[CPMD-list] Gold pseudopotentials and cell parameters

[Axel Kohlmeyer]

On Fri, 10 Dec 2004, Stefano Piana wrote:

dear stefano,

what about the goedecker pbe pseudopotential i sent you? 
does it not work for you? is it much worse or only slightly?

it has been used successfully with a cutoff as little as 45ryd.
as a benefit, you don't have to worry about the treatment of
the non-local part, as it is already included in the pseudopotential.

SP> Somebody suggested me that I should use a _LARGE_ number of K-Points for the 
SP> bulk gold. Although I tried to run with up to 40 K-Points, I still have 
SP> convergence problems and moreover it is not possible to optimize the cell with 
SP> K-POints, therefore I cannot calculate the bulk properties of gold with my 

to be frank, the k-point support in CPMD is not as good as the support
for gamma-point calculations and the direct minimizers don't work as
well with k-points. convergence is (sometimes) better when using 
lanczos, but then again, lanczos is much slower.

SP> pseudo. An alternative is to use a bigger cell. I tried to run with 2x2x2 
SP> supercell containing 32 gold atoms, but still convergence problems arise (i.e. 
SP> the wavefunction never converges). I tried also INITIALIZE WAVEFUNCTION RANDOM 
SP> and to run a few steps of PCG MINIMIZE before switching to ODIIS. The second 
SP> method sometimes works, but convergence seems to be a bit erratic.

you can also switch to lanczos for a few rounds and then go back to
odiis. for the gamma point case, this can be automated with the following
input:

ODIIS NO_RESET=50
 5
LANCZOS DIAGONALIZATION RESET=2

which will run a round of lanczos after you had two successive
diis resets (which are set to happen not earlier than 50 steps apart).

what also happens occasionally, is that you converge to the
wrong state. one way to get out of this, is to do a car-parrinello run 
with a high electron mass, ANNEALING ELECTRONS and _all_ atom positions 
fixed to get the electron structure optimized. i've tried that recently
quite successfully on some otherwise pathological cases.

SP> At the moment I am including only the 6S and 5D in the valence. IF I include 
SP> the 5P as well I get very good results for the gold dimer (as expected from 
SP> the literature), but the calculation is more expensive and it looks like the 
SP> convergence problems for the solid state worsen.

yep, that is to be expected. as ari already pointed out: gold is a
metal, and using the free energy functional is a method implemented
in cpmd to deal with that. however, there have been meaningful 
simulations of metals (e.g. gold, check out the mailing list archives),
but you need to carefully tune your thermostats and watch your simulation
like a raw egg.

as for optimizing the cell: generally the best way of calculating
the equilibrium lattice constant and the bulk modulus,
is to perform a series of single point calculations with varying
lattice constant and then do a Murnaghan fit.

with a variable cell calculation in CPMD you don't have (in general)
a constant cutoff, but constant number of plane waves, so when the
cell changes, your 'effective' cutoff changes. which in turn requires
that the stress tensor is not only reasonably converged (which already 
requires a much higher cutoff than for converged forces) but also
_very_ well converged, so that the change in the effective cutoff has
no side effect. in short: don't do it unless you really, really need it.

there is a CONSTANT CUTOFF keyword, that tries to avoid that 
constant plane wave problem, but i don't know where and how well this 
works. you may want to check out the referred literature. but the 
requirement of having a converged stress tensor still holds.
try it out for a simple case (e.g. bulk silicon), you will be 
surprised (i was).

best regards,
	axel.

SP> I would like to have some suggestions. If anybody has already generated a MT 
SP> pseudopotential for gold could he send it to me so that I can compare with my 
SP> pseudos?
SP> 
SP> Stefano
SP> 

-- 

=======================================================================
Dr. Axel Kohlmeyer   e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.