[Coin-discuss] OsiClp* interior point stability problem

[Sebastian Nowozin]

Hello Laszlo,

On Dec 26, 2007 9:21 PM, Laszlo Ladanyi <ladanyi at us.ibm.com> wrote:

> I don't know how stable the Clp interior solver is (John Forrest is the man to
> answer that question), but you might want to try to use the Volume Algorithm.
> It does have an OSI interface (OsiVol). It gets near the center of the optimal
> face (within 2-5%) much faster than any interior point solver, and that
> accuracy is usually enough to do column generation. And when convergence
> starts to slow down you can switch to Clp's simplex solver.

Thanks a lot for this information, I will definitely try out the
volume algorithm, it sounds really promising for this purpose (and its
a single line source code change, thanks to OSI).

One question: does the OsiVol* interface and the volume algorithm
implementation support warmstarts? (primal infeasible warm starts,
after a violated constraint/cut has been added)
And is this warm start capability accessible from the OsiVol*
interface?  Is there some example in the source code I can take a look
at?  The OsiVol* doxygen documentation is very brief on that.

Thanks,
Sebastian