Solution Counting for CSP and SAT with Large Tree-Width

Рассмотрена проблема подсчета количества решений задачи совместимости ограничений (Constraint Satisfaction Problem). Для ее решения был адаптирован метод обратного прослеживания с ацикличным представлением графа ограничений (Backtracking with Tree-Decomposition). Предложен точный алгоритм, сложность...

Full description

Saved in:
Bibliographic Details
Published in:Управляющие системы и машины
Date:2011
Main Authors: Favier, A., de Givry, S., Jégou, P.
Format: Article
Language:English
Published: Міжнародний науково-навчальний центр інформаційних технологій і систем НАН та МОН України 2011
Subjects:
Online Access:https://nasplib.isofts.kiev.ua/handle/123456789/82919
Tags: Add Tag
No Tags, Be the first to tag this record!
Journal Title:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Cite this:Solution Counting for CSP and SAT with Large Tree-Width / A. Favier, S. de Givry, Ph. Jégou // Управляющие системы и машины. — 2011. — № 2. — С. 4-13, 70. — Бібліогр.: 36 назв. — англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
_version_ 1859633660017246208
author Favier, A.
de Givry, S.
Jégou, P.
author_facet Favier, A.
de Givry, S.
Jégou, P.
citation_txt Solution Counting for CSP and SAT with Large Tree-Width / A. Favier, S. de Givry, Ph. Jégou // Управляющие системы и машины. — 2011. — № 2. — С. 4-13, 70. — Бібліогр.: 36 назв. — англ.
collection DSpace DC
container_title Управляющие системы и машины
description Рассмотрена проблема подсчета количества решений задачи совместимости ограничений (Constraint Satisfaction Problem). Для ее решения был адаптирован метод обратного прослеживания с ацикличным представлением графа ограничений (Backtracking with Tree-Decomposition). Предложен точный алгоритм, сложность которого экспоненциально зависит от ширины дерева, и приближенный алгоритм, экспоненциально зависящий от размера максимальной клики. The problem of counting the number of solutions of a CSP is considered. For solving the problem the Backtracking with a Tree-Decomposition method was adapted. The exact algorithm is suggested which has the worst-time complexity exponential in a tree width, as well as iterative algorithm that has computational complexity exponential in a maximum clique size. Розглянуто проблему підрахунку кількості розв’язків задачі сумісності обмежень (Constraint Satisfaction Problem). Для її розв’язку було адаптовано метод зворотного простеження з ациклічним поданням графа обмежень (Backtracking with Tree-Decomposition). Запропоновано точний алгоритм, складність якого експоненційно залежить від ширини дерева, і наближений алгоритм, експоненційно залежний від розміру максимальної кліки.
first_indexed 2025-12-07T13:13:54Z
format Article
fulltext 4 УСиМ, 2011, № 2 UDC 519.157 A. Favier, S. de Givry, Ph. Jégou Solution Counting for CSP and SAT with Large Tree-Width Рассмотрена проблема подсчета количества решений задачи совместимости ограничений (Constraint Satisfaction Problem). Для ее решения был адаптирован метод обратного прослеживания с ацикличным представлением графа ограничений (Backtracking with Tree-Decomposition). Предложен точный алгоритм, сложность которого экспоненциально зависит от ширины дерева, и приближенный алгоритм, экспоненциально зависящий от размера максимальной клики. The problem of counting the number of solutions of a CSP is considered. For solving the problem the Backtracking with a Tree- Decomposition method was adapted. The exact algorithm is suggested which has the worst-time complexity exponential in a tree width, as well as iterative algorithm that has computational complexity exponential in a maximum clique size. Розглянуто проблему підрахунку кількості розв’язків задачі сумісності обмежень (Constraint Satisfaction Problem). Для її розв’язку було адаптовано метод зворотного простеження з ациклічним поданням графа обмежень (Backtracking with Tree- Decomposition). Запропоновано точний алгоритм, складність якого експоненційно залежить від ширини дерева, і наближений алгоритм, експоненційно залежний від розміру максимальної кліки. Abstract This paper deals with the challenging problem of counting the number of solutions of a CSP, de- noted #CSP. Recent progress has been made using search methods, such as Backtracking with Tree- Decomposition (BTD) [Jégou and Terrioux, 2003], which exploit the constraint graph structure in or- der to solve CSPs. We propose to adapt BTD for solving the #CSP problem. The resulting exact counting method has a worst-case time complex- ity exponential in a specific graph parameter, called tree-width. For problems with a sparse con- straint graph but a large tree-width, we propose an iterative method which approximates the number of solutions by solving a partition of the set of constraints into a collection of partial chordal sub- graphs. Its time complexity is exponential in the maximum clique size – the clique number – of the original problem, which can be much smaller than its tree-width. Experiments on CSP and SAT benchmarks show the practical efficiency of our proposed approaches1. 1. Introduction The Constraint Satisfaction Problem (CSP) for- malism offers a powerful framework for repre- senting and solving efficiently many problems. Finding a solution is NP-complete. A more diffi- cult problem consists in counting the number of Keywords: solution counting, constraint satisfaction prob- lem, dynamic programming, approximate counting, chordal graph, graph coloring. 1A preliminary version appears in [Favier et al., 2009]. solutions. This problem, denoted #CSP, is known to be #P-complete [Valiant, 1979]. This problem has numerous applications in computer science, particularly in AI, e.g. in approximate reasoning [Roth, 1996], in belief revision [Darwiche, 2001], in diagnosis [Kumar, 2002], in guiding backtrack search heuristics to find solutions to CSPs [Zanarini and Pesant, 2009], and in other domains outside computer science such as in statistical physics [Burton and Steif, 1994] or in computa- tional biology for protein structure prediction [Mann et al., 2007]. In the literature, two principal classes of ap- proaches have been proposed. In the first class, methods find exactly the number of solutions in exponential time. In the second class, methods give approximations in a reasonable time. For the first class, a natural approach consists in extend- ing classical search algorithms such as FC or MAC in order to enumerate all solutions. But the more solutions there are, the longer it takes to enumerate them. Here, we are interested in search methods that exploit the problem structure, providing time and space complexity bounds. This is the case for the d-DNNF compiler c2d [Darwiche, 2004] and AND/OR graph search [Dechter and Mateescu, 2004, 2007] for counting. We propose to adapt Backtracking with Tree-Decomposition (BTD) [Jégou and Terrioux, 2003] to #CSP. This method was initially proposed for solving structured CSPs. Our modifications to BTD, resulting in