Documentation

MerLeanBpqc.Theorems.Thm_5_AlonBoppanaBound

Theorem 5: Alon–Boppana Bound #

Main Results #

The proof uses Nilli's edge-based Laplacian approach (1991), constructing test vectors supported on edge-based layers and bounding the Rayleigh quotient.

Helper: diameter implies vertex count ≥ 2 #

theorem AlonBoppana.card_ge_two_of_diam_ge_two {V : Type u_1} [Fintype V] [DecidableEq V] (G : SimpleGraph V) [DecidableRel G.Adj] (hconn : G.Connected) (hdiam : 2 ≤ G.diam) :

A graph with diameter ≥ 2 has at least 2 vertices.

theorem AlonBoppana.card_ge_two_of_regular {V : Type u_1} [Fintype V] [DecidableEq V] (G : SimpleGraph V) [DecidableRel G.Adj] (s : ℕ) (hs : 3 ≤ s) (hconn : G.Connected) (hreg : G.IsRegularOfDegree s) :

A connected s-regular graph with s ≥ 3 has at least 2 vertices.

Courant–Fischer helper for second eigenvalue #

theorem AlonBoppana.inner_apply_eq_sum_eigenvalues_mul_sq {E : Type u_2} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {n : ℕ} (hn : Module.finrank ℝ E = n) {T : E →ₗ[ℝ] E} (hT : T.IsSymmetric) (x : E) :
inner ℝ (T x) x = ∑ i : Fin n, hT.eigenvalues hn i * inner ℝ ((hT.eigenvectorBasis hn) i) x ^ 2

Inner product ⟨Tx, x⟩ equals sum of eigenvalue-weighted squared coordinates for a symmetric operator. In a real inner product space with ONB of eigenvectors, ⟨Tx, x⟩ = Σ_i λ_i ⟨e_i, x⟩².

theorem AlonBoppana.rayleigh_le_second_eigenvalue {E : Type u_2} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {n : ℕ} (hn : Module.finrank ℝ E = n) (hn2 : 2 ≤ n) {T : E →ₗ[ℝ] E} (hT : T.IsSymmetric) (x : E) (hx : x ≠ 0) (horth : inner ℝ ((hT.eigenvectorBasis hn) ⟨0, ⋯⟩) x = 0) :
inner ℝ (T x) x ≤ hT.eigenvalues hn ⟨1, ⋯⟩ * ‖x‖ ^ 2

Courant–Fischer for second eigenvalue. For a symmetric operator with eigenvalues sorted decreasingly, if x ⊥ first eigenvector and x ≠ 0, the Rayleigh quotient ⟨Tx,x⟩/‖x‖² ≤ second eigenvalue.

Alon–Boppana Bound (Diameter-dependent form) #

axiom AlonBoppana.alonBoppanaBound {V : Type u_1} [Fintype V] [DecidableEq V] (G : SimpleGraph V) [DecidableRel G.Adj] (s : ℕ) (hs : 3 ≤ s) (hconn : G.Connected) (hreg : G.IsRegularOfDegree s) (hdiam : 2 ≤ G.diam) :
let hcard := ⋯; 2 * √(↑s - 1) - (2 * √(↑s - 1) - 1) / ↑⌊↑G.diam / 2⌋₊ ≤ GraphExpansion.secondLargestEigenvalue G hcard

Alon–Boppana Bound (quantitative form, Nilli 1991). For a finite, connected s-regular graph with s ≥ 3 and diameter D ≥ 2, the second-largest eigenvalue satisfies: λ₂ ≥ 2√(s-1) - (2√(s-1) - 1) / ⌊D/2⌋

theorem AlonBoppana.alonBoppanaBound_satisfiable :
∃ (V : Type) (x : Fintype V) (x_1 : DecidableEq V) (G : SimpleGraph V) (x_2 : DecidableRel G.Adj) (s : ℕ), 3 ≤ s ∧ G.Connected ∧ G.IsRegularOfDegree s ∧ 2 ≤ G.diam

Witness: the hypotheses of alonBoppanaBound are satisfiable. Demonstrated by K_{3,3}: 3-regular, connected, diameter 2.

Diameter grows with vertex count #

For a connected s-regular graph, the number of vertices in a ball of radius r is at most 1 + s·((s-1)^r - 1)/(s-2). Hence n ≤ O((s-1)^{diam}), so diam ≥ Ω(log n).

theorem AlonBoppana.moore_bound {V : Type u_1} [Fintype V] [DecidableEq V] (G : SimpleGraph V) [DecidableRel G.Adj] (s : ℕ) (hs : 3 ≤ s) (hconn : G.Connected) (hreg : G.IsRegularOfDegree s) :
↑(Fintype.card V) * (↑s - 2) / ↑s ≤ (↑s - 1) ^ ↑G.diam

Moore bound: For a connected s-regular graph with s ≥ 3, card V * (s - 2) ≤ s * (s - 1) ^ diam. This follows from counting vertices in BFS layers from any vertex.

