[CPMD-list] Energy components in PIMD

[Axel Kohlmeyer]

On Thu, 15 May 2008, ukasz Walewski wrote:

LW> Dnia 2008-05-12, pon o godzinie 10:10 -0400, Axel Kohlmeyer pisze:
LW> 
LW> Dear Axel,

dear lukasz,

[...]

[...]

LW> I think I tracked down the problem. The power spectra (that I finally
LW> managed to reproduce - thanks for the fourier.x code) overlap with the
LW> ionic vibrational frequencies! The highest vibrational mode is 3190 cm-1

you have to be careful here. of course the spectrum of EKINC
will show all atomic vibrations, since the electrons have to 
follow the atoms around and thus accelerate as well. if you
want to know the intrinsic frequencies of the electronic DOF,
you have to compute the power spectrum of the ionic kinetic
energy as well and compare the two.

LW> while the EKINC spectrum starts around 1000 cm-1 (see the first attached
LW> graph). I will try to reduce the hydrogen oscillations by deutering the
LW> outer protons with the hope that this will not perturb the inner proton
LW> transfer - will I have to correct for kinetic isotope effects in this
LW> case ? (No, I guess.)

well, if you run CP dynamics you _will_ already perturb the 
inner proton motion significantly (-> drag effect). 
for a simpler example have a look at:
http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/cpmd-tutor/part8.html#chap9_sect2

you can take the provided data files and see how much the fictitious
electron mass affects your dynamics by comparing the power spectra
of BO and CP dynamics with a regular vibrational specta run...
the results speak for themselves.

the problem with BO-dynamics is that usually you have to converge
the wavefunction very tightly to avoid a drift in the kinetic energy
of the atoms (CONVERGENCE ORBITALS set to 1.e-8 to 5.e-8). however, 
changing the initial guess from using the previous wavefunction to
an extrapolation (3rd or 4th order) over several older wavefunctions
removes the systematic loss of energy and thus allows a "sloppier"
(and thus much faster) convergence, on top of that you get a better
initial guess which will reduce the computational effort even more.
the input example i sent you was still rather conservative and takes
less than twice as long as a similar CP trajectory and has by 
construction no red shift due from "drag".

LW> However I have two more observations:
LW> 
LW> 1) Reducing the electron mass while keeping the ionic masses unchanged
LW> reduces the EKINC drift anyway - see the second graph. This must happen
LW> regardless of the apparent coupling btw. electronic and ionic DOF.
LW> Moreover switching from MT to VDBÿÿ pseudopotentials and reducing the
LW> r-cutoff from 70 to 35 Ry while keeping the other parameters fixed does
LW> actually improve the adiabaticity (cf. green and blue plot on the second
LW> attached graph).

reducing the cutoff should also reduce the "drag" (less wavefunction
coefficients coupled to the ionic DOFs). never tested that one though.
you could try to verify this rather quickly with the example from the
tutorial webpages.

LW> 2) The power spectra of electronic DOF do not change much upon change of
LW> the electron mass - on the first attached graph there are in fact two
LW> plots: MT, 70 Ry, mu=400 a.u., dt=3 a.u. (black) and VDB, 35 Ry, mu=200,
LW> dt=3 a.u. (red). The two spectra are almost identical.

you have to look at the high frequency components!
note that electronic thermostat is typically coupled 
to 10000cm-1 to 15000cm-1.

cheers,
   axel.

LW> Trying to join those two observations I conclude that there must be some
LW> mechanism, that prevents energy transfer from ionic subsystem to
LW> the ÿÿelectronic one, concurrent to the frequency coupling. What can it
LW> be ?
LW> 
LW> Thank you guys for the patience...
LW> 
LW> Lukasz
LW> 

-- 
=======================================================================
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.