PINNeAPPle-Labs/blackhole-riaf-density-256

Accretion onto a Schwarzschild black hole: 2.5-D viscous hydro density frames (256 x 128)

Radiatively inefficient accretion flow grown from an equilibrium torus (Paczynski-Wiita potential, Shakura-Sunyaev alpha viscosity, gamma 5/3, r = 4-400 GM/c^2, log-spaced 256 x 128 grid): the set-up of Duarte, Nemmen & Navarro (2022), regenerated with pinneapple_physics.blackhole. Density in the forecaster's window (r < 126.5 GM/c^2, theta in [pi/8, 7 pi/8]) every 20 GM/c^3 for t = 2000-10000 GM/c^3, whole trajectories per split. The training data of the black-hole weather forecaster (#399).

blackhole_riaf_density_256 v1.0.0 (revision v1.0.0.post1) · licence cc-by-4.0 · part of the PINNeAPPle zoo

Configs and splits

Config	Split	Samples	Parquet rows	Parquet files	Content
default	train	1604	1604	1	seeds 0-3, alpha 0.1: 4 whole trajectories, 401 frames each
default	val	401	401	1	seed 4, alpha 0.1 (whole trajectory)
default	test	401	401	1	seed 5, alpha 0.1 (whole trajectory, held out)
default	ood_alpha0.3	401	401	1	alpha 0.3, seed 0: other physics, never used in training
convergence	convergence	12	12	1	resolution study: one row per grid and seed (90x45 to 256x128, seeds 0-2)
grid_r	grid_r	192	192	1	radial cells of the window
grid_theta	grid_theta	96	96	1	theta cells of the window

Split rule: by trajectory (seed), never by time inside a trajectory: train seeds 0-3, val 4, test 5

Columns of config default

Column	Unit	Type	Description
alpha	-	float32	Shakura-Sunyaev viscosity parameter
log10_rho_q	-	uint16[192, 96]	uint16 (frames, 192 r, 96 theta): log10 rho = -8 + q * 9 / 65535 in the window (pinneapple_physics.blackhole.dataset.decode)
mdot	code units	float32	accretion rate through r_in = 4 GM/c^2 (density maximum of the initial torus = 1)
run	-	int16	trajectory: the seed of the initial torus perturbation
t	GM/c^3	float32	time

Columns of config convergence

Column	Unit	Type	Description
ell_over_kepler_r10	-	float64	time-mean specific angular momentum / Keplerian (Paczynski-Wiita), r = 10
ell_over_kepler_r30	-	float64	same at r = 30 GM/c^2
mdot_detrended_rms	-	float64	rms of mdot / (500 GM/c^3 running mean) - 1
mdot_mean	code units	float64	time-mean accretion rate
mdot_psd_slope	-	float64	log-log slope of the accretion-rate power spectrum, 1/2000 to 1/50 per GM/c^3
mdot_rms_over_mean	-	float64	rms / mean of the accretion rate (includes the secular drain of the torus)
nr	-	int32	radial cells of the grid
ntheta	-	int32	theta cells of the grid
persistence_mae_1000M	-	float64	same, 1000 GM/c^3
persistence_mae_100M	-	float64	same, 100 GM/c^3
persistence_mae_20M	-	float64	MAE of persistence at 20 GM/c^3 (normalised log density, log10 rho in [-6, 0])
rho_slope_r10_60	-	float64	d ln rho / d ln r of the time-mean shell density, r = 10-60 GM/c^2
seed	-	int32	seed of the initial perturbation
window_mass_mean	code units	float64	time-mean mass in the window

Columns of config grid_r

Column	Unit	Type	Description
r	GM/c^2	float64	radius of the cell centres (volume weighted)
r_face_hi	GM/c^2	float64	outer radial face
r_face_lo	GM/c^2	float64	inner radial face

Columns of config grid_theta

Column	Unit	Type	Description
theta	rad	float64	polar angle of the cell centres
theta_face_hi	rad	float64	upper theta face
theta_face_lo	rad	float64	lower theta face

Load

With the library (the npz files, SHA-256 checked; arrays with their full shape):

pip install "pinneapple @ git+https://github.com/PINNeAPPle-Labs/PINNeAPPle"

import pinneapple as pp

d = pp.datasets.load("blackhole_riaf_density_256", split="test")
print(d.n_rows, d.columns)
print(d["log10_rho_q"].shape)

With datasets (the Parquet files of this repo):

from datasets import load_dataset

ds = load_dataset("PINNeAPPle-Labs/blackhole-riaf-density-256", split="test", revision="v1.0.0.post1")
print(ds)
ds = ds.with_format("numpy")
print(ds[0]["log10_rho_q"].shape)

Splits with other columns are separate configs (convergence, grid_r, grid_theta); pass the config name as the second argument of load_dataset.

