ? & 0.28 & 20 09 38.3 & -44 44 52 ft 69 & tyc7959 - 1706 - 1 & 11 & s & ? & 0.19 & 20 08 59.3 & -44 30 19 ft 70 & hic99671 & 7.79 & s & ap & 0.00 & 20 13 34.8 & -45 35 26 ft 71 & tyc8409 - 1802 - 1 & 9.78 & s & ? & 0.05 & 20 21 56.8 & -47 43 25 ft 72 & sao230183 & 8.92 & s & f0 & 0.30 & 20 24 00.7 & -48 13 43 ft 73 & tyc8414 - 154 - 1 & 8.93 & s & ? & 0.15 & 20 30 20.6 & -49 26 29 ft 74 & tyc7959 - 1973 - 1 & 10.57 & s & ? & 0.34 & 20 09 39.5 & -44 18 56 ft 75 & tyc8405 - 1915 - 1 & 10.85 & s & ? & 0.10 & 20 18 15.0 & -46 31 36 ft 76 & hic100872 & 7.84 & s & f3iv & 0.43 & 20 27 11.1 & -48 35 35 ft 77 & sao230082 & 9.58 & s & a5 & 0.18 & 20 10 30.7 & -44 08 04 ft 78 & tyc8410 - 1739 - 1 & 8.41 & s & a7iv & 0.26 & 20 26 38.3 & -47 57 42 ft 79 & sao230096 & 9.26 & s & a0 & 0.10 & 20 12 03.3 & -44 15 42 ft 80 & hic99426 & 7.84 & s & a3 m & 0.25 & 20 10 50.2 & -43 36 11 ft 81 & usno - a2 - 0375 - 38789037 & b13.6 & u & ? & * -3.0 & 20 26 22.0 & -47 30 44 ft 82 & hic99448 & 7.71 & s & a4v & 0.22 & 20 11 04.7 & -43 28 28 ft 83 & hic100467 & 9.39 & s & f0v & 0.32 & 20 22 26.3 & -46 26 58 ft 84 & usno0450 - 38842338 & b14.4 & u & ? & * -0.4 & 20 12 41.5 & -43 50 53 ft 85 & hic99722 & 8.42 & s & f8iv & 0.44 & 20 14 06.4 & -44 10 24 ft 86 & sao230212 & 7.65 & s & f0 & 0.33 & 20 27 00.9 & -47 15 34 ft 87 & usno0375 - 39045904 & b13.3 & u & ? & * -0.8 & 20 29 34.1 & -47 47 26 ft 88 & usno0450 - 38892109 & b12.5 & u & ? & * 0.7 & 20 15 38.7 & -44 29 59 ft 89 & tyc7960 - 215 - 1 & 9.7 & s & a3 & 0.17 & 20 12 04.3 & -43 23 53 ft 90 & tyc8405 - 1234 - 1 & 9.78 & s & ? & 0.33 & 20 20 44.3 & -45 39 49 ft 91 & hic100787 & 6.73 & s & a9iii & 0.29 & 20 26 04.8 & -46 39 36 ft 92 & sao230170 & 8.65 & s & a0 & 0.09 & 20 21 57.2 & -45 38 20 ft 93 & hic99806 & 8.19 & s & a7v & 0.21 & 20 14 59.8 & -43 40 42 ft 94 & sao230230 & 9.36 & s & a2 & 0.20 & 20 28 23.9 & -46 57 21 ft 95 & tyc8410 - 743 - 1 & 11.44 & s & ? & 0.22 & 20 33 45.99 & -47 56 38 ft 96 & sao230200 & 8.95 & s & ? & 0.33 & 20 26 04.0 & -46 09 29 ft 97 & sao230238 & 7.42 & s & a0 & 0.05 & 20 29 38.6 & -46 09 05 ft 98 & hic100487 & 7.94 & s & a5 m & 0.34 & 20 22 38.1 & -44 35 14 ft 99 & tyc8406 - 29 - 1 & 10.71 & s & ? & 1.06 & 20 34 00.4 & -46 51 16 ft 100 & tyc8406 - 795 - 1 & 9.57 & s & ? & 0.33 & 20 29 53.0 & -45 54 48 ft 101 & sao230237 & 8.35 & s & ? & 0.23 & 20 29 30.5 & -45 22 58 ft 102 & sao230215 & 8.16 & s & f0 & 0.33 & 20 27 13.6 & -44 45 31 ft 103 & hic100528 & 8.39 & s & a6iv & 0.23 & 20 23 02.1 & -43 40 14 ft 104 & usno - a2 - 0450 - 38843012 & b13.6 & u & ? & * 1.1 & 20 22 45.4 & -43 29 43 ft 106 & tyc7961 - 435 - 1 & 9.89 & s & ? & 0.14 & 20 23 54.1 & -43 39 11 ft 107 & tyc7961 - 1763 - 1 & 10.5 & s & ? & 0.17 & 20 29 19.0 & -44 57 18 ft 108 & ngc6902 & 11.64 & ig & sa(r)b & 0.62 & 20 24 27.7 & -43 39 09 ft 109 & sao230295 & 10.23 & s & a0 & 0.03 & 20 36 54.6 & -46 27 22 ft 110 & tyc8406 - 1679 - 1 & 8.00 & s & f0 & 0.3 & 20 32 09.7 & -45 21 14 ft 111 & tyc8406 - 457 - 1/2 & 7.52 & s & a7iii & 0.28 & 20 36 30.5 & -45 33 28 ft 112 & usno0450 - 39114689 & b12.2 & u & ? & * 1.0 & 20 30 14.1 & -44 00 30 ft 113 & tyc8399 - 463 - 1 & 10.52 & s & ? & 0.17 & 20 01 32.4 & -50 17 46 ft 114 & usno0375 - 38638499 & b11.9 & u & ? & * 0.7 & 20 02 59.4 & -50 17 45 ft 115 & usno0375 - 38534372 & b13.7 & u & ? & * 1.4 & 19 57 01.8 & -48 18 58 ft 116 & usno - a2 - 0375 - 38402237 & b12.3 & u & ? & * 0.7 & 20 01 19.7 & -49 33 04 ft 117 & usno0375 - 38485210 & b13.9 & u & ? & ? & 19 54 22.9 & -47 37 44 ft 118 & usno0375 - 38502981 & b11.9 & u & ? & ? & 19 55 19.0 & -47 48 48 ft 119 & tyc8395 - 953 - 1 & 11.49 & s & ? & 0.04 & 19 57 16.1 & -48 21 29 ft 120 & tyc8400 - 336 - 1 & 10.97 & s & ? & 0.34 & 20 03 56.4 & -49 48 25 ft 121 & sao22994 & 9.61 & s & ? & 0.21 & 19 56 57.0 & -47 25 22 ft 122 & usno - a2 - 0375 - 38326701 & b12.4 & u & ? & * 1.0 & 19 57 01.1 & -47 03 45 ft 123 & tyc8391 - 843 - 1 & 10.47 & s & ? & 0.24 & 19 55 20.5 & -46 47 57 ft 124 & usno0375 - 38606749 & b13.8 & u & ? & * 0.5 & 20 01 08.3 & -48 28 31 ft 125 & tyc8391 - 2312 - 1 & 10.04 & s & ? & 0.40 & 19 54 35.9 & -46 30 37 ft 126 & usno0375 - 38593648 & b15.4 & u & ? & ? & 20 00 22.2 & -47 55 22 ft 127 & tyc8391 - 1762 - 1 & 10.69 & s & ? & 0.28 & 19 54 37.7 & -46 13 35 ft 128 & usno0375 - 38709646 & b12.3 & u & ? & * -0.4 & 20 07 11.4 & -49 30 04 ft 129 & tyc8404 - 1258 - 1 & 8.53 & s & ? & 0.56 & 20 12 02.9 & -50 43 23 ft 130 & tyc8395 - 1998 - 1 & 10.5 & s & ? & 0.83 & 19 59 33.0 & -47 04 30 ft 131 & ngc6845/a / b / c / fd & 14.28 & ig & sbb?pec & ? & 20 01 05.8 & -47 03 34 ft 132 & pgc063901 & 14.27 & gp & sab & * 0.2 & 19 58 17.6 & -46 06 46 ft 133 & sao98233 & 8.64 & s & f5iv & 0.01 & 19 57 45.2 & -45 49 39 ft 134 & usno0375 - 38806618 & b15.3 & u & ? & * 0.1 & 20 13 16.6 & -49 36 20 ft 135 & tyc8396 - 154 - 1 & 10.66 & s & ? & 0.24 & 20 09 52.3 & -48 34 30 ft 136 & tyc8391 - 677 - 1 & 10.85 & s & ? & 0.59 & 19 58 53.2 & -45 05 24 ft 137 & usno - a2 - 0375 - 38425789 & b11.1 & u & ? & * 0.5 & 20 02 43.4 & -45 57 22 ft 138 & usno - a2 - 0375 - 38637967 & b13.6 & u & ? & * 0.2 & 20 15 59.7 & -48 51 28 ft 139 & tyc8392 - 28 - 1 & 10.59 & s & ? & 0.15 & 20 07 57.6 & -46 43 45 ft 140 & tyc8413 - 614 - 1 & 9.98 & s & ? & 0.35 & 20 20 38.8 & -49 49 00 ft 141 & usno0375 - 38888302 & b12.7 & u & ? & * -0.8 & 20 18 36.2 & -49 13 07 ft 142 & tyc8396 - 418 - 1 & 10.63 & s & ? & 0.91 & 20 14 43.0 & -48 12 50 ft 143 & usno0375 - 38910011 & b13.4 & u & ? & * 0.8 & 20 20 01.1 & -49 19 57 ft 144 & tyc8413 - 1738 - 1 & 10 & s & ? & 0.34 & 20 24 36.6 & -50 25 20 ft 145 & tyc8392 - 1262 - 1 & 10.06 & s & ? & 0.65 & 20 06 16.8 & -45 54 16 ft 146 & eso234-g004 & 15.27 & gp & sbpec & ? & 20 20 43.4 & -49 07 29 ft 147 & usno - a2 - 0375 - 38528298 & b15.8 & u & ? & * -0.2 & 20 08 55.8 & -46 05 47 ft 148 & usno0375 - 38748771 & b11.3 & u & ? & * 0.3 & 20 09 35.9 & -46 07 28 ft 149 & usno0375 - 38753431 & b12.6 & u & ? & * 1.0 & 20 09 53.4 & -46 03 28 ft 150 & tyc7959 - 742 - 1 & 11.12 & s & ? & 0.15 & 20 05 11.9 & -44 39 32 ft 151 & usno0450 - 38690235 & b13.5 & u & ? & * 1.2 & 20 04 58.4 & -44 29 13 ft 152 & usno - a2 - 0375 - 38561392 & b15.7 & u & ? & * 0.0 & 20 10 59.6 & -45 53 50 ft 153 & usno0375 - 38758273 & 13.3 & u & ? & * 0.8 & 20 10 11.8 & -45 41 40 ft 154 & usno0375 - 38732664 & b15.9 & u & ? & * -0.5 & 20 08 36.3 & -45 05 16 ft 155 & usno0375 - 38963244 & b12.7 & u & ? & * 1.4 & 20 23 35.0 & -48 49 06 ft 156 & usno - a2 - 0450 - 38583089 & b17.7 & u & ? & * 0.1 & 20 07 44.6 & -44 37 38 ft 157 & usno0450 - 38718073 & b14.7 & u & ? & * -0.2 & 20 06 21.4 & -44 03 01 ft 158 & tyc8405 - 885 - 1 & 10.36 & s & ? & 0.37 & 20 15 24.7 & -46 19 50 ft 159 & usno0450 - 38794373 & b14.1 & u & ? & * 1.0 & 20 10 16.8 & -44 52 15 ft 160 & sao230055 & 8.49 & s & f5 & 0.441 & 20 08 01.3 & -44 11 33 ft 161 & eso284-ig045 & 14 & ig & sabbc & ? & 20 16 48.6 & -46 31 55 ft 162 & sao230119 & 9.75 & s & f5 & 0.38 & 20 15 12.5 & -46 03 12 ft 163 & tyc7959 - 962 - 1 & 10.18 & s & ? & 0.44 & 20 07 49.5 & -43 37 54 ft 164 & usno0375 - 39013302 & b14.4 & u & ? & * 0.9 & 20 27 11.4 & -48 19 47 ft 165 & usno - a2 - 0375 - 38776135 & b12.0 & u & ? & * 0.0 & 20 25 27.5 & -47 51 23 ft 166 & tyc7960 - 1677 - 1 & 9.77 & s & ? & 0.31 & 20 14 09.1 & -44 51 02 ft 167 & tyc8414 - 685 - 1 & 10.21 & s & ? & 0.36 & 20 34 24.2 & -49 05 05 ft 168 & usno0375 - 38887188 & b15.1 & u & ? & ? & 20 18 32.0 & -45 20 59 ft 169 & pgc064404 & 14.31 & pm & sabbcp & 0.95 & 20 16 45.5 & -44 41 43 ft 169 & pgc064399 & 14.51 & pm & sbdm & 1.17 & 20 16 31.0 & -44 42 32 ft 170 & sao230206 & 9.23 & s & f2 & 0.37 & 20 26 48.6 & -46 26 34 ft 171 & tyc8410 - 1555 - 1 & 11.42 & s & ? & 0.37 & 20 35 23.8 & -48 19 20 ft 172 & usno0375 - 38983401 & b15.2 & u & ? & * 1.1 & 20 25 03.5 & -45 42 55 ft 173 & tyc8406 - 1146 - 1 & 11.11 & s & ? & 0.07 & 20 29 17.2 & -46 27 58 ft 174 & hic101168 & 9.17 & s & f5v & 0.41 & 20 30 22.4 & -46 27 20 ft 175 & usno0375 - 39019592 & b14.2 & u & ? & * 1.3 & 20 27 39.3 & -45 40 49 ft 176 & usno0450 - 38948511 & b12.6 & u & ? & * 0.5 & 20 19 12.5 & -43 34 02 ft 177 & usno0450 - 39036632 & b13.6 & u & ? & * 1.4 & 20 24 53.5 & -44 37 31 ft 178 & sao230167 & 9.75 & s & f0 & 0.47 & 20 21 34.2 & -43 49 17 ft 179 & usno0375 - 39120597 & b12.8 & u & ? & * 0.8 & 20 35 14.8 & -46 58 10 ft 180 & usno0450 - 39039977 & b15.9 & u & ? & * 0.8 & 20 25 06.8 & -44 27 39 ft 181 & usno - a2 - 0375 - 38811091 & b13.9 & u & ? & * 1.2 & 20 27 58.9 & -45 00 52 ft 182 & tyc8406 - 452 - 1 & 10.59 & s & ? & 0.36 & 20 29 12.5 & -45 07 14 ft 183 & tyc7962 - 710 - 1 & 9.85 & s & ? & 0.52 & 20 31 12.8 & -44 45 57 ft 184 & tyc8414 - 669 - 1 & 8.46 & s & a2vn & 0.10 & 20 29 26.78 & -49 46 59 ft 185 & tyc8414 - 387 - 1 & 10.66 & s & ? & -0.12 & 20 29 48.9 & -49 52 24 ft 186 & hic97756 & 8.31 & s & a7iiim & 0.28 & 19 51 58.9 & -46 51 42 ft 187 & tyc7960 - 817 - 1 & 9.72 & s & ? & 0.38 & 20 15 08.2 & -43 34 14 ft 188 & sao230029 & 9.67 & s & f0 & 0.43 & 20 05 08.9 & -47 21 42 ft 189 & usno0450 - 39025207 & b13.2 & u & ? & * 2.0 & 20 24 07.9 & -43 41 10 lllccccc fc 27 & eso 323-g025 & sa(rs)b & 12.16@xmath230.20 & 13.39 & 1.17 & fc 32 & ngc 4553 & sa(r)0 & 12.51@xmath230.58 & 12.17 & 1.02 & fc 44 & usno & sab(r)cd & 10.66@xmath230.26 & 9.9 & * -1.3 & fc 61 & pgc 042966 & e ? & 13.06@xmath230.57 & 12.72 & 1.08 & fc 71 & ngc 4683 & sb(s)0 & 12.58@xmath230.43 & 13.8 & 0.93 & fc 75 & ngc 4930 & sb(rs)bc & 12.65@xmath230.44 & 12.0 & 0.76 & fc 161 & ngc 4679 & sa(s)bc & 13.32@xmath230.45 & 11.94 & 1.38 & fc 186 & eso 323-g054 & s & 13.68@xmath230.61 & 14.86 & & fc 229 & eso 269-ig056 & sb(s)m pec & 11.92@xmath230.46 & 14.01 & & fc 235 & eso 269-ig022/3 & s0+e & 12.74@xmath230.34 & 14.75 + 16.17 & & int.pair fc 238 & pgc044774 & sab(s ) pec & 12.58@xmath230.61 & 16.0 & & fd 41 & ngc 1566 & sab(rs)bc sy1 & 9.68@xmath230.19 & 10.33 & 0.56 & xrays fd 53 & ngc 1602 & ib(s)m pec & 12.29@xmath230.48 & 12.98 & 0.35 & fd 62 & euvej0425 - 57.1 & sy1 & 12.03@xmath230.36 & 14.1 & * -0.6 & xrays fd 77 & eso 118-g034 & s0 pec & 11.64@xmath230.39 & 13.04 & 0.44 & hii gal . fd 82 & ngc 1522 & s0 pec & 12.62@xmath230.57 & 13.56 & 0.27 & fd 86 & ngc 1515 & sab(s)bc & 12.90@xmath230.67 & 11.20 & 0.85 & fd 93 & ic 2032 & iab(s)m pec & 12.89@xmath230.68 & 14.73 & * -3.6 & fd 97 & ngc 1536 & sb(s)c pec & 12.92@xmath230.58 & 12.52 & 0.63 & fd 100 & ic 2073 & sb(s)cd ? pec & 11.73@xmath230.37 & 13.78 & 0.77 & fd 111 & pgc 015445 & e & 14.16@xmath232.76 & 15.0 & * 0.7 & fd 118 & eso 157-g044 & ib(s)m & 12.19@xmath230.60 & 14.87 & & fd 134 & he0442 - 5652 & qso z=0.34 & 12.69@xmath230.59 & 16.3 & * 0.2 & fd 142b & usno & qso z=0.29 & 11.55@xmath230.25 & 15.6 & * -0.2 & iras fsc 25 + 60 fm 28 & ngc 5291/b & s pec & 12.57@xmath230.47 & 14.0 & & int.pair fm 49 & ngc 5264 & ib(s)m & 11.14@xmath230.23 & 12.00 & 0.60 & ddo 242 fm 55 & he1337 - 2919 & qso & 12.02@xmath230.32 & 16.06 & & candidate fm 58 & ngc 5253 & i m pec & 9.89@xmath230.19 & 10.87 & 0.38 & * burst , iue fm 69 & he1334 - 2739 & qso & 12.59@xmath230.46 & 16.41 & & candidate fm 70 & ngc 5236=m83 & sab(s)c & 7.92@xmath230.17 & 7.54 & 0.66 & * burst ; xray ; iue fm 71 & ic 4293 & sa0 & 12.21@xmath230.38 & 13.50 & * 1.5 & fm 104 & gsc07265 - 02231 & em.ln.gal . & 11.77@xmath230.29 & 12.7 & * 0.0 & abell 3558 z@xmath160.04 ? fm 112 & ic 4248 & s ? & 12.87@xmath230.50 & 13.85 & * 0.7 & fm 124 & eso 445-g076 & s & 12.83@xmath230.69 & 14.66 & * 0.6 & fm 126 & he1348 - 2751 & qso & 12.22@xmath230.43 & 16.62 & & candidate & fm 130 & he1348 - 2751 & qso & 12.45@xmath230.46 & 16.62 & & candidate & fm 132 & ic 4327 & sb(s)c ? & 13.62@xmath230.85 & 13.89 & 0.61 & fm 141 & ic 4318 & s ? & 13.50@xmath230.46 & 14.54 & & fm 142 & eso - lv4450171 & s & 14.07@xmath230.71 & 17.03 & & fm 152b & he1336 - 2624 & qso & 13.04@xmath230.60 & 16.53 & & candidate fm 163 & eso 444-g075 & i ? & 13.74@xmath230.58 & 14.27 & * 0.7 & fm 175 & eso 444-g046 & cd & 13.91@xmath230.56 & 13.85 & & abell 3558 z@xmath160.04 ? fm 176 & eso 444-g037 & sb(s)dm : & 13.63@xmath230.79 & 14.82 & * 0.6 & lsb ; ugca 358 fm 181 & ngc 5135 & sb(l)ab & 13.84@xmath231.01 & 12.11 & 0.77 & tol 34 ; * burst+sy2 ; iue fm 185 & he1317 - 2937 & em.ln.gal . & 13.64@xmath231.23 & 17.79 & & ft 32 & ngc 6851 a+b & sb ( rs)d+sb(s)m & 10.08@xmath230.20 & 14.81 + 16.67 & 0.84 & ft 34 & euvej2009 - 48.8 & qso & 11.56@xmath230.34 & 15.3 & * 0.6 & bl lac z=0.071 ; xray ft 108 & ngc 6920 & sa0 & 11.75@xmath230.33 & 12.35 & 1.22 & ft 131 & ngc 6845 abcd & group & 12.50@xmath230.54 & 14.28 & & interacting ft 132 & pgc 063901 & sab(rs)bc & 13.46@xmath230.73 & 14.27 & * 0.2 & ft 146 & eso 234-g004 & sb(s)dm pec & 12.77@xmath230.66 & 15.27 & & ft 161 & eso 284-ig045 & sab(rs)bc : & 13.05@xmath230.74 & 13.32 & 0.68 & ft 169a & pgc 064404 & sab(s)bc pec & 12.44@xmath230.52 & 14.31 & 0.95 & nelg ft 169b & pgc 064399 & sb(s)dm & 12.44@xmath230.52 & 14.51 & 1.17 & Abstract:

Write an abstract for the following acadamic paper.

Paper: we have recorded the flight of a fly during take off and landing using digital high speed photography . it is shown that the dynamics of flexible wings are different for these procedures . during this observation fly flew freely in a big box and it was not tethered . Abstract: in this fluid dynamics video , we demonstrated take off and landing of a fly . the deformation of wings is in focus in this video . Paper: the origin of graph theory dates back to euler s solution of the puzzle of knigsberg s bridges in 1736 @xcite . since then a lot has been learned about graphs and their mathematical properties @xcite . in the 20th century they have also become extremely useful as representation of a wide variety of systems in different areas . biological , social , technological , and information networks can be studied as graphs , and graph analysis has become crucial to understand the features of these systems . for instance , social network analysis started in the 1930 s and has become one of the most important topics in sociology @xcite . in recent times , the computer revolution has provided scholars with a huge amount of data and computational resources to process and analyze these data . the size of real networks one can potentially handle has also grown considerably , reaching millions or even billions of vertices . the need to deal with such a large number of units has produced a deep change in the way graphs are approached @xcite . graphs representing real systems are not regular like , e. g. , lattices . they are objects where order coexists with disorder . the paradigm of disordered graph is the random graph , introduced by p. erds and a. rnyi @xcite . in it , the probability of having an edge between a pair of vertices is equal for all possible pairs ( see appendix ) . in a random graph , the distribution of edges among the vertices is highly homogeneous . for instance , the distribution of the number of neighbours of a vertex , or _ degree _ , is binomial , so most vertices have equal or similar degree . real networks are not random graphs , as they display big inhomogeneities , revealing a high level of order and organization . the degree distribution is broad , with a tail that often follows a power law : therefore , many vertices with low degree coexist with some vertices with large degree . furthermore , the distribution of edges is not only globally , but also locally inhomogeneous , with high concentrations of edges within special groups of vertices , and low concentrations between these groups . this feature of real networks is called _ community structure _ @xcite , or _ clustering _ , and is the topic of this review ( for earlier reviews see refs . communities , also called _ clusters _ or _ modules _ , are groups of vertices which probably share common properties and/or play similar roles within the graph . in fig . [ figure1 ] a schematic example of a graph with communities is shown . a simple graph with three communities , enclosed by the dashed circles . reprinted figure with permission from ref . @xcite . 2009 by springer . ] society offers a wide variety of possible group organizations : families , working and friendship circles , villages , towns , nations . the diffusion of internet has also led to the creation of virtual groups , that live on the web , like online communities . indeed , social communities have been studied for a long time @xcite . communities also occur in many networked systems from biology , computer science , engineering , economics , politics , etc . in protein - protein interaction networks , communities are likely to group proteins having the same specific function within the cell @xcite , in the graph of the world wide web they may correspond to groups of pages dealing with the same or related topics @xcite , in metabolic networks they may be related to functional modules such as cycles and pathways @xcite , in food webs they may identify compartments @xcite , and so on . communities can have concrete applications . clustering web clients who have similar interests and are geografically near to each other may improve the performance of services provided on the world wide web , in that each cluster of clients could be served by a dedicated mirror server @xcite . identifying clusters of customers with similar interests in the network of purchase relationships between customers and products of online retailers ( like , e. g. , www.amazon.com ) enables to set up efficient recommendation systems @xcite , that better guide customers through the list of items of the retailer and enhance the business opportunities . clusters of large graphs can be used to create data structures in order to efficiently store the graph data and to handle navigational queries , like path searches @xcite . _ ad hoc networks _ @xcite , i. e. self - configuring networks formed by communication nodes acting in the same region and rapidly changing ( because the devices move , for instance ) , usually have no centrally maintained routing tables that specify how nodes have to communicate to other nodes . grouping the nodes into clusters enables one to generate compact routing tables while the choice of the communication paths is still efficient @xcite . community detection is important for other reasons , too . identifying modules and their boundaries allows for a classification of vertices , according to their structural position in the modules . so , vertices with a central position in their clusters , i. e. sharing a large number of edges with the other group partners , may have an important function of control and stability within the group ; vertices lying at the boundaries between modules play an important role of mediation and lead the relationships and exchanges between different communities ( alike to csermely s `` creative elements '' @xcite ) . such classification seems to be meaningful in social @xcite and metabolic networks @xcite . finally , one can study the graph where vertices are the communities and edges are set between clusters if there are connections between some of their vertices in the original graph and/or if the modules overlap . in this way one attains a coarse - grained description of the original graph , which unveils the relationships between modules . recent studies indicate that networks of communities have a different degree distribution with respect to the full graphs @xcite ; however , the origin of their structures can be explained by the same mechanism @xcite . another important aspect related to community structure is the hierarchical organization displayed by most networked systems in the real world . real networks are usually composed by communities including smaller communities , which in turn include smaller communities , etc . the human body offers a paradigmatic example of hierarchical organization : it is composed by organs , organs are composed by tissues , tissues by cells , etc . another example is represented by business firms , who are characterized by a pyramidal organization , going from the workers to the president , with intermediate levels corresponding to work groups , departments and management . herbert a. simon has emphasized the crucial role played by hierarchy in the structure and evolution of complex systems @xcite . the generation and evolution of a system organized in interrelated stable subsystems are much quicker than if the system were unstructured , because it is much easier to assemble the smallest subparts first and use them as building blocks to get larger structures , until the whole system is assembled . in this way it is also far more difficult that errors ( mutations ) occur along the process . the aim of community detection in graphs is to identify the modules and , possibly , their hierarchical organization , by only using the information encoded in the graph topology . the problem has a long tradition and it has appeared in various forms in several disciplines . the first analysis of community structure was carried out by weiss and jacobson @xcite , who searched for work groups within a government agency . the authors studied the matrix of working relationships between members of the agency , which were identified by means of private interviews . work groups were separated by removing the members working with people of different groups , which act as connectors between them . this idea of cutting the bridges between groups is at the basis of several modern algorithms of community detection ( section [ sec5 ] ) . research on communities actually started even earlier than the paper by weiss and jacobson . already in 1927 , stuart rice looked for clusters of people in small political bodies , based on the similarity of their voting patterns @xcite . two decades later , george homans showed that social groups could be revealed by suitably rearranging the rows and the columns of matrices describing social ties , until they take an approximate block - diagonal form @xcite . this procedure is now standard . meanwhile , traditional techniques to find communities in social networks are hierarchical clustering and partitional clustering ( sections [ sec4_2 ] and [ sec4_3 ] ) , where vertices are joined into groups according to their mutual similarity . identifying graph communities is a popular topic in computer science , too . in parallel computing , for instance , it is crucial to know what is the best way to allocate tasks to processors so as to minimize the communications between them and enable a rapid performance of the calculation . this can be accomplished by splitting the computer cluster into groups with roughly the same number of processors , such that the number of physical connections between processors of different groups is minimal . the mathematical formalization of this problem is called _ graph partitioning _ ( section [ sec4_1 ] ) . the first algorithms for graph partitioning were proposed in the early 1970 s . in a seminal paper appeared in 2002 , girvan and newman proposed a new algorithm , aiming at the identification of edges lying between communities and their successive removal , a procedure that after some iterations leads to the isolation of the communities @xcite . the intercommunity edges are detected according to the values of a centrality measure , the edge betweenness , that expresses the importance of the role of the edges in processes where signals are transmitted across the graph following paths of minimal length . the paper triggered a big activity in the field , and many new methods have been proposed in the last years . in particular , physicists entered the game , bringing in their tools and techniques : spin models , optimization , percolation , random walks , synchronization , etc . , became ingredients of new original algorithms . the field has also taken advantage of concepts and methods from computer science , nonlinear dynamics , sociology , discrete mathematics . in this manuscript we try to cover in some detail the work done in this area . we shall pay a special attention to the contributions made by physicists , but we shall also give proper credit to important results obtained by scholars of other disciplines . section [ sec2 ] introduces communities in real networks , and is supposed to make the reader acquainted with the problem and its relevance . in section [ sec3 ] we define the basic elements of community detection , i. e. the concepts of community and partition . traditional clustering methods in computer and social sciences , i. e. graph partitioning , hierarchical , partitional and spectral clustering are reviewed in section [ sec4 ] . modern methods , divided into categories based on the type of approach , are presented in sections [ sec5 ] to [ sub_sec6_2 ] . algorithms to find overlapping communities , multiresolution and hierarchical techniques , are separately described in sections [ sec45 ] and [ sub_sec6_20 ] , respectively , whereas section [ sec7_1_2 ] is devoted to the detection of communities evolving in time . we stress that our categorization of the algorithms is not sharp , because many algorithms may enter more categories : we tried to classify them based on what we believe is their main feature / purpose , even if other aspects may be present . sections [ sec6_4 ] and [ sec6 ] are devoted to the issues of defining when community structure is significant , and deciding about the quality of algorithms performances . in sections [ sec7_1 ] and [ sec7_2 ] we describe general properties of clusters found in real networks , and specific applications of clustering algorithms . section [ sec8 ] contains the summary of the review , along with a discussion about future research directions in this area . the review makes use of several concepts of graph theory , that are defined and explained in the appendix . readers not acquainted with these concepts are urged to read the appendix first . in this section we shall present some striking examples of real networks with community structure . in this way we shall see what communities look like and why they are important . social networks are paradigmatic examples of graphs with communities . the word community itself refers to a social context . people naturally tend to form groups , within their work environment , family , friends . in fig . [ fig3 ] we show some examples of social networks . the first example ( fig . [ fig3]a ) is zachary s network of karate club members @xcite , a well - known graph regularly used as a benchmark to test community detection algorithms ( section [ sec6_1 ] ) . it consists of @xmath0 vertices , the members of a karate club in the united states , who were observed during a period of three years . edges connect individuals who were observed to interact outside the activities of the club . at some point , a conflict between the club president and the instructor led to the fission of the club in two separate groups , supporting the instructor and the president , respectively ( indicated by squares and circles ) . the question is whether from the original network structure it is possible to infer the composition of the two groups . indeed , by looking at fig . [ fig3]a one can distinguish two aggregations , one around vertices @xmath1 and @xmath0 ( @xmath0 is the president ) , the other around vertex @xmath2 ( the instructor ) . one can also identify several vertices lying between the two main structures , like @xmath3 , @xmath4 , @xmath5 ; such vertices are often misclassified by community detection methods . [ fig3]b displays the largest connected component of a network of collaborations of scientists working at the santa fe institute ( sfi ) . there are @xmath6 vertices , representing resident scientists at sfi and their collaborators . edges are placed between scientists that have published at least one paper together . the visualization layout allows to distinguish disciplinary groups . in this network one observes many cliques , as authors of the same paper are all linked to each other . there are but a few connections between most groups . in fig . [ fig3]c we show the network of bottlenose dolphins living in doubtful sound ( new zealand ) analyzed by lusseau @xcite . there are @xmath7 dolphins and edges were set between animals that were seen together more often than expected by chance . the dolphins separated in two groups after a dolphin left the place for some time ( squares and circles in the figure ) . such groups are quite cohesive , with several internal cliques , and easily identifiable : only six edges join vertices of different groups . due to this natural classification lusseau s dolphins network , like zachary s karate club , is often used to test algorithms for community detection ( section [ sec6_1 ] ) . protein - protein interaction ( ppi ) networks are subject of intense investigations in biology and bioinformatics , as the interactions between proteins are fundamental for each process in the cell @xcite . [ fig4 ] illustrates a ppi network of the rat proteome @xcite . each interaction is derived by homology from experimentally observed interactions in other organisms . in our example , the proteins interact very frequently with each other , as they belong to metastatic cells , which have a high motility and invasiveness with respect to normal cells . communities correspond to functional groups , i. e. to proteins having the same or similar functions , which are expected to be involved in the same processes . the modules are labeled by the overall function or the dominating protein class . most communities are associated to cancer and metastasis , which indirectly shows how important detecting modules in ppi networks is . community structure in technological networks . sample of the web graph consisting of the pages of a web site and their mutual hyperlinks , which are directed . communities , indicated by the colors , were detected with the algorithm of girvan and newman ( section [ subsec5_1 ] ) , by neglecting the directedness of the edges . reprinted figure with permission from ref . @xcite . 2004 by the american physical society . ] relationships / interactions between elements of a system need not be reciprocal . in many cases they have a precise direction , that needs to be taken into account to understand the system as a whole . as an example we can cite predator - prey relationships in food webs . in fig . [ fig5 ] we see another example , taken from technology . the system is the world wide web , which can be seen as a graph by representing web pages as vertices and the hyperlinks that make users move from one page to another as edges @xcite . hyperlinks are directed : if one can move from page a to page b by clicking on a hyperlink of a , one usually does not find on b a hyperlink taking back to a. in fact , very few hyperlinks ( less than @xmath8 ) are reciprocal . communities of the web graph are groups of pages having topical similarities . detecting communities in the web graph may help to identify the artificial clusters created by link farms in order to enhance the pagerank @xcite value of web sites and grant them a higher google ranking . in this way one could discourage this unfair practice . one usually assumes that the existence of a hyperlink between two pages implies that they are content - related , and that this similarity is independent of the hyperlink direction . therefore it is customary to neglect the directedness of the hyperlinks and to consider the graph as undirected , for the purpose of community detection . on the other hand , taking properly into account the directedness of the edges can considerably improve the quality of the partition(s ) , as one can handle a lot of precious information about the system . moreover , in some instances neglecting edge directedness may lead to strange results @xcite . developing methods of community detection for directed graphs is a hard task . for instance , a directed graph is characterized by asymmetrical matrices ( adjacency matrix , laplacian , etc . ) , so spectral analysis is much more complex . only a few techniques can be easily extended from the undirected to the directed case . otherwise , the problem must be formulated from scratch . edge directedness is not the only complication to deal with when facing the problem of graph clustering . in many real networks vertices may belong to more than one group . in this case one speaks of _ overlapping communities _ and uses the term _ cover _ , rather than partition , whose standard definition forbids multiple memberships of vertices . classical examples are social networks , where an individual usually belongs to different circles at the same time , from that of work colleagues to family , sport associations , etc .. traditional algorithms of community detection assign each vertex to a single module . in so doing , they neglect potentially relevant information . vertices belonging to more communities are likely to play an important role of intermediation between different compartments of the graph . in fig . [ fig6 ] we show a network of word association derived starting from the word `` bright '' . the network builds on the university of south florida free association norms @xcite . an edge between words @xmath9 and @xmath10 indicates that some people associate @xmath10 to the word @xmath9 . the graph clearly displays four communities , corresponding to the categories _ intelligence _ , _ astronomy _ , _ light _ and _ colors_. the word `` bright '' is related to all of them by construction . other words belong to more categories , e.g. `` dark '' ( colors and light ) . accounting for overlapping communities introduces a further variable , the membership of vertices in different communities , which enormously increases the number of possible covers with respect to standard partitions . therefore , searching for overlapping communities is usually much more computationally demanding than detecting standard partitions . overlapping communities in a network of word association . the groups , labeled by the colors , were detected with the clique percolation method by palla et al . ( section [ sec45_1 ] ) . reprinted figure with permission from ref . 2005 by the nature publishing group . ] so far we have discussed examples of unipartite graphs . however , it is not uncommon to find real networks with different classes of vertices , and edges joining only vertices of different classes . an example is a network of scientists and papers , where edges join scientists and the papers they have authored . here there is no edge between any pair of scientists or papers , so the graph is bipartite . for a multipartite network the concept of community does not change much with respect to the case of unipartite graphs , as it remains related to a large density of edges between members of the same group , with the only difference that the elements of each group belong to different vertex classes . multipartite graphs are usually reduced to unipartite projections of each vertex class . for instance , from the bipartite network of scientists and papers one can extract a network of scientists only , who are related by coauthorship . in this way one can adopt standard techniques of network analysis , in particular standard clustering methods , but a lot of information gets lost . detecting communities in multipartite networks can have interesting applications in , e.g. , marketing . large shopping networks , in which customers are linked to the products they have bought , allow to classify customers based on the types of product they purchase more often : this could be used both to organize targeted advertising , as well as to give recommendations about future purchases @xcite . the problem of community detection in multipartite networks is not trivial , and usually requires _ ad hoc _ methodologies . [ fig7 ] illustrates the famous bipartite network of southern women studied by davis et al . there are @xmath11 vertices , representing @xmath12 women from the area of natchez , mississippi , and @xmath13 social events . edges represent the participation of the women in the events . from the figure one can see that the network has a clear community structure . community structure in multipartite networks . this bipartite graph refers to the southern women event participation data set . women are represented as open symbols with black labels , events as filled symbols with white labels . the illustrated vertex partition has been obtained by maximizing a modified version of the modularity by newman and girvan , tailored on bipartite graphs @xcite ( section [ sub_sec6_01 ] ) . reprinted figure with permission from ref . 2007 by the american physical society . ] in some of the previous examples , edges have ( or can have ) weights . for instance , the edges of the collaboration network of fig . [ fig3]b could be weighted by the number of papers coauthored by pairs of scientists . similarly , the edges of the word association network of fig . [ fig6 ] are weighted by the number of times pairs of words have been associated by people . weights are precious additional information on a graph , and should be considered in the analysis . in many cases methods working on unweighted graphs can be simply extended to the weighted case . the problem of graph clustering , intuitive at first sight , is actually not well defined . the main elements of the problem themselves , i. e. the concepts of community and partition , are not rigorously defined , and require some degree of arbitrariness and/or common sense . indeed , some ambiguities are hidden and there are often many equally legitimate ways of resolving them . therefore , it is not surprising that there are plenty of recipes in the literature and that people do not even try to ground the problem on shared definitions . it is important to stress that the identification of structural clusters is possible only if graphs are _ sparse _ , i. e. if the number of edges @xmath14 is of the order of the number of nodes @xmath15 of the graph . if @xmath16 , the distribution of edges among the nodes is too homogeneous for communities to make sense . in this case the problem turns into something rather different , close to data clustering @xcite , which requires concepts and methods of a different nature . the main difference is that , while communities in graphs are related , explicitly or implicitly , to the concept of edge density ( inside versus outside the community ) , in data clustering communities are sets of points which are `` close '' to each other , with respect to a measure of distance or similarity , defined for each pair of points . some classical techniques for data clustering , like _ hierarchical _ , _ partitional _ and _ spectral clustering _ will be discussed later in the review ( sections [ sec4_2 ] , [ sec4_3 ] and [ sec4_4 ] ) , as they are sometimes adopted for graph clustering too . other standard procedures for data clustering include _ neural network clustering _ techniques like , e. g. , _ self - organizing maps _ and _ multi - dimensional scaling _ techniques like , e. g. , _ singular value decomposition _ and _ principal component analysis _ @xcite . in this section we shall attempt an ordered exposition of the fundamental concepts of community detection . after a brief discussion of the issue of computational complexity for the algorithms , we shall review the notions of community and partition . the massive amount of data on real networks currently available makes the issue of the efficiency of clustering algorithms essential . the _ computational complexity _ of an algorithm is the estimate of the amount of resources required by the algorithm to perform a task . this involves both the number of computation steps needed and the number of memory units that need to be simultaneously allocated to run the computation . such demands are usually expressed by their scalability with the size of the system at study . in the case of a graph , the size is typically indicated by the number of vertices @xmath15 and/or the number of edges @xmath14 . the computational complexity of an algorithm can not always be calculated . in fact , sometimes this is a very hard task , or even impossible . in these cases , it is however important to have at least an estimate of the _ worst - case _ complexity of the algorithm , which is the amount of computational resources needed to run the algorithm in the most unfavorable case for a given system size . the notation @xmath17 indicates that the computer time grows as a power of both the number of vertices and edges , with exponents @xmath18 and @xmath19 , respectively . ideally , one would like to have the lowest possible values for the exponents , which would correspond to the lowest possible computational demands . samples of the web graph , with millions of vertices and billions of edges , can not be tackled by algorithms whose running time grows faster than @xmath20 or @xmath21 . algorithms with polynomial complexity form the class * p*. for some important decision and optimization problems , there are no known polynomial algorithms . finding solutions of such problems in the worst - case scenario may demand an exhaustive search , which takes a time growing faster than any polynomial function of the system size , e.g. exponentially . problems whose solutions can be verified in a polynomial time span the class * np * of _ non - deterministic polynomial time _ problems , which includes * p*. a problem is * np*-hard if a solution for it can be translated into a solution for any * np*-problem . however , a * np*-hard problem needs not be in the class * np*. if it does belong to * np * it is called * np*-complete . the class of * np*-complete problems has drawn a special attention in computer science , as it includes many famous problems like the travelling salesman , boolean satisfiability ( sat ) , linear programming , etc . the fact that * np * problems have a solution which is verifiable in polynomial time does not mean that * np * problems have polynomial complexity , i. e. , that they are in * p*. in fact , the question of whether * np*=*p * is the most important open problem in theoretical computer science . * np*-hard problems need not be in * np * ( in which case they would be * np*-complete ) , but they are at least as hard as * np*-complete problems , so they are unlikely to have polynomial complexity , although a proof of that is still missing . many clustering algorithms or problems related to clustering are * np*-hard . in this case , it is pointless to use exact algorithms , which could be applied only to very small systems . moreover , even if an algorithm has a polynomial complexity , it may still be too slow to tackle large systems of interest . in all such cases it is common to use _ approximation algorithms _ , i. e. methods that do not deliver an exact solution to the problem at hand , but only an approximate solution , with the advantage of a lower complexity . approximation algorithms are often non - deterministic , as they deliver different solutions for the same problem , for different initial conditions and/or parameters of the algorithm . the goal of such algorithms is to deliver a solution which differs by a constant factor from the optimal solution . in any case , one should give provable bounds on the goodness of the approximate solution delivered by the algorithm with respect to the optimal solution . in many cases it is not possible to approximate the solution within any constant , as the goodness of the approximation strongly depends on the specific problem at study . approximation algorithms are commonly used for optimization problems , in which one wants to find the maximum or minimum value of a given cost function over a large set of possible system configurations . the first problem in graph clustering is to look for a quantitative definition of community . no definition is universally accepted . as a matter of fact , the definition often depends on the specific system at hand and/or application one has in mind . from intuition and the examples of section [ sec2 ] we get the notion that there must be more edges `` inside '' the community than edges linking vertices of the community with the rest of the graph . this is the reference guideline at the basis of most community definitions . but many alternative recipes are compatible with it . moreover , in most cases , communities are algorithmically defined , i. e. they are just the final product of the algorithm , without a precise _ a priori _ definition . let us start with a subgraph @xmath22 of a graph @xmath23 , with @xmath24 and @xmath25 vertices , respectively . we define the _ internal _ and _ external _ degree of vertex @xmath26 , @xmath27 and @xmath28 , as the number of edges connecting @xmath29 to other vertices of @xmath22 or to the rest of the graph , respectively . if @xmath30 , the vertex has neighbors only within @xmath22 , which is likely to be a good cluster for @xmath29 ; if @xmath31 , instead , the vertex is disjoint from @xmath22 and it should better be assigned to a different cluster . the _ internal degree _ @xmath32 of @xmath22 is the sum of the internal degrees of its vertices . likewise , the _ external degree _ @xmath33 of @xmath22 is the sum of the external degrees of its vertices . the _ total degree _ @xmath34 is the sum of the degrees of the vertices of @xmath22 . by definition , @xmath35 . we define the _ intra - cluster density _ @xmath36 of the subgraph @xmath22 as the ratio between the number of internal edges of @xmath22 and the number of all possible internal edges , i. e. @xmath37 similarly , the _ inter - cluster density _ @xmath38 is the ratio between the number of edges running from the vertices of @xmath22 to the rest of the graph and the maximum number of inter - cluster edges possible , i. e. @xmath39 for @xmath22 to be a community , we expect @xmath36 to be appreciably larger than the average link density @xmath40 of @xmath23 , which is given by the ratio between the number of edges of @xmath23 and the maximum number of possible edges @xmath41 . on the other hand , @xmath38 has to be much smaller than @xmath40 . searching for the best tradeoff between a large @xmath36 and a small @xmath38 is implicitly or explicitly the goal of most clustering algorithms . a simple way to do that is , e. g. , maximizing the sum of the differences @xmath42 over all clusters of the partition$ ] , but essentially equivalent . ] @xcite . a required property of a community is _ connectedness_. we expect that for @xmath22 to be a community there must be a path between each pair of its vertices , running only through vertices of @xmath22 . this feature simplifies the task of community detection on disconnected graphs , as in this case one just analyzes each connected component separately , unless special constraints are imposed on the resulting clusters . with these basic requirements in mind , we can now introduce the main definitions of community . social network analysts have devised many definitions of subgroups with various degrees of internal cohesion among vertices @xcite . many other definitions have been introduced by computer scientists and physicists . we distinguish three classes of definitions : local , global and based on vertex similarity . other definitions will be given in the context of the algorithms for which they were introduced . communities are parts of the graph with a few ties with the rest of the system . to some extent , they can be considered as separate entities with their own autonomy . so , it makes sense to evaluate them independently of the graph as a whole . local definitions focus on the subgraph under study , including possibly its immediate neighborhood , but neglecting the rest of the graph . we start with a listing of the main definitions adopted in social network analysis , for which we shall closely follow the exposition of ref . @xcite . there , four types of criteria were identified : _ complete mutuality _ , _ reachability _ , _ vertex degree _ and the _ comparison of internal versus external cohesion_. the corresponding communities are mostly _ maximal subgraphs _ , which can not be enlarged with the addition of new vertices and edges without losing the property which defines them . social communities can be defined in a very strict sense as subgroups whose members are all `` friends '' to each other @xcite ( complete mutuality ) . in graph terms , this corresponds to a _ clique _ , i. e. a subset whose vertices are all adjacent to each other . in social network analysis , a clique is a maximal subgraph , whereas in graph theory it is common to call cliques also non - maximal subgraphs . triangles are the simplest cliques , and are frequent in real networks . but larger cliques are less frequent . moreover , the condition is really too strict : a subgraph with all possible internal edges except one would be an extremely cohesive subgroup , but it would not be considered a community under this recipe . another problem is that all vertices of a clique are absolutely symmetric , with no differentiation between them . in many practical examples , instead , we expect that within a community there is a whole hierarchy of roles for the vertices , with core vertices coexisting with peripheral ones . we remark that vertices may belong to more cliques simultaneously , a property which is at the basis of the clique percolation method of palla et al . @xcite ( see section [ sec45_1 ] ) . from a practical point of view , finding cliques in a graph is an * np*-complete problem @xcite . the bron - kerbosch method @xcite runs in a time growing exponentially with the size of the graph . it is however possible to relax the notion of clique , defining subgroups which are still clique - like objects . a possibility is to use properties related to reachability , i. e. to the existence ( and length ) of paths between vertices . an _ @xmath15-clique _ is a maximal subgraph such that the distance of each pair of its vertices is not larger than @xmath15 @xcite . for @xmath43 one recovers the definition of clique , as all vertices are adjacent , so each geodesic path between any pair of vertices has length @xmath2 . this definition , more flexible than that of clique , still has some limitations , deriving from the fact that the geodesic paths need not run on the vertices of the subgraph at study , but may run on vertices outside the subgraph . in this way , there may be two disturbing consequences . first , the diameter of the subgraph may exceed @xmath15 , even if in principle each vertex of the subgraph is less than @xmath15 steps away from any of the others . second , the subgraph may be disconnected , which is not consistent with the notion of cohesion one tries to enforce . to avoid these problems , mokken @xcite has suggested two possible alternatives , the _ @xmath15-clan _ and the _ @xmath15-club_. an @xmath15-clan is an @xmath15-clique whose diameter is not larger than @xmath15 , i. e. a subgraph such that the distance between any two of its vertices , computed over shortest paths within the subgraph , does not exceed @xmath15 . an @xmath15-club , instead , is a maximal subgraph of diameter @xmath15 . the two definitions are quite close : the difference is that an @xmath15-clan is maximal under the constraint of being an @xmath15-clique , whereas an @xmath15-club is maximal under the constraint imposed by the length of the diameter . another criterion for subgraph cohesion relies on the adjacency of its vertices . the idea is that a vertex must be adjacent to some minumum number of other vertices in the subgraph . in the literature on social network analysis there are two complementary ways of expressing this . @xmath44-plex _ is a maximal subgraph in which each vertex is adjacent to all other vertices of the subgraph except at most @xmath44 of them @xcite . similarly , a _ @xmath44-core _ is a maximal subgraph in which each vertex is adjacent to at least @xmath44 other vertices of the subgraph @xcite . so , the two definitions impose conditions on the minimal number of absent or present edges . the corresponding clusters are more cohesive than @xmath15-cliques , just because of the existence of many internal edges . in any graph there is a whole hierarchy of cores of different order , which can be identified by means of a recent efficient algorithm @xcite . a @xmath44-core is essentially the same as a _ @xmath45-quasi complete _ subgraph , which is a subgraph such that the degree of each vertex is larger than @xmath46 , where @xmath45 is a real number in @xmath47 $ ] and @xmath44 the order of the subgraph @xcite . determining whether a graph includes a @xmath48-quasi complete subgraph of order at least @xmath44 is * np*-complete . as cohesive as a subgraph can be , it would hardly be a community if there is a strong cohesion as well between the subgraph and the rest of the graph . therefore , it is important to compare the internal and external cohesion of a subgraph . in fact , this is what is usually done in the most recent definitions of community . the first recipe , however , is not recent and stems from social network analysis . an _ @xmath49-set _ @xcite , or _ strong community _ @xcite , is a subgraph such that the internal degree of each vertex is greater than its external degree . this condition is quite strict and can be relaxed into the so - called _ weak _ definition of community @xcite , for which it suffices that the internal degree of the subgraph exceeds its external degree . an @xmath49-set is also a weak community , while the converse is not generally true . hu et al . @xcite have introduced alternative definitions of strong and weak communities : a community is strong if the internal degree of any vertex of the community exceeds the number of edges that the vertex shares with any other community ; a community is weak if its total internal degree exceeds the number of edges shared by the community with the other communities . these definitions are in the same spirit of the _ planted partition model _ ( section [ sec6 ] ) . an @xmath49-set is a strong community also in the sense of hu et al .. likewise , a weak community according to radicchi et al . is also a weak community for hu et al .. in both cases the converse is not true , however . another definition focuses on the robustness of clusters to edge removal and uses the concept of _ edge connectivity_. the edge connectivity of a pair of vertices in a graph is the minimal number of edges that need to be removed in order to disconnect the two vertices , i. e. such that there is no path between them . lambda set _ is a subgraph such that any pair of vertices of the subgraph has a larger edge connectivity than any pair formed by one vertex of the subgraph and one outside the subgraph @xcite . however , vertices of a lambda - set need not be adjacent and may be quite distant from each other . communities can also be identified by a _ fitness measure _ , expressing to which extent a subraph satisfies a given property related to its cohesion . the larger the fitness , the more definite is the community . this is the same principle behind _ quality functions _ , which give an estimate of the goodness of a graph partition ( see section [ sec3_2_2 ] ) . the simplest fitness measure for a cluster is its intra - cluster density @xmath36 . one could assume that a subgraph @xmath22 with @xmath44 vertices is a cluster if @xmath36 is larger than a threshold , say @xmath50 . finding such subgraphs is an * np*-complete problem , as it coincides with the * np*-complete clique problem when the threshold @xmath51 it is better to fix the size of the subgraph because , without this conditions , any clique would be one of the best possible communities , including trivial two - cliques ( simple edges ) . variants of this problem focus on the number of internal edges of the subgraph @xcite . another measure of interest is the _ relative density _ @xmath52 of a subgraph @xmath22 , defined as the ratio between the internal and the total degree of @xmath22 . finding subgraphs of a given size with @xmath52 larger than a threshold is * np*-complete @xcite . fitness measures can also be associated to the connectivity of the subgraph at study to the other vertices of the graph . a good community is expected to have a small cut size ( see section [ sec1_1 ] ) , i. e. a small number of edges joining it to the rest of the graph . this sets a bridge between community detection and graph partitioning , which we shall discuss in section [ sec4_1 ] . communities can also be defined with respect to the graph as a whole . this is reasonable in those cases in which clusters are essential parts of the graph , which can not be taken apart without seriously affecting the functioning of the system . the literature offers many global criteria to identify communities . in most cases they are indirect definitions , in which some global property of the graph is used in an algorithm that delivers communities at the end . however , there is a class of proper definitions , based on the idea that a graph has community structure if it is different from a random graph . a random graph la erds - rnyi ( section [ sec1_3 ] ) , for instance , is not expected to have community structure , as any two vertices have the same probability to be adjacent , so there should be no preferential linking involving special groups of vertices . therefore , one can define a _ null model _ , i. e. a graph which matches the original in some of its structural features , but which is otherwise a random graph . the null model is used as a term of comparison , to verify whether the graph at study displays community structure or not . the most popular null model is that proposed by newman and girvan and consists of a randomized version of the original graph , where edges are rewired at random , under the constraint that the expected degree of each vertex matches the degree of the vertex in the original graph @xcite . this null model is the basic concept behind the definition of _ modularity _ , a function which evaluates the goodness of partitions of a graph into clusters . modularity will be discussed at length in this review , because it has the unique privilege of being at the same time a global criterion to define a community , a quality function and the key ingredient of the most popular method of graph clustering . in the standard formulation of modularity , a subgraph is a community if the number of edges inside the subgraph exceeds the expected number of internal edges that the same subgraph would have in the null model . this expected number is an average over all possible realizations of the null model . several modifications of modularity have been proposed ( section [ sub_sec6_01 ] ) . a general class of null models , including modularity as a special case , has been designed by reichardt and bornholdt @xcite ( section [ sub_sec6_01 ] ) . it is natural to assume that communities are groups of vertices similar to each other . one can compute the similarity between each pair of vertices with respect to some reference property , local or global , no matter whether they are connected by an edge or not . each vertex ends up in the cluster whose vertices are most similar to it . similarity measures are at the basis of traditional methods , like hierarchical , partitional and spectral clustering , to be discussed in sections [ sec4_2 ] , [ sec4_3 ] and [ sec4_4 ] . here we discuss some popular measures used in the literature . if it is possible to embed the graph vertices in an @xmath15-dimensional euclidean space , by assigning a position to them , one could use the _ distance _ between a pair of vertices as a measure of their similarity ( it is actually a measure of dissimilarity because similar vertices are expected to be close to each other ) . given the two data points @xmath53 and @xmath54 , one could use any norm @xmath55 , like the _ euclidean distance _ ( @xmath56-norm ) , @xmath57 the _ manhattan distance _ ( @xmath58-norm ) @xmath59 and the @xmath60-norm @xmath61}|a_k - b_k| . \label{eq5}\ ] ] another popular spatial measure is the _ cosine similarity _ , defined as @xmath62 where @xmath63 is the dot product of the vectors * a * and * b*. the variable @xmath64 is defined in the range @xmath65 . if the graph can not be embedded in space , the similarity must be necessarily inferred from the adjacency relationships between vertices . a possibility is to define a distance @xcite between vertices like @xmath66 where * a * is the adjacency matrix . this is a dissimilarity measure , based on the concept of structural equivalence @xcite . two vertices are structurally equivalent if they have the same neighbors , even if they are not adjacent themselves . if @xmath67 and @xmath68 are structurally equivalent , @xmath69 . vertices with large degree and different neighbours are considered very `` far '' from each other . alternatively , one could measure the _ overlap _ between the neighborhoods @xmath70 and @xmath71 of vertices @xmath67 and @xmath68 , given by the ratio between the intersection and the union of the neighborhoods , i. e. @xmath72 another measure related to structural equivalence is the pearson correlation between columns or rows of the adjacency matrix , @xmath73 where the averages @xmath74 and the variances @xmath75 . an alternative measure is the number of edge- ( or vertex- ) independent paths between two vertices . independent paths do not share any edge ( vertex ) , and their number is related to the maximum flow that can be conveyed between the two vertices under the constraint that each edge can carry only one unit of flow ( max - flow / min - cut theorem @xcite ) . the maximum flow can be calculated in a time @xmath21 , for a graph with @xmath14 edges , using techniques like the augmenting path algorithm @xcite . similarly , one could consider all paths running between two vertices . in this case , there is the problem that the total number of paths is infinite , but this can be avoided if one performs a weighted sum of the number of paths . for instance , paths of length @xmath76 can be weighted by the factor @xmath77 , with @xmath78 . another possibility , suggested by estrada and hatano @xcite , is to weigh paths of length @xmath76 with the inverse factorial @xmath79 . in both cases , the contribution of long paths is strongly suppressed and the sum converges . another important class of measures of vertex similarity is based on properties of random walks on graphs . one of this properties is the _ commute - time _ between a pair of vertices , which is the average number of steps needed for a random walker , starting at either vertex , to reach the other vertex for the first time and to come back to the starting vertex . saerens and coworkers @xcite have extensively studied and used the commute - time ( and variants thereof ) as ( dis)similarity measure : the larger the time , the farther ( less similar ) the vertices . the commute - time is closely related @xcite to the _ resistance distance _ introduced by klein and randic @xcite , expressing the effective electrical resistance between two vertices if the graph is turned into a resistor network . white and smyth @xcite and zhou @xcite used instead the average first passage time , i. e. the average number of steps needed to reach for the first time the target vertex from the source . harel and koren @xcite proposed to build measures out of quantities like the probability to visit a target vertex in no more than a given number of steps after it leaves a source vertex ) and in a recent analysis by nadler et al . @xcite , one defined a dissimilarity measure called `` diffusion distance '' , which is derived from the probability that the walker visits the target after a fixed number of steps . ] and the probability that a random walker starting at a source visits the target exactly once before hitting the source again . another quantity used to define similarity measures is the _ escape probability _ , defined as the probability that the walker reaches the target vertex before coming back to the source vertex @xcite . the escape probability is related to the effective conductance between the two vertices in the equivalent resistor network . other authors have exploited properties of modified random walks . for instance , the algorithm by gori and pucci @xcite and that by tong et al . @xcite used similarity measures derived from google s pagerank process @xcite . a _ partition _ is a division of a graph in clusters , such that each vertex belongs to one cluster . as we have seen in section [ sec2 ] , in real systems vertices may be shared among different communities . a division of a graph into overlapping ( or _ fuzzy _ ) communities is called _ cover_. the number of possible partitions in @xmath44 clusters of a graph with @xmath15 vertices is the _ stirling number of the second kind _ @xmath80 @xcite . the total number of possible partitions is the _ n - th bell number _ @xmath81 @xcite . in the limit of large @xmath15 , @xmath82 has the asymptotic form @xcite @xmath83^{n+1/2}e^{\lambda(n)-n-1 } , \label{eqpart1}\ ] ] where @xmath84 , @xmath85 being the _ lambert @xmath86 function _ @xcite . therefore , @xmath82 grows faster than exponentially with the graph size @xmath15 , which means that an enumeration and/or evaluation of all partitions of a graph is impossible , unless the graph consists of very few vertices . partitions can be _ hierarchically ordered _ , when the graph has different levels of organization / structure at different scales . in this case , clusters display in turn community structure , with smaller communities inside , which may again contain smaller communities , and so on ( fig . [ fig8 ] ) . as an example , in a social network of children living in the same town , one could group the children according to the schools they attend , but within each school one can make a subdivision into classes . hierarchical organization is a common feature of many real networks , where it is revealed by a peculiar scaling of the clustering coefficient for vertices having the same degree @xmath44 , when plotted as a function of @xmath44 @xcite . a natural way to represent the hierarchical structure of a graph is to draw a _ dendrogram _ , like the one illustrated in fig . [ fig9 ] . here , partitions of a graph with twelve vertices are shown . at the bottom , each vertex is its own module ( the `` leaves '' of the tree ) . by moving upwards , groups of vertices are successively aggregated . mergers of communities are represented by horizontal lines . the uppermost level represents the whole graph as a single community . cutting the diagram horizontally at some height , as shown in the figure ( dashed line ) , displays one partition of the graph . the diagram is hierarchical by construction : each community belonging to a level is fully included in a community at a higher level . dendrograms are regularly used in sociology and biology . the technique of hierarchical clustering , described in section [ sec4_2 ] , lends itself naturally to this kind of representation . a dendrogram , or hierarchical tree . horizontal cuts correspond to partitions of the graph in communities . reprinted figure with permission from ref . 