Copyright (C) 2010-2012 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
February 2012 and November 2012. These release notes
are relative to improvements of ABINITv7.0 with respect to v6.12.
The list of contributors includes :
F. Abreu, B. Amadon, G. Antonius, T. Applencourt, A. Berger,
J.M. Beuken, J. Bieder, E. Bousquet, F. Bruneval, D. Caliste, R. Cohen, F. Dahm, M. Delaveau,
G. Franco, G. Geneste, M. Giantomassi, Y. Gillet, Jiawang Hong, A. Jacques, F. Jollet,
M. Mancini, M. Marques, C. Martins, L. Nguyen, S. Poncé, Y. Pouillon, T. Rangel, A.H. Romero,
M. Stankovski, M. Stengel, M. Torrent, D. Vanderbilt, M. Verstraete, D. Waroquiers,
Bin 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 7.0, released on November 26, 2012.
Changes with respect to version 6.12 :
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 The GW part of ABINIT uses now k-centered sets of wavefunctions (_WFK files),
except for GW_EET formalism.
This implies some numerical differences with respect to calculations with ABINITv6, for GW only.
They should be smaller than 0.1 eV, though.
All the GW reference files have been updated.
A.2 The input variable cd_customimfrqs has been renamed cd_customnimfrqs
A.3 The input variable fixmom has been renamed spinmagntarget
A.4 The capabilities of pawprtden haven been transferred to prtden .
Work by D. Waroquiers
A.5 The input variables npwwfn and nshwfn are not supported anymore (use ecutwfn instead)
__________________________________________________________________
B. Most noticeable achievements
B.1. Fixed electric field calculations : other flavours than simply fixing the electric field
are now available, and the geometry can be relaxed (non-zero optcell) in these calculations,
following M. Stengel, N.A. Spaldin and D. Vanderbilt, Nat. Phys. 5,304 (2009).
One can do e.g. calculations with fixed E/D-field or fixed reduced ebar/d field
or mixed electric boundary condition (such as fixed D/d in one direction and fixed E/ebar
in other directions which is useful for the calculation of sandwiched capacitor).
The implementation supports fixed three-dimensional E/ebar or D/d.
The variables to be used in finite field calculations have been
gathered in a new html file doc/input_variables/varrf.html.
New input variables : dfield, red_efieldbar, red_dfield, red_efield,
as well as jfielddir, ddamp, polcen, and maxestep
See the new options in berryopt input variable.
New test cases : seq/v6_121, v6_122, v6_123, v6_124, v6_125, v6_126
More implementation notes can be found in doc/theory/fixd_theory.pdf,
tests of the current ABINIT implementation were presented in J Hong and D Vanderbilt, Phys. Rev. B 84, 115107 (2011).
Other improvements for finite electric field calculations include the possibility to keep the same "branch"
from one iteration to the next one, thanks to the input variable berrysav,
and for finite electric field calculations
starting from a centrosymmetric geometry with non-zero polarization (like AlAs), the
non-zero polarization should be specified thanks to the input variable polcen.
Contribution by Jianwang Hong (+D. Vanderbilt, M. Stengel, R. Cohen).
B.2. Hybrid functional calculations can now be performed, at fixed geometry.
This is implemented in the GW part of ABINIT : either perturbatively (like a G0_W0) calculation,
or with self-consistency (like scQPGW).
Three functionals are available : HSE06, PBE0, and B3LYP.
However, with the current production version of libxc (as of Sept 30, 2012, libxc 1.1.0), only the
B3LYP functional is available. HSE06 and PBE0 are available with the development version of libxc, not yet
delivered on the Web, but only from the repository.
Related input variable : gwcalctyp=105, 125, 205, 225, 305 and 325 , but also icutcoul and rcut.
Related test case : libxc#41 and 42.
Contributed by F. Bruneval (also with help by M. Marques).
Another implementation, more similar to usual ABINIT ground-state calculations is now in progress.
This will allow to compute forces and stresses, and thus, to perform geometry optimizations.
