Publikationen

A central task in knowledge compilation is to compile a CNF-SAT instance into a succinct representation format that allows efficient operations such as testing satisfiability, counting, or enumerating all solutions. Useful representation formats studied in this area range from ordered binary decision diagrams (OBDDs) to circuits in decomposable negation normal form (DNNFs). While it is known that there exist CNF formulas that require exponential size representations, the situation is less well studied for other types of constraints than Boolean disjunctive clauses. The constraint satisfaction problem (CSP) is a powerful framework that generalizes CNF-SAT by allowing arbitrary sets of constraints over any finite domain. The main goal of our work is to understand for which type of constraints (also called the constraint language) it is possible to efficiently compute representations of polynomial size. We answer this question completely and prove two tight characterizations of efficiently compilable constraint languages, depending on whether target format is structured. We first identify the combinatorial property of "strong blockwise decomposability" and show that if a constraint language has this property, we can compute DNNF representations of linear size. For all other constraint languages we construct families of CSP-instances that provably require DNNFs of exponential size. For a subclass of "strong uniformly blockwise decomposable" constraint languages we obtain a similar dichotomy for structured DNNFs. In fact, strong (uniform) blockwise decomposability even allows efficient compilation into multi-valued analogs of OBDDs and FBDDs, respectively. Thus, we get complete characterizations for all knowledge compilation classes between O(B)DDs and DNNFs.

The task of computing homomorphisms between two finite relational structures A and B is a well-studied question with numerous applications. Since the set Hom(A, B) of all homomorphisms may be very large having a method of representing it in a succinct way, especially one which enables us to perform efficient enumeration and counting, could be extremely useful. One simple yet powerful way of doing so is to decompose Hom(A, B) using union and Cartesian product. Such data structures, called d-representations, have been introduced by Olteanu and Závodný [Olteanu and Závodný, 2015] in the context of database theory. Their results also imply that if the treewidth of the left-hand side structure A is bounded, then a d-representation of polynomial size can be found in polynomial time. We show that for structures of bounded arity this is optimal: if the treewidth is unbounded then there are instances where the size of any d-representation is superpolynomial. Along the way we develop tools for proving lower bounds on the size of d-representations, in particular we define a notion of reduction suitable for this context and prove an almost tight lower bound on the size of d-representations of all k-cliques in a graph.

As data analytics becomes more crucial to digital systems, so grows the importance of characterizing the database queries that admit a more efficient evaluation. We consider the tractability yardstick of answer enumeration with a polylogarithmic delay after a linear-time preprocessing phase. Such an evaluation is obtained by constructing, in the preprocessing phase, a data structure that supports polylogarithmic-delay enumeration. In this article, we seek a structure that supports the more demanding task of a “random permutation”: polylogarithmic-delay enumeration in truly random order. Enumeration of this kind is required if downstream applications assume that the intermediate results are representative of the whole result set in a statistically meaningful manner. An even more demanding task is that of “random access”: polylogarithmic-time retrieval of an answer whose position is given. We establish that the free-connex acyclic CQs are tractable in all three senses: enumeration, random-order enumeration, and random access; and in the absence of self-joins, it follows from past results that every other CQ is intractable by each of the three (under some fine-grained complexity assumptions). However, the three yardsticks are separated in the case of a union of CQs (UCQ): while a union of free-connex acyclic CQs has a tractable enumeration, it may (provably) admit no random access. We identify a fragment of such UCQs where we can guarantee random access with polylogarithmic access time (and linear-time preprocessing) and a more general fragment where we can guarantee tractable random permutation. For general unions of free-connex acyclic CQs, we devise two algorithms with relaxed guarantees: one has logarithmic delay in expectation, and the other provides a permutation that is almost uniformly distributed. Finally, we present an implementation and an empirical study that show a considerable practical superiority of our random-order enumeration approach over state-of-the-art alternatives.

As data analytics becomes more crucial to digital systems, so grows the importance of characterizing the database queries that admit a more efficient evaluation. We consider the tractability yardstick of answer enumeration with a polylogarithmic delay after a linear-time preprocessing phase. Such an evaluation is obtained by constructing, in the preprocessing phase, a data structure that supports polylogarithmic-delay enumeration. In this paper, we seek a structure that supports the more demanding task of a "random permutation": polylogarithmic-delay enumeration in truly random order. Enumeration of this kind is required if downstream applications assume that the intermediate results are representative of the whole result set in a statistically valuable manner. An even more demanding task is that of a "random access": polylogarithmic-time retrieval of an answer whose position is given. We establish that the free-connex acyclic CQs are tractable in all three senses: enumeration, random-order enumeration, and random access; and in the absence of self-joins, it follows from past results that every other CQ is intractable by each of the three (under some fine-grained complexity assumptions). However, the three yardsticks are separated in the case of a union of CQs (UCQ): while a union of free-connex acyclic CQs has a tractable enumeration, it may (provably) admit no random access. For such UCQs we devise a random-order enumeration whose delay is logarithmic in expectation. We also identify a subclass of UCQs for which we can provide random access with polylogarithmic access time. Finally, we present an implementation and an empirical study that show a considerable practical superiority of our random-order enumeration approach over state-of-the-art alternatives.

