How to read those information SeDuMi print in Matlab

when I use SDPT3 to slove my question ,print NAN.
but use SeDuMi can get the optimal value.
Which answer should I believe!

1 Like

You haven’t shown us the SDPT3 output (solver and CVX). There are at least three different reasons, which should be clear from the CVX output, why NaNs occur - either the solver fails (due to numerical trouble) or CVX declares, based on solver results, that the problem is infeasible or unbounded.

If you have Mosek available, try that, and show us the solver and CVX output.

1 Like


Dear Mr. Mark, this is the result I got after using the SDPT3 optimizer. The Matlab console prints NAN. I don’t know why this happens. The SeDuMi optimizer can get the optimal solution. These results are obtained after installing CVXQUAD and replacing the exponential.m file. The result I got using the Mosek optimizer did not converge. What I did was a concave-convex procedure, which is a concave-concave function, so I used the first-order Taylor formula to approximate it. This bothers me because I don’t know which optimizer gets the right result.

1 Like

SDPT3 failed due to numerical troubles.

If you are going to use Mosek for a problem with exponential cone functions, you should swap back in the original CVX exponential.m file and run it in a new MATLAB session.

it is possible your problem has bad scaling. Perhaps some very large input numbers. The concave-convex procedure can be unstable, and might lad to numerically bad optimization problems even id the starting problem was not badly scaled.