Prolog - simplify derivative

问题:

so I just got started with Prolog this semester, and got the homework to implement a pretty basic d(function, variable, derivative) which I did like this:

d(X,X,1) :- !.
d(C,X,0) :- atomic(C). %, (C \= X).
d(X**E,X,E*X**(E-1)).
d(U+V,X,A+B) :- d(U,X,A), d(V,X,B).
d(U-V,X,A-B) :- d(U,X,A), d(V,X,B).
d(U*V,X,DU*V+U*DV) :- d(U,X,DU), d(V,X,DV).
d(U/V,X,(DU*V-U*DV)/(V*V)) :- d(U,X,DU), d(V,X,DV).

I know this is not complete, but it covers all the tasks required in the exercise.

However, ?- d((x*x+2*x+3)/(3*x),x,R). leads to

R = ((1*x+x*1+ (0*x+2*1)+0)* (3*x)- (x*x+2*x+3)* (0*x+3*1))/ (3*x* (3*x)). which doesn't look pretty at all. is/2 unfortunately doesn't like my x as it is not a number...

Is there a simple solution to achieve a cleaner result?

回答1:

I would rather see this as two separate problems:

First, get derivation right (you're probably getting close, depending on your concrete requirements).

Then, work on simplifying expressions on an algebraic level. Exploit algebraic identities, see if applying the laws of commutativity / associativity / distributivity on some subexpressions enable their rewriting into something equivalent (but simpler / more compact).

As a starting point, you may want to look at the somewhat related question "Replacing parts of expression in prolog".

---

Here's a simplistic sketch how you could do the simplification—using iwhen/2 to safeguard against insufficient instantiation:

expr_simplified(A, B) :-
   iwhen(ground(A), xpr_simplr(A,B)).

xpr_simplr(A, B) :-
   (  atomic(A)
   -> A = B
   ;  ( A = X+0 ; A = 0+X ; A = 1*X ; A = X*1 )
   -> xpr_simplr(X, B)
   ;  ( A = 0*_ ; A = _*0 )
   -> B = 0
   ;  A = X+X
   -> B = X*2
   ;  A = X*X
   -> B = X**2
   ;  A = X**1
   -> B = X
   ;  A =.. [F|Xs0],                          % defaulty catch-all
      maplist(xpr_simplr, Xs0, Xs),
      B =.. [F|Xs]
   ).

Let's see what it does with the expression you gave. We apply expr_simplified/2 until we reach a fixed point:

?- A = ((1*x+x*1+(0*x+2*1)+0)*(3*x)-(x*x+2*x+3)*(0*x+3*1))/(3*x*(3*x)),
   expr_simplified(A,B),
   expr_simplified(B,C),
   expr_simplified(C,D).
A = ((1*x+x*1+(0*x+2*1)+0)*(3*x)-(x*x+2*x+3)*(0*x+3*1))/(3*x*(3*x)),
B = ((x+x+(0+2))*(3*x)-(x**2+2*x+3)*(0+3))/(3*x)**2,
C = ((x*2+2)*(3*x)-(x**2+2*x+3)*3)/(3*x)**2,
D = C.                                        % fixed point reached

As imperfect as the simplifier is, the expression got a lot more readable.

回答2:

a possibility to get a number is to replace each instance of variable x with a value, visiting the derived tree. You should do writing a clause to match each binary operator, or use a generic visit, like

set_vars(E, Vs, Ev) :-
    E =.. [F,L,R],
    set_vars(L, Vs, Lv),
    set_vars(R, Vs, Rv),
    Ev =.. [F,Lv,Rv].
set_vars(V, Vs, N) :- memberchk(V=N, Vs).
set_vars(V, _, V).

that yields

?- d((x*x+2*x+3)/(3*x),x,R), set_vars(R,[x=5],E), T is E.
R = ((1*x+x*1+ (0*x+2*1)+0)* (3*x)- (x*x+2*x+3)* (0*x+3*1))/ (3*x* (3*x)),
E = ((1*5+5*1+ (0*5+2*1)+0)* (3*5)- (5*5+2*5+3)* (0*5+3*1))/ (3*5* (3*5)),
T = 0.29333333333333333 

but, there is a bug in your first clause, that once corrected, will allow to evaluate directly the derived expression:

d(X,V,1) :- X == V, !.
...

now, we can throw away the utility set_vars/3, so

?- d((T*T+2*T+3)/(3*T),T,R), T=8, V is R.
T = 8,
R = ((1*8+8*1+ (0*8+2*1)+0)* (3*8)- (8*8+2*8+3)* (0*8+3*1))/ (3*8* (3*8)),
V = 0.3177083333333333.