[Clp] Did CLP produce an incorrect solution?

[sergey pankov]

Hi John,

Thank you very much, it all now makes sense. Great relief!

I have been checking the primary status, but will start checking the
secondary status as well. Indeed, the secondary status is 2 in my example,
unless I turn scaling off. I may go with this fix. My data is produced by a
(simulated) physical system. Small numbers correspond to small changes in
the state of the system. It is the errors in absolute values of the
parameters, not relative values, that matter. So turning scaling off is OK.

I so far had great time using CLP, it is very fast for my purpose. Thanks a
lot for your great work!

Sergey


2016-10-07 1:36 GMT-07:00 John Forrest <john.forrest at fastercoin.com>:

> Sergey,
>
> A failing in standalone Clp - it should have warned you that the scaled
> problem was optimal while the unscaled one was not feasible - the solution
> from that does indicate column 0 is infeasible.  If you are calling clp
> then check primary status and also secondary status - it would have been 2 -
>
>          2 - scaled problem optimal - unscaled problem has primal
> infeasibilities
>
> Clp saw that the elements of the matrix were very very small and did some
> scaling to make the problem more stable.  Unhappily the scale factors for
> the first two columns were greater than 1.0e9!!  So it thought that
> -53.8/1.0e9 was a small negative value within tolerance.
>
> The quick fix is to turn scaling off.  However it is never a good idea to
> have elements such as 1.2e-11 - maybe better to change units to multiply
> all constraints by 1.0e3 or 1.0e6.
>
> John Forrest
>
> On 07/10/16 02:41, sergey pankov wrote:
>
> Hi,
>
> I am new to CLP (and to this mailing list). I have been using CLP in my
> code for several days and encountered a LP program that CLP thinks it
> solved successfully, but the solution is out of bounds. I am using a
> ClpSimplex model (as in minimum.cpp example). Here is the solution
> (columns):
>
>                 Primal          Dual         Lower         Upper
> Cost
>      0 c      -53.8172       0.00000       0.00000  1.79769e+308
> 1.32630e-06
>      1 c       49.9853       0.00000       0.00000  1.79769e+308
> 1.01542e-06
>      2 c     0.0943604       0.00000       0.00000  1.79769e+308
> 0.00758213
>      3 c     0.0731115       0.00000       0.00000  1.79769e+308
> 0.0170572
>      4 c       1.61406       0.00000       0.00000  1.79769e+308
> 0.000855447
>      5 c   7.37729e-07       0.00000       0.00000  1.79769e+308
> 50.0000
>
> As you can see, the first value is negative, below the lower bound. I
> saved the model to a .mps file, it is attached. Am I doing something wrong?
> Is it me or CLP?
>
> Any help is highly appreciated, thank you very much in advance.
>
> Sergey
>
>
> _______________________________________________
> Clp mailing listClp at list.coin-or.orghttps://urldefense.proofpoint.com/v2/url?u=http-3A__list.coin-2Dor.org_mailman_listinfo_clp&d=CwICAg&c=Ngd-ta5yRYsqeUsEDgxhcqsYYY1Xs5ogLxWPA_2Wlc4&r=js2M0T-3OIMIVDvokcKjokJbk0F8QOCd0mT4FsVFE88&m=D3zjoubDmAt3DDkzmmw59m0w3-ML57HiYbB3joLhYQo&s=Pn6zzITq6UsgiBWWjy2K5lZ5ihqVlPECnn9Kdyiw9AA&e=
>
>
>
> _______________________________________________
> Clp mailing list
> Clp at list.coin-or.org
> https://urldefense.proofpoint.com/v2/url?u=http-3A__list.
> coin-2Dor.org_mailman_listinfo_clp&d=CwICAg&c=Ngd-
> ta5yRYsqeUsEDgxhcqsYYY1Xs5ogLxWPA_2Wlc4&r=_Kk1SZ84F4GzRFbFYzN-
> 7vUZcYEXtVdaNEKqx0mUQHQ&m=-Zn7cZqLC1wotP7-I2_GLAbjErAg3veFIM0EwLd63Ac&s=
> c3c0Bixe0Vz-DhzTrpYmNi2r_Io4_hfXysiseT1oei8&e=
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/clp/attachments/20161007/632b7b17/attachment.html>