an УСиМ, 2011, № 2 5 algorithm called #BTD, are similar to what has been done in the AND/OR context [Dechter and Mateescu, 2004, 2007], except that #BTD is based on a cluster tree-decomposition instead of a pseudo-tree, which naturally enables #BTD to ex- ploit dynamic variable orderings inside clusters whereas AND/OR search uses a static ordering. Most of the recent work on counting has been realized on a specific case of #CSP called #SAT, the model counting problem associated with SAT [Valiant, 1979]. Exact methods for #SAT extend systematic SAT solvers, adding component analy- sis [Bayardo and Pehoushek, 2000] (Relsat solver) and caching [Sang et al., 2004] (Cachet, further improved by sharpSAT [Thurley, 2006] ) for effi- ciency purposes. Approaches using approximations estimate the number of solutions. They propose poly-time or exponential time algorithms which should offer reasonably good approximations of the number of solutions, with theoretical guarantees about the quality of the approximation, or not. Most of the work has been done by sampling either the origi- nal OR search space [Wei and Selman, 2005, Gomes et al., 2007a, Gogate and Dechter, 2007, Kroc et al., 2008], or the original AND/OR search space [Gogate and Dechter, 2008]. All these methods except that in [Gogate and Dechter, 2008] provide a lower bound on the number of solutions with a high-confidence interval obtained by randomly assigning variables until solutions are found. A possible drawback of these ap- proaches is that they might find no solution within a given time limit due to inconsistent partial as- signments. For large and complex problems, this results in zero lower bounds or it requires time- consuming parameter (e.g. sample size) tuning in order to avoid this problem. Another solution is to rely on a complete search method, withdrawing any time limit, in order to check whether every variable assignment made during the sampling process is globally consistent or not and then backtrack as done in [Gogate and Dechter, 2007]. Another approach involves reducing the search space by adding streamlining XOR constraints [Gomes et al., 2006, 2007b]. However, it does not guarantee that the resulting problem is easier to solve. A good overview of state-of-the-art exact and approximate counting methods for #SAT is given in [Gomes et al., 2009]. In this paper, we propose to relax the problem, by partitioning the set of constraints into a collec- tion of structured chordal subproblems. Each sub- problem is then solved using #BTD. Finally, an approximate number of solutions on the whole pro- blem is obtained by combining the results of each subproblem. The resulting approximate method is called Approx#BTD. The task of counting the number of solutions of each subproblem should be relatively easy if the original instance has a sparse graph. In fact, it depends on the tree-width of the subproblems, which is bounded by the maximum clique size of the original instance called the clique number. In the case of a sparse graph, we expect this number to be small. This also forbids using our approach for CSPs with global constraints (i.e. having a complete constraint graph) or proposi- tional CNF formulae with very large clauses. Approx#BTD gives also a trivial upper bound on the number of solutions. Other relaxation-based counting methods have been tried in the literature such as mini-bucket elimination and iterative join-graph propagation [Kask et al., 2004], or in the related context of Bayesian inference, iterative belief propagation and the edge deletion framework [Choi and Dar- wiche, 2006]2. These approaches do not exploit the local structure of the instances as it is done by search methods such as #BTD , thanks to local consistency and dynamic variable ordering. In the next section, we introduce notation and the notion of a tree-decomposition. Section 3 de- scribes #BTD for exact counting and Section 4 presents Approx#BTD for approximate counting. Experimental results are given in Section 5, then we conclude. 2. Preliminaries A Constraint Satisfaction Problem [Montanari, 1974] is a quadruplet = ( , , , )X D C R . X is a set 2It starts by solving an initial polytree-structured sub- problem, further augmented by progressively recovering some edges, until the whole problem is solved. Approx#BTD starts directly with a possibly larger chordal subproblem. 6 УСиМ, 2011, № 2 of n variables with finite domains D. The domain of variable xi  X with [1, ]i n is denoted xi d D . The maximum domain size is [1, ]= | |max i n xi d d . C is a set of m constraints. Each constraint c C is a set 1 { , , }c ck x x X of variables. The prob- lem is called a binary CSP if all the constraints have 2k  . A relation cr R is associated with each constraint c such that cr represents the set of allowed tuples over 1c ck x xd d  for the assignment of the variables in c. Note that we can also define constraints by using functions or predicates for instance. An assignment of 1= { , , }kY x x X is a tuple 1= ( , , )kv v  from 1x xk d d  . We note the assignment 1( , , )kv v in the more meaningful form 1 1( , , )k kx v x v  . The projection of a tup- le  over a subset of variables c Y is denoted [ ]c . A constraint c is said satisfied by  if c Y and [ ] cc r , violated otherwise.  is said con- sistent with respect to a given subproblem if it sat- isfies all its constraints. A solution is an assign- ment of all the variables satisfying all the con- straints. The structure of a CSP can be represented by the graph (X, C), called the constraint graph, whose vertices are the variables of X and with an edge between two vertices if the corresponding variables share a constraint in C. A graph is chor- dal if every cycle of length at least four has a chord, i.e. an edge joining two non-consecutive vertices along the cycle. A tree-decomposition [Robertson and Seymour, 1986] of a CSP  is a pair ( , )  with = ( , )I F a tree with vertices I and edges F and = { : }i i I  a family of subsets of X, such that each cluster i is a node of  and satisfies: (1) =i I i X  , (2) for each constraint c C , there exists i I with ic   , (3) for all , ,i j k I , if k is on a path from i to j in , then i j k    . The width of a tree-decomposition ( , )  is equal to | | 1max i I i  . The tree-width of  is the mini- mum width over all its tree-decompositions. Find- ing an optimal tree-decomposition is NP-Hard [Arnborg et al., 1987]. A tree-decomposition can be found by triangu- lation of (i.e. adding edges to) the constraint graph such that it becomes chordal and then by search- ing the maximal cliques of the triangulated con- straint graph (resulting in the clusters ) and fi- nally by selecting a maximum spanning tree  on the cluster graph with edges between i and j if i j   and edge weights equal to | |i j  . In the experiments, we used the Min-Fill greedy heuristic (it locally adds the minimum number of edges to the constraint graph), a very usual heuris- tic aimed at the production