Datasets:

roseking
/

openAI-math-dataset

import OAI.Combinatorics.CycleDecomposition.EdgeSplitting

namespace OAI

universe cycleUniverse1 cycleUniverse2

section
open Filter Asymptotics Real
open scoped Topology
noncomputable section
open MeasureTheory ProbabilityTheory Finset
noncomputable section
namespace ErdosGallai.Representatives
noncomputable section
open Finset
variable {I : Type cycleUniverse1} {V : Type cycleUniverse2} [DecidableEq I] [DecidableEq V] [Nonempty V]

lemma choose_with_mates (A : Finset I) (mate : I → I)
    (hmate : Function.Involutive mate) (hfix : ∀ i, mate i ≠ i)
    (N : I → Finset V) (k : ℕ)
    (hroom : ∀ i ∈ A, A.card ≤ ((N i).card - 1) * k) :
    ∃ f : I → V, (∀ i ∈ A, f i ∈ N i) ∧
      (∀ i ∈ A, mate i ∈ A → f i ≠ f (mate i)) ∧
      (∀ v, (A.filter (fun i => f i = v)).card ≤ k) := by
  induction A using Finset.induction_on with
  | empty => exact ⟨fun _ => Classical.choice inferInstance, by simp⟩
  | @insert i A hi ih =>
    obtain ⟨f,hf,hsep,hcap⟩ := ih (by
      intro j hj
      exact (Finset.card_le_card (Finset.subset_insert i A)).trans (hroom j (by simp [hj])))
    have hlt : A.card < ((N i).erase (f (mate i))).card * k := by
      have h := hroom i (by simp)
      have hcard := Finset.pred_card_le_card_erase (s := N i) (a := f (mate i))
      have hmul := Nat.mul_le_mul_right k hcard
      rw [Finset.card_insert_of_notMem hi] at h
      omega
    obtain ⟨v,hv,hload⟩ := Finset.exists_card_fiber_lt_of_card_lt_mul (f := f) hlt
    have hvN := (Finset.mem_erase.mp hv).2
    have hvne := (Finset.mem_erase.mp hv).1
    let g := Function.update f i v
    have gi : g i = v := by simp [g]
    have gold : ∀ j, j ≠ i → g j = f j := by
      intro j hj
      simp [g, hj]
    refine ⟨g, ?_, ?_, ?_⟩
    · intro j hj
      rcases Finset.mem_insert.mp hj with rfl | hj
      · simpa only [gi] using hvN
      · rw [gold j (ne_of_mem_of_not_mem hj hi)]
        exact hf j hj
    · intro j hj hmj
      by_cases hji : j = i
      · subst j
        rw [gi, gold (mate i) (hfix i)]
        exact hvne
      · rw [gold j hji]
        by_cases hmi : mate j = i
        · rw [hmi, gi]
          have hjm : j = mate i := by rw [← hmi, hmate]
          simpa [hjm] using hvne.symm
        · rw [gold (mate j) hmi]
          exact hsep j (Finset.mem_of_mem_insert_of_ne hj hji)
            (Finset.mem_of_mem_insert_of_ne hmj hmi)
    · intro w
      have hold : A.filter (fun j => g j = w) = A.filter (fun j => f j = w) := by
        apply Finset.filter_congr
        intro j hj
        rw [gold j (ne_of_mem_of_not_mem hj hi)]
      rw [Finset.filter_insert, hold]
      by_cases hvw : v = w
      · simp only [gi, hvw, ite_eq_left]
        rw [Finset.card_insert_of_notMem (fun h => hi (Finset.mem_filter.mp h).1)]
        simpa only [hvw] using Nat.succ_le_of_lt hload
      · simp only [gi, hvw, ite_false]
        exact hcap w

