arXiv subjects

Ralf Drautz

Publications and source records attributed to Ralf Drautz.

At least 37 records · Page 2Linked to original sources

arXiv2025

Core structure of dislocations in ordered ferromagnetic FeCo

We elucidated the core structure of screw dislocations in ordered B2 FeCo using a recent magnetic bond-order potential (BOP) [Egorov et al., Phys. Rev. Mater. 7, 044403 (2023)]. We corroborated that dislocations in B2 FeCo exist in pairs separated by antiphase boundaries. The equilibrium separation is about 50 A, which demands large-scale atomistic simulations - inaccessible for density functional theory but attainable with BOP. We performed atomistic simulations of these separated dislocations with BOP and predicted that they reside in degenerate core structures. Also, dislocations induce changes in the local electronic structure and magnetic moments.

cond-mat.mtrl-sci↗

arXiv2025

A practical guide to machine learning interatomic potentials -- Status and future

The rapid development and large body of literature on machine learning interatomic potentials (MLIPs) can make it difficult to know how to proceed for researchers who are not experts but wish to use these tools. The spirit of this review is to help such researchers by serving as a practical, accessible guide to the state-of-the-art in MLIPs. This review paper covers a broad range of topics related to MLIPs, including (i) central aspects of how and why MLIPs are enablers of many exciting advancements in molecular modeling, (ii) the main underpinnings of different types of MLIPs, including their basic structure and formalism, (iii) the potentially transformative impact of universal MLIPs for both organic and inorganic systems, including an overview of the most recent advances, capabilities, downsides, and potential applications of this nascent class of MLIPs, (iv) a practical guide for estimating and understanding the execution speed of MLIPs, including guidance for users based on hardware availability, type of MLIP used, and prospective simulation size and time, (v) a manual for what MLIP a user should choose for a given application by considering hardware resources, speed requirements, energy and force accuracy requirements, as well as guidance for choosing pre-trained potentials or fitting a new potential from scratch, (vi) discussion around MLIP infrastructure, including sources of training data, pre-trained potentials, and hardware resources for training, (vii) summary of some key limitations of present MLIPs and current approaches to mitigate such limitations, including methods of including long-range interactions, handling magnetic systems, and treatment of excited states, and finally (viii) we finish with some more speculative thoughts on what the future holds for the development and application of MLIPs over the next 3-10+ years.

cond-mat.mtrl-sci↗

arXiv2024

Charge-constrained Atomic Cluster Expansion

The atomic cluster expansion (ACE) efficiently parameterizes complex energy surfaces of pure elements and alloys. Due to the local nature of the many-body basis, ACE is inherently local or semilocal for graph ACE. Here, we employ descriptor-constrained density functional theory for parameterizing ACE with charge or other degrees of freedom, thereby transfering the variational property of the density functional to ACE. The descriptors can be of scalar, vectorial or tensorial nature. From the simplest case of scalar atomic descriptors we directly obtain charge-dependent ACE with long-range electrostatic interactions between variable charges. We observe that the variational properties of the charges greatly help in training, avoiding the need for charge-constrained DFT calculations.

cond-mat.mtrl-sci↗

arXiv2024

Adaptive-precision potentials for large-scale atomistic simulations

Large-scale atomistic simulations rely on interatomic potentials providing an efficient representation of atomic energies and forces. Modern machine-learning (ML) potentials provide the most precise representation compared to electronic structure calculations while traditional potentials provide a less precise, but computationally much faster representation and thus allow simulations of larger systems. We present a method to combine a traditional and a ML potential to a multi-resolution description, leading to an adaptive-precision potential with an optimum of performance and precision in large complex atomistic systems. The required precision is determined per atom by a local structure analysis and updated automatically during simulation. We use copper as demonstrator material with an embedded atom model as classical force field and an atomic cluster expansion (ACE) as ML potential, but in principle a broader class of potential combinations can be coupled by this method. The approach is developed for the molecular-dynamics simulator LAMMPS and includes a load-balancer to prevent problems due to the atom dependent force-calculation times, which makes it suitable for large-scale atomistic simulations. The developed adaptive-precision copper potential represents the ACE-forces and -energies with a precision of 10 meV/Å and 0 meV for the precisely calculated atoms in a nanoindentation of 4 million atoms calculated for 100 ps and shows a speedup of 11.3 compared with a full ACE simulation.