B.3. For DFT+DMFT calculations, the combined parallelization over k-point, plane waves and bands is now available,
and tested, but only without self-consistency of electronic density. Also MPIIO can be used.
Spin-orbit coupling has been implemented.
One can read/write occupations (including non-diagonal ones) as computed in DMFT,
using the input variable dmft_read_occnd.
The tolerance on variation of local charge during iterations can now be specified,
using the input variable dmft_tollc.
See also the new input variables f4of2_sla and f6of2_sla.
Related test case : mpiio#AG , paral#AF , v7#21.
Contributed by B. Amadon
B.4. Parallelisation of the response-function part over different perturbations is now available.
A good balancing scheme is however not yet worked out.
Related input variables : paral_rf, nppert
Related test case : mpiio#AI .
Contributed by F. Jollet (with help of M. Torrent).
B.5 The GW part of ABINIT is able to start from _WFK files instead of _KSS files (except the EET).
Such _WFK files can be generated in parallel, and written/read using MPIIO. Thus there is no
more such a sequential bottleneck for GW calculations.
Work by M. Giantomassi.
B.6. New integration methods and grid options for the integral
along the imaginary axis in GW contour deformation calculations are available.
This is especially useful for very accurate convergence studies.
See the input variable cd_frqim_method : histogram method, trapezoid method,
natural spline, Gauss-Kronrod with different orders.
See also the input variables freqim_alpha, gw_frqim_inzgrid, gw_frqre_inzgrid,
gw_freqre_tangrid. gw_customnfreqsp, nfreqsp, gw_freqsp.
New test : Tv67mbpt#22 .
From M. Stankovski.
B.7 Calculation of polarization and computation under finite electric field are now available
with spinors and spin-orbit coupling.
Test for norm-conserving pseudopotentials : Tv7#03 .
From J. Zwanziger.
B.8 The U value for DFT+U calculations can now be computed in the cRPA method,
for non entangled correlated bands.
Test Tv7#22,23,24,25 (for SrVO3), and new input variables ucrpa and ucrpa_bands.
Contributed by T. Applencourt and B. Amadon.
B.9 The nudged-elastic band (NEB) method to find path between local minima has been coded.
This include also the "Climbing-image nudged elastic nband" method.
Different solvers are available for the solution for the corresponding ordinary differential equation.
Local and (highly experimental) global BFGS for NEB Variable spring constants for the NEB.
Related work on the string method. Runge-Kutta 4 available for both NEB and string method.
The string methods with equal arc length, and with energy-weighted arc lengths
are available, thanks to the input variable string_algo, with values 1 and 2.
Previously, only equal arc length method was available.
No test available yet for string_algo=2 .
See the input variables : imgmov=5, neb_algo (CI_NEB with neb_algo=2), neb_spring,
mep_mxstep, mep_solver.
Contributed by M. Torrent
B.10 The genetic algorithm to find global optima thanks to a search based on a population
of images, has been implemented.
Call the genetic algorithm with imgmov=4 .
Related input variables : ga_rules, ga_n_rules, ga_opt_percent, ga_fitness. Also,
random_atpos.
Contributed by A.H. Romero and X. Gonze
B.11 The vdW-DFT-D2 (Grimme) semi-empirical potential to mimick the Van De Waals interaction is available.
See Tvdwxc#04, and input variable vdw_xc (that must be equal to 5) and vdw_tol.
From M. Torrent
The Ambrosetti and Silvestrelli http://arxiv.org/abs/1111.6737 van der waals functional
is available, see Tv7#11 .
From C. Espejo, Y. Pouillon and A. Romero.
The tests in tests/vdwxc have been activated (they were not previously).
B.12 Significant progresses have been done for the usage of OpenMP, especially for FFTs.
In particular, using OpenMP for the FFT for GW allows to bypass
memory problems for the largest calculations on massively parallel machines.
[Vnl,r] is parallelized using OpenMP, Hilbert method as well..
Work by M. Giantomassi
Activation of openMP for non-local operation in the ground-state part of ABINIT
in the PAW case (useylm=1).
Activation of openMP in the linear algebra of the LOBPCG algorithm.
