From: Axel Kohlmeyer (
Date: Thu Sep 24 2009 - 21:49:01 CDT

On Thu, 2009-09-24 at 18:23 -0700, Goutham wrote:
> Can you tell me what is the physical basis for using dipole moment to
> calculate the spectral density? Can you give me any pointers to

please look up what how you (properly) compute IR spectra
from quantum mechanics. that will help you understand why
the dipole moment.

> resources that I can read to understand better? Also should I be using
> the position dcds or the velocity dcds?

if you use the dipole moment, you have to use the position dcd,
for velocity ACF based spectral densities, you need to use the
velocity dcd, of course, but then you don't use the charge information.

> One can calculate spectral densities from different quantities, (like
> dipole moment, or using velocity of each atom and then averaging the
> spectral density over the atoms).. I am trying to understand, how the
> spectral densities will change depending on the kind of quantity that
> we use... Should they all be similar? Any pointers will be very
> helpful..

the position based spectra make mosst sense in crystals, or else
you will see spurious peaks. velocity based spectra will be very
similar to dipole based spectra, i.e. the peak positions should be
identical, but the peak heights can be very different. as you don't
consider that motions that don't change the dipole moment, are not
IR active. with the dipole moment based spectrum, you get closer,
but you still don't consider the symmetry of the wavefunction, which
imposes additional rules about what transitions are allowed and
what not (you'd have to compute the born effective charges or
transition moments for that).

i don't recall exactly, but the two main references that have
helped me a lot when i was a graduate student are the book
"Theory of Simple Liquids" by Hansen & Macdonald, and the
review on dielectric properties from simulations by
Madden and Kivelson from the mid 80s. i don't have my phd thesis
at hand to look up the exact reference, but you should find
it through literature search engine easily.


> Thanks
> Goutham
> > Now if I wanted to calculate the spectrum as a Fourier
> Transform of
> > Velocity auto correlation, then should I use similar to
> example 2 in :
> >
> >
> > (i.e.) find the spectrum for each atom in the system, and
> then get an
> > average over all the atoms.
> right, you can take about any property, create a list, and
> then feed it
> to the code and it will produce the spectral densities, _not_
> the
> spectrum, as the code has no knowledge of the transition
> moments.
> please also note that the implemented algorithm computes the
> spectral
> densities directly in fourier space, i.e. the explicit
> auto-correlation
> will not be computed as an intermediate result. this bypasses
> a lot
> of the arbitrariness and undesirable scaling behavior of
> computing
> auto-correlation functions.
> cheers,
> axel.
> >
> >
> > Thanks
> > Goutham
> >
> >
> >
> >
> >
> --
> Dr. Axel Kohlmeyer
> Institute for Computational Molecular Science
> College of Science and Technology
> Temple University, Philadelphia PA, USA.

Dr. Axel Kohlmeyer 
Institute for Computational Molecular Science
College of Science and Technology
Temple University, Philadelphia PA, USA.