2004 by the american physical society . ] reliable algorithms are supposed to identify _ good _ partitions . but what is a good clustering ? in order to distinguish between `` good '' and `` bad '' clusterings , it would be useful to require that partitions satisfy a set of basic properties , intuitive and easy to agree upon . in the wider context of data clustering , this issue has been studied by jon kleinberg @xcite , who has proved an important _ impossibility theorem_. given a set @xmath87 of points , a _ distance function _ @xmath88 is defined , which is positive definite and symmetric ( the triangular inequality is not explicitly required ) . one wishes to find a clustering @xmath89 based on the distances between the points . kleinberg showed that no clustering satisfies at the same time the three following properties : 1 . _ scale - invariance _ : given a constant @xmath18 , multiplying _ any _ distance function @xmath88 by @xmath18 yields the same clustering . richness _ : any possible partition of the given point set can be recovered if one chooses a suitable distance function @xmath88 . consistency _ : given a partition , any modification of the distance function that does not decrease the distance between points of different clusters and that does not increase the distance between points of the same cluster , yields the same clustering . the theorem can not be extended to graph clustering because the distance function can not be in general defined for a graph which is not complete . for weighted complete graphs , like correlation matrices @xcite , it is often possible to define a distance function . on a generic graph , except for the first property , which does not make sense without a distance function , the other two are quite well defined . the property of richness implies that , given a partition , one can set edges between the vertices in such a way that the partition is a natural outcome of the resulting graph ( e.g. , it could be achieved by setting edges only between vertices of the same cluster ) . consistency here implies that deleting inter - cluster edges and adding intra - cluster edges yields the same partition . many algorithms are able to identify a subset of meaningful partitions , ideally one or just a few , whereas some others , like techniques based on hierarchical clustering ( section [ sec4_2 ] ) , deliver a large number of partitions . that does not mean that the partitions found are equally good . therefore it is helpful ( sometimes even necessary ) to have a quantitative _ criterion _ to assess the goodness of a graph partition . quality function _ is a function that assigns a number to each partition of a graph . in this way one can rank partitions based on their score given by the quality function . partitions with high scores are `` good '' , so the one with the largest score is by definition the best . nevertheless , one should keep in mind that the question of when a partition is better than another one is ill - posed , and the answer depends on the specific concept of community and/or quality function adopted . a quality function @xmath90 is _ additive _ if there is an elementary function @xmath91 such that , for any partition @xmath92 of a graph @xmath93 where @xmath22 is a generic cluster of partition @xmath94 . . [ eq10 ] states that the quality of a partition is given by the sum of the qualities of the individual clusters . the function @xmath95 could be any of the cluster fitness functions discussed in section [ sec3_1_2 ] , for instance . most quality functions used in the literature are additive , although it is not a necessary requirement . an example of quality function is the _ performance _ @xmath96 , which counts the number of correctly `` interpreted '' pairs of vertices , i. e. two vertices belonging to the same community and connected by an edge , or two vertices belonging to different communities and not connected by an edge . the definition of performance , for a partition @xmath94 , is @xmath97 by definition , @xmath98 . another example is _ coverage _ , i. e. the ratio of the number of intra - community edges by the total number of edges : by definition , an ideal cluster structure , where the clusters are disconnected from each other , yields a coverage of @xmath2 , as all edges of the graph fall within clusters . the most popular quality function is the modularity of newman and girvan @xcite . it is based on the idea that a random graph is not expected to have a cluster structure , so the possible existence of clusters is revealed by the comparison between the actual density of edges in a subgraph and the density one would expect to have in the subgraph if the vertices of the graph were attached regardless of community structure . this expected edge density depends on the chosen _ null model _ , i. e. a copy of the original graph keeping some of its structural properties but without community structure . modularity can then be written as follows @xmath99 where the sum runs over all pairs of vertices , @xmath9 is the adjacency matrix , @xmath14 the total number of edges of the graph , and @xmath100 represents the expected number of edges between vertices @xmath67 and @xmath68 in the null model . the @xmath101-function yields one if vertices @xmath67 and @xmath68 are in the same community ( @xmath102 ) , zero otherwise . the choice of the null model graph is in principle arbitrary , and several possibilities exist . for instance , one could simply demand that the graph keeps the same number of edges as the original graph , and that edges are placed with the same probability between any pair of vertices . in this case ( bernoulli random graph ) , the null model term in eq . [ eq : mod_0 ] would be a constant ( i. e. @xmath103 $ ] , @xmath104 ) . however this null model is not a good descriptor of real networks , as it has a poissonian degree distribution which is very different from the skewed distributions found in real networks . due to the important implications that broad degree distributions have for the structure and function of real networks @xcite , it is preferable to go for a null model with the same degree distribution of the original graph . the standard null model of modularity imposes that the expected degree sequence ( after averaging over all possible configurations of the model ) matches the actual degree sequence of the graph . this is a stricter constraint than merely requiring the match of the degree distributions , and is essentially equivalent to the _ configuration model _ , which has been subject of intense investigations in the recent literature on networks @xcite . in this null model , a vertex could be attached to any other vertex of the graph and the probability that vertices @xmath67 and @xmath68 , with degrees @xmath105 and @xmath106 , are connected , can be calculated without problems . in fact , in order to form an edge between @xmath67 and @xmath68 one needs to join two _ stubs _ ( i. e. half - edges ) , incident with @xmath67 and @xmath68 . the probability @xmath107 to pick at random a stub incident with @xmath67 is @xmath108 , as there are @xmath105 stubs incident with @xmath67 out of a total of @xmath109 . the probability of a connection between @xmath67 and @xmath68 is then given by the product @xmath110 , since edges are placed independently of each other . the result is @xmath111 , which yields an expected number @xmath112 of edges between @xmath67 and @xmath68 . so , the final expression of modularity reads @xmath113 since the only contributions to the sum come from vertex pairs belonging to the same cluster , we can group these contributions together and rewrite the sum over the vertex pairs as a sum over the clusters @xmath114 . \label{eq : mod1}\ ] ] here , @xmath115 is the number of clusters , @xmath116 the total number of edges joining vertices of module @xmath117 and @xmath118 the sum of the degrees of the vertices of @xmath117 . in eq . [ eq : mod1 ] , the first term of each summand is the fraction of edges of the graph inside the module , whereas the second term represents the expected fraction of edges that would be there if the graph were a random graph with the same expected degree for each vertex . a nice feature of modularity is that it can be equivalently expressed both in terms of the intra - cluster edges , as in eq . [ eq : mod1 ] , and in terms of the inter - cluster edges @xcite . in fact , the maximum of modularity can be expressed as @xmath119\right\}\\ & = & \nonumber \frac{1}{m}\mbox{max}_{\cal p}\left\{\sum_{c=1}^{n_c}\big[l_c-\mbox{ex}(l_c)\big]\right\}\\ & = & -\frac{1}{m}\mbox{min}_{\cal p}\left\{-\sum_{c=1}^{n_c}\big[l_c-\mbox{ex}(l_c)\big]\right\ } , \label{eqr11}\end{aligned}\ ] ] where @xmath120 and @xmath121 indicates the maximum and the minimum over all possible graph partitions @xmath94 and @xmath122 indicates the expected number of links in cluster @xmath117 in the null model of modularity . by adding and subtracting the total number of edges @xmath14 of the graph one finally gets @xmath123\\ & = & -\frac{1}{m}\mbox{min}_{\cal p}(|\mbox{cut}_{\cal p}|-\mbox{excut}_{\cal p } ) . \label{eqr12}\end{aligned}\ ] ] in the last expression @xmath124 is the number of inter - cluster edges of partition @xmath94 , and @xmath125 is the expected number of inter - cluster edges of the partition in modularity s null model . according to eq . [ eq : mod1 ] , a subgraph is a module if the corresponding contribution to modularity in the sum is positive . the more the number of internal edges of the cluster exceeds the expected number , the better defined the community . so , large positive values of the modularity indicate good partitions . ] . the maximum modularity of a graph generally grows if the size of the graph and/or the number of ( well - separated ) clusters increase @xcite . therefore , modularity should not be used to compare the quality of the community structure of graphs which are very different in size . the modularity of the whole graph , taken as a single community , is zero , as the two terms of the only summand in this case are equal and opposite . modularity is always smaller than one , and can be negative as well . for instance , the partition in which each vertex is a community is always negative : in this case the sum runs over @xmath15 terms , which are all negative as the first term of each summand is zero . this is a nice feature of the measure , implying that , if there are no partitions with positive modularity , the graph has no community structure . on the contrary , the existence of partitions with large negative modularity values may hint to the existence of subgroups with very few internal edges and many edges lying between them ( _ multipartite structure _ ) @xcite . holmstrm et al . @xcite have shown that the distribution of modularity values across the partitions of various graphs , real and artificial ( including random graphs with no apparent community structure ) , has some stable features , and that the most likely modularity values correspond to partitions in clusters of approximately equal size . modularity has been employed as quality function in many algorithms , like some of the divisive algorithms of section [ sec5 ] . in addition , modularity optimization is itself a popular method for community detection ( see section [ sub_sec6_0 ] ) . modularity also allows to assess the stability of partitions @xcite ( section [ sec6_4 ] ) , it can be used to design layouts for graph visualization @xcite and to perform a sort of renormalization of a graph , by transforming a graph into a smaller one with the same community structure @xcite . the problem of graph partitioning consists in dividing the vertices in @xmath126 groups of predefined size , such that the number of edges lying between the groups is minimal . the number of edges running between clusters is called _ cut size_. fig . [ figure5 ] presents the solution of the problem for a graph with fourteen vertices , for @xmath127 and clusters of equal size . graph partitioning . the dashed line shows the solution of the minimum bisection problem for the graph illustrated , i. e. the partition in two groups of equal size with minimal number of edges running between the groups . reprinted figure with permission from ref . 2009 by springer.,width=264 ] specifying the number of clusters of the partition is necessary . if one simply imposed a partition with the minimal cut size , and left the number of clusters free , the solution would be trivial , corresponding to all vertices ending up in the same cluster , as this would yield a vanishing cut size . specifying the size is also necessary , as otherwise the most likely solution of the problem would consist in separating the lowest degree vertex from the rest of the graph , which is quite uninteresting . this problem can be actually avoided by choosing a different measure to optimize for the partitioning , which accounts for the size of the clusters . some of these measures will be briefly introduced at the end of this section . graph partitioning is a fundamental issue in parallel computing , circuit partitioning and layout , and in the design of many serial algorithms , including techniques to solve partial differential equations and sparse linear systems of equations . most variants of the graph partitioning problem are np - hard . there are however several algorithms that can do a good job , even if their solutions are not necessarily optimal @xcite . many algorithms perform a bisection of the graph . partitions into more than two clusters are usually attained by iterative bisectioning . moreover , in most cases one imposes the constraint that the clusters have equal size . this problem is called _ minimum bisection _ and is np - hard . the _ kernighan - lin algorithm _ @xcite is one of the earliest methods proposed and is still frequently used , often in combination with other techniques . the authors were motivated by the problem of partitioning electronic circuits onto boards : the nodes contained in different boards need to be linked to each other with the least number of connections . the procedure is an optimization of a benefit function @xmath90 , which represents the difference between the number of edges inside the modules and the number of edges lying between them . the starting point is an initial partition of the graph in two clusters of the predefined size : such initial partition can be random or suggested by some information on the graph structure . then , subsets consisting of equal numbers of vertices are swapped between the two groups , so that @xmath90 has the maximal increase . the subsets can consist of single vertices . to reduce the risk to be trapped in local maxima of @xmath90 , the procedure includes some swaps that decrease the function @xmath90 . after a series of swaps with positive and negative gains , the partition with the largest value of @xmath90 is selected and used as starting point of a new series of iterations . the kernighan - lin algorithm is quite fast , scaling as @xmath128 ( @xmath15 being as usual the number of vertices ) , if only a constant number of swaps are performed at each iteration . the most expensive part is the identification of the subsets to swap , which requires the computation of the gains / losses for any pair of candidate subsets . on sparse graphs , a slightly different heuristic allows to lower the complexity to @xmath129 . the partitions found by the procedure are strongly dependent on the initial configuration and other algorithms can do better . it is preferable to start with a good guess about the sought partition , otherwise the results are quite poor . therefore the method is typically used to improve on the partitions found through other techniques , by using them as starting configurations for the algorithm . the kernighan - lin algorithm has been extended to extract partitions in any number of parts @xcite , however the run - time and storage costs increase rapidly with the number of clusters . another popular technique is the _ spectral bisection method _ @xcite , which is based on the properties of the spectrum of the laplacian matrix . spectral clustering will be discussed more thoroughly in section [ sec4_4 ] , here we focus on its application to graph partitioning . every partition of a graph with @xmath15 vertices in two groups can be represented by an index vector @xmath130 , whose component @xmath131 is @xmath132 if vertex @xmath67 is in one group and @xmath133 if it is in the other group . the cut size @xmath134 of the partition of the graph in the two groups can be written as @xmath135 where @xmath136 is the laplacian matrix and @xmath137 the transpose of vector @xmath130 . vector @xmath130 can be written as @xmath138 , where @xmath139 , @xmath140 are the eigenvectors of the laplacian . if @xmath130 is properly normalized , then @xmath141 where @xmath142 is the laplacian eigenvalue corresponding to eigenvector @xmath139 . it is worth remarking that the sum contains at most @xmath143 terms , as the laplacian has at least one zero eigenvalue . minimizing @xmath134 equals to the minimization of the sum on the right - hand side of eq . this task is still very hard . however , if the second lowest eigenvector @xmath144 is close enough to zero , a good approximation of the minimum can be attained by choosing @xmath130 parallel to the corresponding eigenvector @xmath145 , which is called _ fiedler vector _ @xcite : this would reduce the sum to @xmath144 , which is a small number . but the index vector can not be perfectly parallel to @xmath145 by construction , because all its components are equal in modulus , whereas the components of @xmath145 are not . the best choice is to match the signs of the components . so , one can set @xmath146 ( @xmath133 ) if @xmath147 ( @xmath148 ) . it may happen that the sizes of the two corresponding groups do not match the predefined sizes one wishes to have . in this case , if one aims at a split in @xmath149 and @xmath150 vertices , the best strategy is to order the components of the fiedler vector from the lowest to the largest values and to put in one group the vertices corresponding to the first @xmath149 components from the top or the bottom , and the remaining vertices in the second group . this procedure yields two partitions : the better solution is naturally the one that gives the smaller cut size . the spectral bisection method is quite fast . the first eigenvectors of the laplacian can be computed by using the lanczos method @xcite . the time required to compute the first @xmath44 eigenvectors of a matrix with the lanczos method depends on the size of the eigengap @xmath151 @xcite . if the eigenvalues @xmath152 and @xmath153 are well separated , the running time of the algorithm is much shorter than the time required to calculate the complete set of eigenvectors , which scales as @xmath154 . the method gives in general good partitions , that can be further improved by applying the kernighan - lin algorithm . the well known max - flow min - cut theorem by ford and fulkerson @xcite states that the minimum cut between any two vertices @xmath155 and @xmath156 of a graph , i. e. any minimal subset of edges whose deletion would topologically separate @xmath155 from @xmath156 , carries the maximum flow that can be transported from @xmath155 to @xmath156 across the graph . in this context edges play the role of water pipes , with a given carrying capacity ( e.g. their weights ) , and vertices the role of pipe junctions . this theorem has been used to determine minimal cuts from maximal flows in clustering algorithms . there are several efficient routines to compute maximum flows in graphs , like the algorithm of goldberg and tarjan @xcite . flake et al . @xcite have recently used maximum flows to identify communities in the graph of the world wide web . the web graph is directed but for the purposes of the calculation flake et at . treated the edges as undirected . web communities are defined to be `` strong '' ( ls - sets ) , i. e. the internal degree of each vertex must not be smaller than its external degree @xcite . an artificial sink @xmath156 is added to the graph and one calculates the maximum flows from a source vertex @xmath155 to the sink @xmath156 : the corresponding minimum cut identifies the community of vertex @xmath155 , provided @xmath155 shares a sufficiently large number of edges with the other vertices of its community , otherwise one could get trivial separations and meaningless clusters . other popular methods for graph partitioning include level - structure partitioning , the geometric algorithm , multilevel algorithms , etc . a good description of these algorithms can be found in ref . @xcite . graphs can be also partitioned by minimizing measures that are affine to the cut size , like _ conductance _ the conductance @xmath157 of the subgraph @xmath22 of a graph @xmath23 is defined as @xmath158 where @xmath159 is the cut size of @xmath22 , and @xmath160 , @xmath161 are the total degrees of @xmath22 and of the rest of the graph @xmath162 , respectively . cuts are defined only between non - empty sets , otherwise the measure would not be defined ( as the denominator in eq . [ eq3000 ] would vanish ) . the minimum of the conductance is obtained in correspondence of low values of the cut size and of large values for the denominator in eq . [ eq3000 ] , which peaks when the total degrees of the cluster and its complement are equal . in practical applications , especially on large graphs , close values of the total degrees correspond to clusters of approximately equal size . the problem of finding a cut with minimal conductance is * np*-hard @xcite . similar measures are the _ ratio cut _ @xcite and the _ normalized cut _ @xcite . the ratio cut of a cluster @xmath22 is defined as @xmath163 where @xmath164 and @xmath165 are the number of vertices of the two subgraphs . the normalized cut of a cluster @xmath22 is @xmath166 where @xmath160 is again the total degree of @xmath22 . as for the conductance , minimizing the ratio cut and the normalized cut favors partitions into clusters of approximately equal size , measured in terms of the number of vertices or edges , respectively . on the other hand , graph partitioning requires preliminary assumptions on the cluster sizes , whereas the minimization of conductance , ratio cut and normalized cut does not . the ratio cut was introduced for circuit partitioning @xcite and its optimization is an * np*-hard problem @xcite . the normalized cut is frequently used in image segmentation @xcite and its optimization is * np*-complete @xcite . the cut ratio and the normalized cut can be quite well minimized via spectral clustering @xcite ( section [ sec4_4 ] ) . algorithms for graph partitioning are not good for community detection , because it is necessary to provide as input the number of groups and in some cases even their sizes , about which in principle one knows nothing . instead , one would like an algorithm capable to produce this information in its output . besides , from the methodological point of view , using iterative bisectioning to split the graph in more pieces is not a reliable procedure . for instance , a split into three clusters is necessarily obtained by breaking either cluster of the original bipartition in two parts , whereas in many cases a minimum cut partition is obtained if the third cluster is a merger of parts of both initial clusters . in general , very little is known about the community structure of a graph . it is uncommon to know the number of clusters in which the graph is split , or other indications about the membership of the vertices . in such cases clustering procedures like graph partitioning methods can hardly be of help , and one is forced to make some reasonable assumptions about the number and size of the clusters , which are often unjustified . on the other hand , the graph may have a hierarchical structure , i. e. may display several levels of grouping of the vertices , with small clusters included within large clusters , which are in turn included in larger clusters , and so on . social networks , for instance , often have a hierarchical structure ( section [ sec3_2_1 ] ) . in such cases , one may use _ hierarchical clustering algorithms _ @xcite , i. e. clustering techniques that reveal the multilevel structure of the graph . hierarchical clustering is very common in social network analysis , biology , engineering , marketing , etc . the starting point of any hierarchical clustering method is the definition of a similarity measure between vertices . after a measure is chosen , one computes the similarity for each pair of vertices , no matter if they are connected or not . at the end of this process , one is left with a new @xmath167 matrix @xmath168 , the similarity matrix . in section [ sec3_1_4 ] we have listed several possible definitions of similarity . hierarchical clustering techniques aim at identifying groups of vertices with high similarity , and can be classified in two categories : 1 . _ agglomerative algorithms _ , in which clusters are iteratively merged if their similarity is sufficiently high ; 2 . _ divisive algorithms _ , in which clusters are iteratively split by removing edges connecting vertices with low similarity . the two classes refer to opposite processes : agglomerative algorithms are bottom - up , as one starts from the vertices as separate clusters ( singletons ) and ends up with the graph as a unique cluster ; divisive algorithms are top - down as they follow the opposite direction . divisive techniques have been rarely used in the past ( meanwhile they have become more popular , see section [ sec5 ] ) , so we shall concentrate here on agglomerative algorithms . since clusters are merged based on their mutual similarity , it is essential to define a measure that estimates how similar clusters are , out of the matrix @xmath168 . this involves some arbitrariness and several prescriptions exist . in _ single linkage clustering _ , the similarity between two groups is the minimum element @xmath169 , with @xmath67 in one group and @xmath68 in the other . on the contrary , the maximum element @xmath169 for vertices of different groups is used in the procedure of _ complete linkage clustering_. in _ average linkage clustering _ one has to compute the average of the @xmath169 . the procedure can be better illustrated by means of dendrograms ( section [ sec3_2_1 ] ) , like the one in fig . [ fig9 ] . sometimes , stopping conditions are imposed to select a partition or a group of partitions that satisfy a special criterion , like a given number of clusters or the optimization of a quality function ( e.g. modularity ) . hierarchical clustering has the advantage that it does not require a preliminary knowledge on the number and size of the clusters . however , it does not provide a way to discriminate between the many partitions obtained by the procedure , and to choose that or those that better represent the community structure of the graph . the results of the method depend on the specific similarity measure adopted . the procedure also yields a hierarchical structure by construction , which is rather artificial in most cases , since the graph at hand may not have a hierarchical structure at all . moreover , vertices of a community may not be correctly classified , and in many cases some vertices are missed even if they have a central role in their clusters @xcite . another problem is that vertices with just one neighbor are often classified as separated clusters , which in most cases does not make sense . finally , a major weakness of agglomerative hierarchical clustering is that it does not scale well . if points are embedded in space , so that one can use the distance as dissimilarity measure , the computational complexity is @xmath129 for single linkage , @xmath128 for the complete and average linkage schemes . for graph clustering , where a distance is not trivially defined , the complexity can become much heavier if the calculation of the chosen similarity measure is costly . _ partitional clustering _ indicates another popular class of methods to find clusters in a set of data points . here , the number of clusters is preassigned , say @xmath44 . the points are embedded in a metric space , so that each vertex is a point and a distance measure is defined between pairs of points in the space . the distance is a measure of dissimilarity between vertices . the goal is to separate the points in @xmath44 clusters such to maximize / minimize a given cost function based on distances between points and/or from points to _ centroids _ , i. e. suitably defined positions in space . some of the most used functions are listed below : * _ minimum k - clustering_. the cost function here is the _ diameter _ of a cluster , which is the largest distance between two points of a cluster . the points are classified such that the largest of the @xmath44 cluster diameters is the smallest possible . the idea is to keep the clusters very `` compact '' . * _ k - clustering sum_. same as minimum @xmath44-clustering , but the diameter is replaced by the average distance between all pairs of points of a cluster . * _ k - center_. for each cluster @xmath67 one defines a reference point @xmath170 , the centroid , and computes the maximum @xmath171 of the distances of each cluster point from the centroid . the clusters and centroids are self - consistently chosen such to minimize the largest value of @xmath171 . * _ k - median_. same as k - center , but the maximum distance from the centroid is replaced by the average distance . the most popular partitional technique in the literature is _ @xmath44-means clustering _ @xcite . here the cost function is the total intra - cluster distance , or squared error function @xmath172 where @xmath173 indicates the subset of points of the @xmath67-th cluster and @xmath174 its centroid . the @xmath44-means problem can be simply solved with the lloyd s algorithm @xcite . one starts from an initial distribution of centroids such that they are as far as possible from each other . in the first iteration , each vertex is assigned to the nearest centroid . next , the centers of mass of the @xmath44 clusters are estimated and become a new set of centroids , which allows for a new classification of the vertices , and so on . after a small number of iterations , the positions of the centroids are stable , and the clusters do not change any more . the solution found is not optimal , and it strongly depends on the initial choice of the centroids . nevertheless , lloyd s heuristic has remained popular due to its quick convergence , which makes it suitable for the analysis of large data sets . the result can be improved by performing more runs starting from different initial conditions , and picking the solution which yields the minimum value of the total intra - cluster distance . extensions of @xmath44-means clustering to graphs have been proposed by some authors @xcite . another popular technique , similar in spirit to @xmath44-means clustering , is _ fuzzy k - means clustering _ @xcite . this method accounts for the fact that a point may belong to two or more clusters at the same time and is widely used in pattern recognition . the associated cost function is @xmath175 where @xmath176 is the _ membership matrix _ , which measures the degree of membership of point @xmath67 ( with position @xmath177 ) in cluster @xmath68 , @xmath14 is a real number greater than @xmath2 and @xmath178 is the center of cluster @xmath68 @xmath179 the matrix @xmath176 is normalized so that the sum of the memberships of every point in all clusters yields @xmath2 . the membership @xmath176 is related to the distance of point @xmath67 from the center of cluster @xmath68 , as it is reasonable to assume that the larger this distance , the lower @xmath176 . this can be expressed by the following relation @xmath180 the cost function @xmath181 can be minimized by iterating eqs . [ eq3004 ] and [ eq3005 ] . one starts from some initial guess for @xmath176 and uses eq . [ eq3004 ] to compute the centers , which are then plugged back into eqs . [ eq3005 ] , and so on . the process stops when the corresponding elements of the membership matrix in consecutive iterations differ from each other by less than a predefined tolerance . it can be shown that this procedure indeed delivers a local minimum of the cost function @xmath181 of eq . [ eq3003 ] . this procedure has the same problems of lloyd s algorithm for @xmath44-means clustering , i. e. the minimum is a local minimum , and depends on the initial choice of the matrix @xmath176 . the limitation of partitional clustering is the same as that of the graph partitioning algorithms : the number of clusters must be specified at the beginning , the method is not able to derive it . in addition , the embedding in a metric space can be natural for some graphs , but rather artificial for others . let us suppose to have a set of @xmath15 objects @xmath182 with a pairwise similarity function @xmath87 defined between them , which is symmetric and non - negative ( i. e. , @xmath183 , @xmath184 ) . _ spectral clustering _ includes all methods and techniques that partition the set into clusters by using the eigenvectors of matrices , like @xmath87 itself or other matrices derived from it . in particular , the objects could be points in some metric space , or the vertices of a graph . spectral clustering consists of a transformation of the initial set of objects into a set of points in space , whose coordinates are elements of eigenvectors : the set of points is then clustered via standard techniques , like @xmath44-means clustering ( section [ sec4_3 ] ) . one may wonder why it is necessary to cluster the points obtained through the eigenvectors , when one can directly cluster the initial set of objects , based on the similarity matrix . the reason is that the change of representation induced by the eigenvectors makes the cluster properties of the initial data set much more evident . in this way , spectral clustering is able to separate data points that could not be resolved by applying directly @xmath44-means clustering , for instance , as the latter tends to deliver convex sets of points . the first contribution on spectral clustering was a paper by donath and hoffmann @xcite , who used the eigenvectors of the adjacency matrix for graph partitions . in the same year , fiedler @xcite realized that from the eigenvector of the second smallest eigenvalue of the laplacian matrix it was possible to obtain a bipartition of the graph with very low cut size , as we have explained in section [ sec4_1 ] . for a historical survey see ref . @xcite . in this section we shall follow the nice tutorial by von luxburg @xcite , with a focus on spectral graph clustering . the concepts and methods discussed below apply to both unweighted and weighted graphs . the laplacian is by far the most used matrix in spectral clustering . in section [ sec1_2 ] we see that the unnormalized laplacian of a graph with @xmath44 connected components has @xmath44 zero eigenvalues . in this case the laplacian can be written in block - diagonal form , i. e. the vertices can be ordered in such a way that the laplacian displays @xmath44 square blocks along the diagonal , with ( some ) entries different from zero , whereas all other elements vanish . each block is the laplacian of the corresponding subgraph , so it has the trivial eigenvector with components @xmath185 . therefore , there are @xmath44 degenerate eigenvectors with equal non - vanishing components in correspondence of the vertices of a block , whereas all other components are zero . in this way , from the components of the eigenvectors one can identify the connected components of the graph . for instance , let us consider the @xmath186 matrix , whose columns are the @xmath44 eigenvectors above mentioned . the @xmath67-th row of this matrix is a vector with @xmath44 components representing vertex @xmath67 of the graph . vectors representing vertices in the same connected component of the graph coincide , and their tip lies on one of the axes of a @xmath44-dimensional system of coordinates ( i. e. they are all vectors of the form @xmath187 ) . so , by drawing the vertex vectors one would see @xmath44 distinct points , each on a different axis , corresponding to the graph components . if the graph is connected , but consists of @xmath44 subgraphs which are weakly linked to each other , the spectrum of the unnormalized laplacian will have one zero eigenvalue , all others being positive . now the laplacian can not be put in block - diagonal form : even if one enumerates the vertices in the order of their cluster memberships ( by listing first the vertices of one cluster , then the vertices of another cluster , etc . ) there will always be some non - vanishing entries outside of the blocks . however , the lowest @xmath188 non - vanishing eigenvalues are still close to zero , and the vertex vectors of the first @xmath44 eigenvectors should still enable one to clearly distinguish the clusters in a @xmath44-dimensional space . vertex vectors corresponding to the same cluster are now not coincident , in general , but still rather close to each other . so , instead of @xmath44 points , one would observe @xmath44 groups of points , with the points of each group localized close to each other and far from the other groups . techniques like @xmath44-means clustering ( section [ sec4_3 ] ) can then easily recover the clusters . the scenario we have described is expected from perturbation theory @xcite . in principle all symmetric matrices that can be put in block - diagonal form have a set of eigenvectors ( as many as the blocks ) , such that the elements of each eigenvector are different from zero on the vertices of a block and zero otherwise , just like the laplacian . the adjacency matrix itself has the same property , for example . this is a necessary condition for the eigenvectors to be successfully used for graph clustering , but it is not sufficient . in the case of the laplacian , for a graph with @xmath44 connected components , we know that the eigenvectors corresponding to the @xmath44 lowest eigenvalues come each from one of the components . in the case of the adjacency matrix @xmath189 ( or of its weighted counterpart @xmath190 ) , instead , it may happen that large eigenvalues refer to the same component . so , if one takes the eigenvectors corresponding to the @xmath44 largest eigenvalues , where @xmath191 is the diagonal matrix whose elements are the vertex degrees . ] , some components will be overrepresented , while others will be absent . therefore , using the eigenvectors of @xmath189 ( or @xmath190 ) in spectral graph clustering is in general not reliable . moreover , the elements of the eigenvectors corresponding to the components should be sufficiently far from zero . to understand why , suppose that we take a ( symmetric , block - diagonal ) matrix , and that one or more elements of one of the eigenvectors corresponding to the connected components are very close to zero . if one perturbs the graph by adding edges between different components , all entries of the perturbed eigenvectors will become non - zero and some may have comparable values as the lowest elements of the eigenvectors on the blocks . therefore distinguishing vertices of different components may become a problem , even when the perturbation is fairly small , and misclassifications are likely . on the other hand , the non - vanishing elements of the ( normalized ) eigenvectors of the unnormalized laplacian , for instance , are all equal to @xmath192 , where @xmath193 is the number of vertices in the @xmath67-th component . in this way , there is a gap between the lowest element ( here they are all equal for the same eigenvector ) and zero . this holds as well for the normalized laplacian @xmath194 ( section [ sec1_2 ] ) . for the other normalized laplacian @xmath195 ( section [ sec1_2 ] ) , the non - zero elements of the eigenvectors corresponding to the connected components are proportional to the square root of the degree of the corresponding vertex so , if degrees are very different from each other , and especially if there are vertices with very low degree , some eigenvector elements may be quite small . as we shall see below , in the context of the technique by ng et al . @xcite , a suitable normalization procedure is introduced to alleviate this problem . now that we have explained why the laplacian matrix is particularly suitable for spectral clustering , we proceed with the description of three popular methods : _ unnormalized spectral clustering _ and two _ normalized spectral clustering _ techniques , proposed by shi and malik @xcite and by ng et al . @xcite , respectively . unnormalized spectral clustering uses the unnormalized laplacian @xmath136 . the inputs are the adjacency matrix @xmath189 ( @xmath190 for weighted graphs ) and the number @xmath44 of clusters to be recovered . the first step consists in computing the eigenvectors corresponding to the lowest @xmath44 eigenvalues of @xmath136 . then , one builds the @xmath186 matrix @xmath196 , whose columns are the @xmath44 eigenvectors . the @xmath15 rows of @xmath196 are used to represent the graph vertices in a @xmath44-dimensional euclidean space , through a cartesian system of coordinates . the points are then grouped in @xmath44 clusters by using @xmath44-means clustering or similar techniques ( section [ sec4_3 ] ) . normalized spectral clustering works in the same way . in the version by shi and malik @xcite , one uses the eigenvectors of the normalized laplacian @xmath194 ( section [ sec1_2 ] ) . in the algorithm by ng et al . @xcite one adopts the normalized laplacian @xmath195 ( section [ sec1_2 ] ) . here , however , the matrix @xmath196 is normalized by dividing the elements of each row by their sum , obtaining a new matrix @xmath197 , whose rows are then used to represent the vertices in space , as in the other methods . by doing so , it is much more unlikely that eigenvector components for a well - separated cluster are close to zero , a scenario which would make the classification of the corresponding vertices problematic , as we have said above . however , if the graph has some vertices with low degree , they may still be misclassified . spectral clustering is closely related to graph partitioning . relaxed versions of the minimization of ratio cut and normalized cut ( see section [ sec4_1 ] ) can be turned into spectral clustering problems , by following similar procedures as in spectral graph partitioning . the measure to minimize can be expressed in matrix form , obtaining similar expressions as for the cut size ( see eq . [ eqr ] ) , with index vectors defining the partition of the graph in groups through the values of their entries . for instance , for the minimum cut bipartition of section [ sec4_1 ] , there is only one index vector @xmath130 , whose components equal @xmath198 , where the signs indicate the two clusters . the relaxation consists in performing the minimization over all possible vectors @xmath130 , allowing for real - valued components as well . this version of the problem is exactly equivalent to spectral clustering . the relaxed minimization of ratio cut for a partition in @xmath44 clusters yields the @xmath15 @xmath44-dimensional vertex vectors of unnormalized spectral clustering @xcite ; for normalized cut one obtains the @xmath15 @xmath44-dimensional vertex vectors of normalized spectral clustering , with the normalized laplacian @xmath194 @xcite . the problem is then to turn the resulting vectors into a partition of the graph , which can be done by using techniques like @xmath44-means clustering , as we have seen above . however , it is still unclear what is the relation between the original minimum cut problem over actual graph partitions and the relaxed version of it , in particular how close one can come to the real solution via spectral clustering . random walks on graphs are also related to spectral clustering . in fact , by minimizing the number of edges between clusters ( properly normalized for measures like , e. g. , ratio cut and normalized cut ) one forces random walkers to spend more time within clusters and to move more rarely from one cluster to another . in particular , unnormalized spectral clustering with the laplacian @xmath194 has a natural link with random walks , because @xmath199 ( section [ sec1_2 ] ) , where @xmath200 is the transfer matrix @xmath201 . this has interesting consequences . for instance , meil and shi have proven that the normalized cut for a bipartition equals the total probability that a random walker moves from one of the clusters to the other in either sense @xcite . in this way , minimizing the normalized cut means looking for a partition minimizing the probability of transitions between clusters . spectral clustering requires the computation of the first @xmath44 eigenvectors of a laplacian matrix . if the graph is large , an exact computation of the eigenvectors is impossible , as it would require a time @xmath154 . fortunately there are approximate techniques , like the power method or krylov subspace techniques like the lanczos method @xcite , whose speed depends on the size of the eigengap @xmath151 , where @xmath202 and @xmath152 are the @xmath44-th and @xmath203-th smallest eigenvalue of the matrix . the larger the eigengap , the faster the convergence . in fact , the existence of large gaps between pairs of consecutive eigenvalues could suggest the number of clusters of the graph , an information which is not delivered by spectral clustering and which has to be given as input . we know that , for a disconnected graph with @xmath44 components , the first @xmath44 eigenvalues of the laplacian matrix ( normalized or not ) are zero , whether the @xmath203-th is non - zero . if the clusters are weakly connected to each other , one expects that the first @xmath44 eigenvalues remain close to zero , and that the @xmath203-th is clearly different from zero . by reversing this argument , the number of clusters of a graph could be derived by checking whether there is an integer @xmath44 such that the first @xmath44 eigenvalues are small and the @xmath203-th is relatively large . however , when the clusters are very mixed with each other , it may be hard to identify significant gaps between the eigenvalues . the last issue we want to point out concerns the choice of the laplacian matrix to use in the applications . if the graph vertices have the same or similar degrees , there is no substantial difference between the unnormalized and the normalized laplacians . if there are big inhomogeneities among the vertex degrees , instead , the choice of the laplacian considerably affects the results . in general , normalized laplacians are more promising because the corresponding spectral clustering techniques implicitly impose a double optimization on the set of partitions , such that the intracluster edge density is high and , at the same time , the intercluster edge density is low . on the contrary , the unnormalized laplacian is related to the intercluster edge density only . moreover , unnormalized spectral clustering does not always converge , and sometimes yields trivial partitions in which one or more clusters consist of a single vertex . of the normalized laplacians , @xmath194 is more reliable than @xmath195 because the eigenvectors of @xmath194 corresponding to the lowest eigenvalues are cluster indicator vectors , i. e. , they have equal non - vanishing entries in correspondence of the vertices of each cluster , and zero elsewhere , if the clusters are disconnected . the eigenvectors of @xmath195 , instead , are obtained by ( left- ) multiplying those of @xmath194 by the matrix @xmath204 : in this way , eigenvector components corresponding to vertices of the same cluster are no longer equal , in general , a complication that may induce artefacts in the spectral clustering procedure . a simple way to identify communities in a graph is to detect the edges that connect vertices of different communities and remove them , so that the clusters get disconnected from each other . this is the philosophy of divisive algorithms . the crucial point is to find a property of intercommunity edges that could allow for their identification . divisive methods do not introduce substantial conceptual advances with respect to traditional techniques , as they just perform hierarchical clustering on the graph at study ( section [ sec4_2 ] ) . the main difference with divisive hierarchical clustering is that here one removes inter - cluster edges instead of edges between pairs of vertices with low similarity and there is no guarantee _ a priori _ that inter - cluster edges connect vertices with low similarity . in some cases vertices ( with all their adjacent edges ) or whole subgraphs may be removed , instead of single edges . being hierarchical clustering techniques , it is customary to represent the resulting partitions by means of dendrograms . the most popular algorithm is that proposed by girvan and newman @xcite . the method is historically important , because it marked the beginning of a new era in the field of community detection and opened this topic to physicists . here edges are selected according to the values of measures of _ edge centrality _ , estimating the importance of edges according to some property or process running on the graph . the steps of the algorithm are : 1 . computation of the centrality for all edges ; 2 . removal of edge with largest centrality : in case of ties with other edges , one of them is picked at random ; 3 . recalculation of centralities on the running graph ; 4 . iteration of the cycle from step 2 . girvan and newman focused on the concept of _ betweenness _ , which is a variable expressing the frequency of the participation of edges to a process . they considered three alternative definitions : geodesic edge betweenness , random - walk edge betweenness and current - flow edge betweenness . in the following we shall refer to them as edge betweenness , random - walk betweenness and current - flow betweenness , respectively . edge betweenness is the number of shortest paths between all vertex pairs that run along the edge . it is an extension to edges of the popular concept of site betweenness , introduced by freeman in 1977 @xcite and expresses the importance of edges in processes like information spreading , where information usually flows through shortest paths . historically edge betweenness was introduced before site betweenness in a never published technical report by anthonisse @xcite . it is intuitive that intercommunity edges have a large value of the edge betweenness , because many shortest paths connecting vertices of different communities will pass through them ( fig . [ figure6 ] ) . as in the calculation of site betweenness , if there are two or more geodesic paths with the same endpoints that run through an edge , the contribution of each of them to the betweenness of the edge must be divided by the multiplicity of the paths , as one assumes that the signal / information propagates equally along each geodesic path . edge betweenness is highest for edges connecting communities . in the figure , the edge in the middle has a much higher betweenness than all other edges , because all shortest paths connecting vertices of the two communities run through it . reprinted figure with permission from ref . 