Abstract:Locally Markov walks are natural generalizations of classical Markov chains, where instead of a particle moving independently of the past, it decides where to move next depending on the last action performed at the current location. We introduce the concept of locally Markov walks on finite graphs and investigate several basic properties. For a particular locally Markov walk - the uniform unicycle walk on the complete graph - we investigate the mixing time and we prove that it exhibits cutoff.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about the basic properties and behaviors of locally Markov walks on finite graphs. Specifically, the paper mainly focuses on the following points: 1. **Introducing the concept of locally Markov walks**: The paper defines a new random walk model - locally Markov walks. Different from the classical Markov chain, in a locally Markov walk, the next move of a particle depends not only on the current position but also on the action taken at that position last time. 2. **Studying basic properties**: - **Stationary distribution**: The paper explores whether the stationary distribution of locally Markov walks exists and proves that these processes inherit some ergodic properties of the classical Markov chain. - **Eigenvalues, spectral gap and mixing time**: For locally Markov walks on finite graphs, the paper studies properties such as their stationary distribution, eigenvalues, spectral gap and mixing time. 3. **Research on specific examples**: - **Uniform unicycle walk**: The paper specifically studies the uniform unicycle walk on a complete graph, proves that its mixing time exhibits the "cutoff phenomenon", and calculates the spectrum of the transition matrix. ### Specific problem description The core problem of the paper is to explore whether locally Markov walks retain some typical properties of the classical Markov chain, such as the existence and convergence of the stationary distribution. In addition, the paper also studies the behavioral characteristics of these new processes on finite graphs, especially their mixing time and spectral properties. ### Key conclusions - **Theorem 1.1**: For a locally Markov walk \((X_n)_{n\in\mathbb{N}}\) on a finite strongly - connected graph \(G=(V, E)\), if all local chains have strictly positive transition probabilities, then there exists a stationary distribution and converges to this stationary distribution almost surely. - **Theorem 1.2**: The uniform unicycle walk on the complete graph \(K_m\) exhibits the cutoff phenomenon, and the cutoff time is \(m\log(m)\). Through these studies, the paper provides a new perspective for understanding a wider range of random walk models and reveals the unique properties of locally Markov walks.

Locally Markov walks on finite graphs

Locally Stationary Distributions: A Framework for Analyzing Slow-Mixing Markov Chains

Some properties of Markov chains on the free group $\mathbb F_2$

Random walk on sparse random digraphs

A transient Markov chain with finitely many cutpoints

Random walks on graphs with interval weights and precise marginals

Self-Repellent Random Walks on General Graphs -- Achieving Minimal Sampling Variance via Nonlinear Markov Chains

A local central limit theorem for random walks on expander graphs

Finite-sized one-dimensional lazy random walks

Local stationarity and time-inhomogeneous Markov chains

A random walk on the category of finite abelian $p$-groups

Combinatorics of nondeterministic walks

Bounds on Mixing Time for Time-Inhomogeneous Markov Chains

Markov chains, CAT(0) cube complexes, and enumeration: monotone paths in a strip mix slowly

Path Cohomology of Locally Finite Digraphs,Hodge's Theorem and the $p$-Lazy Random Walk

Local time, upcrossing time and weak cutpoints of a spatially inhomogeneous random walk on the line

Cutoff for mixtures of permuted Markov chains: reversible case

For every quantum walk there is a (classical) lifted Markov chain with faster mixing time

Irreducibility of Markov Chains on simplicial complexes, the Spectrum of the Discrete Hodge Laplacian and Homology

Local Access to Random Walks

Random walks with long-range steps generated by functions of Laplacian matrices