🔬 materials science

Abinit 2025: New Capabilities for the Predictive Modeling of Solids and Nanomaterials

This paper presents the significant scientific and technical advancements in the Abinit software package over the past five years, highlighting new capabilities in ground-state and excited-state methodologies, GPU-accelerated high-performance computing, second-principles modeling, and automated workflows designed to support high-throughput predictive modeling of solids and nanomaterials.

Original authors: Matthieu J. Verstraete, Joao Abreu, Guillaume E. Allemand, Bernard Amadon, Gabriel Antonius, Maryam Azizi, Lucas Baguet, Clementine Barat, Louis Bastogne, Romuald Bejaud, Jean-Michel Beuken, Jordan Bi

Published 2026-01-27

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Matthieu J. Verstraete, Joao Abreu, Guillaume E. Allemand, Bernard Amadon, Gabriel Antonius, Maryam Azizi, Lucas Baguet, Clementine Barat, Louis Bastogne, Romuald Bejaud, Jean-Michel Beuken, Jordan Bieder, Augustin Blanchet, Francois Bottin, Johann Bouchet, Julien Bouquiaux, Eric Bousquet, James Boust, Fabien Brieuc, Veronique Brousseau-Couture, Nils Brouwer Fabien Bruneval, Alois Castellano, Emmanuel Castiel, Jean-Baptiste Charraud, Jean Clerouin, Michel Cote, Clement Duval, Alejandro Gallo, Frederic Gendron, Gregory Geneste, Philippe Ghosez, Matteo Giantomassi, Olivier Gingras, Fernando Gomez-Ortiz, Xavier Gonze, Felix Antoine Goudreault, Andreas Gruneis, Raveena Gupta, Bogdan Guster, Donald R. Hamann, Xu He, Olle Hellman, Natalie Holzwarth, Francois Jollet, Pierre Kestener, Ioanna-Maria Lygatsika, Olivier Nadeau, Lorien MacEnulty, Enrico Marazzi, Maxime Mignolet, David D. O'Regan, Robinson Outerovitch, Charles Paillard, Guido Petretto, Samuel Ponce, Francesco Ricci, Gian-Marco Rignanese, Mauricio Rodriguez-Mayorga, Aldo H. Romero, Samare Rostami, Miquel Royo, Marc Sarraute, Alireza Sasani, Francois Soubiran, Massimiliano Stengel, Christian Tantardini, Marc Torrent, Victor Trinquet, Vasilii Vasilchencko, David Waroquiers, Asier Zabalo, Austin Zadoks, Huazhang Zhang, Josef Zwanziger

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine Abinit as a super-powered, digital microscope that allows scientists to see and predict how atoms and electrons behave inside materials, without ever needing to mix chemicals in a lab. For the past 25 years, this software has been a cornerstone of materials science. This new paper, dated June 2025, is like a "Version 2025" update log, showcasing a massive overhaul that makes the microscope sharper, faster, and capable of seeing things it previously couldn't.

Here is a breakdown of the new capabilities, explained through everyday analogies:

1. The "Ground State" Upgrade: Controlling the Chaos

In physics, the "ground state" is the calm, resting position of a material.

Constrained DFT (The "Traffic Cop"): Previously, if scientists wanted to force a specific atom to hold a specific amount of electric charge or magnetism, the software would guess and often get it slightly wrong. The new Constrained DFT acts like a strict traffic cop. It can now force an atom to hold exactly the right amount of charge or spin, allowing researchers to study "what-if" scenarios, like what happens if you artificially add extra electrons to a material (photodoping) or study specific magnetic states that are hard to reach naturally.
High-Temperature DFT (The "Heat Wave"): Standard simulations usually break down when things get too hot (like inside a star or a nuclear explosion). The new Extended DFT and Thermal Functionals are like adding a heat shield to the microscope. They allow the software to simulate materials in "warm dense matter" states—scorching hot, high-pressure environments where electrons behave like a chaotic gas—without the simulation crashing.
Meta-GGA (The "High-Definition Lens"): The software now uses a more sophisticated mathematical lens called meta-GGA. Think of standard lenses as seeing a blurry picture; this new lens sees the "texture" of the electron clouds (kinetic energy density) in much higher definition, leading to more accurate predictions of how materials hold together.

2. Seeing the Invisible: Responses and Vibrations

Materials don't just sit still; they vibrate and react to fields.

Flexoelectricity (The "Bending Effect"): Imagine bending a ruler. Usually, we think about stretching it. But if you bend it unevenly (creating a gradient), it can generate electricity. The new software can now calculate this flexoelectricity, which is crucial for understanding how tiny, flexible electronics might work.
Phonon Angular Momentum (The "Spinning Dance"): Atoms in a crystal vibrate like dancers. In some crystals, these dancers don't just move up and down; they spin. The software can now calculate this phonon angular momentum, helping scientists understand how light and magnetism interact with these spinning vibrations, especially in "chiral" (handed) crystals.
Polars (The "Self-Trapped Dancers"): Sometimes, an electron gets so excited it drags the atoms around it into a little cloud, trapping itself. This is called a polaron. The new tools can now simulate both weak and strong versions of this "self-trapping," helping scientists understand how electricity moves through materials that are a bit "sticky."

