[Scip] using gurobi as lp solver

Marc Pfetsch m.pfetsch at tu-bs.de
Mon Nov 30 13:22:48 MET 2009


Hi Robert,

Gurobi is not officially supported by SCIP. The interface is beta and 
not really tested. In particular, we have not updated the changes in the 
  interface for the recent versions (Gurobi 2 was not officially 
available at the time of the release of SCIP 1.2).

Anyway, the real problem is that the interface of Gurobi does not 
support some crucial functionality: In particular, it does not support 
infeasibility rays (the "Dual-Farkas" that you mention). Moreover, it 
does not support access to the basis inverse. This is a pity, because I 
am sure that the functionality is there, but the interface does not 
allow to access it.

As a result, using the current version of Gurobi with SCIP is no fun, 
because all the advanced functionality is useless. I hope that the 
Gurobi interface will change soon - we will then update SCIP as well.

In the meantime you have to rely on a different LP-solver like SoPlex or 
Clp (or Cplex if you have a licence).

Best,

Marc


Robert Schwarz schrieb:
> Hey all,
> 
> is Gurobi officially supported to be interfaced with SCIP 1.2.0?
> The Makefile recognizes the LPS=grb option, but I'm getting build errors
> for my application:
> 
> -> linking bin/scip-1.2.0.linux.x86_64.gnu.opt.grb
> lib/liblpigrb-1.2.0.linux.x86_64.gnu.opt.a(lpi_grb.o): In function
> `SCIPlpiWriteLP':
> lpi_grb.c:(.text+0x362): undefined reference to `GRBwrite'
> lpi_grb.c:(.text+0x38a): undefined reference to `GRBgeterrormsg'
> 
> And yet I can build scip itself with
>   make LPS=grb
> and run some of the test IPs with a speed up of 2 compared to soplex.
> Still, there are many errors:
> [src/scip/lpi_grb.c:2796] ERROR: SCIPlpiGetDualfarkas() not supported by
> Gurobi
> 
> I'm using Gurobi 2.0.0 for 64 bit linux. In the Gurobi docs I read that
> one should use gcc 4.3 for interfacing it, so I run g++-4.3 (GCC) 4.3.3
> 
> When building scip and setting up the links, I only link to the .so library.
> 
> Any ideas?
> 
> Thanks,
> 



More information about the Scip mailing list