import field_theory.perfect_closure import for_mathlib.nnreal import valuation.basic lemma iterate_frobenius_apply {α : Type*} [monoid α] (p n : ℕ) (a : α) : (frobenius α p)^[n] a = a^p^n := begin induction n with n ih, {simp}, rw [nat.iterate_succ', ih, frobenius_def, ← pow_mul, nat.pow_succ] end noncomputable theory open_locale classical namespace valuation variables {R : Type*} [comm_ring R] variables (v : valuation R nnreal) namespace perfection open nnreal variables (p : ℕ) [nat.prime p] [char_p R p] private def f₀ : ℕ × R → nnreal := λ (x : ℕ × R), (v x.2)^(p^(-x.1 : ℤ) : ℝ) private def hf₀ : ∀ (x₁ x₂ : ℕ × R) (h : perfect_closure.r R p x₁ x₂), f₀ v p x₁ = f₀ v p x₂ | x₁ x₂ (perfect_closure.r.intro _ n x) := show v x ^ (p ^ (-n : ℤ) : ℝ) = v (x ^ p) ^ (p ^ (-(↑n + 1) : ℤ) : ℝ), from have hp : (p : ℝ) ≠ 0 := by exact_mod_cast nat.prime.ne_zero ‹_›, by rw [valuation.map_pow, ← rpow_nat_cast, ← rpow_mul, neg_add_rev, fpow_add hp, ← mul_assoc, fpow_inv, mul_inv_cancel hp, one_mul] def f : perfect_closure R p → nnreal := quot.lift (f₀ v p) (hf₀ v p) lemma f_zero : f v p (0 : perfect_closure R p) = 0 := calc f v p (0 : perfect_closure R p) = quot.lift (f₀ v p) (hf₀ v p) (0 : perfect_closure R p) : rfl ... = (v (0:R))^(p^(-0:ℤ) : ℝ) : quot.lift_beta (f₀ v p) (hf₀ v p) (0, 0) ... = 0 : by rw [v.map_zero, neg_zero, fpow_zero, rpow_one] lemma f_one : f v p (1 : perfect_closure R p) = 1 := calc f v p (1 : perfect_closure R p) = quot.lift (f₀ v p) (hf₀ v p) (1 : perfect_closure R p) : rfl ... = (v (1:R))^(p^(-0:ℤ) : ℝ) : quot.lift_beta (f₀ v p) (hf₀ v p) (0, 1) ... = 1 : by rw [v.map_one, neg_zero, fpow_zero, rpow_one] lemma f_mul (r s : perfect_closure R p) : f v p (r * s) = f v p r * f v p s := quot.induction_on r $ λ ⟨m,x⟩, quot.induction_on s $ λ ⟨n,y⟩, show f₀ v p (m + n, _) = f₀ v p (m,x) * f₀ v p (n,y), from have hp : p ≠ 0 := nat.prime.ne_zero ‹_›, have hpQ : (p : ℝ) ≠ 0 := by exact_mod_cast hp, begin clear _fun_match _fun_match _x _x, dsimp only [f₀], simp only [iterate_frobenius_apply, v.map_mul, v.map_pow, mul_rpow], congr' 1, all_goals { rw [← rpow_nat_cast, ← rpow_mul, nat.cast_pow, ←fpow_of_nat, ← fpow_add hpQ], { congr, rw [int.coe_nat_add, neg_add], abel }, }, end lemma f_add (r s : perfect_closure R p) : f v p (r + s) ≤ max (f v p r) (f v p s) := quot.induction_on r $ λ ⟨m,x⟩, quot.induction_on s $ λ ⟨n,y⟩, show f₀ v p (m + n, _) ≤ max (f₀ v p (m,x)) (f₀ v p (n,y)), from have hp : p ≠ 0 := nat.prime.ne_zero ‹_›, have hpQ : (p : ℝ) ≠ 0 := by exact_mod_cast hp, begin clear _fun_match _fun_match _x _x, dsimp only [f₀], rw [iterate_frobenius_apply, iterate_frobenius_apply], have h := v.map_add (x^p^n) (y^p^m), rw le_max_iff at h ⊢, cases h with h h; [ {left, rw add_comm m}, right], all_goals { conv_rhs at h { rw v.map_pow }, refine le_trans (rpow_le_rpow _ h (fpow_nonneg_of_nonneg (nat.cast_nonneg p) _)) (le_of_eq _), rw [← rpow_nat_cast, ← rpow_mul, nat.cast_pow, ←fpow_of_nat, ← fpow_add hpQ], { congr, rw [int.coe_nat_add, neg_add], abel }, } end end perfection section variables (p : ℕ) [nat.prime p] [char_p R p] def perfection : valuation (perfect_closure R p) nnreal := { to_fun := perfection.f v p, map_zero' := perfection.f_zero v p, map_one' := perfection.f_one v p, map_mul' := perfection.f_mul v p, map_add' := begin -- TODO(jmc): This is really ugly. But Lean doesn't cooperate. -- It finds two instances that aren't defeq. intros r s, convert perfection.f_add v p r s, delta classical.DLO nnreal.decidable_linear_order, congr, end } end end valuation