[SciPy-user] exponential integral behaviour near -20

[Robert Kern]

On Mon, Dec 15, 2008 at 22:58, David Cournapeau
<david at ar.media.kyoto-u.ac.jp> wrote:
> Robert Kern wrote:
>>
>> I cannot replicate this on OS X with a recent(ish) SVN scipy compiled
>> with gfortran 4.2.1. This function is implemented in FORTRAN, so the
>> FORTRAN compiler used may be relevant.
>
> Isn't exp1 implemented from exp1n in cephes  ?

No, there is no exp1n (or exp1m) actually defined in our Cephes
sources. SPECFUN's E1Z (for the complex input) is the implementation
that is used.

>>  Can you give us more
>> information about the platform, scipy version, and FORTRAN compiler?
>
> FWIW, I can replicate the reported behavior on Linux (x86 - gfortran
> 4.3.2 + recent SVN for both numpy and scipy as well).

Hmm. I dunno. Perhaps gfortran's runtime changed a branch cut for
CDLOG() between 4.2 and 4.3. That would explain the +/- pi*I in the
output.

-- 
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless
enigma that is made terrible by our own mad attempt to interpret it as
though it had an underlying truth."
  -- Umberto Eco