forum.alglib.net

Hi,

The documentation says (https://www.alglib.net/nonlinear-programming/sequential-quadratic-programming.php):



Is there a way of actually specifying the Linear Constraints in a sparse form? In my problem, each constraint only involves a handful of variables, which makes the matrix C very sparse, sometimes leading to memory issues.

Best
Matevž

Hi!

Actually, this feature was recently added. However, I just checked online docs, it turns out that an old version was uploaded to the website :((( really sorry for the mistake, need to refactor release workflow.

What you need is provided by two sets of functions: minnlcsetlc2() that provides "set" semantics (complete rewrite of current constraints) and minnlcaddlc2...() that allows to add a constraint to the current list. The latter can sometimes be more convenient.

Sergey

That's awesome news.
As I see, you have already included this in ALGLIB version 4.02?

Exactly!

ALGLIB nonlinear programming already became much better this year, and we plan to improve it ever more. 4.04, which is planned for December, will definitely see a model-based derivative-free nonlinear programming solver. There is also a good chance that a nonlinear interior point method will be included, on many problems it has less overhead than SQP (although personally I dislike its inability to do warm start). And finally it is planned to introduce sparsity to the conic programming solver.