Johan Löfberg

Smarter handling of quadratics and socps in optimizer

1

This fails with mosek as yalmip will pick the socp version of mosek (why?) which doesn't support quadratic objective sdpvar x y P = optimizer([-1

Compute parametric solution to optimizer object

2

and make it transparant in the way the solution is used, compared to a call to optimizer

Expensive, but should not be standard case

yalmip clear R = sdpvar(1, 1); x = sdpvar(nx, N+1, 'full'); u = sdpvar(nu, N, 'full'); cons = []; obj = 0; cons = x(2)==x(1)+u(1); obj = x(2)'*x(2) + u(1)'*u(1)+R;...

``` X = sdpvar(3,3,'hermitian','complex') e = X(1,2)-(2 + sqrt(-1)*3); Model = [[eye(3) zeros(3);zeros(3) X] >= 0, e == 0]; optimize(Model,trace(X),sdpsettings('dualize',1)) ```

Possible?