Re: [Help-glpk] Working with larger numbers

[Markus Pilz]

Andrew Makhorin wrote:
> [...]
> If large bounds are used to indicate that corresponding arcs have
> unlimited capacity, this is not a good idea; it is more correct to
> remove such bounds at all. If not, and if you really expect that
> some arc flows may reach their upper bounds, which are large, you
> might consider scaling such flows.

All arcs should be bounded, i.e. capacity is always limited.

Can I use glpk routines to perform scaling? Or should I scale my data
beforehand? (Most likely I get the answer "it depends...")

> 
> On the other hand, max flow and min cost flow are pure network
> problems having nice numeric properties. Does your model have only
> flow conservation constraints? If not, which additional constraints
> it has?

Max flow and min cost are the basic problems. In the next step the
approach should is used for multi-commodity scenarios. Therefore, we
started with a LP solver. Do you propose an other direction to solve an
mcf?

> 
> (I also would like to note that the general simplex method implemented
> in glpk is much slower in solving network problems than specific network
> optimization algorithms.)

Absolutely.

> 
>> P.S.: We are using the glpk API to create the problem and to run the
>> simplex method. Therefore, I omitted mathprog code.
> 
> Could you write your model in mps or cplex format, gzip it, and post
> it to me (up to 10 Mb is ok)? I mean the instance, for which you
> obtained inexact results.

I've prepared two small scenarios and send them directely to you.


Markus

> 
> Andrew Makhorin
> 


-- 
______________________________________________________________________
Markus Pilz                          University of Bonn
                                     Institute of Computer Science IV
E-Mail: address@hidden          Roemerstrasse 164
Tel.:   +49 228 73-4549              53117 Bonn
Fax.:   +49 228 73-4571              Germany