Sampling to provide or to bound: With applications to fully dynamic graph algorithms

Dynamic graph algorithms have been extensively studied in the last two decades due to their wide applicability in many contexts. Recently, several implementations and experimental studies have been conducted investigating the practical merits of fundamental techniques and algorithms. In most cases, these algorithms required sophisticated engineering and fine-tuning to be turned into efficient implementations. In this paper, we survey several implementations along with their experimental studies for dynamic problems on undirected and directed graphs. The former case includes dynamic connectivity, dynamic minimum spanning trees, and the sparsification technique. The latter case includes dynamic transitive closure and dynamic shortest paths. We also discuss the design and implementation of a software library for dynamic graph algorithms.

Many dynamic graph algorithms have an amortized update time, rather than a stronger worst-case guarantee. But amortized data structures are not suitable for real-time systems, where each individual operation has to be executed quickly. For this reason, there exist many recent randomized results that aim to provide a guarantee stronger than amortized expected. The strongest possible guarantee for a randomized algorithm is that it is always correct (Las Vegas) and has high-probability worst-case update time, which gives a bound on the time for each individual operation that holds with high probability. In this article, we present the first polylogarithmic high-probability worst-case time bounds for the dynamic spanner and the dynamic maximal matching problem. (1) For dynamic spanner, the only known o ( n ) worst-case bounds were O ( n 3/4 ) high-probability worst-case update time for maintaining a 3-spanner and O ( n 5/9 ) for maintaining a 5-spanner. We give a O (1) k log 3 ( n ) high-probability worst-case time bound for maintaining a ( 2k-1 )-spanner, which yields the first worst-case polylog update time for all constant k . (All the results above maintain the optimal tradeoff of stretch 2k-1 and Õ( n 1+1/k ) edges.) (2) For dynamic maximal matching, or dynamic 2-approximate maximum matching, no algorithm with o(n) worst-case time bound was known and we present an algorithm with O (log 5 ( n )) high-probability worst-case time; similar worst-case bounds existed only for maintaining a matching that was (2+ϵ)-approximate, and hence not maximal. Our results are achieved using a new approach for converting amortized guarantees to worst-case ones for randomized data structures by going through a third type of guarantee, which is a middle ground between the two above: An algorithm is said to have worst-case expected update time ɑ if for every update σ, the expected time to process σ is at most ɑ. Although stronger than amortized expected, the worst-case expected guarantee does not resolve the fundamental problem of amortization: A worst-case expected update time of O(1) still allows for the possibility that every 1/ f(n) updates requires ϴ ( f(n) ) time to process, for arbitrarily high f(n) . In this article, we present a black-box reduction that converts any data structure with worst-case expected update time into one with a high-probability worst-case update time: The query time remains the same, while the update time increases by a factor of O (log 2(n) ). Thus, we achieve our results in two steps: (1) First, we show how to convert existing dynamic graph algorithms with amortized expected polylogarithmic running times into algorithms with worst-case expected polylogarithmic running times. (2) Then, we use our black-box reduction to achieve the polylogarithmic high-probability worst-case time bound. All our algorithms are Las-Vegas-type algorithms.

A (fully) dynamic graph algorithm is a data structure that supports edge insertions, edge deletions, and answers certain queries that are specific to the problem under consideration. There has been a lot of research on dynamic algorithms for graph problems that are solvable in polynomial time by a static algorithm. However, while there is a large body of theoretical work on efficient dynamic graph algorithms, a lot of these algorithms were never implemented and empirically evaluated.In this work, we consider the fully dynamic edge orientation problem, also called fully dynamic Δ-orientation problem, which is to maintain an orientation of the edges of an undirected graph such that the out-degree is low. If edges are inserted or deleted, one may have to flip the orientation of some edges in order to avoid vertices having a large out-degree. While there has been theoretical work on dynamic versions of this problem, currently there is no experimental evaluation available. In this work, we close this gap and engineer a range of new dynamic edge orientation algorithms as well as algorithms from the current literature. The most successful newly proposed algorithms in this paper are based on repeatedly swapping edges with smallest degree neighbors and based on breadth-first search. Moreover, we evaluate these algorithms on real-world dynamic graphs. Generally, these algorithms outperform the algorithms currently proposed in the literature. The best algorithm considered in this paper in terms of quality, based on a simple breadth-first search, computes the optimum result more than 90% of the instances and is on average only 2.4% worse than the optimum solution.*The full version of the paper can be accessed at https://arxiv.org/abs/2301.06968

