Source-linked AI summary
AFLOW: An automatic framework for high-throughput materials discovery
Stefano Curtarolo, Wahyu Setyawan, Gus L. W. Hart, Michal Jahnatek, Roman V. Chepulskii, Richard H. Taylor, Shidong Wang, Junkai Xue, Kesong Yang, Ohad Levy, Michael J. Mehl, Harold T. Stokes, Denis O. Demchenko, Dane Morgan
TL;DR
Aflow addresses the need for systematic high-throughput generation and classification of computational materials data. It provides automated calculations and analysis tools for alloy and compound structures, with software and selected capabilities available for download and online use.
Problem
High-throughput methods are needed to systematically generate and classify computational data for discovering compounds, metastable structures, and materials properties.
Method
Aflow is a software framework that automates calculations of ground-state energies and a broad suite of material properties across specified classes or large databases of structures.
Results
Aflow calculates and analyzes properties including structural stability, phase diagrams, electronic structure, phonons, surfaces, superconductivity, and related material characteristics.
Takeaways & Limitations
The software, manuals, and selected structure-manipulation and analysis capabilities are freely available for download or online operation.
Abstract
from arXiv · showhide
Recent advances in computational materials science present novel opportunities for structure discovery and optimization, including uncovering of unsuspected compounds and metastable structures, electronic structure, surface, and nano-particle properties. The practical realization of these opportunities requires systematic generation and classification of the relevant computational data by high-throughput methods. In this paper we present Aflow (Automatic Flow), a software framework for high-throughput calculation of crystal structure properties of alloys, intermetallics and inorganic compounds. The Aflow software is available for the scientific community on the website of the materials research consortium, aflowlib.org. Its geometric and electronic structure analysis and manipulation tools are additionally available for online operation at the same website. The combination of automatic methods and user online interfaces provide a powerful tool for efficient quantum computational materials discovery and characterization.
I. INTRODUCTION
AFLOW applies high-throughput computation to systematic materials discovery by automating calculations across broad structure libraries and organizing diverse resulting properties. The framework combines database-driven workflows, parallel software, and tools for analyzing stable and metastable materials.
- I. INTRODUCTION: High-throughput methods screen libraries of structures and compositions by calculating many feasible quantities before extracting properties and correlations.This differs from targeting one physical quantity across a large number of structures.
- I. INTRODUCTION: AFLOW is freely downloadable, provides manuals, and exposes selected structure-analysis capabilities through online interfaces.The framework is available through aflowlib.org for both offline and interactive use.
- I. INTRODUCTION: AFLOW automates input setup, calculation execution, result collation, and plotting for large structure databases with little human intervention.The framework is designed to generate and represent data consistently for simultaneous optimization of multiple properties.
- I. INTRODUCTION: The software backbone identifies stable and metastable structures while supporting phase diagrams, electronic bands, phonons, surfaces, superconductivity, local environments, and cluster expansion.New applications are integrated continuously as the code develops.
- I. INTRODUCTION: The database combines approximately 400 experimental prototypes with millions of formally enumerated bcc-, fcc-, and hcp-derived superstructures.Only a few hundred enumerated structures are typically calculated per alloy system, while the broader libraries support cluster-expansion studies.
III. BASIC OPERATION
AFLOW standardizes the workflow from database structure selection through relaxation, static calculations, and electronic-structure analysis. It also supports automated execution beyond DFT and provides recovery and optimization mechanisms for high-throughput runs.
- III. BASIC OPERATION: Aflow generates relaxation, static, and band-structure inputs from a database structure label after adjusting lattice parameters for the selected elements.The starting structural description contains prototype lattice parameters and atomic positions.
- III. BASIC OPERATION: AFLOW standardizes lattice bases and automatically determines Brillouin-zone integration paths across 14 Bravais lattices and 24 Brillouin zones.Standardization enables consistent comparison of electronic-structure data between projects.
- III. BASIC OPERATION: The default workflow uses VASP with PAW pseudopotentials and GGA exchange-correlation, fully relaxing each structure twice to a 1 meV/atom convergence tolerance.Dense reciprocal-space sampling is used during relaxation and later static calculations.
