SpookyNet: Learning Force Fields with Electronic Degrees of Freedom and Nonlocal Effects

Abstract

Machine-learned force fields (ML-FFs) combine the accuracy of ab initio methods with the efficiency of conventional force fields. However, current ML-FFs typically ignore electronic degrees of freedom, such as the total charge or spin state, and assume chemical locality, which is problematic when molecules have inconsistent electronic states, or when nonlocal effects play a significant role. This work introduces SpookyNet, a deep neural network for constructing ML-FFs with explicit treatment of electronic degrees of freedom and quantum nonlocality. Chemically meaningful inductive biases and analytical corrections built into the network architecture allow it to properly model physical limits. SpookyNet improves upon the current state-of-the-art (or achieves similar performance) on popular quantum chemistry data sets. Notably, it is able to generalize across chemical and conformational space and can leverage the learned chemical insights, e.g. by predicting unknown spin states, thus helping to close a further important remaining gap for today's machine learning models in quantum chemistry.

References in corpus (14)

• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• ANI-1: An extensible neural network potential with DFT accuracy at force field computational cost
• On the Convergence of Adam and Beyond
• A Spectral Analysis Method for Automated Generation of Quantum-Accurate Interatomic Potentials
• Unmasking Clever Hans Predictors and Assessing What Machines Really Learn
• Machine learning for molecular simulation
• Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems
• Machine Learning Unifies the Modelling of Materials and Molecules
• Searching for Activation Functions
• Equivariant message passing for the prediction of tensorial properties and molecular spectra
• The Expressive Power of Neural Networks: A View from the Width
• Charge separation in donor-C60 complexes with real-time Green's functions: The importance of nonlocal correlations
• Large yet bounded: Spin gap ranges in carbenes
• Solving the electronic Schrödinger equation for multiple nuclear geometries with weight-sharing deep neural networks

Cited by in corpus (62)

• Deep-Learning Density Functional Theory Hamiltonian for Efficient ab initio Electronic-Structure Calculation
• Neural Network Potentials for Chemistry: Concepts, Applications and Prospects
• Machine Learning Coarse-Grained Potentials of Protein Thermodynamics
• General framework for E(3)-equivariant neural network representation of density functional theory Hamiltonian
• Toward Explainable AI for Regression Models
• Advances of Machine Learning in Materials Science: Ideas and Techniques
• Excited state, non-adiabatic dynamics of large photoswitchable molecules using a chemically transferable machine learning potential
• SchNetPack 2.0: A neural network toolbox for atomistic machine learning
• Transferable E(3) equivariant parameterization for Hamiltonian of molecules and solids
• BIGDML: Towards Exact Machine Learning Force Fields for Materials
• Universal Machine Learning for the Response of Atomistic Systems to External Fields
• The Evolution of Machine Learning Potentials for Molecules, Reactions and Materials
• Graph Neural Networks Accelerated Molecular Dynamics
• Deep-learning electronic-structure calculation of magnetic superstructures
• Deep-learning density functional perturbation theory
• TorchMD-Net 2.0: Fast Neural Network Potentials for Molecular Simulations
• Force-Field-Enhanced Neural Network Interactions: from Local Equivariant Embedding to Atom-in-Molecule properties and long-range effects
• Lifelong Machine Learning Potentials
• Molecular dynamics simulation with finite electric fields using Perturbed Neural Network Potentials
• Towards Linearly Scaling and Chemically Accurate Global Machine Learning Force Fields for Large Molecules
• On the design space between molecular mechanics and machine learning force fields
• Neural network with optimal neuron activation functions based on additive Gaussian process regression
• Machine Learning and Data-Driven Methods in Computational Surface and Interface Science
• Heat flux for semi-local machine-learning potentials
• Modeling electronic response properties with an explicit-electron machine learning potential
• Lightweight and Effective Tensor Sensitivity for Atomistic Neural Networks
• SE(3)-equivariant prediction of molecular wavefunctions and electronic densities
• Reinforcement learning for traversing chemical structure space: Optimizing transition states and minimum energy paths of molecules
• Fast and flexible long-range models for atomistic machine learning
• 3DReact: Geometric deep learning for chemical reactions
• Neural Network Potential with Multi-Resolution Approach Enables Accurate Prediction of Reaction Free Energies in Solution
• Surrogate- and invariance-boosted contrastive learning for data-scarce applications in science
• Machine-Learning Interatomic Potentials for Long-Range Systems
• Machine Learning for Exploring Small Polaron Configurational Space
• Hybrid Classical/Machine-Learning Force Fields for the Accurate Description of Molecular Condensed-Phase Systems
• Accurate quantum Monte Carlo forces for machine-learned force fields: Ethanol as a benchmark
• Stress and heat flux via automatic differentiation
• FeNNol: an Efficient and Flexible Library for Building Force-field-enhanced Neural Network Potentials
• Artificial Intelligence for Science in Quantum, Atomistic, and Continuum Systems
• Algorithmic Differentiation for Automated Modeling of Machine Learned Force Fields
• Super-resolution in Molecular Dynamics Trajectory Reconstruction with Bi-Directional Neural Networks
• How Graph Neural Network Interatomic Potentials Extrapolate: Role of the Message-Passing Algorithm
• Molecular-orbital-based Machine Learning for Open-shell and Multi-reference Systems with Kernel Addition Gaussian Process Regression
• Towards Symbolic XAI -- Explanation Through Human Understandable Logical Relationships Between Features
• Efficient force field and energy emulation through partition of permutationally equivalent atoms
• Equivariant Machine Learning Interatomic Potentials with Global Charge Redistribution
• Efficient Generation of Stable Linear Machine-Learning Force Fields with Uncertainty-Aware Active Learning
• Scaling Molecular Dynamics with ab initio Accuracy to 149 Nanoseconds per Day
• Molecular relaxation by reverse diffusion with time step prediction
• Atomistic Simulations of Oxide-Water Interfaces using Machine Learning Potentials
• A Data-Driven Approach to Coarse-Graining Simple Liquids in Confinement
• Probing the Temporal Response of Liquid Water to a THz Pump Pulse Using Machine Learning-Accelerated Non-Equilibrium Molecular Dynamics
• Machine-Learned Potentials for Solvation Modeling
• FastCHGNet: Training one Universal Interatomic Potential to 1.5 Hours with 32 GPUs
• Orbital Mixer: Using Atomic Orbital Features for Basis Dependent Prediction of Molecular Wavefunctions
• Reconstructing Kernel-based Machine Learning Force Fields with Super-linear Convergence
• Active Learning of a Neural Network Potential for Large-Scale Atomistic Simulations of Polymer Electrolyte Membranes
• Detect the Interactions that Matter in Matter: Geometric Attention for Many-Body Systems
• Faster Molecular Dynamics with Neural Network Potentials via Distilled Multiple Time-Stepping and Non-Conservative Forces
• Neural Polarization: Toward Electron Density for Molecules by Extending Equivariant Networks
• Graph Convolutional Neural Networks for (QM)ML/MM Molecular Dynamics Simulations
• REANN: A PyTorch-based End-to-End Multi-functional Deep Neural Network Package for Molecular, Reactive and Periodic Systems