In 1999, Brodal and Fagerberg (BF) gave an algorithm for maintaining a low outdegree orientation of a dynamic uniformly sparse graph. Specifically, for a dynamic graph on n -vertices, with arboricity bounded by ɑ at all times, the BF algorithm supports edge updates in O (log n ) amortized update time, while keeping the maximum outdegree in the graph bounded by O ( ɑ ). Such an orientation provides a basic data structure for uniformly sparse graphs, which found applications to several dynamic graph algorithms, including adjacency queries and labeling schemes, maximal and approximate matching, approximate vertex cover, forest decomposition, and distance oracles. A significant weakness of the BF algorithm is the possible temporary blowup of the maximum outdegree, following edge insertions. Although BF eventually reduces all outdegrees to O ( ɑ ), some vertices may reach an outdegree of Ω( n ) during the process, and hence local memory usage at the vertices, which is an important quality measure in distributed systems, cannot be bounded. We show how to modify the BF algorithm to guarantee that the outdegrees of all vertices are bounded by O ( ɑ ) at all times, without hurting any of its other properties and present an efficient distributed implementation of the modified algorithm. This provides the first representation of distributed networks in which the local memory usage at all vertices is bounded by the arboricity (which is essentially the average degree of the densest subgraph) rather than the maximum degree. For settings where there is no strict limitation on the local memory, one may take the temporary outdegree blowup to the extreme and allow a permanent outdegree blowup. This allows us to address the second significant weakness of the BF algorithm—its inherently global nature: An insertion of an edge ( u,v ) may trigger changes in the orientations of edges that are arbitrarily far away from u and v . Such a non-local scheme may be prohibitively expensive in various practical applications. We suggest an alternative local scheme, which does not guarantee any outdegree bound on the vertices, yet is just as efficient as the BF scheme for some of the aforementioned applications. For example, we obtain a local dynamic algorithm for maintaining a maximal matching with sub-logarithmic update time in uniformly sparse networks, providing an exponential improvement over the state of the art in this context. We also present a distributed implementation of this scheme and some of its applications.

We combine the work of Garg and Konemann, and Fleischer with ideas from dynamic graph algorithms to obtain faster (1-ε)-approximation schemes for various versions of the multicommodity flow problem. In particular, if ε is moderately small and the size of every number used in the input instance is polynomially bounded, the running times of our algorithms match -- up to poly-logarithmic factors and some provably optimal terms -- the Ω(mn) flow-decomposition barrier for single-commodity flow.

The three-in-a-tree problem is to determine if a simple undirected graph contains an induced subgraph which is a tree connecting three given vertices. Based on a beautiful characterization that is proved in more than twenty pages, Chudnovsky and Seymour [Combinatorica 2010] gave the previously only known polynomial-time algorithm, running in $O(mn^2)$ time, to solve the three-in-a-tree problem on an $n$-vertex $m$-edge graph. Their three-in-a-tree algorithm has become a critical subroutine in several state-of-the-art graph recognition and detection algorithms. In this paper we solve the three-in-a-tree problem in $\tilde{O}(m)$ time, leading to improved algorithms for recognizing perfect graphs and detecting thetas, pyramids, beetles, and odd and even holes. Our result is based on a new and more constructive characterization than that of Chudnovsky and Seymour. Our new characterization is stronger than the original, and our proof implies a new simpler proof for the original characterization. The improved characterization gains the first factor $n$ in speed. The remaining improvement is based on dynamic graph algorithms.

Let H be a random graph on n vertices, grown by adding vertices one at a time, joining each new vertex to a uniformly chosen set of m earlier vertices. If edges of H are deleted independently, each being retained with probability p, then there is a “phase transition”. There is a certain critical value pc of p such that, with high probability, a component of order Θ(n) remains as n → ∞ if and only if p > pc. Among other results, we obtain the exact value of pc, which depends on m in a nontrivial way, and show that the phase transition has “infinite order”; in fact, for p = pc + ϵ, the largest component has order exp(−Θ(1/$\sqrt{\varepsilon}$))n with high probability. Analogous results were proved recently in by Bollobás, Janson, and Riordan [Random Structures Algorithms 26 (2005), 1–36] for a related model in which edges are present independently. The model we study is considerably more difficult to analyze, since the dependence between the edges is very important, affecting the value of pc, so many new complications arise. In overcoming these complications we make use of the techniques developed by the authors [Internet Math 1 (2003), 1–35] to analyze a very different model. © 2005 Wiley Periodicals, Inc. Random Struct. Alg., 2005

