New way to study patterns in bipartite and directed graphs
A new approach uses full graphicality to show that certain graph sampling methods mix rapidly.
Every fully graphic region of a specific form is -stable. This applies to bidegree sequence regions for bipartite and directed graphs.
The switch Markov chain is a method for sampling graph realizations with a prescribed degree sequence. -stability is a general sufficient condition for its rapid mixing.
A region is fully graphic if every bidegree sequence satisfying certain constraints has a bipartite realization. These constraints fix class sizes and , the common degree sum, and degree bounds.
The perturbation ratio is at most . This result means the switch Markov chain is rapidly mixing on every such region.
Full graphicality can be tested on a canonical extremal bidegree sequence. This allows for new explicit sufficient conditions.
Loopless directed graphs are encoded as bipartite graphs. This encoding develops a parallel theory for directed degree sequences.
Key facts
- Every fully graphic region of the described form is -stable.
- The perturbation ratio is at most .
- Full graphicality can be tested on a canonical extremal bidegree sequence.