Copyright (C) 2010-2013 ABINIT group (XG)
This file is distributed under the terms of the GNU General Public License, see
~abinit/COPYING or
http://www.gnu.org/copyleft/gpl.txt .
For the initials of contributors, see ~abinit/doc/developers/contributors.txt .
Many thanks to the contributors to the ABINIT project between
November 2012 and March 2013. These release notes
are relative to improvements of ABINITv7.2 with respect to v7.0.
The list of contributors includes :
F. Abreu, J. Alberdi, B. Amadon, G. Antonius, A. Berger, J.-M. Beuken, F. Bruneval, D. Caliste, F. Dahm,
M. Delaveau, T. Deutsch, C. Espejo, M. Giantomassi,
Y. Gillet, G. Huhs, F. Jollet, M. Oliveira, S. Ponce, Y. Pouillon, T. Rangel,
G.-M. Rignanese, D. Sangalli, M. Torrent, M. Verstraete, D. Waroquiers, B. Xu, J. Zwanziger
It is worth to read carefully
all the modifications that are mentioned in the present file,
and examine the links to help files or test cases ...
This might take some time ...
Xavier
__________________________________________________________________
Version 7.2, released on March 25, 2013.
Changes with respect to version 7.0 :
A. WARNINGS.
B. Most noticeable achievements (for users)
C. Changes in the package, for developers
D. Other changes (or on-going developments, not finalized)
__________________________________________________________________
A. WARNINGS AND IMPORTANT REMARKS
A.1 With the replacement of LibXC1.1 by LibXC2.0, numerical bugs have been eliminated (thus changing the references) :
- in the evaluation of the Langreth-Mehl
correlation energy (ixc -137 , the error was on the order of 0.1Ha in test libxc#04 and 0.5Ha in test libxc#05 ),
- in the evaluation of exchange in modified form of PW91 by Adamo and Barone (ixc -119, the error was on the order
of 6 mHa in test libxc#04, and 3 mHa in test libxc#05),
- in the Vosko-Wilk-Nussair functional, spin-polarized (ixc -007, the error was on the order of 2 mHa in test libxc#05).
- in the MPWLYP1W functional [EE Dahlke and DG Truhlar, J. Phys. Chem. B 109, 15677 (2005)] (ixc -174, 6 mHa in test libxc#06,
and 4 mHa in test libxc#07)
- in the OPTPBE_VDW, ixc -141 ,PBE reparametrization for vdW
[J Klimes, DR Bowler, and A Michaelides, J. Phys.: Condens. Matter 22, 022201 (2010)
(40 mHa in test libxc#17, 4 mHa in test libxc#18)
- in the XC_GGA_XC_KT2, ixc -146 ! Keal and Tozer version 2 [TW Keal and DJ Tozer, J. Chem. Phys. 119, 3015 (2003)
(7 mHa in test libxc#17, 100 mHa in test libxc#18))
Small numerical differences are to be expected when using some of the other functionals,
as changes were introduced in Libxc2.0 in the
evaluation of the functionals in the limit of vanishing densities (the new version should be more accurate).
__________________________________________________________________
B. Most noticeable achievements
B.1 The ability to read XML PAW data has been implemented and tested.
It allows ABINIT to access all the PAW atomic data prepared for the GPAW code, and to do cross-checkings.
In particular, for the PBE function, there has been an extensive comparison versus all-electron WIEN2K calculations,
whose result is reported in http://molmod.ugent.be/DeltaCodesDFT .
See automatic tests v7#60 to 63.
By the way, the Atompaw software is also able to produce such XML PAW data. In the present version
some restrictions exist. In the next version (3.1.x), XML PAW data produced by AtomPAW will be fully
compatible with ABINIT.
Work by F. Jollet and M. Torrent
B.2 Libxc 2.0.1 is now linked to ABINIT.
Many additional functionals are thus available (compared to v7.0, where LibXC1.1 only was available).
This includes the hybrid functionals HSE06, PBE0, and B3LYP (see the release notes of v7.0).
