npj Computational Materials最新文献

Obtaining microscopic structure-property relationships for grain boundaries is challenging due to their complex atomic structures. Recent efforts use machine learning to derive these relationships, but the way the atomic grain boundary structure is represented can have a significant impact on the predictions. Key steps for property prediction common to grain boundaries and other variable-sized atom clustered structures include: (1) describing the atomic structure as a feature matrix, (2) transforming the variable-sized feature matrix to a fixed length common to all structures, and (3) applying a machine learning algorithm to predict properties from the transformed matrices. We examine how these steps and different combinations of engineered features impact the accuracy of grain boundary energy predictions using a database of over 7000 grain boundaries. Additionally, we assess how different engineered features support interpretability, offering insights into the physics of the structure-property relationships.

Density-functional theory with extended Hubbard functionals (DFT + U + V) provides a robust framework to accurately describe complex materials containing transition-metal or rare-earth elements. It does so by mitigating self-interaction errors inherent to semi-local functionals which are particularly pronounced in systems with partially-filled d and f electronic states. However, achieving accuracy in this approach hinges upon the accurate determination of the on-site U and inter-site V Hubbard parameters. In practice, these are obtained either by semi-empirical tuning, requiring prior knowledge, or, more correctly, by using predictive but expensive first-principles calculations. Here, we present a machine learning model based on equivariant neural networks which uses atomic occupation matrices as descriptors, directly capturing the electronic structure, local chemical environment, and oxidation states of the system at hand. We target here the prediction of Hubbard parameters computed self-consistently with iterative linear-response calculations, as implemented in density-functional perturbation theory (DFPT), and structural relaxations. Remarkably, when trained on data from 12 materials spanning various crystal structures and compositions, our model achieves mean absolute relative errors of 3% and 5% for Hubbard U and V parameters, respectively. By circumventing computationally expensive DFT or DFPT self-consistent protocols, our model significantly expedites the prediction of Hubbard parameters with negligible computational overhead, while approaching the accuracy of DFPT. Moreover, owing to its robust transferability, the model facilitates accelerated materials discovery and design via high-throughput calculations, with relevance for various technological applications.

Biofunctionalized magnetite nanoparticles offer unique multifunctional capabilities that can drive nanomedical innovations. Designing synthetic bioorganic coatings and controlling their molecular behavior is crucial for achieving superior performance. However, accurately describing the interactions between bio-inorganic nanosystem components requires reliable computational tools, with empirical force fields at their core. In this work, we integrate first-principles calculations with mainstream force fields to construct and simulate atomistic models of pristine and biofunctionalized magnetite nanoparticles with quantum mechanical accuracy. The practical implications of this approach are demonstrated through a case study of PEG (polyethylene glycol)-coated magnetite nanoparticles in physiological conditions, where we investigate how polymer chain length, in both heterogeneous and homogeneous coatings, impacts key functional properties in advanced nanosystem design. Our findings reveal that coating morphology controls polymer ordering, conformation, and polymer corona hydrogen bonding, highlighting the potential of this computational toolbox to advance next-generation magnetite-based nanosystems with enhanced performance in nanomedicine.

We conduct a systematic investigation of the role of Hubbard U corrections in electronic structure calculations of two-dimensional (2D) materials containing 3d transition metals. Specifically, we use density functional theory (DFT) with the PBE and PBE+U approximations to calculate the crystal structure, band gaps, and magnetic parameters of 638 monolayers. Based on a comprehensive comparison to experiments we first establish that the inclusion of the U correction worsens the accuracy for the lattice constants. Consequently, PBE structures are used for subsequent property evaluations. The band gaps show a significant dependence on U. In particular, for 134 (21%) of the materials the U parameter induces a metal-to-insulator transition. For the magnetic materials we calculate the magnetic moment, magnetic exchange coupling, and magnetic anisotropy parameters. In contrast to the band gaps, the size of the magnetic moments shows only weak dependence on U. Both the exchange energies and magnetic anisotropy parameters are systematically reduced by the U correction. On this basis we conclude that the Hubbard U correction will lead to lower predicted Curie temperatures in 2D materials. All the calculated properties are available in the Computational 2D Materials Database (C2DB).

Machine learning has advanced the rapid prediction of inorganic materials properties, yet data scarcity for specific properties and capturing thermodynamic stability remains challenging. We propose a framework utilizing a Graph Neural Network with composition-based and crystal structure-based architectures, combined with a transfer learning scheme. This approach accurately predicts energy-related properties (e.g., total energy, energy above the convex hull, energy band gap) and data-scarce mechanical properties (e.g., bulk and shear modulus). Our model incorporates four-body interactions, capturing periodicity and structural characteristics. It outperforms state-of-the-art models in 8 materials property regression tasks. Also, this model predicts local atomic environments and global structural features better than several models. Transfer learning addresses mechanical property data scarcity, while separate architecture analysis allows application to materials lacking crystal structure information. Our framework’s interpretability aids in understanding elemental contributions, enhancing material design and discovery. Continuous advancements promise further performance improvements, driving efficient and accurate materials property prediction.

