Search for: All records

Accurate prediction of electron density is fundamental to estimating material properties at both quantum and macroscopic scales using density functional theory (DFT). However, achieving high accuracy without incurring prohibitive computational cost remains a persistent challenge. Here, we introduce a three-dimensional Special Euclidean group (SE(3))-equivariant steerable convolutional neural network for predicting bulk electron density across a wide range of material systems, including medium- and high-entropy alloys. Built on a group-theoretic framework, the model learns higher-order equivariant features that explicitly preserve the rotational and translational symmetries of physical space. We demonstrate that this approach surpasses a descriptor-based machine learning baseline by 33.6% in predicting 3D bulk electron density for high-entropy alloys when using the superposition of atomic densities (SAD) as input. Moreover, the model consistently delivers improved predictive accuracy and enhanced symmetry representation across diverse systems, including elemental (Al), molecular (water), ternary (SiGeSn), and quaternary (CrFeCoNi) materials. In addition to its accuracy, the model enables significantly faster inference compared to conventional DFT-based electron density calculations, making it well-suited for large-scale and high-throughput material screening. Notably, it also demonstrates strong element-interpolation capability, successfully predicting electron densities for previously unseen systems. For instance, it accurately models CrMnFeCoNi alloys with mono- and diatomic Mn configurations when trained only on CrFeCoNi data, and generalizes to the ternary SiGeSn system when trained exclusively on binary combinations of Si, Ge, and Sn. The model captures latent symmetry structures more effectively than standard convolutional neural networks, maintaining higher-order equivariance throughout the prediction process. 

The Pelona–Orocopia–Rand (POR) schists were emplaced during the Farallon flat subduction in the early Cenozoic and now occupy the root of major strike-slip faults of the San Andreas Fault system. The POR schists are considered frictionally stable at lower temperatures than other basement rocks, limiting the maximum depth of seismicity in Southern California. However, experimental constraints on the composition and frictional properties of POR schists are still missing. Here, we study the frictional behavior of synthetic gouge derived from Pelona, Portal, and Rand Mountain schist wall rocks under hydrothermal, triaxial conditions. We conduct velocity-step experiments from 0.04 to 1 μm/s from room temperature to 500ºC under 200 MPa effective normal stress, including a 30 MPa porefluid pressure. The frictional stability of POR schists in the lower crust is caused by a thermally activated transition from slip-rate- and state-dependent friction to inherently stable, rate-dependent creep between 300ºC and 500ºC, depending on sample composition and slip-rate. The mineralogy of POR schists shows much variability caused by different protoliths and metamorphic grades, featuring various amounts of phyllosilicates, quartz, feldspar, and amphibole. Pelona and Portal schists exhibit a velocity-weakening regime enabling the nucleation and propagation of earthquakes when exhumed in the middle crust, as in the Mojave section of the San Andreas Fault. The contrasted frictional properties of POR schists exemplify the lithological control of seismic processes and associated hazards. 

Online reinforcement learning (RL) enhances policies through direct interactions with the environment, but faces challenges related to sample efficiency. In contrast, offline RL leverages extensive pre-collected data to learn policies, but often produces suboptimal results due to limited data coverage. Recent efforts integrate offline and online RL in order to harness the advantages of both approaches. However, effectively combining online and offline RL remains challenging due to issues that include catastrophic forgetting, lack of robustness to data quality and limited sample efficiency in data utilization. In an effort to address these challenges, we introduce A3RL, which incorporates a novel confidence aware Active Advantage Aligned (A3) sampling strategy that dynamically prioritizes data aligned with the policy's evolving needs from both online and offline sources, optimizing policy improvement. Moreover, we provide theoretical insights into the effectiveness of our active sampling strategy and conduct diverse empirical experiments and ablation studies, demonstrating that our method outperforms competing online RL techniques that leverage offline data. Our code will be publicly available at:this https URL.