ABINIT 2009 - 4th developer workshop

Autrans (France) 24-27th March

Time schedule

The workshop provides the following information:

The abstracts of orals, announces or posters are after the time schedule.
Report any mistake or omission to Thierry Deutsch (thierry.deutsch AT cea.fr).

Click on the name of a speaker to see its abstract. Keys:

Tuesday 24th Wednesday 25th
Morning

Methodological developments:

 9.00
10.15
Tea/coffee break

Links with communities:

 10.45
12.00

Scienomics

 12.30
Lunch time Lunch time
Afternoon 14.00 Introduction + SWOT
Ch.: X. Gonze, T. Deutsch & G.-M. Rignanese

Software engineering:

Ch.: M. Cote 		14.00
14.30

PAW session:

 14.50
15.50

Discussion (30')

Mod.: M. Cote
An.: X. Gonze, J.-M. Beuken & A. Jacques
  • Merge/testing
 15.20
Tea/coffee break Tea/coffee break
16.20

Correlated electrons:

Software engineering:

Ch.: M. Cote 		15.50
17.20

Discussion (30')

Mod.: M. Cote
An.: T. Deutsch & Y. Pouillon
  • Build system
 16.30
17.30

Discussion (30')

Mod.: J.-M. Beuken
  • Web

Software engineering:

Ch.: M. Cote 		17.00
18.00

Discussion (25' + 25')

An.: A. Jacques, D. Berkholz & G.-M. Rignanese 		17.25
19.00

Poster session

 Advisory board 18.15
Thursday 26th Friday 27th
Morning 9.00

Perturbation:

Ch.: R. Caracas

Applications:

Ch.: J. Zwanziger 		9.00
10.25 10.00
Tea/coffee break Tea/coffee break
10.55 10.30
11.30

Discussion (40')

Mod.: R. Caracas
An.: X. Gonze, Y. Pouillon & M. Giantomassi
  • Beautification

Discussion (80')

Mod.: J. Zwanziger
An.: X. Gonze & Y. Pouillon
  • ABINIT6
  • Debriefing
 11.10
12.10 12.30
Lunch time Lunch time
Afternoon 14.00

GW + BS:

Ch.: M. Torrent
15.40
Tea/coffee break
16.10
16.35

High performance:
17.35

Discussion (HPC) (55')

Mod.: M. Torrent
An.: F. Bottin, L. Genovese, M. Oliveira & M. C�t�
  • GPU
  • Multi-parallelism
Evening

Social diner

Talks

Plane wave based electronic structure calculations using DMFT and projected local orbitals

AMADON Bernard - The description of realistic strongly correlated systems has recently advanced through the combination of density functional theory in the local density approximation (LDA) and dynamical mean field theory (DMFT). This LDA+DMFT method is able to treat both strongly correlated insulators and metals. Several interfaces between LDA and DMFT have been used, such as (Nth order) linear muffin-tin orbitals or maximally localized Wannier functions. Such schemes are, however, either complex in use or additional simplifications are often performed (i.e., the atomic sphere approximation). We present an alternative implementation of LDA+DMFT, which keeps the precision of the Wannier implementation, but which is lighter [1]. It relies on the projection of localized orbitals onto a restricted set of Kohn-Sham states to define the correlated subspace. The method is implemented within the projector augmented wave and within the mixed-basis pseudopotential frameworks. This opens the way to electronic structure calculations within LDA+DMFT for more complex structures with the precision of an all-electron method. In this presentation, we will briefly present the theoretical framework, discuss practical details and prospects of the implementation, and give some examples of applications.

  1. B. Amadon, F. Lechermann, A. Georges, F. Jollet, T. O. Wehling, and A. I. Lichtenstein Phys. Rev. B 77, 205112 (2008)

Computational investigation of the atomic structure of Mn-rich nanocolumns: comparison with a possible GeMn ordered compound

ARRAS Emmanuel - High curie temperature (> 400K) GeMn alloys have recently been grown by means of low temperature molecular beam epitaxy. The samples exhibit Mn-rich nanostructures embedded in a nearly pure germanium matrix. The composition of theses nanostructures is close to Ge2Mn but their atomic structure is still puzzling. A recent Extended X-Ray Absorption Fine Structure (EXAFS) study reveals that these nanostructures present a complex local structure that does not correspond to any known stable GeMn compound [2]. Based on electronic structure calculations we have investigated both magnetic and structural properties of a possible GeMn ordered compound found in similar transition-metal metalloid systems. The comparison is based on both magnetic properties of the relaxed compounds as well as their simulated EXAFS spectra. This methodology was first validated on the Ge3Mn5 compound and then applied to other ordered compounds. Moreover the Curie temperature is calculated by means of coupled KKR-Monte Carlo calculations for the compounds that exhibit a stable ferromagnetic ordering.

  1. M. Jamet , A. Barski, T. Devillers, V. Poydenot, R. Dujardin, P. Bayle-Guillemaud, J. Rothman, E. Bellet-Amalric, A. Marty, J. Cibert, R. Mattana, and S. Tatarenko, Nat. Mater. 5, 653 (2006).
  2. M. Rovezzi, T. Devillers, E. Arras, F. d'Acapito, A. Barski, M. Jamet, and P. Pochet, Appl. Phys. Lett. 92, 242510 (2008).

How to be a distribution-friendly project

BERKHOLZ Donnie - Getting ABINIT integrated into Linux distributions will help to increase its adoption and to ensure that users have ABINIT well-integrated into their systems and kept up-to-date easily. ABINIT is already packaged in Gentoo Linux, and Donnie will share his experience creating and maintaining ABINIT's Gentoo package and offer suggestions for improvement. Technical and philosophical questions that determine how easy or difficult it is for distribution packagers to work with ABINIT developers will be discussed. Technical issues include the basic metaphor that ABINIT's build and installation process is an API to distribution packages -- it should be changed carefully and purposefully, and changes should be well-documented. In addition, Donnie will describe the level of control and system integration desired by packagers. Philosophical issues, including user expectations and licensing requirements, differ between distributions and can cause major conflicts with upstream developers. Finally, Donnie will discuss developments toward the future of distribution packaging so that ABINIT developers can consider how this fits into the future of ABINIT.

