[Paper Review] On the spectral gap of the kac walk and other binary collision processes on $d$-dimensional lattice

Makiko Sasada|arXiv (Cornell University)|Aug 23, 2013

Stochastic processes and statistical mechanics8 references3 citations

TL;DR

This paper establishes a comparison technique to bound the spectral gap of binary collision processes on the d-dimensional lattice by reducing them to equivalent processes on the complete graph, enabling new, elementary lower bounds for models like the Kac walk, energy exchange models, and zero-range processes. The key result is a spectral gap lower bound of $\lambda^{*,\text{loc}}(N) \geq \frac{1}{384dN^2}$, which confirms the $N^{-2}$ scaling crucial for hydrodynamic limits.

ABSTRACT

We give a lower bound on the spectral gap for a class of binary collision processes. In 2008, Caputo showed that, for a class of binary collision processes given by simple averages on the complete graph, the analysis of the spectral gap of an $N$-component system is reduced to that of the same system for N=3 In this paper, we give a comparison technique to reduce the analysis of the spectral gap of binary collision processes given by simple averages on $d$-dimensional lattice to that on the complete graph. We also give a comparison technique to reduce the analysis of the spectral gap of binary collision processes which are not given by simple averages to that given by simple averages. Combining them with Caputo's result, we give a new and elementary method to obtain spectral gap estimates. The method applies to a number of binary collision processes on the complete graph and also on d-dimensional lattice, including a class of energy exchange models which was recently introduced in by Grigo et al., and zero-range processes.

Motivation & Objective

• To extend Caputo's spectral gap analysis from the complete graph to the d-dimensional lattice for binary collision processes.
• To develop a general comparison method that reduces spectral gap estimation on the lattice to the simpler complete graph case.
• To apply the method to a broad class of reversible processes, including energy exchange models and zero-range processes.
• To provide elementary, non-recursive lower bounds on the spectral gap that match the $N^{-2}$ scaling required for hydrodynamic limits.
• To establish conditions under which the spectral gap remains positive and bounded away from zero in the thermodynamic limit.

Proposed method

• Introduce a comparison technique to relate the spectral gap of the local (lattice-based) process to that of the global (complete graph) process.
• Use the invariance of the invariant measure under coordinate permutations to simplify the analysis of the generator.
• Apply the 'moving particle lemma' idea to control the mixing time and spectral gap in spatially extended systems.
• Reduce the analysis of non-simple-average processes to equivalent simple-average processes via a comparison argument.
• Leverage Caputo's result that the spectral gap for $N$-particle systems reduces to the $N=3$ case for simple-average models.
• Use unitary scaling and spectral decomposition to relate the gap at different energy levels and establish uniform bounds.

Experimental results

Research questions

• RQ1 Can the spectral gap of the Kac walk on a d-dimensional lattice be bounded below using a reduction to the complete graph?
• RQ2 What conditions ensure that the spectral gap of a binary collision process on the lattice scales as $N^{-2}$?
• RQ3 How can the spectral gap of non-simple-average processes be compared to that of simple-average processes?
• RQ4 Under what conditions is the spectral gap of a zero-range process on the lattice uniformly bounded away from zero?
• RQ5 Can the method be applied to energy exchange models and other gradient-type processes?

Key findings

• The spectral gap of the local Kac walk on the d-dimensional lattice satisfies $\lambda^{*,\text{loc}}(N) \geq \frac{1}{384dN^2}$, confirming the $N^{-2}$ scaling.
• For the Kac walk, the spectral gap on the complete graph is exactly $\lambda^*(N) = \frac{N+2}{4N}$, and this value is used as a reference for comparison.
• The method reduces the analysis of the lattice model to the complete graph model, enabling the use of known results for $N=3$.
• For zero-range processes, if the jump rate function satisfies $\sup_k |g(k+1)-g(k)| < \infty$ and $g(k) - g(j) \geq C$ for $k \geq j + k_0$, then $\lambda(2) > 0$, ensuring a positive spectral gap.
• When the spectral radius of the transition matrix $\mathcal{K}_n$ satisfies $\mu_2 < \frac{1}{2}$, the spectral gap $\lambda^*(3) > \frac{1}{3}$, which implies the $N^{-2}$ scaling.
• The method applies to a wide class of reversible, product-measure-preserving processes, including energy exchange models and exclusion processes, with explicit bounds obtainable via comparison.