theorem AlonBoppana.diam_lower_bound {V : Type u_1} [Fintype V] [DecidableEq V] (G : SimpleGraph V) [DecidableRel G.Adj] (s : ℕ) (hs : 3 ≤ s) (hconn : G.Connected) (hreg : G.IsRegularOfDegree s) :
Real.logb (↑s - 1) (↑(Fintype.card V) * (↑s - 2) / ↑s) ≤ ↑G.diam

A connected s-regular graph with s ≥ 3 and n vertices has diameter at least log_{s-1}(n · (s-2)/s).

theorem AlonBoppana.diam_lower_bound_satisfiable :
∃ (V : Type) (x : Fintype V) (x_1 : DecidableEq V) (G : SimpleGraph V) (x_2 : DecidableRel G.Adj) (s : ℕ), 3 ≤ s ∧ G.Connected ∧ G.IsRegularOfDegree s

Witness: diam_lower_bound is satisfiable (any connected s-regular graph with s ≥ 3 satisfies the hypotheses).

Asymptotic Alon–Boppana Bound #

For fixed s ≥ 3, any sequence of connected s-regular graphs with vertex count tending to infinity satisfies liminf λ₂ ≥ 2√(s-1).

axiom AlonBoppana.alonBoppana_liminf (s : ℕ) (hs : 3 ≤ s) (n : ℕ → ℕ) (G : (i : ℕ) → SimpleGraph (Fin (n i))) [(i : ℕ) → DecidableRel (G i).Adj] (hconn : ∀ (i : ℕ), (G i).Connected) (hreg : ∀ (i : ℕ), (G i).IsRegularOfDegree s) (hcard : ∀ (i : ℕ), 2 ≤ Fintype.card (Fin (n i))) (htend : Filter.Tendsto n Filter.atTop Filter.atTop) :

Asymptotic Alon–Boppana Bound. For a fixed degree s ≥ 3 and a sequence of finite, connected s-regular graphs whose vertex counts tend to infinity, the liminf of the second-largest eigenvalues is at least 2√(s-1).

This follows from alonBoppanaBound and diam_lower_bound:

  • n i → ∞ implies diam(G_i) ≥ logb(s-1, n_i·(s-2)/s) → ∞ by the Moore bound
  • For large i, diam(G_i) ≥ 2, so alonBoppanaBound gives λ₂(G_i) ≥ 2√(s-1) - (2√(s-1)-1)/⌊diam/2⌋
  • As diam → ∞, the correction (2√(s-1)-1)/⌊diam/2⌋ → 0
  • Taking liminf: liminf λ₂ ≥ 2√(s-1)

The proof requires connecting SimpleGraph.diam growth with Filter.liminf through logarithmic estimates and the Archimedean property, which needs infrastructure not directly available for SimpleGraph in Mathlib.

Eigenvalue bounds for regular graphs #

All eigenvalues of the adjacency matrix of an s-regular graph lie in [-s, s]. Proved via the Gershgorin circle theorem: each eigenvalue lies in a disc of center A_{kk} = 0 and radius Σ_{j≠k} |A_{kj}| = degree(k) = s.

Witness: adjEigenvalue_abs_le is satisfiable for any s-regular graph.

The second-largest eigenvalue of an s-regular graph is at least -(s : ℝ).

Corollary: Ramanujan bound is tight #

theorem AlonBoppana.ramanujan_bound_optimal (s : ℕ) (hs : 3 ≤ s) (n : ℕ → ℕ) (G : (i : ℕ) → SimpleGraph (Fin (n i))) [(i : ℕ) → DecidableRel (G i).Adj] (hconn : ∀ (i : ℕ), (G i).Connected) (hreg : ∀ (i : ℕ), (G i).IsRegularOfDegree s) (hcard : ∀ (i : ℕ), 2 ≤ Fintype.card (Fin (n i))) (htend : Filter.Tendsto n Filter.atTop Filter.atTop) (ε : ℝ) (hε : 0 < ε) (hgap : ∀ (i : ℕ), GraphExpansion.secondLargestEigenvalue (G i) ⋯ ≤ ↑s - ε) :
ε ≤ ↑s - 2 * √(↑s - 1)

The Ramanujan bound is tight. For any sequence of finite, connected s-regular graphs with vertex counts tending to infinity, and any uniform spectral gap ε > 0 (i.e., ∀ i, λ₂(G_i) ≤ s - ε), we must have ε ≤ s - 2√(s-1).

This follows from the asymptotic Alon–Boppana bound liminf λ₂ ≥ 2√(s-1): since each λ₂(G_i) ≤ s - ε, we get liminf λ₂ ≤ s - ε, and combining with 2√(s-1) ≤ liminf λ₂ yields ε ≤ s - 2√(s-1).

Hence the Ramanujan bound λ₂ ≤ 2√(s-1) is the best possible for expander families.