Superconductivity near a lattice phase transition: the case of NbN

BLACKBURN Simon - We report the study of the electron-phonon coupling in NbC1-xNx crystals in the rocksalt structure. The Kohn anomaly associated with the topology of the Fermi surface greatly increases the electron-phonon coupling and induces a structural instability when the electronic density of states reaches a critical value. We develop a model of the Eliashberg spectral function where the effect of the unstable phonons is set apart. We show that this model within the McMillan formula can reproduce the increase of Tc near the structural phase transition.

From application-centric to data-centric computation: use of ab initio methods in applied research

BLAISE Philippe - We present some results in applied electronics obtained with ab initio methods: computation of realistic band offsets of a high-K MOS gate stack, using a combination of Siesta (DFT/LDA) and Abinit (G0W0), calculation of electronic levels useful for future development of flash memories. A more general discussion will follow about crucial differences when one uses ab initio methods for fundamental or applied research. Some perspectives will be drawn about the embedding of an application like Abinit in a typical workflow for e-science.

Ab initio calculation on three levels of parallelization

BOTTIN Fran�ois - Three levels of parallelization [1] are now available in the ABINIT code package. Indifferently, these ones can be used separately or coupled the ones with the others. To the parallelization on k-points, previously introduced into the code, we add two others of them acting on bands and plane waves. These two last levels of parallelization are based on two algorithms previously introduced in the code: on one hand, a three-dimensional Fast Fourier Transform (3dim-FFT) [2], which allows a parallelization on plane waves, and on the other hand, a blocked eigensolver (LOBPCG) [3], in order to perform a band parallelization. First, we briefly recall the implementation and the efficiency of this parallelization. In particular, we indicate the last implementations carried out in the code and the new features which are now available. Then we focus on various sudies, involving a large number of CPUs (up to one thousand), bands (a few thousands), atoms (a few hundreds) and/or plane waves [4], in order to illustrate the use of this parallelization. For each system we indicate the CPU time needed to perform such a calculation.

  1. F. Bottin, S. Leroux, A. Knyazev, and G. Zerah, Comput. Mater. Sci. 42, 329 (2008).
  2. S. Goedecker, M. Boulet and T. Deutsch, Comput. Phys. Comm. 154, 105 (2003)
  3. A. Knyazev, SIAM Journal on Scientific Computing 23, 517 (2001).
  4. S. Mazevet, F. Lambert, F. Bottin, G. Zerah and J. Clerouin, Phys. Rev. E 75, 056404 (2007); F. Bottin and G. Zerah, Phys. Rev. B 75, 174114 (2007)

Temperature Dependance of the band energies of semiconductors

BOULANGER Paul - The band energies of semiconductors are known to exhibit significant shifts and broadening with the variation of temperature. At constant volume, this dependance is a consequence of the renormalization of band energies due to electron-phonon interactions. This effect was first fully described in a formalism developed by Cardona and coworkers[1] within a semi-empirical context.

We have reanalyzed this formalism in a density functional perturbation theory approach (DFPT). This new formalism includes an extra term, the non-diagonal Debye-Waller term, which can be calculated with the help of DFPT. We have studied the importance of this contribution on diatomic molecules using frozen phonon and DFPT calculations. For the H2 molecule, it contributes to up to 15% of the total temperature dependence. This formalism might help explain the discrepancy still found between calculations and experiments for solids[2].

In this presentation, the details of this implementation in the Abinit package will be discussed and examples on how it can be used will be given.

  1. P.B Allen and M. Cardona, Physical Review B 23, 1495 (1981)
  2. D. Olguin, M. Cardona and A. Cantarero, Solid State Communications 122, 575 (2002); M. Cardona, Solid State Communications 133, 3 (2004)

Fast GW approximation with only a few empty states

BRUNEVAL Fabien - The GW approximation to the electronic self-energy yields band structures in excellent agreement with experimental data. Unfortunately, this type of calculation is extremely cumbersome even for nowadays computers. The huge number of empty states required both in the calculation of the polarizability and of the self-energy is a major bottleneck in GW calculations. We propose an almost costless scheme, which allows to divide the number of empty states by about a factor of five to reach the same accuracy. The computational cost and the memory requirements are decreased by the same amount, accelerating all calculations from small primitive cells to large supercells. I will describe how this method has been implemented in Abinit.

Parser and visualisation: how to link ABINIT and V_Sim?

CALISTE Damien - Dealing with developments around the main ABINIT loop, this talk will focus on two linked topics. The first one will present how the input file parser has been extracted and is now exposed to third party developers. The second topic will deal with the visualisation capabilities of V_Sim [1] developed these last two years and specifically these related to ABINIT.

Either before the calculation for checking purposes or after the calculation for analysis and sorting purposes, one would like to access the ABINIT input file easily and independently from the language one uses. The ABINIT parser is a quite complex object that would be useless to duplicate and which has interesting capabilities like the unit parameters, the array notation, the data sets... A Fortran90 module has thus been created linking with the ABINIT own routines. The purpose of this module is to expose an object oriented interface to be able to bind easily in scripting languages like Python. This part of the talk will present the available interfaces in Fortran90, C and Python.

The development of V_Sim [1] has tightly followed the development of computational softwares around ABINIT. Indeed, using the C bindings of the ABINIT parser, one can now visualise directly whatever step of a data set directly loading the input file. Beside that, several capabilities have been added, like the box duplication that allows to clearly understand the atomic relations from a file describing a primitive cell; or the density mapping, the interactive geometry modifications...

  1. visit the web site or download Debian packages on Debian servers.

