Generalized Davis-Putnam and Satisfiability Problems in Mathematics Robert Cowen September 1, 2009

Abstract One of the main techniques in automated theorem proving (ATP) is the Davis-Putnam procedure and we offer a version that is independent of mathematical logic.

1 Introduction There are many problems in mathematics where one seeks a particular solution subject to certain constraints. We have introduced in [2] a framework for considering a large class of these problems and then showed in [4] how the ideas of Davis and Putnam can be applied directly, without first converting the problems to Logic, to find solutions. It is our view that the Davis-Putnam procedure is combinatorial in nature and need not be tied to formal logic. Here, we treat a wide range of problems: graph coloring, property B, Ramsey Theory, Sudoku and satisfiability in propositional logic from this point of view.

2 Properties B and S A set of finite subsets of a set is said to have Property B if there is a transversal such that 1) , for each 2) for each This is equivalent to the following statement, as the diagram in Figure 1, below, suggests. such that

There is a partition of

and !V!"!Σ

Σ

C

Figure 1 Department of Mathematics, Emeritus, Queens College, CUNY

1

, for each

Property B was named for Felix Bernstein who first used it in Set Theory and has other uses as well(see [12],[13], [15].) A useful generalization of Property B is the following (see [2],[9]) Definition 2.1 Let be a set and let said to have Property S if there is a set , for each 1) 2) for each

be sets of finite subsets of such that

Property B is the special case of Property S when to the following definition as Figure 2 suggests.

. Then the pair

will be

. The above definition of Property S is equivalent

be sets of finite subsets of Definition 2.2 Let be a set and let said to have Property S if there is a partition of , as , for each 1) 2) , for each In either case we shall say that satisifies the pair is clearly unsatisfiable. However, if either satisfiable. Σ

. If or

. Then the pair and

will be

or contains , the empty set, is the empty set then is

!!V !-!Σ

c f

Figure 2 This property was first considered in print by Schrijver [7] in connection with set theoretic principles equivalent to the Boolean Prime Ideal Theorem. Adam Kolany introduced this property independently in [11] to study satisfiability problems and it was also introduced independently and studied by Peter Erd¨os [9]. We named this property in [2], “Property S” in recognition of Schrijver’s first use. We shall often refer to members of as “clauses” and members of as “forbidden sets.” Another important feature of Property S that is suggested by Figure 2 is Duality. Theorem 2.3 Proof

satisfies

is satisfiable iff iff

is satisfiable.

satisfies

.

3 Examples We give some examples which show the applicability of Property S in mathematics. In the following section we discuss a method for solving such problems. Other methods have been considered in [2], [3], [11]. 2

Example 3.1 Ramsey Theory. It is a well-known result from Ramsey Theory that at any party of six people, there are three mutual friends or three mutual strangers. Equivalently, any coloring of the edges of , the complete graph on six vertices, results in either a red or blue triangle. Let consist of all pairs tices) of the complete graph; let Then it must be shown that

, where ranges over the edges (unordered pairs of verbe all triples, , . does not have Property S.

Example 3.2 Propositional Logic A literal is a statement letter or a negated statement letter. A clause is a set of literals. A set of clauses is satisifiable if it is possible to select a literal from each without choosing some letter and its negation. Let be a set of clauses and let be the set of all pairs , has Property S. for ocurring in . Then is satisfiable iff Example 3.3 List Coloring A list coloring of a graph is a proper coloring of the vertices of the graph with colors for each vertex restricted to be in a list of available colors for that vertex. Consider the graph with vertex color lists shown in Figure 3, below.

1,{r,g}

2, {b,g}

3, {b,r}

4, {b,r}

5, {b,g}

6, {g,r}

Figure 3 Let

consist of the following pairs: . If two vertices are adjacent and allow the same color , we put the pair, into . Thus in the example above , etc., will be added to . Then the graph in Figure 4 is list colorable iff has Property S. Example 3.4 Sudoku. An Sudoku puzzle consists of an grid with integers in some of the boxes, where . It is required to fill the remaining boxes with integers in the same range in such a way that no integer occurs twice in any row, column or block (the grid is divided up into n ). We actually consider a more general puzzle where all the boxes contain sets blocks of size, of integers and it is required to choose one of the integers in the list (similar to the case of list colorings). , as an example. In Figure 4, the first figure is the original We consider a small Sudoku puzzle of size puzzle, the middle figure is the puzzle after some progress toward a solution and the last shows the letter names we shall assign to the 16 boxes.

3

Figure 4 To solve the puzzle we must pick an element out of each box; however, elements of the boxes must indicate in which box the element resides, so we construct lists for each box and refer to each element of box labeled as . We must then pick an element from each of these box lists, subject to and are in the same row, column or block we cannot pick both . To the constraint that if achieve this we let be these box lists,

. .

