Electronic Notes in Theoretical Computer Science 176 (2007) 79–93

www.elsevier.com/locate/entcs

Solving Sudoku Puzzles with Rewriting Rules 

Gustavo Santos-Garcı́a1
Universidad de Salamanca

Miguel Palomino2
Dpto. de Sistemas Informáticos y Programación, Universidad Complutense

Abstract
The aim of the sudoku puzzle (also known as number place in the United States) is to enter a numeral from
1 through 9 in each cell of a grid, most frequently a 9 × 9 grid made up of 3 × 3 subgrids, starting with
various numerals given in some of the cells (the “givens”). Each row, column, and region must contain only
one instance of each numeral. In this paper we show how a sudoku puzzle can be solved with rewriting
rules using Maude. Three processes (scanning, marking up, and analysis) are the classical techniques for
solving sudokus. Elimination is the main strategy that we have employed. The strategy what-if and several
contingencies are also implemented.

Keywords: Rewriting logic, Maude, sudoku, puzzle.

1 Introduction
Rewriting logic [13] is a logic of concurrent change that can naturally deal with states
and with highly nondeterministic concurrent computations. It has good properties
as a flexible and general semantic framework for giving semantics to a wide range of
languages and models of concurrency. Moreover, it allows user-definable syntax with
complete freedom to choose the operators and structural properties appropriate for
each problem.
The naturality of rewriting logic and of its implementation, Maude [9], for mod-
eling and experimenting with mathematical problems, has been illustrated in a
number of works. Our goal with this paper is to further contribute to that pool but
in a rather more recreational context, along the lines in [14].
 Work partially supported by Ministerio de Ciencia y Tecnologı́a I+D+i Projects BFF2003-08998-C03-01
and TIC-2003-01000.
1 Email:[email protected]
2 Email:[email protected]

1571-0661 © 2007 Elsevier B.V. Open access under CC BY-NC-ND license.

doi:10.1016/j.entcs.2007.06.009
80 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93

For that, we present a case study of how to use Maude to execute and solve a
popular kind of puzzles, namely sudokus. In Maude it is very easy to support objects
and distributed object interactions in a completely declarative style with rewrite
rules. We describe how sudokus can be represented in an object-oriented way and
give formal rules that transform an initial sudoku into one in solved form, and how
this representation can be straightforwardly mapped to Maude. The application of
these rules, though always leading to a solution, can do so in many different ways;
some give rise to combinatorial explosions and should thus be avoided. To handle
this matter we have employed strategies to guide Maude’s rewrite engine, which
constitute an example of the use of the recently developed strategy language for
Maude [11].

2 Sudokus
A (standard) sudoku is a 9×9 grid made up of 3×3 subgrids, also called “regions.”
Initially, some cells contain “given” numbers: the goal is to fill in the empty cells,
one number in each, so that each column, row, and grid contains the numbers 1
through 9 exactly once (see an example in Figure 1). Originally called number place,

3 7 2 1 5 8 3 7 2 4 1 9 6
6 9 5 8 1 2 6 9 3 5 8 7 4
4 9 1 5 2 4 9 7 6 1 8 3 5 2
5 6 1 3 5 9 8 4 2 7 6 1
8 4 2 9 8 1 4 5 6 7 2 3 9
7 6 4 7 6 2 3 9 1 5 4 8
2 3 5 1 7 2 3 8 4 5 9 6 1 7
1 2 3 9 6 4 1 2 7 3 9 8 5
5 8 6 4 9 7 5 1 8 6 4 2 3
Fig. 1. A sudoku puzzle (The
c Times, 2005, num. 295) and its solution.

the first such puzzle was created by Howard Garnes, a freelance puzzle constructor,
in 1979 [1]. The puzzle was first published in New York by the specialist puzzle
publisher Dell Magazines in its magazine Math Puzzles and Logic Problems. The
puzzle was introduced in Japan by the publishing company Nikoli in the paper
Monthly Nikolist in April 1984 as “Suji wa dokushin ni kagiru” [2], which can be
translated as “the numbers must occur only once” or “the numbers must be single.”
At a later date, the name was abbreviated to sudoku, pronounced sue-do-koo; su =
number, doku = single; it is a common practice in Japanese to take only the first
kanji of compound words to form a shorter version. In 1986, Nikoli introduced two
innovations which guaranteed the popularity of the puzzle: the number of givens was
restricted to no more than 30 and puzzles became symmetrical (meaning the givens
were distributed in rotationally symmetric cells). It is now published in mainstream
Japanese periodicals, such as the Asahi Shimbun. The surge in popularity of sudoku
in 2005 has led the world media to dub it as “the Rubik’s cube of the 21st century”
[3].
 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93 81

The numerals in sudoku puzzles are used for convenience. It works just as fine if
the numbers are replaced with letters, shapes, colours or some other symbols. Su-
doku is not a mathematical or arithmetical puzzle; arithmetic relationships between
numerals are absolutely irrelevant. Indeed, Penny Press use letters in their version
called scramblets; Knight Features Syndicate also use letters in their Sudoku Word
(see Figure 2).