3. The "Excited State" Suite: Looking Beyond the Calm

Most simulations look at materials at rest. But what happens when you hit them with a laser or a strong electric field?

Real-Time TDDFT (The "Slow-Motion Camera"): Instead of just predicting the final result of a laser pulse, the new Real-Time TDDFT acts like a slow-motion camera. It simulates the electrons moving second-by-second as they react to intense light, allowing scientists to see the dynamic dance of electrons in real-time.
GW and DMFT (The "Expert Consultants"): For materials where electrons are very crowded and interact strongly (like in superconductors), standard rules fail. The software now has better "consultants" (GW approximation and Dynamical Mean Field Theory) that can handle these complex social interactions between electrons, giving a much more accurate picture of the material's true energy levels.
Coupled Cluster (The "Precision Partner"): The software can now talk directly to another specialized program called Cc4s. Think of this as Abinit handing off a difficult math problem to a specialist who solves it with extreme precision, then handing the answer back. This allows for ultra-accurate calculations of solid materials.

4. Speed and Power: The GPU Revolution

The most dramatic change is how fast the software runs.

GPU Acceleration (The "Formula 1 Engine"): For years, these calculations ran on standard computer processors (CPUs). The new version has been completely rewritten to run on Graphics Processing Units (GPUs)—the same chips used for high-end video games.
- The Analogy: If the old CPU version was a single cyclist, the new GPU version is a peloton of 100 cyclists riding in perfect formation.
- The Result: Simulations that used to take days or weeks can now be done in hours or minutes. The paper notes speed-ups of 10 to 40 times faster, allowing scientists to simulate materials with thousands of atoms on a single computer node.

5. Automation and Workflow: The "Factory Line"

Calculating one material is hard; calculating thousands is impossible without help.

High-Throughput Workflows (The "Assembly Line"): The paper introduces new tools (AbiPy, Atomate2, AiiDA) that act like an automated factory line. You can feed a list of 1,000 different materials into the system, and it will automatically:
1. Set up the experiment.
2. Run the simulation.
3. Check for errors.
4. Organize the results.
- This allows researchers to screen massive databases of materials to find the perfect candidate for a battery or solar cell without human intervention.
Machine Learning Sampling (The "Smart Scout"): A new tool called MLACS uses machine learning to act as a "smart scout." Instead of simulating every single moment of a material's movement (which is slow), it learns the patterns and predicts the most important moments, drastically speeding up the study of how materials behave at high temperatures.

Summary

In short, Abinit 2025 is a massive upgrade. It is now hotter (can simulate extreme temperatures), sharper (more accurate math), faster (runs on video game chips), and smarter (can automate the discovery of new materials). It transforms the software from a powerful calculator into a comprehensive, automated discovery engine for the next generation of materials.

Technical Summary: Abinit 2025: New Capabilities for the Predictive Modeling of Solids and Nanomaterials

Problem Statement
The field of computational materials science faces increasing demands to simulate complex systems under extreme conditions (high temperature, strong correlations, defects) and to bridge the gap between first-principles accuracy and mesoscopic scales. While Density Functional Theory (DFT) is a standard tool, traditional implementations often struggle with high-temperature regimes (the "orbital wall"), strongly correlated electron systems, and the computational cost of high-throughput screening. Furthermore, the integration of machine learning, high-performance computing (HPC) on GPU architectures, and automated workflows requires robust, flexible, and interoperable software frameworks. This paper addresses the need to update the Abinit software suite to meet these evolving challenges, enhancing its predictive capabilities for ground states, excited states, and response properties while optimizing performance on modern hardware.

Methodology and Technical Implementation
The paper details a comprehensive suite of methodological and technical advancements implemented in Abinit over the past five years.

