Are there any restrictions on using gurobi in yalmip?

[ellensonny]

hello Prof,
This is the part of my code that uses the gurobi solver：

However, there is a problem here：
Warning: Solver not applicable (gurobi does not support quadratic semidefinite constraints).
Gurobi's Q&A staff told me: 1，gurobi itself can handle this problem
2，python supports all the functions of gurobi, but yalmip does not support the non-linear functions of gurobi.
Is there any way to use the full functionality of gurobi in yalmip without restrictions?

thanks for your reply very much!!!

[johanlofberg]

Gurobi does not support semidefinite programming.

Most likely you've defined a symmetric matrix and by mistake constrained it to be positive semidefinite instead of elementwise non-negative https://yalmip.github.io/tutorial/basics/. You see that if you display the object con

[johanlofberg]

As I said, gurobi is historically a MISOCP solver, i.e. not intended for general nonlinear programming, and requires much more thought to see when things can be cast as SOCP stuff. You cannot just throw arbitrary nonlinear stuff at it as a general local nonlinear solver such as fmincon. Since a couple of generations back it was extended to global nonconvex QCQP which YALMIP supports, and in the most recent version they added support for a restricted set of nonlinear operators, which YALMIP does not interface yet.

Cubic can be handled easily though, simply introduce a new variable z, and then add then constraint z == x1x1 and then use fz, as that will lead to a nonconvex QCQP

[ellensonny]

Yeah, I heard about that they added support for a restricted set of nonlinear operators.Since YALMIP does not interface yet,when I was describing this mathematical model, gurobi had no way to solve it.

Would you give me some suggestions to solve this problem correctly without using gurobi or other methods?

[johanlofberg]

Initial fixes have been pushed to develop branch to try

[johanlofberg]

you're still going to have to massage the model though

the cubic expression xV^2 should be written as Q^2 where you want Q=xV which you most likely do best by logic linearization, i.e implies(x,Q==V) + implies(1-x,Q==0)

in the same way, xV1V2cos(a) you write as Qcos(a) where Q is xQ1V2 which you do by Q==Q2V2 where Q2 is the logic linearization of xV1

[ellensonny]

thanks a lot ! I will have a try.