How to use lapack in matlb to solve large eigenvalue problem
6 vues (au cours des 30 derniers jours)
Afficher commentaires plus anciens
Dear all,
If the matrices K and M are so larger, the computation in matlab will take a long time. How ro solve this problem. Somebody tell me use lapack is ok? is there some tutorials for the use lapack in matlab. I only want to solve thie problem in lapack, and the eigenblued and bector will send back to matlb fot postprocessing.
0 commentaires
Réponses (2)
James Tursa
le 6 Déc 2012
If you are well versed in the LAPACK functions and just need an interface, you can use this package from the FEX by Tim Toolan:
4 commentaires
Jan
le 7 Déc 2012
Matlab calls optimized LAPACK already. It chooses a suiting routine already and handles exceptions efficiently. For large data most of the time is spent inside the LAPACK functions itself, therefore the overhead caused by Matlab is tiny to negligible. Only if you have a lot of experiences with LAPACK and your matrix has a very specific structure (e.g. a sparse block matrix with known distribution of zeros), using specific LAPACK calls to solve the subsystems will be faster. But such techniques are prone to errors and you have to prove the numerical stability at first.
5 commentaires
Jan
le 10 Déc 2012
Solving a large eigenvalue problem consume a lot of time, because a lot of calculations are required. The eig() command suggested by Sean calls LAPACK routines already. As long as the system does not have a well known sparse structure, there is no better way.
It is important to mention the dimensions exactly: While for some scientists a 100x100 matrix is called "large" already, others work with 1e6x1e6 matrices. In the later case it is often not required to know all eigenvalues, but the largest 1000 are sufficient already.
To summarize, when you want to solve a general eigenvalue problem in Matlab, buying a faster processor is the most efficient way for improvements - as long as the computer has enough RAM such that the very expensive disk swapping ("virtual memory") can be avoided.
Voir également
Catégories
En savoir plus sur Linear Least Squares dans Help Center et File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!