theorem two_distinct_representatives (S : Finset I) (N : I → Finset V) (k : ℕ)
    (hroom : ∀ i ∈ S, 2*S.card ≤ ((N i).card-1)*k) :
    ∃ f : I × Fin 2 → V,
      (∀ i ∈ S, ∀ t, f (i,t) ∈ N i) ∧
      (∀ i ∈ S, f (i,0) ≠ f (i,1)) ∧
      (∀ v, ((S ×ˢ Finset.univ).filter (fun i => f i = v)).card ≤ k) := by
  let mate : I × Fin 2 → I × Fin 2 := fun i => (i.1, 1-i.2)
  have hm : Function.Involutive mate := by
    rintro ⟨i,t⟩
    fin_cases t <;> rfl
  have hfix : ∀ i, mate i ≠ i := by
    rintro ⟨i,t⟩
    fin_cases t <;> (intro h; have hh := congrArg Prod.snd h; norm_num [mate] at hh)
  obtain ⟨f,hf,hsep,hcap⟩ := choose_with_mates (S ×ˢ Finset.univ) mate hm hfix
    (fun i => N i.1) k (by
      intro i hi
      simpa [Finset.card_product, mul_comm] using hroom i.1 (Finset.mem_product.mp hi).1)
  refine ⟨f, ?_, ?_, hcap⟩
  · intro i hi t
    exact hf (i,t) (by simp [hi])
  · intro i hi
    exact hsep (i,0) (by simp [hi]) (by simp [mate, hi])

theorem reservoir_representatives (S : Finset I) (N : I → Finset V)
    (r d : ℝ) (hr : (S.card : ℝ) ≤ r) (hd : 4 ≤ d)
    (hN : ∀ i ∈ S, d/2 ≤ ((N i).card : ℝ)) :
    ∃ f : I × Fin 2 → V,
      (∀ i ∈ S, ∀ t, f (i,t) ∈ N i) ∧
      (∀ i ∈ S, f (i,0) ≠ f (i,1)) ∧
      (∀ v, ((S ×ˢ Finset.univ).filter (fun i => f i = v)).card ≤ ⌈8*r/d⌉₊) := by
  apply two_distinct_representatives
  intro i hi
  have hd0 : 0 < d := by linarith
  have hr0 : 0 ≤ r := le_trans (Nat.cast_nonneg _) hr
  have hNi := hN i hi
  have hNi1 : 1 ≤ (N i).card := by exact_mod_cast (by linarith : (1 : ℝ) ≤ (N i).card)
  have hn : d/4 ≤ (((N i).card - 1 : ℕ) : ℝ) := by
    rw [Nat.cast_sub hNi1]
    push_cast
    linarith
  have hk : 8*r/d ≤ (⌈8*r/d⌉₊ : ℝ) := Nat.le_ceil _
  have hk0 : 0 ≤ (⌈8*r/d⌉₊ : ℝ) := Nat.cast_nonneg _
  have hprod : 2*(S.card : ℝ) ≤ (((N i).card - 1 : ℕ) : ℝ)*(⌈8*r/d⌉₊ : ℝ) := by
    calc
      _ ≤ 2*r := by linarith
      _ = (d/4)*(8*r/d) := by field_simp; ring
      _ ≤ (d/4)*(⌈8*r/d⌉₊ : ℝ) := mul_le_mul_of_nonneg_left hk (by positivity)
      _ ≤ _ := mul_le_mul_of_nonneg_right hn hk0
  exact_mod_cast hprod

end
end ErdosGallai.Representatives

namespace ErdosGallai
noncomputable section
open Finset Real Filter
attribute [local instance] Classical.propDecidable

noncomputable def reservoirProbability (η : ℝ) : ℝ := min (η/12) (1/12)
noncomputable def reservoirExpansionCoefficient (c : ℝ) : ℝ := min c 1 / 6

