[Coin-lpsolver] ClpSimplex::crash () aborts on an MPS with free variables

Ariel Landau ariel1127 at gmail.com
Thu Mar 1 17:21:07 EST 2007


Hello!

I've recently come across an MPS file that causes ClpSimplex::crash() to
call abort ().
The exact place of the abort call is shown in the following excerpt from
ClpSimplex.cpp (with row numbers as of March 1st)

6405            switch(getColumnStatus(iColumn)) {
6406
6407            case basic:
6408              thisWay=0;
6409             case isFixed:
6410              break;
6411            case isFree:
6412            case superBasic:
6413              if (djValue<-dualTolerance)
6414            thisWay = 1;
6415              else if (djValue>dualTolerance)
6416            thisWay = -1;
6417              else
6418            { thisWay =0; abort();} <--------------- EXECUTION ABORTS
HERE WITH djValue == 0 && dualTolerance == 1e-07
6419              break;
6420            case atUpperBound:
6421              if (djValue>dualTolerance)
6422            { thisWay =-1; abort();}
6423              else if (djValue<-dualTolerance)
6424            thisWay = -3;
6425              else
6426            thisWay = -2;
6427              break;
6428            case atLowerBound:
6429              if (djValue<-dualTolerance)
6430            { thisWay =1; abort();}
6431              else if (djValue>dualTolerance)
6432            thisWay = 3;
6433              else
6434            thisWay = 2;
6435              break;
6436            }

For anyone that wants to take a look, my MPS is available online at
http://ar.geocities.com/ariellandau/test_mps.zip
As it might be expected from the code excerpt above, my MPS does indeed
contain a bunch of FR variable bounds.
And if I replace these FR bounds by very large negative lower bounds (e.g.,
larger in magnitude than any of my negative variables at the optimal
solution) the abort disappears.

Any insight will be much appreciated.

Sincerely,

Ariel Landau
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/clp/attachments/20070302/e2a1bdcb/attachment.html>


More information about the Clp mailing list