Hi,

The question is X=A/B, while A(m,n),B(n,n),hoping X has(m,n) dimension.
Direct inverse proves to be unstable. However, Eigen's solvers like householderQr() etc. requires m=n.

In Matlab, I use X=A/B, it works great. I need suggestion how should I do in Eigen.

Thank you.

In devel branch you can "solve on the right" as follows:

X = A * B.lu().inverse();

This won't compute the inverse explicitly, but appropriately apply the LU factors to A. You can also transpose the pb:

WhateverDecomposition<MatrixXd> dec(B.transpose());
X.transpose() = dec.solve(A.transpose());

where WhateverDecomposition is PartialPivLU, HouseolderQR, etc.

ggael wrote:In devel branch you can "solve on the right" as follows:

X = A * B.lu().inverse();

This won't compute the inverse explicitly, but appropriately apply the LU factors to A. You can also transpose the pb:

WhateverDecomposition<MatrixXd> dec(B.transpose());
X.transpose() = dec.solve(A.transpose());

where WhateverDecomposition is PartialPivLU, HouseolderQR, etc.

Thank you，ggael,

Prob. solved.