New mathematical bounds for spin systems and random forests

Researchers developed a method to find spectral gap bounds for block dynamics and improved bounds for random forests.

Researchers found new ways to calculate spectral gap bounds for block dynamics in spin systems. They also found new bounds for swap walks and simplicial complexes.

The work uses an extended Bochner method. This method uses an abstract inequality for orthogonal projections on Hilbert spaces. It also uses the analysis of an auxiliary real stable polynomial.

A spin system is a probability distribution μ\mu on nn sites. Block Glauber dynamics are used to find global spectral-gap bounds on marginal distributions. A spectral gap is a mathematical value related to these dynamics.

For the O(N)O(N) model where N≥2N \geq 2, the spectral gap is at least (1−∥Jabs∥op/N)/n(1-\|J_{\mathrm{abs}}\|_{\mathrm{op}}/N)/n. This applies to arbitrary external fields and couplings of either sign. Here, JJ is the interaction matrix and Jabs=(∣Jij∣)J_{\mathrm{abs}}=(|J_{ij}|).

For random forests FF of any finite graph with positive edge weights, a new bound was proven. The probability Pr⁡(e,f∈F)≤43Pr⁡(e∈F)Pr⁡(f∈F)\Pr(e,f\in F)\le \frac{4}{3} \Pr(e\in F)\Pr(f\in F) for distinct edges e,fe,f. This improves the factor 2 from general log-concavity bounds.

Key facts

  • The Gaussian system with covariance (I−A)−1(I-A)^{-1} attains equality for every choice of disjoint blocks.
  • The bound for certain simplicial complexes is 2q/(q+1)2\sqrt q/(q+1).
  • The new forest bound improves upon general log-concavity bounds for matroid independent sets.

All sections

Edition