Ken-Ichi Nomura

Machine-learning interatomic potentials have demonstrated power-law scaling in predictive accuracy as training data and model capacity increase, but it remains unclear whether models trained at scale acquire interpretable chemical concepts. Here we show that an E(3)-equivariant machine-learning interatomic potential learns local bond information without direct supervision of bond properties. To expose this information, we develop an edge-wise emergent energy-decomposition framework and apply it to an Allegro neural-network potential trained on SPICE2, a dataset composed of stable molecular structures. The framework analyzes edge-wise energy contributions obtained from the trained model and their distributions together with information entropy to examine how data size, data composition and model-training scenarios shape the internal representation of chemical bonds. The resulting bond-dissociation energy estimates for archetypal bond types agree quantitatively with literature values and are consistent across models trained on organic and inorganic datasets. We further examine a hybrid training set and find that combining complementary chemical data improves transition-state energy-prediction accuracy while reshaping the learned bond representations. These results indicate that scalable interatomic potentials can acquire transferable bond concepts without explicit bond-energy labels, providing a way to analyze the transition-state energy-prediction problem.

The effects of different interatomic potentials on the amorphous structure of 1.1 nm tobermorite, a representative structural model for calcium–silicate–hydrate (C–S–H), are investigated systematically by molecular dynamics simulations using empirical potential Clayff, density functional theory (DFT), and machine-learning interatomic potential (MLMD) (Allegro-FM). Amorphous structures obtained through DFT and Allegro-FM are in agreement with each other when quenched from the same molten structure. In contrast, the structure generated using Clayff differs significantly from those obtained using these two approaches. The analysis of the radial distribution functions reveals that, except for the Si–O interaction, all interatomic distances obtained with Clayff are longer than those predicted by DFT and Allegro-FM. Consequently, the amorphous structure obtained with Clayff exhibits a lower density than those of the other two models. In addition to static structural analysis, the tensile response of amorphous tobermorite prepared at different quenching rates is examined using the MLMD Allegro-FM, and tensile simulations are performed along the x-, y -, and z -directions. The resulting stress–strain curves exhibit a clear dependence on the quenching rate (708, 70.8, and 7.08 K ps ⁻¹ ). Mechanical anisotropy gradually diminishes with a decreasing quenching rate. This finding demonstrates that even subtle variations in the static structure significantly influence dynamic behavior, and that isotropic mechanical responses emerge when a sufficiently slow quenching is employed, i.e. the elastic anisotropy becomes negligible when the cooling duration is further increased.

Supercritical water is found inside Earth’s mantle, where water is subjected to very high temperatures and pressures. It exhibits extraordinary properties, such as having a low dielectric constant and high reactivity, which stems from the breakdown of the hydrogen bond network in a supercritical state. This makes supercritical water a non-polar solvent and the basis for many innovative technologies. We investigate supercritical water at ten densities (0.1–1.0 gr/cc) at 1000 K to study the structural correlations, such as atom-resolved partial pair distributions, co-ordination numbers, bond-angle distributions and neutron scattering, and x-ray structure factors. Among the dynamical correlations, we investigate the velocity autocorrelation function, current–current correlation function, and their Fourier transforms—vibrational density-of-states and frequency dependent dielectric constant. Structural and dynamical correlations are computed from time-trajectories of the positions and velocities calculated ab initio molecular dynamics within the density functional theory framework using the SCAN exchange–correlation functional. Our results for structural correlations are compared with the neutron scattering experiments on supercritical water by Soper and collaborators [J. Chem. Phys. 106, 247–254 (1997)] and dynamical correlations in the supercritical state are compared with the inelastic neutron scattering results by Car and collaborators [J. Phys. Chem. Lett. 11, 9461–9467 (2020)].

