import free_pfpng.main import invpoly.functor import condensed.condensify import laurent_measures.thm69 import normed_free_pfpng.compare import Lbar.ses import laurent_measures.ses2 universe u noncomputable theory open category_theory open_locale nnreal big_operators namespace invpoly open ProFiltPseuNormGrpWithTinv₁ variables (p : ℝ≥0) [fact (0 < p)] [fact (p < 1)] local notation `r` := @r p /-- The "evaluate at 2" function, sending an S-indexed family of elements of `ℤ[T⁻¹]` to an S-indexed family of integers, via the assignment `T⁻¹ ↦ 2`. Note that this is a strict (i.e. norm-nonincreasing) morphism of normed groups, with the norm on a polynomial `∑aₙT⁻ⁿ` being `∑ₙ∥aₙ∥₊r⁻ⁿ` and the norm on an integer `z` being `∥z∥₊ᵖ`; this works because `r = 2⁻ᵖ`, the crucial points being that `(∑∥cₙ∥)ᵖ ≤ ∑(∥cₙ∥ᵖ)` as `0 < p ≤ 1`, and `∥z∥ᵖ ≤ ∥z∥` for an integer `z`. -/ @[simps] def eval2 (S : Fintype.{u}) : strict_comphaus_filtered_pseudo_normed_group_hom (invpoly r S) (normed_free_pfpng p S) := { to_fun := λ F s, (F s).eval 2, map_zero' := by { ext, simp only [polynomial.eval_zero, pi.zero_apply], }, map_add' := by { intros, ext, simp only [polynomial.eval_add, pi.add_apply], }, strict' := λ c F hF, begin refine (finset.sum_le_sum _).trans hF, rintro s -, have h0p : 0 < p := fact.out _, have hp1 : p ≤ 1 := fact.out _, have h0pinv : 0 ≤ p⁻¹, { rw ← nnreal.inv_pos at h0p, exact h0p.le }, calc ∥(F s).eval 2∥₊ ^ (p:ℝ) = ∥∑ n in finset.range ((F s).nat_degree + 1), (F s).coeff n * 2 ^ n∥₊ ^ (p:ℝ) : _ ... ≤ (∑ n in finset.range ((F s).nat_degree + 1), ∥(F s).coeff n * 2 ^ n∥₊) ^ (p:ℝ) : _ ... ≤ ∑ n in finset.range ((F s).nat_degree + 1), (∥(F s).coeff n * 2 ^ n∥₊ ^ (p:ℝ)) : _ ... ≤ ∑ n in finset.range ((F s).nat_degree + 1), (∥(F s).coeff n∥₊ * r ^ (-n:ℤ)) : _ ... ≤ ∑' n, (∥(F s).coeff n∥₊ * r ^ (-n:ℤ)) : _, { rw polynomial.eval_eq_sum_range, }, { refine nnreal.rpow_le_rpow (nnnorm_sum_le _ _) h0p.le, }, { refine nnreal.rpow_sum_le_sum_rpow _ _ h0p hp1, }, { refine finset.sum_le_sum _, intros n hn, rw [int.nnnorm_mul, nnreal.mul_rpow], refine mul_le_mul' _ (le_of_eq _), { exact nnnorm_int_rpow_le p _ }, { have h2n : (2 ^ n : ℤ) = (2 ^ n : ℕ), { norm_cast, }, simp only [h2n, ← nnreal.coe_nat_abs, int.nat_abs_of_nat], rw [zpow_neg, ← inv_zpow, zpow_coe_nat], calc ((2 ^ n : ℕ) : ℝ≥0) ^ (p:ℝ) = (2 ^ (n:ℝ) : ℝ≥0) ^ (p:ℝ) : _ ... = (2⁻¹ ^ (p:ℝ) : ℝ≥0)⁻¹ ^ (n:ℝ) : _ ... = r⁻¹ ^ n : _, { norm_cast }, { rw [← nnreal.rpow_mul, mul_comm, nnreal.rpow_mul, ← nnreal.inv_rpow, inv_inv], }, { rw [nnreal.rpow_nat_cast], refl, } } }, { refine sum_le_tsum _ _ _, { intros, exact zero_le' }, { exact F.nnreal_summable s } }, end, continuous' := λ c, continuous_of_discrete_topology } /-- The "evaluate a polynomial in `T⁻¹` by mapping `T⁻¹` to `2`" functor from `ℤ[T⁻¹]` to `ℤ`. -/ def eval2_nat_trans : (Fintype_invpoly.{u} r ⋙ PFPNGT₁_to_CHFPNG₁ₑₗ.{u} r) ⟶ (normed_free_pfpng_functor.