Test suite failures with cvs version?



Hello Ray,

yes, you are right. It is better when the numerical routine expintegral-si
itself would work correctly. With your improvements I will have a second look
through the code.

Next I would like to implement the more standardized way of calling the
numerical routines. Then a critical view on the accuracy of the results and the
consistent calcuation of all the related functions including gamma_incomplete,
erf, ... is necessary.

Dieter Kaiser

-----Urspr?ngliche Nachricht-----
Von: raymond.toy at ericsson.com [mailto:raymond.toy at ericsson.com] 
Gesendet: Freitag, 24. Oktober 2008 23:44
An: Dieter Kaiser
Cc: maxima at math.utexas.edu
Betreff: Re: AW: AW: [Maxima] Test suite failures with cvs version?

Dieter Kaiser wrote:
> Hello Ray,
> 
> that is the code in simp-expintegral-si which calls expintegral-si:
> 
>       ((expintegral-numerical-eval-p arg)
>        (let ((carg (complex ($realpart arg) ($imagpart arg))))
>          (complexify (expintegral-si carg))))
> 
> Here a CL complex number is generated and transferred to the numerical routine
> expintegral-si. I am not sure, but I think this would be the best place to
make
> sure that the argument carg has the correct declaration. Is a declaration at
> this point enough to make sure that the numerical routines work as expected?

Declarations don't do anything.  But if you meant that we make sure carg
is a complex flonum, then yes, this would work.

But now, I think making expintegral-si do the right thing might be
better, in case some one later says, hey, here's a Si function I can
use, then passes a fixnum to it and wonders why the precision isn't what
was desired.

> 
> Later I would like to use a different test which I use to standardize the
> numerical evaluation a bit. In other functions I have written code like this:
> 
>       ((complex-float-numerical-eval-p arg)
>        (let ((carg (complex ($float ($realpart arg)) ($float ($imagpart
arg)))))
>          (complexify (expintegral-si carg))))
> 
> The functions float-numerical-eval-p, complex-numerical-eval-p, .. I have
> written to get a standard for a call to the numerical routines. Would the
> additional $float guarantee that the argument carg has the correct and
complete
> declaration?

Yes, this should work.

> 
> This is the equivalent code for the Exponential Integral Ci in
> simp-expintegral-ci which calls the numerical routine expintegral-ci. Again a
CL
> complex number is transferred to the numerical routine.
> 
>       ((expintegral-numerical-eval-p arg)
>        (let ((carg (complex ($realpart arg) ($imagpart arg))))
>          (complexify (expintegral-ci carg))))
> 

This needs to be changed too.  Interestingly, the testsuite didn't
complain about Ci not having enough precision, but the Si test did.
That's how I found the problem with expintegral-si.

Ray