Recent work in scientific machine learning has developed so-called physics-informed neural network (PINN) models. The typical approach is to incorporate physical domain knowledge as soft constraints on an empirical loss function and use existing methodologies train the model. We demonstrate that, while PINN can learn good models for relatively trivial problems, they easily fail relevant phenomena even slightly more complex problems. In particular, we analyze several distinct situations of...

The vastness of the materials design space makes it impractical to explore using traditional brute-force methods, particularly in reticular chemistry. However, machine learning has shown promise expediting and guiding design. Despite numerous successful applications materials, progress field stagnated, possibly because digital chemistry is more an art than a science its limited accessibility inexperienced researchers. To address this issue, we present mofdscribe, software ecosystem tailored...

Machine learning has emerged as an attractive alternative to experiments and simulations for predicting material properties. Usually, such approach relies on specific domain knowledge feature design: each target requires careful selection of features that expert recognizes important the task. The major drawback this is computation only a few structural been implemented so far, it difficult tell priori which are particular application. latter problem empirically observed predictors guest...

Abstract Machine learning has emerged as a powerful approach in materials discovery. Its major challenge is selecting features that create interpretable representations of materials, useful across multiple prediction tasks. We introduce an end-to-end machine model automatically generates descriptors capture complex representation material’s structure and chemistry. This builds on computational topology techniques (namely, persistent homology) word embeddings from natural language processing....

Abstract Dynamical systems that evolve continuously over time are ubiquitous throughout science and engineering. Machine learning (ML) provides data-driven approaches to model predict the dynamics of such systems. A core issue with this approach is ML models typically trained on discrete data, using methodologies not aware underlying continuity properties. This results in often do capture any continuous dynamics—either system interest, or indeed related system. To address challenge, we...

Two-dimensional (2D) materials derived from van der Waals (vdW)-bonded layered crystals have been the subject of considerable research focus, but their one-dimensional (1D) analogues received less attention. These bulk consist covalently bonded multiatom atomic chains with weak bonds between adjacent chains. Using density-functional-theory-based methods, we find binding energies several 1D families to be within typical exfoliation ranges possible for 2D materials. In addition, compute...

The foundation model (FM) paradigm is transforming Machine Learning Force Fields (MLFFs), leveraging general-purpose representations and scalable training to perform a variety of computational chemistry tasks. Although MLFF FMs have begun close the accuracy gap relative first-principles methods, there still strong need for faster inference speed. Additionally, while research increasingly focused on models which transfer across chemical space, practitioners typically only study small subset...

New parametrizations for semiempirical density functional tight binding (DFTB) theory have been developed by the numerical optimization of adjustable parameters to minimize errors in atomization energy and interatomic forces with respect ab initio calculated data. Initial guesses radial dependences Slater-Koster bond integrals overlap were obtained from minimum basis calculations. The pair potentials represented simple analytic functions. these functions optimized simulated annealing...

The "lone" 6s electron pair often plays a key role in determining the structure and physical properties of compounds containing sixth-row elements their lower oxidation states: Tl+, Pb2+, Bi3+ with [Xe]4f145d106s2 electronic configuration. lone pairs on these ions are associated reduced structural symmetries, including ferroelectric instabilities other important phenomena. Here we consider isoelectronic auride Au– ion Ab initio density functional theory methods employed to probe effect...

The electronic structures of four semiconductor compounds BaCu2S2, BaCu2Se2, BaAg2S2, and BaAg2Se2 are studied by density functional theory using both semi-local hybrid functionals. ionization energies electron affinities were determined aligning the states with vacuum level calculating electrostatic profile within a supercell slab model. energy affinity calculated Heyd–Scuseria–Ernzerhof functionals range from 4.5 eV to 5.4 3.1 3.4 eV, respectively. replacement Cu Ag slightly increases...

Optimization, a key tool in machine learning and statistics, relies on regularization to reduce overfitting. Traditional methods control norm of the solution ensure its smoothness. Recently, topological have emerged as way provide more precise expressive over solution, relying persistent homology quantify roughness. All such existing techniques back-propagate gradients through persistence diagram, which is summary features function. Their downside that they information only at critical...

Developing equivariant neural networks for the E(3) group plays an important role in modeling 3D data across real-world applications. Enforcing this equivariance primarily involves tensor products of irreducible representations (irreps). However, computational complexity such operations increases significantly as higher-order tensors are used. In work, we propose a systematic approach to substantially accelerate computation irreps. We mathematically connect commonly used Clebsch-Gordan...

Imposing known physical constraints, such as conservation laws, during neural network training introduces an inductive bias that can improve accuracy, reliability, convergence, and data efficiency for modeling dynamics. While constraints be softly imposed via loss function penalties, recent advancements in differentiable physics optimization performance by incorporating PDE-constrained individual layers networks. This enables a stricter adherence to constraints. However, imposing hard...

Structure-Based Drug Design (SBDD) focuses on generating valid ligands that strongly and specifically bind to a designated protein pocket. Several methods use machine learning for SBDD generate these in 3D space, conditioned the structure of desired Recently, diffusion models have shown success here by modeling underlying distributions atomic positions types. While are effective considering structural details pocket, they often fail explicitly consider binding affinity. Binding affinity...

Scaling has been critical in improving model performance and generalization machine learning. It involves how a model's changes with increases size or input data, as well efficiently computational resources are utilized to support this growth. Despite successes other areas, the study of scaling Neural Network Interatomic Potentials (NNIPs) remains limited. NNIPs act surrogate models for ab initio quantum mechanical calculations. The dominant paradigm here is incorporate many physical domain...

Molecular conformer generation (MCG) is an important task in cheminformatics and drug discovery. The ability to efficiently generate low-energy 3D structures can avoid expensive quantum mechanical simulations, leading accelerated virtual screenings enhanced structural exploration. Several generative models have been developed for MCG, but many struggle consistently produce high-quality conformers meaningful downstream applications. To address these issues, we introduce CoarsenConf, which...

Understanding protein structure-function relationships is a key challenge in computational biology, with applications across the biotechnology and pharmaceutical industries. While it known that structure directly impacts function, many functional prediction tasks use only sequence. In this work, we isolate to make annotations for proteins Protein Data Bank order study expressiveness of different structure-based schemes. We present PersGNN - an end-to-end trainable deep learning model...

In principle, a nearly endless number of unique van der Waals heterostructures can be created through the vertical stacking two-dimensional (2D) materials, resulting in unprecedented potential for material design. However, this widely employed synthetic method generating is slow, imprecise, and prone to introducing interlayer contaminants when compared with synthesis methods that are scalable industrially relevant scales. Herein, we study properties new class layered bulk inorganic materials...

Neural network interatomic potentials (NNIPs) are an attractive alternative to ab-initio methods for molecular dynamics (MD) simulations. However, they can produce unstable simulations which sample unphysical states, limiting their usefulness modeling phenomena occurring over longer timescales. To address these challenges, we present Stability-Aware Boltzmann Estimator (StABlE) Training, a multi-modal training procedure combines conventional supervised from quantum-mechanical energies and...

The threat of geomagnetic disturbances (GMDs) to the reliable operation bulk energy system has spurred development effective strategies for mitigating their impacts. One such approach involves placing transformer neutral blocking devices, which interrupt path geomagnetically induced currents (GICs) limit impact. high cost these devices and sparsity transformers that experience GICs during GMD events, however, calls a sparse placement strategy computational cost. To address this challenge, we...