The degree dependence of Houdré–Tetali isoperimetry: simple circulant counterexamples and a uniform law
Authors/Creators
Description
Unrefereed research candidate. In one sentence: simple graphs break the square-root Cheeger inequality conjectured by Houdré and Tetali by a factor that grows like the logarithm of the degree, and no graph or reversible chain can do worse than that, up to fixed constants.
Context. A random walk on a network mixes quickly unless some region is hard to leave. Cheeger's inequality turns this into a guarantee: the spectral gap λ2 certifies how easy the hardest region is to leave, but it loses a square. Houdré and Tetali (Combinatorica, 2004) introduced boundary functionals φp that add up the p-th powers of the escape probabilities of the states of a set (p = 1: conductance; p = 0: vertex expansion). They asked whether φ1/22 ≤ C λ2 holds with an absolute constant, improving their own bound 2κ−1λ2, where κ is the smallest transition probability. Morris and Peres (Probab. Theory Relat. Fields, 2005) proved a version with a logarithmic loss. Lau and Tjowasi (APPROX/RANDOM 2024) showed with a weighted cycle that the logarithm is needed for weighted chains, proved the inequality for p > 1/2, and asked for a counterexample in which the chain is the simple random walk on a graph.
Results. (1) Theorem 1.1: for every even n ≥ 4, an explicit connected simple circulant graph Gn on n4 + n vertices with degree below 4n3 has φ1/22/λ2 ≥ 9·10−6 log n, and the ratio is of order log n. The graphs replace each state of the Lau–Tjowasi weighted cycle by a clique and each weight by a regular bipartite graph; a deterministic comparison theorem (Theorem 4.1) shows that cuts splitting the cliques cannot do much better than cuts along them. (2) Theorem 1.2: every irreducible reversible chain satisfies φp2 ≤ 4(1 + (1−p)2Ψp(1/κ))λ2, where Ψp(x) = ∫1x t−2p dt; at p = 1/2 this is (4 + log(1/κ))λ2, which answers the quantitative question of Houdré and Tetali about the factor 1/κ, and a level set of a second eigenfunction attains the bound. (3) Theorem 1.3: for every 0 ≤ p ≤ 1 and every maximum degree Δ ≥ 2, the largest value of φp2/λ2 over simple graphs (and over circulants) lies between 10−8Ψp(Δ) and 12Ψp(Δ), uniformly in both parameters.
Limits. The constants are explicit but not sharp: the lower constant 10−8 is far below the upper constant 12, and the exact constants and extremal graphs are open. The effect is numerically mild: for Gn an upper witness for the ratio is 0.50 at 260 vertices, 1.00 at about 1.7·107 vertices and 1.55 at about 1.1·1012 vertices. The lower bounds rely on the weighted estimate of Lau and Tjowasi (Section 4.1 of their paper), restated with an explicit constant and checked numerically for small cycles; no self-contained proof of that estimate is given. The critical logarithmic upper bound for regular graphs and the spectral sweep method were known; the new parts are the simple circulant counterexamples, the comparison theorem, the lower bounds uniform in the exponent and the chain-level upper bound with optimal dependence on κ.
Contents. The paper (LaTeX source and PDF) and its figure; three check scripts with their recorded outputs (construction, spectra and quotient fingerprints of Gn for small n; all cuts of small clique-fibre graphs against the comparison theorem; the scalar constants, the parameter choices, the Lau–Tjowasi inequality for small cycles and a brute-force test of Theorem 1.2 on random reversible chains); the figure script; the bibliography; a machine-readable index; a research-gate record; a review summary; a replay receipt and SHA-256 sums. The checks are finite and use stated floating-point tolerances; every check raises SystemExit and stays active under python -O.
Assurance. Written proofs, with finite computational checks. Internal AI-assisted review only (five reviewer roles, a clean replay, one revision round with extensions, a re-review and a proof audit of the new proofs by two further AI verification passes; an adversarial review by a second AI model was cut short by a usage limit and found no defect before it stopped): not peer review, independent reproduction or formal verification. The question is due to Lau and Tjowasi, the conjecture to Houdré and Tetali, and the logarithmic upper bound to Morris and Peres. Creator: Anonymous. Text and data CC0-1.0; code MIT.