rationaltree_exist.pl : Rational tree with exist. quantifiers CHR solver for existentially quantified conjunctions of non-flat equations overs trees with quadratic complexity. The algorithm includes three parts: * Transforms the original equations into equivalent existentially quantified flat equations. * Applies the classic rationaltree solver (with term-order where existentially quantified variables are smaller than free variables). * The reachability of variables and equations is analysed, unreachable ones are removed.
How to use: Existentially quantified variables are identified by exists(V) where V is a list of variables. This must be the first constraint. Equality between trees A and B is written as A eq B. To trigger elimination of unreachable variables and equations the last constraint must be elim. In the answer the constraints exists_s(V) and A eq_s B (for solved) are used.
Program: Change the code, then submit! /* rationaltree_exist.pl: Rational tree with exist. quantifiers (C) Marc.Meister at uni-ulm.de, 2006/10 (C) Khalil.Djelloul at uni-ulm.de, 2006/10 (C) Thom.Fruehwirth at uni-ulm.de, 1993, 1998, 2006/10 This program is distributed under the terms of the GNU General Public License: http://www.gnu.org/licenses/gpl.html %% DESCRIPTION CHR solver for existentially quantified conjunctions of non-flat equations overs trees with quadratic complexity.# The algorithm includes three parts:# * Transforms the original equations into equivalent existentially quantified flat equations.# * Applies the classic rationaltree solver (with term-order where existentially quantified variables are smaller than free variables).# * The reachability of variables and equations is analysed, unreachable ones are removed.# %% HOW TO USE Existentially quantified variables are identified by *exists(V)* where V is a list of variables. This must be the first constraint.# Equality between trees A and B is written as *A eq B.*# To trigger elimination of unreachable variables and equations the last constraint must be *elim.*# In the answer the constraints *exists_s(V)* and *A eq_s B* (for solved) are used.# %% SAMPLE QUERIES % note that we quantify *all* variables: % the answer "yes" tells us that the problem has a solution Q: exists([X,Y,Z,V]), f(X) eq f(g(X,Y)), Z eq f(V), Z eq f(f(Y)), elim. A: yes. % compared to the first example, variables X and V are now free: % note our clear and explicit answer: % exists Y: V eq f(Y), X eq g(X,Y) with free variables V and X Q: exists([Y,Z]), f(X) eq f(g(X,Y)), Z eq f(V), Z eq f(f(Y)), elim. A: exists_s(Y), V eq_s f(Y), X eq_s g(X,Y). Q: exists([X,Y,U]), h(Y,f(a),g(X,a)) eq h(f(U),Y,g(h(Y),U)), elim. A: yes. Q: f(X) eq X, X eq f(f(X)), elim. A: X eq_s f(X). Q: a eq b, elim. A: no. */ :- module(rationaltree_exist, [exists/1, eq/2, exists_s/1, eq_s/2, elim/0]). :- use_module(library(chr)). :- op(700,xfx,[eq,eq_f,eq_s]). %% Deprecated syntax used for SICStus 3.x %handler rationaltree_exist. %constraints %% Syntax for SWI / SICStus 4.x :- chr_constraint exists/1, % INPUT: varlist is existentially quantified eq/2, % INPUT: equality between terms elim/0, % INPUT: EOL, triggers reachability exists_s/1, % OUTPUT: reachable exist. quantified var eq_s/2, % OUTPUT: reachable flat equality constraint eq_f/2, % INTERNAL: flat equality constraint reach/1, % INTERNAL: reachable variable id/2, % INTERNAL: id(V,(Type,Nr)): Type={exist,free,func} nextid/1, % INTERNAL: next number to use as variable id reg_var/2, % INTERNAL: reg_var(V,exist) or reg_var(V,free) reg_func/1. % INTERNAL: reg_func(F) %%% MINIMAL FLATTENING %% convert equations to flattened form %% A eq B (possibly non-flat equation) converted to form A eq_f B (flat) % Meanwhile variables and functions occuring in the equations are registered: % reg_var(X,exist): register X as existentially quantified variable % reg_var(X,free): register X as free variable (happens to most variables) % reg_func(F): register flat function F, so form is either 'a' or 'f(A,B,C)' % An exist. quantified var is always first registered as reg_var(X,exist), % so a later registration of the same as reg_var(X,free) is ignored! flat_vv @ X eq Y <=> var(X), var(Y) | reg_var(X,free), reg_var(Y,free), X eq_f Y. flat_tv @ T eq X <=> nonvar(T), var(X) | X eq T. flat_tt @ T1 eq T2 <=> nonvar(T1), nonvar(T2) | reg_var(X,exist), X eq T1, X eq T2. flat_vt @ X eq T <=> var(X), nonvar(T) | reg_var(X,free), T=..[F|L], flatten_list(L,L0), T0=..[F|L0], reg_func(T0), X eq_f T0. % flatten_list(L1,L2): flatten terms and introduce new vars and equations % e.g., L1=[f(a),b,X] ==> L2=[_A,_B,X], _A eq f(a), _B eq b flatten_list([], []). flatten_list([H|T],[H|T0]) :- var(H), reg_var(H,free), flatten_list(T,T0). flatten_list([H|T],[X|T0]) :- nonvar(H), reg_var(X,exist), X eq H, flatten_list(T,T0). %%% RATIONAL TREE %% solve rational tree (flat equations) reflexivity @ X eq_f X <=> var(X) | true. orientation @ id(X,IX), id(T,IT) \ T eq_f X <=> var(X), lss(IX,IT) | X eq_f T. decomposition @ T1 eq_f T2 <=> nonvar(T1), nonvar(T2) | same_functions(T1,T2). confrontation @ id(X,IX), id(T1,IT1), id(T2,IT2), X eq_f T1 \ X eq_f T2 <=> var(X), lss(IX,IT1), leq(IT1,IT2) | T1 eq_f T2. % functions must be equal same_functions(T1,T2) :- T1=..[F|L1], T2=..[F|L2], eq_list(L1,L2). % two lists must have same length and elements must be equal eq_list([],[]). eq_list([H1|T1],[H2|T2]) :- H1 eq_f H2, eq_list(T1,T2). %%% ELIMINATION OF UNREACHABLE VARIABLES AND EQUATIONS %% reachable exist. quantified variables and equations are marked with _s %% (meaning solution: exist_s, eq_s); all others are removed %% triggered by elim/0. elim \ id(X,(free,_)) <=> reach(X). elim, reach(X) \ X eq_f T <=> nonvar(T) | X eq_s T, T=..[_|L], reach_list(L). elim, reach(X) \ X eq_f Y <=> var(Y) | X eq_s Y, reach(Y). elim, reach(X) \ id(X,(exist,_)) <=> exists_s(X). %% cleanup elim \ id(_,_) <=> true. elim \ _ eq_f _ <=> true. elim \ reach(_) <=> true. elim \ nextid(_) <=> true. elim <=> true. % all vars in list are reachable reach_list([]). reach_list([H|T]) :- reach(H), reach_list(T). %%% AUXILIARY %% registering variables and functions, and testing for order % For each variable and function from the flat equations one ID is created: % id(X,ID): X is the term, ID=(Type,Nr), Type is one of the following 3 types: % id(X,(exist,N)): X is an existentially quantified variable with number N % id(X,(free,N)): X is a free variable with number N % id(X,(func,N)): X is a function with termdepth N % The variable numbers are unique and assigned sequentially. % exists(Varlist): must be entered before equations, registers vars exists([]) <=> true. exists([V|L]) <=> reg_var(V,exist), exists(L). % register flat function: form is either 'a' or 'f(A,B)', so depth is 1 or 2 reg_func(T) <=> atom(T) | id(T,(func,1)). reg_func(T) <=> compound(T) | id(T,(func,2)). %% register var: exist. quantified (id(V,exist)) or free var (id(V,free)) % only first registration for each variable is effective, later is ignored; % so e.g., exists(X), reg_var(X,free). ==> reg_var(X,exist). id(V,_) \ reg_var(V,_) <=> true. % nextid(N) constraint determines number for variable reg_var(V,T), nextid(N) <=> id(V,(T,N)), N1 is N+1, nextid(N1). % nextid/1 doesn't yet exist, take 0 as first var number reg_var(V,T) <=> id(V,(T,0)), nextid(1). %% term order %% to compare variables and functions the corresponding IDs %% must be passed to lss/2 or leq/2 % lss(X,Y): X is smaller than Y, iff % type of X is 'smaller' than type of Y (exist < free < func) or % X and Y have same type and number for X is smaller than number for Y % (for variables the number is sequentially assigned to all used variables, % for functions the number is the term-depth) lss((exist,_),(free,_)). lss((exist,_),(func,_)). lss((free,_),(func,_)). lss((T,N), (T,M)) :- N < M. % leq(X,Y): X is smaller or equal to Y, iff % X is smaller than Y (lss(X,Y)) or % IDs of X and Y are identical leq(IX,IY) :- lss(IX,IY). leq((T,N),(T,N)).
Console: Enter query or select example from below, then submit and wait for answer! % loading solvers/rationaltree_exist.pl | ?- consult(...). yes [2.558 seconds] | ?-