High-pressure study of water ices VII-VIII-X

CARACAS Razvan - We study high-pressure solid H2O ice: the lattice dynamical properties of ice X and the transition path between molecular ices VII/VIII and the ionic ice X with first-principles calculations using density functional theory in the ABINIT implementation. Our work [1] defines the dynamical stability of ice X between about 120 GPa up to about 400 GPa. Based on phonon band dispersion we show that the phase transition sequence at low temperature and high pressures in ice is ice VIII - disordered ice X - ice X - ice Pbcm. The disordered ice X is due to a phonon collapse in the whole Brillouin zone at pressures below 120 GPa, phonon that corresponds to hydrogen atoms bouncing back and forth between every two oxygen neighbors in a double well potential. Post-ice X is orthorhombic Pbcm and appears due to phonon instability in M at pressures higher than 400 GPa that distorts the bcc cubic sublattice of oxygen atoms into a hcp-like structure. Our calculations validate earlier theoretical predictions for a phase transition to a post-ice X structure in H2O [2]. We also identify and discuss the (meta)stability of several intermediate phases between ice VIII and ice X and try to explain some new experimental data.

abilint: python script to help developers

DEUTSCH Thierry - The python script abilint is used to generate automatically interfaces from Fortran routines and dependencies in the build system. This script was developed to help developers with the goal to be ignored. Due to the diversity of developers, this objective was achieved by developing a Fortran parser!

In this talk, I will present the main characteristics of abilint and gives some ideas for its future use to improve modularity and quality of the ABINIT code.

ab initio simulation of large systems in complex environments : the BigDFT project

GENOVESE Luigi - The european BigDFT project has developed an ab initio Density Functional Theory code based on Daubechies wavelets. Such functions have features which make them a powerful and promising basis set for application in materials science. These are a compact support multiresolution basis, and form one of the few examples of systematic real space basis sets. For these reasons they are an optimal basis for expanding localised information. The real space description they provide allows to build an efficient, clean method to treat systems in complex environments, like surfaces geometries or system with a net charge. The mathematical properties of the formalism are optimal to build a robust, highly optimised code, conceived for systems of few hundred atoms, with excellent efficiency on parallel computers. During this talk we will illustrate the main features of the code, its actual performances and capabilities. We will then conclude by outlining the planned developments and the potentialities of this powerful formalism in the context of electronic structure calculations.

Implementation of the GWA in the framework of PAW

GIANTOMASSI Matteo -

Gathering the contributions of the ABINIT developer : the whole story.

pdf[180k]

GONZE Xavier - Over the years, many different people have submitted contributions to the ABINIT project. Despite their diversity, the vast majority of these contributions have been successfully integrated into ABINIT, without ABINIT falling apart... I will survey the reasons, the tools and software techniques that have made this possible. This will give justification to the presence of the rules to be followed by developers, concerning automatic tests, structure of the files, requested documentation, etc. , and, I hope, also more motivation to follow them !

Local exact exchange in the Projector Augmented-Waves framework: Implementation in the ABINIT code, validation, and application to actinide compounds

JOLLET Francois - Exact exchange is used to describe correlated orbitals in the Projector Augmented Waves (PAW)framework. As suggested recently in another context [1], Hartree-Fock exchange energy is used for strongly correlated electrons only inside the PAW atomic spheres. This is done thanks to the PBE0 exchange-correlation hybrid form functional. This method is tested on NiO and results well agree with already published results and LDA+U calculations. It is then applied to UO2 and PuO2 for which the results well agree too with LDA+U calculations, but without adjustable parameter.

  1. P. Novak et al., Phys. Status Solidi B 243, 563 (2006)

MAPS - A flexible modeling environment for efficient delivery of Open Source code

KROKIDIS Xenophon - The Materials And Processes Simulations Platform (MAPS) is a molecular modeling environment offering a variety of tools allowing the building of molecular models (finite and periodic), the access to simulation codes, job management and analysis of results. Today the technology in MAPS covers the areas of quantum, classical, mesoscale and chemical engineering simulations. Moreover, MAPS offers full Python scripting capability that allows the creation of customized graphical user interfaces and simulation scenarios.

Scienomics likes to promote the MAPS platform as a key tool to help open source solutions to become more popular in the scientific community. The plugin based architecture of MAPS and its robust Application Programming Interfaces (APIs) toolbox make MAPS an excellent tool for developing graphical user interfaces for simulation engines. The usage of a simulation code within MAPS is facilitated even for non expert users since they can benefit also of all auxiliary technology in MAPS (builders, visualization, analysis and other simulation engines). This is particularly important for industrial users who usually perform multi-scale modeling.

MAPS includes an interface to ABINIT and to other open source or free simulation engines. In particular ABINIT users can easily build their models and perform some of the sophisticated calculations allowed by ABINIT with a few steps guided by the graphical user interface. Moreover, the results can be visualized and analyzed within the framework of MAPS.

However, in order to increase efficiency a close collaboration with the communities developing the open source tools and Scienomics is required to deliver (a higher) high end technology to the industrial community.

A Recursion method in abinit

MANCINI Marco - Classical Abinit simulations are limited to temperatures up to approximately 10eV as a consequence of the very rapid growth of the number of electronic states when solving the Mermin-Kohn-Sham effective Schr�dinger equation. The recursion method is based on a direct evaluation of the density matrix in the real space representation. This allows for an orbital free computation of the charge density in the Kohn-Sham formalism at finite temperature whose numeric complexity increases linearly with the size of the system (order N method). In our implementation, the density matrix is computed by a recursion method based on the Trotter formula which allows simulation up to very high temperatures. We present the Recursion method and its Abinit implementation, its advantages (high temperature simulations, good parallelization properties) and limitations (low temperature performance, no yet non-local potential). We illustrate the assessment with numerical tests performed to the computation of the Hugoniot curve of cryogenic Helium using two different pseudo-potentials, and we compare our results with Path Integral Monte Carlo simulations and experimental data.

