[Scip] Could a global cut be discarded ?

[nikolaj@crt.umontreal.ca]

Hi Tobias and everybody,

Thanks Tobias for your help.

I checked numerical instabilities and I didn't find any, i.e. the cut is
really violated (first time by 0,5; second time by 0,15).

When I write the LP with SCIPwriteLP(), I can see that during the search
some cuts are added and removed later. Is this normal behavior?

Could it be that a cut is only valid from a node and its descendants in
the B&B tree and not for all the other nodes treated later on?

Thanks in advance,

Nikolaj


On Thu, July 2, 2009 4:27 am, Tobias Achterberg wrote:
> Hi Nikolaj,
>
>
> what you are doing is correct for what you want. You have set the
> 'removable' flag to
> FALSE in the SCIPcreateEmptyRow() call in order to prevent SCIP from
> removing the row from the LP. Additionally, you have set the 'forcecut'
> flag to TRUE in the SCIPaddCut() call. This should make sure that the cut
> actually enters the LP and survives the cut filtering. Therefore, the cut
> should enter and stay forever in the LP relaxation.
>
> May it be that the cut is not violated, or only violated by a very small
> amount, when you add it a second time?
>
> In order to debug this you could call SCIPwriteLP() from your separator
> at the time the cut is added a second time and verify that the first cut
> is in the LP. Additionally, you should print the violation of the cut.
> Maybe, it is just a numerical issue.
>
>
>
> Tobias
>
>
>
> nikolaj at crt.umontreal.ca wrote:
>> Hello everybody,
>>
>>
>> Thanks for your precious help last time.
>>
>>
>> In a branch and cut, I add several global cuts in a separator. For one
>> graph, I've found that after I add one global cut, it was discovered
>> again two times. When I print the added cuts with SCIPprintRow I get
>> exactly the same cut. Could a global cut be discarded or else where did
>> I go wrong?
>>
>>
>> Here is how I add the global cut (in the method scip_execlp) :
>>
>>
>> ------------------------------------
>> SCIP_ROW* cut;
>> std::ostringstream namebuf;
>>
>>
>> assert(sepa != NULL); assert(strcmp(SCIPsepaGetName(sepa),
>> this->scip_name_) == 0); assert(scip != NULL); assert(result != NULL);
>>
>> *result = SCIP_DIDNOTRUN;
>>
>>
>> //only call separator, if an optimal LP solution is at hand
>> if( SCIPgetLPSolstat(scip) != SCIP_LPSOLSTAT_OPTIMAL ) { return
>> SCIP_OKAY;
>> }
>>
>>
>> *result = SCIP_DIDNOTFIND;
>>
>>
>> //main loop
>> //omitted here
>>
>>
>> //create cut
>> namebuf.str(""); namebuf << "generalized_odd_cut"; SCIP_CALL_EXC(
>> SCIPcreateEmptyRow(scip, &cut, namebuf.str().c_str(), 1.0,
>> SCIPinfinity(scip),FALSE, FALSE, FALSE) );
>> SCIP_CALL_EXC( SCIPcacheRowExtensions(scip, cut) );
>>
>>
>> //populate cut
>> //loops omitted here, I add variables in the cut with:
>> SCIP_CALL_EXC( SCIPaddVarToRow(scip, cut,
>> model_->y_[*u_iter][*v_iter][k], 1.0) );
>>
>>
>> SCIP_CALL_EXC( SCIPflushRowExtensions(scip, cut) );
>> SCIP_CALL_EXC( SCIPaddCut(scip, NULL, cut, TRUE) );
>> SCIP_CALL_EXC( SCIPaddPoolCut(scip, cut) );
>>
>>
>> SCIPprintRow(scip,cut,NULL);
>>
>>
>> *result = SCIP_SEPARATED;
>>
>>
>> SCIP_CALL_EXC( SCIPreleaseRow(scip, &cut) );
>> ------------------------------------
>>
>>
>>
>> Any help would be much appreciated.
>>
>>
>> Best regards,
>>
>>
>> Nikolaj
>>
>>
>>
>>
>> _______________________________________________
>> Scip mailing list
>> Scip at zib.de
>> http://listserv.zib.de/mailman/listinfo/scip
>>
>