Search

In the Max-2Lin(2) problem you are given a system of equations on the form $x_i + x_j \equiv b \pmod{2}$, and your objective is to find an assignment that satisfies as many equations as possible. Let $c \in [0.5, 1]$ denote the maximum fraction of satisfiable equations. In this paper we construct a curve $s (c)$ such that it is NP-hard to find a solution satisfying at least a fraction $s$ of equations. This curve either matches or improves all of the previously known inapproximability NP-hardness results for Max-2Lin(2). In particular, we show that if $c \geqslant 0.9232$ then $\frac{1 - s (c)}{1 - c} > 1.48969$, which improves the NP-hardness inapproximability constant for the min deletion version of Max-2Lin(2). Our work complements the work of O'Donnell and Wu that studied the same question assuming the Unique Games Conjecture. Similar to earlier inapproximability results for Max-2Lin(2), we use a gadget reduction from the $(2^k - 1)$-ary Hadamard predicate. Previous works used $k$ ranging from $2$ to $4$. Our main result is a procedure for taking a gadget for some fixed $k$, and use it as a building block to construct better and better gadgets as $k$ tends to infinity. Our method can be used to boost the result of both smaller gadgets created by hand $(k = 3)$ or larger gadgets constructed using a computer $(k = 4)$., Comment: Source code: https://github.com/bjorn-martinsson/NP-hardness-of-Max-2Lin-2

Solving very large linear systems of equations is a key computational task in science and technology. In many cases, the coefficient matrix of the linear system is rank-deficient, leading to systems that may be underdetermined, inconsistent, or both. In such cases, one generally seeks to compute the least squares solution that minimizes the residual of the problem, which can be further defined as the solution with smallest norm in cases where the coefficient matrix has a nontrivial nullspace. This work presents several new techniques for solving least squares problems involving coefficient matrices that are so large that they do not fit in main memory. The implementations include both CPU and GPU variants. All techniques rely on complete orthogonal decompositions that guarantee that both conditions of a least squares solution are met, regardless of the rank properties of the matrix. Specifically, they rely on the recently proposed "randUTV" algorithm that is particularly effective in strongly communication-constrained environments. A detailed precision and performance study reveals that the new methods, that operate on data stored on disk, are competitive with state-of-the-art methods that store all data in main memory., Comment: 26 pages, 12 figures

For a finite Abelian group $(\Gamma,+)$, let $n(\Gamma)$ denote the smallest positive integer $n$ such that for each labelling of the arcs of the complete digraph of order $n$ using elements from $\Gamma$, there exists a directed cycle such that the total sum of the arc-labels along the cycle equals $0$. Alon and Krivelevich initiated the study of the parameter $n(\cdot)$ on cyclic groups and proved that $n(\mathbb{Z}_q)=O(q\log q)$. Studying the prototypical case when $\Gamma=\mathbb{Z}_p^d$ is a power of a cyclic group of prime order, Letzter and Morrison recently showed that $n(\mathbb{Z}_p^d) \le O(pd(\log d)^2)$ and that $n(\mathbb{Z}_2^d)\le O(d \log d)$. They then posed the problem of proving an (asymptotically optimal) upper bound of $n(\mathbb{Z}_p^d)\le O(pd)$ for all primes $p$ and $d \in \mathbb{N}$. In this paper, we solve this problem for $p=2$ and improve their bound for all primes $p \ge 3$ by proving $n(\mathbb{Z}_2^d)\le 5d$ and $n(\mathbb{Z}_p^d)\le O(pd\log d)$. While the first bound determines $n(\mathbb{Z}_2^d)$ up to a multiplicative error of $5$, the second bound is tight up to a $\log d$ factor. Moreover, our result shows that a tight bound of $n(\mathbb{Z}_p^d)=\Theta(pd)$ for arbitrary $p$ and $d$ would follow from a (strong form) of the well-known conjecture of Jaeger, Linial, Payan and Tarsi on additive bases in $\mathbb{Z}_p^d$. Along the way to proving these results, we establish a generalization of a hypergraph matching result by Haxell in a matroidal setting. Concretely, we obtain sufficient conditions for the existence of matchings in a hypergraph whose hyperedges are labelled by the elements of a matroid, with the property that the edges in the matching induce a basis of the matroid. We believe that these statements are of independent interest., Comment: 6 Pages

A linearly ordered (LO) $k$-colouring of a hypergraph assigns to each vertex a colour from the set $\{0,1,\ldots,k-1\}$ in such a way that each hyperedge has a unique maximum element. Barto, Batistelli, and Berg conjectured that it is NP-hard to find an LO $k$-colouring of an LO 2-colourable 3-uniform hypergraph for any constant $k\geq 2$ [STACS'21] but even the case $k=3$ is still open. Nakajima and \v{Z}ivn\'{y} gave polynomial-time algorithms for finding, given an LO 2-colourable 3-uniform hypergraph, an LO colouring with $O^*(\sqrt{n})$ colours [ICALP'22] and an LO colouring with $O^*(\sqrt[3]{n})$ colours [ACM ToCT'23]. Very recently, Louis, Newman, and Ray gave an SDP-based algorithm with $O^*(\sqrt[5]{n})$ colours. We present two simple polynomial-time algorithms that find an LO colouring with $O(\log_2(n))$ colours, which is an exponential improvement., Comment: This paper is a merger of independent work by H{\aa}stad and Martinsson, and by Nakajima and \v{Z}ivn\'y respectively. A full, slightly improved version of an APPROX'24 paper. A discussion of related work on unique-maximum colourings and other similar notions

We propose an adaptive change point detection method (A-CPD) for machine guided weak label annotation of audio recording segments. The goal is to maximize the amount of information gained about the temporal activations of the target sounds. For each unlabeled audio recording, we use a prediction model to derive a probability curve used to guide annotation. The prediction model is initially pre-trained on available annotated sound event data with classes that are disjoint from the classes in the unlabeled dataset. The prediction model then gradually adapts to the annotations provided by the annotator in an active learning loop. We derive query segments to guide the weak label annotator towards strong labels, using change point detection on these probabilities. We show that it is possible to derive strong labels of high quality with a limited annotation budget, and show favorable results for A-CPD when compared to two baseline query segment strategies., Comment: Accepted at EUSIPCO 2024 (nominated best student paper)

Land cover classification and change detection are two important applications of remote sensing and Earth observation (EO) that have benefited greatly from the advances of deep learning. Convolutional and transformer-based U-net models are the state-of-the-art architectures for these tasks, and their performances have been boosted by an increased availability of large-scale annotated EO datasets. However, the influence of different visual characteristics of the input EO data on a model's predictions is not well understood. In this work we systematically examine model sensitivities with respect to several color- and texture-based distortions on the input EO data during inference, given models that have been trained without such distortions. We conduct experiments with multiple state-of-the-art segmentation networks for land cover classification and show that they are in general more sensitive to texture than to color distortions. Beyond revealing intriguing characteristics of widely used land cover classification models, our results can also be used to guide the development of more robust models within the EO domain.

A graph $G$ is said to be Ramsey for a tuple of graphs $(H_1,\dots,H_r)$ if every $r$-coloring of the edges of $G$ contains a monochromatic copy of $H_i$ in color $i$, for some $i$. A fundamental question at the intersection of Ramsey theory and the theory of random graphs is to determine the threshold at which the binomial random graph $G_{n,p}$ becomes a.a.s. Ramsey for a fixed tuple $(H_1,\dots,H_r)$, and a famous conjecture of Kohayakawa and Kreuter predicts this threshold. Earlier work of Mousset-Nenadov-Samotij, Bowtell-Hancock-Hyde, and Kuperwasser-Samotij-Wigderson has reduced this probabilistic problem to a deterministic graph decomposition conjecture. In this paper, we resolve this deterministic problem, thus proving the Kohayakawa-Kreuter conjecture. Along the way, we prove a number of novel graph decomposition results which may be of independent interest., Comment: 28 pages, updated according to the referee's comments