6 LDA and 39 GGAs have been added and are portable. More GGAs, mGGAs and hybrid should still
be either activated, or tested.
Work by M. Oliveira and Y. Pouillon
B.3 One can compute the STS signal for an STM tip atom placed at a given position,
according to the chemical nature of the tip (s- p- d- wave etc...),
thanks to the combination of new input variables natsph_extra, ratsph_extra, and xredsph_extra.
At the end of a normal GS run this produces the l-decomposed
partial DOS at the point xredsph_extra.
Previously, STM signal analysis was done for a fixed voltage by computing the total density
at the different points where the tip was located,
no partial decomposition was allowed and only limited sets of voltages could be used (no spectroscopic STS mode).
See new test v7#04.
Work by M. Verstraete
B.4 Automatic determination of the best processor repartition is now available for ground-state
and response function calculation:
- by means of a simple heuristics
- by means of a series of small benchmarks using the scheme applied for the LOBPCG algorithm.
The size of communicators for parallel matrix algebra routines can be optimized, as well
as the use (or not) of matrix algebra on GPU.
See the description in the input variable autoparal, values 1, 2 or 3..
Automatic tests should still be written.
Work by F. Dahm, M. Torrent and F. Jollet.
__________________________________________________________________
C. Changes for the developers
C.1 Pruned the build-system support for long-obsolete compilers and added GCC 4.7. Reviewed build-system makefile snippets
Work by Y. Pouillon
__________________________________________________________________
D. Other changes
(or on-going developments, not yet finalized).
D.1 PAW + Bigdft project (still in development)
PAW pseudos are now fitted to Gaussians in case of use on a wavelet basis
(wavelet-based PAW calculations are now available; in a beta stage)
Also, icoulomb > 1 can be used with plane waves and PAW (also in parallel).
New Tests bigdft#23 and 31
Work by T. Rangel + M Torrent, with the help of Luigi Genovese, Damien Caliste
D.2 Large improvement of parallelism (over bands and k points) in the response function part of ABINIT (and corresponding testing)
In particular, MPI-IO can be used with DFPT.
However, one should still set up a test of this feature.
By Matteo Giantomassi (and G. Huhs for the testing)
D.3 Continuing the development of orbital magnetism.
In particular, the Chern number calculation now works in both nspinor=1 and nspinor=2 cases.
Updated documentation for bfield, add documentation for orbmag variable
By J. Zwanziger
D.4 Continued development af a completely new set of scripts for the test suite, by M. Giantomassi.
D.5 Continue to work on the FFTs.
* Including the interface fourdp and fourwf with DFTI wrappers (fftalg=512)
Added unitary tests #fourwf06 to #fourwf13
Added support for DFTI FFT library in the build system (seq and threaded version)
Rationalization of the different FFT wrappers.
Introduction of fftalg 5XX (e.g. 511 or 512)
Code the missing fallbacks in GW FFTs when fftalg=112
Continue to work on MKL FFT interfacing (+FFTW3).
By M. Giantomassi
* Now FFTW3 wrappers can be used with MPI (and MPI+opemMP)
Introduction of fftalg 314
LOBPCG algorithm can now use fftalg=31X (e.g. 312 or 314) (FFTW3+MPI)
in addition to fftalg=40X
By F. Dahm and M. Torrent
D.6 Improvement of the ground-state calculations on massively parallel architectures
(LOBPCG algorithm on several hundreds of processors)
- ELPA library can be used (Eigenvalue soLvers for Petaflop Applications)
See : http://elpa-lib.fhi-berlin.mpg.de/wiki/index.php/Main_Page
The build system has to be adapted (with-linalg-flavor="elpa" does not exist)
- Load balancing over bands has been improved in the LOBPCG algorithm
- Load balancing over plane waves has been improved in the LOBPCG algorithm
(only available with fftalg=314)
- The use of linear algebra libraries has been rationnalized in the LOBPCG
algorithm in order to make it easier to incorporate new libraries.
