[Top][All Lists]

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

Re: [Bug-glpk] bugs

From: Andrew Makhorin
Subject: Re: [Bug-glpk] bugs
Date: Mon, 09 Dec 2002 20:11:18 +0300

Thank you for your bug report!

>- in glplpt.c change of maximum lenght for a line form 155 to 1155 (just
>a big value because i deal with very long constraints)

A description of cplex format I used says that any line can't be longer
than 255 chars. I think this is not very restrictive -- you can split a
constraint into several lines (for example, writing one term per one

>- in glplpt.c change of maximum lenght for a token image to catch
>variables with long name (from 16 to 160)

glpk itself allows identifers be up to 255 chars long. Do you know this
limit for latest version of cplex? (I suspect the description I used is
a bit obsolete.)

>- i've noticed some wrong behaviours with integer section parsing. i've
>changed the T_BINARY  define setting to 66 (maybe other likely defines
>can be wrong but i didn't test them yet)

Could you please send me a file that causes the bug or explain what
happens in more details?

>- the parsing of the model is very sensitive to the blank spaces or line
>feeds or some data card missing for mps files (Cplex read them all with
>no problems), too restrictive because it doesn't support definitions
>like lb<= constraint<= ub .

In principle, it is not very hard to rewrite the parser. The problem is
I havn't got a newer description of cplex format.

>- many times the solver close with messages like
>O : lpx_simplex: numerical instability
>Assertion failed: spx_check_bbar(lp, lp->tol_bnd) >= 0.0; file ...

This error usually occurs when the problem is extremely badly scaled.
If you are using api routines, try to scale the problem by means of
lpx_scale_prob before solving it for the first time.

>spx_invert: the basis matrix is singular
>lpx_simplex: numerical problems with basis matrix
>lpx_simplex: sorry, basis recovery procedure not implemented yet

This can happen due to several reasons. For example, if the problem
being solved for the first time is badly scaled, or if some constraint
coefficints were changed before re-optimization that led to singularity
of the basis matrix. I need more information to give more detailed

>don't you think that could be more helpfull to set an infinite value for
>no upper or lower bound than setting them to 0 when they are not

Once you've obtained the type and bounds of a variable from the problem
object, you can adjust them as you need.

reply via email to

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