It is well-known that exciton effects are determinant to understanding the optical absorption spectrum of low-dimensional materials. However, the role of excitons in nonlinear optical responses has been much less investigated at the experimental level. Additionally, computational methods to calculate nonlinear conductivities in real materials are still not widespread, particularly taking into account excitonic interactions. We present a methodology to calculate the excitonic second-order optical responses in 2D materials relying on: (i) ab initio tight-binding Hamiltonians obtained by Wannier interpolation and (ii) solving the Bethe-Salpeter equation with effective electron-hole interactions. Here, in particular, we explore the role of excitons in the shift current of monolayer materials. Focusing on MoS₂ and GeS monolayer systems, our results show that 2p-like excitons, which are dark in the linear response regime, yield a contribution to the photocurrent comparable to that of 1s-like excitons. Under radiation with intensity ~10⁴W/cm², the excitonic theory predicts in-gap photogalvanic currents of almost ~10 nA in sufficiently clean samples, which is typically one order of magnitude higher than the value predicted by independent-particle theory near the band edge.

Prolonged contact between a corrosive liquid and metal alloys can cause progressive dealloying. For one such process as liquid-metal dealloying (LMD), phase field models have been developed to understand the mechanisms leading to complex morphologies. However, the LMD governing equations in these models often involve coupled non-linear partial differential equations (PDE), which are challenging to solve numerically. In particular, numerical stiffness in the PDEs requires an extremely refined time step size (on the order of 10⁻¹²s or smaller). This computational bottleneck is especially problematic when running LMD simulation until a late time horizon is required. This motivates the development of surrogate models capable of leaping forward in time, by skipping several consecutive time steps at-once. In this paper, we propose a U-shaped adaptive Fourier neural operator (U-AFNO), a machine learning (ML) based model inspired by recent advances in neural operator learning. U-AFNO employs U-Nets for extracting and reconstructing local features within the physical fields, and passes the latent space through a vision transformer (ViT) implemented in the Fourier space (AFNO). We use U-AFNOs to learn the dynamics of mapping the field at a current time step into a later time step. We also identify global quantities of interest (QoI) describing the corrosion process (e.g., the deformation of the liquid-metal interface, lost metal, etc.) and show that our proposed U-AFNO model is able to accurately predict the field dynamics, in spite of the chaotic nature of LMD. Most notably, our model reproduces the key microstructure statistics and QoIs with a level of accuracy on par with the high-fidelity numerical solver, while achieving a significant 11, 200 × speed-up on a high-resolution grid when comparing the computational expense per time step. Finally, we also investigate the opportunity of using hybrid simulations, in which we alternate forward leaps in time using the U-AFNO with high-fidelity time stepping. We demonstrate that while advantageous for some surrogate model design choices, our proposed U-AFNO model in fully auto-regressive settings consistently outperforms hybrid schemes.

Metal halide perovskites (MHPs) exhibit unusual properties and complex dynamics. By combining ab initio time-dependent density functional theory, nonadiabatic molecular dynamics and machine learning, we advance quantum dynamics simulation to nanosecond timescale and demonstrate that large fluctuations of MHP defect energy levels extend light absorption to longer wavelengths and enable trapped charges to escape into bands. This allows low energy photons to contribute to photocurrent through energy up-conversion. Deep defect levels can become shallow transiently and vice versa, altering the traditional defect classification into shallow and deep. While defect levels fluctuate more in MHPs than traditional semiconductors, some levels, e.g., Pb interstitials, remain far from band edges, acting as charge recombination centers. Still, many defects deemed detrimental based on static structures, are in fact benign and can contribute to energy up-conversion. The extended light harvesting and energy up-conversion provide strategies for design of novel solar, optoelectronic, and quantum information devices.

Inspired by recent advancements in the field of single-atom magnets, particularly those involving rare-earth (RE) elements, we present a theoretical exploration employing DFT+U calculations to investigate the magnetic properties of selected 4f atoms, specifically Eu, Gd, and Ho, on a monolayer of the transition-metal dichalcogenide WSe₂ in the 1H-phase. This study comparatively examines RE with diverse 4f orbital fillings and valence chemistry, aiming to understand how different coverage densities atop WSe₂ affect magnetocrystalline anisotropy. We observe that RE lacking 5d occupation exhibit larger magnetic anisotropy energies at high densities, while those with outer 5d electrons show larger anisotropies in dilute configurations. Additionally, even half-filled 4f shell atoms with small orbital magnetic moments can generate substantial energy barriers for magnetization rotation due to prominent orbital hybridizations with WSe₂. Open 4f shell atoms further enhance anisotropy barriers through spin-orbit coupling effects. These aspects are crucial for realizing stable magnetic information units experimentally.

We present a multi-objective Bayesian active learning strategy, which greatly accelerates the discovery of super high-strength and high-ductility lead-free solder alloys. The active learning strategy demonstrates that a machine learning model will have high generalizability if experimental data uncertainty is included, which greatly improves the model prediction or the material design accuracy. The feature-point-start forward method in multi-objective optimization adopts two Gaussian process regression (GPR) models, one for strength and one for elongation, and their outputs build up the acquisition-function-modified objective space of strength and elongation. Then, Bayesian sampling is applied to design the next experiments by balancing exploitation and exploration. Seven multi-objective active learning iterations discovered two novel super high-strength and high-ductility lead-free solder alloys. After that, various material characterizations were conducted on the two novel solder alloys, and the results exhibited their high performances in melting properties, wettability, electrical conductivity, and shear strength of the solder joint and explored the mechanism of high strength and high ductility of the alloys. The present work systematically analyzes the important role of experimental uncertainty in machine learning, especially in the global optimization for material design, which demands high generalizability of predictions.