Re: [Help-glpk] binary variables writing lp format; and MIP bounds

[Andrew Makhorin]

>1.  A student I am working with noticed that when writing output in LP
>format, only "General" integer variables are specified, even if they are
>really binary variables.  Is there a reason for omitting the binary
>case?  If not, I'll go ahead and change the code and submit a possible
>correction.

Glpk does not distinguish between integer and binary variables, so
there is no need to have an additional flag. (Note that in MPS one also
can specify binary variables using intorg/intend markers, i.e. as
general integer variables, rather than specifying BV type.)

>2.  IOS seems much better than the previous branch-and-bound solver.  I
>would like to improve the LP bound, however.

If you mean mip_driver, it uses its own data structures, not IOS
(although these data structures is a simplified version of IOS which
does not allow adding/deleting rows and columns).

>  At present, the LP bound
>at the pseudo-root is used.  In fact, the maximum (or minimum) of the LP
>bounds at active nodes could be used and reported.  I would like to add
>this change as well as including a standard option for specifying an
>integrality gap for termination of the algorithm.  Is there any reason
>not to do this?
>
>If there's no obvious reason not to, then I'll work on these changes and
>send them in,

What does "the maximum/minimum of the LP bounds at active nodes could be
used and reported" mean? Used for what? Please explain.

Andrew Makhorin