Light-matter dynamics in topological quantum materials enables ultralow-power, ultrafast devices. A challenge is simulating multiple field and particle equations for light, electrons, and atoms over vast spatiotemporal scales on Exaflop/s computers with increased heterogeneity and low-precision focus. We present a paradigm shift that solves the multiscale/multiphysics/heterogeneity challenge harnessing hardware heterogeneity and low-precision arithmetic. Divide-conquer-recombine algorithms divide the problem into not only spatial but also physical subproblems of small dynamic ranges and minimal mutual information, which are mapped onto best-characteristics-matching hardware units, while metamodel-space algebra minimizes communication and precision requirements. Using 60,000 GPUs of Aurora, DC-MESH (divide-and-conquer Maxwell-Ehrenfest-surface hopping) and XS-NNQMD (excited-state neural-network quantum molecular dynamics) modules of MLMD (multiscale light-matter dynamics) software were 152- and 3,780-times faster than the state-of-the-art for 15.4 million-electron and 1.23 trillion-atom PbTiO3 material, achieving 1.87 EFLOP/s for the former. This enabled the first study of light-induced switching of topological superlattices for future ferroelectric 'topotronics'.

We present a foundation model for exascale molecular dynamics simulations by leveraging an E(3) equivariant network architecture (Allegro) and a set of large-scale organic and inorganic materials data sets merged by the Total Energy Alignment framework. The obtained model (Allegro-FM) is versatile for various material simulations for diverse downstream tasks covering 89 elements in the training sets. Allegro-FM exhibits excellent agreement with high-level quantum chemistry theories in describing structural, mechanical, and thermodynamic properties, while exhibiting emergent capabilities for structural correlations, reaction kinetics, mechanical strengths, fracture, and solid/liquid dissolution, for which the model has not been trained. Furthermore, we demonstrate the robust predictability and generalizability of Allegro-FM for chemical reactions using Transition1x, which consists of tens of thousands of organic reactions and 9.6 million configurations including transition state data, in addition to reactive simulations using calcium silicate hydrates as a test bed. With its computationally efficient, strictly local network architecture, Allegro-FM scales up to multibillion-atom systems with a parallel efficiency of 0.975 on the exaflop/s Aurora supercomputer at Argonne Leadership Computing Facility. The approach presented in this work demonstrates the potential of the foundation model for novel materials design and discovery based on large-scale atomistic simulations.

Machine Learning Interatomic Potentials (MLIPs) have successfully demonstrated scaling behavior, i.e. the power-law improvement in training performance, however the emergence of novel capabilities at scale remains unexplored. We have developed Edge-wise Emergent Decomposition (E3D) framework to investigate how an MLIP develops the ability to derive physically meaningful local representations of chemical bonds without explicit supervision. Employing an E(3)-equivariant network (Allegro) trained on molecular data (SPICE~2), we found that the trained MLIP spontaneously learned representations of bond dissociation energy (BDE) by decomposing the global potential energy landscape. The learned BDE values quantitatively agree with literature and its scalability are found to be robust across diverse training datasets, suggesting the presence of underlying representation that captures chemical reactions faithfully beyond given training information. Our E3D analysis utilizing Shannon's entropy reveals a close interplay between the decomposability of potential energy learning, scalability of learning, and emergent chemical reactivity, thus providing novel insights of scaling limitations and pathways toward more physically interpretable and predictive simulations.

High electric fields can significantly alter catalytic environments and the resultant chemical processes. Such fields arise naturally in biological systems but can also be artificially induced through localized nanoscale excitations. Recently, strong field excitation of dielectric nanoparticles has emerged as an avenue for studying catalysis in highly ionized environments, producing extreme electric fields. While the dynamics of laser-driven surface ion emission has been extensively explored, understanding the molecular dynamics leading to fragmentation has remained elusive. Here, we employ a multiscale approach performing nonadiabatic quantum molecular dynamics (NAQMD) simulations on hydrogenated silica surfaces in both bare and wetted environments under field conditions mimicking those of an ionized nanoparticle. Our findings indicate that hole localization drives fragmentation dynamics, leading to surface silanol dissociation within 50 fs and charge transfer-induced water splitting in wetted environments within 150 fs. Further insight into such ultrafast mechanisms is critical for the advancement of catalysis on the surface of charged nanosystems.