- III. BASIC OPERATION: AFLOW automatically generates electronic band structures and density-of-states outputs for database structures such as Ag2BaSn2.Figure 1 shows the band structure along an automatically selected Brillouin-zone path alongside the density of states.
- III. BASIC OPERATION: AFLOW’s alien mode executes non-DFT tasks in high-throughput fashion and supports pre- and post-processing scripts for adaptive input generation.These controls also improve crash recovery and handling of unconverged runs.
IV. STRUCTURE ANALYSIS TOOLS
AFLOW’s Aconvasp tools let users analyze and transform database or user-supplied structures through standardized cell representations, symmetry operations, coordinate conversions, and reciprocal-space outputs.
- IV. STRUCTURE ANALYSIS TOOLS: The online and command-line tools accept structures from the database or user-provided input files for analysis and conversion.The interface supports operations on stored structures and structures entered through the POSCAR input box.
- IV. STRUCTURE ANALYSIS TOOLS: Cell transformations include normal and standard primitive, standard conventional, Minkowski-reduced, and Niggli-standardized representations.The standard primitive cell is selected so its reciprocal Wigner-Seitz cell matches one of 24 possible Brillouin zones.
- IV. STRUCTURE ANALYSIS TOOLS: Aconvasp identifies equivalent atoms and site symmetries, remaps atoms into the cell, converts coordinate systems, and locates interstitial cages.These operations provide both structural descriptors and geometric locations for further analysis.
- IV. STRUCTURE ANALYSIS TOOLS: The tools convert between crystallographic file formats and generate lattice parameters, cell geometry, reciprocal-lattice, and symmetry information.Supported conversions include WYCKOFF-CAR/ABCCAR to POSCAR and POSCAR to ABCCAR.
- IV. STRUCTURE ANALYSIS TOOLS: The reciprocal-space tool supplies the k-space directions and Brillouin-zone path needed for band-structure calculations.The interface also provides access to manuals and the current experimental-prototype list.
A. Vibration spectra and free energy
AFLOW automates phonon calculations through three complementary approaches and uses the resulting dispersions to derive vibrational thermodynamic quantities. The methods exploit symmetry and parallel execution to reduce user effort and support different spectral-scanning needs.
- A. Vibration spectra and free energy: AFLOW supports direct force-constant, PAW linear-response, and frozen-phonon approaches for calculating phonon dispersion curves.The three methods are implemented for convenience and cover different computational use cases.
- A. Vibration spectra and free energy: The direct force-constant method automatically prescribes displacements, calculates forces with VASP, and maps them onto symmetry-equivalent directions.Only unique atoms are distorted along the minimum independent directions needed for the force-constant matrix.
- A. Vibration spectra and free energy: AFLOW uses phonon dispersions to calculate vibrational free energy, entropy, and specific heat.The resulting quantities are illustrated in the rhodium example in Figure 3.
- A. Vibration spectra and free energy: Linear response handles the non-analytical dynamical-matrix contribution needed to reproduce longitudinal-transverse optical phonon splitting.AFLOW sets up, runs, and converges the corresponding VASP calculation automatically.
- A. Vibration spectra and free energy: The approaches differ in scope: direct force constants provide information throughout the Brillouin zone, whereas linear response and frozen phonons can rapidly scan selected k points.The latter options are useful when searching for features such as Kohn anomalies.
B. Design of high-index surfaces in complex multicomponent compounds
Aflow constructs surfaces from complex bulk crystal structures specified by database labels and Miller indices. The generated surface file includes complete structural coordinates for the constructed supercell.
- Surface construction is motivated by the need for exact surface and subsurface atom positions in technologically relevant compounds.
- The input identifies an arbitrary complex bulk lattice and can include any number of atom types.
- The generated surface file contains direct and Cartesian coordinates for the constructed surface supercell.
- Miller indices are defined using the Bravais lattice corresponding to the bulk unit cell.
C. Nanoparticle generation
Aconvasp generates nanoparticle structure files from an input crystal structure by specifying particle radius and neighbor separation. This automates manual construction and supports high-throughput nanoparticle studies.
