[Coin-lpsolver] Crash in BCP/CLP

[Jörg Herbers]

You're right. BCP imposes a bound on the objective, but in the next 
iteration, the newly generated columns don't yield an improvement. In this 
case, CLP returns an "empty" dual ray. With BCP_SolveLpToOptimality=1, 
everything works.

Actually, I don't extract the rays for column generation. The problem with 
the "empty" ray was in BCP when BCP_restore_feasibility tries to find an 
indexed variable (which I use to guarantee that a feasible solution can 
always be found) which cuts a dual ray.

Thanks a lot for your help!
Jörg


>John,
>
>I'll drop by tomorrow to discuss this.
>
>Jorg,
>
>In a column generation approach I'd strongly recommend to solve the LPs to
>optimality (add "BCP_SolveLpToOptimality 1" to the parameter file). 
>Otherwise
>the LP will probably stop after the dual objective exceeds the current 
>bound
>(like clp does). Also, it depends on the solver what status it reports in 
>such
>case. In any case, the dual ray you can extract from the solver at that 
>point
>(if you get anything at all, seems like you can't from clp yet) is 
>completely
>worthless for generating new colums. The generated columns are highly 
>unlikely
>to appear in a solution.
>
>--Laci
>
>On Mon, 15 Sep 2003, John J Forrest wrote:
>
>>
>>
>>
>>
>>Dual rays are implemented - I will have to get together with Laci to see
>>what can go wrong.
>>
>>Had you got a valid solution before this happened?  If DualObjectiveLimit
>>is reached then clp would return a status of 1 but could not create a ray
>>as the problem may be feasible but just with an optimal value worse than
>>best solution found so far.  I may have to add something to be able to 
>>tell
>>the two cases apart.
>>
>>John Forrest
>>
>>
>>                                                                           
>>               "Jörg Herbers"                                              
>>                <herbers at hotmail.                                          
>>                 com>                                                      
>>  To              Sent by:                  
>>coin-lpsolver at oss.software.ibm.com               coin-lpsolver-adm         
>>                                  cc              in at www-124.ibm.co        
>>                                                   m                       
>>                               Subject                                     
>>    [Coin-lpsolver] Crash in BCP/CLP                                       
>>                                                      09/15/2003 08:28     
>>                                                       AM
>>
>>
>>
>>
>>In one iteration of my BCP application, the problem becomes primal
>>infeasible (termcode 36). Via indexed variables (create_indexed_var), BCP
>>tries to regain feasibility (in BCP_restore_feasibility). But in a
>>multiplication with the dual rays, the program crashes because the only
>>dual
>>ray returned by CLP (called by BCP_lp_fathom) is 0x0 - even if CLP's
>>problemStatus_ is 1. In fact, CLP's ray_ is 0x0.
>>
>>Since I don't have a deep insight into the CLP code: Isn't the dual ray
>>feature implemented yet? Could I have missed the setting of any 
>>parameters?
>>
>>Or is this a bug?
>>
>>Many thanks,
>>Jörg
>>
>>_________________________________________________________________
>>MSN 8 helps eliminate e-mail viruses. Get 2 months FREE*.
>>http://join.msn.com/?page=features/virus
>>
>>_______________________________________________
>>Coin-lpsolver mailing list
>>Coin-lpsolver at www-124.ibm.com 
>>http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-lpsolver
>>
>>
>>
>
>_______________________________________________
>Coin-lpsolver mailing list
>Coin-lpsolver at www-124.ibm.com
>

_________________________________________________________________
Add photos to your e-mail with MSN 8. Get 2 months FREE*. 
http://join.msn.com/?page=features/featuredemail