[Coin-bcpdiscuss] LP Lower Bound

Laszlo Ladanyi ladanyi at us.ibm.com
Tue Jan 24 22:01:55 EST 2006


Hi Carlos,

I usually do this the following way:

I actually generate the variables in the BCP_lp_user::compute_lower_bound()
method, not in the method asking to generate vars. I store these vars in my
user class derived from BCP_lp_user. This way I can return a true lower bound
in compute_lower_bound(). The simplest is to return -infinity if I could
generate vars, and return the LP value otherwise, but for most problem classes
you can actually derive a bound. Then when the generate_vars() method is
invoked I simply return the vars I have already generated. 

Let me know if this is not clear.

--Laci

On Tue, 24 Jan 2006, Carlos Eduardo de Andrade wrote:

> Hi,
> 
> I have implemented a branch-and-price algo, and I should want to use
> the first LP relaxed solution value as lower bound. This solution is
> obtained when we can't generate any more column and we have to branch.
> How can I use this information as lower bound?
> 
> Bcp call the BCP_lp_user::computer_lower_bound() each iteration, but
> it isn't correct for my objectives. I have to take the lower bound in
> last iteration of root node.
> 
> Can somebody help me???
> 
> Thank you very much,
> 
> Carlos
> 
> --
> Carlos Eduardo de Andrade
> Computing Institute - UNICAMP
> Brazil
> 
> _______________________________________________
> Coin-bcpdiscuss mailing list
> Coin-bcpdiscuss at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-bcpdiscuss
> 




More information about the Coin-bcpdiscuss mailing list