universe u

theorem reservoir_preparation (c η : ℝ) (hc : 0 < c) (hη : 0 < η) :
    ∃ D₀ : ℝ, ∀ D : ℝ, D₀ ≤ D → ∀ (V : Type u) [Fintype V] [DecidableEq V]
      (G P : SimpleGraph V) [DecidableRel P.Adj], P ≤ G →
      D^(9/10 : ℝ) ≤ (Fintype.card V : ℝ) →
      (Fintype.card V : ℝ) ≤ D^(51/50 : ℝ) →
      Splitting.CutExpansion P (c*D^(9/10 : ℝ)) →
      ∃ (ω : Sym2 V → Fin 3) (W : Fin 3 → Finset V) (f : Fin 3 → V × Fin 2 → V),
        (∀ i, Splitting.colorGraph P ω i ≤ G) ∧
        (∀ i j, i ≠ j → Disjoint (Splitting.colorGraph P ω i).edgeSet
            (Splitting.colorGraph P ω j).edgeSet) ∧
        (∀ i, Splitting.CutExpansion (Splitting.colorGraph P ω i)
          (reservoirExpansionCoefficient c * D^(9/10 : ℝ))) ∧
        (∀ i j, i ≠ j → Disjoint (W i) (W j)) ∧
        (∀ i, 2 ≤ (W i).card) ∧
        (∀ i, ((W i).card : ℝ) ≤ 2*reservoirProbability η*Fintype.card V) ∧
        (∀ i, (W i).card < Fintype.card V) ∧
        ((∑ i, (W i).card : ℕ) : ℝ) ≤ η * Fintype.card V ∧
        (∀ i, Sampling.CutExpansionOn (Splitting.colorGraph P ω i) (W i)
          (D^(37/50 : ℝ))) ∧
        (∀ i v, v ∉ W i → ∀ t, f i (v,t) ∈ W i ∧
          (Splitting.colorGraph P ω i).Adj v (f i (v,t))) ∧
        (∀ i v, v ∉ W i → f i (v,0) ≠ f i (v,1)) ∧
        (∀ i w, (((W i)ᶜ ×ˢ Finset.univ).filter (fun x => f i x = w)).card ≤
          ⌈8*(Fintype.card V : ℝ) /
            (reservoirProbability η * (reservoirExpansionCoefficient c * D^(9/10 : ℝ)))⌉₊) := by
  let c₀ := min c 1
  let a := reservoirExpansionCoefficient c
  let q := reservoirProbability η
  have hc₀ : 0 < c₀ := lt_min hc (by norm_num)
  have ha : 0 < a := by dsimp [a, reservoirExpansionCoefficient]; positivity
  have hq : 0 < q := by dsimp [q, reservoirProbability]; positivity
  have hq12 : q ≤ 1/12 := min_le_right _ _
  have hqη : q ≤ η/12 := min_le_left _ _
  have hq3 : 3*q ≤ 1 := by linarith
  have hq6 : 6*q ≤ η := by linarith
  have hq2 : 2*q < 1 := by linarith
  let p : unitInterval := ⟨q, hq.le, by linarith⟩
  obtain ⟨D₁,hD₁⟩ := Splitting.edge_splitting_uniform.{u} c₀ hc₀ 3 (by norm_num)
  obtain ⟨D₂,hD₂⟩ := Sampling.three_reservoirs_uniform.{u} a ha p hq hq3
  have ht : ∀ᶠ D : ℝ in atTop, 4/(q*a) ≤ D^(9/10 : ℝ) :=
    (tendsto_rpow_atTop (by norm_num : (0 : ℝ) < 9/10)).eventually
      (eventually_ge_atTop _)
  obtain ⟨D₃,hD₃⟩ := ht.exists_forall_of_atTop
  refine ⟨max 1 (max D₁ (max D₂ D₃)), ?_⟩
  intro D hD V _ _ G P _ hPG hlo hhi hexp
  have hD1 : 1 ≤ D := (le_max_left _ _).trans hD
  have hDsplit : D₁ ≤ D := (le_trans (le_max_left _ _) (le_max_right _ _)).trans hD
  have hDsample : D₂ ≤ D :=
    (le_trans (le_max_left _ _) (le_trans (le_max_right _ _) (le_max_right _ _))).trans hD
  have hDpower : D₃ ≤ D :=
    (le_trans (le_max_right _ _) (le_trans (le_max_right _ _) (le_max_right _ _))).trans hD
  have hpow : 0 < D^(9/10 : ℝ) := Real.rpow_pos_of_pos (by linarith) _
  have hd : 4 ≤ q*(a*D^(9/10 : ℝ)) := by