We show that there are CNF formulas which can be refuted in resolution in both small space and small width, but for which any small-width proof must have space exceeding by far the linear worst-case upper bound. This significantly strengthens the space-width trade-offs in [E. Ben-Sasson, SIAM J. Comput., 38 (2009), pp. 2511-2525], and provides one more example of trade-offs in the "supercritical" regime above worst case recently identified by [A.A. Razborov, J. ACM, 63 (2016), 16]. We obtain our results by using Razborov's new hardness condensation technique and combining it with the space lower bounds in [E. Ben-Sasson and J. Nordström, Short proofs may be spacious: An optimal separation of space and length in resolution, in Proceedings of the 49th Annual IEEE Symposium on Foundations of Computer Science (FOCS '08), 2008, pp. 709-718].

Marx (STOC 2010, J. ACM 2013) introduced the notion of submodular width of a conjunctive query (CQ) and showed that for any class Phi of Boolean CQs of bounded submodular width, the model-checking problem for Phi on the class of all finite structures is fixed-parameter tractable (FPT). Note that for non-Boolean queries, the size of the query result may be far too large to be computed entirely within FPT time. We investigate the free-connex variant of submodular width and generalise Marx's result to non-Boolean queries as follows: For every class Phi of CQs of bounded free-connex submodular width, within FPT-preprocessing time we can build a data structure that allows to enumerate, without repetition and with constant delay, all tuples of the query result. Our proof builds upon Marx's splitting routine to decompose the query result into a union of results; but we have to tackle the additional technical difficulty to ensure that these can be enumerated efficiently.

We investigate the query evaluation problem for fixed queries over fully dynamic databases, where tuples can be inserted or deleted. The task is to design a dynamic algorithm that immediately reports the new result of a fixed query after every database update. We consider queries in first-order logic (FO) and its extension with modulo-counting quantifiers (FO+MOD) and show that they can be efficiently evaluated under updates, provided that the dynamic database does not exceed a certain degree bound. In particular, we construct a data structure that allows us to answer a Boolean FO+MOD query and to compute the size of the result of a non-Boolean query within constant time after every database update. Furthermore, after every database update, we can update the data structure in constant time such that afterwards we are able to test within constant time for a given tuple whether or not it belongs to the query result, to enumerate all tuples in the new query result, and to enumerate the difference between the old and the new query result with constant delay between the output tuples. The preprocessing time needed to build the data structure is linear in the size of the database. Our results extend earlier work on the evaluation of first-order queries on static databases of bounded degree and rely on an effective Hanf normal form for FO+MOD recently obtained by Heimberg, Kuske, and Schweikardt (LICS 2016).

Establishing arc consistency on two relational structures is one of the most popular heuristics for the constraint satisfaction problem. We aim at determining the time complexity of arc consistency testing. The input structures G and H can be supposed to be connected colored graphs, as the general problem reduces to this particular case. We first observe the upper bound O(e(G)v(H)+v(G)e(H)), which implies the bound O(e(G)e(H)) in terms of the number of edges and the bound O((v(G)+v(H))3) in terms of the number of vertices. We then show that both bounds are tight as long as an arc consistency algorithm is based on constraint propagation (as all current algorithms are). Our lower bounds are based on examples of slow constraint propagation. We measure the speed of constraint propagation observed on a pair G,H by the size of a combinatorial proof that Spoiler wins the existential 2-pebble game on G,H.

We investigate the query evaluation problem for fixed queries over fully dynamic databases, where tuples can be inserted or deleted. The task is to design a dynamic algorithm that immediately reports the new result of a fixed query after every database update. We consider queries in first-order logic (FO) and its extension with modulo-counting quantifiers (FO+MOD), and show that they can be efficiently evaluated under updates, provided that the dynamic database does not exceed a certain degree bound. In particular, we construct a data structure that allows to answer a Boolean FO+MOD query and to compute the size of the query result within constant time after every database update. Furthermore, after every update we are able to immediately enumerate the new query result with constant delay between the output tuples. The time needed to build the data structure is linear in the size of the database. Our results extend earlier work on the evaluation of first-order queries on static databases of bounded degree and rely on an effective Hanf normal form for FO+MOD recently obtained by [Heimberg, Kuske, and Schweikardt, LICS, 2016].

