arXiv++ Combinatorics

Lecture notes of a master course given at Orsay between 2019-2024. Topics covered include Part I: One-dimensional random walks, cycle lemma and Bienaymé--Galton--Watson random trees. Part II: Erdös--Rényi random graphs, three proofs of the emergence of the giant component. Part III: Random recursive tree, random permutations and continuous time embedding techniques. Intended for publication.

Random walks on expanders play a crucial role in Markov Chain Monte Carlo algorithms, derandomization, graph theory, and distributed computing. A desirable property is that they are rapidly mixing, which is equivalent to having a spectral gap $γ$ (asymptotically) bounded away from $0$. Our work has two main strands. First, we establish a dichotomy for the robustness of mixing times on edge-weighted $d$-regular graphs (i.e., reversible Markov chains) subject to a Lipschitz condition, which bounds the ratio of adjacent weights by $β\geq 1$. If $β\ge 1$ is sufficiently small, then $γ\asymp 1$ and the mixing time is logarithmic in $n$. On the other hand, if $β\geq 2d$, there is an edge-weighting such that $γ$ is polynomially small in $1/n$. Second, we apply our robustness result to a time-dependent version of the so-called $\varepsilon$-biased random walk, as introduced in Azar et al. [Combinatorica 1996]. We show that, for any constant $\varepsilon>0$, a bias strategy can be chosen adaptively so that the $\varepsilon$-biased random walk covers any bounded-degree regular expander in $Θ(n)$ expected time, improving the previous-best bound of $O(n \log \log n)$. We prove the first non-trivial lower bound on the cover time of the $\varepsilon$-biased random walk, showing that, on bounded-degree regular expanders, it is $ω(n)$ whenever $\varepsilon = o(1)$. We establish this by controlling how much the probability of arbitrary events can be ``boosted'' by using a time-dependent bias strategy.

Given a family of hypergraphs $\mathcal{H}$, we say that a hypergraph $Γ$ is $\mathcal{H}$-universal if it contains every $H \in \mathcal{H}$ as a subgraph. For $D, r \in \mathbb{N}$, we construct an $r$-uniform hypergraph with $Θ\left(n^{r - r/D} \log^{r/D}(n)\right)$ edges which is universal for the family of all $r$-uniform hypergraphs with $n$ vertices and maximum degree at most $D$. This almost matches a trivial lower bound $Ω(n^{r - r/D})$ coming from the number of such hypergraphs. On a high level, we follow the strategy of Alon and Capalbo used in the graph case, that is $r = 2$. The construction of $Γ$ is deterministic and based on a bespoke product of expanders, whereas showing that $Γ$ is universal is probabilistic. Two key new ingredients are a decomposition result for hypergraphs of bounded density, based on Edmond's matroid partitioning theorem, and a tail bound for branching random walks on expanders.

We consider a collection of independent standard Brownian particles (or random walks), starting from a configuration where at least one particle is positive, and study the first time they all become negative. This is clearly equivalent to studying the first hitting time from the negative orthant or the first exit time from the complement of the negative orthant. While it turns out to be possible to compute the distribution of these hitting times for one and two particles, the distribution (and even its tail asymptotics) is not known in closed form for three or more particles. In this paper we study the tail asymptotics of the distribution as the number of particles tends to infinity. Our main techniques come from spectral geometry: we prove new asymptotic estimates for the principal eigenvalue of the complement of a high-dimensional orthant, which we believe are of independent interest.

We characterize the extremal structure for the exact mixing time for random walks on trees $T_{n,d}$ of order $n$ with diameter $d$. Given a graph $G=(V,E)$, let $H(v,π)$ denote the expected length of an optimal stopping rule from vertex $v$ to the stationary distributon $π$. We show that the quantity $\max_{G \in T_{n,d} } T_{\mbox{mix}}(G) = \max_{G \in T_{n,d} } \max_{v \in V} H(v,π)$ is achieved uniquely by the balanced double broom.

We prove sharp bounds on the virtual degrees introduced by Larsen and Shalev. This leads to improved bounds on characters of symmetric groups. We then sharpen bounds of Liebeck and Shalev concerning the Witten zeta function. Our main application is a characterization of the fixed-point free conjugacy classes whose associated random walk mixes in 2 steps.

This work is devoted to the analysis of a Gibbs partition model, also known as a composition scheme. We consider a natural new condition on the component weights. It leads to a new behavior for the total number of components. We discover a condensation phenomenon, producing a unique giant component comprising almost the entire mass. Additionally, we prove a point process limit describing the asymptotic size of the non-maximal components exhibiting a sublinear power-law growth. A particular motivation for our article stems from applications, ranging from simple random walks in the cube, over lattice paths models in the plane, pairs of directed random walks, over to urn models and card guessing games.

