[Pw_forum] difference between molecular dynamics of PW and that of CPMD

[Axel Kohlmeyer]

On Thu, 4 Jan 2007, Lawrence Lee wrote:

LL> Dear sir / madam,

hi lawrence,

LL> I am a new user of pwscf. I know that pw can do molecular dynamics.
LL> What is the difference between that and CPMD?

please be a bit more precise. do you refer to the CPMD package 
or the CPMD method (as implemented in cp.x )?

in the first case you have two packages that implement first 
principles calculations using DFT and plane wave basis 
sets with pseudopotentials. due to differences in the interests
of the developers, those two packages (quantum espresso and CPMD)
have different strenghts and weaknesses. those can be
mostly deduced from looking at the descriptions of the
input file syntax (plus a few undocumented features, for
which you have to read the source code ;-) ).

in the second case you have to distinguish between the
born-oppenheimer MD in pw.x, where the ground state
wavefunction is fully computed in each MD step, and 
car-parrinello MD in cp.x where, starting from a ground
state wavefunction, the wavefunction for subsequent
steps is propagated along the trajectory of the atoms
by means of an extenden lagrangian scheme (i.e. it is
not exactly the ground state wavefunction, but 'oszillates
around it'). strictly speaking, CP-dynamics are also
using the born-oppenheimer approximation, but BO-MD vs
CP-MD is the established nomenclature.

historically, CP-MD used to be much more efficient than
BO-MD, even though you have to use an about 5 times shorter
time step, but more efficient diagonalization and extrapolation
and the availability of more cpu power make BO-MD competitive.
on top of that, CP-dynamics put a number of restrictions on
the system you want to investigate (it needs a band gap,
your atom dynamics is modified by the fictitious electron 
dynamics).


LL> Besides, do you know how to restart the run in pw.x? For example, if
LL> I have made a run of 400 steps with dt = 20 atomic time unit, but
LL> now I find it insufficient. I need the run to have 2000 steps, say.
LL> Do I have any method to proceed the run without doing it from the
LL> first step again?

have you tried using restart_mode='restart' instead of 'from_scratch' ??

axel.

LL> 
LL> Lawrence.
LL> 
LL>  _______________________________________
LL>  YM - Â÷½u°T®§
LL>  ´Nºâ§A¨S¦³¤Wºô¡A§AªºªB¤Í¤´¥i¥H¯d¤U°T®§µ¹§A¡A·í§A¤Wºô®É´N¯à¥ß§Y¬Ý¨ì¡A¥ô¦ó»¡¸Ü³£ÉN¨«¥¢¡C
LL>  http://messenger.yahoo.com.hk

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.