Cvx optimal value (cvx_optval) +Inf

(hua meng) #1

the formulation as follows :

my matlab code using cvx as follows :slight_smile:

however, the cvx output

,i try the solver sedumi and sdpt3 ,however,i doesn’t work. could anyone help me ?
Thanks in advance for anyone giving tips.

(Mark L. Stone) #2

Either your problem is truly unbounded or CVX/solver is having difficulty due to the combination of CVX’s successive approximation method and numerical difficulties. Status does say "Inaccurate/Unbounded, so I am not sure what confidence should be placed on the result. If you have access to CVX Professional with MOSEK, you can try that, as it is more robust and reliable than sedumi or sdpt3. You cn also check to see how well scaled your data is, and if not well-scaled, try to improve the scaling and re-run.

You can avoid use of CVX’s successive approximation method by using CVX 3.0beta with ECOS or SCS solvers, which you would have to install. That might produce a more reliable resut.

Do you know that the problem really is bounded, and if so, how do you know?

(hua meng) #3

it is not difficult to deduce that the problem is bounded ,actually ,due to the the third and fourth constraints . thanks for you advice ,i will try it using CVX 3.0 beta ECOS or SCS solvers.

(Michael C. Grant) #4

There’s a reason we say that the successive approximation method is unsupported, I’m afraid. Complex models involving log/exp/entropy are really not reliably solvable using CVX.