    have hh := (div_le_iff₀ (mul_pos hq ha)).mp (hD₃ D hDpower)
    nlinarith
  have hr : 0 < (Fintype.card V : ℝ) := hpow.trans_le hlo
  have : Nonempty V := Fintype.card_pos_iff.mp (by exact_mod_cast hr)
  have hexp₀ : Splitting.CutExpansion P (c₀*D^(9/10 : ℝ)) := by
    intro U
    exact (mul_le_mul_of_nonneg_right
      (mul_le_mul_of_nonneg_right (min_le_left c 1) hpow.le)
      (le_min (Nat.cast_nonneg _) (Nat.cast_nonneg _))).trans (hexp U)
  obtain ⟨ω,hcut,hcover,hdisj⟩ := hD₁ D hDsplit V P
    (c₀*D^(9/10 : ℝ)) le_rfl hlo hhi hexp₀
  have hcut' : ∀ i, Sampling.CutExpansion (Splitting.colorGraph P ω i)
      (a*D^(9/10 : ℝ)) := by
    intro i
    have heq : a*D^(9/10 : ℝ) = c₀*D^(9/10 : ℝ)/(2*(3 : ℝ)) := by
      dsimp [a, reservoirExpansionCoefficient, c₀]
      ring
    rw [heq]
    exact hcut i
  obtain ⟨W,hWdisj,hWsize,hWneigh,hWcut⟩ := hD₂ D hDsample V
    (fun i => Splitting.colorGraph P ω i) hlo hhi hcut'
  have hWtwo : ∀ i, 2 ≤ (W i).card := by
    intro i
    let v : V := Classical.choice inferInstance
    have hn := hWneigh i v
    have hsub : Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i) ⊆ W i :=
      Finset.filter_subset _ _
    have hh := Finset.card_le_card hsub
    have hn' : (2 : ℝ) ≤ (Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i)).card := by
      change q*(a*D^(9/10 : ℝ))/2 ≤ _ at hn
      linarith
    have hn'' : 2 ≤ (Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i)).card := by
      exact_mod_cast hn'
    omega
  have hWlt : ∀ i, (W i).card < Fintype.card V := by
    intro i
    have hh := (hWsize i).trans_lt (mul_lt_of_lt_one_left hr hq2)
    exact_mod_cast hh
  have hWsum : ((∑ i, (W i).card : ℕ) : ℝ) ≤ η * Fintype.card V := by
    have hh := Finset.sum_le_sum (fun i (_ : i ∈ (Finset.univ : Finset (Fin 3))) => hWsize i)
    simp only [Finset.sum_const, Finset.card_univ, Fintype.card_fin, nsmul_eq_mul] at hh
    rw [Nat.cast_sum]
    exact hh.trans (by dsimp [p]; nlinarith [mul_le_mul_of_nonneg_right hq6 hr.le])
  have hrep : ∀ i, ∃ f : V × Fin 2 → V,
      (∀ v ∈ (W i)ᶜ, ∀ t, f (v,t) ∈ Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i)) ∧
      (∀ v ∈ (W i)ᶜ, f (v,0) ≠ f (v,1)) ∧
      (∀ w, (((W i)ᶜ ×ˢ Finset.univ).filter (fun x => f x = w)).card ≤
        ⌈8*(Fintype.card V : ℝ)/(q*(a*D^(9/10 : ℝ)))⌉₊) := by
    intro i
    exact Representatives.reservoir_representatives (W i)ᶜ
      (fun v => Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i))
      (Fintype.card V) (q*(a*D^(9/10 : ℝ)))
      (by exact_mod_cast (Finset.card_le_univ ((W i)ᶜ))) hd
      (fun v _ => hWneigh i v)
  choose f hf hsep hcap using hrep
  refine ⟨ω,W,f,?_,hdisj,?_,hWdisj,hWtwo,hWsize,hWlt,hWsum,hWcut,?_,?_,hcap⟩
  · intro i x y hxy
    exact hPG hxy.1
  · exact hcut'
  · intro i v hv t
    exact Finset.mem_filter.mp (hf i v (Finset.mem_compl.mpr hv) t)
  · intro i v hv
    exact hsep i v (Finset.mem_compl.mpr hv)
end
end ErdosGallai
end
end
end
end OAI