The hierarchical matrix ($\mathcal{H}^{2}$-matrix) formalism provides a way to reinterpret the Fast Multipole Method and related fast summation schemes in linear algebraic terms. The idea is to tessellate a matrix into blocks in such as way that each block is either small or of numerically low rank; this enables the storage of the matrix and the application of it to a vector in linear or close to linear complexity. A key motivation for the reformulation is to extend the range of dense matrices that can be represented. Additionally, $\mathcal{H}^{2}$-matrices in principle also extend the range of operations that can be executed to include matrix inversion and factorization. While such algorithms can be highly efficient for certain specialized formats (such as HBS/HSS matrices based on ``weak admissibility''), inversion algorithms for general $\mathcal{H}^{2}$-matrices tend to be based on nested recursions and recompressions, making them challenging to implement efficiently. An exception is the \textit{strong recursive skeletonization (SRS)} algorithm by Minden, Ho, Damle, and Ying, which involves a simpler algorithmic flow. However, SRS greatly increases the number of blocks of the matrix that need to be stored explicitly, leading to high memory requirements. This manuscript presents the \textit{randomized strong recursive skeletonization (RSRS)} algorithm, which is a reformulation of SRS that incorporates the randomized SVD (RSVD) to simultaneously compress and factorize an $\mathcal{H}^{2}$-matrix. RSRS is a ``black box'' algorithm that interacts with the matrix to be compressed only via its action on vectors; this extends the range of the SRS algorithm (which relied on the ``proxy source'' compression technique) to include dense matrices that arise in sparse direct solvers.

This work introduces the kernel-independent multi-level algorithm "skelFMM" for evaluating all pairwise interactions between $N$ points connected through a kernel such as the fundamental solution of the Laplace or the Helmholtz equations. The method is based on linear algebraic tools such as randomized low rank approximation and "skeleton representations" of far-field interactions. The work is related to previously proposed linear algebraic reformulations of the fast multipole method (FMM), but is distinguished by relying on simpler data structures. In particular, skelFMM does not require an "interaction list", as it relies instead on algebraically-modified kernel interactions between near-neighbors at every level. Like other kernel independent algorithms, it only requires evaluation of the kernel function, allowing the methodology to easily be extended to a range of different kernels in 2D and 3D. The simplicity of the algorithm makes it particularly amenable to parallel implementation on heterogeneous hardware architectures. The performance of the algorithm is demonstrated through numerical experiments conducted on uniform and non-uniform point distributions in 2D and 3D, involving Laplace and (low frequency) Helmholtz kernels. The algorithm relies on a precomputation stage that constructs a tailored representation for a given geometry of points. Once the precomputation has completed, the matrix-vector multiplication attains high speed through GPU acceleration that leverages batched linear algebra.

Boundary value problems involving elliptic PDEs such as the Laplace and the Helmholtz equations are ubiquitous in mathematical physics and engineering. Many such problems can be alternatively formulated as integral equations that are mathematically more tractable. However, an integral-equation formulation poses a significant computational challenge: solving large dense linear systems that arise upon discretization. In cases where iterative methods converge rapidly, existing methods that draw on fast summation schemes such as the Fast Multipole Method are highly efficient and well-established. More recently, linear complexity direct solvers that sidestep convergence issues by directly computing an invertible factorization have been developed. However, storage and computation costs are high, which limits their ability to solve large-scale problems in practice. In this work, we introduce a distributed-memory parallel algorithm based on an existing direct solver named ``strong recursive skeletonization factorization.'' Specifically, we apply low-rank compression to certain off-diagonal matrix blocks in a way that minimizes computation and data movement. Compared to iterative algorithms, our method is particularly suitable for problems involving ill-conditioned matrices or multiple right-hand sides. Large-scale numerical experiments are presented to show the performance of our Julia implementation.

Let $r$ be any positive integer. We prove that for every sufficiently large $k$ there exists a $k$-chromatic vertex-critical graph $G$ such that $\chi(G-R)=k$ for every set $R \subseteq E(G)$ with $|R|\le r$. This partially solves a problem posed by Erd\H{o}s in 1985, who asked whether the above statement holds for $k \ge 4$., Comment: 6 pages

Interpolative and CUR decompositions involve "natural bases" of row and column subsets, or skeletons, of a given matrix that approximately span its row and column spaces. These low-rank decompositions preserve properties such as sparsity or non-negativity, and are easily interpretable in the context of the original data. For large-scale problems, randomized sketching to sample the row or column spaces with a random matrix can serve as an effective initial step in skeleton selection to reduce computational cost. A by now well-established approach has been to extract a randomized sketch, followed by column-pivoted QR CPQR) on the sketch matrix. This manuscript describes an alternative approach where CPQR is replaced by LU with partial pivoting (LUPP). While LUPP by itself is not rank-revealing, it is demonstrated that when used in a randomized setting, LUPP not only reveals the numerical rank, but also allows the estimation of the residual error as the factorization is built. The resulting algorithm is both adaptive and parallelizable, and attains much higher practical speed due to the lower communication requirements of LUPP over CPQR. The method has been implemented for both CPUs and GPUs, and the resulting software has been made publicly available.