Let Then the Sudoku puzzle is solvable if and only if

has Property S.

Example 3.5 Steiner Triple System. Suppose C = 1, 2, 3 , 1, 4, 5 , 1, 6, 7 , 2, 4, 7 , 2, 5, 6 , 3, 4, 6 , 3,5,7 . This set is the Steiner Triple System, (For more on Steiner Triple systems, see [14]). Paul Erd¨os [8] reports that it can be shown, by means of a calculation, that this set fails to have Property B. This is equivalent to showing does not have Property S.

4 Generalized Davis-Putnam Rules Martin Davis and Hillary Putnam ([5] [6]) introduced, almost 50 years ago, a procedure for solving SAT problems that has been so successful that it still forms the basis of most programs used today. Recently, A. Kolany and I have published [4] an abstract version that is independent of logic and we will show how it applies to the examples above. Before we start we introduce some notation. is a set of subsets of and , consists of all sets obtained from the elements If of by removing the element ; while consists of all elements of which remain after removing all sets in which contain the element . Thus, if , then ; whereas, . , drop any clause in containing the Unit Prune Rule. If some clause in consists of a unit !), resulting in clause set ; further, remove from any set in , obtaining element (not just the collection, . It is a theorem of [4] that is satisfiable (has Property S) if and only if is satisfiable. A rationale for this rule is the following. Since is a clause, must belong to any hitting set . Then any clause containing will also be “hit” by ; thus the question becomes: “What clauses of are also hit by ?”; further, since , the forbidden sets containing can be replaced by those obtained by removing , that is . Pure Literal Rule. In propositional logic, a pure literal for a set of clauses, , is one whose opposite is missing from ( are opposites of each other). The pure literal rule allows the removal of clauses containing pure literals. These can be selected without worrying about forming forbidden sets since for 4

any such literal selected its opposite won’t be selected since it is completely absent from . . Then is a pure literal for if for any with , there is some Suppose with . The idea is that clauses containing pure literals can be removed from the clauses in without effecting satisfiability, since we could have chosen these pure literals as part of the hitting set without risking the formation of a forbidden set with our choice since there is always some element in each forbidden set containing , that is outside of . and satisfies ; if Splitting Rule. Suppose ; if , , and this implies that satisfies clear that if either satisfies or satisfies Hence we have the following principle.

or

Splitting Rule, I. Suppose that is satisfiable.

. Then

, then surely satisfies . Also, it should be , then satisfies .

is satisfiable if and only if

is a unit clause, we can apply the Unit Prune Rule (and the Duality Principle) to obtain the Since following version of the Splitting Rule. Splitting Rule, II. Suppose that or and only if either

is satisfiable and . Then is satisfiable.

is satisfiable if

5 Application of Davis-Putnam to the Mathematical Problems We now show how the Davis-Putman rules can be used to settle the mathematical problems introduced above. Solving the Sudoku Problem. Recall that the question has been reduced to deciding the satisfiability where, of

, .

and

It turns out that all we shall need is the Unit Prune Rule and Duality. Then, by the Unit Prune Rule, we can drop any set containing an element belonging to a singleton and also remove these singleton , where elements from . This leads to the following sets, .

Next, by Duality, we can use Unit Prune on the new unit clauses in dropping all clauses which contain elements in unit clauses, to obtain, , the empty set. If we remove these unit elements from we get . Hence is satisfiable and thus so are 5

and

.

, define the solution: It is now easy to see that the unit clauses in a=1,b=3, c=4, d=2, e=4, f=2, g=1, h=3, i=3, j=4, k=2, l=1, m=2, n=1, o=3, p=4. Ramsey example does not have Property B Let fails to have Property B, that is, appeal to the Splitting Rule; since one of

. We will show that fails to have property S. Since there are no unit clauses, we , and since Duality holds, with any , we need only consider ; we choose the first.