physics.comp-ph↗

arXiv2024

Development of an Atomic Cluster Expansion potential for iron and its oxides

The combined structural and electronic complexity of iron oxides poses many challenges to atomistic modeling. To leverage limitations in terms of the accessible length and time scales, one requires a physically justified interatomic potential which is accurate to correctly account for the complexity of iron-oxygen systems. Such a potential is not yet available in the literature. In this work, we propose a machine-learning potential based on the Atomic Cluster Expansion for modeling the iron-oxygen system, which explicitly accounts for magnetism. We test the potential on a wide range of properties of iron and its oxides, and demonstrate its ability to describe the thermodynamics of systems spanning the whole range of oxygen content and including magnetic degrees of freedom.

cond-mat.mtrl-sci↗

arXiv2024

Efficient parameterization of transferable Atomic Cluster Expansion for water

We present a highly accurate and transferable parameterization of water using the atomic cluster expansion (ACE). To efficiently sample liquid water, we propose a novel approach that involves sampling static calculations of various ice phases and utilizing the active learning (AL) feature of ACE-based D-optimality algorithm to select relevant liquid water configurations, bypassing computationally intensive ab-initio molecular dynamics (AIMD) simulations. Our results demonstrate that the ACE descriptors enable a potential initially-fitted solely on ice structures which is later upfitted with few configurations of liquid, identified with active learning to provide an excellent description of liquid water. The developed potential exhibits remarkable agreement with first-principles reference, accurately capturing various properties of liquid water, including structural characteristics such as pair correlation functions, covalent bonding profiles, and hydrogen bonding profiles, as well as dynamic properties like the vibrational density of states, diffusion coefficient and thermodynamic properties such as the melting point of the ice Ih. Our research introduces a new and efficient sampling technique for machine learning potentials in water simulations, while also presenting a transferable interatomic potential for water that reveals the accuracy of first principles reference. This advancement not only enhances our understanding the relationship between ice and liquid water at the atomic level, but also opens up new avenues for studying complex aqueous systems.

cond-mat.mtrl-sci↗

arXiv2024

From electrons to phase diagrams with classical and machine learning potentials: automated workflows for materials science with pyiron

We present a comprehensive and user-friendly framework built upon the pyiron integrated development environment (IDE), enabling researchers to perform the entire Machine Learning Potential (MLP) development cycle consisting of (i) creating systematic DFT databases, (ii) fitting the Density Functional Theory (DFT) data to empirical potentials or MLPs, and (iii) validating the potentials in a largely automatic approach. The power and performance of this framework are demonstrated for three conceptually very different classes of interatomic potentials: an empirical potential (embedded atom method - EAM), neural networks (high-dimensional neural network potentials - HDNNP) and expansions in basis sets (atomic cluster expansion - ACE). As an advanced example for validation and application, we show the computation of a binary composition-temperature phase diagram for Al-Li, a technologically important lightweight alloy system with applications in the aerospace industry.

cond-mat.mtrl-sci↗

arXiv2024

Graph Atomic Cluster Expansion for semilocal interactions beyond equivariant message passing

The Atomic Cluster Expansion provides local, complete basis functions that enable efficient parametrization of many-atom interactions. We extend the Atomic Cluster Expansion to incorporate graph basis functions. This naturally leads to representations that enable the efficient description of semilocal interactions in physically and chemically transparent form. Simplification of the graph expansion by tensor decomposition results in an iterative procedure that comprises current message-passing machine learning interatomic potentials. We demonstrate the accuracy and efficiency of the graph Atomic Cluster Expansion for a number of small molecules, clusters and a general-purpose model for carbon. We further show that the graph Atomic Cluster Expansion scales linearly with number of neighbors and layer depth of the graph basis functions.

cond-mat.mtrl-sci↗

arXiv2023

Structural transformations driven by local disorder at interfaces

Despite the fundamental importance of solid-solid transformations in many technologies, the microscopic mechanisms remain poorly understood. Here, we explore the atomistic mechanisms at the migrating interface during solid-solid phase transformations between the topologically closed-packed A15 and body-centred cubic phase in tungsten. The high energy barriers and slow dynamics associated with this transformation require the application of enhanced molecular sampling approaches. To this end, we performed metadynamics simulations in combination with a path collective variable derived from a machine learning classification of local structural environments, which allows the system to freely sample the complex interface structure. A disordered region of varying width forming at the migrating interface is identified as a key physical descriptor of the transformation mechanisms, facilitating the atomic shuffling and rearrangement necessary for structural transformations. Furthermore, this can directly be linked to the differences in interface mobility for distinct orientation relationships as well as the formation of interfacial ledges during the migration along low-mobility directions.