The interpolative decomposition (ID) aims to construct a low-rank approximation formed by a basis consisting of row/column skeletons in the original matrix and a corresponding interpolation matrix. This work explores fast and accurate ID algorithms from five essential perspectives for empirical performance: (a) skeleton complexity that measures the minimum possible ID rank for a given low-rank approximation error, (b) asymptotic complexity in FLOPs, (c) parallelizability of the computational bottleneck as matrix-matrix multiplications, (d) error-revealing property that enables automatic rank detection for given error tolerances without prior knowledge of target ranks, (e) ID-revealing property that ensures efficient construction of the optimal interpolation matrix after selecting the skeletons. While a broad spectrum of algorithms have been developed to optimize parts of the aforementioned perspectives, practical ID algorithms proficient in all perspectives remain absent. To fill in the gap, we introduce robust blockwise random pivoting (RBRP) that is parallelizable, error-revealing, and exactly ID-revealing, with comparable skeleton and asymptotic complexities to the best existing ID algorithms in practice. Through extensive numerical experiments on various synthetic and natural datasets, we demonstrate the appealing empirical performance of RBRP from the five perspectives above, as well as the robustness of RBRP to adversarial inputs.

The burning number of a graph $G$ is the smallest number $b$ such that the vertices of $G$ can be covered by balls of radii $0, 1, \dots, b-1$. As computing the burning number of a graph is known to be NP-hard, even on trees, it is natural to consider polynomial time approximation algorithms for the quantity. The best known approximation factor in the literature is $3$ for general graphs and $2$ for trees. In this note we give a $2/(1-e^{-2})+\varepsilon=2.313\dots$-approximation algorithm for the burning number of general graphs, and a PTAS for the burning number of trees and forests. Moreover, we show that computing a $(\frac53-\varepsilon)$-approximation of the burning number of a general graph $G$ is NP-hard., Comment: 7 pages, no figures. Comments are welcome!

In a recent article by Chapuy and Perarnau, it was shown that a uniformly chosen automaton on $n$ states with a $2$-letter alphabet has a synchronizing word of length $O(\sqrt{n}\log n)$ with high probability. In this note, we improve this result by showing that, for any $\varepsilon>0$, there exists a synchronizing word of length $O(\varepsilon^{-1}\sqrt{n \log n})$ with probability $1-\varepsilon$. Our proof is based on two properties of random automata. First, there are words $\omega$ of length $O(\sqrt{n \log n})$ such that the expected number of possible states for the automaton, after inputting $\omega$, is $O(\sqrt{n/\log n})$. Second, with high probability, each pair of states can be synchronized by a word of length $O(\log n)$., Comment: 10 pages, no figures. comments welcome

An additive Runge-Kutta method is used for the time stepping, which integrates the linear stiff terms by an explicit singly diagonally implicit Runge-Kutta (ESDIRK) method and the nonlinear terms by an explicit Runge-Kutta (ERK) method. In each time step, the implicit solve is performed by the recently developed Hierarchical Poincar\'e-Steklov (HPS) method. This is a fast direct solver for elliptic equations that decomposes the space domain into a hierarchical tree of subdomains and builds spectral collocation solvers locally on the subdomains. These ideas are naturally combined in the presented method since the singly diagonal coefficient in ESDIRK and a fixed time-step ensures that the coefficient matrix in the implicit solve of HPS remains the same for all time stages. This means that the precomputed inverse can be efficiently reused, leading to a scheme with complexity (in two dimensions) $\mathcal{O}(N^{1.5})$ for the precomputation where the solution operator to the elliptic problems is built, and then $\mathcal{O}(N \log N)$ for the solve in each time step. The stability of the method is proved for first order in time and any order in space, and numerical evidence substantiates a claim of stability for a much broader class of time discretization methods. Numerical experiments supporting the accuracy of efficiency of the method in one and two dimensions are presented.

