Skip to content

A Reduced-order Binary Decision Diagram (RoBDD) SAT solver written in Rust

Notifications You must be signed in to change notification settings

timbeurskens/rsbdd

Repository files navigation

RsBDD

Rust

Solving satisfiability problems in Rust

Installation

  1. Make sure to install the Rust toolchain.

  2. Clone the latest version of this repository:

$ git clone [email protected]:timbeurskens/rsbdd.git
  1. Build and install the RsBDD tools:
$ cd rsbdd
$ cargo install --bins --path .

The following tools will be available after installing the RsBDD package:

  • max_clique_gen
  • n_queens_gen
  • random_graph_gen
  • rsbdd
  • sudoku_gen

Syntax

Comments

Characters contained within "..." (excluding the " char itself) are regarded as comments and can be placed at any point in the formula.

Constants

The most basic building blocks of the syntax are 'variables' and 'constants'. A constant can be either 'true' or ' false'. A variable can accept either a 'true' or 'false' value after evaluation depending on its environment.

true
false

Variables

A variable is a single word starting with a non-digit character. Examples of good variable names are:

a
a'
alpha
_x
a1
hello_world

Negation

A variable, constant, or sub-formula can be negated using the negation operator. This operator can be expressed by either !, -, or not.

not true
-false
!variable

Binary operators

RsBDD supports the most common, and some uncommon binary operators, such as conjunction, disjunction, implication and bi-implication.

Most operators have a symbolic and textual representation, e.g. and or &.

Operator Option 1 Option 2
Conjunction and &
Disjunction or |
Implication implies or in =>
Bi-implication iff or eq <=>
Exlusive or xor ^
Joint denial nor N.A.
Alternative denial nand N.A.
true or false
true | false
a | b
a & b
a and b
a => b
hello <=> world
on ^ off

Composition

Larger formulae can be composed using left and right parentheses: (, ):

a | (a & b)
(a)
((a))
!(a & b)
(a & b) | (b & c)

If-then-else

A simplification of a common expression (a => b) & ((!a) => c) can be made using the ternary if-then-else (ite) operator.

if a then b else c
if exists a # a <=> b then b <=> c else false | c

Quantifiers

The RsBDD supports universal and existential quantification using the exists and forall/all keywords: {forall|exists} var_1, var_2, .., var_n # {subformula}

forall a # true
forall a # a | b
forall a, b # exists c # (c | a) & (c | b)

Counting

For some problems it can be beneficial to express properties relating to the number of true or false variables, e.g. "at least 2 of the 4 properties must hold".

The counting operator ([]) in combination with five new equality and inequality operators (=, <=, >=, <, >) can be used to concisely express these properties.

Note: like most operators, the counting operator can be expressed using logic primitives, but this operator simplifies the expression significantly.

A counting comparison can either be made by comparing a set of expressions to a given constant, or an other set of expressions.

"exactly one of a, b, and c holds"
[a, b, c] = 1

"there are strictly less true expressions in a, b, c than d, e, f"
[a, b, c] < [d, e, f]

Counting comparison also allows us to specify optimization problems. Example: the max-clique problem can be described as a clique problem, such that for all satisfiable cliques, the reported result is the largest.

-(a & f) &
-(a & g) &

-(b & d) &
-(b & e) &

-(c & e) &
-(c & g) &

forall _a,_b,_c,_d,_e,_f,_g # (
    -(_a & _f) &
    -(_a & _g) &
    -(_b & _d) &
    -(_b & _e) &
    -(_c & _e) &
    -(_c & _g)
) => [a,b,c,d,e,f,g] >= [_a,_b,_c,_d,_e,_f,_g]

Fixed points

The rsbdd language supports least-fixpoint (lfp / mu) and greatest-fixpoint (gfp / nu) operations to find a respectively minimal or maximal solution by repeatedly applying a given transformer function until the solution is stable.

Only monotonic transformer functions are guaranteed to terminate. Termination of fixed point operations are not checked and will run indefinatedly if not handled correctly.

Its basic properties are defined as follows.

gfp X # X           <=> true
lfp X # X           <=> false

nu X # ...          <=> gfp X # ...
mu X # ...          <=> lfp X # ...

gfp/lfp X # a       <=> a
gfp/lfp X # true    <=> true
gfp/lfp X # false   <=> false

Parse-tree display

Adding the -p {path} argument to rsbdd constructs a graphviz graph of the parse-tree. This can be used to for introspection of the intended formula, or for reporting purposes. An example of the parse-tree output for exists b,c # a | (b ^ c) is displayed below.

parse tree

Experimental and/or upcoming features

Currently the RsBDD language relies heavily on logical primitives. Integer arithmetic could be expressed by manually introducing the primitive 'bits' of a number. Rewrite rules could significantly simplify this process by introducting domains other than boolean variables. Embedding rewrite rules in the BDD could prove to be a challenge.

Examples

Example 1: transitivity of the >= operator

([a1,a2,a3,a4] >= [b1,b2,b3,b4] & [b1,b2,b3,b4] >= [c1,c2,c3,c4]) => [a1,a2,a3,a4] >= [c1,c2,c3,c4]

Example 2: the 4 queens problem

The famous n-queens problem can be expressed efficiently in the RsBDD language. The example below shows a 4-queens variant, which can be solved in roughly 15 milliseconds. The library contains a generator for arbitrary n-queens problems. At this point, the largest verified problem size is n=8, which reports all solutions in less than 20 minutes on modern hardware. The explosive nature of the problem makes n=9 an infeasable problem. Further optimizations (such as multi-processor parallellism, or vertex ordering) could decrease the run-time in the future.

"every row must contain exactly one queen"
[_0x0, _0x1, _0x2, _0x3] = 1 &
[_1x0, _1x1, _1x2, _1x3] = 1 &
[_2x0, _2x1, _2x2, _2x3] = 1 &
[_3x0, _3x1, _3x2, _3x3] = 1 &

"every column must contain exactly one queen"
[_0x0, _1x0, _2x0, _3x0] = 1 &
[_0x1, _1x1, _2x1, _3x1] = 1 &
[_0x2, _1x2, _2x2, _3x2] = 1 &
[_0x3, _1x3, _2x3, _3x3] = 1 & 

"every diagonal must contain at most one queen"
[_0x0] <= 1 &
[_0x1, _1x0] <= 1 &
[_0x2, _1x1, _2x0] <= 1 &
[_0x3, _1x2, _2x1, _3x0] <= 1 &
[_1x3, _2x2, _3x1] <= 1 &
[_2x3, _3x2] <= 1 &
[_3x3] <= 1 &

"the other diagonal"
[_0x3] <= 1 &
[_0x2, _1x3] <= 1 &
[_0x1, _1x2, _2x3] <= 1 &
[_0x0, _1x1, _2x2, _3x3] <= 1 &
[_1x0, _2x1, _3x2] <= 1 &
[_2x0, _3x1] <= 1 &
[_3x0] <= 1

Running this example with the following arguments yields a truth-table showing the queen configuration(s) on a 4x4 chess board.

rsbdd -i examples/4_queens.txt -t -ft
_0x0 _0x1 _0x2 _0x3 _1x0 _1x1 _1x2 _1x3 _2x0 _2x1 _2x2 _2x3 _3x0 _3x1 _3x2 _3x3 *
False False True False True False False False False False False True False True False False True
False True False False False False False True True False False False False False True False True

CLI Usage

rsbdd

A BDD-based SAT solver

Usage: rsbdd [OPTIONS] [FILE]

Arguments:
  [FILE]  The input file containing a logic formula in rsbdd format

Options:
  -p, --parsetree <PARSETREE>            Write the parse tree in dot format to the specified file
  -t, --truthtable                       Print the truth table to stdout
  -d, --dot <DOT>                        Write the bdd to a dot graphviz file
  -m, --model                            Compute a single satisfying model as output
  -v, --vars                             Print all satisfying variables leading to a truth value
  -f, --filter <FILTER>                  Only show true or false entries in the output [default: Any]
  -c, --retain-choices <RETAIN_CHOICES>  Only retain choice variables when filtering [default: Any]
  -b, --benchmark <N>                    Repeat the solving process n times for more accurate performance reports
  -g, --plot                             Use GNUPlot to plot the runtime distribution
  -e, --evaluate <EVALUATE>              Parse the formula as string
  -o, --ordering <ORDERING>              Read a custom variable ordering from file
  -r, --export-ordering                  Export the automatically derived ordering to stdout
  -h, --help                             Print help
  -V, --version                          Print version

max_clique_gen

Converts a graph into a max-clique specification

Usage: max_clique_gen [OPTIONS] [INPUT] [OUTPUT]

Arguments:
  [INPUT]   Input file graph in csv edge-list format
  [OUTPUT]  The output rsbdd file

Options:
  -u, --undirected  Use undirected edges (test for both directions in the set-complement operation)
  -a, --all         Construct a satisfiable formula for all cliques
  -h, --help        Print help
  -V, --version     Print version

random_graph_gen

Generates a random edge list formatted graph

Usage: random_graph_gen [OPTIONS] [VERTICES] [EDGES]

Arguments:
  [VERTICES]  The number of vertices in the output graph
  [EDGES]     The number of edges in the output graph

Options:
  -o, --output <FILE>   The output filename (or stdout if not provided)
  -u, --undirected      Use undirected edges (test for both directions in the set-complement operation)
      --complete        Construct a complete graph
  -d, --dot             Output in dot (GraphViz) format
      --convert <FILE>  If this argument is provided, the provided edge-list will be used to generate a graph
  -c, --colors <N>      Generate a graph-coloring problem with N colors
  -h, --help            Print help
  -V, --version         Print version

n_queens_gen

Generates n-queen formulae for the SAT solver

Usage: n_queens_gen [OPTIONS] [OUTPUT]

Arguments:
  [OUTPUT]  The output rsbdd file

Options:
  -n, --queens <QUEENS>  The number of queens [default: 4]
  -h, --help             Print help
  -V, --version          Print version

sudoku_gen

Generates a random edge list formatted graph

Usage: sudoku_gen [OPTIONS] [INPUT] [OUTPUT]

Arguments:
  [INPUT]   The input sudoku file
  [OUTPUT]  The output rsbdd file

Options:
  -r, --root <N>  The root value of the puzzle. Typically the square root of the largest possible number [default: 3]
  -h, --help      Print help
  -V, --version   Print version