The last decade witnessed several pivotal results on random inference problems where the aim is to learn a hidden ground truth from indirect randomised observations; much of this research has been guided by statistical physics intuition. Prominent examples include the stochastic block model, low-density parity check codes or compressed sensing. In all random inference problems studied so far the posterior distribution of the ground truth given the observations appears to enjoy a key property called “strong replica symmetry”. This means that the overlap of the posterior distribution with the ground truth (basically the number of bits that can be learned correctly) concentrates on a deterministic value. Whether this is generally true has been an open question. In this paper we discover an example of an inference problem based on a very simple random matrix over that fails to exhibit strong replica symmetry. Beyond its impact on random inference problems, the random matrix model, reminiscent of the binomial Erdős-Rényi random graph, gives rise to a natural random constraint satisfaction problem related to the intensely studied random k-XORSAT problem.

Given a large graph $H$, does the binomial random graph $G(n,p)$ contain a copy of $H$ as an induced subgraph with high probability? This classical question has been studied extensively for various graphs $H$, going back to the study of the independence number of $G(n,p)$ by Erdös and Bollobás and by Matula in 1976. In this paper we prove an asymptotically best possible result for induced matchings by showing that if $C/n\le p \le 0.99$ for some large constant $C$, then $G(n,p)$ contains an induced matching of order approximately $2\log_q(np)$, where $q= \frac{1}{1-p}$.

Given integers $k,j$ with $1\le j \le k-1$, we consider the length of the longest $j$-tight path in the binomial random $k$-uniform hypergraph $H^k(n,p)$. We show that this length undergoes a phase transition from logarithmic length to linear and determine the critical threshold, as well as proving upper and lower bounds on the length in the subcritical and supercritical ranges. In particular, for the supercritical case we introduce the \tt Pathfinder algorithm, a depth-first search algorithm which discovers $j$-tight paths in a $k$-uniform hypergraph. We prove that, in the supercritical case, with high probability this algorithm will find a long $j$-tight path.

One of the central topics in the theory of random graphs deals with the phase transition in the order of the largest components. In the binomial random graph $\mathcal{G}(n,p)$, the threshold for the appearance of the unique largest component (also known as the giant component) is $p_g = n^{-1}$. More precisely, when $p$ changes from $(1-\varepsilon)p_g$ (subcritical case) to $p_g$ and then to $(1+\varepsilon)p_g$ (supercritical case) for $\varepsilon>0$, with high probability the order of the largest component increases smoothly from $O(\varepsilon^{-2}\log(\varepsilon^3 n))$ to $\Theta(n^{2/3})$ and then to $(1 \pm o(1)) 2 \varepsilon n$. Furthermore, in the supercritical case, with high probability the largest components except the giant component are trees of order $O(\varepsilon^{-2}\log(\varepsilon^3 n))$, exhibiting a structural symmetry between the subcritical random graph and the graph obtained from the supercritical random graph by deleting its giant component. As a natural generalization of random graphs and connectedness, we consider the binomial random $k$-uniform hypergraph $\mathcal{H}^k(n,p)$ (where each $k$-tuple of vertices is present as a hyperedge with probability $p$ independently) and the following notion of high-order connectedness. Given an integer $1 \leq j \leq k-1$, two sets of $j$ vertices are called $j$-connected if there is a walk of hyperedges between them such that any two consecutive hyperedges intersect in at least $j$ vertices. A $j$-connected component is a maximal collection of pairwise $j$-connected $j$-tuples of vertices. Recently, the threshold for the appearance of the giant $j$-connected component in $\mathcal{H}^k(n,p)$ and its order were determined. In this article, we take a closer look at the subcritical random hypergraph. We determine the structure, order, and size of the largest $j$-connected components, with the help of a certain class of “hypertrees” and related objects. In our proofs, we combine various probabilistic and enumerative techniques, such as generating functions and couplings with branching processes. Our study will pave the way to establishing a symmetry between the subcritical random hypergraph and the hypergraph obtained from the supercritical random hypergraph by deleting its giant $j$-connected component.

We consider k-dimensional random simplicial complexes that are generated from the binomial random (k+1)-uniform hypergraph by taking the downward-closure, where k >= 2. For each 1 <= j <= k-1, we determine when all cohomology groups with coefficients in F_2 from dimension one up to j vanish and the zero-th cohomology group is isomorphic to F_2. This property is not monotone, but nevertheless we show that it has a single sharp threshold. Moreover, we prove a hitting time result, relating the vanishing of these cohomology groups to the disappearance of the last minimal obstruction. Furthermore, we study the asymptotic distribution of the dimension of the j-th cohomology group inside the critical window. As a corollary, we deduce a hitting time result for a different model of random simplicial complexes introduced in [Linial and Meshulam, Combinatorica, 2006], a result which has only been known for dimension two [Kahle and Pittel, Random Structures Algorithms, 2016].

In the planted bisection model a random graph G(n,p_+,p_-) with n vertices is created by partitioning the vertices randomly into two classes of equal size (up to plus or minus 1). Any two vertices that belong to the same class are linked by an edge with probability p_+ and any two that belong to different classes with probability (p_-) <(p_+) independently. The planted bisection model has been used extensively to benchmark graph partitioning algorithms. If (p_+)=2(d_+)/n and (p_-)=2(d_-)/n for numbers 0 <= (d_-) <(d_+) that remain fixed as n tends to infinity, then with high probability the "planted" bisection (the one used to construct the graph) will not be a minimum bisection. In this paper we derive an asymptotic formula for the minimum bisection width under the assumption that (d_+)-(d_-) > c * sqrt((d_+)ln(d_+)) for a certain constant c>0.