- Aconvasp generates a nanoparticle with a specified radius and neighbor separation while preserving the input crystal lattice.
- The particle origin can be placed at the unit-cell origin, a crystal atom, or arbitrary Cartesian or fractional coordinates.
- The radius and separation distance are specified in Å.
- The automated option replaces cumbersome manual nanoparticle-file generation and enables high-throughput investigation of nanoparticle sets.
- Figure 5 illustrates L13-based nanoparticles with radii of 5 Å, 8 Å, and 12 Å.
D. Topological identification of interstitial sites
Aflow identifies interstitial sites from crystallographic structure data and characterizes their locations, radii, coordination, and energetics. In the L13 example, symmetry reduces the search to five irreducible sites of octahedral and tetrahedral coordination.
- The interstitial-site algorithm returns each site’s location, radius, and coordination from a unit-cell atomic-position file.
- Aflow can simulate occupation of irreducible cages and calculate energies, entropy, and dilute-limit solubility.
- The multi-threaded implementation accelerates interstitial calculations in multi-core environments.
- The AgZr3-L13 example contains two octahedral sites with r = 2.2225 Å and three tetrahedral sites with r = 1.9248 Å.
- Figure 6 distinguishes 6-coordinated sites in blue from 4-coordinated sites in pink.
E. Apennsy: automatic analysis
Apennsy processes Aflow outputs into system-specific data files and analyzes binary alloy stability through convex-hull construction. It identifies minimum-energy structures, stable extremal phases, and produces plots or exportable formats for downstream tools.
- Apennsy compiles computed structures, energies, relaxed and input prototypes, space groups, and the binary convex hull into system-specific data files.
- For binary alloys, Apennsy constructs the zero-temperature phase diagram from the convex hull.
- It finds the minimum-energy structure at each composition and calculates formation enthalpies.
- The formation-enthalpy equation uses pure-element energies and the alloy energy at concentration x.
- Structures below tie-lines are identified as stable structures forming the binary convex hull, which Apennsy plots automatically.
- Apennsy exports selected structural data for cluster-expansion, miscibility, and other downstream analyses, either offline or through aflowlib.org.
VI. FUTURE DEVELOPMENTS: HIGH-THROUGHPUT HYBRID FUNCTIONALS CALCULATIONS
Aflow is extending its high-throughput framework to hybrid-functional calculations, addressing cases where LDA/GGA-based methods are inaccurate but many-body methods are costly. Preliminary HSE06 results improve semiconductor band gaps, although the calculations require substantially greater computational resources and the framework extension remains incomplete.
- Motivation: Hybrid functionals offer a pragmatic compromise between inaccurate local or semi-local DFT approximations and expensive many-body methods.The paper identifies defect energetics, optical properties, correlated materials, reaction energetics, and adsorption as problematic areas for LDA/GGA(+U).
- Hybrid functional approach: HSE06 mixes 25% short-range exact exchange with 75% semi-local GGA exchange and screens the long-range exchange interaction.The screening length is reported as approximately 7–10 Å.
- Computational cost: Hybrid-functional calculations are substantially more expensive than LDA/GGA calculations because they require non-local Fock exchange integrals.For bulk ZnO, the reported HSE06 setup takes approximately 5 hours on 8 CPUs versus less than a minute for GGA/PBE(+U), with fivefold higher memory requirements.
- Aflow implementation: Aflow currently employs Dudarev GGA+U with user-selectable parameters, while a fully functional hybrid-functional framework is still being developed.The paper states that a consistent and robust Aflow framework with hybrid functionals was planned for 2012.
- Preliminary results: 2.48 eV is the HSE06 ZnO band gap, compared with 0.7 eV for GGA and 1.82 eV for PBE+U.The calculation used the standard exact-exchange ratio α = 0.25; using α = 0.375 was reported to bring the gap into agreement with experiment.
VII. SUMMARY
Aflow is a software package for high-throughput calculations of material properties, with tools for structure manipulation and analysis also available online.
- Aflow provides a software package for high-throughput calculations of material properties.