Re: [eigen] Blocked QR algorithm - lapack compatible ?

[ Thread Index | Date Index | More Archives ]

Mmh, ok then maybe an explication would be the following: minpack (on which i've based the NonLinearOptimization module) was bundled with its own implementation of QR decompozition. Which i assumed was the lapack one, but maybe it is not, or i dont know which version it was. Instead of having lot of packages with dependencies, they often make self-sufficient packages (as is minpack) by copying files/code.

Thomas Capricelli <orzel@xxxxxxxxxxxxxxx>

In data martedì 08 giugno 2010 10:58:10, Gael Guennebaud ha scritto:
> Really ??? When I look at Lapack documentation our storage should really be the same:
> - the upper triangular part (including the diagonal) represents the R matrix
> - the strictly lower triangular part of the matrix stores the householder vectors v_i, and the householder coefficients h_i are stored in a separate vector (called tau in Lapack) such that each Householder reflector H_i is defined as:
> H_i = I - h_i w_i w_i^T
> where w_i^T = [0 .. 0 1 v_i^T]
> The above is true for both Eigen and Lapack.

Mail converted by MHonArc 2.6.19+