	import OAI.Combinatorics.CycleDecomposition.EdgeSplitting

	namespace OAI

	universe cycleUniverse1 cycleUniverse2

	section
	open Filter Asymptotics Real
	open scoped Topology
	noncomputable section
	open MeasureTheory ProbabilityTheory Finset
	noncomputable section
	namespace ErdosGallai.Representatives
	noncomputable section
	open Finset
	variable {I : Type cycleUniverse1} {V : Type cycleUniverse2} [DecidableEq I] [DecidableEq V] [Nonempty V]

	lemma choose_with_mates (A : Finset I) (mate : I → I)
	(hmate : Function.Involutive mate) (hfix : ∀ i, mate i ≠ i)
	(N : I → Finset V) (k : ℕ)
	(hroom : ∀ i ∈ A, A.card ≤ ((N i).card - 1) * k) :
	∃ f : I → V, (∀ i ∈ A, f i ∈ N i) ∧
	(∀ i ∈ A, mate i ∈ A → f i ≠ f (mate i)) ∧
	(∀ v, (A.filter (fun i => f i = v)).card ≤ k) := by
	induction A using Finset.induction_on with
	\| empty => exact ⟨fun _ => Classical.choice inferInstance, by simp⟩
	\| @insert i A hi ih =>
	obtain ⟨f,hf,hsep,hcap⟩ := ih (by
	intro j hj
	exact (Finset.card_le_card (Finset.subset_insert i A)).trans (hroom j (by simp [hj])))
	have hlt : A.card < ((N i).erase (f (mate i))).card * k := by
	have h := hroom i (by simp)
	have hcard := Finset.pred_card_le_card_erase (s := N i) (a := f (mate i))
	have hmul := Nat.mul_le_mul_right k hcard
	rw [Finset.card_insert_of_notMem hi] at h
	omega
	obtain ⟨v,hv,hload⟩ := Finset.exists_card_fiber_lt_of_card_lt_mul (f := f) hlt
	have hvN := (Finset.mem_erase.mp hv).2
	have hvne := (Finset.mem_erase.mp hv).1
	let g := Function.update f i v
	have gi : g i = v := by simp [g]
	have gold : ∀ j, j ≠ i → g j = f j := by
	intro j hj
	simp [g, hj]
	refine ⟨g, ?_, ?_, ?_⟩
	· intro j hj
	rcases Finset.mem_insert.mp hj with rfl \| hj
	· simpa only [gi] using hvN
	· rw [gold j (ne_of_mem_of_not_mem hj hi)]
	exact hf j hj
	· intro j hj hmj
	by_cases hji : j = i
	· subst j
	rw [gi, gold (mate i) (hfix i)]
	exact hvne
	· rw [gold j hji]
	by_cases hmi : mate j = i
	· rw [hmi, gi]
	have hjm : j = mate i := by rw [← hmi, hmate]
	simpa [hjm] using hvne.symm
	· rw [gold (mate j) hmi]
	exact hsep j (Finset.mem_of_mem_insert_of_ne hj hji)
	(Finset.mem_of_mem_insert_of_ne hmj hmi)
	· intro w
	have hold : A.filter (fun j => g j = w) = A.filter (fun j => f j = w) := by
	apply Finset.filter_congr
	intro j hj
	rw [gold j (ne_of_mem_of_not_mem hj hi)]
	rw [Finset.filter_insert, hold]
	by_cases hvw : v = w
	· simp only [gi, hvw, ite_eq_left]
	rw [Finset.card_insert_of_notMem (fun h => hi (Finset.mem_filter.mp h).1)]
	simpa only [hvw] using Nat.succ_le_of_lt hload
	· simp only [gi, hvw, ite_false]
	exact hcap w

