[Wien] A feature of the new mixer

[Stefaan Cottenier]

This is not really different from the 'golden rule' that applied in the 
past (perhaps not any more in the latest versions, but I kept doing 
this): "whenever you change some technical parameter (Rkmax, k-mesh, 
...), do 'rm *.broy*' as well".

General feedback: this new mixer is great, it really speeds up 
convergence a lot. We had one case where it failed (and where also the 
old mixer failed), but it was a rather uncommon one: the starting values 
for the internal coordinates where close to a local *maximum* of the 
Born-Oppenheimer surface, such that the forces became smaller while the 
energy was increasing -- finally leading to a 'inconsistent forces and 
energy' error.

Stefaan


Laurence Marks wrote:

>The new mixer has a "feature" which makes it behave relatively poorly
>if you change parameters during a run (e.g. GMAX and probably RKMAX)
>without deleting the previous history (case.broydX) files. Very
>concisely, the old method would overwrite prior information with the
>most recent, while the new one uses all the information in the
>previous steps (that you specify in case.inm). If you change
>parameters :DIS will probably get larger and the algorithm believes
>that it has taken some bad steps so will slow down. Eventually the
>steps with the pervious parameters will age out and the algorithm will
>start behaving properly. There is no automatic cure for this (the
>algorithm cannot read your mind, alas), so please delete the history
>files if you change parameters.
>
>N.B., I would be interested in feedback about how the algorithm is
>behaving -- better hopefully.
>
>  
>


Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm