Abstract
In the Fastest Mixing Markov Chain problem, we are given a graph G= (V, E) and desire the discrete-time Markov chain with smallest mixing time τ subject to having equilibrium distribution uniform on V and non-zero transition probabilities only across edges of the graph. It is well-known that the mixing time τ RW of the lazy random walk on G is characterised by the edge conductance Φ of G via Cheeger’s inequality: Φ - 1≲ τ RW≲ Φ - 2log | V| . Analogously, we characterise the fastest mixing time τ ⋆ via a Cheeger-type inequality but for a different geometric quantity, namely the vertex conductance Ψ of G: Ψ - 1≲ τ ⋆≲ Ψ - 2(log | V|) 2 . This characterisation forbids fast mixing for graphs with small vertex conductance. To bypass this fundamental barrier, we consider Markov chains on G with equilibrium distribution which need not be uniform, but rather only ε -close to uniform in total variation. We show that it is always possible to construct such a chain with mixing time τ≲ ε - 1(diam G) 2log | V| . Finally, we discuss analogous questions for continuous-time and time-inhomogeneous chains.
| Original language | English |
|---|---|
| Pages (from-to) | 1017-1062 |
| Number of pages | 46 |
| Journal | Probability Theory and Related Fields |
| Volume | 188 |
| Issue number | 3-4 |
| Early online date | 30 Jan 2024 |
| DOIs | |
| Publication status | Published - 1 Apr 2024 |
Funding
Olesker-Taylor was supported by EPSRC Grant No. (EP/N004566/1).
| Funders | Funder number |
|---|---|
| Engineering and Physical Sciences Research Council | EP/N004566/1 |
Keywords
- 05C81
- 60J10
- 60J20
- 60J27
- Conductance
- Fastest mixing Markov chain
- Mixing time
- Random walks
ASJC Scopus subject areas
- Analysis
- Statistics and Probability
- Statistics, Probability and Uncertainty
Fingerprint
Dive into the research topics of 'Geometric bounds on the fastest mixing Markov chain'. Together they form a unique fingerprint.Cite this
- APA
- Standard
- Harvard
- Vancouver
- Author
- BIBTEX
- RIS