An ab initio tool for excited state calculations

MARINI Andrea - Yambo[1] aims to be a combination of several projects devoted to the device of new theoretical and/or numerical tools for calculating, ab initio, the excited state properties of electronic systems. In practice, Yambo is a mixed Fortran/C code composed of a core part and many extensions. These extensions are pieces of code embodied in the main Yambo trunk, that can be isolated or joint together. The result is one source but many possible executables. The core of yambo calculates quasiparticle energies and optical properties within the framework of many-body perturbation theory and time-dependent density functional theory, and it is released under the GPL license. Quasiparticle energies are calculated within the GW approximation for the self-energy. Optical properties are evaluated either by solving the Bethe-Salpeter equation or by using the adiabatic local density approximation. However yambo features are very much extended by the different projects that cohabit in the code: total energy, electron-phonon coupling, surface spectroscopy, non equilibrium time-dependent dynamics, and so on. Yambo is a plane-wave code that, although particularly suited for calculations of periodic bulk systems, has been applied to a large variety of physical systems. Yambo relies on efficient numerical techniques devised to treat systems with reduced dimensionality, or with a large number of degrees of freedom. The code has a user-friendly command-line based interface, flexible I/O procedures and is interfaced to Abinit, PWscf and with the "ETSF file format"[2]. In this talk I will outline the history of the code and its philosophy. I will briefly describe some of the code's features and specific numerical tools. I will also present the projects that cohabit in the code and some of the most recent and intriguing results[3] we have obtained thanks to yambo.

  1. Yambo: "an ab initio tool for excited state calculations", arXiv:0810.3118. Yambo web site
  2. European Theoretical Spectroscpy Facility(ETSF) Standardization Project, ETSF standardisation page
  3. See Yambo publications

libxc - a library of exchange and correlation functionals

MARQUEZ Miguel - The central quantity of density functional theory is the so-called exchange-correlation energy functional. Many approximations exist for it (probably of the order of 150-200), and they are usually arranged in families (LDA, GGA, meta-GGAs, etc.) In spite of these theoretical advances, most computer codes only include a very limited quantity of functionals, typically around 10-15. This choice usually contains the most popular variants, like the Perdew-Zunger or the Perdew-Wang LDAs, or the PBE or BLYP GGAs, but misses many of the old functionals (that are important to reproduce old results) or some of the most recent developments. With this in mind, we have developed libxc, a library of exchange and correlation functionals. This library already includes more than 100 functionals, is available under the LGPL license, and can be easily called both from C and Fortran. Furthermore, it also particularly adapted not only for ground-state calculations, but also for response calculations, as it includes higher derivatives of the exchange-correlation energy

New Phosphors for White LEDs : Theory and Experiment

MIKAMI Masayoshi - A trend to search for new phosphors for white LEDs has been acceralating, because white LEDs are highly expected as one of the most promising light sources with much less energy consumption (less CO2 gas exhaustion) than conventional light sources such as incandescent lamp. In this presentation, we will present how our nitride/oxynitride phosphors have been developed by close cooperation between theory and experiment. In particular, it will be stressed that ABINIT computation was greatly helpful for the discussion on the feasibility of phosphor synthesis as well as the determination of crystal structures. Outlook and issues will be discussed for possible "ab initio design" of phosphor materials.

The Wannier90 Project: Design, Developments, Directions

MOSTOFI Arash - Wannier90[1] is a code for obtaining maximally-localised Wannier functions[2] (MLWF) in a post-processing procedure following an electronic structure calculation. Written in modern, modular FORTRAN, independence from the basis set and methodology used for the underlying electronic structure calculation is a fundamental design feature of the code. Its structure and current scientific functionality will be presented, with particular emphasis on the elements required for a seamless interface to an electronic structure program of choice. Time permitting, some recent developments will be discussed and directions and challenges for the future will be highlighted.

  1. A. A. Mostofi et al., Comput. Phys. Commun., 178, 685 (2008)
  2. N. Marzari et al., PRB 56, 12847 (1997); I. Souza et al., PRB 65, 035109 (2002)

Running ABINIT in a GRID environment

OLIVEIRA Micael - GRID computing facilities are becoming widely available, as they allow research institutions to share their computational resources in an effective and efficient way. Nevertheless, up to now, usage of GRID facilities to run ab-initio codes, like ABINIT, has not been very popular. One of the reasons for this was the deficient MPI support. Indeed, GRID facilities were initially designed having in mind large amounts of small serial jobs, but this is now changing, as support for MPI jobs started to become widely available. In this talk we will present our experience in running ABINIT in a GRID using the GLite software, both in serial and parallel modes. We will discuss the problems encountered, as well as how these were solved. A short introduction to GRID concepts will also be presented.

The mutation of Abinit

POUILLON Yann - During the five last years Abinit has undergone a comprehensive mutation at various levels, including a complete restructuring of the source code and the website, the introduction of enhanced management tools, as well as a substantial increase in terms of size, versatility, visibility and user base. The lifespan of Abinit 5 has moreover let us know much better about the needs of the community and how to drive a sustainable growth.

In addition to improving the development workflows and the conformance of the source code to modern standards, the roles of the various contributors, i.e. end-users, developers and maintainers, have been clarified. Previously, everyone was modifying the source and build system, while further advances required more specialization in the community, e.g. the management of external libraries - in growing number - which has to be done by skilled maintainers.

To address the issues raised by Fortran compilers, and because the ABINIT developers are mostly scientists, it was decided to provide support beyond the GNU Autotools (nowadays the paradigm for binary/package generation) by developing a new build system on top of it, while keeping backward-compatibility with the features present in Abinit 4 as long as possible. Though it has permitted a smoother transition, the latter constraint has also led to a certain number of limitations. Solutions will be proposed to set Abinit 6 free from them.

The ABINIT-Wannier90 interface.

RANGEL Toniatuh - The numerous applications of the Maximaly-Localized Wannier Functions (MLWFs) have motivated many people to work on the development of an interface between ABINIT [1] and Wannier90 [2]. The implementation was first started by Bernard Amadon, and Fran�ois Jollet some years ago [3]. From the beginning, the interface was developed to work with Projected Augmented Waves (PAW) and Norm Conserving pseudopotentials.

As a first improvement to this existing interface, we have provided a variety of possible initial guess for the MLWFs. Indeed, the choice of this initial guess can considerably speed up the convergence of the localization procedure. In particular, for the conduction bands, the convergence cannot be achieved starting with at random.

More recently, we have extended this interface in order to use it in the framework of Many-Body Perturbation Theory (MBPT) for GW calculations. First, the MLWFs can be used to obtain the complete bandstructure based on the GW corrections computed only for a set of special k-points, hence providing a very efficient interpolation scheme. Second, the MLWFs can also be obtained for the quasiparticle wavefunctions when self-consistent GW calculations are performed [4].

As a result of these developments, ABINIT 5.7 has a very robust and user-friendly interface with Wannier90 which can be used with PAW and Norm Conserving pseudopotentials for ground state and GW calculations.

  1. ABINIT web site
  2. Wannier web site
  3. B. Amadon, et al., Phys. Rev. B 77, 205112 (2008)
  4. D. R. Hamann and D. Vanderbilt, Phys. Rev. B 79, 045109 (2009)

Non-collinear magnetism and spin-orbit coupling within PAW

TORRENT Marc - On top on existing implementation of non-collinear magnetism in ABINIT, we have generalized it to the case of the Projector Augmented-Wave (PAW) formalism. Introduction of wave-functions as spinors was necessary in the PAW part of the code, as well as imaginary "on-site" potentials and densities. I will present first the internal representation of each PAW objects, especially on-site densities, occupation matrix and pseudopotential strengths. In the second part of the presentation, I will detail the implementation of spin-orbit coupling in the PAW on-site terms. I will discuss some PAW specifc aspects of the implementation:

Implementation of the DFPT in the framework of PAW: computation of phonons

TORRENT Marc - Two years ago, I presented the formulae derived to apply the Density-Functional Perturbation Theory (DFPT) approach in the framework of the Projector Augmented-Wave (PAW) method. I will briefly recall the key points of the formalism and present some recently published novelties [1]. Then I will restrict the presentation to the implementation of phonons (response of the system to an atomic displacement) and detail the internal representation of first-order perturbed PAW objects and the content of some new PAW routines. A focus on two specific aspects will be done:

As a conclusion, I will show the first calculations done with the DFPT+PAW and compare them with results obtained with the norm-conserving pseudopotential approach.

Finite temperature phonons and the stability of phases

VERSTRAETE Matthieu - The usefulness and implementation in ABINIT of a finite temperature ("self consistent") phonon scheme, invented by Souvakis et al. in PRL 100 095901, is presented. In many crystallographic phase transitions, the high temperature phase presents imaginary phonon modes at low temperature. This precludes the use of the ab initio vibrational spectrum to calculate thermodynamical properties. Increasing the electronic temperature is usually not enough: it is the phonon frequencies themselves which change with (ionic) temperature. This is exemplified by the transition of Ti, Zr, or Hf to a BCC structure at high temperatures, while the 0K phonons present imaginary modes. Starting from the ground state phonons and a (typically 4x4x4) supercell of atoms, the atomic forces, normal mode displacements, and phonon frequencies are solved for self-consistently, accounting for the desired temperature in the statistical occupation of the modes. In this way the high temperature phase phonons can be stabilized, and compared properly to experiments. First examples on Aluminium present encouraging results.

Perturbations in the PAW Formalism: DDK, Electric, and Magnetic Fields

ZWANZIGER Josef - Building on the general PAW DFPT scheme being implemented, we are working to add response functions to electric and magnetic fields within the PAW formalism. This project is in the development and implementation stages. Earlier work by Umari, Gonze, and Pasquarello (PRB 69, 235102 (2004)) showed how to treat both the DDK and static electric field perturbations in the ultrasoft pseudopotential case, and the DFPT PAW equations are very similar (though their implementation is more complex). I will discuss current progress in implementing these perturbations in abinit, including the additional nonlocal potentials required and the unusual asymmetric derivatives of the projectors. Then I will discuss recent work of Xavier Gonze, Marc Torrent, and myself on implementing the response to a static magnetic field. This has required significantly more new theoretical development, and will ultimately represent a substantial step forward as compared to finite-difference methods for treating this perturbation as present in CASTEP and PWSCF

Properties at the Nuclei: Electric Field Gradients and M�ssbauer Isomer Shifts in the PAW Formalism

ZWANZIGER Josef - Key observables in M�ssbauer spectroscopy include the isomer shift and the electric field gradient at each nucleus. The electric field gradient is also observable in nuclear magnetic resonance and nuclear quadrupole resonance. Since the last workshop Marc Torrent and I have implemented the EFG in abinit, and recently I have implemented the Fermi-contact interaction, which is proportional to the isomer shift. I will briefly describe the theory of both these interactions, and show results on various compounds. In particular, I will show how the EFG acts as a useful probe of electron correlation effects in the Mott-Hubbard insulator LaTiO3, and I will show a variety of results on M�ssbauer spectra, including the effect of pressure on on Fe and ZnO.

Announces

An implementation of a consistent, linear-response approach to DFT+U

ADAMS Donat - Traditional Density functional theory [DFT] fails on electronic systems with strong correlations. The total energy of localized electrons is afflicted with the spurious self-interaction energy. DFT+U can cancel this error by introducing a parametrized potential based on the Hubbard model. The parameters of this approximation correspond to the Coulombic repulsion (U) and the exchange energy (J). We present an implementation of a reliable ab initio method to determine these parameters [1].

Our implementation in the ABINIT [2;3] code is based on the work of Cococcioni and de Gironcoli [1] and takes into account the linear response of the system to a potential shift on a localized site with strong correlations. It allows to determine U in the same framework as later the DFT+U is applied. Thus the DFT+U methodology becomes a consistent method without any free parameters.

We will summarize the theory of the procedure for the determination of U and examine the impact of the following choices in the procedure: the symmetry of the unit cell, the amplitude of the shift, convergence with the system size. We will provide the possible simplifications and give an example of an applications to a reference material (e.g. bcc Fe).