R N B A O R E S NMB A U
N B R NMU A B E R O S
A O S B A O U R MN E
M N B O RM E S AUN
S N E M U S N R A B E MO
E U E AMN O U S B R
A O MN S U R A O E B
B S M A U B E S O NRM
E O M S R E O BMN U S A
Fig. 2. A Sudoku Word puzzle (Knight
c Features Syndicate, 2005) and its solution. Complete the grid so
that every row, column and 3 × 3 box contains a different letter: A, B, E, M, N, O, R, S & U. One row or
column contains a 7–letter word. What is it?

The Guardian calls these godoku and describes them as “devilish”; others name
them wordoku or sudoku word. The required letters are given beneath the puzzle:
once arranged they spell out a topical word between the top left and bottom right
corners. This adds an extra dimension to sudokus as it may be possible to guess
what the word is.
The appealing nature of the puzzle lies in the fact that the completion rules are
simple, yet the reasoning required to reach the solution may be difficult. It also
rises interesting questions, some of which are open: to measure the difficulty of a
puzzle, to construct new sudokus, to establish the number of possible sudokus, to
determine the number of possible puzzles with a single solution, to optimize the
search of the solution and so on.

2.1 An NP Problem
Constraint programming has reached the masses. When solving their daily sudoku
puzzle, thousands of newspaper readers apply classic propagation schemes in con-
straint programming like X-wing and swordfish [4]—patterns that cover several rows
and columns, seeking a candidate number that can be removed from other lists in
the corresponding columns and rows—to find a solution.
The general problem of solving sudoku puzzles on n2 × n2 boards of n × n blocks
is known to be NP-complete [15]. This gives some vague indication of why sudokus
are hard to solve, but for boards of finite size the problem is also finite and can
be solved by a deterministic finite automaton that knows the entire search tree.
However, for a non-trivial starting board the search tree is very large and so this
method is not feasible.
A valid sudoku solution is also a Latin square. A Latin square is an n × n
table filled using different symbols in such a way that each symbol occurs exactly
82 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93

once in each row and exactly once in each column. Sudoku imposes the additional
regional constraint; nonetheless, the number of valid sudoku solution grids for the
standard 9 × 9 grid is 6, 670, 903, 752, 021, 072, 936, 960. This number is equal to
9! × 722 × 27 × 27, 704, 267, 971, the last factor of which is prime. The result was
derived through logic and brute force computation; the methodology for this analysis
can be found at [10].

3 Definitions and Notation

3.1 Solution methods
As described in [12], three processes (scanning, marking up, and analysis) are the
classical techniques for solving sudokus. The application of these processes is not
deterministic so that the real performance of a sudoku solver depends on the ap-
proaches in their combination: if one applies these processes blindly, a combinatorial
blowup may happen. Ideally, then, one needs to find a combination of these tech-
niques which finds a solution in an efficient manner.

Scanning
It is performed at the outset and periodically throughout the procedure. It may
have to be performed several times between analysis periods. Scanning consists of
three basic methods that can be used alternatively: cross-hatching, counting, and
“looking for contingencies.”
Cross-hatching is the scanning of rows to identify which line in a particular grid
may contain a certain number by a process of elimination. This process is then
repeated with the columns. It is important to perform this process systematically,
checking all of the digits 1–9. For fastest results, the numbers are considered in
order of their frequency.
The counting of the occurrences of the numbers 1 to 9 in grids, rows, and columns
tries to identify missing numbers. Counting based upon the last number discovered
may speed up the search.
Advanced solvers look for “contingencies” while scanning, that is, narrowing
a number’s location within a row, column, or grid to two or three cells. When
those cells all lie within the same row (or column) and grid, they can be used for
elimination purposes during cross-hatching and counting. More difficult sudokus, by
definition, cannot be solved only by basic scanning alone and require the detection
of contingencies.

Marking up
Scanning comes to a halt when no more numbers can be discovered. At this
point, it is useful to mark candidate numbers in the blank cells. Subscripts are a
popular notation: the candidate numbers are written as subscripts in the cells. It is
usually difficult to use this method in the newspaper because the cell space is small.
There is a second notation with dots (a dot in the top left corner represents a 1 and
 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93 83

a dot in the bottom right corner represents a 9). This notation has the advantage
that it can be used on the original sudoku.

Analysis
There are two main analysis approaches: elimination and what-if. The elim-
ination of possible numbers from a cell allows to leave the only possible choice.
There are a number of elimination tactics. One of the most common is unmatched
candidate deletion: a collection of n cells with identical possible numbers is said
to be matched if the quantity of candidate numbers in each is equal to n. Then
the numbers appearing as candidates elsewhere in the same row, column, or grid in
unmatched cells can be deleted. For instance, if there are three cells in the same
row with {2, 7, 8} as their set of possible numbers, then 2, 7, and 8 can be discarded
as possible numbers from the remaining cells in that row.
Using the what-if approach (which is called reductio ad absurdum in [12]), a cell
with only two candidate numbers is selected and a guess is made. The procedure
then continues with the resulting sudoku and, if no solution is found, the alternative
number is tried.

3.2 Rules for solving sudokus

We will represent sudokus as a “soup” (formally, a set) of objects, where each object
corresponds to a cell.

Definition 3.1 A sudoku S of order n can be represented as a set of objects Cij ,

1 ≤ i, j ≤ n, one for each cell at row i and column j, where each object has the
following attributes:
√ √ √
• Gij = n · int((i − 1)/ n) + int((j − 1)/ n) + 1 is the grid to which the cell
belongs.
• Pij is the set of possible numbers that may occur in this cell. The sudoku will be
solved when this attribute is a unitary set for all cells; if P becomes empty for
some cell, the sudoku has no solution.
• Nij is the number of elements in Pij .
Note that it is not necessary to consider i and j as additional attributes since they
are part of a cell’s name. The standard sudoku is a 9 × 9 grid, that is, a sudoku of
order n = 9.

Initially Pij is equal to {1, . . . , n} except when Pij is the cell of a given gij , in
which case Pij = {gij }. Attributes Gij and Nij are stored explicitly but can be
computed from the rest of the elements of an object.
We will apply several reduction rules to cells to obtain a solution for a sudoku.
The goal is to reach a unique number in Pij for all the cells.
During the procedure, and due to the what-if rule, a sudoku may be split into
two and to distinguish one from the other we will enclose the corresponding objects
84 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93

in brackets. Thus, for example, the representation of two sudokus of order 9 will
look like
[C11 C12 . . . C99 ] [C 11 C 12 . . . C 99 ].
Among the different processes (scanning, marking up, and analysis) for solving
sudokus, the main strategy that we have considered is that of analysis for elimi-
nation. It is defined by Rule 1 below, which removes an element from the set of
possible numbers in a cell. We have complemented it with Rules 2 and 3, of second
and third order simplification, which respectively consider two and three elements.
Several contingencies corresponding to the scanning process are covered by Rules 4
to 6. The strategy what-if is implemented with the sudoku split rules (Rules 7 and
8).
We next present in detail each rule. They are to be understood as local transition
rules in a possibly concurrent system, that can concurrently be applied to different
fragments of the soup: the cells in the upper part of each sequent become the
ones below while the rest remain unchanged (since the attributes Nij and Gij are
computed from Pij , we do not explicitly mention their new values). We use ( )c to
denote the complement of a set.
Rule 1 (First order Simplification Rule) If only one number is possible in a
cell, then we remove this number from the set of possible numbers in all the other
cells in the same row, column or grid. Symbolically,
Cij Ci j 
Cij C i j 
where:
(i) i = i or j = j  or Gij = Gi j  ,
(ii) Pij = {p} ⊆ Pi j  ,
and the attribute P i j  of C i j  is equal to Pi j  − {p}.
Rule 2 (Second order Simplification Rule) If two cells in the same row (col-
umn or grid) have the same set of possible numbers and its cardinality is 2, then
those numbers can be removed from the sets of possible numbers of every other cell
in the same row (column or grid):
Cij Ci j  Ci j 
Cij Ci j  C i j 
where
(i) i = i = i or j = j  = j  or Gij = Gi j  = Gi j  ,
(ii) Nij = Ni j  = 2,
(iii) Pij = Pi j  ; Pij ∩ Pi j  = ∅,
and the attribute P i j  of C i j  is equal to Pi j  − Pij .
Rule 3 (Third order Simplification Rule) If three cells in the same row (col-
umn or grid) have the same set of possible numbers and its cardinality is 3, then
those numbers can be removed from the sets of possible numbers of every other cell
 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93 85

in the same row (column or grid). Actually, the rule can be slightly generalized by
allowing the cardinality of the sets to be 2 for some of the cells but one:
Cij Ci j  Ci j  Ci j 
Cij Ci j  Ci j  C i j 
where
(i) i = i = i = i or j = j  = j  = j  or Gij = Gi j  = Gi j  = Gi j  ,
(ii) Nij = 3; 2 ≤ Ni j  , Ni j  ≤ 3,
(iii) Pi j  , Pi j  ⊆ Pij ; Pij ∩ Pi j  = ∅,
and the attribute P i j  of C i j  is equal to Pi j  − Pij .
Further simplification rules could be added, but they are not needed and actually
matching is much more expensive for them.
Rule 4 (Only One Number Rule) When a number is not possible in any cell
of a row (column or grid) but one, and the cardinality of the set of possible numbers
for this cell is greater than one, then this set can become a singleton set containing
that number. Symbolically,
Ci1 j1 {Cik jk }2≤k≤n
C i1 j1 {Cik jk }2≤k≤n
where:
(i) There exists a number N , 1 ≤ N ≤ n, such that N = ik or N = jk or
N = Gik jk for all 1 ≤ k ≤ n,
(ii) Ni1 j1 > 1,
 c
(iii) There exists a number p, 1 ≤ p ≤ n, such that p ∈ Pi1 j1 ∩ 2≤l≤n Pil jl ,

and the attribute P i1 j1 of C i1 j1 is equal to {p}.