For any positive edge density $p$, a random graph in the Erd\H{o}s-Renyi $G_{n,p}$ model is connected with non-zero probability, since all edges are mutually independent. We consider random graph models in which edges that do not share endpoints are independent while incident edges may be dependent and ask: what is the minimum probability $\rho(n)$, such that for any distribution $\mathcal{G}$ (in this model) on graphs with $n$ vertices in which each potential edge has a marginal probability of being present at least $\rho(n)$, a graph drawn from $\mathcal{G}$ is connected with non-zero probability? As it turns out, the condition ``edges that do not share endpoints are independent'' needs to be clarified and the answer to the question above is sensitive to the specification. In fact, we formalize this intuitive description into a strict hierarchy of five independence conditions, which we show to have at least three different behaviors for the threshold $\rho(n)$. For each condition, we provide upper and lower bounds for $\rho(n)$. In the strongest condition, the coloring model (which includes, e.g., random geometric graphs), we show that $\rho(n)\rightarrow 2-\phi\approx 0.38$ for $n\rightarrow\infty$, proving a conjecture by Badakhshian, Falgas-Ravry, and Sharifzadeh. This separates the coloring models from the weaker independence conditions we consider, as there we prove that $\rho(n)>0.5-o(n)$. In stark contrast to the coloring model, for our weakest independence condition -- pairwise independence of non-adjacent edges -- we show that $\rho(n)$ lies within $O(1/n^2)$ of the threshold $1-2/n$ for completely arbitrary distributions., Comment: 35 pages, 6 figures. [v2] adds related work and is intended as a full version accompanying the version to appear at RANDOM'23

The burning number of a graph $G$ is the smallest positive integer $k$ such that the vertex set of $G$ can be covered with balls of radii $0, 1, \dots, k-1$. A well-known conjecture by Bonato, Janssen and Roshabin states that any connected graph on $n$ vertices has burning number at most $\lceil \sqrt{n} \rceil$. It was recently shown by Norin and Turcotte that the conjecture holds up to a factor of $1+o(1)$. In this note, we demonstrate how this result can be applied to determine the asymptotic value of the burning number for graph classes with given minimum degree. This is based on an observation about connected $2$-hop dominating sets, which may be of independent interest., Comment: 4 pages, no figures

Parallel optimizations for the 2D Hierarchical Poincar\'e-Steklov (HPS) discretization scheme are described. HPS is a multi-domain spectral collocation scheme that allows for combining very high order discretizations with direct solvers, making the discretization powerful in resolving highly oscillatory solutions to high accuracy. HPS can be viewed as a domain decomposition scheme where the domains are connected directly through the use of a sparse direct solver. This manuscript describes optimizations of HPS that are simple to implement, and that leverage batched linear algebra on modern hybrid architectures to improve the practical speed of the solver. In particular, the manuscript demonstrates that the traditionally high cost of performing local static condensation for discretizations involving very high local order $p$ can be reduced dramatically.

The paper describes a sparse direct solver for the linear systems that arise from the discretization of an elliptic PDE on a two dimensional domain. The solver is designed to reduce communication costs and perform well on GPUs; it uses a two-level framework, which is easier to implement and optimize than traditional multi-frontal schemes based on hierarchical nested dissection orderings. The scheme decomposes the domain into thin subdomains, or "slabs". Within each slab, a local factorization is executed that exploits the geometry of the local domain. A global factorization is then obtained through the LU factorization of a block-tridiagonal reduced coefficient matrix. The solver has complexity $O(N^{5/3})$ for the factorization step, and $O(N^{7/6})$ for each solve once the factorization is completed. The solver described is compatible with a range of different local discretizations, and numerical experiments demonstrate its performance for regular discretizations of rectangular and curved geometries. The technique becomes particularly efficient when combined with very high-order convergent multi-domain spectral collocation schemes. With this discretization, a Helmholtz problem on a domain of size $1000 \lambda \times 1000 \lambda$ (for which $N=100 \mbox{M}$) is solved in 15 minutes to 6 correct digits on a high-powered desktop with GPU acceleration.

