Re: [Help-glpk] glpk 4.49 release information

[Andrew Makhorin]

> The University of Pisa Computer Science Department
> have also reimplemented the original RELAX-IV
> Fortran code, but into C++ instead.  They offer
> their version with warning that the algorithm may
> fail to converge.  The full text is:
> 
>  "RelaxIV is our C++ re-implementation of the
>   Relaxation algorithm by D. Bertsekas, based on
>   his original Fortran code. The code seems to
>   work fairly well as a general-purpose MCF
>   solver, and it is apparently capable of dealing
>   with problems with nonintegral costs and/or
>   capacities, although you have to be warned that
>   Relaxation algorithms may in theory fail to
>   converge with nonintegral data."
> 
>   http://www.di.unipi.it/di/groups/optimize/Software/MCF.html#RelaxIV
> 
> This information may be of interest to those using
> the new 'glp_mincost_relax4' call.
> 

Just like to note that glp_mincost_relax4 (as well as glp_mincost_okalg)
does not allow non-integer data--all flow lower bounds, arc capacities,
and per-unit costs should be integer though are passed as doubles.