We study randomized gossip‐based processes in dynamic networks that are motivated by information discovery in large‐scale distributed networks such as peer‐to‐peer and social networks. A well‐studied problem in peer‐to‐peer networks isresource discovery, where the goal for nodes (hosts with IP addresses) is to discover the IP addresses of all other hosts. Also, some of the recent work on self‐stabilization algorithms for P2P/overlay networks proceed via discovery of the complete network. In social networks, nodes (people) discover new nodes through exchanging contacts with their neighbors (friends). In both cases the discovery of new nodes changes the underlying network — new edges are added to the network — and the process continues in the changed network. Rigorously analyzing such dynamic (stochastic) processes in a continuously changing topology remains a challenging problem with obvious applications.This paper studies and analyzes two natural gossip‐based discovery processes. In thepush discoveryortriangulationprocess, each node repeatedly chooses two random neighbors and connects them (i.e., “pushes” their mutual information to each other). In thepull discoveryprocess or thetwo‐hop walk, each node repeatedly requests or “pulls” a random contact from a random neighbor and connects itself to this two‐hop neighbor. Both processes are lightweight in the sense that the amortized work done per node is constant per round, local, and naturally robust due to the inherent randomized nature of gossip.Our main result is an almost‐tight analysis of the time taken for these two randomized processes to converge. We show that in any undirectedn‐node graph both processes takerounds to connect every node to all other nodes with high probability, whereasis a lower bound. We also study the two‐hop walk in directed graphs, and show that it takestime with high probability, and that the worst‐case bound is tight for arbitrary directed graphs, whereas Ω(n2) is a lower bound for strongly connected directed graphs. A key technical challenge that we overcome in our work is the analysis of a randomized process that itself results in a constantly changing network leading to complicated dependencies in every round. We discuss implications of our results and their analysis to discovery problems in P2P networks as well as to evolution in social networks. © 2016 Wiley Periodicals, Inc. Random Struct. Alg., 48, 565–587, 2016

We discuss a new algorithmic type of problem in random graphs studying the minimum number of queries one has to ask about adjacency between pairs of vertices of a random graph in order to find a subgraph which possesses some target property with high probability. In this paper we focus on finding long paths in when for some fixed constant . This random graph is known to have typically linearly long paths. To have edges with high probability in one clearly needs to query at least pairs of vertices. Can we find a path of length economically, i.e., by querying roughly that many pairs? We argue that this is not possible and one needs to query significantly more pairs. We prove that any randomised algorithm which finds a path of length with at least constant probability in with must query at least pairs of vertices. This is tight up to the factor. © 2016 Wiley Periodicals, Inc. Random Struct. Alg., 50, 71–85, 2017

The size of ordered binary decision diagrams (OBDDs) strongly depends on the chosen variable ordering. It is an obvious heuristic to use symmetric variable orderings, i.e., variable orderings where symmetric variables are arranged adjacently. In order to evaluate this heuristic, methods for estimating the OBDD size for random partially symmetric functions are presented. Characterizations of cases where, with high probability, only symmetric variable orderings and, with high probability, only nonsymmetric variable orderings lead to minimum OBDD size are obtained. For this analysis estimates for the number of different blocks of random Boolean matrices are used. © 1998 John Wiley & Sons, Inc. Random Struct. Alg., 13, 49–70, 1998

We study the largest component of a random (multi)graph on n vertices with a given degree sequence. We let n → ∞. Then, under some regularity conditions on the degree sequences, we give conditions on the asymptotic shape of the degree sequence that imply that with high probability all the components are small, and other conditions that imply that with high probability there is a giant component and the sizes of its vertex and edge sets satisfy a law of large numbers; under suitable assumptions these are the only two possibilities. In particular, we recover the results by Molloy and Reed on the size of the largest component in a random graph with a given degree sequence.We further obtain a new sharp result for the giant component just above the threshold, generalizing the case of G(n,p) with np = 1 + ω(n)n−1/3, where ω(n) → ∞ arbitrarily slowly.Our method is based on the properties of empirical distributions of independent random variables, and leads to simple proofs. © 2008 Wiley Periodicals, Inc. Random Struct. Alg., 2009

We study the space complexity of refuting unsatisfiable random k‐CNFs in the Resolution proof system. We prove that for Δ ≥ 1 and any ϵ > 0, with high probability a random k‐CNF over n variables and Δn clauses requires resolution clause space of Ω(n/Δ1+ϵ). For constant Δ, this gives us linear, optimal, lower bounds on the clause space. One consequence of this lower bound is the first lower bound for size of treelike resolution refutations of random 3‐CNFs with clause density Δ ≫ n. This bound is nearly tight. Specifically, we show that with high probability, a random 3‐CNF with Δn clauses requires treelike refutation size of exp(Ω(n/Δ1+ϵ)), for any ϵ > 0. Our space lower bound is the consequence of three main contributions: (1) We introduce a 2‐player Matching Game on bipartite graphs G to prove that there are no perfect matchings in G. (2) We reduce lower bounds for the clause space of a formula F in Resolution to lower bounds for the complexity of the game played on the bipartite graph G(F) associated with F. (3) We prove that the complexity of the game is large whenever G is an expander graph. Finally, a simple probabilistic analysis shows that for a random formula F, with high probability G(F) is an expander. We also extend our result to the case of G‐PHP, a generalization of the Pigeonhole principle based on bipartite graphs G. © 2003 Wiley Periodicals, Inc. Random Struct. Alg., 23: 92–109, 2003

