Hello Robert,
I have committed the changes to get the expected pure real or imaginary results
and added some tests to rtest_gamma.mac. The changes are done for float and
bigfloat precision.
Remark:
When %iargs is true the trigonometric functions simplify like
sin(%i) -> %i*sinh(1) or
cos(%i) -> cosh(1)
In principle something similiar for the Error functions can be done:
erf(%i) -> %i*erfi(1)
erfi(%i) -> %i*erf(1)
I have not implemented this simplifications. Perhaps this simplifications get
more useful if we have more code in sinint or specint which uses the Error
functions more intense.
Dieter Kaiser
-----Urspr?ngliche Nachricht-----
Von: robert.dodier at gmail.com [mailto:robert.dodier at gmail.com]
Gesendet: Dienstag, 23. Dezember 2008 07:10
An: Dieter Kaiser
Cc: maxima at math.utexas.edu
Betreff: Re: [Maxima] Maxima 5.17.1 regressions for Sage
On 12/22/08, Dieter Kaiser <drdieterkaiser at web.de> wrote:
> Thus, I do not think that we have errors in Maxima, but Sage get unexpected
> results for the erf function with complex arguments.
>
> Remark: We have small realparts in the results. This is due to the limited
> numerical accuracy of the routine.
It seems reasonable that if the result is known to be be real for all
arguments in some set of arguments, that a computed numerical
approximation should likewise be real. Can't the numerical function
just return realpart(whatever) when the argument is known to yield
a real result?
best
Robert