Le 17/07/2010 09:22, Martin Rubey a écrit :
William Stein<wst...@gmail.com> writes:http://wstein.org/talks/stein-sd24/stein-sd24.pdf
Reading this document, I was, among many things, very interested by the 2012 horizon (Sage 7.0) and specially by the Engineering projects.
I am interested by this point and would like to contribute, if possible. My remarks and questions:*During the long time I spent with P. Zimmermann and others writing our book, I tested a large part of the numerical methods. Some things are lacking, others cannot actually be used, and do not provide an alternative to Mat*b. But all this can be improved, and the Python basis is very nice.
Examples:1) linear algebra: the support for sparse matrices cannot be used in real problems: we want to solve large (say 10^5) systems of equations. The lil_matrix class of Scipy is so slow, that it is a pity. 2) ODEs: the support provided by scipy is very basic, and insufficient at least for me. Everything is too old!
I have mostly looked at these points. *Question: will (most of) Sage numerical methods remain based on Scipy? *How I think we can improve points 1) and 2):1) sparse matrices are created using lil_matrix, based on dictionaries. I personally use (out of Sage) a C++ class based on std::map which performs some orders faster. It could be integrated either directly in Sage, or in Scipy.
2) For the ODEs, the best and the most *modern* methods are there: http://www.unige.ch/~hairer/software.htmlH. Hairer, G. Wanner, C. Lubich are among the best specialist of numerical methods for ODE and their programs (even if written in f77) are wonderful, and very robust and efficient. Writing an interface may not be a so large task (there is already a Matlab interface) and we would have the best for stiff systems, Hamiltonian systems (make celestial mechanics with Sage) and so on... This would be in the spirit of Sage, I think: use the best software available, written by true specialists.
But may be there are already projects (for point 2)) ? let me know... Thierry Dumont. -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
<<attachment: tdumont.vcf>>
