New algorithms for counting and sampling on specific graphs

The C-3PO framework provides ways to estimate values and samples for spin models on graphs with low threshold rank or high girth.

Researchers developed the C-3PO framework. This framework provides approximate counting and sampling algorithms for spin models on certain graphs.

The algorithms work on graphs with low threshold rank or high girth. An ε\varepsilon-approximate counting algorithm outputs an estimate Z^\widehat Z of the partition function ZZ.

This estimate satisfies (1−ε)Z≤Z^≤(1+ε)Z(1-\varepsilon)Z \le \widehat Z \le (1+\varepsilon)Z with a probability of at least 3/43/4. An ε\varepsilon-approximate sampling algorithm outputs a configuration within total variation distance ε\varepsilon of the target distribution.

The C-3PO algorithm works by enumerating small cuts. It reconstructs the original measure through cluster expansions. It uses an improvement of subspace enumeration methods via active regression.

The framework applies to the Ferromagnetic Potts model and the bipartite hard-core model. For the Potts model, it works on arbitrary graphs or Δ\Delta-regular graphs of girth gg.

For the bipartite hard-core model, it works on Δ\Delta-regular vertex-transitive bipartite graphs or Δ\Delta-regular bipartite graphs of girth gg.

Key facts

  • The C-3PO framework provides approximate counting and sampling algorithms.
  • The algorithms work on graphs with low threshold rank or high girth.
  • The method uses small cuts and cluster expansions to reconstruct the original measure.

All sections

Edition