Work by L. Nguyen, M. Torrent.
B.13 MPI_IO can be used for the writing/reading of the SCR files.
Work by M. Giantomassi
B.14 Mixalch and amu are now evolving variables . This means that
they can differ in different images. All the images algorithm might
be combined with such possibility.
B.15 The new input variable paral_atom has been introduced, to control
the parallel distribution of memory of PAW on-site data over atomic sites.
It is now activated by default. Calculations are
also distributed using the "kpt-band" communicator.
Only compatible for the time being with ground-state calculations.
No test activates it, however.
Work by M. Delaveau, M. Torrent, F. Jollet.
__________________________________________________________________
C. Changes for the developers
C.1 The directory tests/gwdp has been removed : the duplication of tests was costly to maintain, and
did not appear interesting at present.
C.2 The directory tests/v6 has been split in tests/v6, tests/v67mbpt and tests/v7, for more flexibility in the execution
of tests, and because ABINITv7 had been started.
C.3 Two DFPT tests have been activated in tutoparal, with 24 and 32 processors, on the Cronos machine, that is dedicated
to parallel tests with more than 10 procs. More tests from the Bruyeres 2011 tutorial should be activated in the future ...
C.4 There have been many improvements for the test farm.
Among other developments :
- new nightly builders :
- cronos, to test parallel runs with up to 32 cores
- toum gcc46 to test openmp
- buda_gcc44_cuda to test GPU : 4 different tests have been activated
- bigmac_gcc47_sl for testing scalapack 2.0.2 (no yet nightly)
- petrus_gcc47 under openSUSE 12.1
- switch of the reference compiler from gcc44 to gcc47, on testf.
- splitting of all the tasks previously done by the shiva builder in two different builders : shiva and vishnu.
this is for faster execution of this large set of tests.
The tests in the tests/vdwxc have been activated .
The U determination tutorial is now automatically tested.
Work by JM Beuken and X. Gonze
C.5 Fox4.1.2 works on intel 12 (tikal and coba2)
C.6 NetCDF+ETSF_IO have been activated on all machines, except Fock. Should soon become mandatory.
C.7 Split of test-farm and examples config files generation
Work by Y. Pouillon
C.8 The implementation of parallelism in ABINIT has been considerably cleaned.
As an exemple, routines xme_init and xcomm_world have been removed. File xdef_comm has been removed.
Work by F. Jollet
C.9 Fully activated OpenMP support in the build system
Work by Y. Pouillon
C.10 The old, unused, "newsp" utility has been removed.
C.12 Memory profiling has been coded, see m_profiling module.
Tested automatically on max2_gcc46 .
Work by D. Waroquiers.
C.13 Dead project "rdm" routines have been suppressed.
By M. Giantomassi.
__________________________________________________________________
D. Other changes
(or on-going developments, not yet finalized).
D.1 Cleaned the tutorial basic paral. Removed t_paral#03
Cleaned the following tutorials : spin, PAW1, PAW2, elastic, ffield
Work by M. Torrent, X. Gonze, J. Zwanziger
D.2 Creation of the library abipy-0.1.0.dev .
From M. Giantomassi
D.3 New tests for Bigdft tests/bigdft#06,07,14 : surfaces , also test the diagonalisation scheme with wavelets
From D. Caliste
D.4 Test of parallelisation when a processor is unoccupied, paral#AH
From F. Jollet
D.5 Calculation of quasi-particle energies using the effective energy technique (EET) for metals.
Creation of Tv67mbpt#28, for aluminum.
Test contributed by A. Berger.
D.6 Printing the geometry in the POSCAR format.
Input variable prtposcar. Test v7#01.
From M. Verstraete.
D.7 An include mechanism is now defined for the input file.
See Tv7#02
From M. Torrent
D.8 Continued work on GPU (see point B.1 of release notes for ABINITv6.12),
including MAGMA.
Introduction of the new input variable gpu_linalg_limit.
From M. Torrent and F. Dahm
D.9 The new input variable np_slk has been introduced, to control the
use of scalapack calls in the paral_kgb=1 and LOBPCG algorithm case.