Rule 5 (Only Two Numbers Rule) When two numbers p1 and p2 are not pos-
sible in any cell of a row (column or grid) but two, and the sets of possible numbers
for these cells have cardinality greater than two, then these sets can become {p1 , p2 }.
Ci1 j1 Ci2 j2 {Cik jk }3≤k≤n
C i1 j1 C i2 j2 {Cik jk }3≤k≤n
where:
(i) There exists a number N , 1 ≤ N ≤ n, such that N = ik or N = jk or
N = Gik jk for all 1 ≤ k ≤ n,
(ii) Ni1 j1 , Ni2 j2 > 2,
(iii) 
There exist two
 c numbers p1 , p2 , 1 ≤ p1 , p2 ≤ n, such that p1 , p2 ∈ Pi1 j1 ∩ Pi2 j2 ∩
3≤l≤n Pil jl ,

and the attributes P i1 j1 and P i2 j2 of C ik jk are both equal to {p1 , p2 }.

Rule 6 (Twin Rule) If, in a given grid, a number is only possible in one row (or
column), then that number can be removed from the set of possible numbers in all
86 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93

the cells in that same row (or column) but different grid.
Ci0 j0 {Cik jk }1≤k≤n
C i0 j0 {Cik jk }1≤k≤n
where:
(i) There exists a number N , 1 ≤ N ≤ n, such that N = Gik jk , for all 1 ≤ k ≤ n,
(ii) i0 = ik and j0 = jk for all 4 ≤ k ≤ n,
(iii) i0 = i1 or j0 = j1 ,
 c
(iv) There exists a number p, 1 ≤ p ≤ n, such that p ∈ Pi0 j0 ∩Pi1 j1 ∩ 4≤l≤n Pil jl ,

and the attribute P i0 j0 of the representation of C i0 j0 is equal to Pi0 j0 − {p}.

Rule 7 (General Sudoku Split Rule) This rule splits a sudoku when none of
the other rules can be applied. We select a cell with a minimum number (greater
than 1) of possible numbers. Then a sudoku is created with the first possible number
and another one with the remaining possible numbers:
Cij {Ckl }k=i,l=j
   ,
C ij {Ckl }k=i,l=j C ij {Ckl }k=i,l=j
if these conditions are fulfilled:
(i) Nij ≥ 2 and Nij is minimal (but greater than 1),
(ii) p ∈ Pij ,
and the attribute P ij of C ij is equal to {p} and the attribute P ij of C ij is Pij − {p}.
Rule 8 (Sudoku Split Rule) This rule is the particular case of the previous one
when the number of possible numbers in a cell is equal to 2.
The correctness of each rule is immediate by their definition. Its application,
however, does not lead to a confluent system: in case a sudoku has several solutions,
the one that is reached can vary according to the order of execution of the rules;
when a sudoku has a unique solution, this is obtained. The system is terminating:
even though Rule 7 introduces a new sudoku, every rule decreases the cardinal of
the set of possible numbers in one of the cells.

4 Rewriting logic and Maude

Maude [8,9] is a high performance language and system supporting both equational
and rewriting logic computation for a wide range of applications. The key novelty of
Maude is that besides efficiently supporting equational computation and algebraic
specification it also supports rewriting logic computation.
A rewrite theory is a four-tuple T = (Ω, E, L, R), where (Ω, E) is a theory in an
equational logic, L is a set of labels for the rules, and R is the set of labeled rewrite
rules axiomatizing the local state transitions of the system. Some of the rules in R
may be conditional. Mathematically, a rewrite rule has the form l : t −→ t if C,
with t, t terms of the same kind which may contain variables. Intuitively, a rule
 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93 87

describes a local concurrent transition in a system: anywhere where a substitution

instance σ(t) of t is found, a local transition of that state fragment to the new local
state σ(t ) can take place. We can regard an equational theory as the special case
of a rewrite theory in which the sets of labels L and rules R are both empty; in this
way, equational logic appears naturally as a sublanguage of rewriting logic.
Maude is a language whose modules are theories in rewriting logic. The most
general Maude modules are called system modules and are written as mod T endm,
with T the rewrite theory in question expressed with a syntax quite close to the
corresponding mathematical notation. The equations E in the equational theory
(Ω, E) underlying the rewrite theory T = (Ω, E, L, R) are presented as a union
E = A ∪ E  , with A a set of equational axioms introduced as attributes of certain
operators in the signature Ω—for example, an operator + can be declared asso-
ciative and commutative with keywords assoc and comm—and where E  is a set
of equations that are assumed to be Church-Rosser and terminating modulo the
axioms A. Maude supports rewriting modulo different combinations of such equa-
tional attributes: operators can be declared associative, commutative, with identity,
and idempotent.

