Re: [Help-glpk] Save solution in GLPK

[Andrew Makhorin]

Tuesday, June 5, 2007, 1:37:52 PM, you wrote:

> In fact, I implement a column generation.

> But, when I started again the GLPK simplex, the counter is
> incrementing.

The simplex iteration count is kept in the problem object. (It is the
control parameter LPX_K_ITCNT.)

> But I want to initialize my solver by the precedent solution obtained

> For example, I obtain:

>       0:   objval =   0.000000000e+00   infeas =   1.000000000e+00 (0)
>      15:   objval =   3.497400000e+04   infeas =   0.000000000e+00 (0)
> *    15:   objval =   3.497400000e+04   infeas =   0.000000000e+00 (0)
> *    29:   objval =   2.652371454e+02   infeas =   0.000000000e+00 (0)
> OPTIMAL SOLUTION FOUND
>      29:   objval =   0.000000000e+00   infeas =   1.000000000e+00 (0)
>      44:   objval =   3.497400000e+04   infeas =   0.000000000e+00 (0)
> *    44:   objval =   3.497400000e+04   infeas =   0.000000000e+00 (0)
> *    58:   objval =   6.586143114e+01   infeas =   0.000000000e+00 (0)
> OPTIMAL SOLUTION FOUND

> At the second simplex, GLPK starts with an objval=0 instead of the
> precedent solution.

> Why?

If you add columns to the problem object, they become non-basic,
so the optimal basis must remain at least primal feasible, however,
it is primal infeasible (as seen from the solver output). Which
changes in the problem object did you make before re-optimiziation?
Looks like you changed bounds of some basic rows/columns.

> did I use the routine lpx_warm_up?

It allows computing basic solution components for the current basis.
If you call lpx_simplex, there is no need to use lpx_warm_up.


Andrew Makhorin