SearcharxivSearch

arXiv subjects

Sam Olesker-Taylor

Publications and source records attributed to Sam Olesker-Taylor.

At least 19 recordsLinked to original sources

Sharp character bounds and cutoff for symmetric groups

We develop a flexible technique to bound the characters of symmetric groups, via the Naruse hook length formula, the Larsen--Shalev character bounds, and appropriate diagram slicings. It allows us to prove a uniform exponential character bound with optimal constant $1/2$. We furthermore prove sharp character bounds for conjugacy classes having a macroscopic number of fixed points, and deduce that the random walks on the associated Cayley graphs exhibit a total variation and $L^2$ cutoff.

math.RT

A Randomised Approach to Distributed Sorting

We introduce and analyse a new, extremely simple, randomised sorting algorithm: - choose a pair of indices $\{i, j\}$ according to some distribution $q$; - sort the elements in positions $i$ and $j$ of the array in ascending order. Choosing $q_{\{i,j\}} \propto 1/|j - i|$ yields an order-$n (\log n)^2$ sorting time. We call it the harmonic sorter. The sorter trivially parallelises in the asynchronous setting, yielding a linear speed-up. We also exhibit a low-communication, synchronous version with a linear speed-up. We compare and contrast this algorithm with other sorters, and discuss some of its benefits, particularly its robustness and amenability to parallelisation and distributed computing.

cs.DS

Time-Biased Random Walks and Robustness of Expanders

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 $\gamma$ (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 $\beta \geq 1$. If $\beta \ge 1$ is sufficiently small, then $\gamma \asymp 1$ and the mixing time is logarithmic in $n$. On the other hand, if $\beta \geq 2d$, there is an edge-weighting such that $\gamma$ 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 $\Theta(n)$ 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 $\omega(n)$ 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.

math.PR

Limit Profile for the Bernoulli--Laplace Urn

We analyse the convergence to equilibrium of the Bernoulli--Laplace urn model: initially, one urn contains $k$ red balls and a second $n-k$ blue balls; in each step, a pair of balls is chosen uniform and their locations are switched. Cutoff is known to occur at $\tfrac12 n \log \min\{k, \sqrt n\}$ with window order $n$ whenever $1 \ll k \le \tfrac12 n$. We refine this by determining the limit profile: a function $\Phi$ such that \[ d_\mathsf{TV}\bigl( \tfrac12 n \log \min\{k, \sqrt n\} + \theta n \bigr) \to \Phi(\theta) \quad\text{as}\quad n \to \infty \quad\text{for all}\quad \theta \in \mathbb R. \] Our main technical contribution, of independent interest, approximates a rescaled chain by a diffusion on $\mathbb R$ when $k \gg \sqrt n$, and uses its explicit law as a Gaussian process.

math.PR

Multicoloured Hardcore Model: Fast Mixing and Queueing

We extend the hardcore model to a multicoloured version: a subset of vertices of a graph are coloured such that no vertex is adjacent to one of the same colour; uncoloured vertices do not constrain neighbours. This mathematically models multi-channel resource sharing, such as fibreoptic routing. We analyse certain simple Glauber-type dynamics on such configurations, and find conditions which ensure fast mixing. These dynamics model a queueing system: customers queue for service at vertices, who only serve customers whilst they are coloured in the underlying configuration; uncoloured vertices sit idle. The mixing estimates are applied to control queue lengths in equilibrium.

math.PR

An Analysis of Elo Rating Systems via Markov Chains

We present a theoretical analysis of the Elo rating system, a popular method for ranking skills of players in an online setting. In particular, we study Elo under the Bradley--Terry--Luce model and, using techniques from Markov chain theory, show that Elo learns the model parameters at a rate competitive with the state of the art. We apply our results to the problem of efficient tournament design and discuss a connection with the fastest-mixing Markov chain problem.

math.PR

Catalan percolation

In Catalan percolation, all nearest-neighbor edges $\{i,i+1\}$ along $\mathbb Z$ are initially occupied, and all other edges are open independently with probability $p$. Open edges $\{i,j\}$ are occupied if some pair of edges $\{i,k\}$ and $\{k,j\}$, with $i<k<j$, become occupied. This model was introduced by Gravner and the third author, in the context of polluted graph bootstrap percolation. We prove that the critical $p_{\mathrm c}$ is strictly between that of oriented site percolation on $\mathbb Z^2$ and the Catalan growth rate $1/4$. Our main result shows that an enhanced oriented percolation model, with non-decaying infinite-range dependency, has a strictly smaller critical parameter than the classical model. This is reminiscent of the work of Duminil-Copin, Hil\'ario, Kozma and Sidoravicius on brochette percolation. Our proof differs, however, in that we do not use Aizenman--Grimmett enhancements or differential inequalities. Two key ingredients are the work of Hil\'ario, S\'a, Sanchis and Teixeira on stretched lattices, and the Russo--Seymour--Welsh result for oriented percolation by Duminil-Copin, Tassion and Teixeira.

