[Bug-glpk] Re: [Pkg-scicomp-devel] Bug#490288: glpk: Fails on assertion

[Andrew Makhorin]

Saturday, July 12, 2008, 5:56:31 PM, you wrote:

> * Andrew Makhorin <address@hidden> [2008-07-12 16:06]:

>> The failure appears due to insufficient robustness of the glpk mip
>> solver. It is mainly caused by unbounded integer variables (x1 and x2
>> have no upper bound) having relatively large coefficients in the
>> constraint that leads to excessive round-off errors on solving LP
>> relaxation.
>> 
>> [sinp]
>> 
>> Nevertheless, in both cases the problem is badly formulated. Many
>> reseachers do not recommend to use upper bounds of integer variables
>> which exceed 100.

> Does this mean that the problem is in the formulation of that specific
> example or is it a bug in GLPK? If the later is the case, are you
> planning to fix it?

It is a bug in the sense that the mip solver should return GLP_EFAIL
rather than cause abnormal termination. The bug will be fixed in a next
version of the package.


Andrew Makhorin