Re: [Help-glpk] Speeding up GLPK

[Andrew Makhorin]

>In my main problem I need to solve thousands Linear Programming (LP)
> problems. Lots of these LP problems differ from one to another by the
> Right Hand Side (RHS) vector. Besides, these LP problems have about
> a hundred primal variables and a thousand constraints. Throughout
> the main iterative process, the number of constraints increases. 
>
>What can I do to speed up the main problem execution? Which parameters
> are recommended to try on? How to ensure that the GLPK starts the
> optimization at the last solution found?

On return lpx_simplex saves the basis information in LPX object, so
when it is called once again, it starts from the current basis; thus to
speed up solving you just need to do nothing (however, if you expect the
current basic solution to be dual feasible, you can enable the dual
simplex; see the control parameter LPX_K_DUAL). In initial basis all
auxiliary variables are basic and all structural ones are non-basic (so
the basis matrix is unity). The current basis can be changed only by the
following api routines: lpx_std_basis, lpx_adv_basis, lpx_set_row_stat,
lpx_set_col_stat, lpx_simplex.