Slacktol value ?

I tried the solution mentioned in the following link for changing the eps value to slacktol value, but still not working. My version of optimization toolbox is 5.1 and R2010b:

http://www.mathworks.com/support/solutions/en/data/1-1BYP9/?solution=1-1BYP9

    H = 2.*eye(12); f = rand(12, 1); %Define problem
    A = []; b = []; %Define inequality constraints
    Aeq = [ones(1,6) zeros(1,6); zeros(1,6) ones(1,6)]; beq = [16.8; 175.2]; %Define equality constraints
    lb = [0; [1.9:.95:11.4].'; [8.35:.95:18.8].']; lb(7) = 0; lb(13) = 8.35;
    UB = 292.*ones(24,1); UB(1) = 0; UB(7) = 0; %Define upper and lower bounds
    [set, opt_val] = quadprog(H,f,A,b,Aeq,beq,lb,UB); %Solve problem

After changing also still the same answer: Warning: Large-scale algorithm does not currently solve this problem formulation, using medium-scale algorithm instead. > In quadprog at 293 Exiting: the constraints are overly stringent; no feasible starting point found.

[Samir]

done. thanks.

Answer 1

[Rakesh Kumar]

First suggestion is to use the newer algorithm added in 11a if you can.

The warning message I get should be fixed (easy to fix) Warning: Length of lower bounds is > length(x); ignoring extra bounds.

When I tried with the newer algorithm added to quadprog (interior-point), the solver correctly detects the problem to be infeasible. Unfortunately, the older algorithm (active-set) stops with a cryptic message ('the constraints are overly stringent') but essentially tells me that the constraints cannot be satisfied (hence infeasible).

Rakesh

[Samir]

thanks it worked

Answer 2

[Steve Grikschat]

If anyone else checks this, know that the problem given above is infeasible and the solution is erroneous.

It's not recommended to modify the code. Instead, try what Rakesh has suggested. Run the interior-point-convex algorithm if you have R2011a or later.

options = optimset('Algorithm','interior-point-convex'); quadprog(H,f,A,b,Aeq,beq,lb,ub,x0,options);

[Samir]

thanks it worked.