Re: [Help-glpk] Infinite cycles

[Andrew Makhorin]

Hi Heinrich,

> The dual simplex method, techniques for a fast and stable implementation
> / Achim Koberstein
> Paderborn, Univ., Diss., 2005
> Dissertation (PhD)
> 
> describes on page 89 the possibility of numerical cycling due to the
> Harris test:
> 
> http://digital.ub.uni-paderborn.de/hs/content/pageview/1508326
> 
> Wouldn't it be safer to change the default ratio test to --norelax.
> 
> See
> https://pdfs.semanticscholar.org/2472/518346d657b8126d8ebe8657e2626dff5ef9.pdf
> for ideas how to avoid cycling.
> 

In glpk 4.61 the dual simplex solver was significantly improved by using
bound perturbation technique. In most cases this guarantees that no
basic variables violate their bound during the search (that usually
caused the message "numeric instability" to appear). Unfortunately, the
primal simplex solver is still does not use this technique.


Andrew Makhorin

PS: [off-topic] What about replacing the error message in DllMain?