feasibility check

[mengj...@gmail.com]

Dear guys,

The feasibility status is only result we want for our problems so there is no need to solve the problems completely for the solution. Therefore, if a feasible solution is found, the calculation needs to be stopped as we already have the answer: the problem is feasible.

A simply possible way is to have a zero-objective function. I have tested this method with several problems showing a promising result (a typical solution of the zero -objective function after the correct objective function in Interior-point solution summary) as follows:

Problem status : PRIMAL_AND_DUAL_FEASIBLE

Solution status: OPTIMAL

Primal. obj: -1.1188752757e-04 nrm: 2e+01 Viol. con: 2e-12 var: 0e+00 cones: 0e+00 - iterations: 19 time: 1.66s 

Primal. obj: 0.0000000000e+00 nrm: 3e+52 Viol. con: 2e+10 var: 0e+00 cones: 8e-17 - iterations : 0,  time: 0.39s, pre-solved

My question is if the solution from the Presolve can be used, given that a large solution norm is found. What is your suggestion on speeding up the feasibility check merely? 

Best regards,

Jing

[Michal Adamaszek]

Hi,

I answered also your support ticket.

The solution found in the presolve with zero objective has huge norm but also huge violation. We suspect your problem is a bit on the edge and that causes the numerical issue you see. 

If you are satisfied with the quality of the solution then you are welcome to use it. But I would rather get the solution and check the constraints are satisfied to a level you want yourself, and not "just" trust on the solution status. Maybe the MOSEK solution can be rescaled to a more reasonable size.

Best,

Michal

[mengj...@gmail.com]

Hi Michal, Thank you for explanation.