Setting v = 1, let

; let . The problem has been reduced to show-

is unsatisfiable.

ing

There are no units present, so we again appeal to Splitting, with v = 2; now there are two cases and ; we shall cotinue with the former, to consider: leaving the latter for the reader, so, let , and . has the unit clause, , we apply Unit Prune to . Let and . If we then use Splitting on , with , to obtain and (again we leave the second case for the reader) and we continue working on the first case. Let and C . Now, applying Unit Prune with the unit clause , leads to and ; applying Unit Prune to using the unit clauses gives , which finally proves unsatisfiability; so fails to have property B, as promised. A direct proof that fails to have Property B is not given in [8]; instead it is left for the reader to verify by calculation. This might entail breaking the set into two parts in all possible ways and for each showing that there is some which lies entirely in one of these two parts. If one of the two parts has only one or two elements this is rather easy to do; however this still leaves Binomial(7,3)=35 cases to be checked. Since

Solving the List Coloring Problem We will show that no list coloring is possible. Here,

. for We will need to first use the splitting rule, since there are no unit clauses; we choose the splitting variable. Thus by the Splitting Rule, II we must show that both and are unsatisfiable. We continue with the first; Let be defined as follows. . 6

. Then we claim

is unsatisfiable.

has two unit clauses, . We will only use the Unit Prune Rule to produce in one of the sets of clauses. (To this end just use “ ” to elimate elements; one needn’t bother deleting clauses with , lead to unit clauses, obtained from using Unit “ ”.) The unit clauses in , , when Unit Prune is applied to . However, Prune; which in turn lead to unit clauses is a clause of ; thus one more application of Unit Prune produces the empty set and we are done in case one; we leave it to the reader to provide the argument for case two. Solving the Ramsey Example Going back to our orginal example from Ramsey Theory, the set of 15 Clauses(one for each edge of , , can be represented as indicated above, as: . consists of all triples, , , where . We have written a computer program that solves this problem on a laptop computer in under one second. Alternatively, if we take the first five clauses of : Since they are all pairs, if we split on each with a variable in the clause, at least three of the five must be of the same color, let us assume this color is red; further, since is symmetrical, we can assume these three variables are: . If we use the Splitting Rule with these three variables we drop them from the following clauses in : . This, in turn, creates the following units from these sets in : The Unit Prune Rule, then creates the following units from pairs in : However, contained , which then becomes after another application of Unit Prune. Thus is does not have Property B which implies there is no red-blue edge coloring of without either a red or blue triangle. (The alert reader will probably recognize this as a simple transformation of the standard proof of this theorem (see [10].)

6 Conclusion We have seen how the Davis-Putnam method for SAT testing in formal logic can be given a combinatorial formulation and applied directly to satisfy a variety of combinatorial satisifiability problems without first translating them into logic. We believe the Davis-Putnam procedure is essentially combinatorial and this should be recognized. It also brings the techniques closer to the problem and encourages the use of other combinatorial ideas, such as the Duality Principle. Perhaps other new general principles will be found useful and added to the Davis-Putnam procedure. Finally we ask is their a formulation of the Duality Principle that would be useful in the usual versions of Davis-Putnam in propositional logic?

7

7 References Cowen, R., Some connections between set theory and computer science, Lecture Notes in Computer Science 713, 14-22, 1993. Cowen, R., Property S, Reports on Math. Logic, 35, 61-74, 2001. Cowen, R, Combinatorial analytic tableaux, Reports on Math. Logic, 27(1993), 29-39. Cowen, R., Kolany, A., Davis-Putnam style rules for deciding Property S, Fund. Inform., 79(1-2), 5 15. Davis, M., Longemann, G., and Loveland, D., A machine program for theorem proving, Communications of the ACM, 5(1962), 394-397. Davis, M. and Putnam, H., A computing procedure for quantification theory, J. of the ACM, 7(1960), 201-215. Schrijver, A., The dependence of some some logical axioms on disjoint transversals and linked systems, Colloq. Math., 39(1978), 191-199. Erd¨os , P., On a combinatorial problem, Nordisk Mat. Tidskr. 11 (1963), 5–10. Erd¨os, P.L., Some generalizations of Property B and the splitting property, Annals of Combinatorics , 3(1999), 53-59. Graham, R., Rothschild, B.L., Spencer, J.H., Ramsey Theory, 2nd ed., Wiley Interscience, New York, 1990. Kolany, A., Satis ability on hypergraphs, Studia Logica, 52(1993), 393-404. Miller, E.W., On a property of families of sets, Comptes Rendus Varsovic, 30(1937), 31-38. Stein, S. K., B-sets and planar maps, Paci c J. Math. 37(1971), 217-224. Steiner Triple System, Wofram MathWorld, http://mathworld.wolfram.com/SteinerTripleSystem.html. Woodall, D. R., Property B and the four-colour problem, in Combinatorics, ed. Welsh, D.J. and Woodall, D.R., IMA, 1972.

8

## Generalized Davis-Putnam and Satisfiability ...

Sep 1, 2009 - We consider a small Sudoku puzzle of size .... written a computer program that solves this problem on a laptop computer in under one second.

#### Recommend Documents

Generalized Davis-Putnam and Satisfiability Problems ...
Oct 23, 2009 - The above definition of Property S is equivalent to the following definition as Figure 2 suggests. ..... is symmetrical, we can assume these.

Generalized Davis-Putnam and Satisfiability Problems ...
Mar 30, 2009 - [9]. We named this property in [2], âProperty Sâ in recognition of Schrijver's first use. We shall often refer to members of as âclausesâ and members of as âforbidden sets.â Another important feature of Property S that is s

Generalized Davis-Putnam and Satisfiability Problems ...
Sep 1, 2009 - In propositional logic, a pure literal for a set of clauses, , is one whose opposite ... 5 Application of Davis-Putnam to the Mathematical Problems.

Generalized Davis-Putnam and Satisfiability Problems ...
Oct 23, 2009 - written a computer program that solves this problem on a laptop computer in ... transformation of the standard proof of this theorem (see [10].) 7 ...

Generalized and Doubly Generalized LDPC Codes ...
The developed analytical tool is then exploited to design capacity ... error floor than capacity approaching LDPC and GLDPC codes, at the cost of increased.

Generalized quantifiers and clarification content
(13) Terry: Richard hit the ball on the car. ... Nick: What ball? ... Nick: What car? Terry: The car that was going past. ... Yeah a book, log book. BNC file KNV ...

Generalized Inquisitive Semantics and Logic
Aug 28, 2009 - Of course, this picture is limited in several ways. First ...... dition of Groenendijk's logic of interrogation (Groenendijk, 1999; ten Cate and Shan,.

Generalized Bernstein Property and Gravitational ...
over the hosting complex line bundle. ... equations (1.1) and (1.2) over a simply connected domain are, respectively, equivalent instead to the vector equations.

Generalized Inquisitive Semantics and Logic
Aug 28, 2009 - Submitted for publication, manuscript No. (will be inserted .... It follows from definition 2.3 that the support-conditions for Â¬Ï and !Ï are as ...... We will call a model negative in case its valuation is negative. Observe that if

Percolation and magnetization for generalized ...
clusters become infinite (see [3â6]). It is moreover possible to ... S. Fortunato, H. Satz / Nuclear Physics B 598 (2001) 601â611. Fig. 1. .... This variable, that we shall call percolation cumulant and indicate as Î³r(Îº), is a scaling .... We t

The Quest for Efficient Boolean Satisfiability Solvers
A DPLL-based SAT solver is a relatively small piece of software. Many of the ...... Efficient SAT Solver," presented at 39th Design Automation Conference, 2001.

Multidimensional generalized coherent states
Dec 10, 2002 - Generalized coherent states were presented recently for systems with one degree ... We thus obtain a property that we call evolution stability (temporal ...... The su(1, 1) symmetry has to be explored in a different way from the previo

Part IV. SAT (Boolean Formula Satisfiability Problem) - WordPress.com
Department of Computer Science & Engineering ..... Let somebody gives us a certificate that a hamiltonian cycle, A-C-B-D-F-E-A exists in the above graph, it can ...

GENERALIZED COMMUTATOR FORMULAS ...
To do this, we need to modify conjugation calculus which is used in the literature ...... and N.A. Vavilov, Decomposition of transvections: A theme with variations.

Generalized Anxiety Disorder
1997). Three classes of drugs are commonly used to treat .... of school phobia. Journal of ..... Paradoxical anxiety enhancement due to relaxation training. Journal of ..... Auto- nomic characteristics of generalized anxiety disorder and worry.

Generalized anxiety disorder
Taken together, data suggest that GAD symptoms are likely to ... Buhr, & Ladouceur, 2004); perseverative generation of problem solutions and interpreting ...

Generalized Inquisitive Logic
P to {0, 1}. We denote by Ï the set of all indices. Definition 3 (States). A state is a set of indices. We denote by S the set of all states. Definition 4 (Support).

GENERALIZED MACHINE THEORY.pdf
6. a) Using generalized machine theory obtain the impedance matrix and the. equivalent circuit of a 3Ï induction motor. 10. b) Show that for the stator and rotor ...

Generalized Anxiety Disorder
Aug 12, 2004 - ... at EASTERN VIRGINIA MEDICAL SCHOOL on February 18, 2007 . ..... eds. Massachusetts General Hospital hand- book of general hospital ...

Generalized Silver Codes
gineering, Indian Institute of Science, Bangalore 560012, India (e-mail: [email protected]; [email protected]). Communicated by E. Viterbo, Associate ...... 2006, he was with Robert Bosch India limited, Bangalore. Currently, he is a wor

Generalized Anxiety Disorder
Aug 12, 2004 - Warren 6, 55 Fruit St., Boston, MA 02114, .... though objective data on rates of relapse and remis- sion with longer use of the ..... Recovery rates.