concatenation error in call_mosek_dual

[Michael Hibbard]

Hi Professor,

I am trying to run a toy problem in which the objective function is to maximize the entropy of a probability distribution subject to some quadratic constraint (which, in the case of this toy problem, should be trivially satisfied). I am using YALMIP R20190425 with MOSEK 9.0.94. The code is:

% toy problem 

clc;clear all

x = sdpvar(5,1);

constraints = [];

constraints = [constraints, x >= 0];

constraints = [constraints, x <= 1];

constraints = [constraints, sum(x) == 1];

for i = 1:5

    constraints = [constraints, cpower(x(i,1),2) <= 1];

end

obj = entropy(x);

ops = sdpsettings('solver','mosek','verbose',0,'debug',1);

optimize(constraints,-obj,ops);

sol = value(obj);

xVals = value(x);

If I comment out the for loop, the optimization yields the expected result of x(i) = 0.2 for all i. Furthermore, if I change the objective function to simply sum(x), the optimization returns the expected result of 1. However, including both the quadratic constraints and entropy term yields the following output error:

Error using horzcat

Dimensions of matrices being concatenated are not

consistent.

Error in call_mosek_dual (line 39)

        prob.cones.type = [prob.cones.type 3];

Error in callmosek>call_mosek_lpqpsocpsdp (line

97)

        [x,D_struc,problem,r,res,solvertime,prob]

        = call_mosek_dual(model);

Error in callmosek (line 51)

    [x,D_struc,problem,r,res,solvertime,prob] =

    call_mosek_lpqpsocpsdp(model);

Error in solvesdp (line 361)

    eval(['output = ' solver.call

    '(interfacedata);']);

Error in optimize (line 31)

[varargout{1:nargout}] = solvesdp(varargin{:});

Error in practice (line 17)

optimize(constraints,-obj,ops);

I am not sure how to change the code to not produce this concatenation error, or if I am missing something else completely. Any help would be appreciated, thank you!

[Johan Löfberg]

Yes, it was a bug which has fixed already in develop which you thus will have to use until next official release

BTW, cpower(x(i,1),2) <= 1 can simply be written as x(i,1)^2 <= 1. cpower is intended for more exotic powers