npj Computational Materials最新文献

To advance the development of high-strength polycrystalline metallic materials towards achieving carbon neutrality, it is essential to design materials in which the atomic level control of general grain boundaries (GGBs), which govern the material properties, is achieved. However, owing to the complex and diverse structures of GGBs, there have been no reports on interatomic potentials capable of reproducing them. This accuracy is essential for conducting molecular dynamics analyses to derive material design guidelines. In this study, we constructed a machine learning interatomic potential (MLIP) with density functional theory (DFT) accuracy to model the energy, atomic structure, and dynamics of arbitrary grain boundaries (GBs), including GGBs, in Œ±-Fe. Specifically, we employed a training dataset comprising diverse atomic structures generated based on crystal space groups. The GGB accuracy was evaluated by directly comparing with DFT calculations performed on cells cut near GBs from nano-polycrystals, and extrapolation grades of the local atomic environment based on active learning methods for the entire nano-polycrystal. Furthermore, we analyzed the GB energy and atomic structure in Œ±-Fe polycrystals through large-scale molecular dynamics analysis using the constructed MLIP. The average GB energy of Œ±-Fe polycrystals calculated by the constructed MLIP is 1.57‚ÄâJ/m², exhibiting good agreement with experimental predictions. Our findings demonstrate the methodology for constructing an MLIP capable of representing GGBs with high accuracy, thereby paving the way for materials design based on computational materials science for polycrystalline materials.

Recent advances in deep learning generative models (GMs) have created high capabilities in accessing and assessing complex high-dimensional data, allowing superior efficiency in navigating vast material configuration space in search of viable structures. Coupling such capabilities with physically significant data to construct trained models for materials discovery is crucial to moving this emerging field forward. Here, we present a universal GM for crystal structure prediction (CSP) via a conditional crystal diffusion variational autoencoder (Cond-CDVAE) approach, which is tailored to allow user-defined material and physical parameters such as composition and pressure. This model is trained on an expansive dataset containing over 670,000 local minimum structures, including a rich spectrum of high-pressure structures, along with ambient-pressure structures in Materials Project database. We demonstrate that the Cond-CDVAE model can generate physically plausible structures with high fidelity under diverse pressure conditions without necessitating local optimization, accurately predicting 59.3% of the 3547 unseen ambient-pressure experimental structures within 800 structure samplings, with the accuracy rate climbing to 83.2% for structures comprising fewer than 20 atoms per unit cell. These results meet or exceed those achieved via conventional CSP methods based on global optimization. The present findings showcase substantial potential of GMs in the realm of CSP.

Molecular dynamics (MD) is an indispensable atomistic-scale computational tool widely-used in various disciplines. In the past decades, nearly all ab initio MD and machine-learning MD have been based on the general-purpose central/graphics processing units (CPU/GPU), which are well-known to suffer from their intrinsic “memory wall” and “power wall” bottlenecks. Consequently, nowadays MD calculations with ab initio accuracy are extremely time-consuming and power-consuming, imposing serious restrictions on the MD simulation size and duration. To solve this problem, here we propose a special-purpose MD processing unit (MDPU), which could reduce MD time and power consumption by about 10³ times (10⁹ times) compared to state-of-the-art machine-learning MD (ab initio MD) based on CPU/GPU, while keeping ab initio accuracy. With significantly-enhanced performance, the proposed MDPU may pave a way for the accurate atomistic-scale analysis of large-size and/or long-duration problems which were impossible/impractical to compute before.

In metal additive manufacturing (AM), processing parameters can affect the probability of macroscopic defect formation (lack-of-fusion, keyholing, balling), which can, in turn, jeopardize the final product’s integrity. A printability map classifies regions in the processing space where an alloy can be printed with or without porosity defects. However, the creation of these printability maps is resource-intensive. Previous efforts to generate printability maps have required single-track experiments on pre-alloyed powder, limiting the utilization of these printability maps for the high-throughput design of printable alloys. We address these challenges in the case of Laser Powder Bed Fusion AM (L-PBF-AM) by introducing a fully computational, predictive approach to create printability maps for arbitrary alloys. Our framework uses physics-based thermal models and a variety of defect formation criteria. We benchmark the predictive ability of the proposed framework against literature data for the following commonly printed alloys: 316 Stainless Steel, Inconel 718, Ti-6Al-4V, AF96, and Ni-5Nb. Furthermore, we deploy the framework on NiTi-based Shape Memory Alloys (SMAs) as a case study. We scrutinize the accuracy of various sets of defect criteria and use these accuracy measurements to create an uncertainty-aware probabilistic framework capable of predicting the printability maps of arbitrary alloys. This framework has the potential to guide alloy designers to potentially easy-to-print alloys, enabling the co-design of high-performing printable alloys.

Upgrades to advanced scientific user facilities such as next-generation x-ray light sources, nanoscience centers, and neutron facilities are revolutionizing our understanding of materials across the spectrum of the physical sciences, from life sciences to microelectronics. However, these facility and instrument upgrades come with a significant increase in complexity. Driven by more exacting scientific needs, instruments and experiments become more intricate each year. This increased operational complexity makes it ever more challenging for domain scientists to design experiments that effectively leverage the capabilities of and operate on these advanced instruments. Large language models (LLMs) can perform complex information retrieval, assist in knowledge-intensive tasks across applications, and provide guidance on tool usage. Using x-ray light sources, leadership computing, and nanoscience centers as representative examples, we describe preliminary experiments with a Context-Aware Language Model for Science (CALMS) to assist scientists with instrument operations and complex experimentation. With the ability to retrieve relevant information from facility documentation, CALMS can answer simple questions on scientific capabilities and other operational procedures. With the ability to interface with software tools and experimental hardware, CALMS can conversationally operate scientific instruments. By making information more accessible and acting on user needs, LLMs could expand and diversify scientific facilities’ users and accelerate scientific output.

The quantification of interphase properties between metals and their corresponding hydrides is crucial for modeling the thermodynamics and kinetics of the hydrogenation processes in solid-state hydrogen storage materials. In particular, interphase boundary energies assume a pivotal role in determining the kinetics of nucleation, growth, and coarsening of hydrides, alongside accompanying morphological evolution during hydrogenation. The total interphase energy arises from both chemical bonding and mechanical strains in these solid-state systems. Since these contributions are usually coupled, it is challenging to distinguish via conventional computational approaches. Here, a comprehensive atomistic modeling methodology is developed to decouple chemical and mechanical energy contributions using first-principles calculations, of which feasibility is demonstrated by quantifying chemical and elastic strain energies of key interfaces within the FeTi metal-hydride system. Derived materials parameters are then employed for mesoscopic micromechanical analysis, predicting crystallographic orientations in line with experimental observations. The multiscale approach outlined verifies the importance of the chemo-mechanical interplay in the morphological evolution of growing hydride phases, and can be generalized to investigate other systems. In addition, it can streamline the design of atomistic models for the quantitative evaluation of interphase properties between dissimilar phases and allow for efficient predictions of their preferred phase boundary orientations.

The self-consistent field (SCF) generation of the three-dimensional (3D) electron density distribution (ρ) represents a fundamental aspect of density functional theory (DFT) and related first-principles calculations, and how one can shorten or bypass the SCF loop represents a critical question in electronic structure theory from both practical and fundamental standpoints. Herein, a machine learning strategy, DeepSCF, is presented in which the map between the SCF ρ and the initial guess density (ρ₀) constructed by the summation of neutral atomic densities is learned using 3D convolutional neural networks (CNNs). High accuracy and transferability of DeepSCF are achieved by first encoding ρ₀ on a 3D grid and then expanding the input features to include atomic fingerprints beyond ρ₀. The prediction of the residual density (δρ) rather than ρ itself is targeted, and given that δρ is indicative of chemical bonding information, a dataset of small-sized organic molecules featuring diverse bonding characters is adopted. The fidelity of DeepSCF is finally enhanced by subjecting the atomic geometries of the dataset to random rotations and strains. The effectiveness of DeepSCF is demonstrated using a complex carbon nanotube-based DNA sequencer model. This work evidences that the nearsightedness in electronic structure can be optimally represented via the spatial locality in CNNs, offering insight into the success of various machine learning-based atomistic materials simulations.

The conversion efficiency from charge current to spin current via the spin Hall effect is evaluated by the spin Hall ratio (SHR). Through state-of-the-art ab initio calculations involving both charge conductivity and spin Hall conductivity, we report the SHRs of the III-V monolayer family, revealing an ultrahigh ratio of 0.58 in the hole-doped GaAs monolayer. In order to find more promising 2D materials, a descriptor for high SHR is proposed and applied to a high-throughput database, which provides the fully relativistic band structures and Wannier Hamiltonians of 216 exfoliable monolayer semiconductors and has been released to the community. Among potential candidates for high SHR, the MXene monolayer Sc₂CCl₂ is identified with the proposed descriptor and confirmed by computation, demonstrating the descriptor validity for high SHR materials discovery.

Single-atom catalysts (SACs) with multiple active sites exhibit high activity for a wide range of sluggish reactions, but identifying optimal multimetallic SAC is challenging due to the vast design space. Here, we present a self-driving computational strategy that combines first-principles calculations and equivariant graph neural network (GNN) to explore over 30,000 binary metallic sites with varying combinations of 3d transition metals and different ligand environments for oxygen reduction and evolution reactions (ORR/OER). Active learning facilitates the investigation of the search space by balancing the exploration of unseen atomic structures with the exploitation of the active ones. The GNN learns the chemical environments to capture composition-structure-property relationships for ORR/OER activity and selectivity. The computational predictions of promising Co-Fe, Co-Co, and Co-Zn metal pairs are consistent with the state-of-the-art results of experimental measurements reported in the literature. This approach can be extended to a broader class of multi-element high entropic materials systems.

Materials datasets usually contain many redundant (highly similar) materials due to the tinkering approach historically used in material design. This redundancy skews the performance evaluation of machine learning (ML) models when using random splitting, leading to overestimated predictive performance and poor performance on out-of-distribution samples. This issue is well-known in bioinformatics for protein function prediction, where tools like CD-HIT are used to reduce redundancy by ensuring sequence similarity among samples greater than a given threshold. In this paper, we survey the overestimated ML performance in materials science for material property prediction and propose MD-HIT, a redundancy reduction algorithm for material datasets. Applying MD-HIT to composition- and structure-based formation energy and band gap prediction problems, we demonstrate that with redundancy control, the prediction performances of the ML models on test sets tend to have relatively lower performance compared to the model with high redundancy, but better reflect models’ true prediction capability.