of tree-decompositions with a small tree-width [Rose, 1970]. In the following, from a tree-decomposition, we consider a rooted tree ( , )I F with root 1 and we note ( )iSons  the set of son clusters of i and ( )iDesc  the set of variables which belong to i or to any descendant j of i in the subtree rooted in i . Example 1. Consider the CSP the constraint graph of which is provided in Fig. 1, a. We assume that each domain is { , , , }a b c d and each constraint = { , }ij i jc x x has a relation ijcr such that =i jx x , which defines a graph coloring problem. A – A constraint graph b – Its tree-decomposition Fig. 1. A tree-decomposition of a small problem with 8 variables Fig. 1, b represents an optimal tree-decomposi- tion for the chordal graph of Fig. 1, a. We have 1 1 2 3= { , , }x x x , 2 2 3 4 5={ , , , }x x x x , 3 = 4 5 6{ , , }x x x , and 4 3 7 8={ , , }x x x . For instance, Desc(2)=22= = 2 3 4 5 6{ , , , , }x x x x x . The tree-width is 3. УСиМ, 2011, № 2 7 3. Exact solution counting with #BTD The essential property of a tree decomposition is that assigning ij ( ( )j iSons  ) by the as- signment  separates the initial problem into two subproblems, which can then be solved independ- ently and the product of their number of solutions returned as the total number of solutions. The first subproblem, denoted / [ ]j i j   and rooted in j, is defined by the variables in Desc(j), with variables ij assigned by , and by all the constraints involving at least one variable in Desc(j)\ i. The remaining constraints, together with the variables they involve, define the remaining subproblem. A tree search backtracking algorithm can ex- ploit this property by using a suitable variable as- signment ordering: the variables of any cluster i must be assigned before the variables that remain in its son clusters. In this case, for any cluster ( )j iSons  , once ij is assigned, the sub- problem rooted in j conditioned by the current assignment  of ij can be solved independ- ently of the rest of the problem. The exact number of solutions nb of this subproblem / [ ]j i j   , cal- led a #good and represented by a pair ( [ ],i j   nb), can be recorded, which means it will never be computed again for the same assignment of ij. This is why algorithms such as BTD [Jégou and Terrioux, 2003] and AND/OR graph search [Dechter and Mateescu, 2004, 2007], exploiting the related notions of structural goods and pseudo- tree [Freuder and Quinn, 1985] respectively, are able to keep their time (and space) complexity ex- ponential in the size of the largest cluster only. We denote /j  the number of solutions of sub- problem / [ ]j i j   compatible with an assignment  of (ij)  Y, Y  j. It corresponds to the number of extensions of  on Desc(j) satisfying all the constraints in / [ ]j i j   . The total number of solutions of  is S1/. Example 2. Consider the CSP in Example 1. 1 2 8( , , , )x x x is a suitable variable ordering for the tree-decomposition of Fig. 1, b. Given 1 2 3= ( , , )x a x b x c   , the variable set of 2/ [ ]1 2   is 2( )Desc  , (with 2 3 ={ }, ={ }x xd b d c and 4 5 6 = = = { , , , }x x xd d d a b c d ) and its constraint set is 24 25 34 35 45 46 56{ , , , , , , }c c c c c c c . For instance, 2/ 3/( , ) 3/( , )4 5 4 5 = =x a x d x d x a      = 2 + 2 = 4. And the number of solutions of 4/( )3x c is 4/ = 6 . Thus, there are 2/ 4/ = 24   ex- tensions of  being solutions of . Note that for 4/( )3x c , 3(( ),6)x c is a #good. So, for any other assignment  of 1 with 3x assigned to c, it is not necessary to compute the number of solu- tions of 4/ [ ]1 4    because the #good 3(( ),6)x c will be exploited in this case. The total number of solutions is 1/ = 576 . #BTD is described in Algorithm 1. Given an as- signment , a cluster i , and a set i V of unassig- ned variables of i , #BTD (, i , i V ) looks for the number /i  of extensions  of  on ( )iDesc  such that [ \ ] = [ \ ]i ii i V V     . The first call is #BTD ( 1 1, ,   ) and it returns the number of so- lutions 1/ . Inside a cluster i , it proceeds clas- sically by assigning a value to a variable and by backtracking if any constraint is violated. When every variable in i is assigned, #BTD computes the number of solutions of the subproblem rooted in the first son of i , if there is one (otherwise the current subproblem is totally assigned and con- tains only one solution). More generally, let us consider j , a son of i . Given a current assign- ment  of i , #BTD checks whether the assign- ment [ ]i j   corresponds to a #good. If so, #BTD multiplies the recorded number of solutions with the current number of solutions NbSol ( /i  ). Otherwise, it extends  on ( )jDesc  in 8 УСиМ, 2011, № 2 order to compute its number of consistent exten- sions nb ( /j  ). Then, it records the #good ( [ ], )i j nb   . #BTD computes the number of solutions of the subproblem rooted in the next son of i . Finally, when each son of i has been ex- amined, #BTD tries to modify the current assign- ment of i . The number of solutions of the sub- problem rooted in i is the sum of solution counts for every assignment of i . Theorem 1. #BTD is sound, complete and ter- minates. Proof. #BTD exploits two kinds of problem decomposition. The first one is based on condi- tioning. The second one is based on tree- decomposition. In the first case (Else branch start- ing at line 2 in Algorithm 1), let x a be the sub- problem derived from  by assigning variable x to value a. We have = x x aa d  . We denote Sol the set of solutions of  and =| |ol   . For any two distinct values a and b of xd , we have =x x a b ol ol      . Thus, the set { | }x a xol a d   is a partition of ol and = x axa d    . In the second case of problem decomposition (If branch starting at line 1 in Algorithm 1), we are dealing with independent subproblems. Two CSPs 1 1 1 1 1= ( , , , )X D C R and 2 2 2 2 2= ( , , , )X D C R are independent if and only if 1 2 =X X  . If 1 2=    with 1 and 2 two independent subproblems, then the solutions of  is the Carte- sian product of the solutions of 1 and 2 . There- fore, 1 2 =      . In the case of a tree- decomposition, given a cluster i and an assign- ment  of Y X such that ( ) = i iDesc Y  , we have all the subproblems / [ ] ,j ji j      ( )iSons  mutually independent. Then  has / /( ) = j i i jSons     consistent extensions on ( )iDesc  . If ( , )I nb is a #good of ][/ jij   such that [ ] =i jC C  , then  has nb consistent exten- sions on ( )jDesc C : / =j nb .  