2009 by springer . ] the betweenness of all edges of the graph can be calculated in a time that scales as @xmath205 , or @xmath129 on a sparse graph , with techniques based on breadth - first - search @xcite . in the context of information spreading , one could imagine that signals flow across random rather than geodesic paths . in this case the betweenness of an edge is given by the frequency of the passages across the edge of a random walker running on the graph ( random - walk betweenness ) . a random walker moving from a vertex follows each adjacent edge with equal probability . a pair of vertices is chosen at random , @xmath155 and @xmath156 . the walker starts at @xmath155 and keeps moving until it hits @xmath156 , where it stops . one computes the probability that each edge was crossed by the walker , and averages over all possible choices for the vertices @xmath155 and @xmath156 . it is meaningful to compute the _ net _ crossing probability , which is proportional to the number of times the walk crossed the edge in one direction . in this way one neglects back and forth passages that are accidents of the random walk and tell nothing about the centrality of the edge . calculation of random - walk betweenness requires the inversion of an @xmath167 matrix ( once ) , followed by obtaining and averaging the flows for all pairs of nodes . the first task requires a time @xmath154 , the second @xmath206 , for a total complexity @xmath207 $ ] , or @xmath154 for a sparse matrix . the complete calculation requires a time @xmath154 on a sparse graph . current - flow betweenness is defined by considering the graph a resistor network , with edges having unit resistance . if a voltage difference is applied between any two vertices , each edge carries some amount of current , that can be calculated by solving kirchoff s equations . the procedure is repeated for all possible vertex pairs : the current - flow betweenness of an edge is the average value of the current carried by the edge . it is possible to show that this measure is equivalent to random - walk betweenness , as the voltage differences and the random walks net flows across the edges satisfy the same equations @xcite . therefore , the calculation of current - flow betweenness has the same complexity @xmath207 $ ] , or @xmath154 for a sparse graph . calculating edge betweenness is much faster than current - flow or random walk betweenness [ @xmath129 versus @xmath154 on sparse graphs ] . in addition , in practical applications the girvan - newman algorithm with edge betweenness gives better results than adopting the other centrality measures @xcite . numerical studies show that the recalculation step 3 of girvan - newman algorithm is essential to detect meaningful communities . this introduces an additional factor @xmath14 in the running time of the algorithm : consequently , the edge betweenness version scales as @xmath208 , or @xmath154 on a sparse graph . on graphs with strong community structure , that quickly break into communities , the recalculation step needs to be performed only within the connected component including the last removed edge ( or the two components bridged by it if the removal of the edge splits a subgraph ) , as the edge betweenness of all other edges remains the same . this can help saving some computer time , although it is impossible to give estimates of the gain since it depends on the specific graph at hand . nevertheless , the algorithm is quite slow , and applicable to sparse graphs with up to @xmath209 vertices , with current computational resources . in the original version of girvan - newman s algorithm @xcite , the authors had to deal with the whole hierarchy of partitions , as they had no procedure to say which partition is the best . in a successive refinement @xcite , they selected the partition with the largest value of modularity ( see section [ sec3_2_2 ] ) , a criterion that has been frequently used ever since . the method can be simply extended to the case of weighted graphs , by suitably generalizing the edge betweenness . the betweenness of a weighted edge equals the betweenness of the edge in the corresponding unweighted graph , divided by the weight of the edge @xcite . there have been countless applications of the girvan - newman method : the algorithm is now integrated in well known libraries of network analysis programs . tyler et al . proposed a modification of the girvan - newman algorithm , to improve the speed of the calculation @xcite . the gain in speed was required by the analysis of graphs of gene co - occurrences , which are too large to be analyzed by the algorithm of girvan and newman . algorithms computing site / edge betweenness start from any vertex , taken as center , and compute the contribution to betweenness from all paths originating at that vertex ; the procedure is then repeated for all vertices @xcite . tyler et al . proposed to calculate the contribution to edge betweenness only from a limited number of centers , chosen at random , deriving a sort of monte carlo estimate . numerical tests indicate that , for each connected subgraph , it suffices to pick a number of centers growing as the logarithm of the number of vertices of the component . for a given choice of the centers , the algorithm proceeds just like that of girvan and newman . the stopping criterion is different , though , as it does not require the calculation of modularity on the resulting partitions , but relies on a particular definition of community . according to such definition , a connected subgraph with @xmath210 vertices is a community if the edge betweenness of any of its edges does not exceed @xmath211 . indeed , if the subgraph consists of two parts connected by a single edge , the betweenness value of that edge would be greater than or equal to @xmath211 , with the equality holding only if one of the two parts consists of a single vertex . therefore , the condition on the betweenness of the edges would exclude such situations , although other types of cluster structures might still be compatible with it . in this way , in the method of tyler et al . , edges are removed until all connected components of the partition are `` communities '' in the sense explained above . the monte carlo sampling of the edge betweenness necessarily induces statistical errors . as a consequence , the partitions are in general different for different choices of the set of center vertices . however , the authors showed that , by repeating the calculation many times , the method gives good results on a network of gene co - occurrences @xcite , with a substantial gain of computer time . the technique has been also applied to a network of people corresponding via email @xcite . in practical examples , only vertices lying at the boundary between communities may not be clearly classified , and be assigned sometimes to a group , sometimes to another . this is actually a nice feature of the method , as it allows to identify overlaps between communities , as well as the degree of membership of overlapping vertices in the clusters they belong to . the algorithm of girvan and newman , which is deterministic , is unable to accomplish this . another fast version of the girvan - newman algorithm has been proposed by rattigan et al . @xcite . here , a quick approximation of the edge betweenness values is carried out by using a _ network structure index _ , which consists of a set of vertex annotations combined with a distance measure @xcite . basically one divides the graph into regions and computes the distances of every vertex from each region . in this way rattigan et al . showed that it is possible to lower the complexity of the algorithm to @xmath21 , by keeping a fair accuracy in the estimate of the edge betweenness values . this version of the girvan - newman algorithm gives good results on the benchmark graphs proposed by brandes et al . @xcite ( see also section [ sec6_1 ] ) , as well as on a collaboration network of actors and on a citation network . chen and yuan have pointed out that counting all possible shortest paths in the calculation of the edge betweenness may lead to unbalanced partitions , with communities of very different size , and proposed to count only _ non - redundant _ paths , i. e. paths whose endpoints are all different from each other : the resulting betweenness yields better results than standard edge betweenness for mixed clusters on the benchmark graphs of girvan and newman @xcite . holme et al . have used a modified version of the algorithm in which vertices , rather than edges , are removed @xcite . a centrality measure for the vertices , proportional to their site betweenness , and inversely proportional to their indegree , is chosen to identify boundary vertices , which are then iteratively removed with all their edges . this modification , applied to study the hierarchical organization of biochemical networks , is motivated by the need to account for reaction kinetic information , that simple site betweenness does not include . the indegree of a vertex is solely used because it indicates the number of substrates to a metabolic reaction involving that vertex ; for the purpose of clustering the graph is considered undirected , as usual . the algorithm of girvan and newman is unable to find overlapping communities , as each vertex is assigned to a single cluster . pinney and westhead have proposed a modification of the algorithm in which vertices can be split between communities @xcite . to do that , they also compute the betweenness of all vertices of the graph . unfortunately the values of edge and site betweenness can not be simply compared , due to their different normalization , but the authors remarked that the two endvertices of an inter - cluster edge should have similar betweenness values , as the shortest paths crossing one of them are likely to reach the other one as well through the edge . so they take the edge with largest betweenness and remove it only if the ratio of the betweenness values of its endvertices is between @xmath18 and @xmath212 , with @xmath213 . otherwise , the vertex with highest betweenness ( with all its adjacent edges ) is temporarily removed . when a subgraph is split by vertex or edge removal , all deleted vertices belonging to that subgraph are `` copied '' in each subcomponent , along with all their edges . gregory @xcite has proposed a similar approach , named conga ( cluster overlap newman - girvan algorithm ) , in which vertices are split among clusters if their site betweenness exceeds the maximum value of the betweenness of the edges . a vertex is split by assigning some of its edges to one of its duplicates , and the rest to the other . there are several possibilities to do that , gregory proposed to go for the split that yields the maximum of a new centrality measure , called _ split betweenness _ , which is the number of shortest paths that would run between two parts of a vertex if the latter were split . the method has a worst - case complexity @xmath214 , or @xmath154 on a sparse graph , like the algorithm of girvan and newman . the code can be found at http://www.cs.bris.ac.uk/@xmath215steve/networks/index . + html . another promising track to detect inter - cluster edges is related to the presence of cycles , i. e. closed non - intersecting paths , in the graph . communities are characterized by a high density of edges , so it is reasonable to expect that such edges form cycles . on the contrary , edges lying between communities will hardly be part of cycles . based on this intuitive idea , radicchi et al . proposed a new measure , the edge clustering coefficient , such that low values of the measure are likely to correspond to intercommunity edges @xcite . the edge clustering coefficient generalizes to edges the notion of clustering coefficient introduced by watts and strogatz for vertices @xcite ( fig . [ edgeclus ] ) . schematic illustration of the edge clustering coefficient introduced by radicchi et al . the two grey vertices have five and six other neighbors , respectively . of the five possible triangles based on the edge connecting the grey vertices , three are actually there , yielding an edge clustering coefficient @xmath216 . courtesy by f. radicchi . ] the clustering coefficient of a vertex is the number of triangles including the vertex divided by the number of possible triangles that can be formed ( section [ sec1_1 ] ) . the edge clustering coefficient is defined as @xmath217 where @xmath67 and @xmath68 are the extremes of the edge , @xmath218 the number of cycles of length @xmath126 built upon edge @xmath219 and @xmath220 the possible number of cycles of length @xmath126 that one could build based on the existing edges of @xmath67 , @xmath68 and their neighbors . the number of actual cycles in the numerator is augmented by @xmath2 to enable a ranking among edges without cycles , which would all yield a coefficient @xmath221 equal to zero , independently of the degrees of the extremes @xmath67 and @xmath68 and their neighbors . usually , cycles of length @xmath222 ( triangles ) or @xmath223 are considered . the measure is ( anti)correlated with edge betweenness : edges with low edge clustering coefficient usually have high betweenness and vice versa , although the correlation is not perfect . the method works as the algorithm by girvan and newman . at each iteration , the edge with smallest clustering coefficient is removed , the measure is recalculated again , and so on . if the removal of an edge leads to a split of a subgraph in two parts , the split is accepted only if both clusters are ls - sets ( `` strong '' ) or `` weak '' communities ( see section [ sec3_1_2 ] ) . the verification of the community condition on the clusters is performed on the full adjacency matrix of the initial graph . if the condition were satisfied only for one of the two clusters , the initial subgraph may be a random graph , as it can be easily seen that by cutting a random graph la erds and rnyi in two parts , the larger of them is a strong ( or weak ) community with very high probability , whereas the smaller part is not . enforcing the community condition on both clusters , it is more likely that the subgraph to be split indeed has a cluster structure . therefore , the algorithm stops when all clusters produced by the edge removals are communities in the strong or weak sense , and further splits would violate this condition . the authors suggested to use the same stopping criterion for the algorithm of girvan and newman , to get structurally well - defined clusters . since the edge clustering coefficient is a local measure , involving at most an extended neighborhood of the edge , it can be calculated very quickly . the running time of the algorithm to completion is @xmath224 , or @xmath129 on a sparse graph , if @xmath126 is small , so it is much shorter than the running time of the girvan - newman method . the recalculation step becomes slow if @xmath126 is not so small , as in this case the number of edges whose coefficient needs to be recalculated may reach a sizeable fraction of the edges of the graph ; likewise , counting the number of cycles based on one edge becomes lengthier . if @xmath225 , where @xmath88 is the diameter of the graph ( which is usually a small number for real networks ) , the cycles span the whole graph and the measure becomes global and no more local . the computational complexity in this case exceeds that of the algorithm of girvan and newman , but it can come close to it for practical purposes even at lower values of @xmath126 . so , by tuning @xmath126 one can smoothly interpolate between a local and a global centrality measure . the software of the algorithm can be found in http://filrad.homelinux.org / data/. in a successive paper @xcite the authors extended the method to the case of weighted networks , by modifying the edge clustering coefficient of eq . [ eq3010 ] , in that the number of cycles @xmath218 is multiplied by the weight of the edge @xmath219 . the definitions of strong and weak communities can be trivially extended to weighted graphs by replacing the internal / external degrees of the vertices / clusters with the corresponding strengths . more recently , the method has been extended to bipartite networks @xcite , where only cycles of even length are possible ( @xmath226 , @xmath227 , @xmath228 , etc . ) . the algorithm by radicchi et al . may give poor results when the graph has few cycles , as it happens in some social and many non - social networks . in this case , in fact , the edge clustering coefficient is small and fairly similar for most edges , and the algorithm may fail to identify the bridges between communities . an alternative measure of centrality for edges is information centrality . it is based on the concept of efficiency @xcite , which estimates how easily information travels on a graph according to the length of shortest paths between vertices . the efficiency of a network is defined as the average of the inverse distances between all pairs of vertices . if the vertices are `` close '' to each other , the efficiency is high . the information centrality of an edge is the relative variation of the efficiency of the graph if the edge is removed . in the algorithm by fortunato et al . @xcite , edges are removed according to decreasing values of information centrality . the method is analogous to that of girvan and newman . computing the information centrality of an edge requires the calculation of the distances between all pairs of vertices , which can be done with breadth - first - search in a time @xmath205 . so , in order to compute the information centrality of all edges one requires a time @xmath208 . at this point one removes the edge with the largest value of information centrality and recalculates the information centrality of all remaining edges with respect to the running graph . since the procedure is iterated until there are no more edges in the network , the final complexity is @xmath229 , or @xmath230 on a sparse graph . the partition with the largest value of modularity is chosen as most representative of the community structure of the graph . the method is much slower than the algorithm of girvan and newman . partitions obtained with both techniques are rather consistent , mainly because information centrality has a strong correlation with edge betweenness . the algorithm by fortunato et al . gives better results when communities are mixed , i. e. with a high degree of interconnectedness , but it tends to isolate leaf vertices and small loosely bound subgraphs . a measure of vertex centrality based on loops , similar to the clustering coefficient by watts and strogatz @xcite , has been introduced by vragovi and louis @xcite . the idea is that neighbors of a vertex well inside a community are `` close '' to each other , even in the absence of the vertex , due to the high density of intra - cluster edges . suppose that @xmath68 and @xmath44 are neighbors of a vertex @xmath67 : @xmath231 is the length of a shortest path between @xmath68 and @xmath44 , if @xmath67 is removed from the graph . naturally , the existence of alternative paths to @xmath232 implies the existence of loops in the graph . vragovi and louis defined the _ loop coefficient _ of @xmath67 as the average of @xmath233 over all pairs of neighbors of @xmath67 , somewhat reminding of the concept of information centrality used in the method by fortunato et al . high values of the loop coefficient are likely to identify core vertices of communities , whereas low values correspond to vertices lying at the boundary between communities . clusters are built around the vertices with highest values of the loop coefficient . the method has time complexity @xmath234 ; its results are not so accurate , as compared to popular clustering techniques . newman - girvan modularity @xmath90 ( section [ sec3_2_2 ] ) , originally introduced to define a stopping criterion for the algorithm of girvan and newman , has rapidly become an essential element of many clustering methods . modularity is by far the most used and best known quality function . it represented one of the first attempts to achieve a first principle understanding of the clustering problem , and it embeds in its compact form all essential ingredients and questions , from the definition of community , to the choice of a null model , to the expression of the `` strength '' of communities and partitions . in this section we shall focus on all clustering techniques that require modularity , directly and/or indirectly . we will examine fast techniques that can be used on large graphs , but which do not find good optima for the measure @xcite ; more accurate methods , which are computationally demanding @xcite ; algorithms giving a good tradeoff between high accuracy and low complexity @xcite . we shall also point out other properties of modularity , discuss some extensions / modifications of it , as well as highlight its limits . by assumption , high values of modularity indicate good partitions . ] . so , the partition corresponding to its maximum value on a given graph should be the best , or at least a very good one . this is the main motivation for modularity maximization , by far the most popular class of methods to detect communities in graphs . an exhaustive optimization of @xmath90 is impossible , due to the huge number of ways in which it is possible to partition a graph , even when the latter is small . besides , the true maximum is out of reach , as it has been recently proved that modularity optimization is an np - complete problem @xcite , so it is probably impossible to find the solution in a time growing polynomially with the size of the graph . however , there are currently several algorithms able to find fairly good approximations of the modularity maximum in a reasonable time . the first algorithm devised to maximize modularity was a greedy method of newman @xcite . it is an agglomerative hierarchical clustering method , where groups of vertices are successively joined to form larger communities such that modularity increases after the merging . one starts from @xmath15 clusters , each containing a single vertex . edges are not initially present , they are added one by one during the procedure . however , the modularity of partitions explored during the procedure is always calculated from the full topology of the graph , as we want to find the modularity maximum on the space of partitions of the full graph . adding a first edge to the set of disconnected vertices reduces the number of groups from @xmath15 to @xmath143 , so it delivers a new partition of the graph . the edge is chosen such that this partition gives the maximum increase ( minimum decrease ) of modularity with respect to the previous configuration . all other edges are added based on the same principle . if the insertion of an edge does not change the partition , i. e. the edge is internal to one of the clusters previously formed , modularity stays the same . the number of partitions found during the procedure is @xmath15 , each with a different number of clusters , from @xmath15 to @xmath2 . the largest value of modularity in this subset of partitions is the approximation of the modularity maximum given by the algorithm . at each iteration step , one needs to compute the variation @xmath235 of modularity given by the merger of any two communities of the running partition , so that one can choose the best merger . however , merging communities between which there are no edges can never lead to an increase of @xmath90 , so one has to check only the pairs of communities which are connected by edges , of which there can not be more than @xmath14 . since the calculation of each @xmath235 can be done in constant time , this part of the calculation requires a time @xmath21 . after deciding which communities are to be merged , one needs to update the matrix @xmath236 expressing the fraction of edges between clusters @xmath67 and @xmath68 of the running partition ( necessary to compute @xmath90 ) , which can be done in a worst - case time @xmath20 . since the algorithm requires @xmath143 iterations ( community mergers ) to run to completion , its complexity is @xmath237 , or @xmath129 on a sparse graph , so it enables one to perform a clustering analysis on much larger networks than the algorithm of girvan and newman ( up to an order of @xmath238 vertices with current computers ) . in a later paper @xcite , clauset et al . pointed out that the update of the matrix @xmath236 in newman s algorithm involves a large number of useless operations , due to the sparsity of the adjacency matrix . this operation can be performed more efficiently by using data structures for sparse matrices , like _ max - heaps _ , which rearrange the data in the form of binary trees . clauset et al . maintained the matrix of modularity variations @xmath239 , which is also sparse , a max - heap containing the largest elements of each row of the matrix @xmath239 as well as the labels of the corresponding communities , and a simple array whose elements are the sums of the elements of each row of the old matrix @xmath236 . the optimization of modularity can be carried out using these three data structures , whose update is much quicker than in newman s technique . the complexity of the algorithm is @xmath240 , where @xmath88 is the depth of the dendrogram describing the successive partitions found during the execution of the algorithm , which grows as @xmath241 for graphs with a strong hierarchical structure . for those graphs , the running time of the method is then @xmath242 , which allows to analyse the community structure of very large graphs , up to @xmath243 vertices . the greedy optimization of clauset et al . is currently one of the few algorithms that can be used to estimate the modularity maximum on such large graphs . the code can be freely downloaded from http://cs.unm.edu/@xmath215aaron/research/fastmodulari + ty.htm . this greedy optimization of modularity tends to form quickly large communities at the expenses of small ones , which often yields poor values of the modularity maxima . danon et al . suggested to normalize the modularity variation @xmath235 produced by the merger of two communities by the fraction of edges incident to one of the two communities , in order to favor small clusters @xcite . this trick leads to better modularity optima as compared to the original recipe of newman , especially when communities are very different in size . wakita and tsurumi @xcite have noticed that , due to the bias towards large communities , the fast algorithm by clauset et al . is inefficient , because it yields very unbalanced dendrograms , for which the relation @xmath244 does not hold , and as a consequence the method often runs at its worst - case complexity . to improve the situation they proposed a modification in which , at each step , one seeks the community merger delivering the largest value of the product of the modularity variation @xmath235 times a factor ( _ consolidation ratio _ ) , that peaks for communities of equal size . in this way there is a tradeoff between the gain in modularity and the balance of the communities to merge , with a big gain in the speed of the procedure , that enables the analysis of systems with up to @xmath245 vertices . interestingly , this modification often leads to better modularity maxima than those found with the version of clauset et al . , at least on large social networks . the code can be found at http://www.is.titech.ac.jp/@xmath215wakita/en/software + /community - analysis - software/. another trick to avoid the formation of large communities was proposed by schuetz and caflisch and consists in allowing for the merger of more community pairs , instead of one , at each iteration @xcite . this generates several `` centers '' around which communities are formed , which grow simultaneously so that a condensation into a few large clusters is unlikely . this modified version of the greedy algorithm is combined with a simple refinement procedure in which single vertices are moved to the neighboring community that yields the maximum increase of modularity . the method has the same complexity of the fast optimization by clauset et al . , but comes closer to the modularity maximum . the software is available at http://www.biochem-caflisch.uzh.ch/public/5/net + work-clusterization-algorithm.html . the accuracy of the greedy optimization can be significantly improved if the hierarchical agglomeration is started from some reasonable intermediate configuration , rather than from the individual vertices @xcite . xiang et al . suggested to start from a configuration obtained by merging the original isolated vertices into larger subgraphs , according to the values of a measure of topological similarity between subgraphs @xcite . a similar approach has been described by ye et al . @xcite : here the initial partition is such that no single vertex can be moved from its cluster to another without decreasing @xmath90 . higher - quality modularities can be also achieved by applying refinement strategies based on local search at various steps of the greedy agglomeration @xcite . such refinement procedures are similar to the technique proposed by newman to improve the results of his spectral optimization of modularity ( @xcite and section [ sub_sec6_0_4 ] ) . another good strategy consists in alternating greedy optimization with stochastic perturbations of the partitions @xcite . a different greedy approach has been introduced by blondel et al . @xcite , for the general case of weighted graphs . initially , all vertices of the graph are put in different communities . the first step consists of a sequential sweep over all vertices . given a vertex @xmath67 , one computes the gain in weighted modularity ( eq . [ eq : mod2 ] ) coming from putting @xmath67 in the community of its neighbor @xmath68 and picks the community of the neighbor that yields the largest increase of @xmath90 , as long as it is positive . at the end of the sweep , one obtains the first level partition . in the second step communities are replaced by supervertices , and two supervertices are connected if there is at least an edge between vertices of the corresponding communities . in this case , the weight of the edge between the supervertices is the sum of the weights of the edges between the represented communities at the lower level . the two steps of the algorithm are then repeated , yielding new hierarchical levels and supergraphs ( fig . [ blondelmethod ] ) . we remark that modularity is always computed from the initial graph topology : operating on supergraphs enables one to consider the variations of modularity for partitions of the original graph after merging and/or splitting of groups of vertices . therefore , at some iteration , modularity can not increase anymore , and the algorithm stops . the technique is more limited by storage demands than by computational time . the latter grows like @xmath21 , so the algorithm is extremely fast and graphs with up to @xmath246 edges can be analyzed in a reasonable time on current computational resources . the software can be found at http://findcommunities.googlepages.com/. the modularity maxima found by the method are better than those found with the greedy techniques by clauset et al . @xcite and wakita and tsurumi @xcite . however , closing communities within the immediate neighborhood of vertices may be inaccurate and yield spurious partitions in practical cases . so , it is not clear whether some of the intermediate partitions could correspond to meaningful hierarchical levels of the graph . moreover , the results of the algorithm depend on the order of the sequential sweep over the vertices . we conclude by stressing that , despite the improvements and refinements of the last years , the accuracy of greedy optimization is not that good , as compared with other techniques . simulated annealing @xcite is a probabilistic procedure for global optimization used in different fields and problems . it consists in performing an exploration of the space of possible states , looking for the global optimum of a function @xmath247 , say its maximum . transitions from one state to another occur with probability @xmath2 if @xmath247 increases after the change , otherwise with a probability @xmath248 , where @xmath249 is the decrease of the function and @xmath19 is an index of stochastic noise , a sort of inverse temperature , which increases after each iteration . the noise reduces the risk that the system gets trapped in local optima . at some stage , the system converges to a stable state , which can be an arbitrarily good approximation of the maximum of @xmath247 , depending on how many states were explored and how slowly @xmath19 is varied . simulated annealing was first employed for modularity optimization by guimer et al . its standard implementation @xcite combines two types of `` moves '' : local moves , where a single vertex is shifted from one cluster to another , taken at random ; global moves , consisting of mergers and splits of communities . splits can be carried out in several distinct ways . the best performance is achieved if one optimizes the modularity of a bipartition of the cluster , taken as an isolated graph . this is done again with simulated annealing , where one considers only individual vertex movements , and the temperature is decreased until it reaches the running value for the global optimization . global moves reduce the risk of getting trapped in local minima and they have proven to lead to much better optima than using simply local moves @xcite . in practical applications , one typically combines @xmath250 local moves with @xmath15 global ones in one iteration . the method can potentially come very close to the true modularity maximum , but it is slow . the actual complexity can not be estimated , as it heavily depends on the parameters chosen for the optimization ( initial temperature , cooling factor ) , not only on the graph size . simulated annealing can be used for small graphs , with up to about @xmath251 vertices . extremal optimization ( eo ) is a heuristic search procedure proposed by boettcher and percus @xcite , in order to achieve an accuracy comparable with simulated annealing , but with a substantial gain in computer time . it is based on the optimization of local variables , expressing the contribution of each unit of the system to the global function at study . this technique was used for modularity optimization by duch and arenas @xcite . modularity can be indeed written as a sum over the vertices : the local modularity of a vertex is the value of the corresponding term in this sum . a fitness measure for each vertex is obtained by dividing the local modularity of the vertex by its degree , as in this case the measure does not depend on the degree of the vertex and is suitably normalized . one starts from a random partition of the graph in two groups with the same number of vertices . at each iteration , the vertex with the lowest fitness is shifted to the other cluster . the move changes the partition , so the local fitnesses of many vertices need to be recalculated . the process continues until the global modularity @xmath90 can not be improved any more by the procedure . this technique reminds one of the kernighan - lin @xcite algorithm for graph partitioning ( section [ sec4_1 ] ) , but here the sizes of the communities are determined by the process itself , whereas in graph partitioning they are fixed from the beginning . after the bipartition , each cluster is considered as a graph on its own and the procedure is repeated , as long as @xmath90 increases for the partitions found . the procedure , as described , proceeds deterministically from the given initial partition , as one shifts systematically the vertex with lowest fitness , and is likely to get trapped in local optima . better results can be obtained if one introduces a probabilistic selection , in which vertices are ranked based on their fitness values and one picks the vertex of rank @xmath91 with the probability @xmath252 ( @xmath253-eo , @xcite ) . the algorithm finds very good estimates of the modularity maximum , and performs very well on the benchmark of girvan and newman @xcite ( section [ sec6_1 ] ) . ranking the fitness values has a cost @xmath254 , which can be reduced to @xmath20 if heap data structures are used . choosing the vertex to be shifted can be done with a binary search , which amounts to an additional factor @xmath255 . finally , the number of steps needed to verify whether the running modularity maximum can be improved or not is also @xmath20 . the total complexity of the method is then @xmath128 . we conclude that eo represents a good tradeoff between accuracy and speed , although the use of recursive bisectioning may lead to poor results on large networks with many communities . modularity can be optimized using the eigenvalues and eigenvectors of a special matrix , the modularity matrix @xmath256 , whose elements are @xmath257 here the notation is the same used in eq . [ eq : mod ] . let @xmath130 be the vector representing any partition of the graph in two clusters @xmath258 and @xmath259 : @xmath260 if vertex @xmath67 belongs to @xmath258 , @xmath261 if @xmath67 belongs to @xmath259 . modularity can be written as @xmath262 the last expression indicates standard matrix products . the vector @xmath130 can be decomposed on the basis of eigenvectors @xmath263 ( @xmath140 ) of the modularity matrix @xmath256 : @xmath264 , with @xmath265 . by plugging this expression of @xmath130 into eq . [ eqr5 ] one finally gets @xmath266 where @xmath267 is the eigenvalue of @xmath256 corresponding to the eigenvector @xmath263 . [ eqr6 ] is analogous to eq . [ eqr1 ] for the cut size of the graph partitioning problem . this suggests that one can optimize modularity on bipartitions via spectral bisection ( section [ sec4_1 ] ) , by replacing the laplacian matrix with the modularity matrix @xcite . like the laplacian matrix , @xmath256 has always the trivial eigenvector @xmath268 with eigenvalue zero , because the sum of the elements of each row / column of the matrix vanishes . from eq . [ eqr6 ] we see that , if @xmath256 has no positive eigenvalues , the maximum coincides with the trivial partition consisting of the graph as a single cluster ( for which @xmath269 ) , i. e. it has no community structure . otherwise , one has to look for the eigenvector of @xmath10 with largest ( positive ) eigenvalue , @xmath270 , and group the vertices according to the signs of the components of @xmath270 , just like in section [ sec4_1 ] . here , however , one does not need to specify the sizes of the two groups : the vertices with positive components are all in one group , the others in the other group . if , for example , the component of @xmath270 corresponding to vertex @xmath67 is positive , but we set @xmath261 , the modularity is lower than by setting @xmath260 . the values of the components of @xmath270 are also informative , as they indicate the level of the participation of the vertices to their communities . in particular , components whose values are close to zero lie at the border between the two clusters and can be well considered as belonging to both of them . the result obtained from the spectral bipartition can be further improved by shifting single vertices from one community to the other , such to have the highest increase ( or lowest decrease ) of the modularity of the resulting graph partition . this refinement technique , similar to the kernighan - lin algorithm ( section [ sec4_1 ] ) , can be also applied to improve the results of other optimization techniques ( e.g. greedy algorithms , extremal optimization , etc . ) . the procedure is repeated for each of the clusters separately , and the number of communities increases as long as modularity does . at variance with graph partitioning , where one needs to fix the number of clusters and their size beforehand , here there is a clear - cut stopping criterion , represented by the fact that cluster subdivisions are admitted only if they lead to a modularity increase . we stress that modularity needs to be always computed from the full adjacency matrix of the original graph . the drawback of the method is similar as for spectral bisection , i. e. the algorithm gives the best results for bisections , whereas it is less accurate when the number of communities is larger than two . recently , sun et al . @xcite have added a step after each bipartition of a cluster , in that single vertices can be moved from one cluster to another and even form the seeds of new clusters . we remark that the procedure is different from the kernighan - lin - like refining steps , as here the number of clusters can change . this variant , which does not increase the complexity of the original spectral optimization , leads to better modularity maxima . moreover , one does not need to stick to bisectioning , if other eigenvectors with positive eigenvalues of the modularity matrix are used . given the first @xmath45 eigenvectors , one can construct @xmath15 @xmath45-dimensional vectors , each corresponding to a vertex , just like in spectral partitioning ( section [ sec4_4 ] ) . the components of the vector of vertex @xmath67 are proportional to the @xmath45 entries of the eigenvectors in position @xmath67 . then one can define _ community vectors _ , by summing the vectors of vertices in the same community . it is possible to show that , if the vectors of two communities form an angle larger that @xmath271 , keeping the communities separate yields larger modularity than if they are merged ( fig . [ qspectra_plane ] ) . in this way , in a @xmath45-dimensional space the modularity maximum corresponds to a partition in at most @xmath272 clusters . community vectors were used by wang et al . to obtain high - modularity partitions into a number of communities smaller than a given maximum @xcite . in particular , if one takes the eigenvectors corresponding to the two largest eigenvalues , one can obtain a split of the graph in three clusters : in a recent work , richardson et al . presented a fast technique to obtain graph tripartitions with large modularity along these lines @xcite . the eigenvectors with the most negative eigenvalues can also be used to extract useful information , like the presence of a possible multipartite structure of the graph , as they give the most relevant contribution to the modularity minimum . the spectral optimization of modularity is quite fast . the leading eigenvector of the modularity matrix can be computed with the power method , by repeatedly multiplying @xmath256 by an arbitrary vector ( not orthogonal to @xmath270 ) . the number of required iterations to reach convergence is @xmath20 . each multiplication seems to require a time @xmath129 , as @xmath256 is a complete matrix , but the peculiar form of @xmath256 allows for a much quicker calculation , taking time @xmath273 . so , a graph bipartition requires a time @xmath274 $ ] , or @xmath129 on a sparse graph . to find the modularity optimum one needs a number of subsequent bipartitions that equals the depth @xmath88 of the resulting hierarchical tree . in the worst - case scenario , @xmath275 , but in practical cases the procedure usually stops much before reaching the leaves of the dendrogram , so one could go for the average value @xmath276 , for a total complexity of @xmath128 . the algorithm is faster than extremal optimization and it is also slightly more accurate , especially for large graphs . the modularity matrix and the corresponding spectral optimization can be trivially extended to weighted graphs . spectral optimization of modularity by newman @xcite . by using the first two eigenvectors of the modularity matrix , vertices can be represented as points on a plane . by cutting the plane with a line passing through the origin ( like the dashed line in the figure ) one obtains bipartitions of the graph with possibly high modularity values . reprinted figure with permission from ref . 2006 by the american physical society . ] a different spectral approach had been previously proposed by white and smyth @xcite . let @xmath190 indicate the weighted adjancency matrix of a graph @xmath23 . a partition of @xmath23 in @xmath44 clusters can be described through an @xmath186 assignment matrix @xmath277 , where @xmath278 if vertex @xmath67 belongs to cluster @xmath117 , otherwise @xmath279 . it can be easily shown that , up to a multiplicative constant , modularity can be rewritten in terms of the matrix @xmath277 as @xmath280= -\mbox{tr}[{\bf x}^t{\bf { l}_q}{\bf x } ] , \label{eqr7}\ ] ] where @xmath281 is a diagonal matrix with identical elements , equal to the sum of all edge weights , and the entries of @xmath282 are @xmath283 , where @xmath105 is the degree of vertex @xmath67 . the matrix @xmath284 is called the @xmath90-laplacian . finding the assignment matrix @xmath277 that maximizes @xmath90 is an @xmath285-complete problem , but one can get a good approximation by relaxing the constraint that the elements of @xmath277 have to be discrete . by doing so @xmath90 becomes a sort of continuous functional of @xmath277 and one can determine the extremes of @xmath90 by setting its first derivative ( with respect to @xmath277 ) to zero . this leads to the eigenvalue problem @xmath286 here @xmath287 is a diagonal matrix . [ eqr8 ] turns modularity maximization into a spectral graph partitioning problem ( section [ sec4_4 ] ) , using the @xmath90-laplacian matrix . a nice feature of the @xmath90-laplacian is that , for graphs which are not too small , it can be approximated ( up to constant factors ) by the transition matrix @xmath288 , obtained by normalizing @xmath190 such that the sum of the elements of each row equals one . [ eqr8 ] is at the basis of the algorithms developed by white and smyth , which search for partitions with at most @xmath289 clusters , where @xmath289 is a predefined input parameter that may be suggested by preliminary information on the graph cluster structure . the first @xmath290 eigenvectors of the transition matrix @xmath288 ( excluding the trivial eigenvector with all equal components ) can be computed with a variant of the lanczos method @xcite . since the eigenvector components are not integer , the eigenvectors do not correspond directly to a partition of the graph in clusters . however , as usual in spectral graph partitioning , the components of the eigenvectors can be used as coordinates of the graph vertices in an euclidean space and @xmath44-means clustering is applied to obtain the desired partition . white and smyth proposed two methods to derive the clustering after embedding the graph in space . both methods have a worst - case complexity @xmath291 , which is essentially linear in the number of vertices of the graph if the latter is sparse and @xmath292 . however , on large and sparse graphs , @xmath289 could scale with the number of vertices @xmath15 , so the procedure might become quite slow . in order to speed up calculations without losing much accuracy in the final estimates of the maximum modularity , ruan and zhang @xcite have proposed an algorithm , called _ kcut _ , that applies recursively the method by white and smyth , in a slightly modified form : after a first application to the full graph , in the next iteration the method is applied to all clusters of the first partition , treated as independent networks , and so on . the procedure goes on as long as the modularity of the resulting partitions keeps growing . the advantage of _ kcut _ is that one can play with low values for the ( maximal ) number of clusters @xmath293 at each iteration ; if partitions are balanced , after @xmath294 levels of recursions , the number of clusters of the partition is approximately @xmath295 . therefore the complexity of _ kcut _ is @xmath296 $ ] for a final partition in ( at most ) @xmath289 clusters , which is much lower than the complexity of the algorithm by white and smyth . ruan and zhang tested _ kcut _ on artificial graphs generated with the planted @xmath293-partition model ( section [ sec6 ] ) , and on real networks including zachary s karate club @xcite , the american college football network @xcite and two collaboration networks of jazz musicians @xcite and physicists @xcite : the accuracy of _ kcut _ is comparable to that of the algorithm by white and smyth , though generally lower . agarwal and kempe have suggested to maximize modularity within the framework of mathematical programming @xcite . in fact , modularity optimization can be formulated both as a linear and as a quadratic program . in the first case , the variables are defined on the links : @xmath297 if @xmath67 and @xmath68 are in the same cluster , otherwise @xmath298 . the modularity of a partition , up to a multiplicative constant , can then be written as @xmath299 where @xmath256 is the modularity matrix defined by newman ( see section [ sub_sec6_0_4 ] ) . [ eqr9 ] is linear in the variables @xmath300 , which must obey the constraint @xmath301 , because , if @xmath67 and @xmath68 are in the same cluster , and so are @xmath67 and @xmath44 , then @xmath68 and @xmath44 must be in that cluster too . maximizing the expression in eq . [ eqr9 ] under the above constraint is @xmath285-hard , if the variables have to be integer as required . however , if one relaxes this condition by using real - valued @xmath300 , the problem can be solved in polynomial time @xcite . on the other hand , the solution does not correspond to an actual partition , as the @xmath302 variables are fractional . to get clusters out of the @xmath300 one needs a rounding step . the values of the @xmath302 variables are used as sort of distances in a metric space ( the triangular inequality is satisfied by construction ) : clusters of vertices `` close '' enough to each other ( i. e. whose mutual @xmath302 variables are close to zero ) are formed and removed until each vertex is assigned to a cluster . the resulting partition is further refined with the same post - processing technique used by newman for the spectral optimization of modularity , i. e. by a sequence of steps similar to those of the algorithm by kernighan and lin ( see section [ sub_sec6_0_4 ] ) . quadratic programming can be used to get bisections of graphs with high modularity , that can be iterated to get a whole hierarchy of partitions as in newman s spectral optimization . one starts from one of the identities in eq . [ eqr5 ] @xmath303 where @xmath304 , depending on whether the vertex belongs to the first or the second cluster . since the optimization of the expression in eq . [ eqr10 ] is @xmath285-complete , one must relax again the constraint on the variables @xmath155 being integer . a possibility is to transform each @xmath155 into an @xmath15-dimensional vector @xmath305 and each product in the scalar product between vectors . the vectors are normalized so that their tips lie on the unit - sphere of the @xmath15-dimensional space . this vector problem is polynomially solvable , but one needs a method to associate a bipartition to the set of @xmath15 vectors of the solution . any @xmath306-dimensional hyperplane centered at the origin cuts the space in two halves , separating the vectors in two subsets . one can then choose multiple random hyperplanes and pick the one which delivers the partition with highest modularity . as in the linear program , a post - processing technique la newman ( see section [ sub_sec6_0_4 ] ) is used to improve the results of the procedure . the two methods proposed by agarwal and kempe are strongly limited by their high computational complexity , due mostly to the large storage demands , making graphs with more than @xmath251 vertices intractable . on the other hand , the idea of applying mathematical programming to graph clustering is promising . the code of the algorithms can be downloaded from http://www - scf.usc.edu/@xmath215gaurava/. in a recent work @xcite , xu et al . have optimized modularity using mixed - integer mathematical programming , with both integer and continuous variables , obtaining very good approximations of the modularity optimum , at the price of high computational costs . chen et al . have used integer linear programming to transform the initial graph into an optimal target graph consisting of disjoint cliques , which effectively yields a partition @xcite . berry et al . have formulated the problem of graph clustering as a _ facility location problem _ @xcite , consisting in the minimization of a cost function based on a local variation of modularity @xcite . lehmann and hansen @xcite optimized modularity via mean field annealing @xcite , a deterministic alternative to simulated annealing @xcite . the method uses gibbs probabilities to compute the conditional mean value for the variable of a vertex , which indicates its community membership . by making a mean field approximation on the variables of the other vertices in the gibbs probabilities one derives a self - consistent set of non - linear equations , that can be solved by iteration in a time @xmath307 $ ] . the method yields better modularity maxima than the spectral optimization by newman ( section [ sub_sec6_0_4 ] ) , at least on artificial graphs with built - in community structure , similar to the benchmark graphs by girvan and newman ( section [ sec6_1 ] ) . genetic algorithms @xcite have also been used to optimize modularity . in a standard genetic algorithm one has a set of candidate solutions to a problem , which are numerically encoded as chromosomes , and an objective function to be optimized on the space of solutions . the objective function plays the role of biological fitness for the chromosomes . one usually starts from a random set of candidate solutions , which are progressively changed through manipulations inspired by biological processes regarding real chromosomes , like point mutation ( random variations of some parts of the chromosome ) and crossing over ( generating new chromosomes by merging parts of existing chromosomes ) . then , the fitness of the new pool of candidates is computed and the chromosomes with the highest fitness have the greatest chances to survive in the next generation . after several iterations only solutions with large fitness survive . in a work by tasgin et al . @xcite , partitions are the chromosomes and modularity is the fitness function . with a suitable choice of the algorithm parameters , like the number of chromosomes and the rates of mutation and crossing over , tasgin et al . could obtain results of comparative quality as greedy modularity optimization on zachary s karate club @xcite , the college football network @xcite and the benchmark by girvan and newman ( section [ sec6_1 ] ) . genetic algorithms were also adopted by liu et al . @xcite . here the maximum modularity partition is obtained via successive bipartitions of the graph , where each bipartition is determined by applying a genetic algorithm to each subgraph ( starting from the original graph itself ) , which is considered isolated from the rest of the graph . a bipartition is accepted only if it increases the total modularity of the graph . in section [ sec3_2_2 ] we have seen that the modularity maximum is obtained for the partition that minimizes the difference between the cut size and the expected cut size of the partition ( eq . [ eqr12 ] ) . in the complete weighted graph @xmath308 such that the weight @xmath309 of an edge is @xmath310 , if @xmath67 and @xmath68 are connected in @xmath23 , and @xmath311 if they are not , the difference @xmath312 is simply the cut size of partition @xmath94 . so , maximizing modularity for @xmath313 is equivalent to the problem of finding the partition with minimal cut size of the weighted graph @xmath308 , i. e. to a graph partitioning problem . the problem can then be efficiently solved by using existing software for graph partitioning @xcite . in the most recent literature on graph clustering several modifications and extensions of modularity can be found . they are usually motivated by specific classes of clustering problems and/or graphs that one may want to analyze . modularity can be easily extended to graphs with weighted edges @xcite . one needs to replace the degrees @xmath105 and @xmath106 in eq . [ eq : mod ] with the strengths @xmath314 and @xmath315 of vertices @xmath67 and @xmath68 . we remind that the strength of a vertex is the sum of the weights of edges adjacent to the vertex ( section [ sec1_1 ] ) . for a proper normalization , the number of edges @xmath14 in eq . [ eq : mod ] has to be replaced by the sum @xmath86 of the weights of all edges . the product @xmath316 is now the expected weight of the edge @xmath219 in the null model of modularity , which has to be compared with the actual weight @xmath317 of that edge in the original graph . this can be understood if we consider the case in which all weights are multiples of a unit weight , so they can be rewritten as integers . the weight of the connection between two nodes can then be replaced by as many edges between the nodes as expressed by the number of weight units . for the resulting multigraph we can use the same procedure as in the case of unweighted graphs , which leads to the formally identical expression @xmath318 which can be also written as a sum over the modules @xmath319 , \label{eq : mod20}\ ] ] where @xmath320 is the sum of the weights of the internal edges of module @xmath117 and @xmath321 is the sum of the strengths of the vertices of @xmath117 . if edge weights are not mutually commensurable , one can always represent them as integers with good approximation , provided a sufficiently small weight unit is adopted , so the expressions for weighted modularity of eqs . [ eq : mod2 ] , [ eq : mod20 ] are generally valid . in principle , weights can be assigned to the edges of an undirected graph , by using any measure of similarity / correlation between the vertices ( like , e. g. , the measures introduced in section [ sec3_1_4 ] ) . in this way , one could derive the corresponding weighted modularity and use it to detect communities , with a potentially better exploitation of the structural information of the graph as compared to standard modularity @xcite . modularity has also a straightforward extension to the case of directed graphs @xcite . if an edge is directed , the probability that it will be oriented in either of the two possible directions depends on the in- and out - degrees of the end - vertices . for instance , taken two vertices @xmath9 and @xmath10 , where @xmath9 ( @xmath10 ) has a high ( low ) indegree and low ( high ) outdegree , in the null model of modularity an edge will be much more likely to point from @xmath10 to @xmath9 than from @xmath9 to @xmath10 . therefore , the expression of modularity for directed graphs reads @xmath322 where the factor @xmath323 in the denominator of the second summand has been dropped because the sum of the indegrees ( outdegrees ) equals @xmath14 , whereas the sum of the degrees of the vertices of an undirected graph equals @xmath109 ; the factor @xmath323 in the denominator of the prefactor has been dropped because the number of non - vanishing elements of the adjacency matrix is @xmath14 , not @xmath109 as in the symmetric case of an undirected graph . if a graph is both directed and weighted , formulas [ eq : mod2 ] and [ eq : mod3 ] can be combined as @xmath324 which is the most general ( available ) expression of modularity @xcite . kim et al . @xcite have remarked that the directed modularity of eq . [ eq : mod3 ] may not properly account for the directedness of the edges ( fig . [ directedq ] ) , and proposed a different definition based on diffusion on directed graphs , inspired by google s pagerank algorithm @xcite . rosvall and bergstrom raised similar objections @xcite . problem of the directed modularity introduced by arenas et al . the two situations illustrated are equivalent for modularity , as vertices @xmath9 and @xmath325 , as well as @xmath10 and @xmath326 , have identical indegrees and outdegrees . in this way , the optimization of directed modularity is not able to distinguish a situation in which there is directed flow ( top ) or not ( bottom ) . reprinted figure with permission from ref . @xcite.,width=340 ] if vertices may belong to more clusters , it is not obvious how to find a proper generalization of modularity . in fact , there is no unique recipe . shen et al . @xcite , for instance , suggested the simple definition @xmath327 here @xmath328 is the number of communities including vertex @xmath67 . the contribution of each edge to modularity is then the smaller , the larger the number of communities including its endvertices . nicosia et al . @xcite have made some more general considerations on the problem of extending modularity to the case of overlapping communities . they considered the case of directed unweighted networks , starting from the following general expression @xmath329 , \label{eq : mod8}\ ] ] where @xmath330 and @xmath331 are the indegree and outdegree of vertices @xmath67 and @xmath68 , the index @xmath117 labels the communities and @xmath332 , @xmath333 express the contributions to the sum corresponding to the edge @xmath219 in the network and in the null model , due to the multiple memberships of @xmath67 and @xmath68 . if there is no overlap between the communities , @xmath334 , where @xmath335 and @xmath336 correspond to the communities of @xmath67 and @xmath68 . in this case , the edge @xmath219 contributes to the sum only if @xmath337 , as in the original definition of modularity . for overlapping communities , the coefficients @xmath332 , @xmath333 must depend on the membership coefficients @xmath338 , @xmath339 of vertices @xmath67 and @xmath68 . one can assume that @xmath340 , where @xmath341 is some function . the term @xmath333 is related to the null model of modularity , and it must be handled with care . in modularity s original null model edges are formed by joining two random stubs , so one needs to define the membership of a random stub in the various communities . if we assume that there is no correlation _ a priori _ between the membership coefficients of any two vertices , we can assign to a stub originating from a vertex @xmath67 in community @xmath117 the average membership corresponding to all edges which can be formed with @xmath67 . on a directed graph we have to distinguish between outgoing and incoming stubs , so one has @xmath342 @xmath343 and one can write the following general expression for modularity @xmath344 . \label{eq : mod11}\ ] ] the question now concerns the choice of the function @xmath345 . if the formula of eq . [ eq : mod11 ] is to be an extension of modularity to the case of overlapping communities , it has to satisfy some general properties of classical modularity . for instance , the modularity value of a cover consisting of the whole network as a single cluster should be zero . it turns out that a large class of functions yield an expression for modularity that fulfills this requirement . otherwise , the choice of @xmath341 is rather arbitrary and good choices can be only tested _ a posteriori _ , based on the results of the optimization . membership coefficients are also present in an extension of modularity to overlapping communities proposed by shen et al . @xcite . here the membership coefficient of vertex @xmath29 in community @xmath117 is a sum over the edges of @xmath29 belonging to @xmath117 , where each edge has a weight proportional to the number of maximal cliques of @xmath117 containing the edge . gaertler et al . have introduced quality measures based on modularity s principle of the comparison between a variable relative to the original graph and the corresponding variable of a null model @xcite . they remark that modularity is just the difference between the coverage of a partition and the expected coverage of the partition in the null model . we remind that the coverage of a partition is the ratio between the number of edges within clusters and the total number of edges ( section [ sec3_2_2 ] ) . based on this observation , gaertler et al . suggest that the comparison between the two terms can be done with other binary operations as well . for instance , one could consider the ratio @xmath346 where the notation is the same as in eq . [ eq : mod1 ] . this can be done as well for any variable other than coverage . by using performance , for instance , ( section [ sec3_2_2 ] ) one obtains two new quality functions @xmath347 and @xmath348 , corresponding to taking the difference or the ratio between performance and its null model expectation value , respectively . gaertler et al . compared the results obtained with the four functions @xmath349 , @xmath350 , @xmath347 and @xmath348 , on a class of benchmark graphs with built - in cluster structure ( section [ sec6_1 ] ) and social networks . they found that the `` absolute '' variants @xmath351 and @xmath347 are better than the `` relative '' variants @xmath350 and @xmath348 on the artificial benchmarks , whereas @xmath348 is better on social networks . furthermore @xmath347 is better than the standard modularity @xmath351 . modifications of modularity s null model have been introduced by massen and doye @xcite and muff et al . massen and doye s null model is still a graph with the same expected degree sequence as the original , and with edges rewired at random among the