by Kevin F. Garrity,
Joseph W. Bennett, Karin M. Rabe, and David Vanderbilt
Updated August 26,
2019
Version 1.5 now available. See below, here(pdf) and here(pdf).
Welcome to the
GBRV pseudopotential site. This site hosts the GBRV pseudopotential
library, a highly accurate and computationally inexpensive
open-source pseudopotential library which has been designed and
optimized for use in high-throughput DFT calculations and
released under the gnu public license.
We provide potential files for direct use with the Quantum Espresso,
Abinit, and JDFTx plane-wave
pseudopotential codes, as well as input files for the Vanderbilt
Ultrasoft
pseudopotential generator. Please see our paper: K.F. Garrity, J.W. Bennett, K.M. Rabe and D. Vanderbilt,
Comput. Mater. Sci. 81, 446 (2014) (link),
for more information.
The GBRV pseudopotential library has been tested by comparing to
all-electron LAPW+LO calculations performed with the WIEN2k code in a variety of
chemical environments. The GBRV potential library has been
found to produce lattice constants, bulk moduli, and magnetic
moments which are of comparable or higher overall accuracy than other
comprehensive pseudopotential libraries across a wide variety of
bonding environments, while maintaining a low computational cost.
Please consult our paper
(local preprint) for full details on our
design criteria and testing procedure, and extra notes on the Abinit
potentials and testing data.
While these potentials have been designed for high-throughput
calculations, they should be of general use. Despite our
relatively thorough testing, we cannot guarantee that these
potentials will be appropriate for every application, but
we provide testing data as well as the input files for use with
the Vanderbilt Ultrasoft pseudopotential generator code, which
can be used to modify the potentials to suit your needs.
Please let us know if you improve on any of the potentials.
Kevin F.
Garrity
Research Associate
NIST
Formerly Postdoc at Rutgers University
kevin.garrity@nist.gov
Download Entire
Library:
Generation files for PBE QE USPP
Generation files for PBE
Abinit PAW
All PBE potentials in UPF
format for QE
All PBE potentials in
PAW format for Abinit
All LDA potentials in
UPF format for QE
All LDA potentials in
PAW format for Abinit
All PBESOL
potentials in UPF format for QE
All LDA potentials in
uspp format
All PBE potentials in
uspp format
All PBE potentials in
xml format
All LDA potentials in
xml format
Notes:
- September 2, 2019 J.
W. Bennett et. al. have done a study on the
determination of Hubbard U values using the GBRV
pseudopotential set. It is available at Comput. Mater. Sci. 170, 109137 (2019) (link)
- September 12, 2016 My current email is kevin.garrity@nist.gov
- September 23, 2015 Newly added xml paw files. They should be exactly the same as the Abinit paw files, but in a different format for compatibility with more codes. See here for formatting.
- September 23, 2015 The Na uspp was causing trouble in phonon calculations despite passing earlier tests, it has been updated to version 1.5. Also, several pbesol potentials updated in v1.4 were actually pbe potentials internally but told the code to run pbesol, due to an error in transferring potentials on my end. In practice this should result in only very very small quantitative differences, but it has been corrected. Also, the Cl title has been fixed on some psps.
- June 15, 2015 A slight change. Based on updated all-electron tests, P/Cr/Fe/Mn for QE and P for abinit have been updated to version 1.5. Please see these notes.
- April 15, 2015 Tests now include PSlibrary-low precision here.
- April 13, 2015 Version 1.4 of several USPP's
are now available. The changes were made in order to improve convergence of DFT perturbation theory phonon calculations versus plane-wave and charge density cutoffs. Now, DFT-PT and finite differences phonons are in excellent agreement even for low convergence parameters. The new potentials should not show a noticeable difference for most structural properties. Currently no abinit PAWs are affected. Please see full
discussion and new testing results here.
- April 13, 2015 Updated testing of Version 1.4 of the GBRV library, as well as new versions of the JTH and PSlibrary sets of PAWs. In summary, GBRV offers an excellent combination of accuracy and computational speed. The PSlibrary high precision potential set gives very similar results, but many elements require higher cutoffs (see discussion in their paper). PSlibrary-low gives slightly worse results for convergence parameters similar to GBRV. The JTH set performs similarly to GBRV on elemental tests, but worse on a few multi-atom compounds. Please see full
discussion and new testing results here.
- April 13, 2015 Thanks to I. Castelli and N. Marzari for discussions.
- April 13, 2015 Zincblende WIEN2k testing data for a slightly modified Delta test. Use with the same k-point sampling / non-spin-polarized setup as in our original paper.
Older notes:
- February 3, 2014 Version 1.2 of several potentials
are now available. The changes were made in order to improve convergence versus the
plane-wave and double-grid/charge-density cutoffs, but the new potentials should not
show a noticeable difference for most structural properties. Please see full
discussion and new testing results here.
- February 3, 2014 Updated testing data, including a new zinc blende
test set as well as testing results for the new JTH PAW set.
Results suggest the GBRV potentials provide better performance for robust high-throughput applications.
- February 3, 2014 Older versions of potentials are available under the generation directories.
- Please
see notes on Abinit versus Quantum Espresso versions
of potentials here.
-
See discussion of Hf in paper.
- September
25, 2013- Note: in order to run some
potentials in the uspp code, it is necessary to
increase idim4 to 6 in line 24 of
runatom.f. Otherwise some
input files will produce errors due to insufficient
space for the pseudized valence states.
- September
14, 2013- Including (binary)
.uspp versions of potentials (native Vanderbilt
format), for compatibility with JDFTx,
as well as some older PW-PSP codes.
These should produce essentially identical results
when compared to QE versions, they are simply in a
different format.
- September
13, 2013- Our paper has been published: K.F. Garrity et al.,
Comput. Mater. Sci. 81 446 (2014), (link)
- July
24, 2013- Added lda and pbesol (PRL 100
136406 (2008)) versions of the
potentials. These versions have not been
tested specifically!
They are simply the same input files with
different exc functionals. The pbesol potentials
were created by modifying the two parameters in pbe.f
as necessary to change from pbe to pbesol (um and bet,
see reference).
- July
23, 2013- Very minor correction of title
field in Mg and Hf(4+). This should not affect
any calculations (thanks G. Pizzi).
- June
7, 2013- Updated several 2p
elements B, C, N, and O, to version 1.01. They
were performing poorly in small molecules (e.g. O2
).
The solid state tests are very similar, but have been
updated in the paper preprint/notes. While the
potentials are improved, detailed
studies of small molecules require potentials designed
for somewhat higher plane-wave cutoffs.
- The
PAW potentials were updated on May 25, 2013
- See
supplementary materials for
all testing data (or spreadsheet
version)
- A
preprint of our paper is available at http://arxiv.org/abs/1305.5973, final version at (link)
Links to individual Potentials:
Thanks for discussions:
M. Torrent, I.E. Castelli, N. Marzari, G. Pizzi, N.A.W. Holzwarth, S. Cottenier, K. Lejaeghere, E. Bousquet, and others...
Figure 1: Lattice
constant error for GBRV potentials in the perovskite structure,
versus all-electron (AE) calculations. Blue squares are
USPP's tested with Quantum
Espresso, Red diamonds are PAW's tested with Abinit. The
green square and yellow diamond correspond to the Hf(4+)
alternate potential for oxides.
Note: LDA and PBESOL potentials have not been tested specifically. They are the same input files with different exc functionals.