cond-mat.mtrl-sci↗

arXiv2023

Atomic cluster expansion for quantum-accurate large-scale simulations of carbon

We present an atomic cluster expansion (ACE) for carbon that improves over available classical and machine learning potentials. The ACE is parameterized from an exhaustive set of important carbon structures at extended volume and energy range, computed using density functional theory (DFT). Rigorous validation reveals that ACE predicts accurately a broad range of properties of both crystalline and amorphous carbon phases while being several orders of magnitude more computationally efficient than available machine learning models. We demonstrate the predictive power of ACE on three distinct applications, brittle crack propagation in diamond, evolution of amorphous carbon structures at different densities and quench rates and nucleation and growth of fullerene clusters under high pressure and temperature conditions.

cond-mat.mtrl-sci↗

arXiv2023

Non-collinear Magnetic Atomic Cluster Expansion for Iron

The Atomic Cluster Expansion (ACE) provides a formally complete basis for the local atomic environment. ACE is not limited to representing energies as a function of atomic positions and chemical species, but can be generalized to vectorial or tensorial properties and to incorporate further degrees of freedom (DOF). This is crucial for magnetic materials with potential energy surfaces that depend on atomic positions and atomic magnetic moments simultaneously. In this work, we employ the ACE formalism to develop a non-collinear magnetic ACE parametrization for the prototypical magnetic element Fe. The model is trained on a broad range of collinear and non-collinear magnetic structures calculated using spin density functional theory. We demonstrate that the non-collinear magnetic ACE is able to reproduce not only ground state properties of various magnetic phases of Fe but also the magnetic and lattice excitations that are essential for a correct description of the finite temperature behavior and properties of crystal defects.

cond-mat.mtrl-sci↗

arXiv2023

Magnetic bond-order potential for iron-cobalt alloys

For large-scale atomistic simulations of magnetic materials, the interplay of atomic and magnetic degrees of freedom needs to be described with high computational efficiency. Here we present an analytic bond-order potential (BOP) for iron-cobalt, an interatomic potential based on a coarse-grained description of the electronic structure. We fitted BOP parameters to magnetic and non-magnetic density-functional theory (DFT) calculations of Fe, Co, and Fe-Co bulk phases. Our BOP captures the electronic structure of magnetic and non-magnetic Fe-Co phases. It provides accurate predictions of structural stability, elastic constants, phonons, point and planar defects, and structural transformations. It also reproduces the DFT-predicted sequence of stable ordered phases peculiar to Fe-Co and the stabilization of B2 against disordered phases by magnetism. Our Fe-Co BOP is suitable for atomistic simulations with thousands and millions of atoms.

cond-mat.mtrl-sci↗

arXiv2023

Atomic Cluster Expansion for a General-Purpose Interatomic Potential of Magnesium

We present a general-purpose parameterization of the atomic cluster expansion (ACE) for magnesium. The ACE shows outstanding transferability over a broad range of atomic environments and captures physical properties of bulk as well as defective Mg phases in excellent agreement with reference first-principles calculations. We demonstrate the computational efficiency and the predictive power of ACE by calculating properties of extended defects and by evaluating the P-T phase diagram covering temperatures up to 3000 K and pressures up to 80 GPa. We compare the ACE predictions with those of other interatomic potentials, including the embedded-atom method, an angular-dependent potential, and a recently developed neural network potential. The comparison reveals that ACE is the only model that is able to predict correctly the phase diagram in close agreement with experimental observations.

cond-mat.mtrl-sci↗

arXiv2023

Atomic cluster expansion for Pt-Rh catalysts: From ab initio to the simulation of nanoclusters in few steps

Insight into structural and thermodynamic properties of nanoparticles is crucial for designing optimal catalysts with enhanced activity and stability. We present a semi-automated workflow for parameterizing the atomic cluster expansion (ACE) from ab initio data. The main steps of the workflow are the generation of training data from accurate electronic structure calculations, an efficient fitting procedure supported by active learning and uncertainty indication, and a thorough validation. We apply the workflow to the simulation of binary Pt-Rh nanoparticles that are important for catalytic applications. We demonstrate that the Pt-Rh ACE is able to reproduce accurately a broad range of fundamental properties of the elemental metals as well as their compounds while retaining an outstanding computational efficiency. This enables a direct comparison of simulations to high resolution experiments.

cond-mat.mtrl-sci↗

arXiv2023

Parameterization protocol and refinement strategies for accurate and transferable analytic bond-order potentials: Application to Re