4.1 Specifying sudokus in Maude

Our sudoku system module is called SUDOKU.

mod SUDOKU is

We first import into it some predefined modules that define the natural num-
bers, quoted identifiers, and string and number conversion. Maude provides useful
support for modularity by allowing the definition of module hierarchies; a module
can import other Maude modules as submodules in different modes, in this case
with the keyword protecting (which can be abbreviated to pr):
pr NAT . pr QID . pr CONVERSION .

The framework we have proposed for solving sudokus constitutes an example of

an object-based system [9, Chapter 8]. Maude supports the specification of such sys-
tems in a simple and direct way through a predefined module called CONFIGURATION.
However, instead of using this module, and since we want our output to be format-
ted in a tabular manner and to use different colours, we will explicitly declare all the
operators needed for the specification of such systems; we will use them to introduce
Maude syntax.
As described in the previous section, the cells in a sudoku correspond to
objects and we will use messages to show the current status of the puzzle:
SearchingSolution, NoSolution, FinalSolution. A “soup” of such objects and
messages is called a configuration and will be used to represent sudokus. Thus, we
declare sorts for objects and messages, which are subsorts of configurations, as well
as for object and class identifiers (called Oid and Cid).
sorts Oid Cid Object Msg Configuration .
subsort Object Msg < Configuration .

To create objects we introduce, with the keyword op, an operator < : | > that
88 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93

takes an object identifier, a class identifier, and a set of attributes as arguments.

op <_:_|_> : Oid Cid AttributeSet -> Object
[ctor object format (n r! o g o tm! ot d)] .

The underbars allow the specification of mixfix syntax and are placeholders where
the arguments should be written. (The commands enclosed in the brackets simply
specify the format of the output [9, Chapter 4].)
Configurations are declared with an operator with empty syntax (__) which is
associative, commutative, and has an identity element.
op none : -> Configuration .
op __ : Configuration Configuration -> Configuration
[ctor config assoc comm id: none] .

The current state of the solving procedure will be represented as a term of sort
Sudoku, which consists of a set of sudokus each one enclosed in double angles:
sort Sudoku .
op none : -> Sudoku .
op <<_>> : Configuration -> Sudoku .
op __ : Sudoku Sudoku -> Sudoku [ctor config assoc comm id: none] .

Attributes of objects belong to a new sort Attribute, which is a subsort of

AttributeSet built with _,_.
sorts Attribute AttributeSet .
subsort Attribute < AttributeSet .
op none : -> AttributeSet .
op _,_ : AttributeSet AttributeSet -> AttributeSet
[ctor assoc comm id: none format (m! o sm! o)] .

For our sudoku specification we need to declare the three attributes associated
to each cell: the grid to which it belongs, the set of possible numbers that may
be placed in it, and its cardinality; the column and row will be given by the cell
identifier. Attributes are declared as operators that return a term of sort Attribute.
op grd :_ : Nat -> Attribute .
op pss :_ : Set -> Attribute .
op num :_ : Nat -> Attribute .
op id : Nat Nat -> Oid . --- row and column

All these declarations specify the core syntax needed to represent sudokus and
our solving procedure in Maude. Now, their behaviour is specified by means of
equations and rules. Equations are declared using the keyword eq and variables
with the keyword var. For example, the operator that returns the grid associated
to the cell at column C and row R can be specified as
op grd : Nat Nat -> Nat .
vars R C : Nat .
eq grd(R, C) = (sd(R, 1) quo 3) * 3 + (sd(C, 1) quo 3) + 1 .

where sd and quo are respectively the operators for symmetric difference and integer
quotient.
An initial board for a sudoku can be specified as a constant term sudoku. To
avoid the cumbersome task of explicitly writing the complete representation of a
sudoku (9 × 9 cells with their corresponding rows, columns, numbers, grids, . . . , for
a sudoku of order 9), we use several auxiliar operators that will transform a much
closer representation of a sudoku into the object-based format. For example, for
the sudoku in Figure 1:
op sudoku : -> Sudoku .
eq sudoku =
 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93 89

<< msg(’SearchingSolution)
fill(1, 1,
( 0 ; 0 ; 3 ; 7 ; 2 ; 0 ; 1 ; 0 ; 0 ;
0 ; 0 ; 6 ; 9 ; 0 ; 5 ; 8 ; 0 ; 0 ;
4 ; 9 ; 0 ; 0 ; 1 ; 0 ; 0 ; 5 ; 2 ;
0 ; 5 ; 0 ; 0 ; 0 ; 0 ; 0 ; 6 ; 1 ;
8 ; 0 ; 4 ; 0 ; 0 ; 0 ; 2 ; 0 ; 9 ;
7 ; 6 ; 0 ; 0 ; 0 ; 0 ; 0 ; 4 ; 0 ;
2 ; 3 ; 0 ; 0 ; 5 ; 0 ; 0 ; 1 ; 7 ;
0 ; 0 ; 1 ; 2 ; 0 ; 3 ; 9 ; 0 ; 0 ;
0 ; 0 ; 5 ; 0 ; 8 ; 6 ; 4 ; 0 ; 0 )
) >> .