Provenance

examples/black_hole_weather/simulate.py --nr 256 --ntheta 128 --viscosity SS --perturbation-kind modes --t-end 10000 --every 10 (runs and seeds in pinneapple_physics.blackhole.dataset.SPLITS); scripts/zoo/build_blackhole.py

Verification

Solver: mass and angular momentum conserved to 1e-16 with the boundary fluxes counted, exact Bondi accretion steady within 0.5 %, inviscid torus in equilibrium (tests/test_blackhole_weather.py). Resolution study (examples/black_hole_weather/convergence.py; results/convergence.json): grids 90x45, 128x64, 181x90, 256x128 (ratio sqrt 2), seeds 0-2 on every grid with the same seeded perturbation, time averages over t = 2000-10000 GM/c^3 with 95 % intervals (batch means and seeds). Converged within the statistical band (the two finest grids agree within their intervals): mean accretion rate 2.18 +- 0.58 (181) vs 1.97 +- 0.63 (256), window mass 2.65e4 vs 2.58e4, inner density slope -0.41 vs -0.38, persistence error at 100 GM/c^3 0.0034 +- 0.0011 vs 0.0047 +- 0.0011. Not converged: from 181 to 256 the time-mean profiles change more than between the coarser grids (density 6.3 % vs 0.8-1.6 %, specific angular momentum 4.2 % vs 0.5 %, angular-momentum flux 11 % vs 1.5-2.5 %; l / l_Kepler at r = 10 rises 3.8 % beyond its interval), so the observed order is negative and no GCI applies; the number of cells where the speed / temperature ceilings act grows with resolution and is the likely cause. The 256 x 128 grid is the finest the study could afford (a 362 x 181 run did not finish); the 181 -> 256 changes are the uncertainty band of these data. The published PLUTO runs and the paper's values could not be retrieved for a direct comparison.

Files

The npz files are the library's format and the source of truth: pp.datasets.load downloads them from this repo at the tagged revision and refuses any file whose SHA-256 differs from the manifest of the package. The Parquet files under data/ are derived from them by pinneapple_hub.zoo.parquet.write_dataset and hold the same numbers (a test round-trips npz -> Parquet -> arrays bit for bit); they are what the dataset viewer and load_dataset read.

File	Size	SHA-256
convergence.npz	0.0 MB	8fd34040dd14e7b1801f680370968744bfe6ac5583e8402758490170955fc08d
grid_r.npz	0.0 MB	c8fa013056c2699e3228899c7398858512725471e7af592e8030c5cfdfc4de6e
grid_theta.npz	0.0 MB	7d391c5b8f0ede9f945696a9ca94d2ded84a655efd96824232fdb73e04d39d0f
ood_alpha0.3.npz	12.0 MB	35a4f1522e66d58d05eed6c099a0096696834c97ee60c14b283d40a2c63c8567
test.npz	13.2 MB	2e455010eddec17985f369739a4d4a833735f3a0b8da358e568abd054cf6590a
train.npz	53.0 MB	8afe0c57ac35b1e57e6f0ac7f1188da754a09a71633ebc38ae1becac1be11aaa
val.npz	13.3 MB	5aff6213fd73543076e096540e4052851212246ff753c3fe8a71f4bd0d76f895

Total 91.5 MB of npz.

Citation

Duarte R., Nemmen R., Navarro J. P. (2022) MNRAS 512, 5848 (set-up and forecaster); Almeida I., Nemmen R. (2020) MNRAS 492, 2553 (simulations); data generated with PINNeAPPle, Zenodo DOI 10.5281/zenodo.23343062.

Cite

If you use this dataset, please cite PINNeAPPle (software, Zenodo DOI 10.5281/zenodo.23343062; see CITATION.cff):

@software{pinneapple,
  author = {Barros, Yan and Schiassi, Enrico},
  title  = {PINNeAPPle},
  doi    = {10.5281/zenodo.23343062},
  url    = {https://github.com/PINNeAPPle-Labs/PINNeAPPle}
}