new program for GSL
Ewgenij Hübner
huebner@uni-trier.de
Fri May 28 17:38:00 GMT 2004
Hallo,
I am a PhD-Student at the University Trier (Germany). In the context of my
dissertation I am working with variation inequalities. For my purpose I am
implemented an algorithm for minimization of convex non-differentiable
function (min{f(x): x in R^n}) and I would like to supply this program to
GSL.
I implemented the Kiviel's bundle algorithm (Ref.: "Proximity control in
bundle methods for convex nondifferentiable minimization. Math. Programming
46 (1990)"). I tested other algorithm too (e.g."A version of the bundle idea
for minimizing a nonsmooth function: conceptual idea, convergence analysis,
numerical results. SIAM J. Optim. 2 (1992) "), but the algorithm of Kiviel
seems to be very robust and, from computational point of view, the best for
the praxis.
In each step of the bundle method one have to solve a convex quadratic
problem. For this problem I use the primal-dual interior point method of
Mehrotra. I implemented this algorithm in a separate program und one can use
it to solve the arbitrary problem of type min{xQx : Ax=b, Cx>=d}
or min{xQx: Ax=b, x>=0} (Q positive semidefinit).
(Ref.: "Wright, Stephen J. Primal dual interior point methods")
I send two tar-files with both programs. In these files you can find source
files und Makefiles. I hope you are interested for this. If it is possible to
include the programs in GSL, I will write documentation and test the programs
much more.
Wit best regards
Ewgenij Huebner
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bundle_method.tar.gz
Type: application/x-tgz
Size: 14309 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/gsl-discuss/attachments/20040528/6a5f73b4/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cqp.tar.gz
Type: application/x-tgz
Size: 10104 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/gsl-discuss/attachments/20040528/6a5f73b4/attachment-0001.bin>
More information about the Gsl-discuss
mailing list