[Q-e-developers] Potential Software Development

[Kevin Schmidt]

Hello.

I am a graduate researcher at the University of Nevada, Reno, working on a
Master's Thesis in Chemical Engineering.  The focus of my project is
developing pair-potentials from DFT calculations for use in Molecular
Dynamics simulations.  We are currently working on the Lanthanum Hexaboride
system, simulating various geometries to extract relevant energetic data
related to positional variances of the respective atoms within an infinite
lattice.  I am writing to you because I have developed a significant amount
of software for this purpose, automating all aspects (aside from lattice
geometry input into QE) with bash and python scripts and am currently
placing these into a software package in a similar respect to how QE has
been developed (specific sub-programs called to produce input data to other
programs).  I was curious whether this may be of use to your package or
possibly as an added feature.

To give some more information, the pair-potential development takes a few
steps to extract out the full set of homo-atomic and hetero-atomic
interactions required to describe a system.  As a first step, we calculate
DFT energy of mono-atomic systems for "simple" lattices (BCC, FCC, DHCP,
HCP) and invert these using the Chen-Mobius lattice inversion method to
produce pair-potentials for these (assumed) electrically neutral systems
(the generation of all required input files aside from QE data is
automated).  A least-squares method is then utilized to parameterize the
inverted data to one of the non-bonded potentials used in the MD software.
The parameters for this non-bonded potential are then optimized using a
Monte-Carlo/Simulated Annealing process by reducing the noise between the
full lattice energy (energy vs lattice constant) and that reproduced by the
pair-potential (summing over all interactions for the lattice up until a
specified radial cutoff value).

When the mono-atomic potentials have been relaxed and are able to
sufficiently recreate the full QE energy for their respective systems, the
hetero-atomic potentials are then focused on.  Homo-atomic interactions are
subtracted from the total lattice energy of the hetero-atomic system, with
the assumption that

E_(total) = E_(homo) + E_(hetero) + E_(electrostatics)

giving us E_(hetero) + E_(electrostatics) after subtraction.  The
electrostatics are currently dealt with by subtracting out the
electrostatic energy which is given by the MD software (currently, we are
using DL_POLY, so this would be the SPME method).  This assumption was made
due to the implementation of point charges in most MD packages.  Since it
is our goal to reproduce the QE energy of the system throughout temporal
and spatial variances, this somewhat forces the electrostatics to give the
"correct" DFT energy of the system.  It would not be a difficult task to
implement some other charge-carrying specification, such as the spring
methods or other derived charges (Bader,AIM,etc.).

The hetero-atomic interaction energy is then given as a function of the
lattice constant, and this can be inverted and subsequently relaxed to give
parameter values which reproduce DFT generated data.  A further annealing
process is then performed on all parameters to relax the values of these
parameters (typically only 9 are required for a system with 2 atom types)
so that the lattice energy can be best reproduced to within a certain
threshold.

I understand that there are some significant approximations made, but owing
to the fact that many systems do not have well-suited pair-potentials, it
could be a first-step toward better implementations of this method.  Let me
know if you are interested or have any suggestions for improvements which
could be made to create a software package which would be of use to the QE
community.  I owe a great deal of gratitude for the open software which the
QE community has provided and would like to contribute any way I can.
Thanks, and I look forward to hearing from you.

-- 
Kevin Schmidt
*Chemical Engineering Department*
*University of Nevada, Reno*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20140922/35811853/attachment.html>