A global team of scientists has created a novel technique for parameterizing machine-learning interatomic potentials (MLIP), which allows for simulating magnetic materials with enhanced reliability and precision. The primary characteristic of this innovative method lies in training the interaction models using what are known as “magnetic forces.”
This study paves the way for quicker development of materials aimed at advanced electronics, medical applications, and sensor technologies. The research was
published
in
Computational Materials Science.
Magnetic substances are ubiquitous—ranging from simple items like compass needles and fridge magnets to advanced components found in computers, MRI machines, and factory equipment. Mastering the manipulation of magnetism at the molecular scale will be crucial for upcoming innovations, including spintronic technology that utilizes both the electrical charge and the rotational movement of electrons, precise medication distribution through magnetic nanoparticles, or highly sensitive detectors.
Historically, the characteristics of materials have been investigated through experimental methods. Nonetheless, such studies tend to be costly and necessitate extremely pure specimens (as impurities significantly influence magnetism) along with sophisticated apparatus. That’s precisely when simulations become invaluable.
Among the highly precise simulation techniques, Density Functional Theory (DFT), which relies on principles from quantum mechanics, doesn’t demand extensive computational power for obtaining accurate measurements of a substance’s characteristics. Even though modeling just a couple of thousand atoms presents a significant challenge, numerous critical phenomena like lattice irregularities or changes in phases occur precisely within this range and thus warrant examination here.
To surmount this challenge, scientists are concentrating on developing MLIP—a class of AI-driven models designed to forecast both the energy of a system and the atomic forces based on information derived from precise yet time-consuming DFT computations. These MLIPs operate at speeds several magnitudes greater than those of traditional DFT calculations, thereby facilitating simulations of extensive systems over extended durations.
However, since standard MLIPs are not sufficient to study magnetic materials, the need to explicitly account for the magnetic moments of the atoms in the functional form of potentials brought their magnetic counterparts to life. But here a new problem arises: MLIP training requires much more data from even more expensive spin-polarized DFT calculations, because MLIPs must take into account both the arrangement of the atoms and the magnitude and direction of the magnetic moments.
The researchers successfully developed precise and dependable MLIPs using only a small quantity of costly training data. The main concept they employed involved training these models not just on energies, atomic forces, and stresses but also incorporating magnetic forces—which represent negative derivatives of energy concerning magnetic moments.
The training utilized data calculated from approximately 2,600 distinct atomic arrangements of an iron-aluminum (Fe-Al) alloy featuring varying proportions of components. This material, known for its intriguing magnetic characteristics, finds use in numerous technological fields.
When comparing models trained solely on energies, forces, and stresses to those additionally trained on magnetic forces, the new method demonstrated considerable benefits.
Importantly, the new method showed a tenfold reduction in prediction error for magnetic forces, but virtually no change for energies and conventional forces. The models trained on magnetic forces also proved to be more accurate in predicting the equilibrium magnetic moments of iron atoms.
An improvement in the reliability of the trained MLIPs was also crucial. Geometrical optimizations demonstrated that when modeling iron-aluminum systems without incorporating magnetic forces, the untrained models either could not settle the atomic configuration properly or yielded outcomes that were not physically meaningful.
Magnetic force training showed 100% reliability, with successful relaxation calculations and physically meaningful results, which is crucial for the practical use of MLIPs. In fact, magnetic force training helps to obtain a reliable model even with a relatively small training dataset.
The team successfully applied the best of the generated potentials to simulate the behavior of Fe-Al at room temperature (300 K) using molecular dynamics.
The simulation outcomes perfectly aligned with the thermal expansion witnessed during the experiment, even though there was a minor discrepancy in the numerical values. This variance could likely stem from the constraints of DFT when constructing the training data set. Consequently, this innovative method has proven useful for examining dynamics and temperature-related impacts.
Ivan Novikov, who serves as an associate professor at the HSE Faculty of Computer Science, an associate professor at the MIPT Department of Chemical Physics of Functional Materials, and a senior researcher at Skoltech, remarks, “Our main objective was to demonstrate that magnetic forces, often disregarded during potential training, provide extra insights into atomic interactions within magnetic substances.”
By considering these factors during the training of potential models, we managed to enhance both the accuracy of predicting magnetic properties and, equally important, boost the reliability of the simulations.
Now we can more reliably simulate intricate magnetic systems using the same quantity of costly quantum computing resources, which makes these studies more economical and consistent.
The uniqueness of this research stems from the meticulous creation, implementation, and thorough verification of the novel technique. This investigation offers compelling proof that the suggested strategy not only proves effective but also substantially enhances the dependability and precision of simulations, particularly when operating under constrained resources for quantum computation.
In the future, dependable and swift MLIPs will facilitate efficient virtual screenings and optimizations of the formulations for novel magnetic alloys, permanent magnet materials, magnetocaloric substances (used in magnetic refrigeration), as well as spintronics components.
It has become feasible to model extensive systems comprising tens of thousands of atoms, enabling investigation into how defects, grain boundaries, and nanoscale structuring influence magnetic characteristics. Additionally, researchers can explore magnetic phase transformations, such as identifying the Curie temperature.
Understanding magnetism at the atomic level is essential for improving the performance of electric motors, generators, transformers, data recording devices, and medical diagnostic and therapeutic systems, such as MRI.
The new method can work hand in hand with active learning algorithms that can identify the essential quantum computations required for further refinement of the model while the simulation is running. This will also help to reduce the number of DFT calculations.
The study was carried out by researchers from Skoltech, MIPT, HSE, the Institute of Solid State Chemistry and Mechanochemistry of the Siberian Branch of RAS, the Emanuel Institute of Biochemical Physics of RAS, and their colleagues from Germany, Norway, the United States, and Austria.
More information:
Alexey S. Kotykhov et al, Fitting to magnetic forces improves the reliability of magnetic Moment Tensor Potentials,
Computational Materials Science
(2024).
DOI: 10.1016/j.commatsci.2024.113331
Provided by Skolkovo Institute of Science and Technology
The tale was initially released on
WARINFO
. Subscribe to our
newsletter
For the most recent science and technology news updates.
