Hi Matthew,
You are absolutely correct in that the old adage still applies,
garbage in equals garbage out. There is no substitute for knowing what you
are doing. On the other hand, the great increases in computer power have made
what was a nearly intractable problem much more manageable and it is a distinct
advantage to be able to use ab-initio software to get insight into the
chemistry. Of course, the scaling problem with plane-wave DFT is still there
(memory = cube of the number of atoms), but even on a Mac Pro with 32 GB of
memory you can simulate the ground state of hundreds of atoms in reasonable
times whereas ten years ago, you would have had to use a supercomputer for the
same problem. The codes themselves are much easier to use and their
associated mailing lists make it a great way to learn the details of a given
calculation. It certainly beats writing your own code from scratch (which I
did back in 1990 when I was at the Univ. of Illinois on the NCSA machines)!
On the other hand, I realize that I should have recommended a couple of books
for background reading.
A relatively easy read on solid state physics including some details of DFT
Efthimios Kaxiras: Atomic and Electronic Structure of Solids
My favourite book is a more comprehensive work.
Richard M. Martin: Electronic Structure: Basic Theory and Practical Methods
I should also add that abinit (http://www.abinit.org) is a free (as in gnu
copyleft) code that has a set of very nice tutorials online starting with
bonding two hydrogen atoms with references to appropriate papers to read for
deeper information.
To summarize, my point is that the hardware/software barriers to using DFT
codes have gotten much smaller over the years and then investing the time in
learning them can potentially offer greater insight into hard to understand
problems. It is certainly useful for EXAFS in that one can obtain the relaxed
structures in a 0 K calculation to compare for instance different possible
XANES structures while at the same time potentially offering calculated
material properties (optical reflectivity, Raman, etc.) that can offer
additional results that can be compared to experiment.
Certainly the ASE environment allows one to program a series of calculations
using python and helps shorten the time necessary for writing input files and
keeping track of the voluminous output. I give all of my postdocs extensive
training in using DFT to get insight into material science problems. While it
is not always useful, it can provide useful information in many cases.
Paul
On Aug 17, 2013, at 1:45 AM, Matthew Marcus mamar...@lbl.gov wrote:
Please correct me if I'm wrong, but I get the impression that if you don't
know exactly what you're doing, these programs will cheerfully
return wrong answers. There seem to be many parameters and choices to be
made. I once looked at Quantum Espresso but gave up when I saw that
the script for doing MD on a single water molecule ran for over a page of
incomprehensible code. I didn't see anything that looked like a
step-by-step tutorial or manual. Gaussian with the Gaussview UI is simple
enough for an experimentalist like me to use; are there better
packages out there which are as well?
I also get the impression that you need some pretty hefty compute power. A
Linux system is probably to be preferred over Windows.
One of these days I should learn Python, not just for this stuff.
mam
On 8/16/2013 7:27 AM, Paul Fons wrote:
Hi Scott,
I have been dabbling in DFT for a while. There are many free packages
around, but if you would like to model XAFS as well, I would suggest an all
electron code for accuracy (such as Wien2K). For general purposes, I am
also using VASP and CASTEP. The former uses projector augmented waves and
the other ultrasoft pseudopotentials. VASP is fast and scalable to the
largest machines and is designed from the ground up for quantum molecular
dynamics. Both VASP and CASTEP use pseudopotentials whereas Wien2K uses a
linearized augmented plane wave basis (read as radial wavefunctions and Ylms
in a sphere about each atom and plane waves between spheres with boundary
condition matching at the surface of the spheres. This way it is possible
to model even the 1s electrons for heavier atoms and yes it does affect
valence electron wavefunctions via orthogonality. All of these approaches
are single particle DFT approaches, but it should be good enough for a
start, if you want to go
beyond these, with solutions of the Bethe-Saltpeter equation (electron and a
hole) things get complicated and expensive (in terms of computer time) very
quickly.
If you are interested in free software I would suggest gpaw, quantum
espresso, and abinit. I would also suggest learning the atomic simulation
environment in which you can program multiple codes in python (and even
solve for maximally