[Pw_forum] Possible reason for "ortho went bananas"

Nicola Marzari marzari at MIT.EDU
Fri Apr 3 08:31:11 CEST 2009

Zhiping Xu wrote:
> Dear all:
>     I performed a calculation using cp.x. However I always get "ortho 
> went bananas" errors either using calculation = 'scf' or 'cp'. Previous 
> post on this issue says it's convergence problem. My system converges 
> well using pw.x. I totally have no idea about this bananas error, can 
> someone help me on this? Thanks.
> Best,
> ZP

At the very beginning, dt needs to be small - a few steps with dt=5
or so would be good - this is because the initial random
wavefunctions change a lot in the first steps. Also, your script
has bot ionic dynamics and damped electronic dyanmics; this is not
a car-parrinello procedure.

Suppose you want to do a real cp run, this is what you need to do:

a few electronic steps with ions fixed, small dt (1 to 5) to let
the random wavefunctions get reasonable in shape.

~100 elecronic steps with ions fixed, and either damped electronic
dynamics or conjugate gradients minimization, to let the electrons
converge very weel to their ground state.

start the cp runs, with verlet dynamics on the ions and on the
electrons. To become expert with the stuff, do not use any thermostat
at first - thus you can see how your electronic and ionic energy go.
dt at most 10, probably small.

There are few videos on the web (linked from the page) with lectures on
ab-initio md (mostly by me, not that I am the most qualified person on
the topic).

What is your system, also ? A carbon dimer ? A carobn chain ? why the
other pseudos ???

Note tha tthe carbon dimer is a "metal", since as it vibrates it has
level crossings between homo and lumo - the only way to deal with it is
with a direct minimization (in cp there is something called
ensemble-dft, not too tested, thought).

Please read more on the cp method (e.g. the online paper by Hutter and
Marx) before continuing.

Once used properly, it does work extremely well.


Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu

More information about the users mailing list