theory Sat
imports "../Appendix" "../First_Steps"
begin
section {* SAT Solvers\label{rec:sat} *}
text {*
{\bf Problem:}
You like to use a SAT solver to find out whether
an Isabelle formula is satisfiable or not.\smallskip
{\bf Solution:} Isabelle contains a general interface for
a number of external SAT solvers (including ZChaff and Minisat)
and also contains a simple internal SAT solver that
is based on the DPLL algorithm.\smallskip
The SAT solvers expect a propositional formula as input and produce
a result indicating that the formula is either satisfiable, unsatisfiable or
unknown. The type of the propositional formula is
@{ML_type "Prop_Logic.prop_formula"} with the usual constructors such
as @{ML And in Prop_Logic}, @{ML Or in Prop_Logic} and so on.
The function @{ML Prop_Logic.prop_formula_of_term} translates an Isabelle
term into a propositional formula. Let
us illustrate this function by translating @{term "A \<and> \<not>A \<or> B"}.
The function will return a propositional formula and a table. Suppose
*}
ML %grayML{*val (pform, table) =
Prop_Logic.prop_formula_of_term @{term "A \<and> \<not>A \<or> B"} Termtab.empty*}
text {*
then the resulting propositional formula @{ML pform} is
@{ML [display] "Or (And (BoolVar 1, Not (BoolVar 1)), BoolVar 2)" in Prop_Logic}
where indices are assigned for the variables
@{text "A"} and @{text "B"}, respectively. This assignment is recorded
in the table that is given to the translation function and also returned
(appropriately updated) in the result. In the case above the
input table is empty (i.e.~@{ML Termtab.empty}) and the output table is
@{ML_response_fake [display,gray]
"Termtab.dest table"
"[(Free (\"A\", \"bool\"), 1), (Free (\"B\", \"bool\"), 2)]"}
An index is also produced whenever the translation
function cannot find an appropriate propositional formula for a term.
Attempting to translate @{term "\<forall>x::nat. P x"}
*}
ML %grayML{*val (pform', table') =
Prop_Logic.prop_formula_of_term @{term "\<forall>x::nat. P x"} Termtab.empty*}
text {*
returns @{ML "BoolVar 1" in Prop_Logic} for @{ML pform'} and the table
@{ML table'} is:
@{ML_response_fake [display,gray]
"map (apfst (Syntax.string_of_term @{context})) (Termtab.dest table')"
"(\<forall>x. P x, 1)"}
In the print out of the tabel, we used some pretty printing scaffolding
to see better which assignment the table contains.
Having produced a propositional formula, you can now call the SAT solvers
with the function @{ML "SatSolver.invoke_solver"}. For example
@{ML_response_fake [display,gray]
"SatSolver.invoke_solver \"dpll\" pform"
"SatSolver.SATISFIABLE assg"}
determines that the formula @{ML pform} is satisfiable. If we inspect
the returned function @{text assg}
@{ML_response [display,gray]
"let
val SatSolver.SATISFIABLE assg = SatSolver.invoke_solver \"dpll\" pform
in
(assg 1, assg 2, assg 3)
end"
"(SOME true, SOME true, NONE)"}
we obtain a possible assignment for the variables @{text "A"} and @{text "B"}
that makes the formula satisfiable.
Note that we invoked the SAT solver with the string @{text [quotes] dpll}.
This string specifies which SAT solver is invoked (in this case the internal
one). If instead you invoke the SAT solver with the string @{text [quotes] "auto"}
@{ML [display,gray] "SatSolver.invoke_solver \"auto\" pform"}
several external SAT solvers will be tried (assuming they are installed).
If no external SAT solver is installed, then the default is
@{text [quotes] "dpll"}.
There are also two tactics that make use of SAT solvers. One
is the tactic @{ML sat_tac in SAT}. For example
*}
lemma "True"
apply(tactic {* SAT.sat_tac @{context} 1 *})
done
text {*
However, for proving anything more exciting using @{ML "sat_tac" in SAT} you
have to use a SAT solver that can produce a proof. The internal
one is not usuable for this.
\begin{readmore}
The interface for the external SAT solvers is implemented
in @{ML_file "HOL/Tools/sat_solver.ML"}. This file contains also a simple
SAT solver based on the DPLL algorithm. The tactics for SAT solvers are
implemented in @{ML_file "HOL/Tools/sat.ML"}. Functions concerning
propositional formulas are implemented in @{ML_file
"HOL/Tools/prop_logic.ML"}. The tables used in the translation function are
implemented in @{ML_file "Pure/General/table.ML"}.
\end{readmore}
*}
end