Uniform Spectral Gap

Documentation

Lean 4 Proof

theorem uniform_spectralGap (hJ : 2 ≤ J) (ρ : ℝ) :
    spectralGap (uniformNetwork J ρ) = ↑J * (1 - ρ) := by
  simp only [spectralGap]
  -- Every v ⊥ 1 with ‖v‖² > 0 gives Rayleigh quotient = J*(1-ρ)
  -- by uniform_laplacianQF_on_perp. So the image is {J*(1-ρ)}.
  -- Witness for nonemptiness: e₀ - e₁.
  set i₀ : Fin J := ⟨0, by omega⟩
  set i₁ : Fin J := ⟨1, by omegahave h01 : i₀ ≠ i₁ := by simp [Fin.ext_iff, i₀, i₁]
  -- Helper: for v ⊥ 1, the Rayleigh quotient on the uniform network is constant
  have hquot : ∀ v : Fin J → ℝ, orthToOne J v → vecNormSq J v > 0 →
      laplacianQF (uniformNetwork J ρ) v / vecNormSq J v = ↑J * (1 - ρ) := by
    intro v hv hvn
    rw [uniform_laplacianQF_on_perp ρ v hv]
    field_simp
  -- Construct witness w = e₀ - e₁
  set w : Fin J → ℝ := fun j => if j = i₀ then 1 else if j = i₁ then -1 else 0
  have hw_orth : orthToOne J w := by
    simp only [orthToOne, vecSum, w]
    have hrw : ∀ j : Fin J, (if j = i₀ then (1:ℝ) else if j = i₁ then -1 else 0) =
        (if j = i₀ then (1:ℝ) else 0) + (if j = i₁ then (-1:ℝ) else 0) := by
      intro j; by_cases h1 : j = i₀
      · subst h1; simp [h01]
      · by_cases h2 : j = i₁
        · subst h2; simp [h1]
        · simp [h1, h2]
    simp_rw [hrw, Finset.sum_add_distrib, Finset.sum_ite_eq', Finset.mem_univ, ite_true]
    norm_num
  have hw_norm : vecNormSq J w > 0 := by
    simp only [vecNormSq, w]
    apply Finset.sum_pos'
    · intro j _; exact sq_nonneg _
    · exact ⟨i₀, Finset.mem_univ _, by simp-- Show the image is the singleton {J*(1-ρ)}
  have himg : (fun v => laplacianQF (uniformNetwork J ρ) v / vecNormSq J v) ''
      {v : Fin J → ℝ | orthToOne J v ∧ vecNormSq J v > 0} = {↑J * (1 - ρ)} := by
    ext r; simp only [Set.mem_image, Set.mem_setOf_eq, Set.mem_singleton_iff]
    constructor
    · rintro ⟨v, ⟨hv, hvn⟩, hr⟩; rw [← hr]; exact hquot v hv hvn
    · intro hr; exact ⟨w, ⟨hw_orth, hw_norm⟩, hr ▸ hquot w hw_orth hw_norm⟩
  rw [himg, csInf_singleton]

Dependency Graph

Module Section

CES on Networks: Heterogeneous Pairwise Complementarity