vertices , but one imposes the additional constraint that there can be neither multiple edges between a pair of vertices nor edges joining a vertex with itself ( loops or self - edges ) . this null model is more realistic , as multiple edges and loops are usually absent in real graphs . the maximization of the corresponding modified modularity yields partitions with smaller average cluster size than standard modularity . the latter tends to disfavor small communities , because the actual densities of edges inside small communities hardly exceed the null model densities , which are appreciably enhanced by the contributions from multiple connections and loops . muff et al . proposed a local version of modularity , in which the expected number of edges within a module is not calculated with respect to the full graph , but considering just a portion of it , namely the subgraph including the module and its neighbouring modules . their motivation is the fact that modularity s null model implicitly assumes that each vertex could be attached to any other , whereas in real cases a cluster is usually connected to few other clusters . on a directed graph , their _ localized modularity _ @xmath352 reads @xmath353 . \label{eqr16}\ ] ] in eq . [ eqr16 ] @xmath116 is the number of edges inside cluster @xmath117 , @xmath354 ( @xmath355 ) the total internal ( external ) degree of cluster @xmath117 and @xmath356 the total number of edges in the subgraph comprising cluster @xmath117 and its neighbor clusters . the localized modularity is not bounded by @xmath2 , but can take any value . its maximization delivers more accurate partitions than standard modularity optimization on a model network describing the social interactions between children in a school ( school network ) and on the metabolic and protein - protein interaction networks of _ e. coli_. reichardt and bornholdt have shown that it is possible to reformulate the problem of community detection as the problem of finding the ground state of a spin glass model @xcite . each vertex @xmath67 is labeled by a potts spin variable @xmath357 , which indicates the cluster including the vertex . the basic principle of the model is that edges should connect vertices of the same class ( i. e. same spin state ) , whereas vertices of different classes ( i. e. different spin states ) should be disconnected ( ideally ) . so , one has to energetically favor edges between vertices in the same class , as well as non - edges between vertices in different classes , and penalize edges between vertices of different classes , along with non - edges between vertices in the same class . the resulting hamiltonian of the spin model is @xmath358 where @xmath359 is a constant expressing the coupling strength , @xmath360 are the elements of the adjacency matrix of the graph , @xmath361 a parameter expressing the relative contribution to the energy from existing and missing edges , and @xmath362 is the expected number of links connecting @xmath67 and @xmath68 for a null model graph with the same total number of edges @xmath14 of the graph considered . the system is a spin glass @xcite , as the couplings @xmath363 between spins are both ferromagnetic ( on the edges of the graph , provided @xmath364 ) and antiferromagnetic ( between disconnected vertices , as @xmath365 and @xmath366 ) . the multiplicative costant @xmath359 is irrelevant for practical purposes , so in the following we set @xmath367 . the range of the spin - spin interaction is infinite , as there is a non - zero coupling between any pair of spins . [ eqr13 ] bears a strong resemblance with the expression of modularity of eq . [ eq : mod ] . in fact , if @xmath368 and @xmath369 we recover exactly modularity , up to a factor @xmath370 . in this case , finding the spin configuration for which the hamiltonian is minimal is equivalent to maximizing modularity . [ eqr13 ] is much more general than modularity , though , as both the null model and the parameter @xmath371 can be arbitrarily chosen . in particular , the value of @xmath371 determines the importance of the null model term @xmath362 in the quality function . . [ eqr13 ] can be rewritten as @xmath372=-\sum_{s=1}c_{ss}\\ & = & \sum_{s < r}\left[l_{rs}-\gamma(l_{rs})_{p_{ij}}\right]=\sum_{s < r}a_{rs}. \label{eqr14}\end{aligned}\ ] ] here , the sums run over the clusters : @xmath373 and @xmath374 indicate the number of edges within cluster @xmath155 and between clusters @xmath375 and @xmath155 , respectively ; @xmath376 and @xmath377 are the corresponding null model expectation values . [ eqr14 ] defines the coefficients @xmath378 of _ cohesion _ and @xmath379 of _ adhesion_. if a subset of a cluster @xmath155 has a larger coefficient of adhesion with another cluster @xmath375 than with its complement in @xmath155 , the energy can be reduced by merging the subset with cluster @xmath375 . in the particular case in which the coefficient of adhesion of a subset @xmath380 of a cluster @xmath155 with its complement in the cluster exactly matches the coefficient of adhesion of @xmath380 with another cluster @xmath375 , the partitions in which @xmath380 stays within @xmath155 or is merged with @xmath375 have the same energy . in this case one can say that clusters @xmath375 and @xmath155 are overlapping . in general , the partition with minimum energy has the following properties : 1 ) every subset of each cluster has a coefficient of adhesion with its complement in the cluster not smaller than with any other cluster ; 2 ) every cluster has non - negative coefficient of cohesion ; 3 ) the coefficient of adhesion between any two clusters is non - positive . by tuning the parameter @xmath371 one can vary the number of clusters in the partition with minimum energy , going from a single cluster comprising all vertices ( @xmath381 ) , to @xmath15 clusters with a single vertex ( @xmath382 ) . so , @xmath371 is a resolution parameter that allows to explore the cluster structure of a graph at different scales ( see section [ sub_sec6_1 ] ) . the authors used single spin heatbath simulated annealing algorithms to find the ground state of the hamiltonian of eq . [ eqr13 ] . another generalization of modularity was recently suggested by arenas et al . they remarked that the fundamental unit to define modularity is the edge , but that high edge densities inside clusters usually imply the existence of long - range topological correlations between vertices , which are revealed by the presence of _ motifs _ @xcite , i. e. connected undirected subgraphs , like cycles ( section [ sec1_1 ] ) . for instance , a high edge density inside a cluster usually means that there are also several triangles in the cluster , and comparatively few between clusters , a criterion that has inspired on its own popular graph clustering algorithms @xcite . modularity can then be simply generalized by comparing the density of motifs inside clusters with the expected density in modularity s null model ( _ motif modularity _ ) . as a particular case , the _ triangle modularity _ of a partition @xmath22 reads @xmath383 where @xmath384 ( @xmath385 is the label of the cluster @xmath67 belongs to ) , @xmath386 ( @xmath105 is the degree of vertex @xmath67 ) and @xmath387 . if one chooses as motifs paths with even length , and removes the constraint that all vertices of the motif / path should stay inside the same cluster , maximizing motif modularity could reveal the existence of multipartite structure . for example , if a graph is bipartite , one expects to see many @xmath323-paths starting from one vertex class and returning to it from the other class . motif modularity can be trivially extended to the case of weighted graphs . several graphs representing real systems are built out of correlation data between elements . correlation matrices are very common in the study of complex systems : well - known examples are the correlations of price returns , which are intensively studied by economists and econophysicists @xcite . correlations may be positive as well as negative , so the corresponding weighted edges indicate both attraction and repulsion between pairs of vertices . usually the correlation values are filtered or otherwise transformed such to eliminate the weakest correlations and anticorrelations and to maintain strictly positive weights for the edges , yielding graphs that can be treated with standard techniques . however , ignoring negative correlations means to give up useful information on the relationships between vertices . finding clusters in a graph with both positive and negative weights is called _ correlation clustering problem _ @xcite . according to intuition , one expects that vertices of the same cluster are linked by positive edges , whereas vertices of different clusters are linked by negative edges . the best cluster structure is the partition that maximizes the sum of the strengths ( in absolute value ) of positive edges within clusters and negative edges between clusters , or , equivalently , the partition that minimizes the sum of the strengths ( in absolute value ) of positive edges between clusters and negative edges within clusters . this can be formulated by means of modularity , if one accounts for the contribution of the negative edges . a natural way to proceed is to create two copies of the graph at study : in one copy only the weights of the positive edges are kept , in the other only the weights of the negative edges ( in absolute value ) . by applying eq . [ eq : mod2 ] to the same partition of both graphs , one derives the contributions @xmath388 and @xmath389 to the modularity of that partition for the original graph . gmez et al . define the global modularity as a linear combination of @xmath388 and @xmath389 , that accounts for the relative total strengths of positive and negative edge weights @xcite . kaplan and forrest @xcite have proposed a similar expression , with two important differences . first , they have used the total strength of the graph , i. e. the sum of the absolute values of all weights , to normalize @xmath388 and @xmath389 ; gmez et al . instead have used the positive and the negative strengths , for @xmath388 and @xmath389 , respectively , which seems to be the more natural choice looking at eq . [ eq : mod2 ] . second , kaplan and forrest have given equal weight to the contributions of @xmath388 and @xmath389 to their final expression of modularity , which is just the difference @xmath390 . in another work , traag and bruggeman @xcite have introduced negative links in the general spin glass formulation of modularity of reichardt and bornholdt @xcite . here the relative importance of the contribution of positive and negative edge weights is a free parameter , the tuning of which allows to detect communities of various sizes and densities of positive / negative edges . some authors have pointed out that the original expression of modularity is not ideal to detect communities in bipartite graphs , which describe several real systems , like food webs @xcite , scientific @xcite and artistic @xcite collaboration networks , etc .. expressions of modularity for bipartite graphs were suggested by guimer et al . @xcite and barber @xcite . guimer et al . call the two classes of vertices actors and teams , and indicate with @xmath391 the degree of actor @xmath67 and @xmath392 the degree of team @xmath294 . the null model graphs are random graphs with the same expected degrees for the vertices , as usual . the bipartite modularity @xmath393 for a partition @xmath94 ( of the actors ) has the following expression @xmath394 . \label{eqr17}\ ] ] here , @xmath395 is the number of teams in which actors @xmath67 and @xmath68 are together and the sum @xmath396 gives the number of ordered pairs of actors in the same team . the second ratio of each summand is the null model term , indicating the expected ( normalized ) number of teams for pairs of actors in cluster @xmath117 . the bipartite modularity can also be applied to ( unipartite ) directed graphs : each vertex can be duplicated and assigned to both classes , based on its twofold role of source and target for the edges . another interesting alternative was introduced by barber @xcite and is a simple extension of eq . [ eq : mod ] . let us suppose that the two vertex classes ( red and blue ) are made out of @xmath45 and @xmath91 vertices , respectively . the degree of a red vertex @xmath67 is indicated with @xmath105 , that of a blue vertex @xmath68 with @xmath397 . the adjacency matrix @xmath189 of the graph is in block off - diagonal form , as there are edges only between red and blue vertices . because of that , barber assumes that the null model matrix @xmath398 , whose element @xmath100 indicates as usual the expected number of edges between vertices @xmath67 and @xmath68 in the null model , also has the block off - diagonal form @xmath399 where the @xmath400 are square matrices with all zero elements and @xmath401 , as in the null model of standard modularity ( though other choices are possible ) . the modularity maximum can be computed through the modularity matrix @xmath402 , as we have seen in section [ sub_sec6_0_4 ] . however , spectral optimization of modularity gives excellent results for bipartitions , while its performance worsens when the number of clusters is unknown , as it is usually the case . barber has proposed a different optimization technique , called bipartite recursively induced modules ( brim ) , based on the bipartite nature of the graph . the algorithm is based on the special expression of modularity for the bipartite case , for which once the partition of the red or the blue vertices is known , it is easy to get the partition of the other vertex class that yields the maximum modularity . therefore , one starts from an arbitrary partition in @xmath117 clusters of , say , the blue vertices , and recovers the partition of the red vertices , which is in turn used as input to get a better partition of the blue vertices , and so on until modularity converges . brim does not predict the number of clusters @xmath117 of the graph , but one can obtain good estimates for it by exploring different values with a simple bisection approach . typically , for a given @xmath117 the algorithm needs a few steps to converge , each step having a complexity @xmath21 . an expression of the number of convergence steps in terms of @xmath15 and/or @xmath14 still needs to be derived . in this section we shall discuss some features of modularity , which are crucial to identify the domain of its applicability and ultimately to assess the issue of the reliability of the measure for the problem of graph clustering . an important question concerns the value of the maximum modularity @xmath403 for a graph . we know that it must be non - negative , as there is always at least a partition with zero modularity , consisting in a single cluster with all vertices ( section [ sec3_2_2 ] ) . however , a large value for the modularity maximum does not necessarily mean that a graph has community structure . random graphs are supposed to have no community structure , as the linking probability between vertices is either constant or a function of the vertex degrees , so there is no bias _ a priori _ towards special groups of vertices . still , random graphs may have partitions with large modularity values @xcite . this is due to fluctuations in the distribution of edges in the graph , which in many graph realizations is not homogeneous even if the linking probability is constant , like in erds - rnyi graphs . the fluctuations determine concentrations of links in some subsets of the graph , which then appear like communities . according to the definition of modularity , a graph has community structure with respect to a random graph with equal size and expected degree sequence . therefore , the modularity maximum of a graph reveals a significant community structure only if it is appreciably larger than the modularity maximum of random graphs of the same size and expected degree sequence . the significance of the modularity maximum @xmath403 for a graph can be estimated by calculating the maximum modularity for many realizations of the null model , obtained from the original graph by randomly rewiring its edges . one then computes the average @xmath404 and the standard deviation @xmath405 of the results . the statistical significance of @xmath403 is indicated by the distance of @xmath403 from the null model average @xmath404 in units of the standard deviation @xmath405 , i. e. by the @xmath406-score @xmath407 if @xmath408 , @xmath403 indicates strong community structure . cutoff values of @xmath409 for the @xmath406-scores are customary . this approach has problems , though . it can generate both false positives and false negatives : a few graphs that most people would consider without a significant community structure have a large @xmath406-score ; on the other hand , some graphs that are agreed to display cluster structure have very low values for the @xmath406-score . besides , the distribution of the maximum modularity values of the null model , though peaked , is not gaussian . therefore , one can not attribute to the values of the @xmath406-score the significance corresponding to a gaussian distribution , and one would need instead to compute the statistical significance for the right distribution . reichardt and bornholdt have studied the issue of the modularity values for random graphs in some depth @xcite , using their general spin glass formulation of the clustering problem ( section [ sub_sec6_01 ] ) . they considered the general case of a random graph with arbitrary degree distribution @xmath410 and without degree - degree correlations . they set @xmath368 , so that the energy of the ground state coincides with modularity ( up to a constant factor ) . for modularity s null model graphs , the modularity maximum corresponds to an equipartition of the graph , i. e. the magnetization of the ground state of the spin glass is zero , a result confirmed by numerical simulations @xcite . this is because the distribution of the couplings has zero mean , and the mean is only coupled to magnetization @xcite . for a partition of any graph with @xmath15 vertices and @xmath14 edges in @xmath91 clusters with equal numbers of vertices , there is a simple linear relation between the cut size @xmath411 of the partition and its modularity @xmath412 : @xmath413 $ ] . we remind that the cut size @xmath411 is the total number of inter - cluster edges of the partition ( section [ sec4_1 ] ) . in this way , the partition with maximum modularity is also the one with minimum cut size , and community detection becomes equivalent to graph partitioning . reichardt and bornholdt derived analytically the ground state energy for ising spins ( @xmath414 ) , which corresponds to the following expression of the expected maximum modularity @xmath415 for a bipartition @xcite @xmath416 here @xmath417 and @xmath418 is the ground state energy of the sherrington - kirkpatrick model @xcite . the most interesting feature of eq . [ eqr19 ] is the simple scaling with @xmath419 . numerical calculations show that this scaling holds for both erds - rnyi and scale - free graphs ( section [ sec1_3 ] ) . interestingly , the result is valid for partitions in @xmath91 clusters , where @xmath91 is left free , not only for @xmath414 . the number of clusters of the partition with maximum modularity decreases if the average degree @xmath420 increases , and tends to @xmath421 for large values of @xmath420 , regardless of the degree distribution and the size of the graph . from eq . [ eqr19 ] we also see that the expected maximum modularity for a random graph increases when @xmath420 decreases , i. e. if the graph gets sparser . so it is particularly hard to detect communities in sparse graphs by using modularity optimization . as we shall see in section [ sec6_4 ] , the sparsity of a graph is generally a serious obstacle for graph clustering methods , no matter if one uses modularity or not . a more fundamental issue , raised by fortunato and barthlemy @xcite , concerns the capability of modularity to detect `` good '' partitions . if a graph has a clear cluster structure , one expects that the maximum modularity of the graph reveals it . the null model of modularity assumes that any vertex @xmath67 `` sees '' any other vertex @xmath68 , and the expected number of edges between them is @xmath369 . similarly , the expected number of edges between two clusters @xmath258 and @xmath259 with total degrees @xmath422 and @xmath423 , respectively , is @xmath424 . the variation of modularity determined by the merger of @xmath258 and @xmath259 with respect to the partition in which they are separate clusters is @xmath425 , with @xmath426 number of edges connecting @xmath258 to @xmath259 . if @xmath427 , i. e. there is a single edge joining @xmath258 to @xmath259 , we expect that the two subgraphs will often be kept separated . instead , if @xmath428 , @xmath429 . let us suppose for simplicity that @xmath430 , i. e. that the two subgraphs are of about the same size , measured in terms of edges . we conclude that , if @xmath431 and the two subgraphs @xmath258 and @xmath259 are connected , modularity is greater if they are in the same cluster @xcite . the reason is intuitive : if there are more edges than expected between @xmath258 and @xmath259 , there is a strong topological correlation between the subgraphs . if the subgraphs are sufficiently small ( in degree ) , the expected number of edges for the null model can be smaller than one , so even the weakest possible connection ( a single edge ) suffices to keep the subgraphs together . interestingly , this result holds independently of the structure of the subgraphs . in particular it remains true if the subgraphs are cliques , which are the subgraphs with the largest possible density of internal edges , and represent the strongest possible communities . resolution limit of modularity optimization . the natural community structure of the graph , represented by the individual cliques ( circles ) , is not recognized by optimizing modularity , if the cliques are smaller than a scale depending on the size of the graph . in this case , the maximum modularity corresponds to a partition whose clusters include two or more cliques ( like the groups indicated by the dashed contours ) . reprinted figure with permission from ref . 2007 from the national academy of science of the usa . ] in fig . [ figure7 ] a graph is made out of @xmath115 identical cliques , with @xmath76 vertices each , connected by single edges . it is intuitive to think that the clusters of the best partition are the individual cliques : instead , if @xmath115 is larger than about @xmath432 , modularity would be higher for partitions in which the clusters are groups of cliques ( like the clique pairs indicated by the dashed lines in the figure ) . the conclusion is striking : modularity optimization has a resolution limit that may prevent it from detecting clusters which are comparatively small with respect to the graph as a whole , even when they are well defined communities like cliques . so , if the partition with maximum modularity includes clusters with total degree of the order of @xmath433 or smaller , one can not know _ a priori _ whether the clusters are single communities or combinations of smaller weakly interconnected communities . this resolution problem has a large impact in practical applications . real graphs with community structure usually contain communities which are very diverse in size @xcite , so many ( small ) communities may remain undetected . besides , modularity is extremely sensitive to even individual connections . many real graphs , in biology and in the social sciences , are reconstructed through experiments and surveys , so edges may occasionally be false positives : if two small subgraphs happen to be connected by a few false edges , modularity will put them in the same cluster , inferring a relationship between entities that in reality may have nothing to do with each other . the resolution limit comes from the very definition of modularity , in particular from its null model . the weak point of the null model is the implicit assumption that each vertex can interact with every other vertex , which implies that each part of the graph knows about everything else . this is however questionable , and certainly wrong for large systems like , e.g. , the web graph . it is certainly more reasonable to assume that each vertex has a limited horizon within the graph , and interacts just with a portion of it . however , nobody knows yet how to define such local territories for the graph vertices . the null model of the localized modularity of muff et al . ( section [ sub_sec6_01 ] ) is a possibility , since it limits the horizon of a vertex to a local neighborhood , comprising the cluster of the vertex and the clusters linked to it by at least one edge ( neighboring clusters ) . however , there are many other possible choices . in this respect , the null model of girvan and newman , though unrealistic , is the simplest one can think of , which partly explains its success . quality functions that , like modularity , are based on a null model such that the horizon of vertices is of the order of the size of the whole graph , are likely to be affected by a resolution limit @xcite . the problem is more general , though . for instance , li et al . @xcite have introduced a quality function , called _ modularity density _ , which consists in the sum over the clusters of the ratio between the difference of the internal and external degrees of the cluster and the cluster size . the modularity density does not require a null model , and delivers better results than modularity optimization ( e. g. it correctly recovers the natural partition of the graph in fig . [ figure7 ] for any number / size of the cliques ) . however , it is still affected by a resolution limit . to avoid that , li et al . proposed a more general definition of their measure , including a tunable parameter that allows to explore the graph at different resolutions , in the spirit of the methods of section [ sub_sec6_20 ] . a way to go around the resolution limit problem could be to perform further subdivisions of the clusters obtained from modularity optimization , in order to eliminate possible artificial mergers of communities . for instance , one could recursively optimize modularity for each single cluster , taking the cluster as a separate entity @xcite . however , this is not a reliable procedure , for two reasons : 1 ) the local modularities used to find partitions within the clusters have different null models , as they depend on the cluster sizes , so they are inconsistent with each other ; 2 ) one needs to define a criterion to decide when one has to stop partitioning a cluster , but there is no obvious prescription , so any choice is necessarily based on arbitrary assumptions - score is sufficiently high , the subgraph is supposed to have community structure and one accepts the partition in smaller pieces . the procedure stops when none of the subgraphs of the running partitions has significant community structure , based on modularity . ] . resolution limits arise as well in the more general formulation of community detection by reichardt and bornholt @xcite . here the limit scale for the undetectable clusters is @xmath434 . we remind that @xmath371 weighs the contribution of the null model term in the quality function . for @xmath368 one recovers the resolution limit of modularity . by tuning the parameter @xmath371 it is possible to arbitrarily vary the resolution scale of the corresponding quality function this in principle solves the problem of the resolution limit , as one could adjust the resolution of the method to the actual scale of the communities to detect . the problem is that usually one has no information about the community sizes , so it is not possible to decide _ a priori _ the proper value(s ) of @xmath371 for a specific graph . in the most recent literature on graph clustering quite a few _ multiresolution methods _ have been introduced , addressing this problem in several ways . we will discuss them in some detail in section [ sub_sec6_20 ] . the resolution limit can be easily extended to the case of weighted graphs . in a recent paper @xcite , berry et al . have considered the special case in which intracluster edges have weight @xmath2 , whereas intercluster edges have weight @xmath435 . by repeating the same procedure as in ref . @xcite , they conclude that clusters with internal strength ( i. e. sum of all weights of internal edges ) @xmath436 may remain undetected if @xmath437 , where @xmath86 is the total strength of the graph . so , the resolution limit decreases when @xmath435 decreases . berry et al . use this result to show that , by properly weighting the edges of a given unweighted graph , it becomes possible to detect clusters with very high resolution by still using modularity optimization . low - dimensional visualization of the modularity landscape for the metabolic network of the spirochete _ treponema pallidum_. the big degeneracy of suboptimal high - modularity partitions is revealed by the plateau ( whose shape is detailed in the inset ) , which is large and very irregular . modularity values in the plateau are very close to the absolute maximum , although they may correspond to quite different partitions . reprinted figure with permission from ref . @xcite . ] very recently , good et al . @xcite have made a careful analysis of modularity and its performance . they discovered that the modularity landscape is characterized by an exponential number of distinct states / partitions , whose modularity values are very close to the global maximum ( fig . [ figterr ] ) . this problem is particularly dramatic if a graph has a hierarchical community structure , like most real networks . such enormous number of solutions explains why many heuristics are able to come very close to modularity s global maximum , but it also implies that the global maximum is basically impossible to find . in addition , high - modularity partitions are not necessarily similar to each other , despite the proximity of their modularity scores . the optimal partition from a topological point of view , which usually does not correspond to the modularity maximum due to the resolution limit , may however have a large modularity score . therefore the optimal partition is basically indistinguishable from a huge number of high - modularity partitions , which are in general structurally dissimilar from it . the large structural inhomogeneity of the high - modularity partitions implies that one can not rely on any of them , at least in principle , in the absence of additional information on the particular system at hand and its structure . in sections [ sec4_1 ] and [ sec4_4 ] we have learned that spectral properties of graph matrices are frequently used to find partitions . a paradigmatic example is spectral graph clustering , which makes use of the eigenvectors of laplacian matrices ( section [ sec4_4 ] ) . we have also seen that newman - girvan modularity can be optimized by using the eigenvectors of the modularity matrix ( section [ sub_sec6_0_4 ] ) . most spectral methods have been introduced and developed in computer science and generally focus on data clustering , although applications to graphs are often possible as well . in this section we shall review recent spectral techniques proposed mostly by physicists explicitly for graph clustering . early works have shown that the eigenvectors of the _ transfer matrix _ @xmath438 ( section [ sec1_2 ] ) can be used to extract useful information on community structure . the transfer matrix acts as a time propagator for the process of random walk on a graph . given the eigenvector @xmath439 of the transposed transfer matrix @xmath440 , corresponding to the eigenvalue @xmath441 , @xmath442 is the outgoing current flowing from vertex @xmath67 , corresponding to the eigenmode @xmath18 . the _ participation ratio _ ( pr ) @xmath443^{-1 } \label{eqm19}\ ] ] indicates the effective number of vertices contributing to eigenvector @xmath439 . if @xmath444 receives contributions only from vertices of the same cluster , i. e. eigenvector @xmath439 is `` localized '' , the value of @xmath444 indicates the size of that cluster @xcite . the significance of the cluster can be assessed by comparing @xmath444 with the corresponding participation ratio for a random graph with the same expected degree sequence as the original graph . eigenvectors of the adjacency matrix may be localized as well if the graph has a clear community structure @xcite . a recent comprehensive analysis of spectral properties of modular graphs has been carried out by mitrovi and tadi @xcite . donetti and muoz have devised an elegant method based on the eigenvectors of the laplacian matrix @xcite . the idea is the same as in spectral graph clustering ( section [ sec4_4 ] ) : since the values of the eigenvector components are close for vertices in the same community , one can use them as coordinates , such that vertices turn into points in a metric space . so , if one uses @xmath445 eigenvectors , one can embed the vertices in an @xmath445-dimensional space . communities appear as groups of points well separated from each other , as illustrated in fig . [ figure8 ] . the separation is the more visible , the larger the number of dimensions / eigenvectors @xmath445 . spectral algorithm by donetti and muoz . vertex @xmath67 is represented by the values of the @xmath67th components of laplacian eigenvectors . in this example , the graph has an ad - hoc division in four communities , indicated by the colours . the communities are better separated in two dimensions ( b ) than in one ( a ) . reprinted figure with permission from ref . 2004 by iop publishing and sissa . ] the originality of the method consists in the procedure to group the points and to extract the partition . donetti and muoz used hierarchical clustering ( see section [ sec4_2 ] ) , with the constraint that only pairs of clusters which have at least one interconnecting edge in the original graph are merged . among all partitions of the resulting dendrogram , the one with largest modularity is chosen . for the similarity measure between vertices , donetti and muoz used both the euclidean distance and the angle distance . the angle distance between two points is the angle between the vectors going from the origin of the @xmath445-dimensional space to either point . tests on the benchmark by girvan and newman ( section [ sec6_1 ] ) show that the best results are obtained with complete - linkage clustering . the most computationally expensive part of the algorithm is the calculation of the laplacian eigenvectors . since a few eigenvectors suffice to get good partitions , one can determine them with the lanczos method @xcite . the number @xmath445 of eigenvectors that are needed to have a clean separation of the clusters is not known _ a priori _ , but one can compute a number @xmath446 of them and search for the highest modularity partition among those delivered by the method for all @xmath447 . in a related work , simonsen has embedded graph vertices in space by using as coordinates the components of the eigenvectors of the right stochastic matrix @xcite . eigenvalues and eigenvectors of the laplacian matrix have been used by alves to compute the effective conductances for pairs of vertices in a graph , assuming that the latter is an electric network with edges of unit resistance @xcite . the conductances enable one to compute the transition probabilities for a random walker moving on the graph , and from the transition probabilities one builds a similarity matrix between vertex pairs . hierarchical clustering is applied to join vertices in groups . the method can be trivially extended to the case of weighted graphs . the algorithm by alves is rather slow , as one needs to compute the whole spectrum of the laplacian , which requires a time @xmath154 . moreover , there is no criterion to select which partition(s ) of the dendrogram is ( are ) the best . capocci et al . @xcite used eigenvector components of the right stochastic matrix @xmath448 ( section [ sec1_2 ] ) , that is derived from the adjacency matrix by dividing each row by the sum of its elements . the right stochastic matrix has similar properties as the laplacian . if the graph has @xmath126 connected components , the largest @xmath126 eigenvalues are equal to @xmath2 , with eigenvectors characterized by having equal - valued components for vertices belonging to the same component . in this way , by listing the vertices according to the connected components they belong to , the components of any eigenvector of @xmath448 , corresponding to eigenvalue @xmath2 , display a step - wise profile , with plateaus indicating vertices in the same connected component . for connected graphs with cluster structure , one can still see plateaus , if communities are only loosely connected to each other ( fig . [ capoccifig ] ) . basic principle of the spectral algorithm by capocci et al . the bottom diagram shows the values of the components of the second eigenvector of the right stochastic matrix for the graph drawn on the top . the three plateaus of the eigenvector components correspond to the three evident communities of the graph . reprinted figures with permission from ref . 2005 by elsevier.,title="fig : " ] 0.4 cm basic principle of the spectral algorithm by capocci et al . the bottom diagram shows the values of the components of the second eigenvector of the right stochastic matrix for the graph drawn on the top . the three plateaus of the eigenvector components correspond to the three evident communities of the graph . reprinted figures with permission from ref . 2005 by elsevier.,title="fig : " ] here the communities can be immediately deducted by an inspection of the components of any eigenvector with eigenvalue @xmath2 . in practical cases , plateaus are not clearly visible , and one eigenvector is not enough . however , one expects that there should be a strong correlation between eigenvector components corresponding to vertices in the same cluster . capocci et al . derived a similarity matrix , where the similarity between vertices @xmath67 and @xmath68 is the pearson correlation coefficient between their corresponding eigenvector components , averaged over a small set of eigenvectors . the eigenvectors can be calculated by performing a constrained optimization of a suitable cost function . the method can be extended to weighted and directed graphs . it is useful to estimate vertex similarities , however it does not provide a well - defined partition of the graph . yang and liu @xcite adopted a recursive bisectioning procedure . communities are subgraphs such that the external degree of each vertex does not exceed the internal degree ( _ strong communities _ or _ @xmath49-sets _ , see section [ sec3_1_2 ] ) . in the first step of the algorithm , the adjacency matrix of the graph is put in approximately block - diagonal form . this is done by computing a new centrality measure for the vertices , called _ clustering centrality_. this measure is similar to bonacich s eigenvector centrality @xcite , which is given by the eigenvector of the adjacency matrix corresponding to the largest eigenvalue . the clustering centrality of a vertex basically measures the probability that a random walker starting at that vertex hits a given target . such probability is larger if the origin and the target vertices belong to the same cluster than if they do not . if the graph has well - separated communities , the values of the clustering centrality would be similar for vertices in the same cluster . in this way , one can rearrange the original adjacency matrix by listing the vertices in non - decreasing order of their clustering centralities , and blocks would be visible . the blocks are then identified by iterative bisection : each cluster found at some step is split in two as long as the resulting parts are still communities in the strong sense , otherwise the procedure stops . the worst - case complexity of the method is @xmath449 $ ] , where @xmath289 is the number of clusters of the final partition and @xmath156 the ( average ) number of iterations required to compute the clustering centrality with the power method @xcite . since @xmath156 is fairly independent of the graph size , the method scales quite well on sparse graphs [ @xmath254 ] . the main limit of this technique is the assumption that communities are defined in the strong sense , which is too restrictive . on the other hand , one could think of using alternative definitions . this section describes methods employing processes running on the graph , focusing on spin - spin interactions , random walks and synchronization . the potts model is among the most popular models in statistical mechanics @xcite . it describes a system of spins that can be in @xmath91 different states . the interaction is ferromagnetic , i. e. it favours spin alignment , so at zero temperature all spins are in the same state . if antiferromagnetic interactions are also present , the ground state of the system may not be the one where all spins are aligned , but a state where different spin values coexist , in homogeneous clusters . if potts spin variables are assigned to the vertices of a graph with community structure , and the interactions are between neighbouring spins , it is likely that the structural clusters could be recovered from like - valued spin clusters of the system , as there are many more interactions inside communities than outside . based on this idea , inspired by an earlier paper by blatt et al . @xcite , reichardt and bornholdt proposed a method to detect communities that maps the graph onto a zero - temperature @xmath91-potts model with nearest - neighbour interactions @xcite . the hamiltonian of the model , i. e. its energy , reads @xmath451 where @xmath360 is the element of the adjacency matrix , @xmath101 is kronecker s function , @xmath452 the number of spins in state @xmath155 , @xmath359 and @xmath371 are coupling parameters . the energy @xmath453 is the sum of two competing terms : the first is the classical ferromagnetic potts model energy , and favors spin alignment ; the second term instead peaks when the spins are homogeneously distributed . the ratio @xmath454 expresses the relative importance of the two terms : by tuning @xmath454 one can explore different levels of modularity of the system , from the whole graph seen as a single cluster to clusters consisting of individual vertices . if @xmath454 is set to the value @xmath40 of the average density of edges of the graph @xmath313 , the energy of the system is smaller if spins align within subgraphs such that their internal edge density exceeds @xmath40 , whereas the external edge density is smaller than @xmath40 , i. e. if the subgraphs are clusters ( section [ sec3_1_1 ] ) . the minimization of @xmath453 is carried out via simulated annealing ( @xcite and section [ sub_sec6_0_2 ] ) , starting from a configuration where spins are randomly assigned to the vertices and the number of states @xmath91 is very high . the procedure is quite fast and the results do not depend on @xmath91 ( provided @xmath91 is sufficiently high ) . the method also allows to identify vertices shared between communities , from the comparison of partitions corresponding to global and local energy minima . the hamiltonian @xmath453 can be rewritten as @xmath455 which is the energy of an infinite - range potts spin glass , as all pairs of spins are interacting ( neighboring or not ) and there may be both positive and negative couplings . the method can be simply extended to the analysis of weighted graphs , by introducing spin couplings proportional to the edge weights , which amounts to replacing the adjacency matrix @xmath189 with the weight matrix @xmath190 in eq . [ eqspin1 ] . ispolatov et al . @xcite have adopted a similar hamiltonian as in eq . [ eqspin1 ] , with a tunable antiferromagnetic term interpolating between the corresponding term of eq . [ eqspin1 ] and the entropy term ( proportional to @xmath456 ) of the free energy , whose minimization is equivalent to finding the states of the finite - temperature potts model used by blatt et al . @xcite . [ eqspin2 ] is at the basis of the successive generalization of modularity with arbitrary null models proposed by reichardt and bornholdt , that we have discussed in section [ sub_sec6_01 ] . in another work @xcite , son et al . have presented a clustering technique based on the ferromagnetic random field ising model ( frfim ) . given a weighted graph with weight matrix @xmath190 , the hamiltonian of the frfim on the graph is @xmath457 in eq . [ eqspin3 ] @xmath458 and @xmath459 are the spin and the random magnetic field of vertex @xmath67 , respectively . the frfim has been studied to understand the nature of the spin glass phase transition @xcite and the disorder - driven roughening transition of interfaces in disordered media @xcite . the behavior of the model depends on the choice of the magnetic fields . son et al . set to zero the magnetic fields of all vertices but two , say @xmath155 and @xmath156 , for which the field has infinite strength and opposite signs . this amounts to fix the spins of @xmath155 and @xmath156 to opposite values , introducing frustration in the system . the idea is that , if @xmath155 and @xmath156 are central vertices of different communities , they impose their spin state to the other community members . so , the state of minimum energy is a configuration in which the graph is polarized into a subgraph with all positive spins and a subgraph with all negative spins , coinciding with the communities , if they are well defined . finding the minimum of @xmath453 is equivalent to solving a maximum - flow / minimum - cut problem , which can be done through well known techniques of combinatorial optimization , like the augmenting path algorithm @xcite . for a given choice of @xmath155 and @xmath156 , many ground states can be found . the vertices that end up in the same cluster in all ground states represent the cores of the clusters , which are called _ possible vertices not belonging to the coteries indicate that the two clusters overlap . in the absence of information about the cluster structure of the graph , one needs to repeat the procedure for any pair of vertices @xmath155 and @xmath156 . picking vertices of the same cluster , for instance , would not give meaningful partitions . son et al . distinguish relevant clusters if they are of about the same size . the procedure can be iteratively applied to each of the detected clusters , considered as a separate graph , until all clusters have no community structure any more . on sparse graphs , the algorithm has complexity @xmath460 , where @xmath461 , so it is very slow and can be currently used for graphs of up to few thousands vertices . if one happens to know which are the important vertices of the clusters , e.g. by computing appropriate centrality values ( like degree or site betweenness @xcite ) , the choices for @xmath155 and @xmath156 are constrained and the complexity can become as low as @xmath462 , which enables the analysis of systems with millions of vertices . tests on barabsi - albert graphs ( section [ sec1_3 ] ) show that the latter have no community structure , as expected . random walks @xcite can also be useful to find communities . if a graph has a strong community structure , a random walker spends a long time inside a community due to the high density of internal edges and consequent number of paths that could be followed . here we describe the most popular clustering algorithms based on random walks . all of them can be trivially extended to the case of weighted graphs . zhou used random walks to define a distance between pairs of vertices @xcite : the distance @xmath463 between @xmath67 and @xmath68 is the average number of edges that a random walker has to cross to reach @xmath68 starting from @xmath67 . close vertices are likely to belong to the same community . zhou defines a `` global attractor '' of a vertex @xmath67 to be a closest vertex to @xmath67 ( i. e. any vertex lying at the smallest distance from @xmath67 ) , whereas the `` local attractor '' of @xmath67 is its closest neighbour . two types of communities are defined , according to local or global attractors : a vertex @xmath67 has to be put in the same community of its attractor and of all other vertices for which @xmath67 is an attractor . communities must be minimal subgraphs , i. e. they can not include smaller subgraphs which are communities according to the chosen criterion . applications to real networks , like zachary s karate club @xcite and the college football network compiled by girvan and newman @xcite ( section [ sec6_1 ] ) , along with artificial graphs like the benchmark by girvan and newman @xcite ( section [ sec6_1 ] ) , show that the method can find meaningful partitions . the method can be refined , in that vertex @xmath67 is associated to its attractor @xmath68 only with a probability proportional to @xmath464 , @xmath19 being a sort of inverse temperature . the computation of the distance matrix requires solving @xmath15 linear - algebra equations ( as many as the vertices ) , which requires a time @xmath154 . on the other hand , an exact computation of the distance matrix is not necessary , as the attractors of a vertex can be identified by considering only a localized portion of the graph around the vertex ; therefore the method can be applied to large graphs as well . in a successive paper @xcite , zhou introduced a measure of dissimilarity between vertices based on the distance defined above . the measure resembles the definition of distance based on structural equivalence of eq . [ eq7 ] , where the elements of the adjacency matrix are replaced by the corresponding distances . graph partitions are obtained with a divisive procedure that , starting from the graph as a single community , performs successive splits based on the criterion that vertices in the same cluster must be less dissimilar than a running threshold , which is decreased during the process . the hierarchy of partitions derived by the method is representative of actual community structures for several real and artificial graphs , including zachary s karate club @xcite , the college football network @xcite and the benchmark by girvan and newman @xcite ( section [ sec6_1 ] ) . the time complexity of the procedure is again @xmath154 . the code of the algorithm can be downloaded from http://www.mpikg-golm.mpg.de/theory/people/zhou + /networkcommunity.html . in another work @xcite , zhou and lipowsky adopted biased random walkers , where the bias is due to the fact that walkers move preferentially towards vertices sharing a large number of neighbours with the starting vertex . they defined a proximity index , which indicates how close a pair of vertices is to all other vertices . communities are detected with a procedure called _ netwalk _ , which is an agglomerative hierarchical clustering method ( section [ sec4_2 ] ) , where the similarity between vertices is expressed by their proximity . the method has a time complexity @xmath154 : however , the proximity index of a pair of vertices can be computed with good approximation by considering just a small portion of the graph around the two vertices , with a considerable gain in time . the performance of the method is comparable with that of the algorithm of girvan and newman ( section [ subsec5_1 ] ) . a different distance measure between vertices based on random walks was introduced by latapy and pons @xcite . the distance is calculated from the probabilities that the random walker moves from a vertex to another in a fixed number of steps . the number of steps has to be large enough to explore a significant portion of the graph , but not too long , as otherwise one would approach the stationary limit in which transition probabilities trivially depend on the vertex degrees . vertices are then grouped into communities through an agglomerative hierarchical clustering technique based on ward s method @xcite . modularity ( section [ sec3_2_2 ] ) is used to select the best partition of the resulting dendrogram . the algorithm runs to completion in a time @xmath465 on a sparse graph , where @xmath88 is the depth of the dendrogram . since @xmath88 is often small for real graphs [ @xmath255 ] , the expected complexity in practical computations is @xmath128 . the software of the algorithm can be found at http://www-rp.lip6.fr/@xmath215latapy/pp/walktrap.html . hu et al . @xcite designed a graph clustering technique based on a signaling process between vertices , somewhat resembling diffusion . initially a vertex @xmath155 is assigned one unit of signal , all the others have no signal . in the first step , the source vertex @xmath155 sends one unit of signal to each of its neighbors . next , all vertices send as many units of signals they have to each of their neighbors . the process is continued until a given number of iterations @xmath467 is reached . the intensity of the signal at vertex @xmath67 , normalized by the total amount of signal , is the @xmath67-th entry of a vector @xmath468 , representing the source vertex @xmath155 . the procedure is then repeated by choosing each vertex as source . in this way one can associate an @xmath15-dimensional vector to each vertex , which correspons to a point in an euclidean space . the vector @xmath468 is actually the @xmath155-th column of the matrix @xmath469 , where @xmath470 and @xmath189 are the identity and adjacency matrix , respectively . the idea is that the vector @xmath468 describes the influence that vertex @xmath155 exerts on the graph through signaling . vertices of the same community are expected to have similar influence on the graph and thus to correspond to vectors which are `` close '' in space . the vectors are finally grouped via fuzzy @xmath44-means clustering ( section [ sec4_3 ] ) . the optimal number of clusters corresponds to the partition with the shortest average distance between vectors in the same community and the largest average distance between vectors of different communities . the signaling process is similar to diffusion , but with the important difference that here there is no flow conservation , as the amount of signal at each vertex is not distributed among its neighbors but transferred entirely to each neighbor ( as if the vertex sent multiple copies of the same signal ) . the complexity of the algorithm is @xmath471 $ ] , where @xmath420 is the average degree of the graph . like in the previous algorithm by latapy and pons @xcite , finding an optimal value for the number of iterations @xmath467 is non - trivial . delvenne et al . @xcite have shown that random walks enable one to introduce a general quality function , expressing the persistence of clusters in time . a cluster is persistent with respect to a random walk after @xmath156 time steps if the probability that the walker escapes the cluster before @xmath156 steps is low . such probability is computed via the _ clustered autocovariance matrix _ @xmath472 , which , for a partition of the graph in @xmath117 clusters , is defined as @xmath473 here , @xmath474 is the @xmath475 membership matrix , whose element @xmath476 equals one if vertex @xmath67 is in cluster @xmath68 , zero otherwise ; @xmath477 is the transition matrix of the random walk ; @xmath478 the diagonal matrix whose elements are the stationary probabilities of the random walk , i. e. @xmath479 , @xmath105 being the degree of vertex @xmath67 ; @xmath480 is the vector whose entries are the diagonal elements of @xmath478 . the element @xmath481 expresses the probability for the walk to start in cluster @xmath67 and end up in cluster @xmath68 after @xmath156 steps , minus the stationary probability that two independent random walkers are in @xmath67 and @xmath68 . in this way , the persistence of a cluster @xmath67 is related to the diagonal element @xmath482 . delvenne et al . defined the _ stability of the clustering _ @xmath483 . \label{eqbarahona1}\ ] ] the aim is then , for a given time @xmath156 , finding the partition with the largest value for @xmath484 . for @xmath485 , the most stable partition is that in which all vertices are their own clusters . interestingly , for @xmath486 , maximizing stability is equivalent to maximizing newman - girvan modularity ( section [ sec3_2_2 ] ) . the cut size of the partition ( section [ sec4_1 ] ) equals @xmath487 $ ] , so it is also a one - step measure . in the limit @xmath488 , the most stable partition coincides with the fiedler partition @xcite , i. e. the bipartition where vertices are put in the same class according to the signs of the corresponding component of the fiedler eigenvector ( section [ sec4_1 ] ) . therefore , the measure @xmath484 is very general , and gives a unifying interpretation in the framework of the random walk of several measures that were defined in different contexts . in particular , modularity has a natural interpretation in this dynamic picture @xcite . since the size of stable clusters increases with @xmath156 , time can be considered as a resolution parameter . resolution can be fine tuned by taking time as a continuous variable ( the extension of the formalism is straightforward ) ; the linearization of the stability measure at small ( continuous ) times delivers multiresolution versions of modularity @xcite ( section [ sub_sec6_201 ] ) . in a method by weinan et al . @xcite , the best partition of a graph in @xmath44 clusters is such that the markov chain describing a random walk on the meta - graph , whose vertices are the clusters of the original graph , gives the best approximation of the full random walk dynamics on the whole graph . the quality of the approximation is given by the distance between the left stochastic matrices of the two processes , which thus needs to be minimized . the minimization is performed by using a variant of the @xmath44-means algorithm ( section [ sec4_3 ] ) , and the result is the best obtained out of @xmath76 runs starting from different initial conditions , a strategy that considerably improves the quality of the optimum . the time complexity is @xmath489 $ ] , where @xmath156 is the number of steps required to reach convergence . the optimal number of clusters could in principle be determined by analyzing how the quality of the approximation varies with @xmath44 , but the authors do not give any general recipe . the method is rather accurate on the benchmark by girvan and newman @xcite ( section [ sec6_1 ] ) and on zachary s karate club network . the algorithm by weinan et al . is asymptotically equivalent to spectral graph partitioning ( section [ sec4_4 ] ) when the markov chain describing the random walk presents a sizeable spectral gap between some of the largest eigenvalues of the transfer matrix ( section [ sec1_2 ] ) , approximately equal to one , and the others . we conclude this section by describing the _ markov cluster algorithm ( mcl ) _ , which was invented by van dongen @xcite . this method simulates a peculiar process of flow diffusion in a graph . one starts from the _ transfer matrix _ of the graph @xmath201 ( section [ sec1_2 ] ) . the element @xmath490 of the transfer matrix gives the probability that a random walker , sitting at vertex @xmath68 , moves to @xmath67 . the sum of the elements of each column of @xmath438 is one . each iteration of the algorithm consists of two steps . in the first step , called _ expansion _ , the transfer matrix of the graph is raised to an integer power @xmath45 ( usually @xmath491 ) . the entry @xmath492 of the resulting matrix gives the probability that a random walker , starting from vertex @xmath68 , reaches @xmath67 in @xmath45 steps ( diffusion flow ) . the second step , which has no physical counterpart , consists in raising each single entry of the matrix @xmath477 to some power @xmath18 , where @xmath18 is now real - valued . this operation , called _ inflation _ , enhances the weights between pairs of vertices with large values of the diffusion flow , which are likely to be in the same community . next , the elements of each column must be divided by their sum , such that the sum of the elements of the column equals one and a new transfer matrix is recovered . after some iterations , the process delivers a stable matrix , with some remarkable properties . its elements are either zero or one , so it is a sort of adjacency matrix . most importantly , the graph described by the matrix is disconnected , and its connected components are the communities of the original graph . the method is really simple to implement , which is the main reason of its success : as of now , the mcl is one of the most used clustering algorithms in bioinformatics . the code can be downloaded from http://www.micans.org / mcl/. due to the matrix multiplication of the expansion step , the algorithm should scale as @xmath154 , even if the graph is sparse , as the running matrix becomes quickly dense after a few steps of the algorithm . however , while computing the matrix multiplication , mcl keeps only a maximum number @xmath44 of non - zero elements per column , where @xmath44 is usually much smaller than @xmath15 . so , the actual worst - case running time of the algorithm is @xmath493 on a sparse graph . a problem of the method is the fact that the final partition is sensitive to the parameter @xmath18 used in the inflation step . therefore several different partitions can be obtained , and it is not clear which are the most meaningful or representative . synchronization @xcite is an emergent phenomenon occurring in systems of interacting units and is ubiquitous in nature , society and technology . in a synchronized state , the units of the system are in the same or similar state(s ) at every time . synchronization has also been applied to find communities in graphs . if oscillators are placed at the vertices , with initial random phases , and have nearest - neighbour interactions , oscillators in the same community synchronize first , whereas a full synchronization requires a longer time . so , if one follows the time evolution of the process , states with synchronized clusters of vertices can be quite stable and long - lived , so they can be easily recognized . this was first shown by arenas , daz - guilera and prez - vicente @xcite . they used kuramoto oscillators @xcite , which are coupled two - dimensional vectors endowed with a proper frequency of oscillations . in the kuramoto model , the phase @xmath494 of oscillator @xmath67 evolves according to the following dynamics @xmath495 where @xmath496 is the natural frequency of @xmath67 , @xmath289 the strength of the coupling between oscillators and the sum runs over all oscillators ( mean field regime ) . if the interaction coupling exceeds a threshold , depending on the width of the distribution of natural frequencies , the dynamics leads to synchronization . if the dynamics runs on a graph , each oscillator is coupled only to its nearest neighbors . in order to reveal the effect of local synchronization , arenas et al . introduced the local order parameter @xmath497\rangle , \label{eqsync2}\ ] ] measuring the average correlation between oscillators @xmath67 and @xmath68 . the average is computed over different initial conditions . by visualizing the correlation matrix @xmath498 at a given time @xmath156 , one may distinguish groups of vertices that synchronize together . the groups can be identified by means of the _ dynamic connectivity matrix _ @xmath499 , which is a binary matrix obtained from @xmath498 by thresholding its entries . the dynamic connectivity matrix embodies information about both the synchronization dynamics and the underlying graph topology . from the spectrum of @xmath499 it is possible to derive the number of disconnected components at time @xmath156 . by plotting the number of components as a function of time , plateaus may appear at some characteristic time scales , indicating structural scales of the graph with robust communities ( fig . [ syncarenas ] ) . partitions corresponding to long plateaus are characterized by high values of the modularity of newman and girvan ( section [ sec3_2_2 ] ) on graphs with homogeneous degree distributions , whereas such correlation is poor in the presence of hubs @xcite . indeed , it has been proven that the stability ( eq . [ eqbarahona1 ] ) of the dynamics associated to the standard laplacian matrix , which describes the convergence towards synchronization of the kuramoto model with equal intrinsic frequencies , coincides with modularity only for graphs whose vertices have the same degree @xcite . the appearance of plateaus at different time scales hints to a hierarchical organization of the graph . after a sufficiently long @xmath156 all oscillators are synchronized and the whole system behaves as a single component . interestingly , arenas et al . found that the structural scales revealed by synchronization correspond to groups of eigenvalues of the laplacian matrix of the graph , separated by gaps ( fig . [ syncarenas ] ) . based on the same principle , boccaletti et al . designed a community detection method based on synchronization @xcite . the synchronization dynamics is a variation of kuramoto s model , the opinion changing rate ( ocr ) model @xcite . here the interaction coupling between adjacent vertices is weighted by a term proportional to a ( negative ) power of the betweenness of the edge connecting the vertices ( section [ subsec5_1 ] ) , with exponent @xmath18 . the evolution equations of the model are solved by decreasing the value of @xmath18 during the evolution of the dynamics , starting from a configuration in which the system is fully synchronized ( @xmath500 ) . the graph tends to get split into clusters of synchronized elements , because the interaction strengths across inter - cluster edges get suppressed due to their high betweenness scores . by varying @xmath18 , different partitions are recovered , from the graph as a whole until the vertices as separate communities : the partition with the largest value of modularity is taken as the most relevant . the algorithm scales in a time @xmath205 , or @xmath129 on sparse graphs , and gives good results in practical examples , including zachary s karate club @xcite and the benchmark by girvan and newman @xcite ( section [ sec6_1 ] ) . the method can be refined by homogeneizing the natural frequencies of the oscillators during the evolution of the system . in this way , the system becomes more stable and partitions with higher modularity values can be recovered . in a recent paper by li et al . @xcite , it was shown that synchronized clusters in modular networks are characterized by interfacial vertices , whose oscillation frequency is intermediate between those of two or more clusters , so that they do not belong to a specific community . li et al . used this result to devise a technique able to detect overlapping communities . synchronization - based algorithms may not be reliable when communities are very different in size ; tests in this direction are still missing . _ statistical inference _ @xcite aims at deducing properties of data sets , starting from a set of observations and model hypotheses . if the data set is a graph , the model , based on hypotheses on how vertices are connected to each other , has to _ fit _ the actual graph topology . in this section we review those clustering techniques attempting to find the best fit of a model to the graph , where the model assumes that vertices have some sort of classification , based on their connectivity patterns . we mainly focus on methods adopting _ bayesian inference _ @xcite , in which the best fit is obtained through the maximization of a likelihood ( _ generative models _ ) , but we also discuss related techniques , based on _ blockmodeling _ @xcite , _ model selection _ @xcite and _ information theory _ @xcite . bayesian inference uses observations to estimate the probability that a given hypothesis is true . it consists of two ingredients : the evidence , expressed by the information @xmath501 one has about the system ( e.g. , through measurements ) ; a statistical model with parameters @xmath502 . bayesian inference starts by writing the likelihood @xmath503 that the observed evidence is produced by the model for a given set of parameters @xmath502 . the aim is to determine the choice of @xmath502 that maximizes the posterior distribution @xmath504 of the parameters given the model and the evidence . by using bayes theorem one has @xmath505 where @xmath506 is the prior distribution of the model parameters and @xmath507 unfortunately , computing the integral [ eqstinf01 ] is a major challenge . moreover , the choice of the prior distribution @xmath506 is non - obvious . generative models differ from each other by the choice of the model and the way they address these two issues . bayesian inference is frequently used in the analysis and modeling of real graphs , including social @xcite and biological networks @xcite . graph clustering can be considered a specific example of inference problem . here , the evidence is represented by the graph structure ( adjacency or weight matrix ) and there is an additional ingredient , represented by the classification of the vertices in groups , which is a _ hidden _ ( or _ missing _ ) information that one wishes to infer along with the parameters of the model which is supposed to be responsible for the classification . this idea is at the basis of several recent papers , which we discuss here . in all these works , one essentially maximizes the likelihood @xmath503 that the model is consistent with the observed graph structure , with different constraints . we specify the set of parameters @xmath502 as the triplet @xmath508 , where @xmath509 indicates the community assignment of the vertices , @xmath510 the model parameters , and @xmath44 the number of clusters . in the following we shall stick to the notation of the papers , so the variables above may be indicated by different symbols . however , to better show what each method specifically does we shall refer to our general notation at the end of the section . hastings @xcite chooses as a model of network with communities the _ planted partition model _ ( section [ sec6 ] ) . in it , @xmath15 vertices are assigned to @xmath91 groups : vertices of the same group are linked with a probability @xmath511 , while vertices of different groups are linked with a probability @xmath512 . if @xmath513 , the model graph has a built - in community structure . the vertex classification is indicated by the set of labels @xmath514 . the probability that , given a graph , the classification @xmath514 is the right one according to the model is @xmath515\}^{-1 } , \label{eqstinf1}\ ] ] where @xmath516/[p_{out}(1-p_{in})]\}$ ] , @xmath517 $ ] and the first sum runs over nearest neighboring vertices . maximizing @xmath518 is equivalent to minimizing the argument of the exponential , which is the hamiltonian of a potts model with short- and long - range interactions . for @xmath513 , @xmath519 and @xmath520 , so the model is a spin glass with ferromagnetic nearest - neighbor interactions and antiferromagnetic long - range interactions , similar to the model proposed by reichardt and bornholdt to generalize newman - girvan modularity @xcite ( section [ sub_sec6_01 ] ) . hastings used belief propagation @xcite to find the ground state of the spin model . on sparse graphs , the complexity of the algorithm is expected to be @xmath521 , where @xmath18 needs to be estimated numerically . in principle one needs to input the parameters @xmath511 and @xmath512 , which are usually unknown in practical applications . however , it turns out that they can be chosen rather arbitrarily , and that bad choices can be recognized and corrected . newman and leicht @xcite have recently proposed a similar method based on a mixture model and the expectation - maximization technique @xcite . the method bears some resemblance with an _ a posteriori _ blockmodel previously introduced by snijders and nowicki @xcite . they start from a directed graph with @xmath15 vertices , whose vertices fall into @xmath117 classes . the group of vertex @xmath67 is indicated by @xmath522 , @xmath523 the fraction of vertices in group @xmath375 , and @xmath524 the probability that there is a directed edge from vertices of group @xmath375 to vertex @xmath67 . by definition , the sets @xmath525 and @xmath526 satisfy the normalization conditions @xmath527 and @xmath528 . apart from normalization , the probabilities @xmath526 are assumed to be independent of each other . the best classification of the vertices corresponds to the maximum of the average log - likelihood @xmath529 that the model , described by the values of the parameters @xmath525 and @xmath526 fits the adjacency matrix @xmath189 of the graph . the expression of the average log - likelihood @xmath529 requires the definition of the probability @xmath530 , that vertex @xmath67 belongs to group @xmath375 . by applying bayes theorem the probabilities @xmath531 can be computed in terms of the @xmath525 and the @xmath526 , as @xmath532 while the maximization of the average log - likelihood @xmath529 , under the normalization constraints of the model variables @xmath525 and @xmath526 , yields the relations @xmath533 where @xmath105 is the outdegree of vertex @xmath67 . equations [ eqstinf2 ] and [ eqstinf3 ] are self - consistent , and can be solved by iterating them to convergence , starting from a suitable set of initial conditions . convergence is fast , so the algorithm could be applied to fairly large graphs , with up to about @xmath243 vertices . the method , designed for directed graphs , can be easily extended to the undirected case , whereas an extension to weighted graphs is not straightforwad . a nice feature of the method is that it does not require any preliminary indication on what type of structure to look for ; the resulting structure is the most likely classification based on the connectivity patterns of the vertices . therefore , various types of structures can be detected , not necessarily communities . for instance , multipartite structure could be uncovered , or mixed patterns where multipartite subgraphs coexist with communities , etc .. in this respect , it is more powerful than most methods of community detection , which are bound to focus only on proper communities , i. e. subgraphs with more internal than external edges . in addition , since partitions are defined by assigning probability values to the vertices , expressing the extent of their membership in a group , it is possible that some vertices are not clearly assigned to a group , but to more groups , so the method is able to deal with overlapping communities . the main drawback of the algorithm is the fact that one needs to specify the number of groups @xmath117 at the beginning of the calculation , a number that is typically unknown for real networks . it is possible to derive this information self - consistently by maximizing the probability that the data are reproduced by partitions with a given number of clusters . but this procedure involves some degree of approximation , and the results are often not good . in a recent study it has been shown that the method by newman and leicht enables one to rank vertices based on their degree of influence on other vertices , which allows to identify the vertices responsible for the group structure and its stability @xcite . a very similar technique has also been applied by vzquez @xcite to the problem of population stratification , where animal populations and their attributes are represented as hypergraphs ( section [ sec1_1 ] ) . vzquez also suggested an interesting criterion to decide the optimal number of clusters , namely picking the number @xmath534 whose solution has the greatest similarity with solutions obtained at different values of @xmath117 . the similarity between two partitions can be estimated in various ways , for instance by computing the normalized mutual information ( section [ sec6 ] ) . in a successive paper @xcite , vzquez showed that better results are obtained if the classification likelihood is maximized by using variational bayes @xcite . ramasco and mungan @xcite remarked that the normalization condition on the probabilities @xmath526 implies that each group @xmath375 must have non - zero outdegree and that therefore the method fails to detect the intuitive group structure of ( directed ) bipartite graphs ( fig . [ figramasco ] ) . problem of method by newman and leicht . by applying the method to the illustrated complete bipartite graph ( colors indicate the vertex classes ) the natural group structure c ) is not recovered ; instead , the most likely classifications are a ) and b ) . reprinted figure with permission from ref . @xcite.2008 by the american physical society . ] to avoid this problem , they proposed a modification , that consists in introducing three sets for the edge probabilities @xmath526 , relative to edges going from group @xmath375 to vertex @xmath67 ( as before ) , from @xmath67 to @xmath375 and in both directions , respectively . furthermore , they used the average entropy of the classification @xmath535 , where the @xmath536 are the analogs of the probabilities in eq . [ eqstinf2 ] , to infer the optimal number of groups , that the method of newman and leicht is unable to provide . another technique similar to that by newman and leicht has been designed by ren et al . the model is based on the group fractions @xmath525 , defined as above , and a set of probabilities @xmath537 , expressing the relevance of vertex @xmath67 for group @xmath375 ; the basic assumption is that the probability that two vertices of the same group are connected by an edge is proportional to the product of the relevances of the two vertices . in this way , there is an explicit relation between group membership and edge density , and the method can only detect community structure . the community assignments are recovered through an expectation - maximization procedure that closely follows that by newman and leicht . maximum likelihood estimation has been used by opi et al . to define an axiomatization of the problem of graph clustering and its related concepts @xcite . the starting point is again the planted partition model ( section [ sec6 ] ) , with probabilities @xmath511 and @xmath512 . a novelty of the approach is the introduction of the _ size matrix _ @xmath538 , whose element @xmath539 indicates the maximum strength of interaction between vertices @xmath67 and @xmath68 . for instance , in a graph with unweighted connections , all elements of @xmath538 equal @xmath2 . in this case , the probability that the graph conceals a community structure coincides with the expression ( [ eqstinf1 ] ) by hastings . opi et al . used this probability as a quality function to define rankings between graph partitions ( _ likelihood rankings _ ) . the authors show that the likelihood rankings satisfy a number of general properties , which should be satisfied by any reasonable ranking . they also propose an algorithm to find the maximum likelihood partition , by using the auxiliary concept of _ pseudo - community structure _ , i. e. a grouping of the graph vertices in which it is specified which pairs of vertices stay in the same community and which pairs instead stay in different communities . a pseudo - community may not be a community because the transitive property is not generally valid , as the focus is on pairwise vertex relationships : it may happen that @xmath67 and @xmath68 are classified in the same group , and that @xmath68 and @xmath44 are classified in the same group , but that @xmath67 and @xmath44 are not classified as belonging to the same group . we believe that the work by opi et al . is an important first step towards a more rigorous formalization of the problem of graph clustering . zanghi et al . @xcite have designed a clustering technique that lies somewhat in between the method by hastings and that by newman and leicht . as in ref . @xcite , they use the planted partition model to represent a graph with community structure ; as in ref . @xcite , they maximize the classification likelihood using an expectation - maximization algorithm @xcite . the algorithm runs for a fixed number of clusters @xmath91 , like that by newman and leicht ; however , the optimal number of clusters can be determined by running the algorithm for a range of @xmath91-values and selecting the solution that maximizes the integrated classification likelihood introduced by biernacki et al . the time complexity of the algorithm is @xmath129 . hofman and wiggins have proposed a general bayesian approach to the problem of graph clustering @xcite . like hastings @xcite , they model a graph with community structure as in the planted partition problem ( section [ sec6 ] ) , in that there are two probabilities @xmath540 and @xmath541 that there is an edge between vertices of the same or different clusters , respectively . the unobserved community structure is indicated by the set of labels @xmath542 for the vertices ; @xmath523 is again the fraction of vertices in group @xmath375 . the conjugate prior distributions @xmath543 and @xmath544 are chosen to be beta and dirichlet distributions . the most probable number of clusters @xmath545 maximizes the conditional probability @xmath546 that there are @xmath289 clusters , given the matrix @xmath547 . like hastings , hofman and wiggins assume that the prior probability @xmath548 on the number of clusters is a smooth function , therefore maximizing @xmath546 amounts to maximizing the bayesian evidence @xmath549 , obtained by integrating the joint distribution @xmath550 , which is factorizable , over the model parameters @xmath551 and @xmath552 . the integration can be performed exactly only for small graphs . hofman and wiggins used variational bayes @xcite , in order to compute controlled approximations of @xmath553 . the complexity of the algorithm was estimated numerically on synthetic graphs , yielding @xmath554 , with @xmath555 . in fact , the main limitation comes from high memory requirements . the method is more powerful than the one by hastings @xcite , in that the edge probabilities @xmath551 are inferred by the procedure itself and need not be specified ( or guessed ) at the beginning . it also includes the expectation - maximization approach by newman and leicht @xcite as a special case , with the big advantage that the number of clusters need not be given as an input , but is an output of the method . the software of the algorithm can be found at http://www.columbia.edu/@xmath215chw2/. we conclude with a brief summary of the main techniques described above , coming back to our notation at the beginning of the section . in the method by hastings , one maximizes the likelihood @xmath556 over the set of all possible community assignments @xmath509 , given the number of clusters @xmath44 and the model parameters ( i. e. the linking probabilities @xmath511 and @xmath512 ) . newman and leicht maximize the likelihood @xmath556 for a given number of clusters , over the possible choices for the model parameters and community assignments , by deriving the optimal choices for both variables with a self - consistent procedure . hofman and wiggins maximize the likelihood @xmath557 over the possible choices for the number of clusters . block modeling is a common approach in statistics and social network analysis to decompose a graph in classes of vertices with common properties . in this way , a simpler description of the graph is attained . vertices are usually grouped in classes of equivalence . there are two main definitions of topological equivalence for vertices : _ structural equivalence _ @xcite ( section [ sec3_1_4 ] ) , in which vertices are equivalent if they have the same neighbors ; _ regular equivalence _ @xcite , in which vertices of a class have similar connection patterns to vertices of the other classes ( ex . parents / children ) . regular equivalence does not require that ties / edges are restricted to specific target vertices , so it is a more general concept than structural equivalence . indeed , vertices which are structurally equivalent are also regularly equivalent , but the inverse is not true . the concept of structural equivalence can be generalized to probabilistic models , in which one compares classes of graphs , not single graphs , characterized by a set of linking probabilities between the vertices . in this case , vertices are organized in classes such that the linking probabilities of a vertex with all other vertices of the graph are the same for vertices in the same class , which are called _ stochastically equivalent _ @xcite . a thorough discussion of blockmodeling is beyond the scope of this review : we point the reader to ref . @xcite . here we discuss a recent work by reichardt and white @xcite . let us suppose to have a directed graph with @xmath15 vertices and @xmath14 edges . a classification of the graph is indicated by the set of labels @xmath558 , where @xmath559 is the class of vertex @xmath67 . the corresponding blockmodel , or _ image graph _ , is expressed by a @xmath560 adjacency matrix @xmath256 : @xmath561 if edges between classes @xmath562 and @xmath563 are allowed , otherwise it is zero . the aim is finding the classification @xmath558 and the matrix @xmath10 that best fits the adjacency matrix @xmath189 of the graph . the goodness of the fit is expressed by the quality function @xmath564 , \label{eqstinf4}\ ] ] where @xmath565 ( @xmath566 ) reward the presence ( absence ) of edges between vertices if there are edges ( non - edges ) between the corresponding classes , and @xmath14 is the number of edges of the graph , as usual . eq . [ eqstinf4 ] can be rewritten as a sum over the classes @xmath567)b_{rs } , \label{eqstinf5}\ ] ] by setting @xmath568 and @xmath569=(1/m)\sum_{i\neq j}b_{ij}\delta_{\sigma_ir}\delta_{\sigma_js}$ ] . if one sets @xmath570 and @xmath571 , @xmath362 can be interpreted as the linking probability between @xmath67 and @xmath68 , in some null model . thereof , @xmath572 becomes the number of edges running between vertices of class @xmath375 and @xmath155 , and @xmath569 $ ] the expected number of edges in the null model . reichardt and white set @xmath573 , which defines the same null model of newman - girvan modularity for directed graphs ( section [ sub_sec6_01 ] ) . in fact , if the image graph has only self - edges , i. e. @xmath574 , the quality function @xmath575 exactly matches modularity . other choices for the image graph are possible , however . for instance , a matrix @xmath576 describes the classes of a @xmath91-partite graph ( section [ sec1_1 ] ) . from eq . [ eqstinf5 ] we see that , for a given classification @xmath558 , the image graph that yields the largest value of the quality function @xmath575 is that in which @xmath577 when the term @xmath578 $ ] is non - negative , and @xmath579 when the term @xmath578 $ ] is non - positive . so , the best classification is the one maximizing the quality function @xmath580|| , \label{eqstinf6}\ ] ] where all terms of the sum are taken in absolute value . the function @xmath581 is maximized via simulated annealing . the absolute maximum @xmath403 is obtained by construction when @xmath91 matches the number @xmath582 of structural equivalence classes of the graph . however , the absolute maximum @xmath403 does not have a meaning by itself , as one can achieve fairly high values of @xmath581 also for null model instances of the original graph , i. e. if one randomizes the graph by keeping the same expected indegree and outdegree sequences . in practical applications , the optimal number of classes is determined by comparing the ratio @xmath583 [ @xmath584 is the maximum of @xmath581 for @xmath91 classes ] with the expected ratio for the null model . since classifications for different @xmath91-values are not hierarchically ordered , overlaps between classes may be detected . the method can be trivially extended to the case of weighted graphs . model selection @xcite aims at finding models which are at the same time simple and good at describing a system / process . a basic example of a model selection problem is curve fitting . there is no clear - cut recipe to select a model , but a bunch of heuristics , like akaike information criterion ( aic ) @xcite , bayesian information criterion ( bic ) @xcite , minimum description length ( mdl ) @xcite , minimum message length ( mml ) @xcite , etc .. the modular structure of a graph can be considered as a compressed description of the graph to approximate the whole information contained in its adjacency matrix . based on this idea , rosvall and bergstrom @xcite envisioned a communication process in which a partition of a graph in communities represents a synthesis @xmath585 of the full structure that a signaler sends to a receiver , who tries to infer the original graph topology @xmath168 from it ( fig . [ rosvall1 ] ) . the same idea is at the basis of an earlier method by sun et al . @xcite , which was originally designed for bipartite graphs evolving in time and will be described in section [ sec7_1_2 ] . the best partition corresponds to the signal @xmath585 that contains the most information about @xmath168 . this can be quantitatively assessed by the minimization of the conditional information @xmath586 of @xmath168 given @xmath585 , @xmath587 , \label{eqstinf7}\ ] ] where @xmath91 is the number of clusters , @xmath193 the number of vertices in cluster @xmath67 , @xmath588 the number of edges between clusters @xmath67 and @xmath68 . we remark that , if one imposes no constraints on @xmath91 , @xmath586 is minimal in the trivial case in which @xmath589 ( @xmath590 ) . this solution is not acceptable because it does not correspond to a compression of information with respect to the original data set . one has to look for the ideal tradeoff between a good compression and a small enough information @xmath586 . the minimum description length ( mdl ) principle @xcite provides a solution to this problem , which amounts to the minimization of a function given by @xmath586 plus a function of the number @xmath15 of vertices , @xmath14 of edges and @xmath91 of clusters . the optimization is performed by simulated annealing , so the method is rather slow and can be applied to graphs with up to about @xmath251 vertices . however , faster techniques may in principle be used , even if they imply a loss in accuracy . the method appears superior than modularity optimization , especially when communities are of different sizes . this comes from tests performed on the benchmark of girvan and newman @xcite ( section [ sec6_1 ] ) , both in its original version and in asymmetric versions , proposed by the authors , where the clusters have different sizes or different average degrees . in addition , it can detect other types of vertex classifications than communities , as in eq . [ eqstinf7 ] there are no constraints on the relative importance of the edge densities within communities with respect to the edge densities between communities . the software of the algorithm can be found at http://www.tp.umu.se/@xmath215rosvall/code.html . in a recent paper @xcite , rosvall and bergstrom pursued the same idea of describing a graph by using less information than that encoded in the full adjacency matrix . the goal is to optimally compress the information needed to describe the process of information diffusion across the graph . random walk is chosen as a proxy of information diffusion . a two - level description , in which one gives unique names to important structures of the graph and to vertices within the same structure , but the vertex names are recycled among different structures , leads to a more compact description than by simply coding all vertices with different names . this is similar to the procedure usually adopted in geographic maps , where the structures are cities and one usually chooses the same names for streets of different cities , as long as there is only one street with a given name in the same city . huffman coding @xcite is used to name vertices . for the random walk , the above - mentioned structures are communities , as it is intuitive that walkers will spend a lot of time within them , so they play a crucial role in the process of information diffusion . graph clustering turns then into the following coding problem : finding the partition that yields the minimum description length of an infinite random walk . such description length consists of two terms , expressing the shannon entropy of the random walk within and between clusters . every time the walker steps to a different cluster , one needs to use the codeword of that cluster in the description , to inform the decoder of the transition . clearly , if clusters are well separated from each other , transitions of the random walker between clusters will be unfrequent , so it is advantageous to use the map , with the clusters as regions , because in the description of the random walk the codewords of the clusters will not be repeated many times , while there is a considerable saving in the description due to the limited length of the codewords used to denote the vertices . instead , if there are no well - defined clusters and/or if the partition is not representative of the actual community structure of the graph , transitions between the clusters of the partition will be very frequent and there will be little or no gain by using the two - level description of the map . the minimization of the description length is carried out by combining greedy search with simulated annealing . in a successive paper @xcite , the authors adopted the fast greedy technique designed by blondel et al . for modularity optimization @xcite , with some refinements . the method can be applied to weighted graphs , both undirected and directed . in the latter case , the random walk process is modified by introducing a teleportation probability @xmath253 , to guarantee ergodicity , just like in google s pagerank algorithm @xcite . the partitions of directed graphs obtained by the method differ from those derived by optimizing the directed version of newman - girvan modularity ( section [ sub_sec6_01 ] ) : this is due to the fact that modularity focuses on pairwise relationships between vertices , so it does not capture flows . the code of the method is available at http://www.tp.umu.se/@xmath215rosvall/code.html . chakrabarti @xcite has applied the mdl principle to put the adjacency matrix of a graph into the ( approximately ) block diagonal form representing the best tradeoff between having a limited number of blocks , for a good compression of the graph topology , and having very homogeneous blocks , for a compact description of their structure . the total encoding cost @xmath467 includes the information on the total number of vertices of the graph , on the number of blocks and the number of vertices and edges in each block , along with the adjacency matrices of the blocks . the minimization of @xmath467 is carried out by starting from the partition in which the graph is a single cluster . at each step , one operates a bipartition of the cluster of the partition with the maximum shannon entropy per vertex . the split is carried out in order to remove from the original cluster those vertices carrying the highest contribution to the entropy per vertex of the cluster . then , starting from the resulting partition , which has one more cluster than the previous one , @xmath467 is optimized among those partitions with the same number of clusters . the procedure continues until one reaches a number of clusters @xmath591 , for which @xmath467 can not be further decreased . the method by chakrabarti has complexity @xmath592 $ ] , where @xmath593 is the number of iterations required for the convergence of the optimization for a given number of clusters , which is usually small ( @xmath594 in the experiments performed by the author ) . therefore the algorithm can be applied to fairly large graphs . information theory has also been used to detect communities in graphs . ziv et al . @xcite have designed a method in which the information contained in the graph topology is compressed such to preserve some predefined information . this is the basic principle of the information bottleneck method @xcite . to understand this criterion , we need to introduce an important measure , the _ mutual information _ @xmath595 @xcite of two random variables @xmath168 and @xmath585 . it is defined as @xmath596 where @xmath597 indicates the probability that @xmath598 ( similarly for @xmath599 ) and @xmath600 is the joint probability of @xmath168 and @xmath585 , i. e. @xmath601 . the measure @xmath595 tells how much we learn about @xmath168 if we know @xmath585 , and viceversa . if @xmath168 is the input variable , @xmath602 the variable specifying the partition and @xmath585 the variable encoding the information we want to keep , which is called _ relevant variable _ , the goal is to minimize the mutual information between @xmath168 and @xmath602 ( to achieve the largest possible data compression ) , under the constraint that the information on @xmath585 extractable from @xmath602 be accurate . the optimal tradeoff between the values of @xmath603 and @xmath604 ( i. e. compression versus accuracy ) is expressed by the minimization of a functional , where the relative weight of the two contributions is given by a parameter playing the role of a temperature . in the case of graph clustering , the question is what to choose as relevant information variable . ziv et al . proposed to adopt the structural information encoded in the process of diffusion on the graph . they also introduce the concept of _ network modularity _ , which characterizes the graph as a whole , not a specific partition like the modularity by newman and girvan ( section [ sec3_2_2 ] ) . the network modularity is defined as the area under the _ information curve _ , which essentially represents the relation between the extent of compression and accuracy for all solutions found by the method and all possible numbers of clusters . the software of the algorithm by ziv et al . can be found at http://www.columbia.edu/@xmath215chw2/. in this section we describe some algorithms that do not fit in the previous categories , although some overlap is possible . raghavan et al . @xcite have designed a simple and fast method based on _ label propagation_. vertices are initially given unique labels ( e.g. their vertex labels ) . at each iteration , a sweep over all vertices , in random sequential order , is performed : each vertex takes the label shared by the majority of its neighbors . if there is no unique majority , one of the majority labels is picked at random . in this way , labels propagate across the graph : most labels will disappear , others will dominate . the process reaches convergence when each vertex has the majority label of its neighbors . communities are defined as groups of vertices having identical labels at convergence . by construction , each vertex has more neighbors in its community than in any other community . this resembles the strong definition of community we have discussed in section [ sec3_1_2 ] , although the latter is stricter , in that each vertex must have more neighbors in its community than in the rest of the graph . the algorithm does not deliver a unique solution . due to the many ties encountered along the process it is possible to derive different partitions starting from the same initial condition , with different random seeds . tests on real graphs show that all partitions found are similar to each other , though . the most precise information that one can extract from the method is contained by aggregating the various partitions obtained , which can be done in various ways . the authors proposed to label each vertex with the set of all labels it has in different partitions . aggregating partitions enables one to detect possible overlapping communities . the main advantage of the method is the fact that it does not need any information on the number and the size of the clusters . it does not need any parameter , either . the time complexity of each iteration of the algorithm is @xmath21 , the number of iterations to convergence appears independent of the graph size , or growing very slowly with it . so the technique is really fast and could be used for the analysis of large systems . in a recent paper @xcite , tibly and kertsz showed that the method is equivalent to finding the local energy minima of a simple zero - temperature kinetic potts model , and that the number of such energy minima is considerably larger than the number of vertices of the graph . aggregating partitions as raghavan et al . suggest leads to a fragmentation of the resulting partition in clusters that are the smaller , the larger the number of aggregated partitions . this is potentially a serious problem of the algorithm by raghavan et al . , especially when large graphs are investigated . in order to eliminate undesired solutions , barber and clark introduced some constraints in the optimization process @xcite . this amounts to adding some terms to the objective function @xmath605 whose maximization is equivalent to the original label propagation algorithm , where @xmath189 is the adjacency matrix of the graph and @xmath101 is kronecker s function . it is just the negative of the energy of a zero - temperature potts model , as found by tibly and kertsz @xcite ] . interestingly , if one imposes the constraint that partitions have to be balanced , i. e. that clusters have similar total degrees , the objective function becomes formally equivalent to newman - girvan modularity @xmath90 ( section [ sec3_2_2 ] ) , so the corresponding version of the label propagation algorithm is essentially based on a local optimization of modularity . leung et al . have found that the original algorithm by raghavan et al . , applied on online social networks , often yields partitions with one giant community together with much smaller ones @xcite . in order to avoid this disturbing feature , which is an artefact of the algorithm , leung et al . proposed to modify the method by introducing a score for the labels , which decreases as the label propagates far from the vertex to which the label was originally assigned . when choosing the label of a vertex , the labels of its neighbors are weighted by their scores , therefore a single label can not span too large portions of the graph ( as its weight fades away with the distance from the origin ) , and no giant communities can be recovered . tests of the modified algorithm on the lfr benchmark @xcite ( section [ sub_sec6_201 ] ) give good results and encourage further investigations . bagrow and bollt designed an agglomerative technique , called _ l - shell method _ @xcite . it is a procedure that finds the community of any vertex , although the authors also presented a more general procedure to identify the full community structure of the graph . communities are defined locally , based on a simple criterion involving the number of edges inside and outside a group of vertices . one starts from a vertex - origin and keeps adding vertices lying on successive shells , where a shell is defined as a set of vertices at a fixed geodesic distance from the origin . the first shell includes the nearest neighbours of the origin , the second the next - to - nearest neighbours , and so on . at each iteration , one calculates the number of edges connecting vertices of the new layer to vertices inside and outside the running cluster . if the ratio of these two numbers ( `` emerging degree '' ) exceeds some predefined threshold , the vertices of the new shell are added to the cluster , otherwise the process stops the idea of closing a community by expanding a shell has been previously introduced by costa @xcite , in which shells are centered on hubs . however , in this procedure the number of clusters is preassigned and no cluster can contain more than one hub . because of the local nature of the process , the l - shell method is very fast and can identify communities very quickly . unfortunately the method works well only when the source vertex is approximately equidistant from the boundary of its community . to overcome this problem , bagrow and bollt suggested to repeat the process starting from every vertex and derive a _ membership matrix _ @xmath445 : the element @xmath492 is one if vertex @xmath68 belongs to the community of vertex @xmath67 , otherwise it is zero . the membership matrix can be rewritten by suitably permutating rows and columns based on their mutual distances . the distance between two rows ( or columns ) is defined as the number of entries whose elements differ . if the graph has a clear community structure , the membership matrix takes a block - diagonal form , where the blocks identify the communities . the method enables one to detect overlaps between communities as well @xcite . unfortunately , the rearrangement of the matrix requires a time @xmath154 , so it is quite slow . a variant of the algorithm by bagrow and bollt , in which boundary vertices are examined separately and both first and second nearest neighbors of the running community are simultaneously investigated , was suggested by rodrigues et al . @xcite . a recent methodology introduced by papadopoulos et al . @xcite , called _ bridge bounding _ , is similar to the l - shell algorithm , but here the cluster around a vertex grows until one `` hits '' the boundary edges . such edges can be recognized from the values of various measures , like betweenness @xcite or the edge clustering coefficient @xcite . the problem is that there are often no clear gaps in the distributions of the values of such measures , so one is forced to set a threshold to automatically identify the boundary edges from the others , and there is no obvious way to do it . the best results of the algorithm are obtained by using a measure consisting of a weighted sum of the edge clustering coefficient over a wider neighborhood of the given edge . this version of the method has a time complexity @xmath606 , where @xmath420 is the average degree of the graph . in another algorithm by clauset , local communities are discovered through greedy maximization of a local modularity measure @xcite . given a community @xmath22 , the boundary @xmath259 is the set of vertices of @xmath22 with at least one neighbor outside @xmath22 ( fig . [ clausetloc ] ) . schematic picture of a community @xmath22 used in the definition of local modularity by clauset @xcite . the black area indicates the subgraph of @xmath22 including all vertices of @xmath22 , whose neighbors are also in @xmath22 . the boundary @xmath259 entails the vertices of @xmath22 with at least one neighbor outside the community . reprinted figure with permission from ref . 2005 by the american physical society.,width=340 ] the local modularity @xmath134 by clauset is the ratio of the number of edges having both endpoints in @xmath22 ( but at least one in @xmath259 ) , with the number of edges having at least one endpoint in @xmath259 . it is a measure of the sharpness of the community boundary . its optimization consists of a local exploration of the community starting from a source vertex : at each step the neighboring vertex yielding the largest increase ( smallest decrease ) of @xmath134 is added , until the community has reached a predefinite size @xmath115 . this greedy optimization takes a time @xmath607 , where @xmath420 is the average degree of the graph . the local modularity @xmath134 has been used in a paper by hui et al . @xcite , where methods to find communities in networks of mobile devices are designed . another method , where communities are defined based on a local criterion , was presented by eckmann and moses @xcite . the idea is to use the clustering coefficient @xcite of a vertex as a quantity to distinguish tightly connected groups of vertices . many edges mean many loops inside a community , so the vertices of a community are likely to have a large clustering coefficient . the latter can be related to the average distance between pairs of neighbours of the vertex . the possible values of the distance are @xmath2 ( if neighbors are connected ) or @xmath323 ( if they are not ) , so the average distance lies between @xmath2 and @xmath323 . the more triangles there are in the subgraph , the shorter the average distance . since each vertex always has distance @xmath2 from its neighbours , the fact that the average distance between its neighbours is different from @xmath2 reminds what happens when one measures segments on a curved surface . endowed with a metric , represented by the geodesic distance between vertices / points , and a curvature , the graph can be embedded in a geometric space . communities appear as portions of the graph with a large curvature . the algorithm was applied to the graph representation of the world wide web , where vertices are web pages and edges are the hyperlinks that take users from a page to the other . the authors found that communities correspond to web pages dealing with the same topic . long et al . have devised an interesting technique that is able to detect various types of vertex groups , not necessarily communities @xcite . the method is based on graph approximation , as it tries to match the original graph topology onto a coarse type of graph , the _ community prototype graph _ , which has a clear group structure ( block - diagonal for clusters , block - off - diagonal for classes of multipartite graphs , etc . ) . the goal is to determine the community prototype graph that best approximates the graph at study , where the goodness of the approximation is expressed by the distance between the corresponding matrices . in this way the original problem of finding graph subsets becomes an optimization problem . long et al . called this procedure _ community learning by graph approximation _ sometimes the minimization of the matrix distance can be turned into the maximization of the trace of a matrix . measures like cut size or ratio cut can be also formulated as the trace of matrices ( see for instance eq . [ eqr ] ) . in fact , clga includes traditional graph partitioning as a special case ( section [ sec4_1 ] ) . long et al . designed three algorithms for clga : two of them seek for divisions of the graph into overlapping or non - overlapping groups , respectively ; in the third one an additional constraint is introduced to produce groups of comparable size . the complexity of these algorithms is @xmath608 , where @xmath156 is the number of iterations until the optimization converges and @xmath44 the number of groups . the latter has to be given as an input , which is a serious limit of clga . a fast algorithm by wu and huberman identifies communities based on the properties of resistor networks @xcite . it is essentially a method for partitioning graphs in two parts , similar to spectral bisection , although partitions in an arbitrary number of communities can be obtained by iterative applications . the graph is transformed into a resistor network where each edge has unit resistance . a unit potential difference is set between two randomly chosen vertices . the idea is that , if there is a clear division in two communities of the graph , there will be a visible gap between voltage values for vertices at the borders between the clusters . the voltages are calculated by solving kirchoff s equations : an exact solution would be too time consuming , but it is possible to find a reasonably good approximation in a linear time for a sparse graph with a clear community structure , so the more time consuming part of the algorithm is the sorting of the voltage values , which takes time @xmath254 . any possible vertex pair can be chosen to set the initial potential difference , so the procedure should be repeated for all possible vertex pairs . the authors showed that this is not necessary , and that a limited number of sampling pairs is sufficient to get good results , so the algorithm scales as @xmath254 and is very fast . an interesting feature of the method is that it can quickly find the natural community of any vertex , without determining the complete partition of the graph . for that , one uses the vertex as source voltage and places the sink at an arbitrary vertex . the same feature is present in an older algorithm by flake et al . @xcite , where one uses max - flow instead of current flow ( section [ sec4_1 ] ) . an algorithm by orponen and schaeffer @xcite is based on the same principle , but it does not need the specification of target sources as it is based on diffusion in an unbounded medium . the limit of such methods is the fact that one has to give as input the number of clusters , which is usually not known beforehand . ohkubo and tanaka @xcite pointed out that , since communities are rather compact structures , they should have a small volume , where the volume of a community is defined as the ratio of the number of vertices by the internal edge density of the community . ohkubo and tanaka assumed that the sum @xmath609 of the volumes of the communities of a partition is a reliable index of the goodness of the partition . so , the most relevant partition is the one minimizing @xmath609 . the optimization is carried out with simulated annealing . zarei and samani @xcite remarked that there is a symmetry between community structure and anti - community ( multipartite ) structure , when one considers a graph and its complement , whose edges are the missing edges of the original graph . in fact , if a graph has a well identified communities , the same groups of vertices would be strong anti - communities in the complement graph , i. e. they should have a few intra - cluster edges and many inter - cluster edges . based on this remark , the communities of a graph can be identified by looking for anticommunities in the complement graph , which can sometimes be easier . zarei and samani devised a spectral method using matrices of the complement graph . the results of this technique appear good as compared to other spectral methods on artificial graphs generated with the planted @xmath293-partition model @xcite , as well as on zachary s karate club @xcite , lusseau s dolphins network @xcite and a network of protein - protein interactions . however , the authors have used very small graphs for testing . communities make sense on sparse graphs , but the complements of large sparse graphs would not be sparse , but very dense , and their community ( multipartite ) structure basically invisible . gudkov and montealegre detected communities by means of dynamical simplex evolution @xcite . graph vertices are represented as points in an @xmath306-dimensional space . each point initially sits on the @xmath15 vertices of a simplex , and then moves in space due to forces exerted by the other points . if vertices are neighbors , the mutual force acting on their representative points is attractive , otherwise it is repulsive . if the graph has a clear community structure , the corresponding spatial clusters repel each other because of the few connections between them ( repulsion dominates over attraction ) . if communities are more mixed with each other , clusters are not well separated and they could be mistakenly aggregated in larger structures . to avoid that , gudkov and montealegre defined clusters as groups of points such that the distance between each pair of points does not exceed a given threshold , which can be arbitrarily tuned , to reveal structures at different resolutions ( section [ sub_sec6_201 ] ) . the algorithm consists in solving first - order differential equations , describing the dynamics of mass points moving in a viscous medium . the complexity of the procedure is @xmath129 . differential equations are also at the basis of a recent method designed by krawczyk and kuakowski @xcite . here the equations describe a dynamic process , in which the original graph topology evolves to a disconnected graph , whose components are the clusters of the original graph . despite the significant improvements in computational complexity , it is still problematic to apply clustering algorithms to many large networks available today . therefore narasimhamurthy et al . @xcite proposed a two - step procedure : first , the graph at study is decomposed in smaller pieces by a fast graph partitioning technique ; then , a clustering method is applied to each of the smaller subgraphs obtained [ narasimhamurthy et al . used the clique percolation method ( section [ sec45_1 ] ) ] . the initial decomposition of the graph is carried out through the multilevel method by dhillon et al . it is crucial to verify that the initial partitioning does not split the communities of the graph among the various subgraphs of the decomposition . this can be done by comparing , on artificial graphs , the final clusters obtained with the two - step method with those detected by applying the chosen clustering technique to the entire graph . most of the methods discussed in the previous sections aim at detecting standard partitions , i. e. partitions in which each vertex is assigned to a single community . however , in real graphs vertices are often shared between communities ( section [ sec2 ] ) , and the issue of detecting overlapping communities has become quite popular in the last few years . we devote this section to the main techniques to detect overlapping communities . the most popular technique is the clique percolation method ( cpm ) by palla et al . @xcite . it is based on the concept that the internal edges of a community are likely to form cliques due to their high density . on the other hand , it is unlikely that intercommunity edges form cliques : this idea was already used in the divisive method of radicchi et al . ( section [ subsec5_2 ] ) . palla et al . use the term @xmath44-clique to indicate a complete graph with @xmath44 vertices vertices ( section [ sec1_1 ] ) . ] . notice that a @xmath44-clique is different from the @xmath15-clique ( see section [ sec3_1_2 ] ) used in social science . if it were possible for a clique to move on a graph , in some way , it would probably get trapped inside its original community , as it could not cross the bottleneck formed by the intercommunity edges . palla et al . introduced a number of concepts to implement this idea . two @xmath44-cliques are _ adjacent _ if they share @xmath188 vertices . the union of adjacent @xmath44-cliques is called _ @xmath44-clique chain_. two @xmath44-cliques are connected if they are part of a @xmath44-clique chain . finally , a _ @xmath44-clique community _ is the largest connected subgraph obtained by the union of a @xmath44-clique and of all @xmath44-cliques which are connected to it . examples of @xmath44-clique communities are shown in fig . [ figure10 ] . clique percolation method . the example shows communities spanned by adjacent 4-cliques . overlapping vertices are shown by the bigger dots . reprinted figure with permission from ref . 