Re[3]: [Help-glpk] glpk 4.5 release information

[Andrew Makhorin]

>There are a couple of things I an interested in:
>
>     - the ability to terminate an integer solution when it is, eg, within 
>10% of the real solution. Is this any easier in this version?

Currently this feature is not implemented yet. However, it is easy to
include additional termination tests directly in the b&b solver. If you
need to do that, please inform me, and I will provide you with all
necessary information.

>     - there was a stability patch by Robert Horvath for 4.4. Is it applied 
>and/or still necessary?

I found your mip instance in the mailing list archive, which causes
glpsol to fall into an infinite loop. This happens because the
objective row is badly scaled. (Near optimal point of LP relaxation all
reduced costs are close to zero, so the simplex solver can test only
absolute tolerances. If objective coefficients are very large in
magnitude as in your instance, the test fails though the solution is
actually optimal. This also causes numerical instability in the dual
simplex.) I just scaled the objective row in your instance multiplying
it by 1e-6 (please see the attachment), and the error disappeared.

Unfortunately the patch provided by Robert Horvath cannot be applied
to the version 4.5 due to cardinal changes in api data structures made
in 4.4. This patch, however, resolves another problem which does not
appear in your case.

Andrew Makhorin

last_problem_int.lpt
Description: Binary data