If I keep the presolver on, the basis information from the previous
solve is lost and the 2-phase primal algorithm kicks off from scratch
(instead of the dual simplex using the previous basis). If I turn the
presolver off, I know my problem is bigger than it needs to be, but
the dual simplex does kick in starting with the previously optimal
basis. So I guess my question is: is there a way to use the
presolver AND supply an initial basic solution?
No, the lp presolver does not use the current basis information,
so you should disable it on performing re-optimization.
If not, is there a
mathematical/algorithmic reason this isn't possible?
There is not much sense to do that, because re-optimization needs
much less iterations. Besides, in that case the lp presolver could not
remove redundant basic rows and redundant non-basic columns.
If so, I'd be
interested in hearing it because it will probably change my approach
(and understanding) of my problem. My understanding from the glpk
code is that the presolver uses a copy of the original problem to
perform all of it transformations and then the solution to the
presolved problem is translated back to the original problem when the
simplex completes.
Correct.
This transformation must interfere with any basis
that the original problem had, thus eliminating it I suppose? Hope
all of these questions make sense.
I'll gladly supply more info if it's helpful. Thanks in advance for
any insights.
___
Help-glpk mailing list
Help-glpk@gnu.org
http://lists.gnu.org/mailman/listinfo/help-glpk