Theorem 2. #BTD has time complexity in 1( )wn m d   and space complexity in ( )sn s d  . Proof. Space complexity. #BTD only records #goods. These are assignments on the intersec- tions i j  with j a son of i . Therefore, if s is the size of the largest of these intersections, #BTD has a space complexity of ( )sn s d  be- cause the number of these intersections is bounded by n, while the number of #goods associated to one intersection is bounded by sd and the size of a #good is at most s. Time complexity. In the worst case, #BTD ex- plores all the clusters (at most n) and tries all the values of every variable inside each cluster, each time checking at most m constraints at line 3. УСиМ, 2011, № 2 9 Thanks to its #good recording mechanism, it never explores the same cluster with the same as- signment of its variables twice. The number of assignments of a cluster is bounded by 1wd  with = max | | 1ii w     , the width of the tree-decom- position. Consequently, #BTD has a time com- plexity in 1( )wn m d   . In practice, for problems with a large tree- width, #BTD may run out of time and memory, as shown in Section 5. In this case, we are interested in an approximate method. 4. Approximate solution counting with Approx#BTD We consider here CSPs with a large tree-width but a sparse constraint graph. We define a collec- tion of easy-to-solve subproblems of an original problem  by partitioning the set of constraints, that is the set of edges in the constraint graph in the case of a binary CSP. The constraint graph ( , )X C will be partitioned into k subgraphs 1 1( , )X E ,  , ( , )k kX E , such that =iX X , =iE C and =iE  . We add the extra prop- erty that each ( , )i iX E is chordal (without adding extra edges as for building a tree-decomposition). Thus, each ( , )i iX E will be associated to a chordal subproblem i (with corresponding sets of vari- ables iX and constraints iE ), which should have a small tree-width and be efficiently solved using #BTD. Assume that i  is the number of solutions for each subproblem i , 1 i k  . We will estimate the number of solutions of  exploiting the fol- lowing property. Let  be any assignment of X, we denote ( )  the probability of « is a solution of P». We have ( ) = xx X d    , assuming a uniform prior probability distribution among the different value assignments. We also have 1 2( ) = ( )k           = 1 2 1 1 1 = ( ) ( | ) ( | ).k k                    In order to simplify these conditional probabili- ties, we assume probability independence between the ( )i  terms, which is true only if =iX  . Thus, we have 1 2( ) ( ) ( ) ( ).k            Now, we can easily deduce the following prop- erty in order to estimate  . Property 1. Given a CSP = ( , , , )X D C R and a partition 1{ ,..., }k  of  induced by a partition of C in k elements. =1 i i k x i x Xx x X d d                         . (1) Recall that this approximation returns an exact answer if all the subproblems are independent ( =iX  ) or = 1k ( is already chordal as in Example 1) or if there exists an inconsistent sub- problem i ( has no solution)3. Moreover, we can provide a trivial upper bound on the number of solutions due to the fact that each subproblem i is a relaxation of  (the same argument is used in [Pesant, 2005] to construct an upper bound): [1, ] min i i x i k x Xx x X d d         (2) Approx#BTD is described in Algorithm 2. Applied to a problem  with constraint graph ( , )X C , the method builds a partition 1{ ,..., }kE E of C such that the constraint graph ( , )i iX E is chordal for all 1 i k  . Each chordal subgraph is produced by the MaxChord  algorithm4 [Dear- 3Due to the celling function in Equation 1, if the ap- proximation returns zero then  has no solution. 4MaxChord+ returns a maximal subgraph for binary CSPs. For non-binary CSPs, we do not guarantee subgraph maximality and add to the subproblem all the constraints totally included in the extracted chordal subgraph. In Figure 2, the edge {x3,x5} has been removed from the first part be- 10 УСиМ, 2011, № 2 ing et al., 1988], described in Algorithm 3. An ex- ample of a partition found by Approx#BTD is given in Figure 2. Subproblems associated to (Xi, Ei) are solved with #BTD. The method returns an approximation to the number of solutions of  based on Property 1. Theorem 3. Approx#BTD is sound, complete and terminates. Proof. It suffices to prove that we have a parti- tion of the constraints at the end of the while loop in order to be able to apply Property 1. This can be easily shown by induction using the invariants  =1= i j jX X X  and « 1 2= ( , , , , )iQ C E E E  is a partition of C» inside the loop. Theorem 4. Approx#BTD has time complex- ity in 2 1( )wn m d   and space complexity in ( )sn s d   with < 1 1s w c w n       . Proof. Space complexity. Approx#BTD has the same space complexity as #BTD applied on the larg- est subproblem i w.r.t. the largest cluster inter- section denoted , , , [1, ]= | |max i i i u v i i u v i k u vs         . Time complexity. The number of iterations of Approx#BTD is less than n. At each step, the cause it is associated to the ternary constraint {x3, x4x5} not totally included in this part. first variable considered by MaxChord  at line 4 will have all its constraints totally included in the maximal chordal subgraph. Each chordal subgraph and its associated optimal tree-decomposition can be computed in ( )O nm [Dearing et al., 1988]. Thus, the time complexity of Approx#BTD is in 2 1( )wn m d   with the largest subproblem tree- width , [1, ]= | | 1max i i u i i k uw      . Because each i is a partial chordal subgraph of , its tree-width w is equal to the maximum clique size in its sub- graph [Fulkerson and Gross, 1965] which is by definition less than or equal to the maximum clique size of the original problem, called the clique number c , inferior to the problem tree- width 1w .  5. The experimental results We implemented #BTD and Approx#BTD counting methods on top of toulbar2 C++ solver5. The experimentations were performed on 5http://mulcyber.toulouse.inra.fr/projects/toulbar2/ ver- sion 0.8.1. УСиМ, 2011, № 2 11 a Linux 2.6 GHz Intel Xeon computer with 2GB. Reported times (total CPU times as given by the #SAT solvers or reported by the bash command «time» if not) are in seconds. For #BTD and Approx#BTD, the total time does not include the task of finding a variable elimination ordering6. We limit to one hour the time spent for solving a given instance. Inside #BTD (line 3), we use gen- eralized arc consistency (only for constraints with 2 or 3 unassigned variables) instead of backward checking, for efficiency reasons. The min domain /max degree dynamic variable ordering, modified by a conflict back-jumping heuristic [Lecoutre et al., 2006], is used inside the clusters. Our methods exploit a binary branching scheme. The variable is assigned to its first value or this value is removed from the domain. а – An original problem b – The first part c – The second part Fig. 2. A partition of a CSP with 6 variables found by Approx#BTD. The original problem has 5 binary constraints ({x1,x2}, {x1,x4}, {x1,x6}, {x2,x3}, {x5,x6}), and one ternary constraint {x3,x4,x5}. We have k = 2, w’ = 2, c = w =3 We performed experiments on SAT and CSP benchmarks7. We selected academic (random k- SAT wff, All-Interval Series ais, Towers of Hanoi hanoi) and industrial (circuit fault analysis ssa and bit, logistics planning logistics) satisfiable in- stances. CSP benchmarks are graph coloring instan- ces (counting the number of optimal solutions). For #SAT solvers only, CSP instances are trans- lated into SAT by using the direct encoding (one Boolean variable per domain value, one clause per domain to enforce at least one domain value is selected, and a set of binary clauses to forbid mul- tiple value selection). 