[Maxima] use of $csign

[Barton Willis]

I tried changing sign-log to

(defun sign-log (x)
  (setq x (cadr x)) ;; looking at sign of log(x)
  (cond ((eq t (meqp x 1)) (setf sign '$zero)) ;; log(1) = 0.
 ;; for x in unit circle and x # 1, log(x) is pure imaginary
 ((and  *complexsign* (eq 1 (meqp 1 (take '(mabs) x))) (mnqp x 1))
  (setf sign '$imaginary))
 ;; log(x) is positive for x > 1
 ((eq t (mgrp x 1)) (setf sign '$pos))
 ;; log(x) is negative for 0 < x < 1.
 ((and (eq t (mgrp x 0)) (eq t (mgrp 1 x))) (setf sign '$neg))
 ;; when *complexsign* is true, return $complex, else $pnz.
 (*complexsign* (setf sign '$complex))
 (t (setf sign '$pnz)))
  sign)

The testsuite gets through with a few errors--some integrals in
rtest_integrate
(#270 for example) fail--actually, with my sign-log, test #270 generates a
huge expression that might be correct but it contains many terms such as
atan2(0,h). Changing

  (*complexsign* (setf sign '$complex))

to

  (*complexsign* (setf sign '$pnz))

makes the testsuite more quiet.

What is the meaning of the defmvars minus, evens, and odds in compar.lisp?



Barton

-----maxima-bounces at math.utexas.edu wrote: -----

>To: Barton Willis <willisb at unk.edu>
>From: Dieter Kaiser <drdieterkaiser at web.de>
>Sent by: maxima-bounces at math.utexas.edu
>Date: 09/16/2009 04:29PM
>cc: "maxima at math.utexas.edu" <maxima at math.utexas.edu>
>Subject: Re: [Maxima] use of $csign
>
>Am Mittwoch, den 16.09.2009, 07:24 -0500 schrieb Barton Willis:
>> Should csign(log(x)) --> complex or pnz? More generally, if csign
>returns
>> complex,
>> does this mean that the expression is non-real for all inputs or
>non-real
>> for some
>> inputs?
>
>Hello Barton,
>
>your comments are right.
>
>The handling of functions by $csign is not complete and consistent. More
>work is possible and necessary.
>
>Now, all functions are assumed to be real valued for a real or a complex
>argument.
>
>(%i1) csign(f(x));
>(%o1)                                 pnz
>(%i2) declare(z,complex);
>(%o2)                                done
>(%i3) csign(f(z));
>(%o3)                                 pnz
>
>But not in all cases as you have observed for the log function:
>
>(%i8) csign(log(x));
>(%o8)                                 pnz
>(%i9) csign(log(z));
>(%o9)                               complex
>
>Now, we have to declare the symbol for the function to be complex to get
>a complex answer:
>
>(%i10) declare(f,complex)$
>
>(%i11) csign(f(x));
>(%o11)                              complex
>
>The time I have worked on the complex components I have hesitated to
>change the behavior that every function is assumed to have real values.
>The problem is, that a lot of known results will change. Furthermore I
>have not seen in fully depth all relations between the different
>routines for the complex components which might cause subtle bugs.
>
>But I think we should improve the complex behavior step by step to get
>more nice results.
>
>By the way: I think it is important too to improve the sqrt function,
>see the bugs sqrt(1/x) -> 1/sqrt(x) and sqrt(%i/x) -> -%i/sqrt(x), to
>get more correct results for complex expressions.
>
>Dieter Kaiser
>
>
>_______________________________________________
>Maxima mailing list
>Maxima at math.utexas.edu
>http://www.math.utexas.edu/mailman/listinfo/maxima