Hello Arun,

Sorry for the long silence.

As reported here: https://savannah.gnu.org/bugs/?61762, it turns out that the behaviour of qp, as it is, is not reliable when H is not positive definite.

As long as a better QP solver is not used, checking for positive definiteness before calling qp is therefore the safest choice.

Regarding your patch, it means I think that:

1) "quadratic program" -> "convex quadratic program" is ok, you could even call it a "strictly convex" quadratic program

2) the matrix is expected to be positive definite (semidefinite is not enough)

3) The test for positive definiteness should stay.

Julien, could you check against the bug reported in 61762:  https://savannah.gnu.org/bugs/?61762

It seems allowing non-positive definite matrices caused a crash in qp?

Copying Olaf in either case.

The documentation of qp does not indicate that the matrix H is expected to be (symmetric) positive definite.  If it's not, the function fails with error code 102.

There is nothing obvious about it, since the "quadratic programming" framework allows in general, mathematically speaking, H to be any symmetric matrix.

I suggest the following fixes:

1) quadratic program -> convex quadratic program

2) indicate somewhere explicitely that H is expected to be symmetric positive definite