  1. Cococcioni, M. and de Gironcoli, S. (2005). Linear response approach to the calcu- lation of the effective interaction parameters in the LDA + U method. Phys. Rev. B, 71(3):035105.
  2. Gonze, X., Beuken, J.-M., Caracas, R., Detraux, F., Fuchs, M., Rignanese, G.-M., Sindic, L., Verstraete, M., Zerah, G., Jollet, F., Torrent, M., Roy, A., Mikami, M., Ghosez, P., Raty, J.-Y., and Allan, D. (2002). First-principle computation of material properties: the ABINIT software project. Comput. Mat. Sci., 25(478).
  3. Torrent, M., Jollet, F., Bottin, F., Z�rah, G., and Gonze, X. (2008). Implementation of the projector augmented-wave method in the abinit code: Application to the study of iron under pressure. Comput. Mat. Sci., 42(2):337�351.

Ab-initio GW self-energy corrections in metallic systems

CAZZANIGA Marco - The calculation of GW corrections in metallic systems is prone to the problem ofintraband contributions in the small wavevector limit of the dielectric matrix. We demonstrate that neglecting the intraband term leads to an opening of an unphysical gap in the quasiparticle dispersion, which only disappears using an extremely and prohibitively dense k-point mesh. To overcome this difficulty we propose a simple method based on a fit on the asymptotical trend of the independent particle polarizability [1], correcting the wrong term of the dielectric matrix. We are planning to include this approach in ABINIT. This simple approach preforms well in combination with the plasmon pole model, providing correct bandwidths, and adding no computational overhead with respect to a standard calculation for semiconductors. On the other hand, going beyond the plasmon-pole model with a contour deformation approach reveals problematic for very small frequencies, and different approaches have to be preferred. We discuss the possibility of calculate quasiparticle corrections for metallic systems also with method beyond the Plasmon pole model.

  1. Cazzaniga et al., Phys. Rev. B 77, 035117 (2008)

Implementation of Raman scattering for spin polarized systems

DA PIEVE Fabiana - Raman spectroscopy is one of the most spread tool in order to study lattice dynamics. The intensities of the Raman peaks and thus the calculations of the full Raman spectra depend on the mode frequencies, which are obtained straightforwardly from the diagonalization of the dynamical matrix in the density functional perturbation theory, and on the Raman tensors, which represent the first derivative of the dielectric tensor with respect to the atomic displacement pattern corresponding to the vibrational mode. The Raman tensors can be computed either from finite differences, as the change of the dielectric tensor due to infinitesimal atomic displacements, or from perturbation theory, as the third order derivative of the energy with respect to two electric fields and one atomic displacement. Actually in the ABINIT code it is only possible to calculate the Raman tensor for LDA functionals. However, it is sometimes necessary to go beyond this simple approximation, either by supporting the spin polarized LDA case, or going beyond LDA.

I will report on the current effort going on in order to implement the spin polarized case, which will allow to study Raman scattering from magnetic systems. Work is in progress, and we will give a simple description of the project, as well as the already established formal results.

Implementation of the modern theory of polarization within the PAW method

HERMET Patrick - Nowadays, the modern theory of polarization is intensively used to compute the spontaneous polarization in ferroelectrics or the Born effective charge tensors in materials.

Several years ago, King-Smith and Vanderbilt gave a formulation that allows for the calculation of the electronic polarization of an insulating crystal in the context of a conventional electronic band structure calculation [1]. They showed that the contribution of a given band to the electronic polarization could be expressed as a geometric quantum phase or Berry phase of the cell-periodic Bloch function uk as the wave-vector k is adiabatically transported by a reciprocal lattice vector. Resta has given a review of this theory [2]. This theory can be applied straightforwardly in the context of an all-electron scheme, or any norm-conserving pseudopotential scheme. However, in the case of ultrasoft pseudopotential scheme or projected augmented waves (PAW) method, the pseudo-wavefunctions do not obey a conventional normalization condition, and the Berry phase expression must be modified according to Ref. 3.

In this talk, I present the appropriate formulation of the electronic polarization within the PAW method as implemented in the ABINIT package and I give an example of input file using this implementation. In addition, I compare the Born effective charge tensors of alpha-bithiophene calculated using the Berry phases within the PAW method and the ones calculated using the norm-conserving scheme. The theoretical infrared spectrum of this material will also be confronted to the experimental one.

This implementation is the result of a strong collaboration with the CEA of Bruy�res-Le-Chatel.

  1. R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47 (1993) 1651
  2. R. Resta, Rev. Mod. Phys. 66 (1994) 899
  3. D. Vanderbilt and R. D. King-Smith, unpublished. link

Improving Abinit calculations scalability for large systems : implementation of the RMM-DIIS algorithm

JANSSEN Laflamme Jonathan - The current conjugate gradient method implemented in Abinit for the iterative diagonalisation of the Kohn-Sham hamiltonian allows SCF calculations up to a few hundreds of electrons on current computers. The interest in condensed matter for large simulated systems such as complex nanostructures or amorphous systems requires better scalability with number of electrons in ab initio calculations. The current conjugate gradient method scaling is dominated, in large systems, by the O(N3) orthonormalization step done at each band update. The residual minimization method - direct inversion in the iterative subspace (RMM-DIIS) allows to reduce the number of O(N3) operations by performing orthogonalisation of eigenvectors only after sweeping over all bands, before each charge density update. This poster discusses the project of this implementation in Abinit.

PAW positron lifetime calculation with ABINIT

JOMARD Gerald - Positron Annihilation Spectroscopy is a powerful technique to detect and characterize vacancy-type defects in solids. The repulsive Coulomb potential of the nuclei pushes the positron into vacancies making positrons particularly sensitive to defects. However, experimentally, one can only measure a mean positron lifetime that contains the contributions of all type of defects present in the sample. Consequently, characterizing the defects in a sample requires knowledge of the specific lifetimes for various possible defect structures as well as the undefected bulk. The use of the TCDFT framework (Two Components Density Functional Theory) allows for the calculation of such quantities. As we are interested in MOX nuclear fuels (UO2-PuO2 mixtures) we know that we need to go beyond LDA in order to correctly describe the strong electronic correlations arising in these systems. We choose to work within the LDA+U framework which is available in the PAW method in ABINIT. For that reason we are currently implementing the TCDFT in the PAW formalism. Both formalism and first applications will be presented in this talk.

Improvements in STM calculations and implementation of kinetic energy density in ABINIT

LHERBIER Aur�lien - STM current calculations in ABINIT are presently implemented under the single s-orbital Tersoff-Hamann tip approximation. However, p- and d-orbitals of the STM tip should also present a non-negligible contribution to the total current and are thus sometimes required to obtain a better STM image. This improvements in the tunneling current calculations in ABINIT will be described. As a by-product, such an implementation gives also access to the kinetic energy density. The latter quantity is used in the Electron Localization Function (ELF) theory to track down the localized electrons as well as in the meta-GGA. Its implementation in ABINIT will also be presented.

Exploring energy reaction paths with ABINIT

ROMERO Aldo Humberto -Gathering the contributions of the ABINIT developer : the whole story. We will discuss different implementations to calculate energy reaction paths within ABINIT. In particular, we will focus in the most recent implementation by using the string method, where we are able to find the minimum energy path as well as to locate the transition state to any accuracy. If time allows, we will describe the recent intents to implement the method of metadynamics and to create an interface with ABINIT.

Posters

GW without empty states: Beyond the COHSEX approximation

BERGER Arjan - The GW approximation (GWA) to the self-energy[1] has proved to be very successful in the calculation of quasi-particle energies for a wide range of solids. However, the GWA is computationally expensive which is mainly due to the slow convergence with the number of unoccupied states that have to be taken into account. This has led to to search for simplified approximations such as the COHSEX approximation[1] in which there is no sum over unoccupied states. However, the COHSEX quasi-particle energies are, in general, not close to the GW energies. A different strategy is to look for a more efficient approach to calculate the GW self-energy such that only a small number of unoccupied states are required[2]. In this way the GW accuracy is retained. In this work we will go beyond the COHSEX approximation and obtain an approximation to the GWA without empty states but with a better accuracy than the COHSEX approximation.

  1. L. Hedin, Phys. Rev. 139, A796 (1965)
  2. F. Bruneval and X. Gonze, Phys. Rev. B 78, 085125 (2008)

Raman and infrared spectra of hydrogenated and deuterated talc

BEZACIER Lucile - Raman spectroscopy is a fundamental tool for characterizing crystal symmetry and interatomic bonding. It can also be successfully used as an identification method. Until recently the standard calculations reported only the peak positions for Raman and infrared modes. The very recent algorithmic developments and their implementations in two first-principles packages (ABINIT and PWSCF) allow today the full calculation of Raman spectra.

We use the latest development of the ABINIT package to compute the variation of the Raman and infrared spectra as a function of pressure for hydrogenated and deuterated talc. For the infrared we determine the peak position and for the Raman both peak positions and intensities. We find a good agreement with experiment.

We perform a detailed analysis of the vibrational pattern, with a particular emphasis on the participation of the hydrogen/deuterium atoms. This can be used next as a basis to estimate the isotope partitioning coefficients between various Earth's upper mantle hydrous minerals.

Improved description of Schottky barriers at metal/ferroelectric interfaces

BILC Ioan Daniel - Understanding the behavior of metal/ferroelectric heterostructures and determining the Schottky barrier at the interface is not only theoretically interesting but also crucial for device applications. To quantify the Schottky barriers is however not an easy task, neither at experimental nor at theoretical level. DFT calculations within the usual approximations (LDA/GGA) systematically underestimate the band gaps of typical ferroelectrics by a factor of ~2, hampering therefore further correct description of the barrier with a metal. GW corrections can eventually be considered, but are computationally heavy and do not allow for self-consistent atomic relaxation. To bypass these problems, we consider here the recently proposed B1-WC hybrid functional.[1] This hybrid is able to describe both structural and electronic (band gaps) properties of prototypical ferroelectrics and multiferroics.[2] A comparison between the electronic band structure of BaTiO3, within B1-WC and GW approximation (using ABINIT code) will be shown. B1-WC allows us to provide improved first-principle description of Schottky barriers at the interfaces between different ferroelectrics (BaTiO3, PbTiO3) and metal electrodes (Al, Ag, Au, Cu, Pt, SrRuO3). The differences and main trends will be discussed. These successful results question the relevance and interest of implementing hybrid functional approaches (B1-WC, HSE) in future versions of ABINIT. Work supported by FAME-EMMI and QCN IAP project.

  1. D. I. Bilc, R. Orlando, R. Shaltaf, G. M. Rignanese, J. Iniguez, and Ph. Ghosez, Phys. Rev. B 77, 165107 (2008)
  2. M. Goffinet, P. Hermet, D. I. Bilc, and Ph. Ghosez, Phys. Rev. B 79, 014403 (2009)

DFT Calculations on the Thermoelectric Zn1-xCdxSb and Mg2Si Phases Using Abinit

BOULET Pascal - The ZnSb and Mg2 Si compounds are well-known for their thermoelectric properties. The figure of merit which characterizes the efficiency of the thermoelectric materials is enhanced in ZnSb by substituting Zn for Cd. It was evidenced in the literature that the solid solution Zn1-xCdxSb exists from ZnSb to CdSb. However, some doubts remain concerning the existence of an ordered phase around 50 at.% of cadmium. We therefore performed DFT calculations on the Zn1-xCdxSb solid solution using the PBE functional in order to bring additional information. This theoretical investigation, which was undertaken on this system for the first time, should allow us to know the relative stabilities of the ordered and disordered phases with respect to the composition.

Electronic and structural characterization of the Fe3C compound under Pressure. An ab initio study.

CALDERON Alejandra Vargas -

Minerals under extreme conditions

CARACAS Razvan - We employ state-of-the-art first-principles calculations based on density functional theory and density functional perturbation theory to investigate the behavior of major planetary materials under extreme conditions of pressure. We build phase diagrams and determine relevant physical properties of several selected minerals and ices.

Theoretical mineral spectroscopy

CARACAS Razvan - We use the latest theoretical developments of the density functional perturbation theory and their practical implementations in the ABINIT package to determine the elastic constants and the Raman spectra of several of the major constituents of the Earth's mantle. Finally we present a preliminary sketch of database of computed Raman spectra for minerals - the WURM project.

