Re: [Bug-glpk] [Fwd: preprocessing goes into infinite loop]

[Andrew Makhorin]

> 1) If I set the upper bound to DBL_MAX, I encounter the problem. Am I
>  supposed to set an upper bound of infinity differently?

Using DBL_MAX is a bad idea, because if such value is used as a finite
bound, this immediately causes numerical difficulties.

> 
> 2) If not, are you saying that GLPK cannot solve this kind of unbounded
>  integer problem in general?

Glpk cannot preprocess it correctly due to insufficient robustness of
the mip preprocessor. Usually the preprocessor is able to discover
implied bounds of integer variables, in which case the bug does not
appear.

> 
> 3) Since glp_simplex discovers that there is no solution without
>  crashing, would it be advisable to solve by invoking glp_simplex
>  first, then invoking glp_intopt only if continuous solutions exist? or
>  is this horribly inefficient?
> 

Inefficient, but not horribly, because solution of lp relaxation takes a
small percentage of the total time needed to solve mip.