Here, for example, the auxiliary operator fill places the givens in their corre-
sponding cells, making thus their set of possible numbers a singleton, whereas the
remaining cells will have {1, 2, 3, 4, 5, 6, 7, 8, 9} as their set of possible numbers. The
sort List is an auxiliary sort over which lists of natural numbers are constructed
using the operator _;_ in the usual way [9, Chapter 7].
op fill : Nat Nat List -> Object .
eq fill(R, C, (N ; LL)) =
if C == 9 then
< id(R, C) : cell | grd : grd(R, C), pss : pss(N),
num : num(N) >
fill(s R, 1, LL)
else < id(R, C) : cell | grd : grd(R, C), pss : pss(N),
num : num(N) >
fill(R, s C, LL) fi .
eq fill(R, C, N) =
< id(R, C) : cell | grd : grd(R, C), pss : pss(N),
num : num(N) > .

The operators together with the equations define the static part of the system.
Next we need to define its dynamics, the way the system evolves towards reaching a
solution, by means of rules that capture the processes for solving sudokus explained
in Section 3.1. The goal, of course, is to find a solution whenever one exists (the
set of possible numbers becomes a singleton for every cell) or otherwise to return
a message warning about its non-existence (some set of possible numbers becomes
empty).
Each of our rewrite rules diminishes the number of elements in the set of possible
numbers of some cell, in a way that faithfully mimics the presentation of the solving
rules given in Section 3.2. We illustrate the naturalness with which they are written
in Maude by presenting two of them; for the rest, we refer the reader to the files at
http://maude.sip.ucm.es/~miguelpt/bibliography.
The Sudoku Split rule that splits a sudoku into two when the number of possible
numbers in a cell is equal to two, creating a sudoku with the first possible number
and another with the second, is represented by means of
var VConf : Configuration .
rl [sudokuSplit2] :
<< msg(’SearchingSolution)
VConf < id(R, C) : cell | grd : G, pss : (P1 P2), num : 2 > >>
=> << msg(’SearchingSolution)
VConf < id(R, C) : cell | grd : G, pss : P1, num : 1 > >>
<< msg(’SearchingSolution)
VConf < id(R, C) : cell | grd : G, pss : P2, num : 1 > >> .

This rule can be applied provided that a Sudoku term has a cell with just two
elements P1 P2 in its set of possible numbers. In this case the term will be rewritten
to a term with two concatenated Sudoku terms as expected: one with P1 as its set
of possible numbers and another with P2.
90 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93

As another example, and to illustrate the use of conditional rewriting rules,

let us consider the Second order Simplification rule: if two cells in the same row
(column or grid) have the same set of possible numbers and its cardinality is 2, then
those numbers can be removed from the sets of possible numbers of every other cell
in the same row (column or grid):
crl [simplify2nd] :
< id(R1, C1) : cell | grd : G1, pss : (P P’), num : 2 >
< id(R2, C2) : cell | grd : G2, pss : (P P’), num : 2 >
< id(R3, C3) : cell | grd : G3, pss : (P LP3), num : N3 >
=>
< id(R1, C1) : cell | grd : G1, pss : (P P’), num : 2 >
< id(R2, C2) : cell | grd : G2, pss : (P P’), num : 2 >
< id(R3, C3) : cell | grd : G3, pss : LP3, num : sd(N3,1) >
if ((R1 == R2) and (R1 == R3)) or ((C1 == C2) and (C1 == C3))
or ((G1 == G2) and (G1 == G3)) .

The remaining rules for solving sudokus are represented in a similar manner.
In addition to the rules for solving a sudoku, there are also a number of equations
and rules to take care of “maintenance” issues: to finish the procedure when there
is no possible solution and to stop the application of rules when the final solution
has been reached. For example, we have found a solution and therefore can stop the
solving procedure if both the maximum and the minimum cardinality of the sets of
possible numbers (computed by the auxiliary operators maxCard and minCard) are
equal to 1:
ceq << msg(’SearchingSolution) VConf >> =
<< msg(’FinalSolution) VConf >>
if (maxCard(VConf) == 1) and (minCard(VConf) == 1) .

