Z3 Solver中Tactic的使用

Z3求解器是微软开发的一款定理证明器。可以用来证明定理，也可以用于求解。其基本用法并不复杂，通过现有的教程，结合源码中提供的示例就可以理解。除了基础的用法以外，Z3还提供了一系列Tactic辅助计算和证明。

本文将介绍Tactic的一些用法，以及部分Tactic的效果。本文总体基于Z3的Python API来写。本文的基本材料来自Z3的API Documentation以及代码中提供的C++示例，以及根据以上资料进行的实验。本文使用的Z3版本是4.8.0。

Tactic 简介

什么是 Tactic

Tactic在Z3中可以对一组约束（Goal）进行形式转换、求解或简化，便于对约束进行进一步的处理。

Tactic 的基本使用方法

>>> t = Tactic("simplify")
>>> x = Int('x')
>>> c = x > 0
>>> t(c)
[[Not(x <= 0)]]

可用的 Tactic 列表

本列表通过describe_tactics()获得所有可用的Tactic类型及其描述。

ackermannize_bv : A tactic for performing full Ackermannization on bv instances.
subpaving : tactic for testing subpaving module.
horn : apply tactic for horn clauses.
horn-simplify : simplify horn clauses.
nlsat : (try to) solve goal using a nonlinear arithmetic solver.
qfnra-nlsat : builtin strategy for solving QF_NRA problems using only nlsat.
nlqsat : apply a NL-QSAT solver.
qe-light : apply light-weight quantifier elimination.
qe-sat : check satisfiability of quantified formulas using quantifier elimination.
qe : apply quantifier elimination.
qsat : apply a QSAT solver.
qe2 : apply a QSAT based quantifier elimination.
qe_rec : apply a QSAT based quantifier elimination recursively.
sat : (try to) solve goal using a SAT solver.
sat-preprocess : Apply SAT solver preprocessing procedures (bounded resolution, Boolean constant propagation, 2-SAT, subsumption, subsumption resolution).
ctx-solver-simplify : apply solver-based contextual simplification rules.
smt : apply a SAT based SMT solver.
psmt : builtin strategy for SMT tactic in parallel.
unit-subsume-simplify : unit subsumption simplification.
aig : simplify Boolean structure using AIGs.
add-bounds : add bounds to unbounded variables (under approximation).
card2bv : convert pseudo-boolean constraints to bit-vectors.
degree-shift : try to reduce degree of polynomials (remark: :mul2power simplification is automatically applied).
diff-neq : specialized solver for integer arithmetic problems that contain only atoms of the form (<= k x) (<= x k) and (not (= (- x y) k)), where x and y are constants and k is a numeral, and all constants are bounded.
eq2bv : convert integer variables used as finite domain elements to bit-vectors.
factor : polynomial factorization.
fix-dl-var : if goal is in the difference logic fragment, then fix the variable with the most number of occurrences at 0.
fm : eliminate variables using fourier-motzkin elimination.
lia2card : introduce cardinality constraints from 0-1 integer.
lia2pb : convert bounded integer variables into a sequence of 0-1 variables.
nla2bv : convert a nonlinear arithmetic problem into a bit-vector problem, in most cases the resultant goal is an under approximation and is useul for finding models.
normalize-bounds : replace a variable x with lower bound k <= x with x’ = x - k.
pb2bv : convert pseudo-boolean constraints to bit-vectors.
propagate-ineqs : propagate ineqs/bounds, remove subsumed inequalities.
purify-arith : eliminate unnecessary operators: -, /, div, mod, rem, is-int, to-int, ^, root-objects.
recover-01 : recover 0-1 variables hidden as Boolean variables.
bit-blast : reduce bit-vector expressions into SAT.
bv1-blast : reduce bit-vector expressions into bit-vectors of size 1 (notes: only equality, extract and concat are supported).
bv_bound_chk : attempts to detect inconsistencies of bounds on bv expressions.
propagate-bv-bounds : propagate bit-vector bounds by simplifying implied or contradictory bounds.
propagate-bv-bounds-new : propagate bit-vector bounds by simplifying implied or contradictory bounds.
reduce-bv-size : try to reduce bit-vector sizes using inequalities.
bvarray2uf : Rewrite bit-vector arrays into bit-vector (uninterpreted) functions.
dt2bv : eliminate finite domain data-types. Replace by bit-vectors.
elim-small-bv : eliminate small, quantified bit-vectors by expansion.
max-bv-sharing : use heuristics to maximize the sharing of bit-vector expressions such as adders and multipliers.
blast-term-ite : blast term if-then-else by hoisting them.
cofactor-term-ite : eliminate term if-the-else using cofactors.
collect-statistics : Collects various statistics.
ctx-simplify : apply contextual simplification rules.
der : destructive equality resolution.
distribute-forall : distribute forall over conjunctions.
dom-simplify : apply dominator simplification rules.
elim-term-ite : eliminate term if-then-else by adding fresh auxiliary declarations.
elim-uncnstr : eliminate application containing unconstrained variables.
injectivity : Identifies and applies injectivity axioms.
snf : put goal in skolem normal form.
nnf : put goal in negation normal form.
occf : put goal in one constraint per clause normal form (notes: fails if proof generation is enabled; only clauses are considered).
pb-preprocess : pre-process pseudo-Boolean constraints a la Davis Putnam.
propagate-values : propagate constants.
reduce-args : reduce the number of arguments of function applications, when for all occurrences of a function f the i-th is a value.
reduce-invertible : reduce invertible variable occurrences.
simplify : apply simplification rules.
elim-and : convert (and a b) into (not (or (not a) (not b))).
solve-eqs : eliminate variables by solving equations.
split-clause : split a clause in many subgoals.
symmetry-reduce : apply symmetry reduction.
tseitin-cnf : convert goal into CNF using tseitin-like encoding (note: quantifiers are ignored).
tseitin-cnf-core : convert goal into CNF using tseitin-like encoding (note: quantifiers are ignored). This tactic does not apply required simplifications to the input goal like the tseitin-cnf tactic.
fpa2bv : convert floating point numbers to bit-vectors.
qffp : (try to) solve goal using the tactic for QF_FP.
qffpbv : (try to) solve goal using the tactic for QF_FPBV (floats+bit-vectors).
qffplra : (try to) solve goal using the tact