optimizer with Hermitian sdpvars

[Alastair Abbott]

Hi, 

As it stands, it seems that optimizer doesn't work with Hermitian parameters. As a trivial example:

a = sdpvar(2,2,'hermitian','complex');

optimizer([], trace(a), sdpsettings('solver','mosek'), a, trace(a))

which produces the error

Error using optimizer (line 98)

All parameter arguments have to be simple variables (i.e., not expressions such a+b or 1+a)

I find this a bit cryptic, but I guess the problem is the relation between the complex parts of the upper/lower parts of the matrix.

Is there a way to get around this or to support Hermitian parameters?

Best regards and thanks in advance,

Alastair.

[Johan Löfberg]

Nope, not possible. You would have to hack around that by using the unique elements in the real and complex part as the parameters

a = sdpvar(n,n,'hermitian','complex');

indR =find(triu(ones(n)));
indI =find(triu(ones(n),1));

a_r =real(a(indR));
a_i =imag(a(indI));