[SCIP] Using barrier solver when linked to Gurobi?

[Marc Pfetsch]

Hi Christian,

it seems that the barrier has some problem. You can get more information
by setting "display/lpinfo = true" within SCIP (either through a
parameter file or in the command line). Then the lp-solver will produce
output for each LP. Maybe you can spot what the problem is.

Two more notes:

- The warning "WARNING: LP solution value is above SCIP's infinity
value" is likely to come from the fact that the infinity value of Gurobi
is 1e30, while that for SCIP is 1e20.

- The interface to the barrier is definitely not as well-tested as the
simplex interfaces (because warmstart is more difficult, this option is
rarely chosen without crossover). So, there might still be a problem in
the interface. If this seems to be the case, you can write an email
directly to me or submit a bug report.

Best

Marc

On 28/06/2020 21:12, Franzen, Christian wrote:
> I have compiled SCIP 7.0.0 against Gurobi 9.0.0 and I am trying to use
> Barrier method WITHOUT crossover for solving my node relaxations.
> However, I always receive the following error:
> 
> 
> 1/1 feasible solution given by solution candidate storage, new primal
> bound 2.919805e+10
> 
> WARNING: LP solution value is above SCIP's infinity value
> feasible solution found by completesol heuristic after 1.0 seconds,
> objective value 2.657620e+09
> presolving:
> presolving (1 rounds: 1 fast, 1 medium, 1 exhaustive):
>  0 deleted vars, 0 deleted constraints, 0 added constraints, 0 tightened
> bounds, 0 added holes, 0 changed sides, 0 changed coefficients
>  0 implications, 0 cliques
> presolved problem has 28565 variables (0 bin, 0 int, 0 impl, 28565 cont)
> and 24445 constraints
>   24445 constraints of type <linear>
> Presolving Time: 1.00
> transformed 3/3 original solutions to the transformed problem space
> 
> WARNING: LP solution value is above SCIP's infinity value
> (node 1) unresolved numerical troubles in LP 0 -- using pseudo solution
> instead (loop 1)
> (node 1) forcing the solution of an LP (last LP 0)...
> [solve.c:3910] ERROR: (node 1) unresolved numerical troubles in LP 0
> cannot be dealt with
> [solve.c:4195] ERROR: Error <-6> in function call
> [solve.c:4989] ERROR: Error <-6> in function call
> [scip_solve.c:2653] ERROR: Error <-6> in function call
> 
> Any idea what is going on there? Dual simplex, primal simplex and
> barrier WITH crossover ist working just fine. It seems to be not related
> to my LP files, I have tested at least 5 of them, all with the same error.
> 
> Regards
> Christian
> 
> 
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
>