[Ecls-list] quad_qawf nondeterministic

[Robert Dodier]

On 2014-04-24, Dan Gildea <gildea at users.sourceforge.net> wrote:

> If I run the following command at the maxima command line several
> times in a row, I usually get the first result below, but
> sometimes the second:
>
> (%i46) ev( quad_qawf (exp (-u), u, au, omega, sin, limit=32), au=0, omega=2);
>
> (%o46) [0.4000000000000001,22.165709488159253e-12,175,0]
> (%i47) ev( quad_qawf (exp (-u), u, au, omega, sin, limit=32), au=0, omega=2);
>
> (%o47) quad_qawf(%e^-u,u,0,2,sin,epsabs = 10.e-11,limit = 32,maxp1 = 100,
>                  limlst = 10)

Confirmed w/ ECL 11.1.1 (yes, I know it's out of date) + current Maxima
from Git. Tracing SLATEC:DQAWF shows that its last argument is an array
which seems to contain garbage; it has a lot of very large and very
small values, and sometimes NaN. I'm guessing those values cause DQAWF
to fail.

With Clisp and SBCL, that array contains all nil values and all zeros,
respectively.

best

Robert Dodier