[SCIP] Assertion error with barrier algorithm in debug mode

[Marc Pfetsch]

Hi Selin,

I think we would need more details to help you. For instance:

- What is the final status of Gurobi (lpi->solstat)?

- Can we somehow reproduce your run?

- Can you turn on the output of Gurobi to see whether it gives you an 
indication of what is going wrong. There is a SCIP parameter 
"display/lpinfo" which you can set to "true" to turn the output on.


Please also consider upgrading your SCIP version. This is quite old by now.

Best

Marc

On 17/06/2025 20:12, Bayramoglu, Selin wrote:
> Hi all,
> 
> I'm working on a custom cut selector in SCIP 8.0.0 and using the barrier 
> algorithm in diving mode, following the method from this GitHub repo 
> <https://github.com/Opt-Mucca/Analytic-Center-Cut-Selection>. 
> Specifically, I compute the analytic centers of the optimal faceand the 
> LP polyhedron at the current node by doubling the original objective and 
> setting it to 0 in divingmode, respectively.
> 
> I use Gurobi as the LP solver and apply the following diving settings:
> 
> lp/initalgorithm      = b
> 
> lp/resolvealgorithm   = b
> 
> lp/checkdualfeas      = FALSE
> 
> lp/disablecutoff      = 1
> 
> lp/solutionpolishing  = 0
> 
> lp/checkstability     = FALSE
> 
> lp/checkfarkas        = FALSE
> 
> lp/presolving         = FALSE  // added by me
> 
> lp/scaling            = 0      // added by me
> 
> Everything runs fine in normal mode, but in debug mode I hit this 
> assertion error:
> 
> scip/src/lpi/lpi_grb.c:4116: SCIPlpiGetObjval: Assertion `lpi->solstat ! 
> = GRB_OPTIMAL || oval == obnd' failed.
> 
> oval holds the correct objective value, but obnd is -1e+100.
> 
> Any ideas on what might be causing this? If there is a better selection 
> of diving settings, I’d be happy to learn that as well.
> 
> Best regards,
> 
> *Selin Bayramoglu*
> 
> Ph.D. Student
> 
> *H. Milton Stewart School of Industrial and Systems Engineering*
> Georgia Institute of Technology
> 
> 
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip