September 10th, 2002, 5:41 am
LAPACK++ is basically a C++ interface to the Fortran 77 routines in LAPACK. Now, LAPACK is a very general linear algebra package which supports a variety of matrix types: Real symmetric, banded, tridiagonal, sparse etc. Perfomance is excellent, relying in part on the availability of machine specific optimized BLAS (Basic Linear Algebra Subroutines) level 3 routines. However, given that linear algebra per se is not your point of interest, you are probably not interested in the plethora of routines for eigenvalue calculations for different matrix types but, rather, in a few general routines. Also, sparse matrices are not likely to crop up in financial applications. LAPACK++ routines are scattered over a large number of files. I suggest sticking to a library with few routines which are easy to include in projects and where matrix/vector classes are simple to change. I've been using TNT for a while since I can understand myself what is going on and can modify it myself should the TNT-projct not be supported any more.