	theorem two_distinct_representatives (S : Finset I) (N : I → Finset V) (k : ℕ)
	(hroom : ∀ i ∈ S, 2S.card ≤ ((N i).card-1)k) :
	∃ f : I × Fin 2 → V,
	(∀ i ∈ S, ∀ t, f (i,t) ∈ N i) ∧
	(∀ i ∈ S, f (i,0) ≠ f (i,1)) ∧
	(∀ v, ((S ×ˢ Finset.univ).filter (fun i => f i = v)).card ≤ k) := by
	let mate : I × Fin 2 → I × Fin 2 := fun i => (i.1, 1-i.2)
	have hm : Function.Involutive mate := by
	rintro ⟨i,t⟩
	fin_cases t <;> rfl
	have hfix : ∀ i, mate i ≠ i := by
	rintro ⟨i,t⟩
	fin_cases t <;> (intro h; have hh := congrArg Prod.snd h; norm_num [mate] at hh)
	obtain ⟨f,hf,hsep,hcap⟩ := choose_with_mates (S ×ˢ Finset.univ) mate hm hfix
	(fun i => N i.1) k (by
	intro i hi
	simpa [Finset.card_product, mul_comm] using hroom i.1 (Finset.mem_product.mp hi).1)
	refine ⟨f, ?_, ?_, hcap⟩
	· intro i hi t
	exact hf (i,t) (by simp [hi])
	· intro i hi
	exact hsep (i,0) (by simp [hi]) (by simp [mate, hi])

	theorem reservoir_representatives (S : Finset I) (N : I → Finset V)
	(r d : ℝ) (hr : (S.card : ℝ) ≤ r) (hd : 4 ≤ d)
	(hN : ∀ i ∈ S, d/2 ≤ ((N i).card : ℝ)) :
	∃ f : I × Fin 2 → V,
	(∀ i ∈ S, ∀ t, f (i,t) ∈ N i) ∧
	(∀ i ∈ S, f (i,0) ≠ f (i,1)) ∧
	(∀ v, ((S ×ˢ Finset.univ).filter (fun i => f i = v)).card ≤ ⌈8*r/d⌉₊) := by
	apply two_distinct_representatives
	intro i hi
	have hd0 : 0 < d := by linarith
	have hr0 : 0 ≤ r := le_trans (Nat.cast_nonneg _) hr
	have hNi := hN i hi
	have hNi1 : 1 ≤ (N i).card := by exact_mod_cast (by linarith : (1 : ℝ) ≤ (N i).card)
	have hn : d/4 ≤ (((N i).card - 1 : ℕ) : ℝ) := by
	rw [Nat.cast_sub hNi1]
	push_cast
	linarith
	have hk : 8r/d ≤ (⌈8r/d⌉₊ : ℝ) := Nat.le_ceil _
	have hk0 : 0 ≤ (⌈8*r/d⌉₊ : ℝ) := Nat.cast_nonneg _
	have hprod : 2(S.card : ℝ) ≤ (((N i).card - 1 : ℕ) : ℝ)(⌈8*r/d⌉₊ : ℝ) := by
	calc
	_ ≤ 2*r := by linarith
	_ = (d/4)(8r/d) := by field_simp; ring
	_ ≤ (d/4)(⌈8r/d⌉₊ : ℝ) := mul_le_mul_of_nonneg_left hk (by positivity)
	_ ≤ _ := mul_le_mul_of_nonneg_right hn hk0
	exact_mod_cast hprod

	end
	end ErdosGallai.Representatives

	namespace ErdosGallai
	noncomputable section
	open Finset Real Filter
	attribute [local instance] Classical.propDecidable

	noncomputable def reservoirProbability (η : ℝ) : ℝ := min (η/12) (1/12)
	noncomputable def reservoirExpansionCoefficient (c : ℝ) : ℝ := min c 1 / 6

