I was having the same problem. I'm very grateful too for your anwser, Laszlo (to Mohamad too for raising the question!). I was not able to fix my code yet but at least I have some hints :-)<br><br>If I understood it well, the problem comes from adding "bound" constraints (0 <= x_i <= 1) since eventually one of those became non-basic and bounded to 1 with a non-negative dual and Bcp (Osi?) was not returning it in the dual price vector because it concerned the upper bound? Is my understanding good?<br>
<br>To complement this thread just say that the behavior I was observing was that cplex command line and glpsol (glpk's counterpart) were returning non-zero duals whereas Bcp (Osi?) was returning 0. <br><br>Thanks,<br>
<br>Dani.<br><br><div class="gmail_quote">On Fri, Apr 17, 2009 at 10:35 PM, mohamad reisi <span dir="ltr"><<a href="mailto:mohamad.reisi@in.iut.ac.ir">mohamad.reisi@in.iut.ac.ir</a>></span> wrote:<br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Dear Laci,<br>
<br>
You are right. I have set the upper bound of variables to DBL_MAX, then this problem solved. Nw it is ok.<br>
Thanks a lot.<br>
<div class="im"><br>
Mohamad.<br>
----- Original Message -----<br>
From: Laszlo Ladanyi <<a href="mailto:ladanyi@us.ibm.com">ladanyi@us.ibm.com</a>><br>
To: mohamad reisi <<a href="mailto:mohamad.reisi@in.iut.ac.ir">mohamad.reisi@in.iut.ac.ir</a>><br>
Cc: <a href="mailto:bcp@list.coin-or.org">bcp@list.coin-or.org</a><br>
</div><div><div></div><div class="h5">Sent: Fri, 17 Apr 2009 23:15:57 +0330 (IRST)<br>
Subject: Re: [BCP] Dual ray is wrong!<br>
<br>
Hi Mohamad,<br>
<br>
Then I'm almost sure that my explanation below is the correct one.<br>
Check if you have any variables out of basis at their upper bound. Add<br>
together their reduced costs and see if that is the difference.<br>
<br>
--Laci<br>
<br>
On Fri, 17 Apr 2009, mohamad reisi wrote:<br>
<br>
> Dear Laci,<br>
><br>
> I am using binary variables. The problem is that sum of the all dual<br>
> value must be equal to primal objecctive solution in optimality. But<br>
> in my case, these values are not equal, and somtimes have a large<br>
> difference.<br>
><br>
> Mohamad.<br>
><br>
> ----- Original Message -----<br>
> From: Laszlo Ladanyi <<a href="mailto:ladanyi@us.ibm.com">ladanyi@us.ibm.com</a>><br>
> To: mohamad reisi <<a href="mailto:mohamad.reisi@in.iut.ac.ir">mohamad.reisi@in.iut.ac.ir</a>><br>
> Cc: <a href="mailto:bcp@list.coin-or.org">bcp@list.coin-or.org</a><br>
> Sent: Fri, 17 Apr 2009 23:06:30 +0330 (IRST)<br>
> Subject: Re: [BCP] Dual ray is wrong!<br>
><br>
> Hi Mohamad,<br>
><br>
> Well, I need more information... Wrong in what sense? The negative of<br>
> what you expected? Or you keep regenerating the same column because<br>
> the duals do not change?<br>
><br>
> One common mistake in column generation schemes is that the newly<br>
> generated variables have a non-zero bound that can actually be<br>
> attained (like 0<=x<=1). If in the optimal solution of an LP<br>
> relaxation any such variable is out of basis at its upper bound then<br>
> you may have a nonzero dual value that corresponds to the upper bound<br>
> constraint (in this case the dual value is the reduced cost of the<br>
> variable) and this dual value is not in the returned dual ray. Since<br>
> usually for every generated variable there is a clique constraint that<br>
> contains that variable the clique constraint will implicitly enforce<br>
> the upper bound thus it's much better to just say 0<=x.<br>
><br>
> --Laci<br>
><br>
> On Fri, 17 Apr 2009, mohamad reisi wrote:<br>
><br>
>> Dear Laci,<br>
>><br>
>> As you said, I am doing column generation.<br>
>> I transefered my code to version 1.1.3. it seems that the dual value are a little better. But they are wrong yet. It is so strange for me! What can I do?<br>
>><br>
>> Mohamad.<br>
>> ----- Original Message -----<br>
>> From: Laszlo Ladanyi <<a href="mailto:ladanyi@us.ibm.com">ladanyi@us.ibm.com</a>><br>
>> To: mohamad reisi <<a href="mailto:mohamad.reisi@in.iut.ac.ir">mohamad.reisi@in.iut.ac.ir</a>><br>
>> Cc: <a href="mailto:bcp@list.coin-or.org">bcp@list.coin-or.org</a><br>
>> Sent: Fri, 17 Apr 2009 16:12:37 +0330 (IRST)<br>
>> Subject: Re: [BCP] Dual ray is wrong!<br>
>><br>
>> Hi Mohamad,<br>
>><br>
>> The cvs version is rather old... There has been relatively little<br>
>> change in the interfacesince then, but a fair amount under the hood,<br>
>> and especially clp/OsiClp has changed a lot. If you could port your<br>
>> code to version 1.1.3 and let us know if you still experience<br>
>> problems, that would be great. (Since you get the dual rays, I assume<br>
>> you are doung column generation. versions 1.2.x still have some column<br>
>> generation related bugs, that's why I suggested 1.1.3.)<br>
>><br>
>> Thanks,<br>
>> --Laci<br>
>><br>
>> On Fri, 17 Apr 2009, mohamad reisi wrote:<br>
>><br>
>>> Dear all,<br>
>>><br>
>>> I have developed an example in Bcp-cvs2006 package, nased on AAP example.<br>
>>> But in this example the dual ray and sum of them are wrong. (by using clp)<br>
>>> I want to know that this problem is because of Bcp-CVS package or not?<br>
>>> I have checked every thing but it seems that my codes are right!<br>
>>><br>
>>> Thanks.<br>
>>> Mohamad Reisi.<br>
>>><br>
>>> _______________________________________________<br>
>>> BCP mailing list<br>
>>> <a href="mailto:BCP@list.coin-or.org">BCP@list.coin-or.org</a><br>
>>> <a href="http://list.coin-or.org/mailman/listinfo/bcp" target="_blank">http://list.coin-or.org/mailman/listinfo/bcp</a><br>
>>><br>
>><br>
>><br>
>><br>
><br>
><br>
><br>
<br>
<br>
_______________________________________________<br>
BCP mailing list<br>
<a href="mailto:BCP@list.coin-or.org">BCP@list.coin-or.org</a><br>
<a href="http://list.coin-or.org/mailman/listinfo/bcp" target="_blank">http://list.coin-or.org/mailman/listinfo/bcp</a><br>
</div></div></blockquote></div><br>