{u} p ⋙ PFPNG₁_to_CHFPNG₁ₑₗ) := { app := λ S, eval2 p S, naturality' := λ S T f, begin ext x t, show (map f x t).eval 2 = _, dsimp only [functor.comp_map, normed_free_pfpng_functor_map, CompHausFiltPseuNormGrp₁.comp_apply, map], erw [polynomial.eval_finset_sum], refl, recover, exact 0, end } . section ses open CompHausFiltPseuNormGrp₁ -- move this instance (M N : Type*) [comphaus_filtered_pseudo_normed_group M] [comphaus_filtered_pseudo_normed_group N] : add_monoid_hom_class (comphaus_filtered_pseudo_normed_group_hom M N) M N := { coe := λ f, f, coe_injective' := λ f g h, by { ext, dsimp at h, rw h }, map_add := λ f, f.map_add, map_zero := λ f, f.map_zero } lemma Tinv2_injective (S : Fintype) : function.injective ((Tinv2_nat_trans (Fintype_invpoly r)).app S) := begin rw injective_iff_map_eq_zero, intros f hf, ext s n, apply_fun (λ φ, φ s) at hf, simp only [Tinv2_nat_trans, Tinv_nat_trans, nat_trans.app_nsmul, nat_trans.id_app, sub_apply, nat_trans.app_sub, comphaus_filtered_pseudo_normed_group_hom.sub_apply, pi.nsmul_apply, comphaus_filtered_pseudo_normed_group_hom.nsmul_apply, pi.sub_apply, pi.zero_apply] at hf, simp only [pi.zero_apply, polynomial.coeff_zero], induction n with n ih, { apply_fun (λ φ, φ.coeff 0) at hf, simp only [polynomial.coeff_zero, polynomial.coeff_sub, polynomial.coeff_smul] at hf, erw [invpoly.Tinv_zero, zero_sub, neg_eq_zero, two_nsmul_eq_zero ℤ ℤ] at hf, exact hf, }, { apply_fun (λ φ, φ.coeff (n+1)) at hf, simp only [polynomial.coeff_zero, polynomial.coeff_sub, polynomial.coeff_smul] at hf, erw [invpoly.Tinv_succ, ih, zero_sub, neg_eq_zero, two_nsmul_eq_zero ℤ ℤ] at hf, exact hf } end lemma Tinv2_comp_eval2_eq_zero (S : Fintype) : (Tinv2_nat_trans (Fintype_invpoly r)).app S ≫ (whisker_right (eval2_nat_trans p) CHFPNG₁_to_CHFPNGₑₗ).app S = 0 := begin ext f s, show (polynomial.X * (f s) - _).eval 2 = (0 : ℤ), simp only [nat_trans.app_nsmul, nat_trans.id_app, comphaus_filtered_pseudo_normed_group_hom.nsmul_apply, category_theory.id_apply, pi.smul_apply, nsmul_eq_mul, nat.cast_bit0, nat.cast_one, polynomial.eval_sub, polynomial.eval_mul, polynomial.eval_X, polynomial.eval_bit0, polynomial.eval_one, sub_self], end . lemma theta_eval2_aux {S : Fintype} (s : S) (c : ℝ≥0) {f : invpoly r S} : ∑' (n : ℤ), (to_laurent_measures_fun r S f s n : ℝ) * ((2 : ℝ) ^ n)⁻¹ = ↑(polynomial.eval 2 (f s)) := begin rw polynomial.eval_eq_sum_range, have this : ∀ (n : ℤ), n ∉ finset.Icc (-(f s).nat_degree : ℤ) 0 → (λ (b : ℤ), ↑(to_laurent_measures_fun «r» S f s b) * ((2 : ℝ) ^ b)⁻¹) n = 0, { rintros n hn, rw [finset.mem_Icc, not_and_distrib] at hn, push_neg at hn, dsimp only, convert zero_mul _, norm_cast, cases hn with hn hn, { have : ∃ m : ℕ, n = - m, { use n.nat_abs, rw int.nat_abs_of_nonpos, { simp }, exact le_trans hn.le (by simp), }, rcases this with ⟨m, rfl⟩, rw to_laurent_measures_fun_nonpos, rw polynomial.coeff_eq_zero_of_nat_degree_lt, simpa using hn }, { have : ∃ m : ℕ, n = m + 1, { use (n - 1).nat_abs, rw int.nat_abs_of_nonneg, ring, linarith, }, rcases this with ⟨m, rfl⟩, apply to_laurent_measures_fun_pos' } }, rw tsum_eq_sum this, clear this, push_cast, apply finset.sum_bij (λ (n : ℤ) hn, n.nat_abs), -- this is supposed to be the easy one! all that remains -- is four boring pieces of easy nonsense e.g. --2^{-n}=2^{n.nat_abs} if n < 0 is probably the hardest one { intros a ha, suffices : a.nat_abs < (f s).nat_degree + 1, { simpa }, rw finset.mem_Icc at ha, rw [nat.lt_succ_iff, ← int.coe_nat_le, int.nat_abs_of_nonpos ha.2], linarith [ha.1] }, { intros a ha, rw finset.mem_Icc at ha, simp only, congr, { have : ∃ m : ℕ, a = - m, { use a.nat_abs, rw int.nat_abs_of_nonpos ha.2, simp, }, rcases this with ⟨m, rfl⟩, rw to_laurent_measures_fun_nonpos, congr, rw [← int.coe_nat_eq_coe_nat_iff, int.nat_abs_of_nonpos ha.2, neg_neg], }, { rw [← zpow_neg, ← int.nat_abs_of_nonpos ha.2, zpow_coe_nat], } }, { rintros a b ha hb (habs : a.nat_abs = b.nat_abs), rw finset.mem_Icc at ha hb, rw [← int.coe_nat_eq_coe_nat_iff, int.nat_abs_of_nonpos ha.2, int.nat_abs_of_nonpos hb.2] at habs, exact neg_inj.mp habs }, { intros a ha, rw [finset.mem_range_succ_iff] at ha, refine ⟨-(a : ℤ), _, _⟩, { rw [finset.mem_Icc, neg_le_neg_iff, neg_le, neg_zero], simp [ha], }, { simp } }, end lemma theta_zero_of_eval2_zero {S : Fintype} (c : ℝ≥0) {f : invpoly r S} (hf1 : eval2 p S f = 0) : (laurent_measures.Θ p S) ((to_laurent_measures_addhom r S) f) = 0 := begin ext s, replace hf1 := congr_fun hf1 s, suffices : ∑' (n : ℤ), ↑(to_laurent_measures_fun «r» S f s n) * ((2 : ℝ) ^ n)⁻¹ = 0, { simpa [laurent_measures.Θ, laurent_measures_ses.θ_to_add, θ, theta.ϑ, to_laurent_measures_addhom], }, simp only [theta_eval2_aux p s c, int.cast_eq_zero], exact hf1, end . lemma _root_.int.eq_neg_nat_or_succ_nat (n : ℤ) : ∃ m : ℕ, n = -m ∨ n = m + 1 := begin cases le_or_lt n 0, { use n.nat_abs, left, rw [int.nat_abs_of_nonpos h, neg_neg] }, { use (n - 1).nat_abs, right, rw [int.nat_abs_of_nonneg], ring, linarith }, end lemma phi_eq_mul_Tinv_sub_two {S : Fintype} (g : invpoly r S) : ϕ ((to_laurent_measures_addhom r S) (λ (s : ↥S), g s)) = (to_laurent_measures_addhom r S) (λ (s : ↥S), (polynomial.X - 2) * g s) := begin ext s n, delta ϕ, simp only [laurent_measures.sub_apply, laurent_measures.shift_to_fun_to_fun, laurent_measures.nsmul_apply, nsmul_eq_mul, int.coe_nat_zero, zero_add], obtain ⟨m, (rfl | rfl)⟩ := n.eq_neg_nat_or_succ_nat, { simp only [to_laurent_measures_addhom, to_laurent_measures_fun_nonpos, add_monoid_hom.mk'_apply, to_laurent_measures_to_fun], cases m, { simp only [nat.nat_zero_eq_zero, int.coe_nat_zero, neg_zero', zero_add, polynomial.mul_coeff_zero, polynomial.coeff_sub, polynomial.coeff_X_zero, zero_sub, neg_mul], rw [to_laurent_measures_fun_pos'' r S g s (show (0 : ℤ) < 1, by norm_num)], simp only [zero_sub, neg_inj, mul_eq_mul_right_iff, show (2 : polynomial ℤ) = polynomial.C 2, by simp, polynomial.coeff_C_zero], left, norm_num, }, { simp only [to_laurent_measures_fun_nonpos, sub_mul, int.coe_nat_succ, neg_add_rev, neg_add_cancel_comm, int.coe_nat_bit0, int.coe_nat_zero, zero_add, polynomial.coeff_sub, polynomial.coeff_X_mul, sub_right_inj, show (2 : polynomial ℤ) = polynomial.C 2, by simp, polynomial.coeff_C_mul], } }, { norm_cast }, -- didn't expect that to close it! end lemma psi_eq_div_X_sub_two {S : Fintype} (c : ℝ≥0) {f : invpoly r S} (hf1 : ((eval2_nat_trans p).app S) f = 0) : (to_laurent_measures_addhom r S) (λ (s : ↥S), f s /ₘ (polynomial.X - 2)) = ψ ((to_laurent_measures_addhom «r» S) f) (theta_zero_of_eval2_zero p c hf1) := begin apply injective_ϕ', rw θ_ϕ_split_exact, have hf3 : ∀ s : S, (polynomial.X - 2) * ((f s) /ₘ (polynomial.X - 2)) = f s, { intro s, replace hf1 := congr_fun hf1 s, rw (show (2 : polynomial ℤ) = polynomial.C (2 : ℤ), by simp), rwa polynomial.mul_div_by_monic_eq_iff_is_root, }, suffices : (to_laurent_measures_addhom r S) (λ s, (polynomial.X - 2) * ((f s) /ₘ (polynomial.X - 2))) = (to_laurent_measures_addhom r S) (λ s, f s), { rw ← this, apply phi_eq_mul_Tinv_sub_two, }, simp [hf3], end theorem short_exact (S : Profinite) : short_exact ((condensify_Tinv2 _).app S) ((condensify_map $ eval2_nat_trans p).app S) := begin let κ : ℝ≥0 → ℝ≥0 := λ c, max c (c ^ (↑p⁻¹ : ℝ)), have hκ : id ≤ κ := λ c, le_max_left _ _, have h0p : 0 < p := fact.out _, have hp1 : p ≤ 1 := fact.out _, have h0pinv : 0 ≤ p⁻¹, { rw ← nnreal.inv_pos at h0p, exact h0p.le }, refine condensify_nonstrict_exact _ _ (r⁻¹ + 2) (Tinv2_bound_by _) -- next line can be simplified because 1/2 < r < 1 so the max is always on the right. (λ c, max c (c * (2 - r⁻¹)⁻¹ * (r⁻¹ + 2))) κ (λ c, le_max_left _ _) hκ (Tinv2_injective p) (Tinv2_comp_eval2_eq_zero p) _ _ _, { rintros S c f ⟨hf1, hf2⟩, have hf3 : ∀ s : S, (polynomial.X - 2) * ((f s) /ₘ (polynomial.X - 2)) = f s, { intro s, replace hf1 := congr_fun hf1 s, rw (show (2 : polynomial ℤ) = polynomial.C (2 : ℤ), by simp), rwa polynomial.mul_div_by_monic_eq_iff_is_root, }, refine ⟨λ s, (f s) /ₘ (polynomial.X - 2), _, _⟩, { change ∥_∥₊ ≤ c at hf2, change ∥_∥₊ ≤ _, have hf1' : ((eval2_nat_trans p).app S) f = 0, { simpa using hf1 }, rw ← to_laurent_measures_addhom_isometry at ⊢ hf2, refine le_trans _ (nnreal.mul_le_mul_right (le_max_right _ _) _), rw mul_inv_cancel_right₀ (lt_of_lt_of_le (zero_lt_two : (0 : ℝ≥0) < 2) le_add_self).ne', convert laurent_measures.psi_bound p S c (theta_zero_of_eval2_zero p c hf1') hf2, exact psi_eq_div_X_sub_two p c hf1, }, { ext1 s, convert hf3 s, simp [sub_mul, Tinv2_nat_trans, Tinv_nat_trans], refl } }, { rintro S c g (hg : _ ≤ _), let f : invpoly r S := λ s, polynomial.C (g s), refine ⟨f, _, _⟩, { show _ ≤ _, replace hg := nnreal.rpow_le_rpow hg h0pinv, refine le_trans _ (le_max_right _ _), refine le_trans _ hg, refine le_trans _ (nnreal.rpow_le_rpow (nnreal.rpow_sum_le_sum_rpow _ _ h0p hp1) h0pinv), rw [← nnreal.rpow_mul, ← nnreal.coe_mul, mul_inv_cancel h0p.ne', nnreal.coe_one, nnreal.rpow_one, invpoly.nnnorm_def], refine finset.sum_le_sum _, rintro s -, simp only [f, zpow_neg, zpow_coe_nat, id.def, polynomial.coeff_C], rw tsum_eq_single 0, { rw [if_pos rfl, pow_zero, inv_one, mul_one], }, { intros n hn, rw [if_neg hn, nnnorm_zero, zero_mul], }, { apply_instance }, }, { ext s, apply polynomial.eval_C } }, end end ses end invpoly