Re: [Help-glpk] Initial starting point of simplex algorithm

[Rabih Chaar]

Hello,
Thanks for the quick reply. 
Concerning your recommendation on checking my objective coefficients, I was 
also outputting my relaxed problem in cplex, and everything seems fine. 

I am not having this behavior on all instances of the problem. For others I am 
getting the sinf not null.

My goal is to implement an algorithm giving a minimal iis if the problem is 
infeasible, hence the relaxation. This is done programmatically via the glpk 
API,  making a copy of the problem, adding the additional elasticity variables, 
setting there objective coefficients to one, setting the objective coefficients 
of initial variables to zero, and adapting row constraints with the elasticity 
variables. I will check my code again. 

I wanted to be sure that the primal simplex (on the initial problem) will 
always find a feasible domain if it exists. 

Thanks in advance. 

Rabih

Le 24 janv. 2013 à 14:40, Andrew Makhorin <address@hidden> a écrit :

> 
>> PS: here is a small log of both problems
>> 
>> GLPK Simplex Optimizer, v4.47
>> 191255 rows, 68860 columns, 450946 non-zeros
>> 0: obj = 0.000000000e+000 infeas = 5.436e+009 (28783)
>> ...
>> 17761: obj = -8.927238865e+008 infeas = 2.030e+007 (14520)
>> PROBLEM HAS NO FEASIBLE SOLUTION
>> 
>> Elasticity
>> Addtional constraints 220038 for 191255 rows
>> Current column size is 68860
>> Additional Constraints 220038
>> Current column size after resize is 288898
>> GLPK Simplex Optimizer, v4.47
>> 191255 rows, 288898 columns, 642201 non-zeros
>> 0: obj = 0.000000000e+000 infeas = 5.436e+009 (28783)
>> ...
>> * 1552: obj = 0.000000000e+000 infeas = 1.490e-008 (28782)
>> OPTIMAL SOLUTION FOUND
>> Sinf is 0
> 
> BTW, in the second case the objective (Sinf, I guess) is zero on all
> iterations. Probably you forgot to assign objective coefficients to
> elastic variables.
> 
>