Subject: [ublas] Slightly OT: How to test/compare QR algorithm
From: Jeremy Conlin (jeremit0_at_[hidden])
Date: 2009-04-28 22:35:52
I apologize in advance for being slightly off topic. I was hoping someone
on this list would know the answer to my question.
I have implemented a double-shift QR algorithm in three languages, C++,
Python, and MATLAB. The reason for the Python and MATLAB implementations
are to check my C++ version which uses the boost bindings to LAPACK. My
problem is that none of the implementations give the same result. Even the
QR decompositions are different. I know the QR decomposition isn't unique,
but it is only not-unique up to multiplying by -1. At least that's the way
I understand it.
Is anyone aware of how I might go about testing my algorithm? Is there a
published example I can compare with?