math.PR

Limit Profile for Projections of Random Walks on Groups

Establishing cutoff, an abrupt transition from "not mixed" to "well mixed", is a classical topic in the theory of mixing times for Markov chains. Interest has grown recently in determining not only the existence of cutoff and the order of its mixing time and window, but the exact shape, or profile, of the convergence inside the window. Classical techniques, such as coupling or $\ell_2$-bounds, are typically too crude to establish this and there has been a push to develop general techniques. We build upon this work, extending from conjugacy-invariant random walks on groups to certain projections. We exemplify our method by analysing the $k$-particle interchange process on the complete $n$-graph with $k \asymp n$. This is a projection of the random-transposition card shuffle, which corresponds to $k = n$, analysed by Teyssier.

math.PR

Cutoff for Rewiring Dynamics on Perfect Matchings

We establish cutoff for a natural random walk (RW) on the set of perfect matchings (PMs). An $n$-PM is a pairing of $2n$ objects. The $k$-PM RW selects $k$ pairs uniformly at random, disassociates the corresponding $2k$ objects, then chooses a new pairing on these $2k$ objects uniformly at random. The equilibrium distribution is uniform over the set of all $n$-PM. We establish cutoff for the $k$-PM RW whenever $2 \le k \ll n$. If $k \gg 1$, then the mixing time is $\tfrac nk \log n$ to leading order. The case $k = 2$ was established by Diaconis and Holmes (2002) by relating the $2$-PM RW to the random transpositions card shuffle and also by Ceccherini-Silberstein, Scarabotti and Tolli (2007, 2008) using representation theory. We are the first to handle $k > 2$. Our argument builds on previous work of Berestycki, Schramm, Şengül and Zeitouni (2005, 2011, 2019) regarding conjugacy-invariant RWs on the permutation group.

math.PR

Geometric Bounds on the Fastest Mixing Markov Chain

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 $\tau$ 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 $\tau_\textsf{RW}$ of the lazy random walk on $G$ is characterised by the edge conductance $\Phi$ of $G$ via Cheeger's inequality: $\Phi^{-1} \lesssim \tau_\textsf{RW} \lesssim \Phi^{-2} \log |V|$. Analogously, we characterise the fastest mixing time $\tau^\star$ via a Cheeger-type inequality but for a different geometric quantity, namely the vertex conductance $\Psi$ of $G$: $\Psi^{-1} \lesssim \tau^\star \lesssim \Psi^{-2} (\log |V|)^2$. 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 $\varepsilon$-close to uniform in total variation. We show that it is always possible to construct such a chain with mixing time $\tau \lesssim \varepsilon^{-1} (\operatorname{diam} G)^2 \log |V|$. Finally, we discuss analogous questions for continuous-time and time-inhomogeneous chains.

math.PR

Metastability in Loss Networks with Dynamic Alternative Routing

Consider $N$ stations interconnected with links, each of capacity $K$, forming a complete graph. Calls arrive to each link at rate $λ$ and depart at rate $1$. If a call arrives to a link $x y$, connecting stations $x$ and $y$, which is at capacity, then a third station $z$ is chosen uniformly at random and the call is attempted to be routed via $z$: if both links $x z$ and $z y$ have spare capacity, then the call is held simultaneously on these two; otherwise the call is lost. We analyse an approximation of this model. We show rigorously that there are three phases according to the traffic intensity $α:= λ/K$: for $α\in (0,α_c) \cup (1,\infty)$, the system has mixing time logarithmic in the number of links $n := \binom N2$; for $α\in (α_c,1)$ the system has mixing time exponential in $n$, the number of links. Here $α_c := \tfrac13 (5 \sqrt{10} - 13) \approx 0.937$ is an explicit critical threshold with a simple interpretation. We also consider allowing multiple rerouting attempts. This has little effect on the overall behaviour; it does not remove the metastability phase. Finally, we add trunk reservation: in this, some number $σ$ of circuits are reserved; a rerouting attempt is only accepted if at least $σ+1$ circuits are available. We show that if $σ$ is chosen sufficiently large, depending only on $α$, not $K$ or $n$, then the metastability phase is removed.