First-principles calculations of molecular crystals under pressure

CARACAS Razvan - We employ state-of-the-art first-principles calculations based on density functional theory and density functional perturbation theory to investigate phase diagrams and relevant physical properties of the main components of molecular solids under pressure. We present the basics of the theoretical approach and then focus on nitrogen and on water ice.

Spin and structural transitions in AlFeO3 and FeAlO3 perovskite and post- perovskite

CARACAS Razvan - We use an advanced formalism within the density functional theory to investigate the perovskite and post- perovskite structures with AlFeO3 chemistry. We consider two ordered cases: one with Fe and then one with Al in octahedral coordination. For each case we investigate several spin configurations. We observe that up to 90 GPa the FeAlO3 perovskite structure, i.e. with Fe3+ in the inter-octahedral space, with antiferromagnetic configuration and large local magnetic moment is the most stable one. Beyond 90 GPa the post-perovskite structure of AlFeO3, i.e. with Fe3+ in octahedral coordination, with antiferromagnetic configuration and small local magnetic moment is the most stable one. The perovskite to post-perovskite phase transition at 90GPa triggers a partial collapse of the magnetic moment. The local magnetic moment vanishes beyond 150 GPa in post-perovskite.

SrTiO3 and BaTiO3 surface reconstructions from first principle calculations

ILES Nadia - Using a Norm Conserving Pseudo-Potential method NCPP and the Density Functional Theory DFT implemented in ABINIT code, we present here a comparative study of the thermodynamic stability of (001) reconstructed surfaces of BaTiO3 and SrTiO3 perovskites. For this purpose we compute surface energy and atomic relaxations for the (1x1), (2x1) and (2x2) model reconstructions, with AO (A ? Sr, Ba) and TiO2 as possible terminations. Thereafter, we use a thermodynamic approach based on the calculation of the surface free energy Gs, in order to compute the stability of the previous reconstructions with respect to the chemical potentials of the components A and O. The recently proposed TiO2 double layer (2x2) reconstruction with characteristic TiO2 zigzag chains is analyzed and estimated to be the most stable one for very Ti-rich conditions. We discuss the role of the atomic coordination number and surface stress on the thermodynamic stability of the various models.

Structural characterization of methanol monohydrate under pressure: a first principles study

MARTINEZ Davila - Methanol monohydrate is one of several crystalline phases present in the methanol-water system. At ambient conditions of pressure, methanol monohydrate has a disordered structure, which is also polar. Here we study its behavior under pressure using the density-functional theory in the ABINIT implementation. We propose several initial ideally ordered configurations that we optimize at different pressures. We analyze the energy and compressibility of the different structures under pressure. A new monoclinic structure is proposed with space group Cc (#9), which is competitive (~meV difference) with a similar structure with space group P21 (#4); the two configuration can go from one to the other easily by hydrogen reordering. Hydrogen bonds play a major role in the structures. X-ray diffraction patterns based on the theoretical structures have been generated and compared with experimental neutron diffractions. Next, Raman spectra will be calculated for the optimized structures within abinit to compare and interpret the experimental data.

EXCITON: A Gaussian orbital code for solid state spectroscopy

PATTERSON Charles - EXCITON is a new Gaussian orbital code for periodic systems which is intended primarily for simulation of electron and optical spectroscopies of crystalline solids. It is interfaced to the Crystal code and an SCF module for scHF, hybrid density functionals and scGW is under development. A module for treating excitations via Bethe-Salpeter integral equations is also under development. EXCITON makes full use of crystal space group symmetries in order to minimise computer time and generate irreducible representations of various states. Some of these features of the code are described here. We present results of Bethe-Salpeter equation calculations on solid rare gases, single-particle optical excitations at silicon surfaces reconstructed by Au adsorption, visualisation of wavefunctions and Fermi surfaces in superconductors, all performed using EXCITON.

PAW+U study of the La2-xSrxCuO4 phonon spectra

PESANT Simon - The electronic and vibrational properties of the superconductor La2-xSrxCuO4 were studied using the projected augmented waves (PAW) method and density functional theory. We simulated the strontium hole doping by removing a finite charge in the density of La2CuO4. For the optimal hole doping (1/8), the GGA functional gives a metallic ground state. Adding a Hubbard term to the functional (GGA+U), the ground state becomes a Stripe ordered phase. The effect of stripe ordering on the half breathing modes of the optimally doped La2CuO4 was studied and compared to the GGA phonon frequencies for the same doping.

Band Offsets Prediction From Many Body Perturbation Theory

SHALTAF Riad - We use many-body perturbation theory, the state-of-the-art method for band gap calculations, to compute the band offsets at the Si/SiO2 interface. We examine the adequacy of the usual approximations in this context. We show that (i) the separate treatment of band-structure and potential lineup contributions, the latter being evaluated within density-functional theory, is justified, (ii) most plasmon-pole models lead to inaccuracies in the absolute quasiparticle corrections, (iii) vertex corrections can be neglected, (iv) eigenenergy self-consistency is adequate. Our theoretical offsets agree with the experimental ones within 0.3 eV. On the other hand we show that that the success of DFT- LDA approximation in predicting accurately the band offsets for some systems such as ZrO2/Si, can be traced back to of cancellation errors.

Lattice dynamics and specific heat of alpha - GeTe: a theoretical and experimental study

SHALTAF Riad - We have investigated the dynamical and specific heat of GeTe in its low tem- perature rhombohedral phase using first-principles density functional theory. We report the phonon dispersion curves, the density of phonon states and tempera- ture dependent specific heat. In contrast to results for other IV-VI materials, we found that the inclusion of spin orbit coupling has a small effect on the calculated phonon frequencies and consequently on the heat capacity. On the other hand, we found that ignoring non-stoichiometric effects of GeTe, leads to discrepancies between the calculated and the measured Cp /T 3. These can be resolved by taking into account the existence of free hole carriers which lead to an increase of Cp /T 3 in agreement with experimental findings. Such increase was found to be due to phonon frequency softening and the corresponding down shift of the density of phonon states.