Re: [OctDev] complex error function

[Steven G. Johnson]

Daniel J Sebald wrote:
> Have you sent someone in command of that library an email to inquire?
> [...]
> Because it hasn't been accepted into the standard library doesn't mean 
> it can't be used as an extension.

Not only would it be an extension (which they RARELY accept...it took 
them years to even accept commonplace and desperately needed extensions 
like qsort_r), but it would potentially (at least in theory) conflict 
with future C standards (since the name is reserved but not yet 
defined).  Not only are they almost certain to reject such a 
contribution, I would actually agree with such a rejection: they should 
NOT be implementing reserved C99 identifiers based on "guessed" meanings 
(no matter how reasonable the guess).

>  It sure would be nice though, if it 
> were incorporated, to be just like "erf()" in the C++ code, but instead 
> "cerf()".

(C++ does not support erf for complex arguments either.)

> The other part of this is that I would really like to see the algorithm 
> evaluated before considering for inclusion in the core of Octave.  I'm 
> sure there are at least a few people on this list capable of doing that, 
> but I wonder if you are better served by opening this to a more 
> algorithm-oriented audience or publication in SIAM or some similar journal.

It mainly implements a combination of two previously published 
algorithms (using one or the other algorithm in different regions of the 
complex plane), with a few minor twists, so there is probably not enough 
novelty at this point to merit a paper in a good journal.

However, I have corresponded with the author of one of the algorithms I 
implemented to get his feedback, along with a couple of other 
colleagues, and the SciPy people did their own testing as well.

Of course, I'd be very happy to get any constructive feedback on the 
algorithm or code from the Octave folks as well.  Certainly, you will 
want to do your own testing.

--SGJ