math.PR

Supplementary Material for Random Cayley Graphs Project

This document contains supplementary material for the main articles in our Random Cayley Graphs project. We prove refined results about simple random walks on the integers and on the cycle. We are primarily interested in the entropy of these random walks at certain times and how this entropy changes when the time changes slightly. Additionally, we prove some large deviation and exit time estimates. We prove some results on the size of discrete lattice balls and how this size changes when the radius changes slightly. We do this in a general $L_q$ norm, with $q \in [1,\infty]$. We also prove some other technical results deferred from the main papers. We hope that some of the results, particularly the simple random walk estimates, will be useful in their own right for other researchers.

math.PR

Cutoff for Random Walks on Upper Triangular Matrices

Consider the random Cayley graph of a finite group $G$ with respect to $k$ generators chosen uniformly at random, with $1 \ll \log k \ll \log |G|$ (ie $1 \ll k = |G|^{o(1)}$). A conjecture of Aldous and Diaconis (1985) asserts, for $k\gg\log|G|$, that the random walk on this graph exhibits cutoff. When $\log k \lesssim \log\log|G|$ (ie $k = (\log |G|)^{\mathcal O(1)}$), the only example of a non-Abelian group for which cutoff has been established is the dihedral group. We establish cutoff (as $p\to infty$) for the group of $d \times d$ unit upper triangular matrices with integer entries modulo $p$ (prime), which we denote $U_{p,d}$, for fixed $d$ or $d$ diverging sufficiently slowly. We allow $1 \ll k \lesssim \log |U_{p,d}|$ as well as $k\gg\log|U_{p,d}|$. The cutoff time is $\max\{\log_k |U_{p,d}|, \: s_0 k\}$, where $s_0$ is the time at which the entropy of the random walk on $\mathbb Z$ reaches $(\log |U_{p,d}^\mathrm{ab}|)/k$, where $U_{p,d}^\mathrm{ab} \cong \mathbb Z_p^{d-1}$ is the Abelianisation of $U_{p,d}$. When $1 \ll k \ll \log |U_{p,d}^\mathrm{ab}|$ and $d \asymp 1$, we find the limit profile. We also prove highly related results for the $d$-dimensional Heisenberg group over $\mathbb Z_p$. The Aldous--Diaconis conjecture also asserts, for $k gg\log |G|$, that the cutoff time should depend only on $k$ and $|G|$. This was verified for all Abelian groups. Our result shows that this is not the case for $U_{p,d}$: the cutoff time depends on $k$, $|U_{p,d}| = p^{d(d-1)/2}$ and $|U_{p,d}^\mathrm{ab}|=p^{d-1}$. We also show that all but $o(|U_{p,d}|)$ of the elements of $U_{p,d}$ lie at graph distance $M \pm o(M)$ from the identity, where $M$ is the minimal radius of a ball in $\mathbb Z^k$ of cardinality $|U_{p,d}^\mathrm{ab}| = p^{d-1}$. Finally, we show that the diameter is also asymptotically $M$ when $k \gtrsim \log |U_{p,d}^\textrm{ab}|$ and $d\asymp1$.

math.PR

Further Results and Discussions on Random Cayley Graphs

Consider the random Cayley graph of a finite group $G$ with respect to $k$ generators chosen uniformly at random, with $1 \ll k \lesssim \log |G|$. The results of this article supplement those in the three main papers on random Cayley graphs. The majority of the results are inspired by a `universality' conjecture of Aldous and Diaconis (1985). To start, we study the limit profile of cutoff for the simple random walk on this random graph, as well as a detailed investigation into mixing properties when $G = \mathbb Z_p^d$ with $p$ prime. We then exposit a proof of Diaconis and Saloff-Coste (1994) establishing lack of cutoff when $k \asymp 1$. We move onto discussing material from our companion paper on matrix groups. We then study distance of a typical element of $G$ from the identity in an $L_q$-type graph distance in the Abelian set-up. Finally, we give necessary and sufficient conditions for $k$ independent uniform elements of $G$ to generate $G$, ie for the random Cayley graph to be connected, based on work of Pomerance (2001). The aforementioned results all hold with high probability over the random Cayley graph.

