I am using Eigen3 to calculate the eigenvalues and eigenvectors of lots of matrices. Some of these matrices are a bit big (i.e. 18000x18000) and when I use the eigensolver on them I get an error at runtime.
After some research I found out that it crashes with matrixsize of 16000 and above.
Here is the code I use:

Code: Select all

n = 16000;
MatrixXf L = MatrixXf::Random(n,n);
EigenSolver<MatrixXf> eigensolver(L);
system("PAUSE");
EigenSolver<MatrixXf>::EigenvalueType eigenvalues = eigensolver.eigenvalues();
system("PAUSE");
EigenSolver<MatrixXf>::EigenvectorsType eigenvectors = eigensolver.eigenvectors();


It never reaches the first system("PAUSE"); before that I get:

Debug Error!

Program: F:\<some folders here>\main.exe

R6010
- abort() has been called

(Press Retry to debug the application)

When I press Retry then I only get the message that main.exe has stopped working, so thats not very usefull either.
Is this a known limitation of Eigen3 or am I doing something wrong?

extra info:
Win7 x64
Visual Studio C++ 2010 Express
(Plz let me know if you need more info)

You probably ran out of memory. With such large matrices you should consider using a sparse eigen-decomposiiton, assuming your matrices are sparse. If you are only interested by the largest eigenvalues with a high probability, there is: http://code.google.com/p/redsvd/ which uses Eigen's matrices. Otherwise slepc is a reference but more painful to use.