[boost] Re: Numerical Integration

From: Geoff Leyland (geoff.leyland_at_[hidden])
Date: 2002-10-25 08:00:55

Hi,

I've just seen all the stuff that's been going on discussing 
integration.  I don't have a lot to say about integration itself (to me 
it seems that the problems of integrating a function f(x) and 
integrating a series of data only available at fixed x's are quite 
different, and it's probably best not to try to impose the same 
interface on both of them, but I'm hardly an expert), however there was 
some talk of ODE solving.

I'm surprised that boosters have an interest in this, but if so...

I'd be VERY interested in such a library.  I have aging code 
implementing a number of methods (none of which are particularly 
special), but it's not really what's been done that's interesting, more 
the insight I got into how I should have done it.
What's more I may have a simulation-type project coming up which would 
need both stiff solvers and interpolation within a step (as opposed to 
the runge-kutta, and I think bullrisch-stoer methods I used before).
As far as I can tell, most codes exist already in FORTRAN, so perhaps 
all that is needed is defining an interface to these methods (which I am 
begining to think should be based around using a "step" as the basic 
object), or whether, as some of these codes are (to my not very FORTRAN 
familiar eyes) not very easy to read, whether it wouldn't be worth 
re-writing some of the simpler methods in c++.

So, if there's any interest in ODEs, perhaps there's a need to 
talk/argue about what the interfaces should look like, and then about 
what methods could be re-written in c++ (remembering that we have ublas 
now) and what methods could be implemented with interfaces to existing 
codes.

cheers,
goof

--
"One hundred percent of the shots you don't take don't go in."
- Wayne Gretzky
Geoff Leyland, Village Idiot
Laboratoire d'energetique industrielle
LENI-DGM-EPFL, CH-1015, Lausanne, Switzerland
Phone: [phone], Fax: [phone]