2005 by the nature publishing group . ] one could say that a @xmath44-clique community is identified by making a @xmath44-clique `` roll '' over adjacent @xmath44-cliques , where rolling means rotating a @xmath44-clique about the @xmath188 vertices it shares with any adjacent @xmath44-clique . by construction , @xmath44-clique communities can share vertices , so they can be overlapping . there may be vertices belonging to non - adjacent @xmath44-cliques , which could be reached by different paths and end up in different clusters . unfortunately , there are also vertices that can not be reached by any @xmath44-clique , like , e. g. vertices with degree one ( `` leaves '' ) . in order to find @xmath44-clique communities , one searches first for maximal cliques . then a clique - clique overlap matrix @xmath400 is built @xcite , which is an @xmath610 matrix , @xmath115 being the number of cliques ; @xmath611 is the number of vertices shared by cliques @xmath67 and @xmath68 . to find @xmath44-cliques , one needs simply to keep the entries of @xmath400 which are larger than or equal to @xmath188 , set the others to zero and find the connected components of the resulting matrix . detecting maximal cliques is known to require a running time that grows exponentially with the size of the graph . however , the authors found that , for the real networks they analyzed , the procedure is quite fast , due to the fairly limited number of cliques , and that ( sparse ) graphs with up to @xmath612 vertices can be analyzed in a reasonably short time . the actual scalability of the algorithm depends on many factors , and can not be expressed in closed form . an interesting aspect of @xmath44-clique communities is that they allow to make a clear distinction between random graphs and graphs with community structure . this is a rather delicate issue : we have seen in section [ sub_sec6_1 ] that newman - girvan modularity can attain large values on random graphs . dernyi et al . @xcite have studied the percolation properties of @xmath44-cliques on random graphs , when the edge probability @xmath45 varies . they found that the threshold @xmath613 for the emergence of a giant @xmath44-clique community , i. e. a community occupying a macroscopic portion of the graph , is @xmath614^{-1/(k-1)}$ ] , @xmath15 being the number of vertices of the graph , as usual . for @xmath615 , for which the @xmath44-cliques reduce to edges , one recovers the known expression for the emergence of a giant connected component in erds - rnyi graphs ( section [ sec1_3 ] ) . this percolation transition is quite sharp : if the edge probability @xmath616 , @xmath44-clique communities are rather small ; if @xmath617 there is a giant component and many small communities . to assess the significance of the clusters found with the cpm , one can compare the detected cover with the cover found on a null model graph , which is random but preserves the expected degree sequence of the original graph . the modularity of newman and girvan is based on the same null model ( section [ sec3_2_2 ] ) . the null models of real graphs seem to display the same two scenarios found for erds - rnyi graphs , characterized by the presence of very small @xmath44-clique communities , with or without a giant cluster . therefore , covers with @xmath44-clique communities of large or appreciable size can hardly be due to random fluctuations . palla and coworkers @xcite have designed a software package implementing the cpm , called _ cfinder _ , which is freely available ( www.cfinder.org ) . the algorithm has been extended to the analysis of weighted , directed and bipartite graphs . for weighted graphs , in principle one can follow the standard procedure of thresholding the weights , and apply the method on the resulting graphs , treating them as unweighted . farkas et al . @xcite proposed instead to threshold the weight of cliques , defined as the geometric mean of the weights of all edges of the clique . the value of the threshold is chosen slightly above the critical value at which a giant @xmath44-clique community emerges , in order to get the richest possible variety of clusters . on directed graphs , palla et al . defined _ directed @xmath44-cliques _ as complete graphs with @xmath44 vertices , such that there is an ordering among the vertices , and each edge goes from a vertex with higher order to one with lower order . the ordering is determined from the _ restricted outdegree _ of the vertex , expressing the fraction of outgoing edges pointing to the other vertices of the clique versus the total outdegree . the method has been extended to bipartite graphs by lehmann et al . @xcite . in this case one uses bipartite cliques , or _ bicliques _ : a subgraph @xmath618 is a biclique if each of @xmath294 vertices of one class are connected with each of @xmath619 vertices of the other class . two cliques @xmath618 are adjacent if they share a clique @xmath620 , and a @xmath618 clique community is the union of all @xmath618 cliques that can be reached from each other through a path of adjacent @xmath618 cliques . finding all @xmath621 bicliques of a graph is an @xmath622-complete problem @xcite , mostly because the number of bicliques tends to grow exponentially with the size of the graph . the algorithm designed by lehmann et al . to find biclique communities is similar to the original cpm , and has a total complexity of @xmath623 . on sparse graphs , @xmath621 often grows linearly with the number of edges @xmath14 , yielding a time complexity @xmath624 . bicliques are also the main ingredients of _ bitector _ , a recent algorithm to detect community structure in bipartite graphs @xcite . kumpula et al . have developed a fast implementation of the cpm , called sequential clique percolation algorithm ( scp ) @xcite . it consists in detecting @xmath44-clique communities by sequentially inserting the edges of the graph at study , one by one , starting from an initial empty graph . whenever a new edge is added , one checks whether new @xmath44-cliques are formed , by searching for @xmath625-cliques in the subset of neighboring vertices of the endpoints of the inserted edge . the procedure requires to build a graph @xmath626 , in which the vertices are @xmath627-cliques and edges are set between vertices corresponding to @xmath627-cliques which are subgraphs of the same @xmath44-clique . at the end of the process , the connected components of @xmath626 correspond to the searched @xmath44-clique communities . the technique has a time complexity which is linear in the number of @xmath44-cliques of the graph , so it can vary a lot in practical applications . nevertheless , it turns out to be much faster than the original implementation of the cpm . the big advantage of the scp , however , consists of its implementation for weighted graphs . by inserting edges in decreasing order of weight , one recovers in a single run the community structure of the graph for all possible weight thresholds , by storing every cover detected after the addition of each edge . the standard cpm , instead , needs to be applied once for each threshold . if , instead of edge weight thresholding , one performs @xmath44-clique weight thresholding , as prescribed by farkas et al . @xcite , the scp remains much faster than the cpm , if one applies a simple modification to it , consisting in detecting and storing all @xmath44-cliques on the full graph , sorting them based on their weights , and finding the communities by sequentially adding the @xmath44-cliques in decreasing order of weight . the cpm has the same limit as the algorithm of radicchi et al . @xcite ( section [ subsec5_2 ] ) : it assumes that the graph has a large number of cliques , so it may fail to give meaningful covers for graphs with just a few cliques , like technological networks and some social networks . on the other hand , if there are many cliques , the method may deliver trivial community structure , like a cover consisting of the whole graph as a single cluster . a more fundamental issue is the fact that the method does not look for actual communities , consistent with the shared notion of dense subgraphs , but for subgraphs `` containing '' many cliques , which may be quite different objects than communities ( for instance , they could be `` chains '' of cliques with low internal edge density ) . another big problem is that on real networks there is a considerable fraction of vertices that are left out of the communities , like leaves . one could think of some post - processing procedure to include them in the communities , but for that it is necessary to introduce a new criterion , outside the framework that inspired the method . furthermore it is not clear _ a priori _ which value of @xmath44 one has to choose to identify meaningful structures . finally , the criterion to choose the threshold for weighted graphs and the definition of directed @xmath44-cliques are rather arbitrary . one of the first methods to find overlapping communities was designed by baumes et al . a community is defined as a subgraph which locally optimizes a given function @xmath86 , typically some measure related to the edge density of the cluster ) . ] . different overlapping subsets may all be locally optimal , so vertices can be shared between communities . detecting the cluster structure of a graph amounts to finding the set of all locally optimal clusters . two efficient heuristics are proposed , called iterative scan ( @xmath628 ) and rank removal ( @xmath629 ) . @xmath628 performs a greedy optimization of the function @xmath86 . one starts from a random seed vertex / edge and adds / deletes vertices one by one as long as @xmath86 increases . then another seed is randomly picked and the procedure is repeated . the algorithm stops when , by picking any seed , one recovers a previously identified cluster . @xmath629 consists in removing important vertices such to disconnect the graphs in small components representing the cores of the clusters . the importance of vertices is determined by their centrality scores ( e.g. degree , betweenness centrality @xcite ) , pagerank @xcite ) . vertices are removed until one fragments the graph into components of a given size . after that , the removed vertices are added again to the graph , and are associated to those clusters for which doing so increases the value of the function @xmath86 . the complexity of @xmath628 and @xmath629 is @xmath129 on sparse graphs . the best performance is achieved by using @xmath628 to refine results obtained from @xmath629 . in a successive paper @xcite , baumes et al . further improved such two - step procedure , in that the removed vertices in @xmath629 are reinserted in decreasing order of their centrality scores , and the optimization of @xmath86 in @xmath628 is only extended to neighboring vertices of the running cluster . the new recipe maintains time complexity @xmath129 , but on sparse graphs it requires a time lower by an order of magnitude than the old one , while the quality of the detected clustering is comparable . a different method , combining spectral mapping , fuzzy clustering and the optimization of a quality function , has been presented by zhang et al . the membership of vertex @xmath67 in cluster @xmath44 is expressed by @xmath630 , which is a number between @xmath631 and @xmath2 . the sum of the @xmath630 over all communities @xmath44 of a cover is @xmath2 , for every vertex . this normalization is suggested by the fact that the entry @xmath630 can be thought of as the probability that @xmath67 belongs to community @xmath44 , so the sum of the @xmath630 represents the probability that the vertex belongs to any community of the cover , which is necessarily @xmath2 . if there were no overlaps , @xmath632 , where @xmath105 represents the unique community of vertex @xmath67 . the algorithm consists of three phases : 1 ) embedding vertices in euclidean space ; 2 ) grouping the corresponding vertex points in a given number @xmath115 of clusters ; 3 ) maximizing a modularity function over the set of covers found in step 2 ) , corresponding to different values of @xmath115 . this scheme has been used in other techniques as well , like in the algorithm of donetti and muoz @xcite ( section [ sec43 ] ) . the first step builds upon a spectral technique introduced by white and smyth @xcite , that we have discussed in section [ sub_sec6_0_4 ] . graph vertices are embedded in a @xmath88-dimensional euclidean space by using the top @xmath88 eigenvectors of the right stochastic matrix @xmath190 ( section [ sec1_2 ] ) , derived from the adjacency matrix @xmath189 by dividing each element by the sum of the elements of the same row . the spatial coordinates of vertex @xmath67 are the @xmath67-th components of the eigenvectors . in the second step , the vertex points are associated to @xmath115 clusters by using fuzzy @xmath44-means clustering @xcite ( section [ sec4_3 ] ) . the number of clusters @xmath115 varies from @xmath323 to a maximum @xmath289 , so one obtains @xmath290 covers . the best cover is the one that yields the largest value of the modularity @xmath633 , defined as @xmath634 , \label{eq : mod5}\ ] ] where @xmath635 and @xmath636 the sets @xmath637 and @xmath638 include the vertices of module @xmath117 and of the whole network , respectively . [ eq : mod5 ] is an extension of the weighted modularity in eq . [ eq : mod20 ] , obtained by weighing the contribution of the edges weights to the sums in @xmath320 and @xmath321 by the ( average ) membership coefficients of the vertices of the edge . the determination of the eigenvectors is the most computationally expensive part of the method , so the time complexity is the same as that of the algorithm by white and smyth ( see section [ sub_sec6_0_4 ] ) , i. e. @xmath291 , which is essentially linear in @xmath15 if the graph is sparse and @xmath639 . nepusz et al . proposed a different approach based on vertex similarity @xcite . one starts from the membership matrix @xmath197 , defined as in the previous method by zhang et al . from @xmath197 a matrix @xmath538 is built , where @xmath640 , expressing the similarity between vertices ( @xmath115 is the number of clusters ) . if one assumes to have information about the actual vertex similarity , corresponding to the matrix @xmath641 , the best cover is obtained by choosing @xmath197 such that @xmath538 approximates as closely as possible @xmath641 . this amounts to minimize the function @xmath642 where the @xmath309 weigh the importance of the approximation for each entry of the similarity matrices . in the absence of any information on the community structure of the graph , one sets @xmath643 , @xmath104 ( equal weights ) and @xmath641 equal to the adjacency matrix @xmath189 , by implicitly assuming that vertices are similar if they are neighbors , dissimilar otherwise . on weighted graphs , one can set the @xmath309 equal to the edge weights . minimizing @xmath644 is a nonlinear constrained optimization problem , that can be solved with a gradient - based iterative optimization method , like simulated annealing . the optimization procedure adopted by nepusz et al . , for a fixed number of clusters @xmath115 , has a time complexity @xmath645 , where @xmath646 is the number of iterations leading to convergence , so the method can only be applied to fairly small graphs . if @xmath115 is unknown , as it usually happens , the best cover is the one corresponding to the largest value of the modularity @xmath647 eq . [ eqoverl2 ] is very similar to the expression of newman - girvan modularity ( eq . [ eq : mod_0 ] ) : the difference is that the kronecker s @xmath101 is replaced by the vertices similarity , to account for overlapping communities . once the best cover is identified , one can use the entries of the partition matrix @xmath197 to evaluate the participation of each vertex in the @xmath115 clusters of the cover . nepusz et al . defined the _ bridgeness _ @xmath648 of a vertex @xmath67 as @xmath649 if @xmath67 belongs to a single cluster , @xmath650 . if , for a vertex @xmath67 , @xmath651 , @xmath652 , @xmath653 and @xmath67 is a perfect _ bridge _ , as it lies exactly between all clusters however , a vertex with low @xmath654 may be simply an outlier , not belonging to any cluster . since real bridges are usually rather central vertices , one can identify them by checking for large values of the _ centrality - corrected bridgeness _ , obtained by multiplying the bridgeness of eq . [ eqoverl3 ] by the centrality of the vertex ( expressed by , e.g. , degree , betweenness @xcite , etc . ) . a variant of the algorithm by nepusz et al can be downloaded from http://www.cs.rhul.ac.uk/home/tamas/assets/file + s / fuzzyclust - static.tar.gz . in real networks it is often easier to discriminate between intercluster and intracluster edges than recognizing overlapping vertices . for instance , in social networks , even though many people may belong to more groups , their social ties within each group can be easily spotted . besides , it may happen that communities are joined to each other through their overlapping vertices ( fig . [ figlink ] ) , without intercluster edges . for these reasons , it has been recently suggested that defining clusters as sets of edges , rather than vertices , may be a promising strategy to analyze graphs with overlapping communities @xcite . one has to focus on the _ line graph _ @xcite , i. e. the graph whose vertices are the edges of the original graph ; vertices of the line graph are linked if the corresponding edges in the original graph are adjacent , i. e. if they share one of their endvertices . partitioning the line graph means grouping the edges of the starting graph . evans and lambiotte @xcite introduced a set of quality functions , similar to newman - girvan modularity ( eq . [ eq : mod_0 ] ) , expressing the stability of partitions against random walks taking place on the graph , following the work of delvenne et al . @xcite [ section [ sec44_2 ] ] . they considered a projection of the traditional random walk on the line graph , along with two other diffusion processes , where walkers move between adjacent edges ( rather than between neighboring vertices ) . evans and lambiotte optimized the three corresponding modularity functions to look for partitions in two real networks , zachary s karate club @xcite ( section [ sec6_1 ] ) and the network of word associations derived from the university of south florida free association norms @xcite ( section [ sec2 ] ) . the optimization was carried out with the hierarchical technique by blondel et al . @xcite and the multi - level algorithm by noack and rotta @xcite . while the results for the word association network are reasonable , the test on the karate club yields partitions in more than two clusters . however , the modularities used by evans et lambiotte can be modified to include longer random walks ( just like in ref . @xcite ) , and the length of the walk represents a resolution parameter that can be tuned to get better results . ahn et al . @xcite proposed to group edges with an agglomerative hierarchical clustering technique , called _ hierarchical link clustering _ ( section [ sec4_2 ] ) . they use a similarity measure for a pair of ( adjacent ) edges that expresses the size of the overlap between the neighborhoods of the non - coincident endvertices , divided by the total number of ( different ) neighbors of such endvertices . groups of edges are merged pairwise in descending order of similarity , until all edges are together in the same cluster . the resulting dendrogram provides the most complete information on the community structure of the graph . however , as usual , most of this information is redundant and is an artefact of the procedure itself . so , ahn et al . introduced a quality function to select the most meaningful partition(s ) , called _ partition density _ , which is essentially the average edge density within the clusters . the method is able to find meaningful clusters in biological networks , like protein - protein and metabolic networks , as well as in a social network of mobile phone communications . it can also be extended to multipartite and weighted graphs . the idea of grouping edges is surely interesting . however it is not _ a priori _ better than grouping vertices . in fact , the two situations are somewhat symmetric . edges connecting vertices of different clusters are `` overlapping '' , but they will be assigned just to one cluster ( or else the clusters would be merged ) . communities as sets of edges . in the figure , the graph has a natural division in two triangles , with the central vertex shared between them . if communities are identified by their internal edges , detecting the triangles and their overlapping vertex becomes easier than by using methods that group vertices . reprinted figure with permission from ref . 2009 by the american physical society.,width=264 ] the possibility of having overlapping communities makes most standard clustering methods inadequate , and enforces the design of new _ ad hoc _ techniques , like the ones we have described so far . on the other hand , if it were possible to identify the overlapping vertices and `` separate '' them among the clusters they belong to , the overlaps would be removed and one could then apply any of the traditional clustering methods to the resulting graph . this idea is at the basis of a recent method proposed by gregory @xcite . it is a three - stages procedure : first , one transforms the graph into a larger graph without overlapping vertices ; second , a clustering technique is applied to the resulting graph ; third , one maps the partition obtained into a cover by replacing the vertices with those of the original graph . the transformation step , called _ peacock _ , is performed by identifying the vertices with highest _ split betweenness _ ( section [ subsec5_1 ] ) and splitting them in multiple parts , connected by edges . this is done as long as the split betweenness of the vertices is sufficiently high , which is determined by a parameter @xmath155 . in this way , most vertices of the resulting graph are exactly the same one had initially , the others are multiple copies of the overlapping vertices of the initial graph . the overlaps of the final cover are obtained by checking if copies of the same initial vertex end up in different disjoint clusters . the complexity is dominated by the peacock algorithm , if one computes the exact values of the split betweenness for the vertices , which requires a time @xmath154 on a sparse graph . gregory proposed an approximate local computation , which scales as @xmath254 : in this way the total complexity of the method becomes competitive , if one chooses a fast algorithm for the identification of the clusters . the goodness of the results depends on the specific method one uses to find the clusters after the graph transformation . the software of the version of the method used by gregory in his applications can be found at http://www.cs.bris.ac.uk/@xmath215steve/networks/peaco + ckpaper/. the idea of gregory is interesting , as it allows to exploit traditional methods even in the presence of overlapping communities . the choice of the parameter @xmath155 , which determines whether a vertex is overlapping or not , does not seem to affect significantly the results , as long as @xmath155 is taken sufficiently small . the existence of a resolution limit for newman - girvan modularity ( section [ sub_sec6_1 ] ) implies that the straight optimization of quality functions yields a coarse description of the cluster structure of the graph , at a scale which has _ a priori _ nothing to do with the actual scale of the clusters . in the absence of information on the cluster sizes of the graph , a method should be able to explore all possible scales , to make sure that it will eventually identify the right communities . multiresolution methods are based on this principle . however , many real graphs display _ hierarchical _ cluster structures , with clusters inside other clusters @xcite . in these cases , there are more levels of organization of vertices in clusters , and more relevant scales . in principle , clustering algorithms should be able to identify them . multiresolution methods can do the trick , in principle , as they scan continuously the range of possible cluster scales . recently other methods have been developed , where partitions are by construction hierarchically nested in each other . in this section we discuss both classes of techniques . in general , multiresolution methods have a freely tunable parameter , that allows to set the characteristic size of the clusters to be detected . the general spin glass framework by reichardt and bornholdt ( @xcite and section [ sub_sec6_01 ] ) is a typical example , where @xmath371 is the resolution parameter . the extension of the method to weighted graphs has been recently discussed @xcite . pons has proposed a method @xcite consisting of the optimization of multiscale quality functions , including the _ multiscale modularity _ @xmath655 , \label{eq : mult0}\ ] ] and two other additive quality functions , derived from the _ performance _ ( eq . [ eq10_perf ] ) and a measure based on the similarity of vertex pairs . in eq . [ eq : mult0 ] @xmath656 is the resolution parameter and the notation is otherwise the same as in eq . [ eq : mod ] . we see that , for @xmath657 , one recovers standard modularity . however , since multiplicative factors in @xmath658 do not change the results of the optimization , we can divide @xmath658 by @xmath18 , recovering the same quality function as in eq . [ eqr13 ] , with @xmath659 , up to an irrelevant multiplicative constant . to evaluate the relevance of the partitions , for any given multiscale quality function , pons suggested that the length of the @xmath18-range @xmath660 $ ] , for which a community @xmath22 `` lives '' in the maximum modularity partition , is a good indicator of the stability of the community . he then defined the _ relevance function _ of a community @xmath22 at scale @xmath18 as @xmath661 the relevance @xmath662 of a partition @xmath94 at scale @xmath18 is the average of the relevances of the clusters of the partition , weighted by the cluster sizes . peaks in @xmath18 of @xmath662 reveal the most meaningful partitions . another interesting technique has been devised by arenas et al . @xcite , and consists of a modification of the original expression of modularity . the idea is to make vertices contribute as well to the computation of the edge density of the clusters , by adding a self - loop of strength @xmath375 to each vertex . arenas et al . remarked that the parameter @xmath375 does not affect the structural properties of the graph in most cases , which are usually determined by an adjacency matrix without diagonal elements . with the introduction of the vertex strength @xmath375 , modularity reads @xmath663 , \label{eq : mult1}\ ] ] for the general case of a weighted graph . the notation is the same as in eq . [ eq : mod20 ] , @xmath621 is the number of vertices in cluster @xmath117 . we see that now the relative importance of the two terms in each summand depends on @xmath375 , which can take any value in @xmath664 - 2w / n , \infty[$ ] . arenas et al . made a sweep in the range of @xmath375 , and determined for each @xmath375 the maximum modularity with extremal optimization ( section [ sub_sec6_0_3 ] ) and _ tabu search _ meaningful cluster structures correspond to plateaus in the plot of the number of clusters versus @xmath375 ( fig . [ figarenasplat ] ) . the length of a plateau gives a measure of the _ stability _ of the partition against the variation of @xmath375 . the procedure is able to disclose the community structure of a number of real benchmark graphs . as expected , the most relevant partitions can be found in intervals of @xmath375 not including the value @xmath665 , which corresponds to the case of standard modularity ( fig . [ figarenasplat ] ) . analysis of zachary s karate club with the multiresolution method by arenas et al . the plot shows the number of clusters obtained in correspondence of the resolution parameter @xmath375 . the longest plateau @xmath666 indicates the most stable partition , which exactly matches the social fission observed by zachary . the partition obtained with straight modularity optimization ( @xmath665 ) consists of four clusters and is much less stable with respect to @xmath666 , as suggested by the much shorter length of its plateau . reprinted figure with permission from ref . 2008 by iop publishing . ] a drawback of the method is that it is very slow , as one has to compute the modularity maximum for many values of @xmath375 in order to discriminate between relevant and irrelevant partitions . if the modularity maximum is computed with precise methods like simulated annealing and/or extremal optimization , as in ref . @xcite , only graphs with a few hundred vertices can be analyzed on a single processor . on the other hand the algorithm can be trivially parallelized by running the optimization for different values of @xmath375 on different processors . this is a common feature of all multiresolution methods discussed in this section . in spite of the different formal expressions of modularity , the methods by arenas et al . and reichardt and bornholdt are somewhat related to each other and yield similar results @xcite on zachary s karate club @xcite ( section [ sec6_1 ] ) , synthetic graphs la ravasz - barabsi @xcite and on a model graph with the properties of real weighted social networks . in fact , their modularities can be both recovered from the continuous - time version of the stability of clustering under random walk , introduced by delvenne et al . @xcite ( section [ sec44_2 ] ) . lancichinetti et al . have designed a multiresolution method which is capable of detecting both the hierarchical structure of graphs and overlapping communities @xcite . it is based on the optimization of a fitness function , which estimates the strength of a cluster and entails a resolution parameter @xmath18 . the function could in principle be arbitrary , in their applications the authors chose a simple ansatz based on the tradeoff between the internal and the total degree of the cluster . the optimization procedure starts from a cluster with a single vertex , arbitrarily selected . given a cluster core , one keeps adding and removing neighboring vertices of the cluster as long as its fitness increases . the fitness is recalculated after each addition / removal of a vertex . at some point one reaches a local maximum and the cluster is `` closed '' . then , another vertex is chosen at random , among those not yet assigned to a cluster , a new cluster is built , and so on , until all vertices have been assigned to clusters . during the buildup of a cluster , vertices already assigned to other clusters may be included , i. e. communities may overlap . the computational complexity of the algorithm , estimated on sparse erds - rnyi random graphs , is @xmath667 , with @xmath668 for small values of the resolution parameter @xmath18 , and @xmath669 if @xmath18 is large . for a complete analysis , the worst - case computational complexity is @xmath128 , where the factor @xmath241 comes from the minimum number of different @xmath18-values which are needed to resolve the actual community structure of the graph . relevant partitions are revealed by pronounced spikes in the histogram of the fitness values of covers obtained for different @xmath18-values , where the fitness of a cover is defined as the average fitness of its clusters . a technique based on the potts model , similar to that of reichardt and bornholdt @xcite , has been suggested by ronhovde and nussinov @xcite . the energy of their spin model is @xmath670\delta(\sigma_i,\sigma_j ) . \label{eq : mult2}\ ] ] the big difference with eq . [ eqr13 ] is the absence of a null model term . the model considers pairs of vertices in the same community : edges between vertices are energetically rewarded , whereas missing edges are penalized . the parameter @xmath371 fixes the tradeoff between the two contributions . the energy is minimized by sequentially shifting single vertices / spins to the communities which yield the largest decrease of the system s energy , until convergence . if , for each vertex , one just examines the communities of its neighbors , the energy is minimized in a time @xmath671 , where @xmath19 turns out to be slightly above @xmath2 in most applications , allowing for the analysis of large graphs . this essentially eliminates the problem of limited resolution , as the criterion to decide about the merger or the split of clusters only depends on local parameters . still , for the detection of possible hierarchical levels tuning @xmath371 is mandatory . in a successive paper @xcite , the authors have introduced a new stability criterion for the partitions , consisting of the computation of the similarity of partitions obtained for the same @xmath371 and different initial conditions . the idea is that , if a partition is robust in a given range of @xmath371-values , most replicas delivered by the algorithm will be very similar . on the other hand , if one explores a region of resolutions in between two strong partitions , the algorithm will deliver the one or the other partition and the individual replicas will be , on average , not so similar to each other . so , by plotting the similarity as a function of the resolution parameter @xmath371 , stable communities are revealed by peaks . ronhovde and nussinov adopted similarity measures borrowed from information theory ( section [ sec6_2 ] ) . their criterion of stability can be adopted to determine the relevance of partitions obtained with any multiresolution algorithm . a general problem of multiresolution methods is how to assess the stability of partitions for large graphs . the rapidly increasing number of partitions , obtained by minimal shifts of vertices between clusters , introduces a large amount of noise , that blurs signatures of stable partitions like plateaus , spikes , etc . that one can observe in small systems . in this respect , it seems far more reliable focusing on correlations between partitions ( like the average similarity used by ronhovde and nussinov @xcite ) than on properties of the individual partitions ( like the measures of occurrence used by arenas et al . @xcite and by lancichinetti et al . @xcite ) . the natural procedure to detect the hierarchical structure of a graph is hierarchical clustering , that we have discussed in section [ sec4_2 ] . there we have emphasized the main weakness of the procedure , which consists of the necessity to introduce a criterion to identify relevant partitions ( hierarchical levels ) out of the full dendrogram produced by the given algorithm . furthermore , there is no guarantee that the results indeed reflect the actual hierarchical structure of the graph , and that they are not mere artefacts of the algorithm itself . scholars have just started to deal with these problems . sales - pardo et al . have proposed a top - down approach @xcite . their method consists of two steps : 1 ) measuring the similarity between vertices ; 2 ) deriving the hierarchical structure of the graph from the similarity matrix . the similarity measure , named _ node affinity _ , is based on newman - girvan modularity . basically the affinity between two vertices is the frequency with which they coexist in the same community in partitions corresponding to local optima of modularity . the latter are configurations for which modularity is stable , i. e. it can not increase if one shifts one vertex from one cluster to another or by merging or splitting clusters . the set of these partitions is called @xmath672 . before proceeding with the next step , one verifies whether the graph has a significant community structure or not . this is done by calculating the @xmath406-score ( eq . [ eq : zscore ] ) for the average modularity of the partitions in @xmath672 with respect to the average modularity of partitions with local modularity optima of the equivalent ensemble of null model graphs , obtained as usual by randomly rewiring the edges of the original graph under the condition that the expected degree sequence is the same as the degree sequence of the graph . large @xmath406-scores indicate meaningful cluster structure : sales - pardo et al . used a threshold corresponding to the @xmath673 significance level - score has to be computed with respect to the actual distribution of the maximum modularity for the null model graphs , as the latter is not gaussian ( section [ sub_sec6_1 ] ) . ] . if the graph has a relevant cluster structure , one proceeds with the second step , which consists in putting the affinity matrix in a form as close as possible to block - diagonal , by minimizing a cost function expressing the average distance of connected vertices from the diagonal . the blocks correspond to the communities and the recovered partition represents the uppermost organization level . to determine lower levels , one iterates the procedure for each subgraph identified at the previous level , which is treated as an independent graph . the procedure stops when all blocks found do not have a relevant cluster structure , i. e. their @xmath406-scores are lower than the threshold . the partitions delivered by the method are hierarchical by construction , as communities at each level are nested within communities at higher levels . however , the method may find no relevant partition ( no community structure ) , a single partition ( community structure but no hierarchy ) or more ( hierarchy ) and in this respect it is better than most existing methods . the algorithm is not fast , as both the search of local optima for modularity and the rearrangement of the similarity matrix are performed with simulated annealing , but delivers good results for computer generated networks , and meaningful partitions for some real networks , like the world airport network @xcite , an email exchange network of a catalan university @xcite , a network of electronic circuits @xcite and metabolic networks of _ e. coli _ @xcite . clauset et al . @xcite described the hierarchical organization of a graph by introducing a class of _ hierarchical random graphs_. a hierarchical random graph is defined by a dendrogram @xmath674 , which is the natural representation of the hierarchy , and by a set of probabilities @xmath675 associated to the @xmath143 internal nodes of the dendrogram . ancestor _ of a vertex @xmath67 is any internal node of the dendrogram that is encountered by starting from the `` leaf '' vertex @xmath67 and going all the way up to the top of the dendrogram . the probability that vertices @xmath67 and @xmath68 are linked to each other is given by the probability @xmath676 of the lowest common ancestor of @xmath67 and @xmath68 . clauset et al . searched for the model @xmath677 that best fits the observed graph topology , by using bayesian inference ( section [ subsec_statinf1 ] ) . the probability that the model fits the graph is proportional to the likelihood @xmath678 here , @xmath679 is the number of edges connecting vertices whose lowest common ancestor is @xmath375 , @xmath680 and @xmath681 are the numbers of graph vertices in the left and right subtrees descending from the dendrogram node @xmath375 , and the product runs over all internal dendrogram nodes . for a given dendrogram @xmath674 , the maximum likelihood @xmath682 corresponds to the set of probabilities @xmath683 , where @xmath684 equals the actual density of edges @xmath685 between the two subtrees of @xmath375 ( fig . [ clausethier ] ) . hierarchical random graphs by clauset et al . the picture shows two possible dendrograms for the simple graph on the top . the linking probabilities on the internal nodes of the dendrograms yield the best fit of the model graphs to the graph at study . reprinted figure with permission from ref . 2008 by the nature publishing group . ] one can define the statistical ensemble of hierarchical random graphs describing a given graph @xmath23 , by assigning to each model graph @xmath686 a probability proportional to the maximum likelihood @xmath682 . the ensemble can be sampled by a markov chain monte carlo method @xcite . the procedure suggested by clauset et al . seems to converge to equilibrium roughly in a time @xmath129 , although the actual complexity may be much higher . still , the authors were able to investigate graphs with a few thousand vertices . from sufficiently large sets of model configurations sampled at equilibrium , one can compute average properties of the model , e. g. degree distributions , clustering coefficients . etc . , and compare them with the corresponding properties of the original graph . tests on real graphs reveal that the model is indeed capable to describe closely the graph properties . furthermore , the model enables one to predict missing connections between vertices of the original graph . this is a very important problem @xcite : edges of real graphs are the result of observations / experiments , that may fail to discover some relationships between the units of the system . from the ensemble of the hierarchical random graphs one can derive the average linking probability between all pairs of graph vertices . by ranking the probabilities corresponding to vertex pairs which are disconnected in the original graph , one may expect that the pairs with highest probabilities are likely to be connected in the system , even if such connections are not observed . clauset et al . pointed out that their method does not deliver a sharp hierarchical organization for a given graph , but a class of possible organizations , with well - defined probabilities . it is certainly reasonable to assume that many structures are compatible with a given graph topology . in the case of community structure , it is not clear which information one can extract from averaging over the ensemble of hierarchical random graphs . moreover , since the hierarchical structure is represented by a dendrogram , it is impossible to rank partitions according to their relevance . in fact , the work by clauset et al . questions the concept of `` relevant partition '' , and opens a debate in the scientific community about the meaning itself of graph clustering . the software of the method can be found at http://www.santafe.edu/@xmath215aaronc / hierarchy/. the analysis of dynamic communities is still in its infancy . studies in this direction have been mostly hindered by the fact that the problem of graph clustering is already controversial on single graph realizations , so it is understandable that most efforts still concentrate on the `` static '' version of the problem . another difficulty is represented by the dearth of timestamped data on real graphs . recently , several data sets have become available , enabling to monitor the evolution in time of real systems @xcite . so it has become possible to investigate how communities form , evolve and die . the main phenomena occurring in the lifetime of a community are ( fig . [ commevol ] ) : birth , growth , contraction , merger with other communities , split , death . possible scenarios in the evolution of communities . reprinted figure with permission from ref . 2007 by the nature publishing group . ] the first study was carried out by hopcroft et al . @xcite , who analyzed several snapshots of the citation graph induced by the nec citeseer database @xcite . the snapshots cover the period from @xmath687 to @xmath688 . communities are detected by means of ( agglomerative ) hierarchical clustering ( section [ sec4_2 ] ) , where the similarity between vertices is the _ cosine similarity _ of the vectors describing the corresponding papers , a well known measure used in information retrieval @xcite . in each snapshot hopcroft et al . identified the _ natural communities _ , defined as those communities of the hierarchical tree that are only slightly affected by minor perturbations of the graph , where the perturbation consists in removing a small fraction of the vertices ( and their edges ) . such natural communities are conceptually similar to the _ stable _ communities we will see in section [ sec6_4 ] . hopcroft et al . found the best matching natural communities across different snapshots , and in this way they could follow the history of communities . in particular they could see the emergence of new communities , corresponding to new research topics . the main drawback of the method comes from the use of hierarchical clustering , which is unable to sort out meaningful communities out of the hierarchical tree , which includes many different partitions of the graph . more recently , palla et al . performed a systematic analysis of dynamic communities @xcite . they studied two social systems : 1 ) a graph of phone calls between customers of a mobile phone company in a year s time ; 2 ) a collaboration network between scientists , describing the coauthorship of papers in condensed matter physics from the electronic e - print archive ( cond - mat ) maintained by cornell university library , spanning a period of 142 months . the first problem is identifying the image of a community @xmath689 at time @xmath690 among the communities of the graph at time @xmath156 . a simple criterion , used in other works , is to measure the relative overlap ( eq . [ eqt20 ] ) of @xmath689 with all communities at time @xmath156 , and pick the community which has the largest overlap with @xmath689 . this is intuitive , but in many cases it may miss the actual evolution of the community . for instance , if @xmath691 at time @xmath690 grows considerably and overlaps with another community @xmath692 ( which at the previous time step was disjoint from @xmath691 ) , the relative overlap between @xmath689 and @xmath693 may be larger than the relative overlap between @xmath689 and @xmath691 . it is not clear whether there is a general prescription to avoid this problem . palla et al . solved it by exploiting the features of the clique percolation method ( cpm ) ( section [ sec45_1 ] ) , that they used to detect communities . the idea is to analyze the graph @xmath694 , obtained by merging the two snapshots @xmath695 and @xmath696 of the evolving graph , at times @xmath156 and @xmath690 ( i. e. , by putting together all their vertices and edges ) . any cpm community of @xmath695 and @xmath696 does not get lost , as it is included within one of the cpm communities of @xmath694 . for each cpm community @xmath697 of @xmath694 , one finds the cpm communities @xmath698 and @xmath699 ( of @xmath695 and @xmath696 , respectively ) which are contained in @xmath697 . the image of any community in @xmath699 at time @xmath156 is the community of @xmath698 that has the largest relative overlap with it . the age @xmath253 of a community is the time since its birth . it turns out that the age of a community is positively correlated with its size @xmath700 , i. e. that older communities are also larger ( on average ) . the time evolution of a community @xmath22 can be described by means of the relative overlap @xmath701 between states of the community separated by a time @xmath156 : @xmath702 one finds that , in both data sets , @xmath701 decays faster for larger communities , so the composition of large communities is rather variable in time , whether small communities are essentially static . another important question is whether it is possible to predict the evolution of communities from information on their structure or on their vertices . in fig . [ evolpred]a the probability @xmath703 that a vertex will leave the community in the next step of the evolution is plotted as a function of the relative external strength of the vertex , indicating how much of the vertex strength lies on edges connecting it to vertices outside its community . the plot indicates that there is a clear positive correlation : vertices which are only loosely connected to vertices of their community have a higher chance ( on average ) to leave the community than vertices which are more `` committed '' towards the other community members . the same principle holds at the community level too . relation between structural features and evolution of a community . a ) relation between the probability that a vertex will abandon the community in the next time step and its relative external strength . b ) relation between the probability of disintegration of a community in the next time step and its relative external strength . reprinted figure with permission from ref . 2007 by the nature publishing group . ] fig . [ evolpred]b shows that the probability @xmath704 that a community will disintegrate in the next time step is positively correlated with the relative external strength of the community . finally , palla et al . have found that the probability for two communities to merge increases with the community sizes much more than what one expects from the size distribution , which is consistent with the faster dynamics observed for large communities . palla et al . analyzed two different real systems , a network of mobile phone communications and a coauthorship network , to be able to infer general properties of community evolution . however , communities were only found with the cpm , so their results need to be cross - checked by employing other clustering techniques . asur et al . @xcite explored the dynamic relationship between vertices and communities . communities were found with the mcl method by van dongen @xcite ( section [ sec44_2 ] ) , by analyzing the graph at different timestamps . asur et al . distinguished events involving communities and events involving the vertices . events involving communities are _ continue _ ( the community keeps most of its vertices in consecutive time steps ) , _ @xmath705-merge _ ( two clusters merge into another ) , _ @xmath705-split _ ( two clusters split in two parts ) , _ form _ ( no pair of vertices of the cluster at time @xmath690 were in the same cluster at time @xmath156 ) and _ dissolve _ ( opposite of _ form _ ) . events involving vertices are _ appear _ ( if a vertex joins the graph for the first time ) , _ disappear _ ( if a vertex of the graph at time @xmath156 is no longer there at time @xmath690 ) , _ join _ ( if a vertex of a cluster at time @xmath690 was not in that cluster at time @xmath156 ) and _ leave _ ( if a vertex which was in a cluster at time @xmath156 is not in that cluster at time @xmath690 ) . based on such events , four measures are defined in order to catch the behavioral tendencies of vertices contributing to the evolution of the graph : the _ stability index _ ( measuring the tendency of a vertex to interact with the same vertices over time ) , the _ sociability index _ ( measuring the number of different interactions of a vertex , basically the number of _ join _ and _ leave _ events ) , the _ popularity index _ ( measuring the number of vertices attracted by a cluster in a given time interval ) and the _ influence index _ ( measuring the influence a vertex has on the others , which is computed from the number of vertices that leave or join a cluster together with the vertex ) . applications on a coauthorship network of computer scientists and on a network of subjects for clinical trials show that the behavioral measures above enable one to make reliable predictions about the time evolution of such graphs ( including , e. g. , the inference of missing links @xcite ) . dynamic communities can be as well detected with methods of information compression , such as some of those we have seen in section [ subsec_statinf2 ] . sun et al . @xcite applied the minimum description length ( mdl ) principle @xcite to find the minimum encoding cost for the description of a time sequence of graphs and their partitions in communities . the method is quite similar to that successively developed by rosvall and bergstrom @xcite , which is however defined only for static graphs ( section [ subsec_statinf2 ] ) . here one considers bipartite graphs evolving in time . the time sequence of graphs can be separated in segments , each containing some number of consecutive snapshots of the system . the graphs of each segment are supposed to have the same modular structure ( i. e. they represent the same phase in the history of the system ) , so they are characterized by the same partition of the two vertex classes . for each graph segment it is possible to define an encoding cost , which combines the encoding cost of the partition of the graphs of the segment with the entropy of compression of the segment in the subgraph segments induced by the partition . the total encoding cost @xmath706 of the graph series is given by the sum of the encoding costs of its segments . minimizing @xmath706 enables one to find not only the most modular partition for each graph segment ( high modularity corresponds to low encoding costs for a partition ) , but also the most compact subdivision of the snapshots into segments , such that graphs in the same segment are strongly correlated with each other . the latter feature allows to identify _ change points _ in the time history of the system , i. e. short periods in which the dynamics produces big changes in the graph structure ( corresponding to , e.g. , extreme events ) . the minimization of @xmath706 is @xmath622-hard , so the authors propose an approximation method called _ graphscope _ , which consists of two steps : first , one looks for the best partition of each graph segment ; second , one looks for the best division in segments . in both cases the `` best '' result corresponds to the minimal encoding cost . the best partition within a graph segment is found by local search . graphscope has the big advantage not to require any input , like the number and sizes of the clusters . it is also suitable to operate in a streaming environment , in which new graph configurations are added in time , following the evolution of the system : the computational complexity required to process a snapshot ( on average ) is stable over time . tests on real evolving data sets show that graphscope is able to find meaningful communities and change points . since keeping track of communities in different time steps is not a trivial problem , as we have seen above , it is perhaps easier to adopt a vertex - centric perspective , in which one monitors the community of a given vertex at different times . for any method , given a vertex @xmath67 and a time @xmath156 , the community to which @xmath67 belongs at time @xmath156 is well defined . fenn et al . @xcite used the multiresolution method by reichardt et al . @xcite ( section [ sub_sec6_01 ] ) and investigated a fully connected graph with time - dependent weights , representing the correlations of time series of hourly exchange rate returns . the resolution parameter @xmath371 is fixed to the value that occurs in most stability plateaus of the system at different time steps . motivated by the work of guimer and amaral @xcite ( section [ sec7_1 ] ) , fenn et al . identify the role of individual vertices in their community through the pair @xmath707 , where @xmath708 is the @xmath406-score of the internal strength ( weighted degree , section [ sec1_1 ] ) , defined in eq . [ eqtrol1 ] , and @xmath709 the @xmath406-score of the site betweenness , defined by replacing the internal degree with the site betweenness of freeman @xcite in eq . [ eqtrol1 ] . we remind that the site betweenness is a measure of the number of shortest paths running through a vertex . the variable @xmath709 expresses the importance of a vertex in processes of information diffusion with respect to the other members of its community . another important issue regards the _ persistence _ of communities in time , i. e. how stable they are during the evolution . as a measure of persistence , fenn et al . introduced a vertex - centric version of the relative overlap of eq . [ eqtrol3 ] @xmath710 where @xmath67 is the vertex and @xmath711 , @xmath712 the communities of @xmath67 at times @xmath156 , @xmath713 , respectively . the decay of @xmath714 depends on the type of vertex . in particular , if the vertex is strongly connected to its community ( @xmath708 large ) , @xmath714 decays quite slowly , meaning that it tends to stay attached to a stable core of vertices . the methods described above are basically two - stage approaches , in which clusters are detected at each timestamp of the graph evolution , independently of the results at different times , and relationships between partitions at different times are inferred successively . however , this often produces significant variations between partitions close in time , especially when , as it usually happens , the datasets are noisy . in this case one would be forced to introduce _ ad hoc _ hypotheses on the graph evolution to justify the variability of the modular structure , whereas such variability is mostly an artefact of the approach . it would be desirable instead to have a unified framework , in which clusters are deduced both from the current structure of the graph and from the knowledge of the cluster structure at previous times . this is the principle of _ evolutionary clustering _ , a framework introduced by chakrabarti et al . @xcite and successively adopted and refined by other authors . let @xmath715 be the partition of the graph at time @xmath156 . the _ snapshot quality _ of @xmath715 measures the goodness of the partition with respect to the graph structure at time @xmath156 . the _ history cost _ is a measure of the distance / dissimilarity of @xmath715 with respect to the partition @xmath716 at the previous time step . the overall quality of @xmath715 is given by a combination of the snapshot quality and the history cost at each time step . ideally , a good partition should have high snapshot quality ( i. e. it should cluster well the data at time @xmath156 ) and low history cost ( i. e. it should be similar to the partition at the previous time step ) . in order to find @xmath715 from @xmath716 and the relational data at time @xmath156 chakrabarti et al . suggested to minimize the difference between the snapshot quality and the history cost , with a relative weight @xmath717 that is a tunable parameter . the input of the procedure consists in the sequence of adjacency / similarity matrices at different time steps . in practice , one could use modified versions of such matrices , obtained by performing ( weighted ) averages of the data over some time window , in order to make the relational data more robust against noise and the results of the clustering procedure more reliable . one can adopt arbitrary measures to compute the snapshot quality and the historical cost . besides , several known clustering techniques used for static graphs can be reformulated within this evolutionary framework . chakrabarti et al . derived evolutionary versions of hierarchical clustering ( section [ sec4_2 ] ) and @xmath44-means clustering ( section [ sec4_3 ] ) , whereas chi et al . @xcite designed two implementations for spectral clustering ( section [ sec4_4 ] ) . based on evolutionary clustering , lin et al . @xcite introduced a framework , called _ facetnet _ , that allows vertices to belong to more communities at the same time . here the snapshot cost is the kullback - leibler ( kl ) divergence @xcite between the adjacency / similarity matrix at time @xmath156 and the matrix describing the community structure of the graph at time @xmath156 ; the history cost is the kl divergence between the matrices describing the community structure of the graph at times @xmath718 and @xmath156 . facetnet can be extended to handle adding and removing of vertices as well as variations of the number of clusters in consecutive time steps . however , it is not able to account for the creation and the disintegration of communities and not scalable to large systems due to the high number of iterations necessary for the matrix computations to reach convergence . these issues have been addressed in a recent approach by kim and han @xcite . naturally , what one hopes to achieve at the end of the day is to see how real groups form and evolve in time . backstrom et al . @xcite have carried out an analysis of group dynamics in the free online community of livejournal ( http://www.livejournal.com/ ) and in a coauthorship network of computer scientists . here the groups are identified through the declared memberships of users ( for livejournal ) and conferences attended by computer scientists , respectively . backstrom and coworkers have found that the probability that an individual joins a community grows with the number of friends / coauthors who are already in the community and ( for livejournal ) with their degree of interconnectedness . moreover , the probability of growth of livejournal communities is positively correlated to a combination of factors including the community size , the number of friends of community members which are not in the community and the ratio of these two numbers . a high density of triads within a community appears instead to hinder its growth . given a network , many partitions could represent meaningful clusterings in some sense , and it could be difficult for some methods to discriminate between them . quality functions evaluate the goodness of a partition ( section [ sec3_2_2 ] ) , so one could say that high quality corresponds to meaningful partitions . but this is not necessarily true . in section [ sub_sec6_1 ] we have seen that high values of the modularity of newman and girvan do not necessarily indicate that a graph has a definite cluster structure . in particular we have seen that partitions of random graphs may also achieve considerably large values of @xmath90 , although we do not expect them to have community structure , due to the lack of correlations between the linking probabilities of the vertices . the optimization of quality functions , like modularity , delivers the best partition according to the criterion underlying the quality function . but is the optimal clustering also _ significant _ , i. e. a relevant feature of the graph , or is it just a byproduct of randomness and basic structural properties like , e. g. , the degree sequence ? little effort has been devoted to this crucial issue , that we discuss here . in some works the concept of significance has been related to that of _ robustness _ or _ stability _ of a partition against random perturbations of the graph structure . the basic idea is that , if a partition is significant , it will be recovered even if the structure of the graph is modified , as long as the modification is not too extensive . instead , if a partition is not significant , one expects that minimal modifications of the graph will suffice to disrupt the partition , so other clusterings are recovered . a nice feature of this approach is the fact that it can be applied for any clustering technique . gfeller et al . @xcite considered the general case of weighted graphs . a graph is modified , in that its edge weights are increased or decreased by a relative amount @xmath719 . this choice also allows to account for the possible effects of uncertainties in the values of the edge weights , resulting from measurements / experiments carried out on a given system . after fixing @xmath720 ( usually to @xmath721 ) , multiple realizations of the original graph are generated . the best partition for each realization is identified and , for each pair of adjacent vertices @xmath67 and @xmath68 , the _ in - cluster _ probability @xmath362 is computed , i. e. the fraction of realizations in which @xmath67 and @xmath68 were classified in the same cluster . edges with in - cluster probability smaller than a threshold @xmath722 ( usually @xmath723 ) are called _ external edges_. the stability of a partition is estimated through the _ clustering entropy _ @xmath724 , \label{sign1}\ ] ] where @xmath14 is , as usual , the number of graph edges , and the sum runs over all edges . the most stable partition has @xmath725 along inter - cluster edges and @xmath726 along intra - cluster edges , which yields @xmath727 ; the most unstable partition has @xmath728 on all edges , yielding @xmath729 . the absolute value of @xmath87 is not meaningful , though , and needs to be compared with the corresponding value for a null model graph , similar to the original graph , but with supposedly no cluster structure . gfeller et al . adopted the same null model of newman - girvan modularity , i. e. the class of graphs with