mirror of https://gitlab.com/QEF/q-e.git
123 lines
5.5 KiB
Plaintext
123 lines
5.5 KiB
Plaintext
|
|
*** How to set up the parameters for minimization/dynamics ***
|
|
|
|
Variable Meaning Typical Range
|
|
|
|
DELTAT Time Step (au) 1 - 20
|
|
EMASS Electron Mass (au) 100 - 1000
|
|
EMAEC Preconditioning Mass 2 - 3
|
|
FRICE Electronic Friction 0.2- 0.02 (minim. only)
|
|
FRICP Ionic Friction 1/10 of FRICE (minim. only)
|
|
QNP Ionic thermostat mass
|
|
TEMPW Desired Ionic temperature (you set it)
|
|
QNE Electronic thermostat mass
|
|
EKINCW Desired Electronic kin.energy <0.001/atom ?
|
|
|
|
*** Electronic minimization at fixed ions
|
|
|
|
Performed if TFOR=.false. TNOSEP=.false. TNOSEE=.false. FRICE>0
|
|
Requires: DELTAT EMASS EMAEC FRICE
|
|
|
|
The parameter that determines the evolution of the electronic wave
|
|
functions is Lambda=(DELTAT**2/EMASS). Its maximum value is fixed
|
|
by the requirements that
|
|
- iterative orthonormalization works (it will not if Lambda is too big)
|
|
- the minimization works (as above).
|
|
If this does not happen, DELTAT must be reduced, or EMASS increased.
|
|
|
|
For the very first (10-20) steps, Lambda should be small, EMAEC should
|
|
be set to a large value (large wrt ECUTW) so that preconditioning is
|
|
effectively turned off; FRICE should be set to a large value (up to
|
|
FRICE=1, corresponding to steepest descent).
|
|
EMASS should be assigned a value that is good for subsequent ionic
|
|
minimization/dynamics (see below; typical values 300-700).
|
|
DELTAT should be given a value that produces a working minimization
|
|
(typical values 5 or less).
|
|
|
|
For subsequent steps, EMAEC should be set to 2 or 3 (an optimal value
|
|
can be determined by trial and error: the best value is the one that
|
|
allows the larger time step).
|
|
FRICE should be set to 0.2-0.1, and reduced if the minimization
|
|
becomes sluggish.
|
|
DELTAT should be given the largest value that produces a working
|
|
minimization (typical values 10-15, max. up to 20).
|
|
|
|
*** Ionic minimization
|
|
|
|
Performed if TFOR=.true. TNOSEP=.false. TNOSEE=.false. FRICE or FRICP>0
|
|
Requires: DELTAT EMASS EMAEC FRICE FRICP
|
|
|
|
As above, with FRICP about 1/10 of FRICE. If the minimization becomes
|
|
sluggish, reducing FRICP to < 0.01 (in some cases even setting it to
|
|
zero) usually help. Also, setting ionic masses to the same value
|
|
usually helps. See below for bounds on DELTAT and EMASS.
|
|
|
|
*** Dynamics with electrons and ions
|
|
|
|
Performed if TFOR=.true. TNOSEP=.false. TNOSEE=.false. FRICE=0 FRICP=0
|
|
Requires: DELTAT EMASS EMAEC
|
|
|
|
In the combined electronic/ionic dynamics, one has to chose DELTAT
|
|
as big as possible, but ensuring that
|
|
- energy is conserved during the dynamics
|
|
- the dynamics is adiabatic: there is no transfer of energy between
|
|
electronic and ionic degrees of freedom. The adiabaticity is better
|
|
conserved in systems with a large electronic energy gap.
|
|
|
|
A typical time step for classical MD is given by 1/50 to 1/100 of the
|
|
inverse of the maximum ionic frequency (1/10 is an upper limit). As
|
|
an example: Si, omega_0=15 THz, 1/omega_0=60 ps, time step=1 fs or so.
|
|
We assume here that the ionic masses are set to their physical values.
|
|
The typical DELTAT for Car-Parrinello MD is a few times smaller than
|
|
the time step for a classical MD (in the example above, 0.2-0.3 fs,
|
|
or DELTAT=~10).
|
|
Once DELTAT is fixed, a suitable value for EMASS is determined by
|
|
the above conditions. If iterative orthonormalization gives problem
|
|
or the energy is not conserved, reduce DELTAT or try a different value
|
|
of EMAEC for optimal preconditioning. If there is energy transfer,
|
|
reducing EMASS will give a more adiabatic behavior.
|
|
|
|
*** Nose' thermostats
|
|
|
|
If a simulation on the canonical ensemble - system in contact with
|
|
a thermal bath at temperature T - is desired, one has to use a Nose'
|
|
thermostat on ions, setting TNOSEP=.true., TEMPW=T desired (Kelvin).
|
|
The kinetic energy will make wide oscillations [***how wide???]
|
|
around the mean value gKbT/2, where g=number of ionic degrees of
|
|
freedom, Kb=Boltzmann constant.
|
|
The value of QNP must be set so that the frequency of such oscillations
|
|
is of the same order of magnitude of the typical ionic frequencies of
|
|
the systems, so as to maximize the interaction of the thermostat with
|
|
the ionic degrees of freedom. This can be estimated by inspection or
|
|
from the approximate formula: QNP=2*g*Kb/Omega^2, where Omega is the
|
|
typical ionic frequency (for instance, the Debye frequency, or half
|
|
the maximum phonon frequency). The value of QNP can also be estimated
|
|
on a small system and extrapolated to a larger cell.
|
|
|
|
In low-gap or zero-gap systems, it may be needed to have a Nose'
|
|
thermostat on both ions and electrons (TNOSEE=.true.) in order
|
|
to obtain an adiabatic dynamics.
|
|
EKINCW must be set to approx. twice the value of the adiabatic
|
|
contribution to the electronic kinetic energy Ekad (that is:
|
|
the part of the electronic kinetic energy that comes from the
|
|
adiabatic motion of electrons following the atomic motion).
|
|
Assuming that all atoms have mass M, the following estimate hold:
|
|
Ekad=2KbT*EMASS/M*Ekin , where Ekin=electronic kinetic energy.
|
|
Ekad can also be estimated from the energy gained by minimizing wrt
|
|
electronic degrees of freedom, starting from a given ionic and
|
|
electronic configuration found during the dynamics.
|
|
QNE must be set to a value such that the associate frequency
|
|
Omega=sqrt(4Ekin/QNE) is larger than the highest phonon.
|
|
|
|
*** Relevant Theory
|
|
|
|
Adiabaticity in CPMD:
|
|
G. Pastore, F. Buda, M. Smargiassi, PRA 44, 6334 (1991)
|
|
|
|
Nose' Thermostats:
|
|
P. E. Bloechl and M. Parrinello, PRB 45, 9413 (1992)
|
|
|
|
Second-order damped dynamics, preconditioning:
|
|
F. Tassone, F. Mauri, R. Car, PRB 50, 10561 (1994).
|
|
|