	universe u

	theorem reservoir_preparation (c η : ℝ) (hc : 0 < c) (hη : 0 < η) :
	∃ D₀ : ℝ, ∀ D : ℝ, D₀ ≤ D → ∀ (V : Type u) [Fintype V] [DecidableEq V]
	(G P : SimpleGraph V) [DecidableRel P.Adj], P ≤ G →
	D^(9/10 : ℝ) ≤ (Fintype.card V : ℝ) →
	(Fintype.card V : ℝ) ≤ D^(51/50 : ℝ) →
	Splitting.CutExpansion P (c*D^(9/10 : ℝ)) →
	∃ (ω : Sym2 V → Fin 3) (W : Fin 3 → Finset V) (f : Fin 3 → V × Fin 2 → V),
	(∀ i, Splitting.colorGraph P ω i ≤ G) ∧
	(∀ i j, i ≠ j → Disjoint (Splitting.colorGraph P ω i).edgeSet
	(Splitting.colorGraph P ω j).edgeSet) ∧
	(∀ i, Splitting.CutExpansion (Splitting.colorGraph P ω i)
	(reservoirExpansionCoefficient c * D^(9/10 : ℝ))) ∧
	(∀ i j, i ≠ j → Disjoint (W i) (W j)) ∧
	(∀ i, 2 ≤ (W i).card) ∧
	(∀ i, ((W i).card : ℝ) ≤ 2reservoirProbability ηFintype.card V) ∧
	(∀ i, (W i).card < Fintype.card V) ∧
	((∑ i, (W i).card : ℕ) : ℝ) ≤ η * Fintype.card V ∧
	(∀ i, Sampling.CutExpansionOn (Splitting.colorGraph P ω i) (W i)
	(D^(37/50 : ℝ))) ∧
	(∀ i v, v ∉ W i → ∀ t, f i (v,t) ∈ W i ∧
	(Splitting.colorGraph P ω i).Adj v (f i (v,t))) ∧
	(∀ i v, v ∉ W i → f i (v,0) ≠ f i (v,1)) ∧
	(∀ i w, (((W i)ᶜ ×ˢ Finset.univ).filter (fun x => f i x = w)).card ≤
	⌈8*(Fintype.card V : ℝ) /
	(reservoirProbability η * (reservoirExpansionCoefficient c * D^(9/10 : ℝ)))⌉₊) := by
	let c₀ := min c 1
	let a := reservoirExpansionCoefficient c
	let q := reservoirProbability η
	have hc₀ : 0 < c₀ := lt_min hc (by norm_num)
	have ha : 0 < a := by dsimp [a, reservoirExpansionCoefficient]; positivity
	have hq : 0 < q := by dsimp [q, reservoirProbability]; positivity
	have hq12 : q ≤ 1/12 := min_le_right _ _
	have hqη : q ≤ η/12 := min_le_left _ _
	have hq3 : 3*q ≤ 1 := by linarith
	have hq6 : 6*q ≤ η := by linarith
	have hq2 : 2*q < 1 := by linarith
	let p : unitInterval := ⟨q, hq.le, by linarith⟩
	obtain ⟨D₁,hD₁⟩ := Splitting.edge_splitting_uniform.{u} c₀ hc₀ 3 (by norm_num)
	obtain ⟨D₂,hD₂⟩ := Sampling.three_reservoirs_uniform.{u} a ha p hq hq3
	have ht : ∀ᶠ D : ℝ in atTop, 4/(q*a) ≤ D^(9/10 : ℝ) :=
	(tendsto_rpow_atTop (by norm_num : (0 : ℝ) < 9/10)).eventually
	(eventually_ge_atTop _)
	obtain ⟨D₃,hD₃⟩ := ht.exists_forall_of_atTop
	refine ⟨max 1 (max D₁ (max D₂ D₃)), ?_⟩
	intro D hD V _ _ G P _ hPG hlo hhi hexp
	have hD1 : 1 ≤ D := (le_max_left _ _).trans hD
	have hDsplit : D₁ ≤ D := (le_trans (le_max_left _ _) (le_max_right _ _)).trans hD
	have hDsample : D₂ ≤ D :=
	(le_trans (le_max_left _ _) (le_trans (le_max_right _ _) (le_max_right _ _))).trans hD
	have hDpower : D₃ ≤ D :=
	(le_trans (le_max_right _ _) (le_trans (le_max_right _ _) (le_max_right _ _))).trans hD
	have hpow : 0 < D^(9/10 : ℝ) := Real.rpow_pos_of_pos (by linarith) _
	have hd : 4 ≤ q(aD^(9/10 : ℝ)) := by