Finally, to show the final term (a solved sudoku) in the desired format, a new
attribute is added to the objects representing cells which is assigned the list of all
the values in a given row:
op val‘:_ : List -> Attribute .
eq msg(’FinalSolution)
< id(R,1) : cell | pss: N1, At1 > < id(R,2) : cell | pss : N2, At2 >
< id(R,3) : cell | pss: N3, At3 > < id(R,4) : cell | pss : N4, At4 >
< id(R,5) : cell | pss: N5, At5 > < id(R,6) : cell | pss : N6, At6 >
< id(R,7) : cell | pss: N7, At7 > < id(R,8) : cell | pss : N8, At8 >
< id(R,9) : cell | pss: N9, At9 >
= msg(’FinalSolution)
< id(R,0) : rows | val: (N1 ; N2 ; N3 ; N4 ; N5 ; N6 ; N7 ; N8 ; N9) > .

Then, when a solution exists, the final term has the following form:
<< msg(’FinalSolution)
< id(1, 0) : rows | val : (2 ; 9 ; 5 ; 7 ; 4 ; 3 ; 8 ; 6 ; 1) >
< id(2, 0) : rows | val : (4 ; 3 ; 1 ; 8 ; 6 ; 5 ; 9 ; 2 ; 7) >
< id(3, 0) : rows | val : (8 ; 7 ; 6 ; 1 ; 9 ; 2 ; 5 ; 4 ; 3) >
< id(4, 0) : rows | val : (3 ; 8 ; 7 ; 4 ; 5 ; 9 ; 2 ; 1 ; 6) >
< id(5, 0) : rows | val : (6 ; 1 ; 2 ; 3 ; 8 ; 7 ; 4 ; 9 ; 5) >
< id(6, 0) : rows | val : (5 ; 4 ; 9 ; 2 ; 1 ; 6 ; 7 ; 3 ; 8) >
< id(7, 0) : rows | val : (7 ; 6 ; 3 ; 5 ; 2 ; 4 ; 1 ; 8 ; 9) >
< id(8, 0) : rows | val : (9 ; 2 ; 8 ; 6 ; 7 ; 1 ; 3 ; 5 ; 4) >
< id(9, 0) : rows | val : (1 ; 5 ; 4 ; 9 ; 3 ; 8 ; 6 ; 7 ; 2) > >>

4.2 Running the sudokus

The previous section has described how a procedure for solving sudokus can be
specified in Maude. As discussed in Section 3.1, however, one cannot simply apply
the rewrite rules in it to obtain a solution lest a combinatorial explosion is produced.
Then, in order to avoid this it will be necessary to apply the rules according to some
 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93 91

suitable strategy.
Maude has been extended with a powerful strategy language that allows a user
to specify the ways in which the rewrite rules in a module will be applied [11].
This language is itself built on top of another extension, called Full Maude (see [9])
which adds to Maude a richer set of primitives for working with parameterization
and object-oriented modules. We will not use any of the advanced features of Full
Maude and for our purposes it will be enough to know that to load a module it
is necessary to enclose it in parentheses. Regarding the strategy language, we will
limit ourselves to a very limited subset; we refer the reader to [11] for the complete
details.
A strategy module is declared with the keyword stratdef, strategy operators
with sop, and the equations that define these operators are introduced with seq:
(stratdef STRAT is
sop rules .
seq rules = (simplify1st orelse
(simplify2nd orelse
(onlyOneNumber orelse
(simplify3rd orelse
(onlyTwoNumbers orelse
twins))))) .
sop split .
seq split = (sudokuSplit2 orelse sudokuSplitN) .
sop solve .
seq solve = (rules orelse split) ! .
endsd)

This module defines three strategies. The first one, rules, simply tries to apply
one of the first six rules, in order: it tries with simplify1st; if it is not possi-
ble, it tries with simplify2nd; and so on. The second one, split, tries to apply
sudokuSplit and sudokuSplitN rules in order. The last one, solve, applies the
first strategy and, only if it is not possible, tries to rewrite using splitting rules;
the bang ! at the end asks to continue with the application of the strategy while
possible.
To illustrate its use, after loading Maude with Full-Maude and the module with
the strategies we can solve the sudoku in Figure 1 by means of:
Maude> (srew sudoku using solve .)
rewrite with strategy :
result Sudoku :
<< msg(’FinalSolution)
< id(1,0): rows | val :(5 ; 8 ; 3 ; 7 ; 2 ; 4 ; 1 ; 9 ; 6) >
< id(2,0): rows | val :(1 ; 2 ; 6 ; 9 ; 3 ; 5 ; 8 ; 7 ; 4) >
< id(3,0): rows | val :(4 ; 9 ; 7 ; 6 ; 1 ; 8 ; 3 ; 5 ; 2) >
< id(4,0): rows | val :(3 ; 5 ; 9 ; 8 ; 4 ; 2 ; 7 ; 6 ; 1) >
< id(5,0): rows | val :(8 ; 1 ; 4 ; 5 ; 6 ; 7 ; 2 ; 3 ; 9) >
< id(6,0): rows | val :(7 ; 6 ; 2 ; 3 ; 9 ; 1 ; 5 ; 4 ; 8) >
< id(7,0): rows | val :(2 ; 3 ; 8 ; 4 ; 5 ; 9 ; 6 ; 1 ; 7) >
< id(8,0): rows | val :(6 ; 4 ; 1 ; 2 ; 7 ; 3 ; 9 ; 8 ; 5) >
< id(9,0): rows | val :(9 ; 7 ; 5 ; 1 ; 8 ; 6 ; 4 ; 2 ; 3) > >>