6Which was always fast to compute (linear complexity). 7From www.satlib.org, www.satcompetition.org and mat.gsia.cmu.edu/COLOR02/. 5.1. The evaluation of exact methods We compared #BTD with state-of-the-art #SAT solvers Relsat [Bayardo and Pehoushek, 2000] v2.02, Cachet [Sang et al., 2004] v1.22 (with a default memory limit of 5 MB), sharpSAT [Thur- ley, 2006] v1.1, and also c2d [Darwiche, 2004] v2.20 which also exploits the problem structure (with a default memory limit of 512 MB and a limit of 64 MB for storing its d-DNNF). c2d and #BTD methods use the Min-Fill variable elimina- tion ordering heuristic (except for hanoi where we used the default file order) to construct a tree- decomposition / d-DNNF. Our results are summarized in Table 1. The co T a b l e 1. Comparison of exact methods. Legend: mem: out of memory, – : out of time (for c2d, * : out of memory for storing NNF) c2d s h a r p S A T C a c h e t R e l s a t #B T D  n (Bool- vars) d w  Time Time Time Time Time SAT wff.3.100.150 100 39 1.8e21 * mem – – mem wff.3.150.525 150 92 1.4e14 * mem 266. 2509 mem wff.4.100.500 100 80 – * mem – – mem ssa7552-038 1501 25 2.84e40 0.15 0.06 0.22 67 0.65 ssa7552-158 1363 9 2.56e31 0.10 0.03 0.07 3 0.19 ssa7552-159 1363 11 7.66e33 0.09 0.04 0.07 4 0.27 ssa7552-160 1391 12 7.47e32 0.12 0.04 0.08 5 0.30 2bitcomp_5 125 36 9.84e15 0.43 0.05 0.14 1 16.24 2bitmax_6 252 2 58 2.10e29 17.00 0.87 1.51 20 mem ais6 61 41 24 0.05 0.01 0.03 <1 0.08 ais8 113 77 40 0.51 0.17 0.58 <1 3.27 ais10 181 116 296 16.64 4.13 29.19 6 543 ais12 265 181 1328 1147 161. 2173 229 – logistics.a 828 116 3.8e14 – 0.17 3.78 10 mem logistics.b 843 107 2.3e23 – 1.38 12.34 433 mem hanoi4 718 46 1 7.18 1.11 32.64 3 1.87 hanoi5 1931 58 1 – mem – – 26.75 CSP (Graph Coloring) 2-Insertions_3 37 (148) 4 9 6.84e13 235. mem – – 7.80 2-Insertions_4 149 (596) 4 38 – – mem – – – DSJC125.1 125 (625) 5 65 – – mem – – mem games120 120 (1080) 9 41 – – mem – – mem GEOM30a 30 (180) 6 6 4.98e14 0.86 5.53 – – 0.10 GEOM40 40 (240) 6 5 4.1e23 1.00 mem – – 0.09 le450_5a 450 (2250) 5 315 3840 – 32.31 318 326 1100 le450_5b 450 (2250) 5 318 120 – 13.12 227 187 1364 le450_5c 450 (2250) 5 315 120 – 2.18 19.09 57 47.53 le450_5d 450 (2250) 5 299 960 – 4.40 14.60 36 92.03 Mug100_1 100 (400) 4 3 1.3e37 0.19 23.88 – – 0.02 myciel5 47 (282) 6 21 – – mem – – mem lumns are: instance name, number of variables (and also the number of Boolean variables on translated 12 УСиМ, 2011, № 2 CSP instances), maximum domain size, width of the tree-decomposition, exact number of solutions if known, time for c2d, sharpSAT, Cachet, Relsat, and #BTD. We noticed that #BTD can solve instances with relatively small tree-widths (except for ais and le450 which have few solu- tions). Exact #SAT solvers generally perform bet- ter than #BTD on SAT instances (except for ha- noi5), with sharpSAT obtaining the best results, but have difficulties on translated CSP instances. Here, #BTD maintaining arc consistency performed better than #SAT solvers using unit propagation. 5.2. Evaluation of approximate methods Table 2 gives an analysis of Approx#BTD on the tested instances. The columns are: instance name, T a b l e 2. Analysis of Approx#BTD performance and subprob- lem features  n d  w w k  Time SAT wff.3.100.150 100 1.80e21 39 3 6  3.10e21  5.05e27 0.03 wff.3.150.525 150 1.14e14 92 3 13  1.58e15  1.13e42 0.18 wff.4.100.500 100 – 80 6 48  1.59e16  4.87e29 0.47 ssa7552-038 1501 2.84e40 25 6 4  9.33e38  3.79e142 1.34 ssa7552-158 1363 2.56e31 9 5 3  2.22e25  1.41e79 0.77 ssa7552-159 1363 7.66e33 11 5 3  6.53e27  6.33e88 0.85 ssa7552-160 1391 7.47e32 12 5 3  4.50e26  3.36e106 1.09 2bitcomp_5 125 9.84e15 36 6 4  8.61e16  6.81e27 0.02 2bitmax_6 252 2 2.10e29 58 6 4  4.53e29  7.19e45 0.10 ais6 61 24 41 12 8  1  1.81e9 0.04 ais8 113 40 77 16 11  1  3.49e15 0.20 ais10 181 296 116 21 23  1  2.06e22 0.84 ais12 265 1328 181 26 23  1  3.19e29 2.47 logistics.a 828 3.8e14 116 10 24  1  4.91e180 14.85 logistics.b 843 2.3e23 107 13 25  1  2.15e169 14.08 hanoi4 718 1 46 10 8  1  4.26e106 1.40 hanoi5 1931 1 58 12 11  1  4.48e309 16.05 CSP (Graph Coloring) 2-Insertions_3 37 4 6.84e13 9 1 3  1.91e13  6.00e17 0.01 2-Insertions_4 149 4 – 38 1 6  1.30e22  1.64e71 0.07 DSJC125.1 125 5 – 65 3 7  1.23e13  2.27e70 0.12 games120 120 9 – 41 8 6  1.12e78  1.92e99 9.84 GEOM30a 30 6 4.98e14 6 5 2  7.29e14  1.81e15 0.04 GEOM40 40 6 4.1e23 5 5 2  4.8e23  1.72e24 0.01 le450_5a 450 5 3840 315 4 13  1  2.41e216 3.17 le450_5b 450 5 120 318 4 13  1  5.72e213 3.23 le450_5c 450 5 120 315 4 20  1  1.49e201 7.42 le450_5d 450 5 960 299 4 20  1  8.58e200 7.38 mug100_1 100 4 1.3e37 3 2 2  5.33e37  7.18e41 0.01 myciel5 47 6 – 21 1 8  7.70e17  8.53e32 0.03 myciel6 95 7 – 35 1 13  5.49e29  9.80e73 0.19 myciel7 191 8 – 66 1 21  4.25e35  2.96e161 1.23 number of variables, maximum domain size, exact number of solutions if known, width of the tree- decomposition for the original problem, maximum width of the tree-decomposition for all the chordal subproblems, number of subproblems in the parti- tion, approximate number of solutions and upper- bound on the number of solutions as given by Equation 2, and time for Approx#BTD. Our approximate method exploits a partition of the constraint graph in such a way that the result- ing subproblems to solve have a small tree-width on these instances (w  26). It has the practical effect that the method is relatively fast whatever the original tree-width. Notice that the upper- bound is generally very poor even with a small number of subproblems (e.g. ssa). We also compared Approx#BTD with the ap- proximation method SampleCount [Gomes et al., 2007a]. With parameters (s = 20, t = 7,  = 1), SampleCount-LB provides an estimated lower bound on the number of solutions with a high- confidence interval (99% confidence), after seven runs. With parameters (s = 20, t = 1,  = 0), called SampleCount-A in the following table, it gives only an approximation without any guarantee, af- ter the first run of SampleCount-LB. Our results are summarized in Table 3. The co- lumns are : instance name, exact number of soluti- ons if known, approximate number of solutions and time for Approx#BTD and SampleCount-A, and estimated lower bound on the number of so- lutions and time for SampleCount-LB. The qua- T a b l e 3. Comparison of approximate methods. Legend: – : out of time (1 hour) Approx#BTD Sample Count-A Sample Count-LB   ˆ  T im e ˆ  T im e ˆ  T im e SAT wff.3.100.150 1.8e21  3.10e21 0.1  1.37e21 959.8 – – wff.3.150.525 1.4e14  1.58e15 0.2  3.80e14 0.7  2.53e12 4.6 wff.4.100.500 –  1.59e16 0.5  4.15e16 2045.0 – – ssa7552-038 2.84e40  9.33e38 1.3  1.11e40 134.0  3.54e38 1162.0 ssa7552-158 2.56e31  2.22e25 0.8  1.43e30 14.1  1.43e30 177.0 ssa7552-159 7.66e33  6.53e27 