Solving QCQP using gurobi and numerical trouble


(Qingqing Zhao) #1

I am trying to optimize a high dimension sparse QCQP problem and I have encountered the numerical issue.

May I ask where do those statistics come from?
Whats the Matrix in “Matrix coefficient range” and what is the RHS coefficient range?

Linear constraint matrix : 82820 Constrs, 132512 Vars, 393928 NZs
Quadratic constraints : 33128 Constrs, 99384 NZs
Matrix coefficient range : [ 4.30793e-06, 39.5285 ]
Objective coefficient range : [ 0.5, 400 ]
Variable bound range : [ 0, 0 ]
RHS coefficient range : [ 1, 225000 ]


(Mark L. Stone) #2

These numbers are reported by Gurobi for the model sent to it by CVX (after whatever transformations CVX has applied). https://www.gurobi.com/documentation/8.1/refman/attributes.html .

If that does not adequately address your questions, perhaps you can inquire at https://groups.google.com/forum/#!forum/gurobi .


(Qingqing Zhao) #3

Thanks a lot! I have already posted in the Gurobi google group!


(Mark L. Stone) #4

Have you read http://files.gurobi.com/Numerics.pdf ?

You can set NumericFocus to 3 as shown in my answer at How can I make the solver settings to act? .


(Qingqing Zhao) #5

Thanks a lot for your reply. It is really helpful.