On Sun, Jul 10, 2011 at 11:34 AM, Andrew Makhorin
<address@hidden> wrote:
> I changed the code and all the variables now have fixed upper bound
> ( 1000 units). But I am still getting the Numerical instability
> warnings.
Numerical instability may happen because of badly scaled problems, so
try to scale your initial lp instance with glp_scale_prob (do not use
glp_scale_prob before reoptimization). Please note that the message
about numerical instability is a warning; if glp_simplex solves to the
end, you may ignore that message.
> In my program, the size of the model increases in each solve. Is the
> instability because of the increasing problem size or accumulated
> round-off errors. If latter is the case, is there any way to determine
> the instability and reset the model when instability is present.
>
In the simplex method round-off errors are not accumulated. However,
almost linearly dependent rows added to the current formulation may
cause numerical instability.