5 Final remarks
We have presented in this paper a case study of how to use Maude to execute
and solve sudokus. We have first shown a representation of sudokus in an object-
oriented way and introduced rules for solving them, whose implementation in Maude
is straightforward. Since the blind application of these rules would give rise to
a combinatorial explosion, we have explained how to take advantage of Maude’s
92 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93

strategy language to apply the rules in a non-expensive way.

The main strength of our approach is the naturalness with which sudokus are
represented in Maude and the ease with which the solving procedure is imple-
mented. The specification can be easily modified to deal with sudokus of arbitrary
order just by extending the rules with additional objects to represent the extra
cells. However, it is precisely this need to add additional subterms to the rules
which prevents the use of a single specification to cover all orders; such specification
could be written by resorting to Maude’s metalevel [9], but that would make the
specification more obscure. To illustrate how the extension works, the maude files
for the specification to solve sudoku monsters (sudokus of order 4), along with the
complete specification of the sudoku solver presented in the paper, are available at
http://maude.sip.ucm.es/~miguelpt.
On the other hand, the weakness of our implementation lies in its efficiency.
Even with the use of strategies to prune the search tree, our implementation cannot
compete with the numerous solvers available in the web (e.g. [5,6,7]).

Acknowledgement
We thank Alberto Verdejo and Narciso Martı́-Oliet for helpful comments and dis-
cussions (specially about the strategy language), and the anonymous referees for
their suggestions.

References
[1] History of Dell Magazines. http://www.dellmagazines.com/.

[2] Rules and history from the Nikoli website. http://www.nikoli.co.jp/puzzles/

[3] The History of Sudoku, Conceptis Editoria. http://www.conceptispuzzles.com/articles/sudoku/

[4] Sudoku Solver, Scanraid Ltd. http://www.scanraid.com/AdvanStrategies.htm

[5] SuDoku Solver by Logic v1.4. http://www.sudokusolver.co.uk/ . (Javascript)

[6] Sudokulist. http://www.sudokulist.com/. (Step by step solver, with hints at each step.)

[7] A Su Doku solver. http://sudoku.sourceforge.net/ . (Java)

[8] Manuel Clavel, Francisco Durán, Steven Eker, Patrick Lincoln, Narciso Martı́-Oliet, José Meseguer and
José F. Quesada. Maude: Specification and Programming in Rewriting Logic, Theoretical Computer
Science, 285(2):187–243, 2002.

[9] Manuel Clavel, Francisco J. Durán, Steven Eker, Patrick Lincoln, Narciso Martı́-Oliet, José Meseguer,
and Carolyn Talcott. Maude Manual (version 2.2). http://maude.cs.uiuc.edu, 2006.

[10] Bertram Felgenhauer and Frazer Jarvis, Enumerating possible Sudoku grids, Univ. Sheffield and
Univ. Dresden, 2005 http://www.shef.ac.uk/~ pm1afj/sudoku/sudoku.pdf

[11] Narciso Martı́-Oliet, José Meseguer and Alberto Verdejo. Towards a strategy language for Maude. In N.
Martı́-Oliet, editor, Proceedings Fifth International Workshop on Rewriting Logic and its Applications,
WRLA 2004, volume 117 of Electronic Notes in Theoretical Computer Science. Elsevier, 2004.

[12] Michael Mepham. Solving Sudoku, Daily Telegraph (2005).

[13] José Meseguer, Conditional Rewriting Logic as a Unified Model of Concurrency, Theoretical Computer
Science, 96(1):73–155, 1992.
 G. Santos-García, M. Palomino / Electronic Notes in Theoretical Computer Science 176 (2007) 79–93 93

[14] Miguel Palomino, Narciso Martı́-Oliet, and Alberto Verdejo. Playing with Maude. In Slim Abdennadher
and Christophe Ringeissen, editors, Fifth International Workshop on Rule-Based Programming, RULE
2004, volume 124 of Electronic Notes in Theoretical Computer Science. Elsevier, 2005.

[15] Takayuki Yato and Takahiro Seta. Complexity and Completeness of Finding Another Solution and Its
Application to Puzzles. IEICE Trans. Fundamentals, E86-A (5):1052–1060, 2003.