The results obtained by using quadprog and gurobi are quite different

Answered

la La

September 17, 2023 09:44

• 
• 

I have an optimization problem, but the results obtained by using quadprog and gurobi are quite different. I debugged for a long time but couldn't find the reason, so I came to ask for help.

See the mat file for the data of optimization problem (In the below link).

https://drive.google.com/file/d/1Xf1AtQ6x3wTIl3xO07yXrn8Rn4VjtcNE/view?usp=sharing

 

The procedure of quadprog is as follows:

qpoptions = optimset('Display','off');

[x, fval, qpexitflag, ~, Lagmultipliers] = quadprog(qpinfo.H, qpinfo.f,...

                        qpinfo.A, qpinfo.b, qpinfo.Aeq, qpinfo.beq, [], [], [], qpoptions);     

 

The procedure of Gurobi is as follows:

clear params;

params.outputflag = 0;

params.method = -1;

% full problem

clear model;

model.Q = 0.5*sparse(qpinfo.H);

model.obj = full(qpinfo.f);

model.A = sparse([qpinfo.A;qpinfo.Aeq]);

model.rhs = full([qpinfo.b;qpinfo.beq]);

model.sense = [char('<'*ones(size(qpinfo.A,1),1));char('='*ones(size(qpinfo.Aeq,1),1))];

model.modelsense = 'min';

%%

result = gurobi(model, params);

 

 

0

• 

  Eli Towle

  • Gurobi Staff

  September 18, 2023 20:24

  In Gurobi, the lower bounds on variables are \( 0 \) by default. To solve the same problem as \(\texttt{quadprog}\), explicitly remove the lower bounds on the variables:

  model.lb = -Inf(size(qpinfo.f));

  You can read more about this in the The model argument section of the Gurobi MATLAB documentation.

  0

Please sign in to leave a comment.