	have hh := (div_le_iff₀ (mul_pos hq ha)).mp (hD₃ D hDpower)
	nlinarith
	have hr : 0 < (Fintype.card V : ℝ) := hpow.trans_le hlo
	have : Nonempty V := Fintype.card_pos_iff.mp (by exact_mod_cast hr)
	have hexp₀ : Splitting.CutExpansion P (c₀*D^(9/10 : ℝ)) := by
	intro U
	exact (mul_le_mul_of_nonneg_right
	(mul_le_mul_of_nonneg_right (min_le_left c 1) hpow.le)
	(le_min (Nat.cast_nonneg _) (Nat.cast_nonneg _))).trans (hexp U)
	obtain ⟨ω,hcut,hcover,hdisj⟩ := hD₁ D hDsplit V P
	(c₀*D^(9/10 : ℝ)) le_rfl hlo hhi hexp₀
	have hcut' : ∀ i, Sampling.CutExpansion (Splitting.colorGraph P ω i)
	(a*D^(9/10 : ℝ)) := by
	intro i
	have heq : aD^(9/10 : ℝ) = c₀D^(9/10 : ℝ)/(2*(3 : ℝ)) := by
	dsimp [a, reservoirExpansionCoefficient, c₀]
	ring
	rw [heq]
	exact hcut i
	obtain ⟨W,hWdisj,hWsize,hWneigh,hWcut⟩ := hD₂ D hDsample V
	(fun i => Splitting.colorGraph P ω i) hlo hhi hcut'
	have hWtwo : ∀ i, 2 ≤ (W i).card := by
	intro i
	let v : V := Classical.choice inferInstance
	have hn := hWneigh i v
	have hsub : Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i) ⊆ W i :=
	Finset.filter_subset _ _
	have hh := Finset.card_le_card hsub
	have hn' : (2 : ℝ) ≤ (Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i)).card := by
	change q(aD^(9/10 : ℝ))/2 ≤ _ at hn
	linarith
	have hn'' : 2 ≤ (Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i)).card := by
	exact_mod_cast hn'
	omega
	have hWlt : ∀ i, (W i).card < Fintype.card V := by
	intro i
	have hh := (hWsize i).trans_lt (mul_lt_of_lt_one_left hr hq2)
	exact_mod_cast hh
	have hWsum : ((∑ i, (W i).card : ℕ) : ℝ) ≤ η * Fintype.card V := by
	have hh := Finset.sum_le_sum (fun i (_ : i ∈ (Finset.univ : Finset (Fin 3))) => hWsize i)
	simp only [Finset.sum_const, Finset.card_univ, Fintype.card_fin, nsmul_eq_mul] at hh
	rw [Nat.cast_sum]
	exact hh.trans (by dsimp [p]; nlinarith [mul_le_mul_of_nonneg_right hq6 hr.le])
	have hrep : ∀ i, ∃ f : V × Fin 2 → V,
	(∀ v ∈ (W i)ᶜ, ∀ t, f (v,t) ∈ Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i)) ∧
	(∀ v ∈ (W i)ᶜ, f (v,0) ≠ f (v,1)) ∧
	(∀ w, (((W i)ᶜ ×ˢ Finset.univ).filter (fun x => f x = w)).card ≤
	⌈8(Fintype.card V : ℝ)/(q(a*D^(9/10 : ℝ)))⌉₊) := by
	intro i
	exact Representatives.reservoir_representatives (W i)ᶜ
	(fun v => Sampling.neighborsIn (Splitting.colorGraph P ω i) v (W i))
	(Fintype.card V) (q(aD^(9/10 : ℝ)))
	(by exact_mod_cast (Finset.card_le_univ ((W i)ᶜ))) hd
	(fun v _ => hWneigh i v)
	choose f hf hsep hcap using hrep
	refine ⟨ω,W,f,?_,hdisj,?_,hWdisj,hWtwo,hWsize,hWlt,hWsum,hWcut,?_,?_,hcap⟩
	· intro i x y hxy
	exact hPG hxy.1
	· exact hcut'
	· intro i v hv t
	exact Finset.mem_filter.mp (hf i v (Finset.mem_compl.mpr hv) t)
	· intro i v hv
	exact hsep i v (Finset.mem_compl.mpr hv)

	end
	end ErdosGallai

	end
	end
	end

	end OAI