0.8  6.49e34 32.8  1.64e32 182.0 ssa7552-160 7.47e32  4.50e26 1.1  5.08e32 144.0  2.31e31 1293.0 2bitcomp_5 9.84e15  8.61e16 0.1  4.37e15 0.2  3.26e15 1.2 2bitmax_6 2.10e29  4.53e29 0.1  1.62e29 1.7  2.36e26 10.3 ais6 24  1 0.1  12 0.1  12 0.2 ais8 40  1 0.2  16 1.5  12 15.9 ais10 296  1 0.8  124 45.9  20 312.0 ais12 1328  1 2.5  0 9.2  0 9.2 logistics.a 3.8e14  1 14.8  7.25e11 171.0  0 605.0 logistics.b 2.3e23  1 14.1  2.13e23 199.0  0 229.0 hanoi4 1  1 1.4  0 5.2  0 5.2 hanoi5 1  1 16.0  0 6.1  0 6.2 УСиМ, 2011, № 2 13 Approx#BTD SampleCount-A SampleCount-LB   ˆ  T im e ˆ  T im e ˆ  T im e CSP (Graph Coloring) 2-Insertions_3 6.84e13  1.91e13 0.1  4.73e12 1.0  4.73e12 7.4 2-Insertions_4 –  1.30e22 0.1  0 3.8  0 3.8 DSJC125.1 –  1.23e13 0.1  0 73.1  0 73.2 games120 –  1.12e78 9.8  7.33e64 13.8  1.35e61 91.1 GEOM30a 4.98e14  7.29e14 0.1  1.23e13 0.4  3.28e12 3.7 GEOM40 4.1e23  4.8e23 0.1  2.14e20 1.5  6.50e19 9.3 le450_5a 3840  1 3.2  0 8.6  0 8.6 le450_5b 120  1 3.2  0 8.6  0 8.6 le450_5c 120  1 7.4  0 110.0  0 111.0 le450_5d 960  1 7.4  0 4.6  0 54.6 mug100_1 1.3e37  5.33e37 0.1  4.2e34 2.1  4.20e34 15.6 myciel5 –  7.69e17 0.1  7.29e17 0.9  7.29e17 6.4 myciel6 –  5.49e29 0.2  9.38e40 4.5  7.42e36 30.7 myciel7 –  4.26e35 1.2  1.37e80 27.7  5.56e74 163.0 lity of the approximation found by Approx#BTD is relatively good and it is comparable (except for ssa, logistics, and my- ciel6-7 benchmarks) to SampleCount-A, which takes more time. For graph coloring, Approx#BTD outperforms also a de- dicated CSP approach producing an estimated lower bound: 2_Insertion_3  2.3e12, games120  4.5e42, and mug100_1  1.0e28 in 1 minute each; myciel5  4.1e17 in 12 minutes, times were measured on a 3.8GHz Xeon as reported in [Gomes et al., 2007b]. 6. Conclusion In this paper, we have proposed two methods for counting solutions of CSPs. These methods are based on a structural decomposition of CSPs. We have presented an exact method, which is adapted to problems with a small tree-width. For prob- lems with a large tree-width and a sparse constraint graph, we have presented a new approximate method whose quality is comparable with existing methods, which is much faster than other approaches, and which requires no parameter tuning (ex- cept for the choice of a tree decomposition heuristic). Exploring other structural parameters [Nishimura et al., 2007, Samer and Szeider, 2010] should deserve future work. A practical im- provement of our approach would be to impose a limit on the maximum clique size of the extracted chordal subproblems when the original problem has large arity constraints or a large clique number. Conversely, denser non-chordal subproblems could be produced and solved in an anytime manner as done in [Choi and Darwiche,2006] when the original problem has a small clique number. A direction of future work is also to extend our approach to the problem of (approximate) inference in probabilistic discrete graphical models. 1. Arnborg S., Corneil D., Proskurowski A. Complexity of finding embeddings in a k-tree // SIAM J. of Dis- crete Mathematics. – 1987. – N 8. – P. 277–284. 2. Bayardo R., Pehoushek J. Counting models using con- nected components // Proc. of AAAI-00. – Austin, Texas. – 2000. – P. 157–162. 3. Burton R., Steif J. Nonuniqueness of measures of maxi- mal entropy for subshifts of finite type // Ergodic theory and dynamical system, 1994. 4. Choi A., Darwiche A. An edge deletion semantics for be- lief propagation and its practical impact on approxima- tion quality // Proc. of AAAI-06. – 2006. – P. 1107–1114. 5. Darwiche A. On the tractable counting of theory mod- els and its applications to truth maintenance and belief revision // J. of Applied Non-classical Logic. – 2001. – N 11. – P. 11–34. 6. Darwiche A. New advances in compiling cnf to decom- posable negation normal form // Proc. of ECAI-04. Valencia, Spain. – 2004. – P. 328–332. 7. Dearing P., Shier D., Warner D. Maximal chordal sub- graphs // Discrete Applied Mathematics. – 1988. – 20(3). – P. 181–190. 8. Dechter R., Mateescu R. And/or search spaces for graphi- cal models // Artif. Intell. – 2007. – 171(2–3). – P. 73–106. 9. Dechter R., Mateescu R. The impact of and/or search spaces on constraint satisfaction and counting // Proc. of CP-04. Toronto, CA. – 2004. – P. 731–736. 10. Favier A., de Givry S., Jégou P. Exploiting problem structure for solution counting // Proc. of CP-09, Lis- bon, Portugal. – 2009. – P. 335–343. 11. Freuder E., Quinn M. Taking advantage of stable sets of variables in constraint satisfaction problems // Proc. of IJCAI-85. Los Angeles, CA. – 1985. – P. 1076–1078. 12. Fulkerson D., Gross O. Incidence matrices and interval graphs // Pacific J. Math. – 1965. – 15(3). – P. 835–855. 13. Gogate V., Dechter R. Approximate counting by sam- pling the backtrack-free search space // Proc. of AAAI-07. Vancouver, CA. – 2007. – P. 198–203. 14. Gogate V., Dechter R. Approximate Solution Sampling (and Counting) on AND/OR search spaces // Proc. of CP-08. Sydney, AU. – 2008. – P. 534–538. 15. Gomes C., Sabharwal A., Selman B. Model counting: A new strategy for obtaining good bounds // Proc. of AAAI-06. Boston, MA. – 2006. – P. 534–538. 16. From sampling to model counting / C. Gomes, J. Hoff- mann, A. Sabharwal et al. // Proc. of IJCAI-07. Hydera- bad, India. – 2007. – P. 2293–2299. 17. Counting CSP solutions using generalized XOR constraints / C. Gomes,W-J. van Hoeve, A. Sabharwal et al. // Proc. of AAAI-07. Vancouver, BC. – 2007. – P. 204–209. 18. Gomes C., Sabharwal A., Selman B. Handbook of Satis- fiability, chapter 20, Model Counting. IOS Press, 2009. 19. Jégou P., Terrioux C. Hybrid backtracking bounded by tree-decomposition of constraint networks // Artificial Intelligence. – 2003. – 146. – P. 43–75. 20. Kask K., Dechter R., Gogate V. New look-ahead schemes for constraint satisfaction // Proc. of AI&M, 2004. 21. Kroc L., Sabharwal A., Selman B. Leveraging belief propagation, backtrack search, and statistics for model counting // Proc. of CPAIOR-08. Paris, France. – 2008. – P. 127–141. End on the page 70 70 УСиМ, 2011, № 2 Ending of the article by A. Favier et al. 22. Kumar T. A model counting characterization of diag- noses // Proc. of the 13th International Work-shop on Principles of Diagnosis, 2002. 23. Last conflict based reasoning / C. Lecoutre, L. Sais, S. Tabary, V. Vidal // Proc. of ECAI-06. Trento, It- aly. – 2006. – P. 133–137. 24. Mann M., Tack G., Will S. Decomposition during search for propagation-based constraint solvers. CoRR, abs/ 0712.2389, 2007. 25. Montanari U. Network of constraints: Fundamental properties and applications to picture processing // Inf. Sci. – 1974. – N 7. – P. 95–132. 26. Nishimura N., Ragde P., Szeider S. Solving #SAT us- ing vertex covers // Acta Inf. – 2007. – 44(7). – P. 509–523. 27. Pesant G. Counting solutions of CSPs: A structural ap- proach // Proc. of IJCAI-05. Edinburgh, Scotland. – 2005. – P. 260–265. 28. Robertson N., Seymour P. Graph minors II: Algo- rithmic aspects of tree-width // Algorithms. – 1986. – N 7. – P. 309–322. 