Interatomic potentials provide a means to simulate extended length and time scales that are outside the reach of ab initio calculations. The development of an interatomic potential for a particular material requires the optimization of the parameters of the functional form of the potential. We present a parameterization protocol for analytic bond-order potentials (BOP) that provide a physically transparent and computationally efficient description of the interatomic interaction. The parameterization protocol of the BOP follows the derivation of the BOP along the coarse-graining of the electronic structure from density-functional theory (DFT) to the tight-binding (TB) bond model to analytic BOPs. In particular, it starts from TB parameters that are obtained by downfolding DFT eigenstates of two-atomic molecules to an $sd$-valent minimal basis. This $sd$-valent Hamiltonian is combined with a pairwise repulsion to obtain an initial binding energy relation. The $s$ electrons are then removed from the Hamiltonian and instead represented by an isotropic embedding term. In the final step, the parameters of the remaining $d$-$d$ interaction, the pair repulsion and the embedding term are optimized simultaneously. We demonstrate that the application of this parameterization protocol leads to a basic BOP for Re with good transferability. We discuss different strategies to refine the basic BOP towards global transferability or towards local accuracy. We demonstrate that homogeneous samplings of the structural phase-space in a map of local atomic environments can be used to systematically increase the global transferability. We also demonstrate the influence of training data-weighting on local accuracy refinements with a Pareto-front analysis and suggest further requirements to select a final BOP. The local accuracy and global transferability of the final BOP is also shown and compared to DFT.

cond-mat.mtrl-sci↗

arXiv2023

Off-stoichiometric softening and polytypic transformations in the plastic deformation of the C14 Fe$_2$Nb Laves phase

Plastic deformation of the brittle C14-Fe$_2$Nb Laves phase occurs mostly by basal slip due to the complex crystal structure. Here, we compare the barriers for basal slip for the known mechanisms of direct slip, synchroshear and undulating slip using density functional theory calculations. According to our calculated generalized stacking fault (SF) energies, the most favorable mechanisms are synchroshear and undulating slip. Both mechanisms lead to stable SF with a formation energy of 50 $mJ/m^2$ through the same unstable SF configuration at the transition. The energy barrier of approximately 3 $J/m^2$ indicates a low dislocation mobility as expected from the brittle character. We also determine the influence of vacancies and antisite defects on the formation energy of stable and unstable SF. Both kinds of point defects tend to lower the energy barrier on both sides of 2:1 stoichiometry. This explains the experimentally observed off-stoichiometric softening of C14-Fe$_2$Nb. The small energy differences between the Fe$_2$Nb Laves phase polytypes raises the question if there are further deformation mechanisms with low barrier. Therefore, we additionally consider transformations between C14, C15 and C36 Laves phases as further deformation mechanism. Our calculations for polytypic transformations by successive synchroshear steps show that the corresponding energy barriers are in fact very similar to the energy barrier for basal slip in C14. This suggests that the energy needed to create a stable SF in C14 by synchroshear is also sufficient to initiate polytypic transformations where existing SFs in C14 are further transformed to form C15 or C36 Laves phases.

cond-mat.mtrl-sci↗

arXiv2022

Active learning strategies for atomic cluster expansion models

The atomic cluster expansion (ACE) was proposed recently as a new class of data-driven interatomic potentials with a formally complete basis set. Since the development of any interatomic potential requires a careful selection of training data and thorough validation, an automation of the construction of the training dataset as well as an indication of a model's uncertainty are highly desirable. In this work, we compare the performance of two approaches for uncertainty indication of ACE models based on the D-optimality criterion and ensemble learning. While both approaches show comparable predictions, the extrapolation grade based on the D-optimality (MaxVol algorithm) is more computationally efficient. In addition, the extrapolation grade indicator enables an active exploration of new structures, opening the way to the automated discovery of rare-event configurations. We demonstrate that active learning is also applicable to explore local atomic environments from large-scale MD simulations.

cond-mat.mtrl-sci↗

arXiv2022

The Design Space of E(3)-Equivariant Atom-Centered Interatomic Potentials

The rapid progress of machine learning interatomic potentials over the past couple of years produced a number of new architectures. Particularly notable among these are the Atomic Cluster Expansion (ACE), which unified many of the earlier ideas around atom density-based descriptors, and Neural Equivariant Interatomic Potentials (NequIP), a message passing neural network with equivariant features that showed state of the art accuracy. In this work, we construct a mathematical framework that unifies these models: ACE is generalised so that it can be recast as one layer of a multi-layer architecture. From another point of view, the linearised version of NequIP is understood as a particular sparsification of a much larger polynomial model. Our framework also provides a practical tool for systematically probing different choices in the unified design space. We demonstrate this by an ablation study of NequIP via a set of experiments looking at in- and out-of-domain accuracy and smooth extrapolation very far from the training data, and shed some light on which design choices are critical for achieving high accuracy. Finally, we present BOTNet (Body-Ordered-Tensor-Network), a much-simplified version of NequIP, which has an interpretable architecture and maintains accuracy on benchmark datasets.

stat.ML↗