2
$\begingroup$

In the convex optimization community, there has been a growing concern regarding the search for a reliable open-source solver (especially a quadratic programming solver) that can effectively deliver accurate results in reasonable times. This search has become quite challenging, requiring a substantial investment of time and effort.

What is a reliable open-source quadratic solver?

$\endgroup$
3
  • 3
    $\begingroup$ The benchmarks here and here might be of interest. $\endgroup$ Commented May 18, 2023 at 11:04
  • 2
    $\begingroup$ Clarify (in the title) whether you are looking for Quadratic Programming (QP) solver, i.e.. linear constraints; or Quadratically-Constrained Quadratic Programming (QCQP), i.e., quadratic constrraints. Furthermore, specify whether the solver need only handle convex problems, or must also handle non-convex problems. If non-convex capability, do you want just local optimization, just global optimization, or ability for uses to specify which of local or global optimization to perform.. Unfortunately, you might be difficult to choose more than two out of "reliable", "scalable", "open source". $\endgroup$ Commented May 18, 2023 at 12:55
  • 3
    $\begingroup$ The situation is even worse for MIQP solvers. $\endgroup$ Commented May 18, 2023 at 14:01

1 Answer 1

5
$\begingroup$

I'll take the bullet and give you the honest answer, there aren't any.

That's not to say that we don't have perfectly fine free solvers, it's just that if our frame of reference about "reliable", "scalable", or performant, is the very best stuff out there, the best stuff is always commercial and it is many times better than any free alternative, especially on non-library problems.

This often comes as a surprise for people unfamiliar with the field, but it takes hundreds of thousands of hours of development to write a solver that performs well, which is time open-source developers don't really have (since they can't get funding for it).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.