FHI-aims: Full-Potential, All-Electron Electronic Structure Theory with Numeric Atom-Centered Basis Functions

Methods and code development · Fritz-Haber-Institut der Max-Planck-Gesellschaft · Theory Department

Stable Features at a glance:

FHI-aims is an all-electron electronic structure code based on numeric atom-centered orbitals.

It enables first-principles simulations with very high numerical accuracy for production calculations, with excellent scalability up to very large system sizes (thousands of atoms) and up to very large, massively parallel supercomputers (ten thousands of CPU cores).

Broad accuracy on par with the best available benchmarks is demonstrated, for example, in

Science 351 (2016), DOI: 10.1126/science.aad3000,

J. Phys. Chem. Lett. 8, 1449-1457 (2017)
and

Phys. Rev. Materials 1, 033803 (2017).

Scalability to very large systems is demonstrated, for example, in

J. Phys.: Condens. Matter 26, 213201 (2014),

Phys. Rev. Lett. 111, 065502 (2013) and

Comp. Phys. Commun. 192, 60-69 (2015).

FHI-aims is developed by an active, globally distributed community at FHI, Duke University, TU Munich, USTC Hefei, Aalto University, University of Luxembourg, TU Graz, Cardiff University and many others.

We here provide a list of some of the functionality that is now available. If a specific item is missing that should be here, or if you would like to know about a specific (future) item, feel free to ask us directly.

- Periodic geometries on equal footing with non-periodic
systems for LDA, GGA, and hybrid functionals, including:
- dipole corrections for surface slab calculations
- well-converged large vacuum regions for surface slabs with practically no overhead

- Non-relativistic and scalar relativistic total energies (scalar relativistic energy differences match full-potential LAPW benchmark accuracy)
- Extensively validated second-variational spin-orbit coupling (see Phys. Rev. Materials 1, 033803 (2017)) for band structures, densities of states, "fat band" Mulliken decompoosition of bands and states, macroscopic dielectric tensor and absorption coefficients
- Total energy gradients (forces) [LDA, GGA, and hybrid functionals]
- Efficient structure optimization (BFGS or Trust Radius optimization, tunable initial Hessian matrix including scheme by Lindh and coworkers)
- Stress tensor based unit cell optimization for periodic systems (including hybrid functionals)
- Charged excitations (
*GW*, currently for non-periodic systems) - LR-TDLDA and GW-Bethe-Salpeter Equation, currently for non-periodic systems
- Born-Oppenheimer molecular dynamics:
*NVE*ensemble: Optional wave function extrapolation or higher-order integrator beyond Verlet*NVT*ensemble: Andersen, Nose-Hoover, Bussi-Donadio-Parrinello thermostats

- Infrastructure for vibrations and phonons (from finite
differences)
- Harmonic "computational spectroscopy" for infrared intensities
- Anharmonic infrared intensities from Born-Oppenheimer molecular dynamics
- Phonon dispersion relations for periodic systems
- Much more phonon-related functionality by interface to (external) phonopy package

- Density-functional perturbation theory
- NMR chemical shieldings and J-couplings for light-element molecules
- Free-energy methods:
- Harmonic free energy from vibrations / phonons
- Thermodynamic integration for explicit anharmonic free energies
- Interface to (external) Plumed plugin

- Molecular transport infrastructure "aitranss" (by Alexej Bagrets and Ferdinand Evers group, Regensburg)
- Kubo-Greenwood electronic thermal transport
- Output: electron densities, Kohn-Sham orbitals, band structures, densities of states, ...

(ready for visualization with standard visualization tools) - QM/MM embedding infrastructure with norm-conserving pseudopotentials (see J. Chem. Phys. 141, 024105 (2014))
- Connection to i-Pi code for advanced molecular dynamics and path integral molecular dynamics
- Implicit solvation formalisms for molecules (see J. Chem. Theory. Comput, 12, 4052-4066 (2016) and J. Chem. Theor. Comput. 13, 5582-5603 (2017) for two implemented schemes)

- Preconstructed, hierarchical numeric atom-centered basis sets for elements 1-102, for systematic convergence from fast qualitative to meV-level total energy convergence
- Preconstructed default settings for the most important numerical choices (grids, Hartree potential, basis cutoff etc.): "light", "tight", "really_tight"
- Numeric atom-centered valence correlation consistent basis sets for systematic convergence of many-body perturbation methods (currently, for H-Ar)
- Empty sites for counterpoise-corrected binding energies (especially for many-body methods MP2, RPA)

- Local-density and generalized gradient approximations for
molecules and solids, including gradients (forces):
- LDA functionals: Perdew/Wang (1992), Perdew/Zunger (1981), Vosko/Wilk/Nusair (1980)
- GGA (like) functionals: AM05, BLYP, PBE, PBEsol, RPBE, revPBE, PBEint

- Hartree-Fock and hybrid functionals (non-periodic and periodic, including gradients): Hartree-Fock, PBEh (incl. PBE0), HSE03, HSE06, B3LYP, M06 and M11. Non-periodic: screened exchange, COHSEX
- Periodic version of exchange-correlation "beyond" LDA/GGA (Hartree-Fock and hybrid functionals)
- Tkatchenko-Scheffler interatomic dispersion (van der Waals) correction for PBE, BLYP, PBE0, B3LYP
- Many-body dispersion correction for PBE, BLYP, PBE0, B3LYP
- Locally constrained DFT (charge and spin) for the above functionals (non-periodic systems)
- meta-GGA: M06-L, TPSS, revTPSS, TPSSloc, SCAN
- "XYG3" doubly-hybrid functional (currently, post-scf only)
- Self-consistent version of Langreth-Lundvist
non-local vdW-DF (Ville Havu
*et al.*, Helsinki) - Non-local DFT (currently, post-SCF total energy correction by Monte Carlo integration scheme): Langreth-Lundqvist vdW-DF
- Many-body perturbation theory (post-SCF correction for
single-point non-periodic geometries):

Second-order Moller-Plesset theory (MP2), Random Phase Approximation (RPA) and renormalized second-order perturbation theory

*G*approximation for single-carrier like excited energy levels_{0}W_{0} - Self-consistent
*GW*

- DFT-LDA/GGA: All-electron accuracy at a computational cost comparable to plane-wave/pseudopotential implementations
- System size range up to thousand(s) of atoms, with O(N)
like scaling for the most expensive operations [limiting
factor: Conventional O(N
^{3}) eigensolver beyond this range] - Seamlessly parallel (time and memory) from desktop up to currently (ten)thousands of CPUs
- Specifically optimized, massively parallel conventional eigensolver ELPA to minimize scalability barriers independent of type of system treated (see J. Phys.: Condens. Matter 26, 213201 (2014))
- Per-process memory saving options, efficient run-time communication for large, low-memory massively parallel platforms
- Partial support for GPU acceleration (CUDA)
- Accurate, fast and more memory-efficient "localized" resolution of identity method for exchange-correlation beyond LDA/GGA (currently functional as base method for Hartree-Fock and hybrid functionals, for more than 1,000 heavy atoms on large supercomputers) (see New J. Phys. 17, 093020 (2015) and Comp. Phys. Commun. 192, 60-69 (2015))