English

Parameter estimation for Gibbs distributions

Data Structures and Algorithms 2025-04-04 v9 Computational Complexity Discrete Mathematics Probability

Abstract

We consider Gibbs distributions, which are families of probability distributions over a discrete space Ω\Omega with probability mass function of the form μβΩ(ω)eβH(ω)\mu^\Omega_\beta(\omega) \propto e^{\beta H(\omega)} for β\beta in an interval [βmin,βmax][\beta_{\min}, \beta_{\max}] and H(ω){0}[1,n]H( \omega ) \in \{0 \} \cup [1, n]. The partition function is the normalization factor Z(β)=ωΩeβH(ω)Z(\beta)=\sum_{\omega \in\Omega}e^{\beta H(\omega)}. Two important parameters of these distributions are the log partition ratio q=logZ(βmax)Z(βmin)q = \log \tfrac{Z(\beta_{\max})}{Z(\beta_{\min})} and the counts cx=H1(x)c_x = |H^{-1}(x)|. These are correlated with system parameters in a number of physical applications and sampling algorithms. Our first main result is to estimate the counts cxc_x using roughly O~(qε2)\tilde O( \frac{q}{\varepsilon^2}) samples for general Gibbs distributions and O~(n2ε2)\tilde O( \frac{n^2}{\varepsilon^2} ) samples for integer-valued distributions (ignoring some second-order terms and parameters), and we show this is optimal up to logarithmic factors. We illustrate with improved algorithms for counting connected subgraphs, independent sets, and perfect matchings. As a key subroutine, we also develop algorithms to compute the partition function ZZ using O~(qε2)\tilde O(\frac{q}{\varepsilon^2}) samples for general Gibbs distributions and using O~(n2ε2)\tilde O(\frac{n^2}{\varepsilon^2}) samples for integer-valued distributions.

Keywords

Cite

@article{arxiv.2007.10824,
  title  = {Parameter estimation for Gibbs distributions},
  author = {David G. Harris and Vladimir Kolmogorov},
  journal= {arXiv preprint arXiv:2007.10824},
  year   = {2025}
}

Comments

This is a longer version which extends two previous papers "A Faster Approximation Algorithm for the Gibbs Partition Function" (arXiv:1608.04223), published in COLT 2018, and "Parameter estimation for Gibbs distributions" published in ICALP 2023

R2 v1 2026-06-23T17:16:55.735Z