[Scip] Replications

[Benjamin Müller]

Dear John,

if your NLP relaxation has a unique optimum, you will always end up with 
the same fractional solution. So maybe a possible way to go is to change 
your relaxation in each replication slightly.

I could think of considering an MINLP relaxation of your problem. If x 
is an integer variable in the original problem and has a fractional 
value in your NLP solution, then you could add the integrality 
requirement for this variable and try to solve the resulting MINLP 
relaxation. Maybe it is possible to receive some good candidates from 
your heuristic?

Alternatively, if x* is the optimal fractional solution with objective 
value U, then you could add a constraint c'x = U and set a different 
objective function to hopefully find a different fractional solution.


Best regards,

Benny



On 10.05.2015 09:26, John Beasley wrote:
>
> I have a hard MINLP for which I explicitly relax the integrality 
> constraints and terminate SCIP after a set time limit, taking the best 
> (fractional) solution found and apply a heuristic to (hopefully) 
> generate a feasible integer solution.
>
> Obviously running replications (so repeating the process) may improve 
> the (heuristic integer) solution found. What would be a good way to 
> replicate the situation, ideally forcing SCIP to follow a different 
> solution path from the previous replication to arrive at a different 
> fractional solution?
>
> John Beasley
>
> ------------------------------
>
> John Beasley
>
> Emeritus Professor
> Mathematical Sciences
> Brunel University
> Uxbridge UB8 3PH
> UK
>
> http://people.brunel.ac.uk/~mastjjb/jeb/jeb.html
>
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20150510/5bf01f24/attachment.html>