density functional theory
PulseAugur coverage of density functional theory — every cluster mentioning density functional theory across labs, papers, and developer communities, ranked by signal.
10 day(s) with sentiment data
-
Tessera runtime boosts AI-driven molecular dynamics simulations
Researchers have developed Tessera, a distributed runtime system designed to enhance the efficiency of large-scale molecular dynamics simulations. Tessera addresses communication bottlenecks by optimizing workload shape…
-
Tessera runtime enables molecular dynamics simulations at tens of millions of atoms
Researchers have developed Tessera, a distributed runtime system designed to significantly enhance the scale and efficiency of molecular dynamics simulations. By employing an equivariant message-passing Transformer arch…
-
New methods enhance reliability and safety of machine-learned interatomic potentials
Two new research papers explore methods for improving the reliability and safety of machine-learned interatomic potentials (MLIPs). The first paper proposes pre-training uncertainty estimators on large datasets to provi…
-
New training signal improves machine-learned Hamiltonian models for DFT
Researchers have developed a new training signal called cumulative local density of states (cumulative LDOS) to improve machine-learned Hamiltonian models used in density functional theory (DFT). This method addresses l…
-
AI distillation creates faster MLIPs for battery electrolyte simulation
Researchers have developed a knowledge distillation technique to create faster, more accurate machine learning interatomic potentials (MLIPs) for simulating liquid electrolytes. These "student" models, distilled from la…
-
New AI framework UniTB accelerates materials science discovery
Researchers have developed UniTB, a universal E(3) equivariant Hamiltonian representation framework that integrates physical priors from density-functional tight-binding theory with deep learning. This framework aims to…
-
New dataset SoLiD26 advances MLIPs for solid-liquid interface simulations
Researchers have introduced SoLiD26, a new dataset designed to train machine-learned interatomic potentials (MLIPs) for simulating solid-liquid interfaces. This dataset comprises 15.4 million first-principles structures…
-
Machine learning models phase equilibria of ammonia and hydrogen in planets
Researchers have utilized machine learning-based molecular dynamics simulations, trained on density functional theory calculations, to map the phase equilibria of ammonia and hydrogen under extreme pressures and tempera…
-
LatticeMind framework improves reliability of AI-assisted scientific computing
Researchers have developed LatticeMind, a novel agentic framework designed to enhance the reliability of complex scientific computing workflows, particularly in density functional theory (DFT). This framework uses natur…
-
New Conformal Risk Control method optimizes DFT selection for MLIPs
Researchers have developed a novel approach using Conformal Risk Control (CRC) to optimize the selection of configurations for density functional theory (DFT) calculations in materials discovery. This method aims to red…
-
Machine learning potentials can exceed training accuracy, researchers propose
A new perspective paper argues that machine learning potentials (MLPs) can surpass the accuracy of the methods used to train them. The authors propose that by generalizing across different quantum-chemical methods, MLPs…
-
Study reveals grokking in MLPs sharpens pre-existing algorithms
A new study published on arXiv investigates the phenomenon of 'grokking' in neural networks, specifically focusing on ReLU MLPs trained on modular arithmetic. Researchers found that instead of sinusoidal weight distribu…
-
New MLIP methods enhance atomistic material modeling scalability and accuracy
Researchers have developed new methods to improve the accuracy and scalability of machine learning interatomic potentials (MLIPs) for materials modeling. One approach, DPA4C-LR, analytically separates long-range electro…
-
Periodic Labs builds AI scientists for physical world experimentation
Periodic Labs is pioneering a new frontier in AI by developing autonomous laboratories capable of conducting physical experiments. Their vision of "synthesis superintelligence" aims to accelerate scientific discovery by…
-
CNet framework enables complex-valued deep learning with Wirtinger derivatives
Researchers have developed CNet, a C++/CUDA framework designed for training deep complex-valued neural networks (CVNNs) and optimizing complex-valued functions using Wirtinger derivatives. This framework adopts a physic…
-
Machine learning model predicts molecular spectra using electron density for higher accuracy
Researchers have developed a machine-learning model that predicts molecular absorption spectra more accurately by using the ground-state electron density as input, rather than molecular geometry. This density-based appr…
-
New OrbFlow model advances electron density prediction in quantum chemistry
Researchers have developed OrbFlow, a new SE(3)-equivariant generative model designed to predict electron densities more efficiently and accurately. This model utilizes flow matching to predict Gaussian-type orbital coe…
-
Language Models Rewrite Molecular Optimizer for Faster Calculations
Researchers have developed AutoSella, a novel molecular geometry optimizer that utilizes language models to rewrite existing optimization algorithms. This agent-based approach aims to reduce the computational cost of fo…
-
New GS-DFT method optimizes molecular orbitals for quantum mechanics calculations
Researchers have developed a new method called Gaussian Splatting for Density Functional Theory (GS-DFT) that optimizes molecular orbitals using a cloud of Gaussians. This approach aims to improve the accuracy and compu…
-
Machine learning models achieve 92% accuracy in classifying magnetic order in materials
Researchers have developed machine-learning classifiers capable of identifying magnetic order in materials with over 92% accuracy. These models, trained on experimental data and utilizing descriptors from the Materials …