YALMIP gets mad with quadratically constrained uncertain variables

[dnmiller]

Hi,

I'm trying to solve a robust optimization problem in which the uncertain variable is a unit vector. When I run optimize, I get an error message that says Report bug in problem classification (linear constraint). So here I am.

Simplest code I can find to reproduce:

x = sdpvar(3, 1);
e = sdpvar(2, 1);
A = randn(2, 3);
V = randn(2, 2);

W = [0 <= x <= 1, e' * e == 1, uncertain(e)];
J = norm(A*x - V*e);
sol = optimize(W, J);

If I try to use robustify instead, I get the error message Some uncertain variables are unconstrained.

So, I'm guessing YALMIP doesn't support quadratic constraints for uncertain variables? Any ideas on how I can translate this into a problem that would make YALMIP happy? I tried parameterizing the problem with a single angle, but ran into the same error.

[johanlofberg]

thanks for reporting. For the underlying modelling issue, please post a question on the google groups forum for suggestions