Initial guess

[optimizer]

Dear Professor Johan,

I am trying to verify the feasibility of a problem by:

diag_cvx=optimize([s==rvec]+ConstrTRelax+ConstrBounds+ConstrSpillover,[],sdpsettings('solver','fmincon','usex0',0,'verbose',0,'fmincon.MaxFunEvals',Inf,'fmincon.MaxIter',Inf,'savedebug',1))

diag_cvx = 

  struct with fields:

    yalmiptime: 0.1569

    solvertime: 0.0431

          info: 'Successfully solved (FMINCON)'

       problem: 0

If i do:

SpEig=recover(U_res);

[EigRes,omega_useless]=eig(A,B);

EigVecRes=[];

for i = 1 : size(M_res,1)    

    EigVecRes = [EigVecRes;EigRes(:, i)];       

end

assign([x;SpEig],[0.75186;eps;eps;0.7678;0.83798;39402;67486;34565;23609;35852;eps;eps;32180;24198;EigVecRes]);

diag_cvx=optimize([s==rvec]+ConstrTRelax+ConstrBounds+ConstrSpillover,[],sdpsettings('solver','fmincon','usex0',1,'verbose',0,'fmincon.MaxFunEvals',Inf,'fmincon.MaxIter',Inf,'savedebug',1))

diag_cvx = 

  struct with fields:

    yalmiptime: 0.1058

    solvertime: 0.0342

          info: 'Infeasible problem (FMINCON)'

       problem: 1

Why if i give an initial guess the problem becomes infeasible? If my initial guess is wrong i would expect fmincon to mode inside the bounds the x0 but not to have an Infeasible problem error.

Best Regards

[Johan Löfberg]

I guess it is a nonconvex problem, so all bets are off. fmincon simply performs better when starting from 0, than from some other supplied point

or your initial point is not exactly feasible, and the solver gets fooled somehow. or you are only initializing the variables which you know of, but there are internal variables introduced by YALMIP which cannot be initialized and they lead to problems, etc. Impossible to say without reproducible code