We consider multifold convolutions of a combinatorial sequence $(a_n)_{n=0}^{\infty}$: namely, for each $k \in \N$ the $k$-fold convolution is $\mathcal{M}^{(k)}_n(\boldsymbol{a}) = \sum_{j_1+\dots+j_k=n} a_{j_1} \cdots a_{j_k}$. Let $C_n$ be the Catalan numbers, and let $B_n$ be the central binomial coefficients. Then for random Dyck paths or simple random walk bridges, the multifold convolutions give moments of returns to the origin, using the stars-and-bars problem. There are well-known explicit formulas for the multifold convolutions of $C_n$ and $B_n$. But even for combinatorial sequences $B_n^2$ and $B_n^3$, one may determine asymptotics of multifold convolutions for large $n$. We also discuss large deviations: In a second part of the paper we consider an elementary version of the circle method for calculating asymptotics using complex analysis.

Consider the Euclidean traveling salesman problem with $n$ random points on the plane. Suppose that one of the points is shifted to a new random location. This gives us a new optimal path. Consider such shifts for each of the $n$ points. Do we get $n$ very different optimal paths? In this article, we show that this is not the case - in fact, the number of truly different paths can be at most $\mathcal{O}(1)$ as $n\to \infty$. The proof is based on a general argument which allows us to prove similar stability results in a number of other settings, such as branching random walk, the Sherrington-Kirkpatrick model of mean-field spin glasses, the Edwards-Anderson model of short-range spin glasses, and the Wigner ensemble of random matrices.

Phylogenetic trees constitute an interesting class of objects for stochastic processes due to the non-standard nature of the space they inhabit. In particular, many statistical applications require the construction of Markov processes on the space of trees, whose cardinality grows superexponentially with the number of leaves considered. We investigate whether certain lower-dimensional projections of tree space preserve the Markov property in tree-valued Markov processes. We study exact lumpability of tree shapes and $\varepsilon$-lumpability of clades, exploiting the combinatorial structure of the SPR graph to obtain bounds on the lumping error under the random walk and Metropolis-Hastings processes. Finally, we show how to use these results in empirical investigation, leveraging exact and $\varepsilon$-lumpability to improve Monte Carlo estimation of tree-related quantities.

This work studies the relation between two graph parameters, $ρ$ and $Λ$. For an undirected graph $G$, $ρ(G)$ is the growth rate of its universal covering tree, while $Λ(G)$ is a weighted geometric average of the vertex degree minus one, corresponding to the rate of entropy growth for the non-backtracking random walk (NBRW). It is well known that $ρ(G) \geq Λ(G)$ for all graphs, and that graphs with $ρ=Λ$ exhibit some special properties. In this work we derive an easy to check, necessary and sufficient condition for the equality to hold. Furthermore, we show that the variance of the number of random bits used by a length $\ell$ NBRW is $O(1)$ if $ρ= Λ$ and $Ω(\ell)$ if $ρ> Λ$. As a consequence we exhibit infinitely many non-trivial examples of graphs with $ρ= Λ$.

This paper concerns discrete-time occupancy processes on a finite graph. Our results can be formulated in two theorems, which are stated for vertex processes, but also applied to edge process (e.g., dynamic random graphs). The first theorem shows that concentration of local state averages is controlled by a random walk on the graph. The second theorem concerns concentration of polynomials of the vertex states. For dynamic random graphs, this allows to estimate deviations of edge density, triangle density, and more general subgraph densities. Our results only require Lipschitz continuity and hold for both dense and sparse graphs.

We present a relation between volumes of certain lower dimensional simplices associated to a full-dimensional primal and polar dual polytope in R^k. We then discuss an application of this relation to a geometric construction of a Colin de Verdiere matrix by Ivan Izmestiev. In the second part of the paper, we introduce a variation of vertex transitive polytopes, translate their associated Colin de Verdiere matrices into random walk matrices, and investigate extremality properties of the spectral gaps of these random walk matrices in two concrete examples - permutahedra of Coxeter groups and polytopes associated to the pure rotational tetrahedral group - where maximal spectral gaps correspond to equilateral polytopes.

When we want to simulate the realization of a symmetric simple random walk on $\mathbb Z^d$, we use $(2d)$-side fair dice to decide to which neighbor it jumps at each step if $d\geq 2$ or we simply use a fair coin when $d=1$. Assume that instead of using a dice or a coin we want to do a simulation using a well shuffled deck with $K$ cards of each of the $2d$ suits. In the first step the probability of jumping to each neighbor is $(2d)^{-1}$, but from the second step it becomes biased. Of course if we continue performing this simulation, the total variation distance between its law and the law of the random walk will increase until all cards are used. In this paper we investigate the minimum number of cards $N=2d K$ that a deck must contain so that the total variation distance between the law of a $n$-step simulation and the law of a $n$-step realization of the random walk is smaller than a chosen threshold $\varepsilon \in (0,1)$. More generally, we prove that when $N=cn$ this distance converges, as $n \to \infty$, to a Gaussian profile which depends on $c\geq 2d$. Furthermore, our analysis shows that this Gaussian profile vanishes as $c \to \infty$, proving the convergence of a multivariate hypergeometric distribution to a multinomial distribution in total variation.

Motivated by the Asymptotic Equipartition Property and its recently discovered role in the cutoff phenomenon, we initiate the systematic study of varentropy on discrete groups. Our main result is an approximate tensorization inequality which asserts that the varentropy of any conjugacy-invariant random walk is, up to a universal multiplicative constant, at most that of the free Abelian random walk with the same jump rates. In particular, it is always bounded by the number d of generators, uniformly in time and in the size of the group. This universal estimate is sharp and can be seen as a discrete analogue of a celebrated result of Bobkov and Madiman concerning random d-dimensional vectors with a log-concave density (AOP 2011). A key ingredient in our proof is the fact that conjugacy-invariant random walks have non-negative Bakry-Émery curvature, a result which seems new and of independent interest.

A recent paper by Bhatia, Chin, Mani, and Mossel (2026) defined stochastic processes aimed at modeling the game of War for {\em two players} with $n$ cards. That paper showed that these models, assuming uniform random decks, are equivalent to the Gambler's Ruin problem and therefore have an expected termination time of $Θ(n^2)$. In this paper, we generalize these models to {\em any number of players} $m$. We prove that the game with $m$ players is equivalent to a sticky random walk on an $m$-simplex; therefore, the termination time is the same as the absorption time of the sticky random walk. Interestingly, it seems that this absorption time has not been analyzed before. We show that the absorption time of the walk and the termination time of the game are both $Θ(n^2)$ for any number of players.

In this paper we study the asymptotic behavior of the (skew) Macdonald and Jack symmetric polynomials as the number of variables grows to infinity. We characterize their limits in terms of certain variational problems. As an intermediate step, we establish a large deviation principle for the $θ$ analogue of non-intersecting Bernoulli random walks. When $θ=1$, these walks are equivalent to random Lozenges tilings of strip domains, where the variational principle (with general domains and boundary conditions) has been proven in the seminal work by Cohn, Kenyon, and Propp. Our result gives a new argument of this variational principle, and also extends it to non-intersecting $θ$-Bernoulli random walks for any $θ\in (0,\infty)$. Remarkably, the rate functions remain identical, differing only by a factor of $1/θ$.

Bourgain's symmetrization theorem is a powerful technique reducing boolean analysis on product spaces to the cube. It states that for any product $Ω_i^{\otimes d}$, function $f: Ω_i^{\otimes d} \to \mathbb{R}$, and $q > 1$: $$||T_{\frac{1}{2}}f(x)||_q \leq ||\tilde{f}(r,x)||_{q} \leq ||T_{c_q}f(x)||_q$$ where $T_ρf = \sum\limits ρ^Sf^{=S}$ is the noise operator and $\widetilde{f}(r,x) = \sum\limits r_Sf^{=S}(x)$ `symmetrizes' $f$ by convolving its Fourier components $\{f^{=S}\}_{S \subseteq [d]}$ with a random boolean string $r \in \{\pm 1\}^d$. In this work, we extend the symmetrization theorem to high dimensional expanders (HDX). Building on (O'Donnell and Zhao 2021), we show this implies nearly-sharp $(2{\to}q)$-hypercontractivity for partite HDX. This resolves the main open question of (Gur, Lifshitz, and Liu STOC 2022) and gives the first fully hypercontractive subsets $X \subset [n]^d$ of support $n\cdot\exp(\text{poly}(d))$, an exponential improvement over Bafna, Hopkins, Kaufman, and Lovett's $n\cdot\exp(\exp(d))$ bound (BHKL STOC 2022). Adapting (Bourgain JAMS 1999), we also give the first booster theorem for HDX, resolving a main open question of BHKL. Our proof is based on two elementary new ideas in the theory of high dimensional expansion. First we introduce `$q$-norm HDX', generalizing standard spectral notions to higher moments, and observe every spectral HDX is a $q$-norm HDX. Second, we introduce a simple method of coordinate-wise analysis on HDX which breaks high dimensional random walks into coordinate-wise components and allows each component to be analyzed as a $\textit{$1$-dimensional}$ operator locally within $X$. This allows for application of standard tricks such as the replacement method, greatly simplifying prior analytic techniques.

Eppstein and Frishberg recently proved that the mixing time for the simple random walk on the $1$-skeleton of the associahedron is $O(n^3\log^3 n)$. We obtain similar rapid mixing results for the simple random walks on the $1$-skeleta of the type-$B$ and type-$D$ associahedra. We adapt Eppstein and Frishberg's technique to obtain the same bound of $O(n^3\log^3 n)$ in type $B$ and a bound of $O(n^{13} \log^2 n)$ in type $D$; in the process, we establish an expansion bound that is tight up to logarithmic factors in type $B$.

We construct an infinite family of triples (G,S1, S2) each consisting of a group G and a pair (S1, S2) of distinct subsets of G with the following properties. i The two Cayley graphs Cay(G, S1) and Cay(G,S2) are non-isomorphic. ii The distributions of the simple random walks on Cay(G,S1) and Cay(G,S2) are the same if one takes an appropriate correspondence between the two vertex sets at each step. iii The spectral set of Cay(G, Si) is decomposed into a disjoint union of two subsets A and B_i of the equal size which satisfies B1 = -B2.