Stephen Langdell of NAG and myself recently collaborated on an article in The Actuary magazine, where we discussed some issues that frequently crop up when dealing with correlation matrices. You can read the article here, or a slightly longer version on NAG's site here.
NAG are the Numerical Algorithms Group, a not-for-profit organisation which originated in academia several decades ago. They are specialists in numeric computing and their flagship product is a vast software library of numeric routines with bindings to C, .NET, FORTRAN and various other languages. I recently used the library extensively when building model calibration routines for a client and I am impressed. Readers will know that I am a big fan of R, but when you need speed there is nothing like compiled code.
NAG don't have a very high profile in the insurance industry currently, but with companies looking at taking greater ownership of their models and calibrations that should change.