29. Rose D. Tringulated graphs and the elimination process // J. of Mathematical Analysis and its Applications. – 1970. – N 32. 30. Roth D. On the hardness of approximate reasoning // Artificial Intelligence. – 1996. – 82(1–2). – P. 273–302. 31. Samer M., Szeider S. Algorithms for propositional model counting // J. of Discrete Algorithms. – 2010. – N 8. – P. 50–64. 32. Combining component caching and clause learning for effective model counting / T. Sang, F. Bacchus, P. Beame et al. // In SAT-04, Vancouver, Canada, 2004. 33. Thurley M. SharpSAT – counting models with advan- ced component caching and implicit BCP // Proc. of SAT-06. Seattle, WA. – 2006. – P. 424–429. 34. Valiant L. The complexity of computing the permanent // Theoretical Computer Sciences. – 1979. – N 8. – P. 189–201. 35. Wei W., Selman B. A new approach to model counting // Proc. of SAT-05. St. Andrews, UK. – 2005. – P. 324– 339. 36. Zanarini A., Pesant G. Solution counting algorithms for constraint-centered search heuristics // Constraints. – 2009. – 14(3). – P. 392–413. © A. Favier, S.de Givry, INRA MIA Toulouse, France; Ph. Jégou, Universitè Paul Cèzanne, Marseille, France, 2011  3.pdf 70.pdf << /ASCII85EncodePages false /AllowTransparency false /AutoPositionEPSFiles true /AutoRotatePages /None /Binding /Left /CalGrayProfile (Dot Gain 20%) /CalRGBProfile (sRGB IEC61966-2.1) /CalCMYKProfile (U.S. Web Coated \050SWOP\051 v2) /sRGBProfile (sRGB IEC61966-2.1) /CannotEmbedFontPolicy /Error /CompatibilityLevel 1.4 /CompressObjects /Tags /CompressPages true /ConvertImagesToIndexed true /PassThroughJPEGImages true /CreateJobTicket false /DefaultRenderingIntent /Default /DetectBlends true /DetectCurves 0.0000 /ColorConversionStrategy /CMYK /DoThumbnails false /EmbedAllFonts true /EmbedOpenType false /ParseICCProfilesInComments true /EmbedJobOptions true /DSCReportingLevel 0 /EmitDSCWarnings false /EndPage -1 /ImageMemory 1048576 /LockDistillerParams false /MaxSubsetPct 100 /Optimize true /OPM 1 /ParseDSCComments true /ParseDSCCommentsForDocInfo true /PreserveCopyPage true /PreserveDICMYKValues true /PreserveEPSInfo true /PreserveFlatness true /PreserveHalftoneInfo false /PreserveOPIComments true /PreserveOverprintSettings true /StartPage 1 /SubsetFonts true /TransferFunctionInfo /Apply /UCRandBGInfo /Preserve /UsePrologue false /ColorSettingsFile () /AlwaysEmbed [ true ] /NeverEmbed [ true ] /AntiAliasColorImages false /CropColorImages true /ColorImageMinResolution 300 /ColorImageMinResolutionPolicy /OK /DownsampleColorImages true /ColorImageDownsampleType /Bicubic /ColorImageResolution 300 /ColorImageDepth -1 /ColorImageMinDownsampleDepth 1 /ColorImageDownsampleThreshold 1.50000 /EncodeColorImages true /ColorImageFilter /DCTEncode /AutoFilterColorImages true /ColorImageAutoFilterStrategy /JPEG /ColorACSImageDict << /QFactor 0.15 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >> /ColorImageDict << /QFactor 0.15 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >> /JPEG2000ColorACSImageDict << /TileWidth 256 /TileHeight 256 /Quality 30 >> /JPEG2000ColorImageDict << /TileWidth 256 /TileHeight 256 /Quality 30 >> /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 300 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict << /QFactor 0.15 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >> /GrayImageDict << /QFactor 0.15 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >> /JPEG2000GrayACSImageDict << /TileWidth 256 /TileHeight 256 /Quality 30 >> /JPEG2000GrayImageDict << /TileWidth 256 /TileHeight 256 /Quality 30 >> /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 1200 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 1200 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict << /K -1 >> /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile () /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False /CreateJDFFile false /Description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> /CHS <FEFF4f7f75288fd94e9b8bbe5b9a521b5efa7684002000410064006f006200650020005000440046002065876863900275284e8e9ad88d2891cf76845370524d53705237300260a853ef4ee54f7f75280020004100630072006f0062006100740020548c002000410064006f00620065002000520065006100640065007200200035002e003000204ee553ca66f49ad87248672c676562535f00521b5efa768400200050004400460020658768633002> /CHT <FEFF4f7f752890194e9b8a2d7f6e5efa7acb7684002000410064006f006200650020005000440046002065874ef69069752865bc9ad854c18cea76845370524d5370523786557406300260a853ef4ee54f7f75280020004100630072006f0062006100740020548c002000410064006f00620065002000520065006100640065007200200035002e003000204ee553ca66f49ad87248672c4f86958b555f5df25efa7acb76840020005000440046002065874ef63002> /CZE <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> /DAN <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> /DEU <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> /ESP <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> /ETI <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> /FRA <FEFF005500740069006c006900730065007a00200063006500730020006f007000740069006f006e00730020006100660069006e00200064006500200063007200e900650072002000640065007300200064006f00630075006d0065006e00740073002000410064006f00620065002000500044004600200070006f0075007200200075006e00650020007100750061006c0069007400e90020006400270069006d007000720065007300730069006f006e00200070007200e9007000720065007300730065002e0020004c0065007300200064006f00630075006d0065006e00740073002000500044004600200063007200e900e90073002000700065007500760065006e0074002000ea0074007200650020006f007500760065007200740073002000640061006e00730020004100630072006f006200610074002c002000610069006e00730069002000710075002700410064006f00620065002000520065006100640065007200200035002e0030002000650074002000760065007200730069006f006e007300200075006c007400e90072006900650075007200650073002e> /GRE <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a stvaranje Adobe PDF dokumenata najpogodnijih za visokokvalitetni ispis prije tiskanja koristite ove postavke. Stvoreni PDF dokumenti mogu se otvoriti Acrobat i Adobe Reader 5.0 i kasnijim verzijama.) /HUN <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> /ITA <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> /JPN <FEFF9ad854c18cea306a30d730ea30d730ec30b951fa529b7528002000410064006f0062006500200050004400460020658766f8306e4f5c6210306b4f7f75283057307e305930023053306e8a2d5b9a30674f5c62103055308c305f0020005000440046002030d530a130a430eb306f3001004100630072006f0062006100740020304a30883073002000410064006f00620065002000520065006100640065007200200035002e003000204ee5964d3067958b304f30533068304c3067304d307e305930023053306e8a2d5b9a306b306f30d530a930f330c8306e57cb30818fbc307f304c5fc59808306730593002> /KOR <FEFFc7740020c124c815c7440020c0acc6a9d558c5ec0020ace0d488c9c80020c2dcd5d80020c778c1c4c5d00020ac00c7a50020c801d569d55c002000410064006f0062006500200050004400460020bb38c11cb97c0020c791c131d569b2c8b2e4002e0020c774b807ac8c0020c791c131b41c00200050004400460020bb38c11cb2940020004100630072006f0062006100740020bc0f002000410064006f00620065002000520065006100640065007200200035002e00300020c774c0c1c5d0c11c0020c5f40020c2180020c788c2b5b2c8b2e4002e> /LTH <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> /LVI <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> /NLD (Gebruik deze instellingen om Adobe PDF-documenten te maken die zijn geoptimaliseerd voor prepress-afdrukken van hoge kwaliteit. De gemaakte PDF-documenten kunnen worden geopend met Acrobat en Adobe Reader 5.0 en hoger.) /NOR <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> /POL <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> /PTB <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> /RUM <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> /RUS <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> /SKY <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> /SLV <FEFF005400650020006e006100730074006100760069007400760065002000750070006f0072006100620069007400650020007a00610020007500730074007600610072006a0061006e006a006500200064006f006b0075006d0065006e0074006f0076002000410064006f006200650020005000440046002c0020006b006900200073006f0020006e0061006a007000720069006d00650072006e0065006a016100690020007a00610020006b0061006b006f0076006f00730074006e006f0020007400690073006b0061006e006a00650020007300200070007200690070007200610076006f0020006e00610020007400690073006b002e00200020005500730074007600610072006a0065006e006500200064006f006b0075006d0065006e0074006500200050004400460020006a00650020006d006f0067006f010d00650020006f0064007000720065007400690020007a0020004100630072006f00620061007400200069006e002000410064006f00620065002000520065006100640065007200200035002e003000200069006e0020006e006f00760065006a01610069006d002e> /SUO <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> /SVE <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> /TUR <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> /UKR <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> /ENU (Use these settings to create Adobe PDF documents best suited for high-quality prepress printing. Created PDF documents can be opened with Acrobat and Adobe Reader 5.0 and later.) >> /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ << /AsReaderSpreads false /CropImagesToFrames true /ErrorControl /WarnAndContinue /FlattenerIgnoreSpreadOverrides false /IncludeGuidesGrids false /IncludeNonPrinting false /IncludeSlug false /Namespace [ (Adobe) (InDesign) (4.0) ] /OmitPlacedBitmaps false /OmitPlacedEPS false /OmitPlacedPDF false /SimulateOverprint /Legacy >> << /AddBleedMarks false /AddColorBars false /AddCropMarks false /AddPageInfo false /AddRegMarks false /ConvertColors /ConvertToCMYK /DestinationProfileName () /DestinationProfileSelector /DocumentCMYK /Downsample16BitImages true /FlattenerPreset << /PresetSelector /MediumResolution >> /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles false /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /DocumentCMYK /PreserveEditing true /UntaggedCMYKHandling /LeaveUntagged /UntaggedRGBHandling /UseDocumentProfile /UseDocumentBleed false >> ] >> setdistillerparams << /HWResolution [2400 2400] /PageSize [612.000 792.000] >> setpagedevice
id nasplib_isofts_kiev_ua-123456789-82919
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
issn 0130-5395
language English
last_indexed 2025-12-07T13:13:54Z
publishDate 2011
publisher Міжнародний науково-навчальний центр інформаційних технологій і систем НАН та МОН України
record_format dspace
spelling Favier, A.
de Givry, S.
Jégou, P.
2015-06-11T20:00:01Z
2015-06-11T20:00:01Z
2011
Solution Counting for CSP and SAT with Large Tree-Width / A. Favier, S. de Givry, Ph. Jégou // Управляющие системы и машины. — 2011. — № 2. — С. 4-13, 70. — Бібліогр.: 36 назв. — англ.
0130-5395
https://nasplib.isofts.kiev.ua/handle/123456789/82919
519.157
Рассмотрена проблема подсчета количества решений задачи совместимости ограничений (Constraint Satisfaction Problem). Для ее решения был адаптирован метод обратного прослеживания с ацикличным представлением графа ограничений (Backtracking with Tree-Decomposition). Предложен точный алгоритм, сложность которого экспоненциально зависит от ширины дерева, и приближенный алгоритм, экспоненциально зависящий от размера максимальной клики.
The problem of counting the number of solutions of a CSP is considered. For solving the problem the Backtracking with a Tree-Decomposition method was adapted. The exact algorithm is suggested which has the worst-time complexity exponential in a tree width, as well as iterative algorithm that has computational complexity exponential in a maximum clique size.
Розглянуто проблему підрахунку кількості розв’язків задачі сумісності обмежень (Constraint Satisfaction Problem). Для її розв’язку було адаптовано метод зворотного простеження з ациклічним поданням графа обмежень (Backtracking with Tree-Decomposition). Запропоновано точний алгоритм, складність якого експоненційно залежить від ширини дерева, і наближений алгоритм, експоненційно залежний від розміру максимальної кліки.
en
Міжнародний науково-навчальний центр інформаційних технологій і систем НАН та МОН України
Управляющие системы и машины
Оптимизационные задачи структурного распознавания образов
Solution Counting for CSP and SAT with Large Tree-Width
Подсчет количества решений задач CSP и SAT на деревьях большой ширины
Підрахунок кількості розв’язків задач CSP та SAT на деревах великої ширини
Article
published earlier
spellingShingle Solution Counting for CSP and SAT with Large Tree-Width
Favier, A.
de Givry, S.
Jégou, P.
Оптимизационные задачи структурного распознавания образов
title Solution Counting for CSP and SAT with Large Tree-Width
title_alt Подсчет количества решений задач CSP и SAT на деревьях большой ширины
Підрахунок кількості розв’язків задач CSP та SAT на деревах великої ширини
title_full Solution Counting for CSP and SAT with Large Tree-Width
title_fullStr Solution Counting for CSP and SAT with Large Tree-Width
title_full_unstemmed Solution Counting for CSP and SAT with Large Tree-Width
title_short Solution Counting for CSP and SAT with Large Tree-Width
title_sort solution counting for csp and sat with large tree-width
topic Оптимизационные задачи структурного распознавания образов
topic_facet Оптимизационные задачи структурного распознавания образов
url https://nasplib.isofts.kiev.ua/handle/123456789/82919
work_keys_str_mv AT faviera solutioncountingforcspandsatwithlargetreewidth
AT degivrys solutioncountingforcspandsatwithlargetreewidth
AT jegoup solutioncountingforcspandsatwithlargetreewidth
AT faviera podsčetkoličestvarešeniizadačcspisatnaderevʹâhbolʹšoiširiny
AT degivrys podsčetkoličestvarešeniizadačcspisatnaderevʹâhbolʹšoiširiny
AT jegoup podsčetkoličestvarešeniizadačcspisatnaderevʹâhbolʹšoiširiny
AT faviera pídrahunokkílʹkostírozvâzkívzadačcsptasatnaderevahvelikoíširini
AT degivrys pídrahunokkílʹkostírozvâzkívzadačcsptasatnaderevahvelikoíširini
AT jegoup pídrahunokkílʹkostírozvâzkívzadačcsptasatnaderevahvelikoíširini