[SCIP] SCIPgetVarRedcost

[Stefan Vigerske]

Hi,

the reason that reduced costs are available for an LP when presolve is 
disabled is that in this case, the LP relaxation in SCIP is equivalent 
to the problem. So, you can take the dual solution that were computed by 
the LP solver for the LP relaxation as dual solution for the problem itself.
If you don't give an LP to SCIP, then the reduced costs you may be able 
to obtain in some ways are the ones of an LP relaxation of your problem. 
They are very like not connected to what you expect the reduced costs 
for a MIQCP to be, whatever that is.
The reason you may be able to obtain values is that some plugins of SCIP 
want to access the reduced costs in the LP relaxation and there are no 
checks that prevent you from using the same calls.

Stefan

On 22/08/2023 17:18, Peter Notebaert wrote:
> I have a small model that contains one integer variable and it contains one
> quadratic constraint along with other real variables and linear constraints.
> 
> When solved via the scip command and asking to display the dual solution it
> says that this information is not available when presolve is active. When
> presolve is disabled (set presolving emphasis off) it says that dual
> information is only available for pure LPs.
> 
> However I am now interested in the variable reduced costs. I don't see a
> way to display this via the scip command. Is that correct?
> 
> Anyway in fact the model is build in code and I use SCIPgetVarRedcost to
> get the reduced costs.
> I see that this information is returned even with presolve on and with the
> integer variable.
> I also note that I get totally different values when presolve is on or off
> And I even get non-zero values for basic variables which is also quite
> strange I think.
> 
> So my question. Is this call possible if there are integer variables /
> quadratic constraints and is it possible when presolve is active and if
> not, why do I get values? Oh and is there a way to show these values via
> the scip command?
> 
> Thanks,
> 
> Peter
> 
> 
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip