Re: [ublas] BOOST_UBLAS_USE_ITERATING vs lu_factorize: broken?

Subject: Re: [ublas] BOOST_UBLAS_USE_ITERATING vs lu_factorize: broken?
From: Paul C. Leopardi (paul.leopardi_at_[hidden])
Date: 2009-08-07 20:58:52

On Wed, 5 Aug 2009, Gunter Winkler wrote:
> Paul C. Leopardi schrieb:
> > Has anyone checked lu_factorize with -DBOOST_UBLAS_USE_ITERATING?
> > It's giving me stranger results. I'm using Boost 1.37.0 with g++ 4.4.0.
> > Best, Paul
>
> A similar problem occurred some time ago, see
> http://osdir.com/ml/lib.boost.ublas/2008-07/msg00000.html maybe the
> patch did not fix everything?
>
> The output is difficult to understand. What is the expected result?

Thanks for your reply. 

The expected result is the result without BOOST_UBLAS_USE_ITERATING. Notice 
that without BOOST_UBLAS_USE_ITERATING, the expression B/B causes debug 
prints of AT, BT, LU and XT (twice), with a result of 1 (ie. the identity 
matrix). 

With BOOST_UBLAS_USE_ITERATING, the only debug print is for AT and the result 
is NaN. This means that ublas::lu_factorize(LU, pvector) is returning a 
non-zero result, indicating that the matrix is singular.
Best, Paul