Re: [Help-glpk] Regarding understanding MIP output

[Andrew Makhorin]

> I used glpsol to solve the linear relaxed version of a MIP problem ( all
> variables in my problem are 0 or 1) and  output the solution using -o
> flag. What I want  is to get all those variables which have been set to
> either 0 or 1 as a part of the solution and do not have  a fractional
> value. In the output file , I extracted all the column ids having value 0
> or 1 and added these as a new set of constraints to the original MIP
> problem by fixing the variables as 0 or 1 according to the relaxed
> solution.
> 
> 
> Thus,  this new MIP problem should surely be feasible right?

Not mip. Its relaxation must remain optimal (ergo, feasible).

> 
> I did that but I found the new MIP problem to be infeasible. I dont know
> why this is happening. The problem instance is present at
> http://www.cse.ohio-state.edu/~khannag/bip_8_20_new.lp.
> 
> The solution I get when I run the relaxed version of the problem is
> present at http://www.cse.ohio-state.edu/~khannag/solution
> 
> In this solution file,  I took the value for column name under the
> activity label as their solution value. Based on my
> interpretation of the output file, 9409 variables out of 10244 were either
> 0 or 1 according to the number under the Activity Label in the solution
> file. I added these as constraints to the original MIP but it becomes
> infeasible.
> 
> Could you help me figure out the problem?

Probably you made a mistake somewhere. Try checking your code.
If you fix some variables at the values they have in the optimal point,
the lp solution must remain optimal.