Not all MPI processors should be used together for the linear algebra.
From M.Torrent and F. Dahm
D.10 The distance (in configuration space) used is the String Method has been
changed. A true L2 norm is now used.
Contributed by A. Romero
D.11 Added Entropy record in MOLDYN and HIST files.
Work by M. Mancini.
D.12 One can have a finer control on the conjugate-gradient algorithm used to minimize the residual
of the wavefunction, thanks to the input variable tolrde. Previously, it was implicitly
set to 0.005, while it is now adjustable.
See test atompaw#04.
D.13 Implementation of the path-integral molecular dynamics with the (N,P,T) ensemble
is going on. Velocities of the cell parameters can be specified with vel_cell (evolving variables)..
From G. Geneste and M. Torrent.
D.14 The c parameter in the meta-GGA TB09 functional can be fixed, instead of adjusted, like in the original
functional. Use the input variable xc_tb09_c to do this.
See test Tlibxc#13.
From D. Waroquiers.
D.15 Many minor fixes of the Web tutorial have been made, thanks to problems identified by
the tutors and the students at the CECAM school Zurich 22-26 October 2012
(E. Bousquet, M. Torrent, J. Zwanziger).
D.16 Development af a completely new set of scripts for the test suite, by M. Giantomassi.
D.17 Development of the Continuous Time - Quantum Monte Carlo and its interfacing with DMFT.
This work by J. Bieder and B. Amadon is going on, and has not yet been merged in the trunk.
D.18 The ability to read XML PAW data has been implemented (still to be tuned, though).
Work by F. Jollet
D.19 The new input variable mdf_epsinf has been introduced, to specify the longwavelength of the
model dielectric function for BSE calculations.
Work by M. Giantomassi
D.20 Continued work on the recursion method (high-temperature DFT), by M. Mancini
D.21 Interfaced fourdp and fourwf with DFTI wrappers (fftalg=512)
Work by M. Giantomassi
D.22 Developed the polyfit script to plot GW full band structures, starting from DFT band structure
+ a few GW corrections.
Also for the spin band structures.
Work by D. Waroquiers
D.23 Switch to atompaw 3.1.0.9
From Y. Pouillon and M. Torrent
D.24 The size of DDK files can be decreased (no storing of the wavefunctions !), to initialize optic runs.
This is activated by the use of prtwf=3 .
D.25 Introduced diagonalisation scheme for wavelets.
Update BigDFT version : now bigdft-1.7-dev.12
Work by D. Caliste.
D.26 Work on the electron-phonon coupling part of ABINIT:
decomposition with spin; tetrahedron weights for computation of transport properties ;
elastic LOVA : computation of electron life time ;
beyond LOVA : computation of electrical resistivity (with Fermi-smearing) ;
parallelism of elphon computation over k points.
Still buggy, should not be used in production.
Work by Bin Xu and M. Verstraete.
D.27 Implementation of the monopole correction for charged supercells.
This correction is automatically computed whenever the input variable "charge"
is non-zero. It is printed in the output file, and proposed to the user, to be
included as a correction to the total energy.
Such correction is computed in tests v2#42, v3#54, v5#42, v6#67.
Work by F. Bruneval
D.28 New option enable_mpi2_inplace to enable in-place MPI_ALLREDUCE
Work by M. Giantomassi
D.29 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.30 With prtfsurf=1 in ANADDB, an additional output of eigen energies and velocities is written in the VTK format,
which allows the visualization of Fermi velocities on the Fermi surface
Work by Bin Xu
D.31 The capability to compute M-resolved partial DOS with prtdosm=2 is now available.
Work by B. Amadon
D.32 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
G. Antonius, J.M. Beuken, E. Bousquet, F. Bruneval, D. Caliste, M. Delaveau,
G. Franco, M. Giantomassi, Y. Gillet, A. Jacques, F. Jollet,
S. Poncé, T. Rangel, A.H. Romero, M. Stankovski, M. Torrent, M. Verstraete, Bin Xu, J. Zwanziger, D. Waroquiers.