expected degree sequence coinciding with that of the original graph . since the null model is defined on unweighted graphs , the significance of @xmath87 can be assessed only in this case , although it would not be hard to think of a generalization to weighted graphs . the approach enables one as well to identify _ unstable vertices _ , i. e. vertices lying at the boundary between clusters . in order to do that , the external edges are removed and the connected components of the resulting disconnected graph are associated with the clusters detected in the original graph , based on their relative overlap ( computed through eq . [ eqt20 ] ) . unstable vertices end up in components that are not associated to any of the initial clusters . a weakness of the method by gfeller et al . is represented by the two parameters @xmath720 and @xmath722 , whose values are in principle arbitrary . more recently , karrer et al . @xcite adopted a similar strategy to unweighted graphs . here one performs a sweep over all edges : the perturbation consists in removing each edge with a probability @xmath18 and replacing it with another edge between a pair of vertices @xmath730 , chosen at random with probability @xmath369 , where @xmath105 and @xmath106 are the degrees of @xmath67 and @xmath68 . we recognize the probability of the null model of newman - girvan modularity . indeed , by varying the probability @xmath18 from @xmath631 to @xmath2 one smoothly interpolates between the original graph ( no perturbation ) and the null model ( maximal perturbation ) . the degree sequence of the graph remains invariant ( on average ) along the whole process , by construction . the idea is that the perturbation affects solely the organization of the vertices , keeping the basic structural properties . for a given value of @xmath18 , many realizations of the perturbed graph are generated , their cluster structures are identified with some method ( karrer et al . used modularity optimization ) and compared with the partition obtained from the original unperturbed graph . the partitions are compared by computing the variation of information @xmath638 ( section [ sec6_2 ] ) . from the plot of the average @xmath731 versus @xmath18 one can assess the stability of the cluster structure of the graph . if @xmath732 changes rapidly for small values of @xmath18 the partition is likely to be unstable . as in the approach by gfeller et al . the behaviour of the function @xmath732 does not have an absolute meaning , but needs to be compared with the corresponding curve obtained for a null model . for consistency , the natural choice is again the null model of modularity , which is already used in the process of graph perturbation . the approaches by gfeller et al . and karrer et al . , with suitable modifications , can also be used to check for the stability of the cluster structure in parts of a graph , up to the level of individual communities . this is potentially important as it may happen that some parts of the graph display a strong community structure and other parts weak or no community structure at all . rosvall and bergstrom @xcite defined the significance of clusters with the bootstrap method @xcite , which is a standard procedure to check for the accuracy of a measurement / estimate based on resampling from the empirical data . the graph at study is supposed to be generated by a parametric model , which is used to create many samples . this is done by assigning to each edge a weight taken by a poisson distribution with mean equal to the original edge weight . for the initial graph and each sample one identifies the community structure with some method , that can be arbitrary . for each cluster of the partition of the original graph one determines the largest subset of vertices that are classified in the same cluster in at least @xmath733 of all bootstrap samples . identifying such cluster cores enables one to track the evolution in time of the community structure , as we explained in section [ sec7_1_2 ] . a different approach has been proposed by massen and doye @xcite . they analyzed an equilibrium canonical ensemble of partitions , with @xmath734 playing the role of the energy , @xmath90 being newman - girvan modularity . this means that the probability of occurrence of a partition at temperature @xmath467 is proportional to @xmath735 . the idea is that , if a graph has a significant cluster structure , at low temperatures one would recover essentially the same partition , corresponding to the modularity maximum , which is separated by an appreciable gap from the modularity values of the other partitions . on the contrary , graphs with no community structure , e. g. random graphs , have many competing ( local ) maxima , and the corresponding configurations will emerge already at low temperatures , since their modularity values are close to the absolute maximum , good et al . @xcite have actually shown that the modularity landscape has a huge degeneracy of states with high modularity values , close to the global maximum , especially on graphs with community structure . so the results of the method by massen and doye may be misleading . ] . these distinct behaviors can manifest themselves in various ways . for instance , if one considers the variation of the specific heat @xmath736 with @xmath467 , the gap in the modularity landscape is associated to a sharp peak of @xmath706 around some temperature value , like it happens in a phase transition . if the gap is small and there are many partitions with similar modularity values , the peak of @xmath706 becomes broad . another strategy to assess the significance of the maximum modularity partition consists of the investigation of the similarity between partitions recovered at a given temperature @xmath467 . this similarity can be expressed by the _ frequency matrix _ , whose element @xmath737 indicates the relative number of times vertices @xmath67 and @xmath68 have been classified in the same cluster . if the graph has a clear community structure , at low temperatures the frequency matrix can be put in block - diagonal form , with the blocks corresponding to the communities of the best partition ; if there is no significant community structure , the frequency matrix is rather homogeneous . the fiedler eigenvalue @xcite @xmath144 , the second smallest eigenvalue of the laplacian matrix associated to the frequency matrix , allows to estimate how `` block - diagonal '' the matrix is ( see section [ sec4_1 ] ) . at low temperatures @xmath738 if there is one ( a few ) partitions with maximum or near to maximum modularity ; if there are many ( almost ) degenerate partitions , @xmath144 is appreciably different from zero even when @xmath739 . a sharp transition from low to high values of @xmath144 by varying temperature indicates significant community structure . another clear signature of significant community structure is the observation of a rapid drop of the average community size with @xmath467 , as `` strong '' communities break up in many small pieces for a modest temperature increase , while the disintegration of `` weak '' communities takes place more slowly . in scale - free graphs ( section [ sec1_3 ] ) clusters are often not well separated , due to the presence of the hubs ; in these cases the above - mentioned transitions of ensemble variables are not so sharp and take place over a broader temperature range . the canonical ensemble of partitions is generated through single spin heatbath simulated annealing @xcite , combined with parallel tempering @xcite . the approach by massen and doye could be useful to recognize graphs without cluster structure , if the modularity landscape is characterized by many maxima with close values ( but see footnote ) . however , it can happen that gaps between the absolute modularity maximum and the rest of the modularity values are created by fluctuations , and the method is unable to identify these situations . furthermore , the approach heavily relies on modularity and on a costly technique like simulated annealing : extensions to other quality functions and/or optimization procedures do not appear straightforward . in a recent work by bianconi et al . @xcite the notion of entropy of graph ensembles @xcite is employed to find out how likely it is for a cluster structure to occur on a graph with a given degree sequence . the entropy is computed from the number of graph configurations which are compatible with a given classification of the vertices in @xmath91 groups . the clustering is quantitatively described by fixing the number of edges @xmath740 running between clusters @xmath562 and @xmath563 , for all choices of @xmath741 . bianconi et al . proposed the following indicator of clustering significance @xmath742 where @xmath743 is the entropy of the graph configurations with given degree sequence @xmath744 and clustering @xmath745 ( with fixed numbers of inter - cluster edges @xmath740 ) , and @xmath746 is the average entropy of the configurations with the same degree sequence and a random permutation @xmath747 of the cluster labels . the absolute value of the entropy @xmath743 is not meaningful , so the comparison of @xmath743 and @xmath746 is crucial , as it tells how relevant the actual cluster structure is with respect to a random classification of the vertices . however , different permutations of the assignments @xmath745 yield different values of the entropy , which can fluctuate considerably . therefore one has to compute the standard deviation @xmath748 of the entropy corresponding to all random permutations @xmath747 , to estimate how significant the difference between @xmath743 and @xmath746 is . in this way , if @xmath749 , the entropy of the given cluster structure is of the same order as the entropy of some random permutation of the cluster labels , so it is not relevant . instead , if @xmath750 , the cluster structure is far more likely than a random classification of the vertices , so the clustering is relevant . the indicator @xmath751 can be simply generalized to the case of directed and weighted graphs . lancichinetti et al . @xcite as well addressed the issue by comparing the cluster structure of the graph with that of a random graph with similar properties . an important novelty of this approach is the fact that it estimates the significance of single communities , not of partitions . in fact , not all communities are equally significant , in general , so it makes a lot of sense to check them individually . in particular , it may happen that real networks are not fully modular , due to their particular history or generating mechanisms , and that only portions of them display community structure . the main idea is to verify how likely it is that a community @xmath22 is a subgraph of a random graph with the same degree sequence of the original graph . this likelihood is called @xmath22-score , and is computed by examining the vertex @xmath752 of @xmath22 , with the lowest internal degree @xmath753 in @xmath22 ( the `` worst '' vertex ) . the @xmath22-score is defined as the probability that the internal degree of the worst vertex in the corresponding community of the null model graphs is larger than or equal to @xmath753 . this probability is computed by using tools from extreme and order statistics @xcite . a low value of the @xmath22-score ( @xmath754 ) is a strong indication that the group of vertices at study is a community and not the product of random fluctuations . in addition , the measure can be used to check whether a subgraph has an internal modular structure . for that , one removes the vertices of the subgraph one at a time , starting from the worst and proceeding in increasing order of the internal degree , and observes how the @xmath22-score varies at each vertex removal : sharp drops indicate the presence of dense subgraphs ( fig . [ cscore ] ) . therefore , one could think of using the @xmath22-score as ingredient of new clustering techniques . as we have seen , the @xmath22-score is based on the behavior of the vertex with lowest internal degree of the subgraph . real networks are characterized by noise , which could strongly affect the structural relationships between vertices and clusters . for this reason , relying on the properties of a single vertex to evaluate the significance of a subgraph could be a problem for applications to real networks . lancichinetti et al . have shown that the @xmath22-score can be easily extended to consider the @xmath156 vertices with lowest internal degree , with @xmath755 ( @xmath259-score ) . the main limit of the @xmath22-score is the fact that its null model is the same as that of newman - girvan modularity . according to this null model , each vertex can in principle be connected to any other , no matter how large the system is . this is however not realistic , especially for large graphs , where it is much more reasonable to assume that each vertex has its own `` horizon '' , i.e. a subset of other vertices with which it can interact , which is usually much smaller than the whole system ( see section [ sub_sec6_1 ] ) . how to define such `` horizons '' and , more in general , realistic null models is still an open problem . however , the @xmath22-score could be easily reformulated with any null model , so one could readily derive more reliable definitions . we conclude with a general issue which is related to the significance of community structure . the question is : given a cluster structure in a graph , can it be recovered _ a priori _ by an algorithm ? in a recent paper @xcite , reichardt and leone studied under which conditions a special built - in cluster structure can be recovered . the clusters have equal size and a pair of vertices is connected with probability @xmath45 if they belong to the same cluster , with probability @xmath756 otherwise . in computer science this is known as the _ planted partitioning problem _ the goal is to propose algorithms that recover the planted partition for any choice of @xmath45 and @xmath375 . for dense graphs , i. e. graphs whose average degree grows with the number @xmath15 of vertices , algorithms can be designed that find the solution with a probability which equals @xmath2 minus a term that vanishes in the limit of infinite graph size , regardless of the difference @xmath757 , which can then be chosen arbitrarily small . since many real networks are not dense graphs , as their average degree @xmath758 is usually much smaller than @xmath15 and does not depend on it , reichardt and leone investigated the problem in the case of fixed @xmath758 and infinite graph size . we indicate with @xmath91 the number of clusters and with @xmath511 the probability that a randomly selected edge of the graph lies within any of the @xmath91 clusters . in this way , if @xmath759 , the inter - cluster edge density matches the intra - cluster edge density ( i. e. @xmath760 ) , and the planted partition would not correspond to a recoverable clustering , whereas for @xmath761 , there are no inter - cluster edges and the partition can be trivially recovered . the value of @xmath511 is in principle unknown , so one has to detect the cluster structure ignoring this information . reichardt and leone proposed to look for a minimum cut partition , i. e. for the partition that minimizes the number of inter - cluster edges , as it is usually done in the graph partitioning problem ( discussed in section [ sec4_1 ] ) . clearly , for @xmath761 the minimum cut partition trivially coincides with the planted partition , whereas for @xmath762 there should be some overlap , which is expected to vanish in the limit case @xmath759 . the minimum cut partition corresponds to the minimum of the following ferromagnetic potts model hamiltonian @xmath763 over the set of all spin configurations with zero magnetization . here the spin @xmath357 indicates the cluster vertex @xmath67 belongs to , and the coupling matrix @xmath363 is just the adjacency matrix of the graph . the constraint of zero magnetization ensures that the clusters have all the same size , as required by the planted partitioning problem . the energy of a spin configuration , expressed by eq . [ eq : mod12 ] , is the negative of the number of edges that lie within clusters : the minimum energy corresponds to the maximum number of intra - cluster edges , which is coupled to the minimum number of inter - cluster edges . the minimum energy can be computed with the cavity method , or belief propagation , at zero temperature @xcite . the accuracy of the solution with respect to the planted partition is expressed by the fraction of vertices which are put in the same class in both partitions . the analysis yields a striking result : the planted clustering is accurately recovered for @xmath511 larger than a critical threshold @xmath764 . so , there is a range of values of @xmath511 , @xmath765 , in which the clustering is not recoverable , as the minimum cut partition is uncorrelated with it . the threshold @xmath766 depends on the degree distribution @xmath767 of the graph . when a clustering algorithm is designed , it is necessary to test its performance , and compare it with that of other methods . in the previous sections we have said very little about the performance of the algorithms , other than their computational complexity . indeed , the issue of testing algorithms has received very little attention in the literature on graph clustering . this is a serious limit of the field . because of that , it is still impossible to state which method ( or subset of methods ) is the most reliable in applications , and people rely blindly on some algorithms instead of others for reasons that have nothing to do with the actual performance of the algorithms , like . e.g. popularity ( of the method or of its inventor ) . this lack of control is also the main reason for the proliferation of graph clustering techniques in the last few years . virtually in any paper , where a new method is introduced , the part about testing consists in applying the method to a small set of simple benchmark graphs , whose cluster structure is fairly easy to recover . because of that , the freedom in the design of a clustering algorithm is basically infinite , whereas it is not clear what a new procedure is adding to the field , if anything . in this section we discuss at length the issue of testing . first , we describe the fundamental ingredients of any testing procedure , i. e. benchmark graphs with built - in community structure , that methods have to identify ( section [ sec6_1 ] ) . we proceed by reviewing measures to compare graph partitions with each other ( section [ sec6_2 ] ) . in section [ sec6_3 ] we present the comparative evaluations of different methods that have been performed in the literature . testing an algorithm essentially means applying it to a specific problem whose solution is known and comparing such solution with that delivered by the algorithm . in the case of graph clustering , a problem with a well - defined solution is a graph with a clear community structure . this concept is not trivial , however . many clustering algorithms are based on similar intuitive notions of what a community is , but different implementations . so it is crucial that the scientific community agrees on a set of reliable benchmark graphs . this mostly applies to computer - generated graphs , where the built - in cluster structure can be arbitrarily designed . in the literature real networks are used as well , in those cases in which communities are well defined because of information about the system . we start our survey from computer - generated benchmarks . a special class of graphs has become quite popular in the last years . they are generated with the so - called _ planted @xmath293-partition model _ the model partitions a graph with @xmath768 vertices in @xmath293 groups with @xmath126 vertices each . vertices of the same group are linked with a probability @xmath511 , whereas vertices of different groups are linked with a probability @xmath512 . each subgraph corresponding to a group is then a random graph la erds - rnyi with connection probability @xmath769 ( section [ sec1_3 ] ) . the average degree of a vertex is @xmath770 . if @xmath513 the intra - cluster edge density exceeds the inter - cluster edge density and the graph has a community structure . this idea is quite intuitive and we have encountered it in several occasions in the previous sections . girvan and newman considered a special case of the planted @xmath293-partition model @xcite . they set @xmath771 , @xmath772 ( so the number of graph vertices is @xmath773 ) and fixed the average total degree @xmath420 to @xmath774 . this implies that @xmath775 , so the probabilities @xmath511 and @xmath512 are not independent parameters . in calculation it is common to use as parameters @xmath776 and @xmath777 , indicating the expected internal and external degree of a vertex , respectively . these particular graphs have by now gained the status of standard benchmarks @xcite ( fig . [ figure11 ] ) . in the first applications of the graphs one assumed that communities are well defined when @xmath778 , corresponding to the situation in which the internal degree exceeds the external degree . however , the threshold @xmath779 implies @xmath780 and @xmath781 , so it is not the actual threshold of the model , where @xmath782 , corresponding to @xmath783 . so , one expects to be able to detect the planted partition up until @xmath783 . testing a method against the girvan - newman benchmark consists in calculating the similarity between the partitions determined by the method and the natural partition of the graph in the four equal - sized groups . several measures of partitions similarity may be adopted ; we describe them in section [ sec6_2 ] . one usually builds many graph realizations for a particular value of @xmath784 and computes the average similarity between the solutions of the method and the built - in solution . the procedure is then iterated for different values of @xmath784 . the results are usually represented in a plot , where the average similarity is drawn as a function of @xmath784 . most algorithms usually do a good job for small @xmath784 and start to fail when @xmath784 approaches @xmath228 . fan et al . @xcite have designed a weighted version of the benchmark of girvan and newman , in that one gives different weights to edges inside and between communities . one could pick just two values , one for intra- and the other for inter - community edges , or uniformly distributed values in two different ranges . for this benchmark there are then two parameters that can be varied : @xmath784 and the relative importance of the internal and the external weights . typically one fixes the topological structure and varies the weights . this is particularly insightful when @xmath785 , which delivers graphs without topological cluster structure : in this case , the question whether there are clusters or not depends entirely on the weights . as we have remarked above , the planted @xmath293-partition model generates mutually interconnected random graphs la erds - rnyi . therefore , all vertices have approximately the same degree . moreover , all communities have exactly the same size by construction . these two features are at odds with what is observed in graph representations of real systems . degree distributions are usually skewed , with many vertices with low degree coexisting with a few vertices with high degree . a similar heterogeneity is also observed in the distribution of cluster sizes , as we shall see in section [ sec7_1 ] . so , the planted @xmath293-partition model is not a good description of a real graph with community structure . however , the model can be modified to account for the heterogeneity of degrees and community sizes . a modified version of the model , called _ gaussian random partition generator _ , was designed by brandes et al . @xcite . here the cluster sizes have a gaussian distribution , so they are not the same , although they do not differ much from each other . the heterogeneity of the cluster sizes introduces a heterogeneity in the degree distribution as well , as the expected degree of a vertex depends on the number of vertices of its cluster . still , the variability of degree and cluster size is not appreciable . besides , vertices of the same cluster keep having approximately the same degree . a better job in this direction has been recently done by lancichinetti et al . ( lfr benchmark ) @xcite . they assume that the distributions of degree and community size are power laws , with exponents @xmath786 and @xmath787 , respectively . each vertex shares a fraction @xmath788 of its edges with the other vertices of its community and a fraction @xmath789 with the vertices of the other communities ; @xmath790 is the _ mixing parameter_. the graphs are built as follows : 1 . a sequence of community sizes obeying the prescribed power law distribution is extracted . this is done by picking random numbers from a power law distribution with exponent @xmath787 . each vertex @xmath67 of a community receives an internal degree @xmath791 , where @xmath105 is the degree of vertex @xmath67 , which is taken by a power law distribution with exponent @xmath786 . in this way , each vertex @xmath67 has a number of stubs @xmath791 . all stubs of vertices of the same community are randomly attached to each other , until no more stubs are `` free '' . in this way the sequence of internal degrees of each vertex in its community is maintained . each vertex @xmath67 receives now an additional number of stubs , equal to @xmath792 ( so that the final degree of the vertex is @xmath105 ) , that are randomly attached to vertices of different communities , until no more stub is `` free '' . numerical tests show that this procedure has a complexity @xmath21 , where @xmath14 is as usual the number of edges of the graph , so it can be used to create graphs of sizes spanning several orders of magnitude . [ fignewbench ] shows an example of a lfr benchmark graph . recently the lfr benchmark has been extended to directed and weighted graphs with overlapping communities @xcite . the software to create the lfr benchmark graphs can be freely downloaded at http://santo.fortunato.googlepages.com/inthepre + ss2 . a class of benchmark graphs with power law degree distributions had been previously introduced by bagrow @xcite . the construction process starts from a graph with a power - law degree distribution . bagrow used barabsi - albert scale free graphs @xcite . then , vertices are randomly assigned to one of four equally - sized communities . finally , pairs of edges between two communities are rewired so that either edge ends up within the same community , without altering the degree sequence ( on average ) . this is straightforward : suppose that the edges join the vertex pairs @xmath793 , @xmath794 and @xmath795 , @xmath796 , where @xmath793 , @xmath795 belong to community @xmath9 and @xmath794 , @xmath796 to community @xmath10 . it the edges are replaced by @xmath793-@xmath795 and @xmath794-@xmath796 ( provided they do not exist already ) , all vertices keep their degrees . with this rewiring procedure one can arbitrarily vary the edge density within and , consequently , between clusters . in this class of benchmarks , however , communities are all of the same size by construction , although one can in principle relax this condition . a ( seemingly ) different benchmark is represented by the class of _ relaxed caveman graphs _ , which were originally introduced to explain the clustering properties of social networks @xcite . the starting point is a set of disconnected cliques . with some probability edges are rewired to link different cliques . such model graphs are interesting as they are smooth variations of the ideal graph with `` perfect '' communities , i. e. disconnected cliques . on the other hand the model is equivalent to the planted @xmath293-partition model , where @xmath797 and @xmath512 is proportional to @xmath45 , with coefficient depending on the size of the clusters . benchmark graphs have also been introduced to deal with special types of graphs and/or cluster structures . for instance , arenas et al . @xcite have introduced a class of benchmark graphs with embedded hierarchical structure , which extends the class of graphs by girvan and newman . here there are @xmath798 vertices and two hierarchical levels , corresponding to a partition in @xmath774 groups ( microcommunities ) with @xmath774 vertices and a partition in @xmath223 larger groups of @xmath799 vertices ( macrocommunities ) , comprising each @xmath223 of the smaller groups . the edge densities within and between the clusters are indicated by three parameters @xmath800 , @xmath801 and @xmath784 : @xmath800 is the expected internal degree of a vertex within its microcommunity ; @xmath801 is the expected number of edges that the vertex shares with the vertices of the other microcommunities within its macrocommunity ; @xmath784 is the expected number of edges connecting the vertex with vertices of the other three macrocommunities . the average degree @xmath802 of a vertex is fixed to @xmath12 . [ fig8 ] shows an example of hierarchical graph constructed based on the same principle , with @xmath803 vertices and an average degree of @xmath11 . guimer et al . @xcite have proposed a model of bipartite graphs with built - in communities . they considered a bipartite graph of actors and teams , here we describe how to build the benchmarks for general bipartite graphs . one starts from a bipartite graph whose vertex classes @xmath9 and @xmath10 are partitioned into @xmath115 groups , @xmath804 and @xmath805 ( @xmath806 ) . each cluster @xmath807 comprises all vertices of the subgroups @xmath804 and @xmath805 , respectively . with probability @xmath45 edges are placed between vertices of subgroups @xmath804 and @xmath805 ( @xmath806 ) , i. e. within clusters . with probability @xmath808 , edges are placed between vertices of subgroups @xmath804 and @xmath809 , where @xmath67 and @xmath68 are chosen at random , so they can be equal or different . by construction , a non - zero value of the probability @xmath45 indicates a preference by vertices to share links with vertices of the same cluster , i. e. for @xmath810 the graph has a built - in community structure . for @xmath811 there would be edges only within clusters , i. e. the graph has a perfect cluster structure . finally , sawardecker et al . introduced a general model , that accounts for the possibility that clusters overlap @xcite . the model is based on the reasonable assumption that the probability @xmath362 that two vertices are connected by an edge grows with the number @xmath210 of communities both vertices belong to . for vertices in different clusters , @xmath812 , if they are in the same cluster ( and only in that one ) @xmath813 , if they belong to the same two clusters @xmath814 , etc .. by hypothesis , @xmath815 . the planted @xmath293-partition model is recovered when @xmath816 . as we have seen , nearly all existing benchmark graphs are inspired by the planted @xmath293-partition model , to some extent . however , the model needs to be refined to provide a good description of real graphs with community structure . the hypothesis that the linking probabilities of each vertex with the vertices of its community or of the other communities are constant is not realistic . it is more plausible that each pair of vertices @xmath67 and @xmath68 has its own linking probability @xmath362 , and that such probabilities are correlated for vertices in the same cluster . tests on real networks usually focus on a very limited number of examples , for which one has precise information about the vertices and their properties . in section [ sec2 ] we have introduced two popular real networks with known community structure , i. e. the social network of zachary s karate club and the social network of bottlenose dolphins living in doubtful sound ( new zealand ) , studied by lusseau . here , the question is whether the actual separation in two social groups could be predicted from the graph topology . zachary s karate club is by far the most investigated system . several algorithms are actually able to identify the two classes , modulo a few intermediate vertices , which may be misclassified . other methods are less successful : for instance , the maximum of newman - girvan modularity corresponds to a split of the network in four groups @xcite . another well known example is the network of american college football teams , derived by girvan and newman @xcite . there are @xmath817 vertices , representing the teams , and two vertices are connected if their teams play against each other . the teams are divided into @xmath818 conferences . games between teams in the same conference are more frequent than games between teams of different conferences , so one has a natural partition where the communities correspond to the conferences . when dealing with real networks , it is useful to resolve their community structure with different clustering techniques , to cross - check the results and make sure that they are consistent with each other , as in some cases the answer may strongly depend on the specific algorithm adopted . however , one has to keep in mind that there is no guarantee that `` reasonable '' communities , defined on the basis of non - structural information , must coincide with those detected by methods based only on the graph structure . checking the performance of an algorithm involves defining a criterion to establish how `` similar '' the partition delivered by the algorithm is to the partition one wishes to recover . several measures for the similarity of partitions exist . in this section we present and discuss the most popular measures . a thorough introduction of similarity measures for graph partitions has been given by meil @xcite and we will follow it closely . let us consider two generic partitions @xmath819 and @xmath820 of a graph @xmath313 , with @xmath821 and @xmath822 clusters , respectively . we indicate with @xmath15 the number of graph vertices , with @xmath823 and @xmath824 the number of vertices in clusters @xmath825 and @xmath826 and with @xmath827 the number of vertices shared by clusters @xmath825 and @xmath826 : @xmath828 . in the first tests using the benchmark graphs by girvan and newman ( section [ sec6_1 ] ) scholars used a measure proposed by girvan and newman themselves , the _ fraction of correctly classified vertices_. a vertex is correctly classified if it is in the same cluster with at least half of its `` natural '' partners . if the partition found by the algorithm has clusters given by the merging of two or more natural groups , all vertices of the cluster are considered incorrectly classified . the number of correctly classified vertices is then divided by the total size of the graph , to yield a number between @xmath631 and @xmath2 . the recipe to label vertices as correctly or incorrectly classified is somewhat arbitrary , though . apart from the fraction of correctly classified vertices , which is somewhat _ ad hoc _ and distinguishes the roles of the natural partition and of the algorithm s partition , most similarity measures can be divided in three categories : measures based on _ pair counting _ , _ cluster matching _ and _ information theory_. measures based on pair counting depend on the number of pairs of vertices which are classified in the same ( different ) clusters in the two partitions . in particular @xmath829 indicates the number of pairs of vertices which are in the same community in both partitions , @xmath830 ( @xmath831 ) the number of pairs of elements which are put in the same community in @xmath832 ( @xmath833 ) and in different communities in @xmath833 ( @xmath832 ) and @xmath834 the number of pairs of vertices that are in different communities in both partitions . wallace @xcite proposed the two indices @xmath835 @xmath836 and @xmath837 represent the probability that vertex pairs in the same cluster of @xmath832 are also in the same cluster for @xmath833 , and viceversa . these indices are asymmetrical , as the role of the two partitions is not the same . fowlkes and mallows @xcite suggested to use the geometric mean of @xmath836 and @xmath837 , which is symmetric . the _ rand index _ @xcite is the ratio of the number of vertex pairs correctly classified in both partitions ( i. e. either in the same or in different clusters ) , by the total number of pairs @xmath838 a measure equivalent to the rand index is the _ mirkin metric _ @xcite @xmath839 . \label{eqt03}\ ] ] the _ jaccard index _ is the ratio of the number of vertex pairs classified in the same cluster in both partitions , by the number of vertex pairs which are classified in the same cluster in at least one partition , i. e. @xmath840 adjusted versions of both the rand and the jaccard index exist , in that a null model is introduced , corresponding to the hypothesis of independence of the two partitions @xcite . the null model expectation value of the measure is subtracted from the unadjusted version , and the result is normalized by the range of this difference , yielding @xmath2 for identical partitions and @xmath631 as expected value for independent partitions ( negative values are possible as well ) . unadjusted indices have the drawback that they are not local , i. e. the result depends on how the whole graph is partitioned , even when the partitions differ only in a small region of the graph . similarity measures based on cluster matching aim at finding the largest overlaps between pairs of clusters of different partitions . for instance , the _ classification error _ @xmath841 is defined as @xcite @xmath842 where @xmath843 is an injective mapping from the cluster indices of partition @xmath833 to the cluster indices of partition @xmath832 . the maximum is taken over all possible injections @xmath510 . in this way one recovers the maximum overlap between the clusters of the two partitions . an alternative measure is the _ normalized van dongen metric _ , defined as @xcite @xmath844 . \label{eqt06}\ ] ] a common problem of this type of measures is that some clusters may not be taken into account , if their overlap with clusters of the other partition is not large enough . therefore if we compute the similarity between two partitions @xmath845 and @xmath846 and partition @xmath847 , with @xmath845 and @xmath846 differing from each other by a different subdivision of parts of the graph that are not used to compute the measure , one would obtain the same score . the third class of similarity measures is based on reformulating the problem of comparing partitions as a problem of message decoding within the framework of information theory @xcite . the idea is that , if two partitions are similar , one needs very little information to infer one partition given the other . this extra information can be used as a measure of dissimilarity . to evaluate the shannon information content @xcite of a partition , one starts by considering the community assignments @xmath848 and @xmath849 , where @xmath170 and @xmath850 indicate the cluster labels of vertex @xmath67 in partition @xmath832 and @xmath833 , respectively . one assumes that the labels @xmath302 and @xmath851 are values of two random variables @xmath168 and @xmath585 , with joint distribution @xmath852 , which implies that @xmath853 and @xmath854 . the _ mutual information _ @xmath595 of two random variables has been previously defined [ eq . ( [ eqstinf8 ] ) ] , and can be applied as well to partitions @xmath832 and @xmath833 , since they are described by random variables . actually @xmath855 , where @xmath856 is the shannon entropy of @xmath168 and @xmath857 is the conditional entropy of @xmath168 given @xmath585 . the mutual information is not ideal as a similarity measure : in fact , given a partition @xmath832 , all partitions derived from @xmath832 by further partitioning ( some of ) its clusters would all have the same mutual information with @xmath832 , even though they could be very different from each other . in this case the mutual information would simply equal the entropy @xmath858 , because the conditional entropy would be systematically zero . to avoid that , danon et al . adopted the _ normalized mutual information _ @xcite @xmath859 which is currently very often used in tests of graph clustering algorithms . the normalized mutual information equals @xmath2 if the partitions are identical , whereas it has an expected value of @xmath631 if the partitions are independent . the measure , defined for standard partitions , in which each vertex belongs to only one cluster , has been recently extended to the case of overlapping clusters by lancichinetti et al . the extension is not straightforward as the community assignments of a partition are now specified by a vectorial random variable , since each vertex may belong to more clusters simultaneously . in fact , the definition by lancichinetti et al . is not a proper extension of the normalized mutual information , in the sense that it does not recover exactly the same value of the original measure for the comparison of proper partitions without overlap , even though the values are close . meil @xcite introduced the _ variation of information _ @xmath860 which has some desirable properties with respect to the normalized mutual information and other measures . in particular , it defines a metric in the space of partitions as it has the properties of distance . it is also a local measure , i. e. the similarity of partitions differing only in a small portion of a graph depends on the differences of the clusters in that region , and not on the partition of the rest of the graph . the maximum value of the variation of information is @xmath241 , so similarity values for partitions of graphs with different size can not be compared with each other . for meaningful comparisons one could divide @xmath861 by @xmath241 , as suggested by karrer et al . @xcite . a concept related to similarity is that of _ distance _ , which indicates basically how many operations need to be performed in order to transform a partition to another . gustafsson et al . defined two distance measures for partitions @xcite . they are both based on the concept of _ meet _ of two partitions , which is defined as @xmath862 . \label{eqt1}\ ] ] the distance measures are @xmath863 and @xmath864 . in both cases they are determined by summing the distances of @xmath832 and @xmath833 from the meet @xmath865 . for @xmath863 the distance of @xmath832 ( @xmath833 ) from the meet is the minimum number of elements that must be moved between @xmath832 and @xmath833 so that @xmath832 ( @xmath833 ) and @xmath865 coincide @xcite . for @xmath864 the distance of @xmath832 ( @xmath833 ) from the meet is the minimum number of divisions that must be done in @xmath832 ( @xmath833 ) so that @xmath832 ( @xmath833 ) and @xmath865 coincide @xcite . such distance measures can easily be transformed in similarity measures , like @xmath866 identical partitions have zero mutual distance and similarity @xmath2 based on eqs . [ eqt2 ] . finally an important problem is how to define the similarity between clusters . if two partitions @xmath832 and @xmath833 of a graph are similar , each cluster of @xmath832 will be very similar to one cluster of @xmath833 , and viceversa , and it is important to identify the pairs of corresponding clusters . for instance , if one has information about the time evolution of a graph , one could monitor the dynamics of single clusters as well , by keeping track of each cluster at different time steps @xcite . given clusters @xmath825 and @xmath826 , their similarity can be defined through the _ relative overlap _ @xmath867 @xmath868 in this way , looking for the cluster of @xmath833 corresponding to @xmath825 means finding the cluster @xmath826 that maximizes @xmath867 . the index @xmath867 can be used to define similarity measures for partitions as well @xcite . an interesting discussion on the problem of comparing partitions , along with further definitions of similarity measures not discussed here , can be found in ref . @xcite . the first systematic comparative analysis of graph clustering techniques has been carried out by danon et al . they compared the performances of various algorithms on the benchmark graphs by girvan and newman ( section [ sec6_1 ] ) . the algorithms examined are listed in table [ table_orders ] , along with their complexity . [ cols="^,^,^,^",options="header " , ] fig . [ figdanon ] shows the performance of all algorithms . instead of showing the whole curves of the similarity versus @xmath784 ( section [ sec6_1 ] ) , which would display a fuzzy picture with many strongly overlapping curves , difficult to appreciate , danon et al . considered three values for @xmath784 ( @xmath227 , @xmath869 and @xmath228 ) , and represented the result for each algorithm as a group of three columns , indicating the average value of the similarity between the planted partition and the partition found by the method for each of the three @xmath784-values . the similarity was measured in terms of the fraction of correctly classified vertices ( section [ sec6_1 ] ) . the comparison shows that modularity optimization via simulated annealing ( section [ sub_sec6_0_2 ] ) yields the best results , although it is a rather slow procedure , that can not be applied to graphs of size of the order of @xmath612 vertices or larger . on the other hand , we have already pointed out that the benchmark by girvan and newman is not a good representation of real graphs with community structure , which are characterized by heterogeneous distributions of degree and community sizes . in this respect , the class of graphs designed by lancichinetti et al . ( lfr benchmark ) @xcite ( section [ sec6_1 ] ) poses a far more severe test to clustering techniques . for instance , many methods have problems to detect clusters of very different sizes ( like most methods listed in table [ table_orders ] ) . for this reason , lancichinetti and fortunato have carried out a careful comparative analysis of community detection methods on the much more restrictive lfr benchmark @xcite . the algorithms chosen are listed in table [ table2 ] . in fig . [ companalysis ] the performances of the algorithms on the lfr benchmark are compared . whenever possible , tests on the versions of the lfr benchmark with directed edges , weighted edges and/or overlapping communities @xcite were carried out . lancichinetti and fortunato also tested the methods on random graphs , to check whether they are able to notice the absence of community structure . from the results of all tests , the infomap method by rosvall and bergstrom @xcite appears to be the best , but also the algorithms by blondel et al . @xcite and by ronhovde and nussinov @xcite have a good performance . these three methods are also very fast , with a complexity which is essentially linear in the system size , so they can be applied to large systems . on the other hand , modularity - based methods ( with the exception of the method by blondel et al . ) have a rather poor performance , which worsens for larger systems and smaller communities , due to the well known resolution limit of modularity @xcite . the performance of the remaining methods worsens considerably if one increases the system size ( dm and infomod ) or the community size ( cfinder , mcl and method by radicchi et al . ) . performances of several algorithms on the lfr benchmark @xcite . the plots show the normalized mutual information ( in the version proposed in ref . @xcite ) as a function of the mixing parameter of the benchmark graphs . the different curves for each method refer to different system sizes ( @xmath870 and @xmath871 vertices ) and community size ranges [ ( s)=from @xmath5 to @xmath872 vertices , ( b)=from @xmath873 to @xmath874 vertices ] . for the gn algorithm only the smaller graph size was adopted , due to the high complexity of the method , whereas for the em method there are eight curves instead of four because for each set of benchmark graphs the algorithm was run starting from two different initial conditions . reprinted figure with permission from ref . 2009 by the american physical society.,title="fig:",width=302 ] performances of several algorithms on the lfr benchmark @xcite . the plots show the normalized mutual information ( in the version proposed in ref . @xcite ) as a function of the mixing parameter of the benchmark graphs . the different curves for each method refer to different system sizes ( @xmath870 and @xmath871 vertices ) and community size ranges [ ( s)=from @xmath5 to @xmath872 vertices , ( b)=from @xmath873 to @xmath874 vertices ] . for the gn algorithm only the smaller graph size was adopted , due to the high complexity of the method , whereas for the em method there are eight curves instead of four because for each set of benchmark graphs the algorithm was run starting from two different initial conditions . reprinted figure with permission from ref . 2009 by the american physical society.,title="fig:",width=302 ] performances of several algorithms on the lfr benchmark @xcite . the plots show the normalized mutual information ( in the version proposed in ref . @xcite ) as a function of the mixing parameter of the benchmark graphs . the different curves for each method refer to different system sizes ( @xmath870 and @xmath871 vertices ) and community size ranges [ ( s)=from @xmath5 to @xmath872 vertices , ( b)=from @xmath873 to @xmath874 vertices ] . for the gn algorithm only the smaller graph size was adopted , due to the high complexity of the method , whereas for the em method there are eight curves instead of four because for each set of benchmark graphs the algorithm was run starting from two different initial conditions . reprinted figure with permission from ref . 2009 by the american physical society.,title="fig:",width=302 ] fan et al . have evaluated the performance of some algorithms to detect communities on weighted graphs @xcite . the algorithms are : modularity maximization , carried out with extremal optimization ( weo ) ( section [ sub_sec6_0_3 ] ) ; the girvan - newman algorithm ( wgn ) ( section [ subsec5_1 ] ) ; the potts model algorithm by reichardt and bornholdt ( potts ) ( section [ sec44_1 ] ) . all these techniques have been originally introduced for unweighted graphs , but we have shown that they can easily be extended to weighted graphs . the algorithms were tested on the weighted version of the benchmark of girvan and newman , that we discussed in section [ sec6_1 ] . edge weights have only two values : @xmath875 for inter - cluster edges and @xmath876 for intra - cluster edges . such values are linked by the relation @xmath877 , so they are not independent . for testing one uses realizations of the benchmark with fixed topology ( i. e. fixed @xmath784 ) and variable weights . in fig . [ fantest ] the comparative performance of the three algorithms is illustrated . the topology of the benchmark graphs corresponds to @xmath878 , i. e. to graphs in which each vertex has approximately the same number of neighbors inside and outside its community . by varying @xmath875 from @xmath631 to @xmath323 one goes smoothly from a situation in which most of the weight is concentrated inside the clusters , to a situation in which instead the weight is concentrated between the clusters . from fig . [ fantest ] we see that weo and potts are more reliable methods . comparative evaluation of the performances of algorithms to find communities in weighted graphs . tests are carried out on a weighted version of the benchmark of girvan and newman . the two plots show how good the algorithms are in terms of the precision and accuracy with which they recover the planted partition of the benchmark . precision indicates how close the values of similarity between the planted and the model partition are after repeated experiments with the same set of parameters ; accuracy indicates how close the similarity values are to the ideal result ( 1 ) after repeated experiments with the same set of parameters . the similarity measure adopted here is based on the relative overlap of clusters of eq . [ eqt20 ] . we see that the maximization of modularity with extremal optimization ( weo ) and the potts model algorithm ( potts ) are both precise and accurate as long as the weight of the inter - cluster edges @xmath875 remains lower than the weight of the intra - cluster edges ( @xmath879 ) . reprinted figures with permission from ref . 2007 by elsevier.,title="fig : " ] comparative evaluation of the performances of algorithms to find communities in weighted graphs . tests are carried out on a weighted version of the benchmark of girvan and newman . the two plots show how good the algorithms are in terms of the precision and accuracy with which they recover the planted partition of the benchmark . precision indicates how close the values of similarity between the planted and the model partition are after repeated experiments with the same set of parameters ; accuracy indicates how close the similarity values are to the ideal result ( 1 ) after repeated experiments with the same set of parameters . the similarity measure adopted here is based on the relative overlap of clusters of eq . [ eqt20 ] . we see that the maximization of modularity with extremal optimization ( weo ) and the potts model algorithm ( potts ) are both precise and accurate as long as the weight of the inter - cluster edges @xmath875 remains lower than the weight of the intra - cluster edges ( @xmath879 ) . reprinted figures with permission from ref . 2007 by elsevier.,title="fig : " ] sawardecker et al . have tested methods to detect overlapping communities @xcite . they considered three algorithms : modularity optimization , the clique percolation method ( cpm ) ( section [ sec45_1 ] ) and the modularity landscape surveying method by sales - pardo et al . @xcite ( section [ sub_sec6_202 ] ) . for testing , sawardecker et al . defined a class of benchmark graphs in which the linking probability between vertices is an increasing function of the number of clusters the vertices belong to . we have described this benchmark in section [ sec6_1 ] . it turns out that the modularity landscape surveying method is able to identify overlaps between communities , as long as the fraction of overlapping vertices is small . curiously , the cpm , designed to find overlapping communities , has a poor performance , as the overlapping vertices found by the algorithm are in general different from the overlapping vertices of the planted partition of the benchmark . the authors also remark that , if the overlap between two clusters is not too small , it may be hard ( for any method ) to recognize whether the clusters are overlapping or hierarchically organized , i. e. loosely connected clusters within a large cluster . we close the section with some general remarks concerning testing . we have seen that a testing procedure requires two crucial ingredients : benchmark graphs with built - in community structure and clustering algorithms that try to recover it . such two elements are not independent , however , as they are both based on the concept of community . if the underlying notions of community for the benchmark and the algorithm are very different , one can hardly expect that the algorithm will do a good job on the benchmark . furthermore , there is a third element , i. e. the quality of a partition . all benchmarks start from a situation in which communities are clearly identified , i. e. connected components of the graph , and introduce some amount of noise , that eventually leads to a scenario where clusters are hardly or no longer detectable . it is then important to keep track of how the quality of the natural partition of the benchmark worsens as the amount of noise increases , in order to distinguish configurations in which the graphs have a cluster structure , that an algorithm should then be able to resolve , from configurations in which the noise prevails and the natural clusters are not meaningful . moreover , quality functions are important to evaluate the performance of an algorithm on graphs whose community structure is unknown . quality functions are strongly related to the concept of community as well , as they are supposed to evaluate the goodness of the clusters , so they require a clear quantitative concept of what a cluster is . it is very important for any testing framework to check for the mutual dependencies between the benchmark , the quality function used to evaluate partitions , and the clustering algorithm to be tested . this issue has so far received very little attention @xcite . finally , empirical tests are also very important , as one ultimately wishes to apply clustering techniques to real graphs . therefore , it is crucial to collect more data sets of graphs whose community structure is known or deducible from information on the vertices and their edges . what are the general properties of partitions and clusters of real graphs ? in many papers on graph clustering applications to real systems are presented . in spite of the variety of clustering methods that one could employ , in many cases partitions derived from different techniques are rather similar to each other , so the general properties of clusters do not depend much on the particular algorithm used . the analysis of clusters and their properties delivers a _ mesoscopic description _ of the graph , where the communities , and not the vertices , are the elementary units of the topology . the term mesoscopic is used because the relevant scale here lies between the scale of the vertices and that of the full graph . one of the first issues addressed was whether the communities of a graph are usually about of the same size or whether the community sizes have some special distribution . most clustering techniques consistently find skewed distributions of community sizes , with a tail described with good approximation by a power law ( at least , a sizeable portion of the curve ) with exponents in the range between @xmath2 and @xmath3 @xcite . so , there seems to be no characteristic size for a community : small communities usually coexist with large ones . cumulative distribution of community sizes for the amazon purchasing network . the partition is derived by greedy modularity optimization . reprinted figure with permission from ref . 