[Q-e-developers] minimizzazione inefficiente

Paolo Giannozzi giannozz at democritos.it
Wed May 11 20:46:50 CEST 2011


On May 9, 2011, at 16:33 , Nicola Manini wrote:

> At Paolo's request, I attach the input file of a typical example
> of problematic damp minimization

I don't see any problem here: the system is already minimized.
Really: forces are very small and energy is almost constant.

!    total energy              =    -138.30526161 Ry
!    total energy              =    -138.30526161 Ry
!    total energy              =    -138.30526161 Ry
!    total energy              =    -138.30526161 Ry
!    total energy              =    -138.30526161 Ry
!    total energy              =    -138.30526140 Ry
!    total energy              =    -138.30526161 Ry
!    total energy              =    -138.30526140 Ry
!    total energy              =    -138.30526161 Ry
!    total energy              =    -138.30526140 Ry

I think that trying to further minimize is a waste of time. In
principle minimization could go on forever, but I am afraid
that at these levels, the noise on the energy and on the forces
due to scf convergence, to the discreteness of the FFT grid
(making the system not translationally invariant!), to all kind
of hidden and forgotten approximations, will conspire to make
any minimization algorithm unusable. I don't know what the
origin of the above "oscillations" in the energy is: maybe the
discreteness of the FFT. In any event, I would say that there
is nothing wrong in the damped dynamics (or in bfgs) itself:
there is some numerical noise in forces and energy calculation,
somewhere, God knows where, and whether it can be removed.

By the way: some years ago the energy was printed with 6
decimals. Increasing the number of decimals to 8 was in
my opinion a very bad idea: it gives the illusion that those
last figures are "significant", but in fact they change with the
phase of the moon.

Paolo
---
Paolo Giannozzi, Dept of Chemistry&Physics&Environment,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222







More information about the developers mailing list