Re: [Reduce-algebra-developers] Issue with num_min

[Rainer S. <rai...@gm...>]

The accuracy a must be a positive integer, interpreted as 10^(-a), but this is
not checked in the code. I will commit a correction with an additional error
check in a moment.

  Rainer


On Fri, 4 Apr 2025 at 22:32 +0100, Arthur Norman via Reduce-algebra-develop...:

 > Until this evening if you passed non-numeric values such as
 > (!:rd!: . 1.0e-36) to expt in CSL it crashed out in an embarassing way, but I
 > have now made that function check its args a bit better. I hit this when
 > trying to do a numerical minimization, and sometimes that seems to pass a
 > Reduce wrapped up float not a number. Well also the value 10^1.0e-36 looks a
 > pretty odd thing to be calculating! I have not (yet) looked inside the code
 > to understand what is going on - this is posted so that others can comment
 > too....
 >
 >
 > 1: on backtrace;
 >
 > 2: num_min(cos x^2 + cos y^2, {x=1, y=2}, accuracy=1.0e-36, iterations=100);
 > +++ Error: expt 10 (!:rd!: . 1.0e-36)
 > Inside: steepdeceval1
 > Arg1: (plus (expt (cos x) 2) (expt (cos y) 2))
 > Arg2: (x y)
 > Arg3: (1 2)
 > Arg4: num_min
 > Inside: rdmineval
 > Arg1: ((plus (expt (cos x) 2) (expt (cos y) 2)) (list (equal x 1) (equal y
 > 2)) (
 > equal accuracy (!:dn!: 10 . -37)) (equal iterations 100))
 > Inside: lispapply
 > ...
 >
 > Arthur
 >
 >
 >
 > _______________________________________________
 > Reduce-algebra-developers mailing list
 > Red...@li...
 > https://lists.sourceforge.net/lists/listinfo/reduce-algebra-developers
 >

   Rainer Schöpf