Randomized subspace approximation with "matrix sketching" is an effective approach for constructing approximate partial singular value decompositions (SVDs) of large matrices. The performance of such techniques has been extensively analyzed, and very precise estimates on the distribution of the residual errors have been derived. However, our understanding of the accuracy of the computed singular vectors (measured in terms of the canonical angles between the spaces spanned by the exact and the computed singular vectors, respectively) remains relatively limited. In this work, we present practical bounds and estimates for canonical angles of randomized subspace approximation that can be computed efficiently either a priori or a posteriori, without assuming prior knowledge of the true singular subspaces. Under moderate oversampling in the randomized SVD, our prior probabilistic bounds are asymptotically tight and can be computed efficiently, while bringing a clear insight into the balance between oversampling and power iterations given a fixed budget on the number of matrix-vector multiplications. The numerical experiments demonstrate the empirical effectiveness of these canonical angle bounds and estimates on different matrices under various algorithmic choices for the randomized SVD., Comment: Accepted at SIAM Journal on Matrix Analysis and Applications

We introduce stronger notions for approximate single-source shortest-path distances, show how to efficiently compute them from weaker standard notions, and demonstrate the algorithmic power of these new notions and transformations. One application is the first work-efficient parallel algorithm for computing exact single-source shortest paths graphs -- resolving a major open problem in parallel computing. Given a source vertex in a directed graph with polynomially-bounded nonnegative integer lengths, the algorithm computes an exact shortest path tree in $m \log^{O(1)} n$ work and $n^{1/2+o(1)}$ depth. Previously, no parallel algorithm improving the trivial linear depths of Dijkstra's algorithm without significantly increasing the work was known, even for the case of undirected and unweighted graphs (i.e., for computing a BFS-tree). Our main result is a black-box transformation that uses $\log^{O(1)} n$ standard approximate distance computations to produce approximate distances which also satisfy the subtractive triangle inequality (up to a $(1+\varepsilon)$ factor) and even induce an exact shortest path tree in a graph with only slightly perturbed edge lengths. These strengthened approximations are algorithmically significantly more powerful and overcome well-known and often encountered barriers for using approximate distances. In directed graphs they can even be boosted to exact distances. This results in a black-box transformation of any (parallel or distributed) algorithm for approximate shortest paths in directed graphs into an algorithm computing exact distances at essentially no cost. Applying this to the recent breakthroughs of Fineman et al. for compute approximate SSSP-distances via approximate hopsets gives new parallel and distributed algorithm for exact shortest paths.

This paper describes a trapezoidal quadrature method for the discretization of singular and hypersingular boundary integral operators (BIOs) that arise in solving boundary value problems for elliptic partial differential equations. The quadrature is based on a uniform grid in parameter space coupled with the standard punctured Trapezoidal rule. A key observation is that the error incurred by the singularity in the kernel can be expressed exactly using generalized Euler-Maclaurin formulae that involve the Riemann zeta function in 2D and the Epstein zeta functions in 3D. These expansions are exploited to correct the errors via local stencils at the singular point using a novel systematic moment-fitting approach. This new method provides a unified treatment of all common BIOs (Laplace, Helmholtz, Stokes, etc.). We present numerical examples that show convergence of up to 32nd-order in 2D and 9th-order in 3D with respect to the mesh size., Comment: 26 pages, 7 figures, 1 table

Hadwiger's famous coloring conjecture states that every $t$-chromatic graph contains a $K_t$-minor. Holroyd [Bull. London Math. Soc. 29, (1997), pp. 139--144] conjectured the following strengthening of Hadwiger's conjecture: If $G$ is a $t$-chromatic graph and $S \subseteq V(G)$ takes all colors in every $t$-coloring of $G$, then $G$ contains a $K_t$-minor rooted at $S$. We prove this conjecture in the first open case of $t=4$. Notably, our result also directly implies a stronger version of Hadwiger's conjecture for $5$-chromatic graphs as follows: Every $5$-chromatic graph contains a $K_5$-minor with a singleton branch-set. In fact, in a $5$-vertex-critical graph we may specify the singleton branch-set to be any vertex of the graph., Comment: 10 pages, no figures