The contributions of this paper are both of theoretical and of experimental nature. From the experimental point of view, we conduct an empirical study on some dynamic connectivity algorithms which where developed recently. In particular, the following implementations were tested and compared with simple algorithms: simple sparsification by Eppstein et al. and the recent randomized algorithm by Henzinger and King. In our experiments, we considered both random and non-random inputs. Moreover, we present a simplified variant of the algorithm by Henzinger and King, which for random inputs was always faster than the original implementation. For non-random inputs, simple sparsification was the fastest algorithm for small sequences of updates; for medium and large sequences of updates, the original algorithm by Henzinger and King was faster.From the theoretical point of view, we analyze the average case running time of simple sparsification and prove that for dynamic random graphs its logarithmic overhead vanishes.

We study differentially private algorithms for analyzing graph databases in the challenging setting of continual release with fully dynamic updates, where edges are inserted and deleted over time, and the algorithm is required to update the solution at every time step. Previous work has presented differentially private algorithms for many graph problems that can handle insertions only or deletions only (called partially dynamic algorithms) and obtained some hardness results for the fully dynamic setting. The only algorithms in the latter setting were for the edge count, given by Fichtenberger, Henzinger, and Ost (ESA '21), and for releasing the values of all graph cuts, given by Fichtenberger, Henzinger, and Upadhyay (ICML '23). We provide the first differentially private and fully dynamic graph algorithms for several other fundamental graph statistics (including the triangle count, the number of connected components, the size of the maximum matching, and the degree histogram), analyze their error, and show strong lower bounds on the error for all algorithms in this setting. Previously, only lower bounds for purely differentially private algorithms were known; our lower bounds give an exponential improvement in terms of the dependence on the number of time steps, while applying to algorithms satisfying pure as well as approximate differential privacy. We study two variants of edge differential privacy for fully dynamic graph algorithms: event-level and item-level. Under the former notion, two graph database update sequences are considered neighboring if, roughly speaking, they differ in at most one update; under the latter notion, they can differ only in updates pertaining to one edge. Differential privacy requires that for any two neighboring inputs, the output distributions of the algorithm are close. We give upper and lower bounds on the error of both---event-level and item-level---fully dynamic algorithms for several fundamental graph problems. No fully dynamic algorithms that are private at the item-level (the more stringent of the two notions) were known before. In the case of item-level privacy, for several problems, our algorithms match our lower bounds.

As graphs continue to grow in size, we seek ways to effectively process such data at scale. The model of streaming graph processing, in which a compact summary is maintained as each edge insertion/deletion is observed, is an attractive one. However, few results are known for optimization problems over such dynamic graph streams.In this paper, we introduce a new approach to handling graph streams, by instead seeking solutions for the parameterized versions of these problems. Here, we are given a parameter k and the objective is to decide whether there is a solution bounded by k. By combining kernelization techniques with randomized sketch structures, we obtain the first streaming algorithms for the parameterized versions of Maximal Matching and Vertex Cover. We consider various models for a graph stream on n nodes: the insertion-only model where the edges can only be added, and the dynamic model where edges can be both inserted and deleted. More formally, we show the following results:•In the insertion only model, there is a one-pass deterministic algorithm for the parameterized Vertex Cover problem which computes a sketch using Õ(k2) space1 such that at each timestamp in time Õ(2k) it can either extract a solution of size at most k for the current instance, or report that no such solution exists. We also show a tight lower bound of Ω(k2) for the space complexity of any (randomized) streaming algorithms for the parameterized Vertex Cover, even in the insertion-only model.•In the dynamic model, and under the promise that at each timestamp there is a maximal matching of size at most k, there is a one-pass Õ(k2)-space (sketch-based) dynamic algorithm that maintains a maximal matching with worst-case update time Õ(k2). This algorithm partially solves Open Problem 64 from [1]. An application of this dynamic matching algorithm is a one-pass Õ(k2)-space streaming algorithm for the parameterized Vertex Cover problem that in time Õ(2k) extracts a solution for the final instance with probability 1 – δ/no(1), where δ < 1. To the best of our knowledge, this is the first graph streaming algorithm that combines linear sketching with sequential operations that depend on the graph at the current time.•In the dynamic model without any promise, there is a one-pass randomized algorithm for the parameterized Vertex Cover problem which computes a sketch using Õ(nk) space such that in time Õ(nk + 2k) it can either extract a solution of size at most k for the final instance, or report that no such solution exists.