possible bug findelements

[Abhishek B]

Hi Prof Logberg,

I think I've found a bug in Yalmip. 

I'm trying to solve SOS-relaxation problems and the following error in matlab has occured (see also attached screenshot);

-----------------------------------------------------------------------------------------------------------------------------------------------------------------

Undefined function 'findelements' for input arguments of type 'double'.

Error in solvesos (line 321)

                            h{constraint} = h{constraint}(findelements(sum(h{constraint},2)),:);

Error in step3_HilbertB (line 34)

[sol, u, Q]=solvesos(prog, [], ops, coeff2);                    % Running optimizer

Error in debug (line 14)

[s, G, res] = step3_HilbertB(n,m,v, [v0 V], 1,1/(2^1), solver); 

-----------------------------------------------------------------------------------------------------------------------------------------------------------------

Since this problem is occuring with mosek and not sedumi, I think its a bug in Yalmip code. 

This is done with random input data to construct the functions. I've had a look at the solvesos.m file, and my guess is that 

in the mosek formulation of the problem, the constraint variables become small and sos.clean makes them 0.

Is this a known issue with some simple fix?

I'd be happy to send you the data that generates this problem if you'd like, I've copied my code for "step3_HilbertB" below;

Best Regards, 

Abhishek B.

=================================================================================================

function [success, G, res] = step3_HilbertB(n,m,v,V,d,delta,solver)

%UNTITLED Summary of this function goes here

%   Detailed explanation goes here

%% sdpvar's

X = sdpvar(n,1);

Y = sdpvar(m,1);

Z = kron(X,Y);

monlist = monolist([X;Y], d); 

monlist = monlist(2:end);

G = sdpvar(size(monlist,1), size(monlist,1),'symmetric');       % matrix of coefficients

phi = monlist'*G*monlist;                                       % vector of decision functions phi

%% linear and quadratic forms

f = v'*kron(Z,Z);                                             % defining the non-sos quad form f

h = V'*Z;                                                      % The linear forms h constructed in step 1

F = delta*f + (10)*(h'*h);                                    % non-negative but not sos form

%% sos program

prog = [sos(phi*F);trace(G)==1];

%% decision variables

coeff2 = reshape(G, [1, size(monlist,1)*size(monlist,1)]);   % reshaping coefficient into vector

%% sdp solver

ops = sdpsettings('solver', solver);

ops.verbose = 0;

[sol, u, Q]=solvesos(prog, [], ops, coeff2);                    % Running optimizer

if size(Q)~=0    

    G1 = reshape(value(coeff2),[size(monlist,1),size(monlist,1)]);

    res = max(coefficients((monlist'*G1*monlist)*F - u{1,1}'*Q{1,1}*u{1,1}));

    

    if sol.problem==0 && res<=1e-4

        %disp('SOS decomposition is possible.')

        G = G1;

        success = 1;

    else

        %disp('SOS decomposition not found.')

        success = 0;

        G = zeros(size(monlist,1), size(monlist,1));

    end

else

    success = 0;

    G = zeros(size(monlist,1), size(monlist,1));

    res = 0;

end

end

[Johan Löfberg]

you would have to supply reproducible code