2004 by the american physical society.,width=302 ] as an example , fig . [ figure13 ] shows the cumulative distribution of community sizes for a recommendation network of the online vendor amazon.com . vertices are products and there is a connection between item @xmath9 and @xmath10 if @xmath10 was frequently purchased by buyers of @xmath9 . recall that the cumulative distribution is the integral of the probability distribution : if the cumulative distribution is a power law @xmath880 , the probability distribution is also a power law with exponent @xmath881 . leskovec et al . @xcite have gone one step further . they carried out a systematic analysis of communities in large real networks , including traditional and on - line social networks , technological , information networks and web graphs . the main goal was to assess the quality of communities at various sizes . as a quality function the _ conductance _ of the cluster was chosen . we remind that the conductance of a cluster is the ratio between the cut size of the cluster and the minimum between the total degree of the cluster and that of the rest of the graph ( section [ sec4_1 ] ) . so , if the cluster is much smaller than the whole graph , the conductance equals the ratio between the cut size and the total degree of the cluster . since a `` good '' cluster is characterized by a low cut size and a large internal density of edges , low values of the conductance indicate good clusters . for each real network leskovec derived the _ network community profile plot _ ( ncpp ) , showing the minimum conductance score among subgraphs of a given size as a function of the size . interestingly , they found that the ncpps of all networks they studied have a characteristic shape : they go downwards up until subgraphs with about @xmath874 vertices , and then they rise monotonically for larger subgraphs ( fig . [ figlesk ] ) . this seems to suggest that communities are well defined only when they are fairly small in size . analysis of communities in large real networks by leskovec et al . ( left ) typical shape of the network community profile plot ( ncpp ) , showing how the minimum conductance of subgraphs of size @xmath15 varies with @xmath15 . the plot indicates that the `` best '' communities have a size of about @xmath874 vertices ( minimum of the curve ) , whereas communities of larger sizes are not well - defined . in the plot two other ncpps are shown : the one labeled _ rewired network _ corresponds to a randomized version of the network , where edges are randomly rewired by keeping the degree distribution ; the one labeled _ bag of whiskers _ gives the minimum conductance scores of clusters composed of disconnected pieces . ( right ) scheme of the core - periphery structure of large social and information networks derived by leskovec et al . based on the results of their empirical analysis . most of the vertices are in a central core , which does not have a clear community structure , whereas the best communities , which are rather small , are weakly connected to the core . reprinted figure with permission from ref . @xcite.,title="fig : " ] 0.5 cm analysis of communities in large real networks by leskovec et al . ( left ) typical shape of the network community profile plot ( ncpp ) , showing how the minimum conductance of subgraphs of size @xmath15 varies with @xmath15 . the plot indicates that the `` best '' communities have a size of about @xmath874 vertices ( minimum of the curve ) , whereas communities of larger sizes are not well - defined . in the plot two other ncpps are shown : the one labeled _ rewired network _ corresponds to a randomized version of the network , where edges are randomly rewired by keeping the degree distribution ; the one labeled _ bag of whiskers _ gives the minimum conductance scores of clusters composed of disconnected pieces . ( right ) scheme of the core - periphery structure of large social and information networks derived by leskovec et al . based on the results of their empirical analysis . most of the vertices are in a central core , which does not have a clear community structure , whereas the best communities , which are rather small , are weakly connected to the core . reprinted figure with permission from ref . @xcite.,title="fig : " ] such small clusters are weakly connected to the rest of the graph , often by a single edge ( in this case they are called _ whiskers _ ) , and form the _ periphery _ of the network . the other vertices form a big _ core _ , in which larger clusters are well connected to each other , and are therefore barely distinguishable ( fig . [ figlesk ] ) . leskovec et al . performed low - conductance cuts with several methods , to ensure that the result is not a simple artefact of a particular chosen technique . moreover , they have also verified that , for large real networks with known community structure ( such as , e.g. , the social network of the on - line blogging site _ livejournal _ , with its user groups ) , the ncpp has the same qualitative shape if one takes the real communities instead of low - conductance subgraphs . the analysis by leskovec et al . may shed new light on our understanding of community structure and its detection in large networks . the fact that the `` best '' communities appear to have a characteristic size of about @xmath874 vertices is consistent with dunbar conjecture that @xmath882 is the upper size limit for a working human community @xcite . on the other hand , if large communities are very mixed with each other , as leskovec et al . claim , they could hardly be considered communities , and the alleged `` community structure '' of large networks would be limited to their peripheral region . the results by leskovec et al . may be affected by the properties of conductance , and need to be validated with alternative approaches . in any case , whatever the value of the quality score of a cluster may be ( low or high ) , it is necessary to estimate the significance of the cluster ( section [ sec6_4 ] ) , before deciding whether it is a meaningful structure or not . if the community structure of a graph is known , it is possible to classify vertices according to their roles within their community , which may allow to infer individual properties of the vertices . a promising classification has been proposed by guimer and amaral @xcite . the role of a vertex depends on the values of two indices , the _ within - module degree _ and the _ participation ratio _ ( though other variables may be chosen , in principle ) . the within - module degree @xmath883 of vertex @xmath67 is defined as @xmath884 where @xmath885 is the internal degree of @xmath67 in its cluster @xmath314 , @xmath886 and @xmath887 the average and standard deviation of the internal degrees for all vertices of cluster @xmath314 . the within - module degree is then defined as the @xmath406-score of the internal degree @xmath885 . large values of @xmath406 indicate that the vertex has many more neighbors within its community than most other vertices of the community . vertices with @xmath888 are classified as _ hubs _ , if @xmath889 they are _ non - hubs_. the participation ratio @xmath890 of vertex @xmath67 is defined as @xmath891 here @xmath892 is the internal degree of @xmath67 in cluster @xmath155 , @xmath105 the degree of @xmath67 . values of @xmath96 close to @xmath2 indicate that the neighbors of the vertex are uniformly distributed among all clusters ; if all neighbors are within the cluster of the vertex , instead , @xmath893 . based on the values of the pair @xmath894 , guimer and amaral distinguished seven roles for the vertices . non - hub vertices can be _ ultra - peripheral _ ( @xmath895 ) , _ peripheral _ ( @xmath896 ) , _ connectors _ ( @xmath897 ) and _ kinless vertices _ ( @xmath898 ) . hub vertices are classified in _ provincial hubs _ ( @xmath899 ) , _ connector hubs _ ( @xmath900 ) and _ kinless hubs _ ( @xmath901 ) . the regions of the @xmath902 plane corresponding to the seven roles are highlighted in fig . [ figroles ] . we stress that the actual boundaries of the regions can be chosen rather arbitrarily . regions of the @xmath902 plane defining the roles of vertices in the modular structure of a graph , according to the scheme of guimer and amaral @xcite . reprinted figure with permission from ref . 2005 by the nature publishing group . ] on graphs without community structure , like erds - rnyi @xcite random graphs and barabsi - albert @xcite graphs ( section [ sec1_3 ] ) , non - hubs are mostly kinless vertices . in addition , if there are hubs , like in barabsi - albert graphs , they are kinless hubs . kinless hubs ( non - hubs ) vertices have less than half ( one third ) of their neighbors inside any cluster , so they are not clearly associated to a cluster . on real graphs , the topological roles can be correlated to functions of vertices : in metabolic networks , for instance , connector hubs , which share most edges with vertices of other clusters than their own , are often metabolites which are more conserved across species than other metabolites , i. e. they have an evolutionary advantage @xcite . if communities are overlapping , one can explore other statistical properties , like the distributions of the overlaps and of the vertex memberships . the overlap is defined as the number of vertices shared by each pair of overlapping clusters ; the membership of a vertex is the number of communities including the vertex . both distributions turn out to be skewed , so there seem to be no characteristic values for the overlap and the membership . moreover , one could derive a network , where the communities are the vertices and pairs of vertices are connected if their corresponding communities overlap @xcite . such networks seem to have some special properties . for instance , the degree distribution is a particular function , with an initial exponential decay followed by a slower power law decay ) and a coauthorship network of physicists . there is no _ a priori _ reason to believe that this result is general . ] . we stress that the above results have been obtained with the clique percolation method by palla et al . ( section [ sec45_1 ] ) and it is not clear whether other techniques would confirm them or not . in a recent analysis it has been shown that the degree distribution of the network of communities can be reproduced by assuming that the graph grows according to a simple preferential attachment mechanism , where communities with large degree have an enhanced chance to interact / overlap with new communities @xcite . the ultimate goal of clustering algorithms is trying to infer properties of and relationships between vertices , that are not available from direct observation / measurement . if the scientific community agrees on a set of reliable techniques , one could then proceed with careful investigations of systems in various domains . so far , most works in the literature on graph clustering focused on the development of new algorithms , and applications were limited to those few benchmark graphs that one typically uses for testing ( section [ sec6_1 ] ) . still , there are also applications aiming at understanding real systems . some results have been actually mentioned in the previous sections . this section is supposed to give a flavor of what can be done by using clustering algorithms . therefore , the list of works presented here is by no means exhaustive . most studies focus on biological and social networks . we mention a few applications to other types of networks as well . the recent abundance of genomic data has allowed us to explore the cell at an unprecedented depth . a wealth of information is available on interactions involving proteins and genes , metabolic processes , etc . in order to study cellular systems , the graph representation is regularly used . protein - protein interaction networks ( pin ) , gene regulatory networks ( grn ) and metabolic networks ( mn ) are meanwhile standard objects of investigation in biology and bioinformatics @xcite . biological networks are characterized by a remarkable modular organization , reflecting functional associations between their components . for instance , proteins tend to be associated in two types of cellular modules : _ protein complexes _ and _ functional modules_. a protein complex is a group of proteins that mutually interact at the same time and space , forming a sort of physical object . examples are transcription factor complexes , protein transport and export complexes , etc . functional modules instead are groups of proteins taking place in the same cellular process , even if the interactions may happen at different times and places . examples are the cdk / cyclin module , responsible for cell - cycle progression , the yeast pheromone response pathway , etc .. identifying cellular modules is fundamental to uncover the organization and dynamics of cell functions . however , the information on cell units ( e. g. proteins , genes ) and their interactions is often incomplete , or even incorrect , due to noise in the data produced by the experiments . therefore , inferring modules from the topology of cellular networks enables one to restrict the set of possible scenarios and can be a safe guide for future experiments . rives and galitski @xcite studied the modular organization of a subset of the pin of the yeast ( _ saccharomyces cerevisiae _ ) , consisting of the ( signaling ) proteins involved in the processes leading the microorganism to a filamentous form . the clusters were detected with a hierarchical clustering technique . proteins mostly interacting with members of their own cluster are often essential proteins ; edges between modules are important points of communication . spirin and mirny @xcite identified protein complexes and functional modules in yeast with different techniques : clique detection , superparamagnetic clustering @xcite and optimization of cluster edge density . they estimated the statistical significance of the clusters by computing the @xmath45-values of seeing those clusters in random graphs with the same expected degree sequence as the original network . from the known functional annotations of yeast genes one can see that the modules usually group proteins with the same or consistent biological functions . indeed , in many cases , the modules exactly coincide with known protein complexes . the results appear robust if noise is introduced in the system , to simulate the noise present in the experimental data . functional modules in yeast were also found by chen and yuan @xcite , who applied the algorithm by girvan and newman with a modified definition of edge betweenness ( section [ subsec5_1 ] ) . the standard girvan - newman algorithm has proved to be reliable to detect functional modules in pins @xcite . the novelty of the work by chen and yuan is its focus on weighted pins , where the weights come from information derived through microarray expression profiles . weights add information about the system and should lead to a more reliable modular structure . by knocking out genes in the same structural cluster similar phenotypes appeared , suggesting that the genes have similar biological roles . moreover , the clusters often contained known protein complexes , either entirely or to a large extent . finally , chen and yuan were able to make predictions of the unknown function of some genes , based on the structural module they belong to : gene function prediction is the most promising outcome deriving from the application of clustering techniques to pins . farutin et al . @xcite have adopted a local concept of community , and derived a hierarchical decomposition of pins , in that the modules identified at some level become the vertices of a network at the higher level . communities are overlapping , to account for the fact that proteins ( and whole modules ) may have diverse biological functions . high level structures detected in a human pin correspond to general biological concepts like signal transduction , regulation of gene expression , intercellular communication . sen et al . @xcite identified protein clusters for yeast from the eigenvectors of the laplacian matrix ( section [ sec1_2 ] ) , computed via singular value decomposition . in a recent analysis , lewis et al . @xcite carefully explored the relationship between structural communities of pins and their biological function . communities were detected with the multiresolution approach by reichardt and bornholt @xcite ( section [ sub_sec6_01 ] ) . a community is considered biologically homogeneous if the functional similarity between protein pairs of the community ( extracted through the gene ontology database @xcite ) is larger than the functional similarity between all protein pairs of the network . lewis et al . also specified the comparison to interacting and non - interacting protein pairs . as a result , many communities turn out to be biologically homogeneous , especially if they are not too small . moreover , some topological attributes of communities , like the within - community clustering coefficient ( i.e. the average value of the clustering coefficients of the vertices of a community , computed by considering just the neighbors belonging to the community ) and link density ( density of internal edges ) , are good indicators of biological homogeneity : the former is strongly correlated with biological homogeneity , independently of the community size , whereas for the latter the correlation is strong for large communities . metabolic networks have also been extensively investigated . we have already discussed the `` functional cartography '' designed by guimer and amaral @xcite , which applies to general types of networks , not necessarily metabolic . a hierarchical decomposition of metabolic networks has been derived by holme et al . @xcite , by using a hierarchical clustering technique inspired by the algorithm by girvan and newman ( section [ subsec5_1 ] ) . here , vertices are removed based on their betweenness values , which are obtained by dividing the standard site betweenness scores @xcite by the indegree of the respective vertices . a picture of metabolic network emerges , in which there are core clusters centered at hub - substances , surrounded by outer shells of less connected substances , and a few other clusters at intermediate scales . in general , clusters at different scales seem to be meaningful , so the whole hierarchy should be taken into account . wilkinson and huberman @xcite analyzed a network of gene co - occurrence to find groups of related genes . the network is built by connecting pairs of genes that are mentioned together in the abstract of articles of the medline database ( http://medline.cos.com/ ) . clusters were found with a modified version of the algorithm by girvan and newman , in which edge betweenness is computed by considering the shortest paths of a small subset of all vertex pairs , to gain computer time ( section [ subsec5_1 ] ) . as a result , genes belonging to the same cluster turn out to be functionally related to each other . co - occurrence of terms is also used to extract associations between genes and diseases , to find out which genes are relevant for a specific disease . communities of genes related to colon cancer can be helpful to identify the function of the genes . networks depicting social interactions between people have been studied for decades @xcite . recently the modern information and communication technology ( ict ) has opened new interaction modes between individuals , like mobile phone communications and online interactions enabled by the internet . such new social exchanges can be accurately monitored for very large systems , including millions of individuals , whose study represents a huge opportunity for social science . communities of social networks can be friendship circles , groups of people sharing common interests and/or activities , etc .. blondel et al . have analyzed a network of mobile phone communications between users of a belgian phone operator @xcite . the vertices of the graph are 2.6 millions and the edges are weighted by the cumulative duration of phone calls between users in the observation time frame . the clustering analysis , performed with a fast hierarchical modularity optimization technique developed by the authors ( discussed in section [ sub_sec6_0_1 ] ) , delivers six hierarchical levels . the highest level consists of @xmath903 groups with more than @xmath874 vertices , which are clearly arranged in two main groups , linguistically homogeneous , reflecting the linguistic split of belgian population ( fig . [ blondphone ] ) . community structure of a social network of mobile phone communication in belgium . dots indicate subcommunities at the lower hierarchical level ( with more than @xmath874 people ) and are colored in a red - green scale to represent the level of representation of the two main languages spoken in belgium ( red for french and green for dutch ) . communities of the two larger groups are linguistically homogeneous , with more than @xmath904 of people speaking the same language . only one community ( zoomed ) , which lies at the border between the two main aggregations , has a more balanced distribution of languages . reprinted figure with permission from ref . 2008 by iop publishing and sissa.,width=302 ] tyler et al . @xcite studied a network of e - mail exchanges between people working at hp labs . they applied the same modified version of girvan - newman algorithm that two of the authors have used to find communities of related genes @xcite ( section [ sec7_2_1 ] ) . the method enables one to measure the degree of membership of each vertex in a community and allows for overlaps between communities . the detected clusters matched quite closely the organization of the labs in departments and project groups , as confirmed by interviews conducted with researchers . communities in social networking sites . ( top ) visualization of a network of friendships between students at caltech , constructed from facebook data ( september 2005 ) . the colors / shapes indicate the dormitories ( houses ) of the students . ( bottom ) topological communities of the network , which are quite homogeneous with respect to house affiliation . reprinted figures with permission from refs . @xcite and @xcite.,title="fig:",width=264 ] communities in social networking sites . ( top ) visualization of a network of friendships between students at caltech , constructed from facebook data ( september 2005 ) . the colors / shapes indicate the dormitories ( houses ) of the students . ( bottom ) topological communities of the network , which are quite homogeneous with respect to house affiliation . reprinted figures with permission from refs . @xcite and @xcite.,title="fig:",width=264 ] social networking sites , like _ myspace _ ( www.myspace.com ) , _ friendster _ ( www.friendster.com ) , _ facebook _ ( www.facebook.com ) , etc . have become extremely popular in the last years . they are online platforms that allow people to communicate with friends , send e - mails , solicit opinions on specific issues , spread ideas and/or fads , etc . traud et al . @xcite used anonymous facebook data to create networks of friendships between students of different american universities , where vertices / students are connected if they are friends on facebook . communities were detected by applying a variant of newman s spectral optimization of modularity ( section [ sub_sec6_0_4 ] ) : the results were further refined through additional steps la kernighan - lin ( section [ sec4_1 ] ) . one of the goals of the study was to infer relationships between the online and offline lives of the students . by using demographic information on the students populations , one finds that communities are organized by class year or by house ( dormitory ) affiliation , depending on the university ( fig . [ figcaltech ] ) . yuta et al . @xcite observed a gap in the community size distribution of a friendship network extracted from _ mixi _ ( mixi.jp ) , the largest social networking site in japan ( as of december 2006 ) . communities were identified with the fast greedy modularity optimization by clauset et al . the gap occurs in the intermediate range of sizes between @xmath873 and @xmath905 , where but a few communities are observed . yuta et al . introduced a model where people form new friendships both by `` closing '' ties with people who are friends of friends , and by setting new links with individuals having similar interests . in this way most groups turn out to be either small or large , and medium size groups are rare . collaboration networks , in which individuals are linked if they are ( were ) involved in a common activity , have been often studied because they embed an implicit objective concept of acquaintance , that is not easy to capture in direct social experiments / interviews . for instance , somebody may consider another individual a friend , while the latter may disagree . a collaboration instead is a proof of a social relationship between individuals . the analysis of the structure of scientific collaboration networks @xcite has exerted a big influence on the development of the modern network science . scientific collaboration is associated to coauthorship : two scientists are linked if they have coauthored at least one paper together . information about coauthorships can be extracted from different databases of research papers . communities indicate groups of people with common research interests , i. e. topical or disciplinary groups . in the seminal paper by girvan and newman @xcite , the authors applied their method on a collaboration network of scientists working at the santa fe institute , and were able to discriminate between research divisions ( fig . [ fig3]b ) . the community structure of scientific collaboration networks has been investigated by many authors @xcite . other types of collaboration networks have been studied too . gleiser and danon @xcite considered a collaboration network of jazz musicians . vertices are either musicians , connected if they played in the same band , or bands , connected if they have a musician in common . by applying the algorithm of girvan and newman they found that communities reflect both racial segregation ( with two main groups comprising only black or white players ) and geographical separation , due to the different recording locations . citation networks @xcite have been regularly used to understand the citation patterns of authors and to disclose relationships between disciplines . rosvall and bergstrom @xcite used a citation network of over @xmath906 scientific journals to derive a map of science . they used a clustering technique based on compressing the information on random walks taking place on the graph ( section [ subsec_statinf2 ] ) . a random walk follows the flow of citations from one field to another , and the fields emerge naturally from the clustering analysis ( fig . [ maprosvall ] ) . the structure of science resembles the letter @xmath197 , with the social sciences and engineering at the terminals , joined through a chain including medicine , molecular biology , chemistry and physics . reichardt and bornholdt @xcite performed a clustering analysis on a network built from bidding data taken from the german version of ebay ( www.ebay.de ) , the most popular online auction site . the vertices are bidders and two vertices are connected if the corresponding bidders have expressed interest for the same item . clusters were detected with the multiresolution modularity optimization developed by the authors themselves @xcite ( section [ sub_sec6_01 ] ) . in spite of the variety of items that it is possible to purchase through ebay , about @xmath904 of bidders were classified into a few major clusters , reflecting bidders broad categories of interests . ebay data were also examined by jin et al . @xcite , who considered bidding networks where the vertices are the individual auctions and edges are placed between auctions having at least one common bidder . communities , detected with greedy modularity optimization @xcite ( section [ sub_sec6_0_1 ] ) , allow to identify substitute goods , i. e. products that have value for the same bidder , so that they can be purchased together or alternatively . legislative networks enable one to deduce associations between politicians through their parliamentary activity , which may be related or not to party affiliation . porter and coworkers have carried out numerous studies on the subject @xcite , by using data on the congress of the united states . in refs . @xcite , they examined the community structure of networks of committees in the us house of representatives . committees sharing common members are connected by edges , which are weighted by dividing the number of common members by the number one would expect to have if committee memberships were randomly assigned . hierarchical clustering ( section [ sec4_2 ] ) reveals close connections between some of the committees . in another work @xcite , zhang et al . analyzed networks of legislation cosponsorship , in which vertices are legislators and two legislators are linked if they support at least one common bill . communities , identified with a modification of newman s spectral optimization of modularity ( section [ sub_sec6_0_4 ] ) , are correlated with party affiliation , but also with geography and committee memberships of the legislators . networks of correlations between time series of stock returns have received a growing attention in the past few years @xcite . in early studies , scholars found clusters of correlated stocks by computing the _ maximum spanning tree _ of the network @xcite ( section [ sec1_1 ] ) , and realized that such clusters match quite well the economic sectors of the stocks . more recently , the community structure of the networks has been investigated by means of proper clustering algorithms . farkas et al . @xcite have applied the weighted version of the clique percolation method ( section [ sec45_1 ] ) and found that the presence of two strong ( i. e. carrying high correlation ) edges in triangles is usually accompanied by the presence of a strong third edge . heimo et al . @xcite used the weighted version of the multiresolution method by reichardt and bornholdt @xcite ( section [ sub_sec6_01 ] ) . clusters correspond to relevant business sectors , as indicated by forbes classification ; moreover , smaller clusters at lower hierarchical levels seem to correspond to ( economically ) meaningful substructures of the main clusters . despite the remote origins and the great popularity of the last years , research on graph clustering has not yet given a satisfactory solution of the problem and leaves us with a number of important open issues . from our exposition it appears that the field has grown in a rather chaotic way , without a precise direction or guidelines . in some cases , interesting new ideas and tools have been presented , in others existing methods have been improved , becoming more accurate and/or faster . what the field lacks the most is a theoretical framework that defines precisely what clustering algorithms are supposed to do . everybody has his / her own idea of what a community is , and most ideas are consistent with each other , but , as long as there is still disagreement , it remains impossible to decide which algorithm does the best job and there will be no control on the creation of new methods . therefore , we believe that the first and foremost task that the scientific community working on graph clustering has to solve in the future is defining a set of reliable benchmark graphs , against which algorithms should be tested ( section [ sec6_1 ] ) . defining a benchmark goes far beyond the issue of testing . it means designing practical examples of graphs with communities , and , in order to do that , one has to agree on the fundamental concepts of community and partition . clustering algorithms have to be devised consistently with such definitions , in order to give the best performance on the set of designated benchmarks , which represent a sort of ground truth . the explosion in the number of algorithms we have witnessed in recent times is due precisely to the present lack of reliable mechanisms of control of their quality and comparison of their performances . if the community agrees on a benchmark , the future development of the field will be more coherent and the progress brought by new methods can be evaluated in an unbiased manner . the planted @xmath293-partition model @xcite is the easiest recipe one can think of when it comes to defining clusters , and is the criterion underlying well - known benchmarks , like that by girvan and newman . we believe that the new benchmarks have to be defined along the same lines . the benchmark graphs recently introduced by lancichinetti et al . @xcite and by sawardecker et al . @xcite are an important step in this direction . defining a benchmark implies specifying the `` natural '' partition of a graph , the one that any algorithm should find . this issue in turn involves the concept of quality of a partition , that has characterized large part of the development of the field , in particular after the introduction of newman - girvan modularity ( section [ sec3_2_2 ] ) . estimating the quality of a partition allows to discriminate among the large number of partitions of a graph . in some cases this is not difficult . for instance , in the benchmark by girvan and newman there is a single meaningful partition , and it is hard to argue with that . but most graphs of the real world have a hierarchical structure , with communities including smaller communities and so on . hence there are several meaningful partitions , corresponding to different hierarchical levels , and discriminating among them is hard , as they may be all relevant , in a sense . if we consider the human body , we can not say that the organization in tissues of the cells is more important than the organization in organs . we have seen that there are recent methods dealing with the problem of finding meaningful hierarchical levels ( section [ sub_sec6_20 ] ) . such methods rank the hierarchical partitions based on some criterion and one can assess their relevance through the ranking . one may wonder whether it makes sense sorting out levels , which means introducing a kind of threshold on the quality index chosen to rank partitions ( to distinguish `` good '' from `` bad '' partitions ) , or whether it is more appropriate to keep the information given by the whole set of hierarchical partitions . the work by clauset et al . on hierarchical random graphs @xcite , discussed in section [ sub_sec6_202 ] , indirectly raises this issue . there it was shown that the ensemble of model graphs , represented by dendrograms , encodes most of the information on the structure of the graph at study , like its degree distribution , transitivity and distribution of shortest path lengths . at the same time , by construction , the model reveals the whole hierarchy of communities , without any distinction between good and bad partitions . the information given by a dendrogram may become redundant and confusing when the graph is large , as then there is a big number of partitions this is actually the reason why quality functions were originally introduced . however , in that case , one was dealing with artificial hierarchies , produced by techniques that systematically yield a dendrogram as a result of the analysis ( like , e. g. , hierarchical clustering ) , regardless of whether the graph actually has a hierarchical structure or not . here instead we speak of real hierarchy , which is a fundamental element of real graphs and , as such , it must be considered in any serious approach to graph clustering . any good clustering method must be able to tell whether a graph has community structure or not , and , in the first case , whether the community structure is hierarchical ( i. e. with two or more levels ) or flat ( one level ) . we expect that the concept of hierarchy will become a key ingredient of future clustering techniques . in particular , assessing the consistence of the concepts of partitions quality and hierarchical structure is a major challenge . a precise definition of null models , i. e. of graphs without community structure , is also missing . this aspect is extremely important , though , as defining communities also implies deciding whether or not they exist in a specific graph . at the moment , it is generally accepted that random graphs have no communities . the null model of modularity ( section [ sec3_2_2 ] ) , by far the most popular , comprises all graphs with the same expected degree sequence of the original graph and random rewiring of edges . this class of graphs is characterized , by construction , by the fact that any vertex can be linked to any other , as long as the constraint on the degree sequence is satisfied . but this is by no means the only possibility . a community can be generically defined as a subgraph whose vertices have a higher probability to be connected to the other vertices of the subgraph than to external vertices . the planted @xmath293-partition model @xcite is based on this principle , as we have seen . however , this does not mean that the linking probabilities of a vertex with respect to the other vertices in its community or in different communities be constant ( or simply proportional to their degrees , as in the configuration model @xcite ) . in fact , in large networks it is reasonable to assume that the probability that a vertex is linked to most vertices is zero , as the vertex `` ignores '' their existence . this does not exclude that the probability that the vertex gets connected to the `` known '' vertices is the same ( or proportional to their degrees ) , in which case the graph would still be random and have no communities . we believe that we are still far from a precise definition and a complete classification of null models . this represents an important research line for the future of the field , for three main reasons : 1 ) to better disentangle `` true '' communities from byproducts of random fluctuations ; 2 ) to pose a stringent test to existing and future clustering algorithms , whose reliability would be questionable if they found `` false positives '' in null model graphs ; 3 ) to handle `` hybrid '' scenarios , where a graph displays community structure on some portions of it , while the rest is essentially random and has no communities . in the previous chapters we have seen a great number of clustering techniques . which one(s ) shall we use ? at the moment the scientific community is unable to tell . modularity optimization is probably the most popular method , but the results of the analysis of large graphs are likely to be unreliable ( section [ sub_sec6_1 ] ) . nevertheless , people have become accustomed to use it , and there have been several attempts to improve the measure . a newcomer , who wishes to find clusters in a given network and is not familiar with clustering techniques , would not know , off - hand , which method to use , and he / she would hardly find indications about good methods in any single paper on graph clustering , except perhaps on the method presented in the paper . so , people keep using algorithms because they have heard of them , or because they know that other people are using them , or because of the reputation of the scientists who designed them . waiting for future reliable benchmarks , that may give an objective assessment of the quality of the algorithms , there are at the moment hardly solid reasons to prefer an algorithm to another : the comparative analyses by danon et al . @xcite and by lancichinetti and fortunato @xcite ( section [ sec6_3 ] ) represent a first serious assessment of this issue . however , we want to stress that there is no such thing as the perfect method , so it is pointless to look for it . among the other things , if one tries to look for a very general method , that should give good results on any type of graphs , one is inevitably forced to make very general assumptions on the structure of the graph and on the properties of communities . in this way one neglects a lot of specific features of the system , that may lead to a more accurate detection of the clusters . informing a method with features characterizing some types of graphs makes it far more reliable to detect the community structure of those graphs than a general method , even if its applicability may be limited . therefore in the future we envision the development of domain - specific clustering techniques . the challenge here is to identify the peculiar features of classes of graphs , which are bound to become crucial ingredients in the design of suitable algorithms . some of the methods available today are actually based on assumptions that hold only for some specific categories of graphs . the clique percolation method by palla et al . @xcite , for instance , may work well for graphs characterized by a large number of cliques , like certain social networks , whereas it may give poor results otherwise . moving one step further , one should learn how to use specific information about a graph , whenever available , e. g. properties of vertices and/or partial information about their classification . for instance , it may be that one has some information on a subset of vertices , like demographic data on people of a social network , and such data may highlight relationships between people that are not obvious from the network of social interactions . in this case , using only the social network may be reductive and ideally one should exploit both the structural and the non - structural information in the search of clusters , as the latter should be consistent with both inputs . how to do this is an open problem . the scientific community has just begun to study this aspect @xcite . most algorithms in the literature deal with the `` classical '' case of a graph with undirected and unweighted edges . this is certainly the simplest case one could think of , and graph clustering is already a complex task on such types of graphs . we know that real networks may be directed , have weighted connections , be bipartite . methods to deal with such systems have been developed , as we have seen , especially in the most recent literature , but they are mostly preliminary attempts and there is room for improvement . another situation that may occur in real systems is the presence of edges with positive and negative weights , indicating attractive and repulsive interactions , respectively . this is the case , for instance , of correlation data @xcite . in this case , ideal partitions would have positively weighted intracluster edges and negatively weighted intercluster edges . we have discussed some studies in this direction @xcite , but we are just at the beginning of this endeavour . instead , there are no algorithms yet which are capable to deal with graphs in which there are edges of several types , indicating different kinds of interactions between the vertices ( _ multigraphs _ ) . agents of social networks , for instance , may be joined by working relationships , friendship , family ties , etc . at the moment there are essentially two ways of proceeding in these instances : 1 ) keeping edges of one type and forgetting the others , repeating the analysis for each type of edges and eventually comparing the results obtained ; 2 ) analyzing a single ( weighted ) graph , obtained by `` combining '' the contributions of the different types of edges in some way . finally , since most real networks are built through the results of experiments , which carry errors in their estimates , it would be useful to consider as well the case in which edges have not only associated weights , but also errors on their values . since the paper by palla et al . @xcite , overlapping communities have received a lot of attention ( section [ sec45 ] ) . however , there is still no consensus about a quantitative definition of the concept of overlapping community , and most definitions depend on the method adopted . intuitively , one would expect that clusters share vertices lying at their borders , and this idea has inspired most algorithms . however , clusters detected with the clique percolation method ( section [ sec45_1 ] ) often share central vertices of the clusters , which makes sense in specific instances , especially in social networks . so , it is still unclear how to characterize overlapping vertices . moreover , the concept of overlapping clusters seems at odds with that of hierarchical structure . no dendrogram can be drawn if there are overlapping vertices , at least in the standard way . due to the relevance of both features in real networks , it is necessary to adapt them to each other in a consistent way . overlapping vertices pose problems as well when it comes to comparing the results of different methods on the same graph . most similarity measures are defined only in the case of partitions , where each vertex is assigned to a single cluster ( section [ sec6_2 ] ) . it is then necessary to extend such definitions to the case of overlapping communities , whenever possible . another issue that is getting increasingly more popular is the study of graphs evolving in time . this is now possible due to the availability of timestamped network data sets . tracking the evolution of community structure in time is very important , to uncover how communities are generated and how they interact with each other . scholars have just begun to study this problem @xcite ( section [ sec7_1_2 ] ) . typically one analyzes separately snapshots at different times and checks what happened at time @xmath690 to the communities at time @xmath156 . it would be probably better to use simultaneously the whole dynamic data set , and future work shall aim at defining proper ways to do that . in this respect , the evolutionary clustering framework by chakrabarti et al . @xcite is a promising starting point . the computational complexity of graph clustering algorithms has improved by at least one power in the graph size ( on average ) in just a couple of years . due to the large size of many systems one wishes to investigate , the ultimate goal would be to design techniques with linear or even sublinear complexity . nowadays partitions in graphs with up to millions of vertices can be found . however , the results are not yet very reliable , as they are usually obtained by greedy optimizations , which yield rough approximations of the desired solution . in this respect the situation could improve by focusing on the development of efficient local methods , for two reasons : 1 ) they enable analyses of portions of the graph , independently of the rest ; 2 ) they are often suitable for parallel implementations , which may speed up considerably the computation . finally , if there has been a tremendous effort in the design of clustering algorithms , basically nothing has been done to make sense of their results . what shall we do with communities ? what can they tell us about a system ? the hope is that they will enable one to disclose `` hidden '' relationships between vertices , due to features that are not known , because they are hard to measure , for instance . it is quite possible that the scientific community will converge sooner or later to a definition _ a posteriori _ of community . already now , most algorithms yield similar results in practical applications . but what is the relationship between the vertex classification given by the algorithms and real classifications ? this is the main question beneath the whole endeavor . 0.5 cm * acknowledgments * i am indebted to these people for giving useful suggestions and advice to improve this manuscript at various stages : a. arenas , j. w. berry , a. clauset , p. csermely , s. gmez , s. gregory , v. gudkov , r. guimer , y. ispolatov , r. lambiotte , a. lancichinetti , j .- onnela , g. palla , m. a. porter , f. radicchi , j. j. ramasco , c. wiggins . i gratefully acknowledge ictecollective , grant number 238597 of the european commission . a _ graph _ @xmath313 is a pair of sets @xmath907 , where @xmath638 is a set of _ vertices _ or _ nodes _ and @xmath908 is a subset of @xmath909 , the set of unordered pairs of elements of @xmath638 . the elements of @xmath908 are called _ edges _ or _ links _ , the two vertices that identify an edge are called _ endpoints_. an edge is _ adjacent _ to each of its endpoints . if each edge is an ordered pair of vertices one has a _ directed _ graph ( or _ digraph _ ) . in this case an ordered pair @xmath910 is an edge directed from @xmath29 to @xmath752 , or an edge beginning at @xmath29 and ending at @xmath752 . a graph is visualized as a set of points connected by lines , as shown in fig . [ fig1 ] . in many real examples , graphs are _ weighted _ , i. e. a real number is associated to each of the edges . graphs do not include _ loops _ , i. e. edges connecting a vertex to itself , nor multiple edges , i. e. several edges joining the same pair of vertices . graphs with loops and multiple edges are called _ multigraphs_. generalizations of graphs admitting edges between any number of vertices ( not necessarily two ) are called _ hypergraphs_. a graph @xmath911 is a _ subgraph _ of @xmath912 if @xmath913 and @xmath914 . if @xmath380 contains all edges of @xmath313 that join vertices of @xmath915 one says that the subgraph @xmath380 is induced or spanned by @xmath915 . a partition of the vertex set @xmath638 in two subsets @xmath87 and @xmath916 is called a _ cut _ ; the _ cut size _ is the number of edges of @xmath313 joining vertices of @xmath87 with vertices of @xmath916 . we indicate the number of vertices and edges of a graph with @xmath15 and @xmath14 , respectively . the number of vertices is the _ order _ of the graph , the number of edges its _ size_. the maximum size of a graph equals the total number of unordered pairs of vertices , @xmath41 . if @xmath917 and @xmath918 , the graph is a _ clique _ ( or _ complete graph _ ) , and is indicated as @xmath919 . two vertices are _ neighbors _ ( or _ adjacent _ ) if they are connected by an edge . the set of neighbors of a vertex @xmath29 is called _ neighborhood _ , and we shall denote it with @xmath920 . the _ degree _ @xmath921 of a vertex @xmath29 is the number of its neighbors . the _ degree sequence _ is the list of the degrees of the graph vertices , @xmath922 . on directed graphs , one distinguishes two types of degree for a vertex @xmath29 : the _ indegree _ , i. e. the number of edges beginning at @xmath29 and the _ outdegree _ , i. e. the number of edges ending at @xmath29 . the analogue of degree on a weighted graph is the _ strength _ , i. e. the sum of the weights of the edges adjacent to the vertex . another useful local property of graphs is _ transitivity _ or _ clustering _ @xcite , which indicates the level of cohesion between the neighbors of a vertex . the clustering coefficient @xmath923 of vertex @xmath29 is the ratio between the number of edges joining pairs of neighbors of @xmath29 and the total number of possible edges , given by @xmath924 , @xmath925 being the degree of @xmath29 . according to this definition , @xmath923 measures the probability that a pair of neighbors of @xmath29 are connected . since all neighbors of @xmath29 are connected to @xmath29 by definition , edges connecting pairs of neighbors of @xmath29 form triangles with @xmath29 . this is why the definition is often given in terms of number of triangles . a _ path _ is a graph @xmath926 , with @xmath927 and @xmath928 . the vertices @xmath929 and @xmath930 are the _ endvertices _ of @xmath92 , whereas @xmath76 is its _ length_. given the notions of vertices , edges and paths , one can define the concept of _ independence_. a set of vertices ( or edges ) of a graph are independent if no two elements of them are adjacent . similarly , two paths are independent if they only share the endvertices . a _ cycle _ is a closed path whose vertices and edges are all distinct . cycles of length @xmath76 are indicated with @xmath931 . the smallest non - trivial cycle is the _ triangle _ , @xmath932 . paths allow to define the concept of connectivity and distance in graphs . a graph is _ connected _ if , given any pair of vertices , there is at least one path going from one vertex to the other . in general , there may be multiple paths connecting two vertices , with different lengths . shortest path _ , or _ geodesic _ , between two vertices of a graph , is a path of minimal length . such minimal length is the _ distance _ between the two vertices . the _ diameter _ of a connected graph is the maximal distance between two vertices . if there is no path between two vertices , the graph is divided in at least two connected subgraphs . each maximal connected subgraph of a graph is called _ connected component_. a graph without cycles is a _ forest_. a connected forest is a _ tree_. trees are very important in graph theory and deserve some attention . in a tree , there can be only one path from a vertex to any other . in fact , if there were at least two paths between the same pair of vertices they would form a cycle , while the tree is an acyclic graph by definition . further , the number of edges of a tree with @xmath15 vertices is @xmath143 . if any edge of a tree is removed , it would get disconnected in two parts ; if a new edge is added , there would be at least one cycle . this is why a tree is a minimally connected , maximally acyclic graph of a given order . every connected graph contains a _ spanning tree _ , i. e. a tree sharing all vertices of the graph . on weighted graphs , one can define a _ minimum _ ( _ maximum _ ) _ spanning tree _ , i. e. a spanning tree such that the sum of the weights on the edges is minimal ( maximal ) . minimum and maximum spanning trees are often used in graph optimization problems , including clustering . a graph @xmath313 is _ bipartite _ if the vertex set @xmath638 is separated in two disjoint subsets @xmath933 and @xmath934 , or _ classes _ , and every edge joins a vertex of @xmath933 with a vertex of @xmath934 . the definition can be extended to that of _ @xmath375-partition _ , where the vertex classes are @xmath375 and no edge joins vertices within the same class . in this case one speaks of _ multipartite _ graphs . the whole information about the topology of a graph of order @xmath15 is entailed in the _ adjacency matrix _ * a * , which is an @xmath167 matrix whose element @xmath360 equals @xmath2 if there is an edge joining vertices @xmath67 and @xmath68 , otherwise it is zero . due to the absence of loops the diagonal elements of the adjacency matrix are all zero . for an undirected graph * a * is a symmetric matrix . the sum of the elements of the @xmath67-th row or column yields the degree of node @xmath67 . if the edges are weighted , one defines the _ weight matrix _ @xmath935 , whose element @xmath317 expresses the weight of the edge between vertices @xmath67 and @xmath68 . the _ spectrum _ of a graph @xmath313 is the set of eigenvalues of its adjacency matrix * a*. spectral properties of graph matrices play an important role in the study of graphs . for instance , the _ stochastic matrices _ rule the process of diffusion ( random walk ) on a graph . right stochastic matrix _ @xmath448 is obtained from * a * by dividing the elements of each row @xmath67 by the degree of vertex @xmath67 . the _ left stochastic matrix _ @xmath201 , or _ transfer matrix _ , is the transpose of @xmath448 . the spectra of stochastic matrices allow to evaluate , for instance , the mixing time of the random walk , i. e. the time it takes to reach the stationary distribution of the process . the latter is obtained by computing the eigenvector of the transfer matrix corresponding to the largest eigenvalue . another important matrix is the _ @xmath936 , where * d * is the diagonal matrix whose element @xmath937 equals the degree of vertex @xmath67 . the matrix @xmath136 is usually referred to as _ unnormalized laplacian_. in the literature one often uses _ normalized laplacians _ @xcite , of which there are two main forms : @xmath938 and @xmath939 . the matrix @xmath195 is symmetric ; @xmath194 is not symmetric and is closely related to a random walk taking place on the graph . all laplacian matrices have a straightforward extension to the case of weighted graphs . the laplacian is one of the most studied matrices and finds application in many different contexts , like graph connectivity @xcite , synchronization @xcite , diffusion @xcite and graph partitioning @xcite . by construction , the sum of the elements of each row of the laplacian ( normalized or unnormalized ) is zero . this implies that * l * always has at least one zero eigenvalue , corresponding to the eigenvector with all equal components , such as @xmath940 . eigenvectors corresponding to different eigenvalues are all orthogonal to each other . interestingly , * l * has as many zero eigenvalues as there are connected components in the graph . so , the laplacian of a connected graph has but one zero eigenvalue , all others being positive . eigenvectors of laplacian matrices are regularly used in _ spectral clustering _ ( section [ sec4_4 ] ) . in particular , the eigenvector corresponding to the second smallest eigenvalue , called _ fiedler vector _ @xcite , is used for graph bipartitioning , as described in section [ sec4_1 ] . in this section we present the most popular models of graphs introduced to describe real systems , at least to some extent . such graphs are useful null models in community detection , as they do not have community structure , so they can be used for negative tests of clustering algorithms . basic models of complex networks . ( top ) erds - rnyi random graph with @xmath874 vertices and a link probability @xmath941 . ( center ) small world graph la watts - strogatz , with @xmath874 vertices and a rewiring probability @xmath942 . ( bottom ) barabsi - albert scale - free network , with @xmath874 vertices and an average degree of @xmath323 . courtesy by j. j. ramasco.,title="fig:",width=340 ] basic models of complex networks . ( top ) erds - rnyi random graph with @xmath874 vertices and a link probability @xmath941 . ( center ) small world graph la watts - strogatz , with @xmath874 vertices and a rewiring probability @xmath942 . ( bottom ) barabsi - albert scale - free network , with @xmath874 vertices and an average degree of @xmath323 . courtesy by j. j. ramasco.,title="fig:",width=340 ] basic models of complex networks . ( top ) erds - rnyi random graph with @xmath874 vertices and a link probability @xmath941 . ( center ) small world graph la watts - strogatz , with @xmath874 vertices and a rewiring probability @xmath942 . ( bottom ) barabsi - albert scale - free network , with @xmath874 vertices and an average degree of @xmath323 . courtesy by j. j. ramasco.,title="fig:",width=340 ] the oldest model is that of _ random graph _ , proposed by solomonoff and rapoport @xcite and independently by erds and rnyi @xcite . there are two parameters : the number of vertices @xmath15 and the connection probability @xmath45 . each pair of vertices is connected with equal probability @xmath45 independently of the other pairs . the expected number of edges of the graph is @xmath943 , and the expected mean degree @xmath944 . the degree distribution of the vertices of a random graph is binomial , and in the limit @xmath945 , @xmath946 for fixed @xmath420 it converges to a poissonian . therefore , the vertices have all about the same degree , close to @xmath420 ( fig . [ fig2 ] , top ) . the most striking property of this class of graphs is the phase transition observed by varying @xmath420 in the limit @xmath945 . for @xmath947 , the graph is separated in connected components , each of them being microscopic , i. e. occupying but a vanishing portion of the system size . for @xmath948 , instead , one of the components becomes macroscopic ( _ giant component _ ) , i. e. it occupies a finite fraction of the graph vertices . the diameter of a random graph with @xmath15 vertices is very small , growing only logarithmically with @xmath15 . this property ( _ small - world effect _ ) is very common in many real graphs . the first evidence that social networks are characterized by paths of small length was provided by a series of famous experiments conducted by the phychologist stanley milgram @xcite . the expected clustering coefficient of a vertex of a random graph is @xmath45 , as the probability for two vertices to be connected is the same whether they are neighbors of the same vertex or not . real graphs , however , are characterized by far higher values of the clustering coefficient as compared to random graphs of the same size . watts and strogatz @xcite showed that the small world property and high clustering coefficient can coexist in the same system . they designed a class of graphs which result from an interpolation between a regular lattice , which has high clustering coefficient , and a random graph , which has the small - world property . one starts from a ring lattice in which each vertex has degree @xmath44 , and with a probability @xmath45 each edge is rewired to a different target vertex ( fig . [ fig2 ] , center ) . it turns out that low values of @xmath45 suffice to reduce considerably the length of shortest paths between vertices , because rewired edges act as shortcuts between initially remote regions of the graph . on the other hand , the clustering coefficient remains high , since few rewired edges do not perturb appreciably the local structure of the graph , which remains similar to the original ring lattice . for @xmath811 all edges are rewired and the resulting structure is a random graph la erds and rnyi . the seminal paper of watts and strogatz triggered a huge interest towards the graph representation of real systems . one of the most important discoveries was that the distribution of the vertex degree of real graphs is very heterogeneous @xcite , with many vertices having few neighbors coexisting with some vertices with many neighbors . in several cases the tail of this distribution can be described as a power law with good approximation , hence the expression _ scale - free networks_. such degree heterogeneity is responsible for a number of remarkable features of real networks , such as resilience to random failures / attacks @xcite , and the absence of a threshold for percolation @xcite and epidemic spreading @xcite . the most popular model of a graph with a power law degree distribution is the model by barabsi and albert @xcite . a version of the model for directed graphs had been proposed much earlier by de solla price @xcite , building up on previous ideas developed by simon @xcite . the graph is created with a dynamic procedure , where vertices are added one by one to an initial core . the probability for a new vertex to set an edge with a preexisting vertex is proportional to the degree of the latter . in this way , vertices with high degree have large probability of being selected as neighbors by new vertices ; if this happens , their degree further increases so they will be even more likely to be chosen in the future . in the asymptotic limit of infinite number of vertices , this _ rich - gets - richer _ strategy generates a graph with a degree distribution characterized by a power - law tail with exponent @xmath3 . in fig . [ fig2 ] ( bottom ) we show an example of barabsi - albert ( ba ) graph . the clustering coefficient of a ba graph decays with the size of the graph , and it is much lower than in real networks . moreover , the power law decays of the degree distributions observed in real networks are characterized by a range of exponents values ( usually between @xmath323 and @xmath3 ) , whereas the ba model yields a fixed value . however , many refinements of the ba model as well as plenty of different models have been later introduced to account more closely for the features observed in real systems ( for details see @xcite ) . Abstract: