English

A Faster Approximation Algorithm for the Gibbs Partition Function

Data Structures and Algorithms 2017-12-29 v4

Abstract

We consider the problem of estimating the partition function Z(β)=xexp(β(H(x))Z(\beta)=\sum_x \exp(-\beta(H(x)) of a Gibbs distribution with a Hamilton H()H(\cdot), or more precisely the logarithm of the ratio q=lnZ(0)/Z(β)q=\ln Z(0)/Z(\beta). It has been recently shown how to approximate qq with high probability assuming the existence of an oracle that produces samples from the Gibbs distribution for a given parameter value in [0,β][0,\beta]. The current best known approach due to Huber [9] uses O(qlnn[lnq+lnlnn+ε2])O(q\ln n\cdot[\ln q + \ln \ln n+\varepsilon^{-2}]) oracle calls on average where ε\varepsilon is the desired accuracy of approximation and H()H(\cdot) is assumed to lie in {0}[1,n]\{0\}\cup[1,n]. We improve the complexity to O(qlnnε2)O(q\ln n\cdot\varepsilon^{-2}) oracle calls. We also show that the same complexity can be achieved if exact oracles are replaced with approximate sampling oracles that are within O(ε2qlnn)O(\frac{\varepsilon^2}{q\ln n}) variation distance from exact oracles. Finally, we prove a lower bound of Ω(qε2)\Omega(q\cdot \varepsilon^{-2}) oracle calls under a natural model of computation.

Keywords

Cite

@article{arxiv.1608.04223,
  title  = {A Faster Approximation Algorithm for the Gibbs Partition Function},
  author = {Vladimir Kolmogorov},
  journal= {arXiv preprint arXiv:1608.04223},
  year   = {2017}
}