In recent years, we have seen several approaches to the graph isomorphism problem based on “generic” mathematical programming or algebraic (Grobner basis) techniques. For most of these, lower bounds have been established. In fact, it has been shown that the pairs of non-isomorphic CFI-graphs (introduced by Cai, FUrer, and Immerman in 1992 as hard examples for the combinatorial Weisfeiler-Leman algorithm) cannot be distinguished by these mathematical algorithms. A notable exception were the algebraic algorithms over the field 2, for which no lower bound was known. Another, in some way even stronger, approach to graph isomorphism testing is based on solving systems of linear Diophantine equations (that is, linear equations over the integers), which is known to be possible in polynomial time. So far, no lower bounds for this approach were known. Lower bounds for the algebraic algorithms can best be proved in the framework of proof complexity, where they can be phrased as lower bounds for algebraic proof systems such as Nullstellensatz or the (more powerful) polynomial calculus. We give new hard examples for these systems: families of pairs of non-isomorphic graphs that are hard to distinguish by polynomial calculus proofs simultaneously over all prime fields, including 2, as well as examples that are hard to distinguish by the systems-of-linear-Diophantine- equations approach. In a previous paper, we observed that the CFI-graphs are closely related to what we call “group CSPs”: constraint satisfaction problems where the constraints are membership tests in some coset of a subgroup of a cartesian power of a base group (Z2 in the case of the classical CFI-graphs). Our new examples are also based on group CSPs (for Abelian groups), but here we extend the CSPs by a few non-group constraints to obtain even harder instances for graph isomorphism.

We show that there are CNF formulas which can be refuted in resolution in both small space and small width, but for which any small-width resolution proof must have space exceeding by far the linear worst-case upper bound. This significantly strengthens the space-width trade-offs in [Ben-Sasson 2009], and provides one more example of trade-offs in the "supercritical" regime above worst case recently identified by [Razborov 2016]. We obtain our results by using Razborov’s new hardness condensation technique and combining it with the space lower bounds in [Ben-Sasson and Nordström 2008].

We investigate the power of graph isomorphism algorithms based on algebraic reasoning techniques like Gröbner basis computation. The idea of these algorithms is to encode two graphs into a system of equations that are satisfiable if and only if if the graphs are isomorphic, and then to (try to) decide satisfiability of the system using, for example, the Gröbner basis algorithm. In some cases this can be done in polynomial time, in particular, if the equations admit a bounded degree refutation in an algebraic proof systems such as Nullstellensatz or polynomial calculus. We prove linear lower bounds on the polynomial calculus degree over all fields of characteristic ≠2 and also linear lower bounds for the degree of Positivstellensatz calculus derivations. We compare this approach to recently studied linear and semidefinite programming approaches to isomorphism testing, which are known to be related to the combinatorial Weisfeiler-Lehman algorithm. We exactly characterise the power of the Weisfeiler-Lehman algorithm in terms of an algebraic proof system that lies between degree-k Nullstellensatz and degree-k polynomial calculus.

Zsfassung in dt. u. engl. Sprache

We establish optimal bounds on the number of nested propagation steps in k-consistency tests. It is known that local consistency algorithms such as arc-, path- and k-consistency are not efficiently parallelizable. Their inherent sequential nature is caused by long chains of nested propagation steps, which cannot be executed in parallel. This motivates the question “What is the minimum number of nested propagation steps that have to be performed by k-consistency algorithms on (binary) constraint networks with n variables and domain size d?” It was known before that 2-consistency requires Ω(nd) and 3-consistency requires Ω(n 2) sequential propagation steps. We answer the question exhaustively for every k ≥ 2: there are binary constraint networks where any k-consistency procedure has to perform Ω(n k − 1 d k − 1) nested propagation steps before local inconsistencies were detected. This bound is tight, because the overall number of propagation steps performed by k-consistency is at most n k − 1 d k − 1.

An assignment of colours to the vertices of a graph is stable if any two vertices of the same colour have identically coloured neighbourhoods. The goal of colour refinement is to find a stable colouring that uses a minimum number of colours. This is a widely used subroutine for graph isomorphism testing algorithms, since any automorphism needs to be colour preserving. We give an O((m+n)log n) algorithm for finding a canonical version of such a stable colouring, on graphs with n vertices and m edges. We show that no faster algorithm is possible, under some modest assumptions about the type of algorithm, which captures all known colour refinement algorithms.

The existential k-pebble game characterizes the expressive power of the existential-positive k-variable fragment of first-order logic on finite structures. The winner of the existential k-pebble game on two given finite structures can be determined in time O(n2k) by dynamic programming on the graph of game configurations. We show that there is no O(n(k-3)/12)-time algorithm that decides which player can win the existential k-pebble game on two given structures. This lower bound is unconditional and does not rely on any complexity-theoretic assumptions. Establishing strong k-consistency is a well-known heuristic for solving the constraint satisfaction problem (CSP). By the game characterization of Kolaitis and Vardi our result implies that there is no O(n(k-3)/12)-time algorithm that decides if strong k-consistency can be established for a given CSP-instance.

We study the complexity of the following "resolution width problem": Does a given 3-CNF formula have a resolution refutation of width k? For fixed k, refutations of width k can easily be found in polynomial time. We prove a matching polynomial lower bound for the resolution width problem that shows that there is no significant faster way to decide the existence of a width-k refutation than exhaustively searching for it. This lower bound is unconditional and does not rely on any unproven complexity theoretic assumptions. We also prove that the resolution width problem is EXPTIME-complete (if k is part of the input). This confirms a conjecture by Vardi, who has first raised the question for the complexity of the resolution width problem. Furthermore, we prove that the variant of the resolution width problem for regular resolution is PSPACE-complete, confirming a conjecture by Urquhart.