On Wed, 28 Oct 2009 at 04:40PM +0100, Thierry Dumont wrote: > I have looked at the quadrature routines: > > -in gsl: it seems that qags routine is called: this is a sophisticated > procedure with step adaptation, and convergence acceleration with the > epsilon-algorithm. This should integrate some singular functions and > discontinuous functions. > -in scipy: things look not so sophisticated, but it's difficult to > understand what is really used! > > *All* this is from quadpack (you can find quadpack on netlib), > transcripted to C for the gsl. The gsl routine is certainly expensive, > but robust. This is more or less the most "universal" method (but a > universal method of integration cannot exist...), certainly too > powerfull in many cases... I would recommend to keep it.
Let me also add mpmath's quad routine here. A while back, I was doing some numerical integrals. The standard gsl routine was a bit slow, and also had a huge memory leak (I posted about this, but I can't find it). I had much better success with mpmath. If we're reworking our numerical integration, I think we should consider adding in mpmath's quad, since it seems pretty good. Dan -- --- Dan Drake ----- http://mathsci.kaist.ac.kr/~drake -------
signature.asc
Description: Digital signature