- openMP is now available for PAW calculations (non-local operator
has been adapted)
By F. Dahm, L. NGuyen and M. Torrent
D.7 Continued efforts on the development of the abipy library
Abitools project from G. Antonius has been merged in abipy
From M. Giantomassi and G. Antonius
D.8 Continued work on GPU and hybrid architecture (see point B.1 of release notes for ABINITv6.12,
and point D.8 or release notes for ABINITv7.0),
- Now ABINIT+Cuda works on Fermi and Kepler NVidia GPU.
- Interfaces for PLASMA library (matrix algebra)
(Parallel Linear Algebra for Scalable Multi-core Architectures, http://icl.cs.utk.edu/plasma/)
Plasma is usuable but not yet efficient because it misses the computation of eigenvectors
(for this, standard CPU routines are used).
Also, the user must be careful in the distribution of nodes and cores (one full node must be
reserved for PLASMA)
By F. Dahm, L. NGuyen and M. Torrent
D.9 A patch for improved KSS files (allowing to treat pseudopotentiels
with several projectors, and also spin-orbit coupling), readable by YAMBO,
has been produced by Davide Sangalli. See patch.Yambo and patch.YamboHowTo
in the top abinit directory.
D.10 Further work on the ABINIT GUI. Drag and drop graphics utility,
more windows, correction of bugs. Not committed though.
Work by F. Abreu and Y. Gillet.
D.11 Added outphbtrap to output BoltzTraP format file of frequencies on a regular grid for interpolation and transport.
Also tests the *atprj_bs variables, for atomic projections of the phonon band structure
Test v6#76 (not entirely new, but changed).
By M. Verstraete
D.12 Introduction of a double grid integration, for the computation of the temperature dependence of the electronic structure.
New input variables and values : thmflag=7, getkfine, irdwfkfine
Tests v7#50 to 54
Work by S. Ponce
D.13 The vdW-DF1 and DF2 are now in debugging phase.
Many new input variables have been introduced : irdvdw, prtvdw,
vdw_df_acutmin, vdw_df_aratio, vdw_df_damax,
vdw_df_dcut, vdw_df_dratio, vdw_df_dsoft,
vdw_df_gcut, vdw_df_ndpts, vdw_df_nqpts, vdw_df_nrpts,
vdw_df_nsmooth, vdw_df_qcut, vdw_df_qratio,
vdw_df_rcut, vdw_df_threshold, vdw_df_zab .
From Y. Pouillon and C. Espejo
D.14 Upgrade the Python parser bindings, enable to use MPI compiled code with bindings
From D. Caliste
D.15 Pspio library in development, interfaced with ABINIT.
From M. Oliveira, Y. Pouillon, M. Verstraete, J. Alberdi
Interfacing by Y. Pouillon
D.16 Update atompaw lib : ABINIT is now interfaced with version 3.1.0.9
Work by J.-M. Beuken
D.17 Fixed misconfiguration of the build system for external dependencies
From Y. Pouillon
D.18 Update bigdft lib : ABINIT is now interfaced with bigdft-abi-1.0.4
Work by T. Rangel, J.-M. Beuken
D.19 Further work on the parallelisation over atomic sites of DFPT calculation within PAW
Many improvements / bug fixes for DFPT + PAW
From M. Torrent and M. Delaveau
D.20 Work on abilint by T. Deutsch.
D.21 Work on BSE part of ABINIT, trying to improve the scaling, by Y. Gillet.
D.22 Parallelisation of the computation of the Bader charge computation
By X. Gonze
D.23 Added writing/reading of polyfits.
By D. Waroquiers
D.24 Numerous miscellaneous additional bug fixes (to the sources, as well as to the build system,
including patches for the fallbacks), and improvements of documentation by :
B. Amadon, A. Berger, J.-M. Beuken, F. Bruneval, D. Caliste,
F. Dahm, M. Delaveau, M. Giantomassi,
Y. Gillet, F. Jollet, Y. Pouillon, G.-M. Rignanese,
M. Torrent, M. Verstraete, D. Waroquiers, B. Xu