[Clp] bestBasicSequence, pivotRow and theta

[John J Forrest]

I sent this a few days ago - but it did not seem to appear.

---------
The problem is mainly one of standard form of input for a quadratic
objective - cTx + 1/2 Tx Q x

MPS format seems to be different from LP format.

So in this case Clp is given an unbounded  problem which is not positive
definite - and makes a mess of it.

If I scale the off diagonal elements in the problem by 0.5 then barrier
gives the same correct answer quickly.  Simplex still does not like it and
I may look into that.  If I scale back objective by 1.0e-6 then simplex
also solves it.

John Forrest





                                                                                                                 
  From:       Hok Go <Hok.Go at qrm.com>                                                                            
                                                                                                                 
  To:         "'clp at list.coin-or.org'" <clp at list.coin-or.org>                                                    
                                                                                                                 
  Date:       05/14/2009 11:50 AM                                                                                
                                                                                                                 
  Subject:    [Clp] bestBasicSequence, pivotRow and theta                                                        
                                                                                                                 





Hi all,

Running into a problem in ClpSimplexNonlinear where pivotColumn finds a
theta of 1e30, bestBasicSequence of -1 and pivotRow of -1. This also causes
problems when we start reading from these locations. I’ve attempted
modifications to the perturbation without success and was wondering whether
anyone’s run into this problem and knows a way around it. Another thing
I’ve tried is scaling the coefficients.

On a side note the barrier solver does not run into memory problems but
ends after 201 iterations with a  stopped objective. I’ve also verified
that other solvers can produce a correct answer to this problem and I’ve
appended that solution.

Anyone got any ideas?

Cheers,
Hok

Presolve 2 (0) rows, 21 (0) columns and 42 (0) elements
0  Obj 0 Primal inf 55.6677 (2) Dual inf 0.153707 (21)
2  Obj 0
End of values pass after 2 iterations
2  Obj 0
2  Obj 0
Optimal - objective value 0
0  Obj -5.71266e+017 Dual inf 1.40189e+015 (19)


STATE
optimal
OBJECTIVE
5483597535069.56
SOLUTION
-173.941591973461
252.228299131887
645.768536126273
-479.002236942252
0.372648640617555
-12.3137632778437
1.09779599200074
-0.0512981461901408
-4.95954705298369
-0.662623008913045
-2.48548076636163
-1.3484270133586
0.149873781527723
-0.0198378539000891
4.477576268655
0.47567076116259
0.14752851966193
0.0421460826706324
-0.748317699220338
0.33517340397477
0.300487415965592
 [attachment "badlyformed.mps" deleted by John J Forrest/Watson/IBM]
_______________________________________________
Clp mailing list
Clp at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/clp

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/clp/attachments/20090519/03e4a981/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: graycol.gif
Type: image/gif
Size: 105 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/clp/attachments/20090519/03e4a981/attachment.gif>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ecblank.gif
Type: image/gif
Size: 45 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/clp/attachments/20090519/03e4a981/attachment-0001.gif>