Hello,

First of all, thanks for Eigen! It's a very nice package, and its interface is much more "native" to the C++ programmer who simply wants to solve a single set of linear equations than any of the alternatives I looked at - which all seem to assume you're a Matrix Expert.

That said, I'm trying to solve a system of linear equations using least squares fit. However, the resulting vector must not contain negative values (they're a physical impossibility). I can promise that all values in my matrix are positive. Furthermore, my matrix is diagonally symmetric with 1's on the diagonal, some interesting values two or three positions off the diagonal, and zeros in the upper right and lower left triangles (not sure how such a matrix is called - banded?). Maybe that helps in picking the right algorithm. The matrix is in the order of 100x100.

Ideally, I would love to use a much larger matrix and use all measured data as resulting vector (4096 data points) and fill the matrix with rendered base functions so it would become something like 100x4096 in size. It this still feasible for a sub-second calculation on "normal" PC hardware?

Many thanks for any pointers. I browsed the forum up to one year back but could not find this item addressed (by the way, it seems that when I use the "search" field, I look through the entire KDE forum; can I limit my search to just the Eigen forum?)

Kind regards,
Simone

The quick search form searches across the entire forum. If you use the normal search form (search.php) then you can limit your search to the Eigen forum (among other forms of limits).

Thanks, I tried the search you suggested. I looked for "nonnegative least squares" but only my own post turned up...

So as to the original question, any help?