Problem when trying to translate a generalized eigenvalue solution in Matlab into using LAPACK directly

Hi,

I have recently had a problem related to trying to move a working solution that use the generalized eigenvectors as computed by Matlab into a Java program. I am then trying to call the same routines in LAPACK that I assume Matlab is using (both matrices will be real and symmetric), but then get the eigenvectors scaled by different factors, i.e., if I divide the Matlab eigenvectors with the ones I get from LAPACK when called directly, I get a different scaling for each of the vectors. Regrettably, this makes my solution give a different result - and a wrong one. I would therefore like to know how I should call the LAPACK functions to get the same results as if using Matlab, or any fix that I can use to solve the problem.

As the vectors are not normalized, one is not able to correct this by simply normalizing them.

Thank you for your time.

Regards, Andreas