Ground State and Constrained DFT:
- Constrained DFT (cDFT): A new Potential-based Lagrange Multiplier (PLM-cDFT) algorithm was implemented to impose precise constraints on atomic charges and magnetization vectors. This allows for the study of excited states, charge transfers, and localized excitations with arbitrary numerical precision, resolving discrepancies in magnetic materials like $YBaCo_4O_7$ .
- Photodoping: A scheme to mimic photo-induced structural deformations via constrained occupation numbers (quasi-Fermi energies) was introduced, enabling structural relaxation in excited states.
- High-Temperature DFT: To address the "orbital wall" in warm dense matter, an extended plane-wave model was implemented, treating high-energy states as free electrons. Additionally, thermal exchange-correlation functionals (e.g., corr-KSDT, KDT16) were integrated to account for electronic temperature effects.
- Meta-GGA Functionals: Full support for meta-GGA functionals (including SCAN, rSCAN, r2SCAN) was extended to both Projector Augmented Wave (PAW) and norm-conserving pseudopotential frameworks, utilizing kinetic energy density and Laplacian terms.
- Orbital Magnetism: Implementations for calculating orbital magnetism and magnetic shielding (NMR) were added, including complete treatments for diamagnetic insulators and atom-centered approaches for antiferromagnets, with support for spin-orbit coupling via ZORA.
Response Properties and Perturbations:
- Spatial Dispersion: New capabilities allow for the calculation of natural optical activity (NOA) and flexoelectricity (both clamped-ion and lattice-mediated) using norm-conserving pseudopotentials.
- Phonon Angular Momentum: The Anaddb utility was updated to compute phonon angular momentum in systems lacking inversion symmetry.
- Electron-Phonon Interaction: The framework was expanded to include transport properties via the Boltzmann Transport Equation (BTE) using both Relaxation Time Approximation (RTA) and Iterative BTE (IBTE), as well as the Kubo-Greenwood formalism for systems with strong coupling. New methods for calculating polaron binding energies (weak and strong coupling regimes) and band-gap renormalization were introduced.
- Databases: The derivative database (DDB) format was updated to NetCDF, including metadata and support for higher-order derivatives (e.g., Berry curvature).
Excited States and Many-Body Perturbation Theory:
- Real-Time TDDFT: Implementation of real-time time-dependent DFT (RT-TDDFT) allows for the simulation of electron dynamics under strong perturbations (e.g., laser pulses) and non-adiabatic molecular dynamics.
- GW and DMFT: The GW approximation now includes the computation of the one-body reduced density matrix (using linearized Dyson equations) and low-scaling GW (GWR) in real space/imaginary time. Dynamical Mean-Field Theory (DMFT) implementations were optimized, including faster Green's function calculations and integration with CT-QMC solvers (TRIQS).
- Coupled Cluster: An interface to the Cc4s code was established to export Coulomb integrals for high-accuracy post-Hartree-Fock calculations (CCSD, CCSD(T)) in periodic systems.
- Effective Interactions: Automated calculation of Hubbard $U$ and Hund's $J$ parameters via linear response and constrained random phase approximation (cRPA) was refactored.
High-Performance Computing (HPC) and Optimization:
- GPU Acceleration: A complete rewrite of the GPU code using OpenMP offload directives enables ground-state and DFPT calculations on NVIDIA and AMD accelerators. This includes optimized iterative diagonalization (Chebyshev filtering, LOBPCG) and non-local operations.
- Parallelization: New band-parallelized DFPT algorithms reduce memory footprints for large systems. The RMM-DIIS eigen-solver was integrated to accelerate SCF convergence.
- Build Systems: The build system now supports CMake alongside Autotools, and pre-compiled binaries are available via package managers (Spack, EasyBuild, etc.).
Multiscale and Workflows:
- Second-Principles: The Multibinit package (MB-EAP, MB-LWF, MB-SPIN) and the TDEP/aTDEP frameworks were enhanced to construct effective models for atomistic and spin dynamics at mesoscopic scales, including temperature-dependent effective potentials and substitutional alloys.
- Machine Learning: The Mlacs package was integrated to perform machine-learning-assisted canonical sampling, generating equilibrium configurations and training interatomic potentials (MLIPs) using Abinit data.
- Automation: Workflows for high-throughput calculations were developed within AbiPy, Atomate2, and AiiDA, automating tasks from input generation to post-processing (e.g., luminescence spectra, thermal expansion, SHG tensors).

Key Results and Capabilities

Performance: GPU-accelerated versions of Abinit demonstrate speed-ups of 10–30x for LOBPCG and up to 40x for Chebyshev filtering compared to CPU-only runs on equivalent node counts, enabling the convergence of ground states for systems with thousands of atoms.
Accuracy: The PLM-cDFT implementation successfully reproduced experimental magnetic moments in $YBaCo_4O_7$ where conventional DFT+U failed. The extended DFT model successfully generated equation-of-state tables for materials like hydrogen and iron up to ideal plasma temperatures.
New Physics: The code successfully computed phonon angular momentum, flexoelectric coefficients, and polaronic states (e.g., in LiF) from first principles.
Workflow Integration: Automated workflows have been validated for computing luminescence spectra of defects, thermal expansion coefficients via Quasi-Harmonic Approximation (QHA), and second-harmonic generation (SHG) tensors for large datasets.

Significance and Claims
The paper claims that these updates significantly expand the scope of Abinit, reinforcing its position as a versatile tool for computational materials science. Key contributions include:

Bridging Scales: The integration of second-principles methods (Multibinit, TDEP) and machine learning (Mlacs) allows for the simulation of phenomena (phase transitions, domain walls) at length and time scales inaccessible to direct DFT, while retaining quantum mechanical accuracy.
Extreme Conditions: The implementation of thermal functionals and extended DFT models enables the predictive modeling of warm dense matter and plasmas, a regime previously difficult to access with standard plane-wave codes.
High-Throughput and Reproducibility: The development of robust interfaces with AbiPy, Atomate2, and AiiDA facilitates the automation of thousands of calculations, ensuring reproducibility and enabling large-scale materials screening.
Hardware Adaptation: The modernization of the code for GPU accelerators ensures that Abinit remains competitive in the era of exascale computing, offering substantial energy and time savings for large systems.

The authors emphasize that the software remains open-source, fostering a collaborative environment for developers and users to further extend these capabilities into relativistic treatments, non-adiabatic dynamics, and advanced many-body techniques.