[CPMD-list] highest levels of theory

[Axel Kohlmeyer]

>>> "RW" == Robert Williams <bob at bob.usuhs.mil> writes:

RW> Dear Axel, Juerg, cpmd-list,

RW> I am looking at table 1 in
RW> X. Xu and W.A. Goddard III, PNAS 101, 2673-2677 (2004)
RW> showing how various methods and functionals perform
RW> compared to experiment, particularly wrt dispersion
RW> forces.   This paper shows that reasonable results for
RW> dispersion interactions can be obtained even
RW> without HF exchange, in particular for the GGA PBEPBE
RW> (M. Ernzerhof & G.E. Scuseria, J. Chem. Phys,
RW> 110, 5029-5036 (1999), and - more so - for the GGA XLYP:

RW> 1.0 Ex(Slater) + 0.722\Delta Ex(Bech88) + 0.347\Delta Ex(PW91) + 1.0 Ec(LYP)

RW> where \Delta E for He2 and Ne2 are very close indeed to experiment.

RW> So, perhaps my question should be:

RW> How might I implement one of these, PBEPBE or XLYP,
RW> in CPMD?

dear bob,
please have a look at the file dftin.F, where you can see
how cpmd reads the &DFT section and then follow the flow
of the defined variables through the files, e.g. functionals.F,
gcener.F, lsd_func.F, and so on.

RW> It is not at all clear to me how I should relate
RW> various values of the CUTOFF in cpmd to the levels of theory
RW> I am more accustomed to thinking about.  The

the cutoff required is mainly determined by the type
of pseudopotential. since plane waves are 'delocalized'
in space, you can increase the basis set almost linearly
by increasing the number of plane waves through the 
energy cutoff. please check out, e.g. page 39-41 of

Ab initio molecular dynamics: Theory and Implementation by 
Dominik Marx and J"urg Hutter.
http://www.theochem.ruhr-uni-bochum.de/research/marx/cprev.en.html

there is also a lot of helpful material on juerg hutters
homepage. check out:
http://www.pci.unizh.ch/gruppe.hutter/e/information.html

RW> advice given in the list archive is to perform
RW> a series of calculations to determine a cutoff that
RW> gives acceptable results.  In writing a grant application,
RW> proposing to perform a series of calculations,
RW> it is useful to be able to say with some confidence
RW> that the work will be performed at a level
RW> equivalent to or better than ... double numeric
RW> with polarization - for example.

there are quite a large number of CPMD calculations 
published to whom you can compare your intended calculations
to. check out: http://www.cpmd.org/cpmd_publications.html

RW> My calculations on N-methylacetamide
RW> in one unit cell were initially performed with a CUTOFF
RW> of 120 Ry, and I've move to 150 Ry,
RW> but it seems to me that I should be
RW> able to report the results in a context that
RW> doesn't depend on how much personal experience
RW> the reader has with CPMD.

having next to no cutoff recommendations for many of the 
available pseudopotentials and little or no documentation 
of the performed test calculations is a big problem for 
beginners in CPMD, or plane wave pseudopotential based 
electron structure calculations in general. the best way
to deal with it, would be to document pseudopotentials
as they have been created and used and collect them in
a central place, e.g. on cpmd.org. but this needs somebody
to step up and actually do this. i would be willing to
collect the pseudopotentials, the test results or references
to publications, where these pseudopotentials were used,
and organize them, e.g. in the form of a webpage, but i 
currently don't have the time to do anything beyond that.

RW> Is the appropriate CUTOFF related to the
RW> size of the molecule or cell?

neither. the appropriate cutoff is determined by the 'hardest' 
pseudopotential and the property you want to calculate,
e.g. you need a much higher cutoff to get a reasonably 
converged stress tensor than to get converged forces.

RW> Has anyone worked out a general relationship between
RW> the CUTOFF and the approximate accuracy of the results
RW> in terms that might be related to previous ab initio codes?

i don't think you can draw direct correlation between those
two approaches. similar to the kind of quantum chemical codes,
that you are referring to, you need to get some experience.
imagine you have never used a code with a gaussian basis set
before. how should you know whether Aug-cc-pVTZ provides a
'better' result than 6-31G or which one of them is adequate 
for a hartree-fock or a coupled-cluster calculation?

best,
        axel.

RW> Best wishes,
RW> Bob

[...]

--

=======================================================================
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.