math.PR

Cutoff for Almost All Random Walks on Abelian Groups

Consider the random Cayley graph of a finite group $G$ with respect to $k$ generators chosen uniformly at random, with $1 \ll \log k \ll \log |G|$; denote it $G_k$. A conjecture of Aldous and Diaconis (1985) asserts, for $k \gg \log |G|$, that the random walk on this graph exhibits cutoff. Further, the cutoff time should be a function only of $k$ and $|G|$, to sub-leading order. This was verified for all Abelian groups in the '90s. We extend the conjecture to $1 \ll k \lesssim \log |G|$. We establish cutoff for all Abelian groups under the condition $k - d(G) \gg 1$, where $d(G)$ is the minimal size of a generating subset of $G$, which is almost optimal. The cutoff time is described (abstractly) in terms of the entropy of random walk on $\mathbb Z^k$. This abstract definition allows us to deduce that the cutoff time can be written as a function only of $k$ and $|G|$ when $d(G) \ll \log |G|$ and $k - d(G) \asymp k \gg 1$; this is not the case when $d(G) \asymp \log |G| \asymp k$. For certain regimes of $k$, we find the limit profile of the convergence to equilibrium. Wilson (1997) conjectured that $\mathbb Z_2^d$ gives rise to the slowest mixing time for $G_k$ amongst all groups of size at most $2^d$. We give a partial answer, verifying the conjecture for nilpotent groups. This is obtained via a comparison result of independent interest between the mixing times of nilpotent $G$ and a corresponding Abelian group $\overline G$, namely the direct sum of the Abelian quotients in the lower central series of $G$. We use this to refine a celebrated result of Alon and Roichman (1994): we show for nilpotent $G$ that $G_k$ is an expander provided $k - d(\overline G) \gtrsim \log |G|$. As another consequence, we establish cutoff for nilpotent groups with relatively small commutators, including high-dimensional special groups, such as Heisenberg groups.

math.PR

Geometry of Random Cayley Graphs of Abelian Groups

Consider the random Cayley graph of a finite Abelian group $G$ with respect to $k$ generators chosen uniformly at random, with $1 \ll \log k \ll \log |G|$. Draw a vertex $U \sim \operatorname{Unif}(G)$. We show that the graph distance $\operatorname{dist}(\mathsf{id},U)$ from the identity to $U$ concentrates at a particular value $M$, which is the minimal radius of a ball in $\mathbb Z^k$ of cardinality at least $|G|$, under mild conditions. In other words, the distance from the identity for all but $o(|G|)$ of the elements of $G$ lies in the interval $[M - o(M), M + o(M)]$. In the regime $k \gtrsim \log |G|$, we show that the diameter of the graph is also asymptotically $M$. In the spirit of a conjecture of Aldous and Diaconis (1985), this $M$ depends only on $k$ and $|G|$, not on the algebraic structure of $G$. Write $d(G)$ for the minimal size of a generating subset of $G$. We prove that the order of the spectral gap is $|G|^{-2/k}$ when $k - d(G) \asymp k$ and $|G|$ lies in a density-$1$ subset of $\mathbb N$ or when $k - 2 d(G) \asymp k$. This extends, for Abelian groups, a celebrated result of Alon and Roichman (1994). The aforementioned results all hold with high probability over the random Cayley graph.

math.PR

Cutoff for Random Walk on Dynamical Erdős--Rényi Graph

We consider dynamical percolation on the complete graph $K_n$, where each edge refreshes its state at rate $μ\ll 1/n$, and is then declared open with probability $p = λ/n$ where $λ> 1$. We study a random walk on this dynamical environment which jumps at rate $1/n$ along every open edge. We show that the mixing time of the full system exhibits cutoff at $\log n/μ$. We do this by showing that the random walk component mixes faster than the environment process; along the way, we control the time it takes for the walk to become isolated.

math.PR

Chen--Stein Method for the Uncovered Set of Random Walk on $\mathbb Z_n^d$ for $d \ge 3$

Let $X$ be a simple random walk on $\mathbb{Z}_n^d$ with $d\geq 3$ and let $t_{\rm{cov}}$ be the expected cover time. We consider the set of points $\mathcal{U}_α$ of $\mathbb{Z}_n^d$ that have not been visited by the walk by time $αt_{\rm{cov}}$ for $α\in (0,1)$. It was shown in [MS17] that there exists $α_1(d)\in (0,1)$ such that for all $α>α_1(d)$ the total variation distance between the law of the set $\mathcal{U}_α$ and an i.i.d. sequence of Bernoulli random variables indexed by $\mathbb{Z}_n^d$ with success probability $n^{-αd}$ tends to $0$ as $n \to \infty$. In [MS17] the constant $α_1(d)$ converges to $1$ as $d\to\infty$. In this short note using the Chen--Stein method and a concentration result for Markov chains of Lezaud we greatly simplify the proof of [MS17] and find a constant $α_1(d)$ which converges to $3/4$ as $d\to\infty$.

math.PR