bug-glpk
[Top][All Lists]
Advanced

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

[Bug-glpk] numerical instability - infinite loop


From: Ali Baharev
Subject: [Bug-glpk] numerical instability - infinite loop
Date: Tue, 27 Oct 2009 20:28:53 +0100

Dear Andrew,

This seems to me the same issue we discussed:

http://www.mail-archive.com/address@hidden/msg00374.html

Although i have removed the tiny coefficients as you proposed and i
have also switched to equilibration scaling, i still have a bunch of
problems where this issue remains and i get into an infinite loop.

For example the problem here:

http://reliablecomputing.eu/glpk/

was originally obtained on a Linux machine (but i am unable to
reproduce it in the same environment), however i can reproduce the
infinite loop in the Windows environment...

Even if you cannot reproduce the problem, please have a look at the
the problem instance given at the above link.

Please help me:

What else should i remove / fix to zero / do whatever is needed to
either resolve this issue or realize that i will get into an infinite
loop and abandon the procedure?

For example i see int the dump.txt file that there is a row bound
2.01e-17 and an other row with the bound 1.75e+03. Can it cause this
strange behaviour?

Is it possible to set a predefined limit on the number of switches
back to the primal simplex method (at least i would not get into an
infinite loop)? I think i can implement this but it is a rather ugly
"fix"...

A feature request: could you please implement a function to dump the
lp object in hexadecimal format?

It is supported by the ISO/IEC 9899:1999 standard (7.19.6.1 The
fprintf function and 7.19.6.2 The fscanf function; the a conversion
specifier), or

http://www.dinkumware.com/manuals/?manual=compleat&page=lib_prin.html

Now, it is totally random if i can reproduce the issue using the
decimal dump even in the same software and hardware environment. The
chances would be higher with a hexadecimal dump.

Many thanks in advance!

Ali




reply via email to

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