Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
Drexel University or The Math Forum.



Q: Speed of compilied eig() vs. MEXfile
Posted:
Sep 17, 1996 10:41 AM


I'm computing the eigenvalues of 1000 by 1000 covariance matrix and need to speed up the process. The matrix is real, symmetric, Toeplitz and positive definite.
Which would be faster:
(1) Compiling the eig() function to a MEXfile with array bounds checking off, all numbers assumed real and the 'nobalance' option enabled, (2) Taking the eigenvalue code from Numerical Recipes in C (specifically routines tred2 and tqli) and making it into a MEXfile.
Does anyone have experience with this?



