arXiv subjects

Petr Grigorev

Publications and source records attributed to Petr Grigorev.

4 recordsLinked to original sources

arXiv2024

Exploring parameter dependence of atomic minima with implicit differentiation

Interatomic potentials are essential to go beyond ab initio size limitations, but simulation results depend sensitively on potential parameters. Forward propagation of parameter variation is key for uncertainty quantification, whilst backpropagation has found application for emerging inverse problems such as fine-tuning or targeted design. Here, the implicit derivative of functions defined as a fixed point is used to Taylor expand the energy and structure of atomic minima in potential parameters, evaluating terms via automatic differentiation, dense linear algebra or a novel sparse operator approach. The latter allows efficient forward and backpropagation through relaxed structures of arbitrarily large systems. The implicit expansion accurately predicts lattice distortion and defect formation energies and volumes with classical and machine-learning potentials, enabling high-dimensional uncertainty propagation without prohibitive overhead. We then show how the implicit derivative can be used to solve challenging inverse problems, minimizing an implicit loss to fine-tune potentials and stabilize solute-induced structural rearrangements at dislocations in tungsten.

cond-mat.mtrl-sci↗

arXiv2023

A foundation model for atomistic materials chemistry

Atomistic simulations of matter, especially those that leverage first-principles (ab initio) electronic structure theory, provide a microscopic view of the world, underpinning much of our understanding of chemistry and materials science. Over the last decade or so, machine-learned force fields have transformed atomistic modeling by enabling simulations of ab initio quality over unprecedented time and length scales. However, early ML force fields have largely been limited by: (i) the substantial computational and human effort of developing and validating potentials for each particular system of interest; and (ii) a general lack of transferability from one chemical system to the next. Here we show that it is possible to create a general-purpose atomistic ML model, trained on a public dataset of moderate size, that is capable of running stable molecular dynamics for a wide range of molecules and materials. We demonstrate the power of the MACE-MP-0 model - and its qualitative and at times quantitative accuracy - on a diverse set of problems in the physical sciences, including properties of solids, liquids, gases, chemical reactions, interfaces and even the dynamics of a small protein. The model can be applied out of the box as a starting or "foundation" model for any atomistic system of interest and, when desired, can be fine-tuned on just a handful of application-specific data points to reach ab initio accuracy. Establishing that a stable force-field model can cover almost all materials changes atomistic modeling in a fundamental way: experienced users get reliable results much faster, and beginners face a lower barrier to entry. Foundation models thus represent a step towards democratising the revolution in atomic-scale modeling that has been brought about by ML force fields.

physics.chem-ph↗

arXiv2022

Optically-triggered strain-driven N\'{e}el vector manipulation in a metallic antiferromagnet

The absence of stray fields, their insensitivity to external magnetic fields, and ultrafast dynamics make antiferromagnets promising candidates for active elements in spintronic devices. Here, we demonstrate manipulation of the N\'{e}el vector in the metallic collinear antiferromagnet Mn$_2$Au by combining strain and femtosecond laser excitation. Applying tensile strain along either of the two in-plane easy axes and locally exciting the sample by a train of femtosecond pulses, we align the N\'{e}el vector along the direction controlled by the applied strain. The dependence on the laser fluence and strain suggests the alignment is a result of optically-triggered depinning of 90$^{\mathrm{o}}$ domain walls and their sliding in the direction of the free energy gradient, governed by the magneto-elastic coupling. The resulting, switchable, state is stable at room temperature and insensitive to magnetic fields. Such an approach may provide ways to realize robust high-density memory device with switching timescales in the picosecond range.

cond-mat.mtrl-sci↗

arXiv2021

Synergistic coupling in ab initio-machine learning simulations of dislocations

Ab initio simulations of dislocations are essential to build quantitative models of material strength, but the required system sizes are often at or beyond the limit of existing methods. Many important structures are thus missing in the training or validation of interatomic potentials, whilst studies of dislocation-defect interactions must mitigate the effect of strong periodic image interactions along the line direction. We show how these restrictions can be lifted through the use of linear machine learning potentials in hybrid simulations, where only a subset of atoms are governed by ab initio forces. The linear form is exploited in a constrained retraining procedure, qualitatively expanding the range of training structures for learning and giving precise matching of dislocation core structures, such that lines can cross the quantum/classical boundary. We apply our method to fully three dimensional studies of impurity segregation to edge and screw dislocations in tungsten. Our retrained potentials give systematically improved accuracy to QM/ML reference data and the three dimensional geometry allows for long-range relaxations that qualitatively change impurity-induced core reconstructions compared to simulations using short periodic supercells. More generally, the ability to treat arbitrary sub-regions of large scale simulations with ab initio accuracy opens a vast range of previously inaccessible extended defects to quantitative investigation.

cond-mat.mtrl-sci↗