help-glpk
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Help-glpk] Re: Arbitrary floating point precision in GLPK


From: Andrew Makhorin
Subject: [Help-glpk] Re: Arbitrary floating point precision in GLPK
Date: Thu, 20 Sep 2007 13:11:03 +0400

Hi Serge,

> A couple of years ago, I asked a question on the mailing list about the
> arbitrary precision (floating point and integer/rational) in GLPK (see
> http://osdir.com/ml/gnu.glpk/2005-10/msg00001.html).

> I received no answer (probably no one was interested )and I moved to
> other solutions for my problem (I switched to PIP and help to implement
> GMP types input and output).

> But I'm faced again with the problem of linear programming with high
> precision floating points numbers.

> My question now is not whether this will be implemented in GLPK or not,
> it turns out to the following one  should we (I work in a the ARENAIRE
> research team in Lyon) roll up our sleeves, is it reasonably possible to
> make an adaptation of GLPK to the MPFR floating point types ?

> MIP is another beast and would possibly take a much bigger effort 
> (though it may be of some use...).

> If we should try, could you give us some guidance on :
> - where to start,
> - on the shortest path to get something usefull working (like a basic
>    simplex).

Starting from glpk 4.13 it is possible to solve LP in exact (that is
rational) arithmetic based on GNU MP.

The following glpk components might be interesting to you (file names
correspond to the most recent version, which is 4.22):

glpgmp.c    glpk version of GNU MP routines
glplux.c    LU-factorization based on exact arithmetic
glpssx01.c, glpssx02.c
            primal simplex method based on exact arithmetic
glplpx10.c  driver to the exact simplex solver

The glpk exact lp solver is available on api level (lpx_exact) as well
as in the solver glpsol (options --exact, --xcheck). For more details
please see the reference manual included in the glpk distribution and
comments in the components' source code.


Andrew Makhorin





reply via email to

[Prev in Thread] Current Thread [Next in Thread]