Complex reaction processes in combustion unraveled by neural network-based molecular dynamics simulation

Nature Communications, Journal Year: 2020, Volume and Issue: 11(1)

Published: Nov. 11, 2020

Abstract Combustion is a complex chemical system which involves thousands of reactions and generates hundreds molecular species radicals during the process. In this work, neural network-based dynamics (MD) simulation carried out to simulate benchmark combustion methane. During MD simulation, detailed reaction processes leading creation specific including various intermediate products are intimately revealed characterized. Overall, total 798 different were recorded some new pathways discovered. We believe that present work heralds dawn era in reactive can be practically applied simulating important systems at ab initio level, provides atomic-level understanding as well discovery an unprecedented level detail beyond what laboratory experiments could accomplish.

Language: Английский

---

SpookyNet: Learning force fields with electronic degrees of freedom and nonlocal effects

Nature Communications, Journal Year: 2021, Volume and Issue: 12(1)

Published: Dec. 14, 2021

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

Language: Английский

---

Machine Learning in Chemical Engineering: Strengths, Weaknesses, Opportunities, and Threats

Engineering, Journal Year: 2021, Volume and Issue: 7(9), P. 1201 - 1211

Published: July 29, 2021

Chemical engineers rely on models for design, research, and daily decision-making, often with potentially large financial safety implications. Previous efforts a few decades ago to combine artificial intelligence chemical engineering modeling were unable fulfill the expectations. In last five years, increasing availability of data computational resources has led resurgence in machine learning-based research. Many recent have facilitated roll-out learning techniques research field by developing databases, benchmarks, representations applications new frameworks. Machine significant advantages over traditional techniques, including flexibility, accuracy, execution speed. These strengths also come weaknesses, such as lack interpretability these black-box models. The greatest opportunities involve using time-limited real-time optimization planning that require high accuracy can build self-learning ability recognize patterns, learn from data, become more intelligent time. threat today is inappropriate use because most had limited training computer science analysis. Nevertheless, will definitely trustworthy element toolbox engineers.

Language: Английский

---

Machine Learning Methods for Small Data Challenges in Molecular Science

Chemical Reviews, Journal Year: 2023, Volume and Issue: 123(13), P. 8736 - 8780

Published: June 29, 2023

Small data are often used in scientific and engineering research due to the presence of various constraints, such as time, cost, ethics, privacy, security, technical limitations acquisition. However, big have been focus for past decade, small their challenges received little attention, even though they technically more severe machine learning (ML) deep (DL) studies. Overall, challenge is compounded by issues, diversity, imputation, noise, imbalance, high-dimensionality. Fortunately, current era characterized technological breakthroughs ML, DL, artificial intelligence (AI), which enable data-driven discovery, many advanced ML DL technologies developed inadvertently provided solutions problems. As a result, significant progress has made decade. In this review, we summarize analyze several emerging potential molecular science, including chemical biological sciences. We review both basic algorithms, linear regression, logistic regression (LR),

Language: Английский

---

GPUMD: A package for constructing accurate machine-learned potentials and performing highly efficient atomistic simulations

The Journal of Chemical Physics, Journal Year: 2022, Volume and Issue: 157(11)

Published: Aug. 24, 2022

We present our latest advancements of machine-learned potentials (MLPs) based on the neuroevolution potential (NEP) framework introduced in [Fan et al., Phys. Rev. B 104, 104309 (2021)] and their implementation open-source package GPUMD. increase accuracy NEP models both by improving radial functions atomic-environment descriptor using a linear combination Chebyshev basis extending angular with some four-body five-body contributions as atomic cluster expansion approach. also detail efficient approach graphics processing units well workflow for construction models, we demonstrate application large-scale atomistic simulations. By comparing to state-of-the-art MLPs, show that not only achieves above-average but is far more computationally efficient. These results GPUMD promising tool solving challenging problems requiring highly accurate, To enable MLPs minimal training set, propose an active-learning scheme latent space pre-trained model. Finally, introduce three separate Python packages, GPYUMD, CALORINE, PYNEP, which integration into workflows.

Language: Английский

---

Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems

Chemical Reviews, Journal Year: 2021, Volume and Issue: 121(16), P. 9816 - 9872

Published: July 7, 2021

Machine learning models are poised to make a transformative impact on chemical sciences by dramatically accelerating computational algorithms and amplifying insights available from chemistry methods. However, achieving this requires confluence coaction of expertise in computer science physical sciences. This Review is written for new experienced researchers working at the intersection both fields. We first provide concise tutorials machine methods, showing how involving can be achieved. follow with critical review noteworthy applications that demonstrate used together insightful (and useful) predictions molecular materials modeling, retrosyntheses, catalysis, drug design.

Language: Английский

---

TorchMD: A Deep Learning Framework for Molecular Simulations

Journal of Chemical Theory and Computation, Journal Year: 2021, Volume and Issue: 17(4), P. 2355 - 2363

Published: March 17, 2021

Molecular dynamics simulations provide a mechanistic description of molecules by relying on empirical potentials. The quality and transferability such potentials can be improved leveraging data-driven models derived with machine learning approaches. Here, we present TorchMD, framework for molecular mixed classical All force computations including bond, angle, dihedral, Lennard-Jones, Coulomb interactions are expressed as PyTorch arrays operations. Moreover, TorchMD enables simulating neural network We validate it using standard Amber all-atom simulations, an ab initio potential, performing end-to-end training, finally coarse-grained model protein folding. believe that provides useful tool set to support Code data freely available at github.com/torchmd.

Language: Английский

---

Unified representation of molecules and crystals for machine learning

Machine Learning Science and Technology, Journal Year: 2022, Volume and Issue: 3(4), P. 045017 - 045017

Published: Nov. 3, 2022

Accurate simulations of atomistic systems from first principles are limited by computational cost. In high-throughput settings, machine learning can reduce these costs significantly accurately interpolating between reference calculations. For this, kernel approaches crucially require a representation that accommodates arbitrary systems. We introduce many-body tensor is invariant to translations, rotations, and nuclear permutations same elements, unique, differentiable, represent molecules crystals, fast compute. Empirical evidence for competitive energy force prediction errors presented changes in molecular structure, crystal chemistry, dynamics using regression symmetric gradient-domain as models. Applicability demonstrated phase diagrams Pt-group/transition-metal binary

Language: Английский

---

Development of Multimodal Machine Learning Potentials: Toward a Physics-Aware Artificial Intelligence

Accounts of Chemical Research, Journal Year: 2021, Volume and Issue: 54(7), P. 1575 - 1585

Published: March 13, 2021

ConspectusMachine learning interatomic potentials (MLIPs) are widely used for describing molecular energy and continue bridging the speed accuracy gap between quantum mechanical (QM) classical approaches like force fields. In this Account, we focus on out-of-the-box to developing transferable MLIPs diverse chemical tasks. First, introduce "Accurate Neural Network engine Molecular Energies," ANAKIN-ME, method (or ANI short). The model utilizes Justin Smith Symmetry Functions (JSSFs) realizes training vast data sets. set of several orders magnitude larger than before has become key factor knowledge transferability flexibility MLIPs. As quantity, quality, types interactions included in will dictate MLIPs, task proper selection could be assisted with advanced methods active (AL), transfer (TL), multitask (MTL).Next, describe AIMNet "Atoms-in-Molecules Network" that was inspired by theory atoms molecules. architecture lifts multiple limitations It encodes long-range learnable representations elements. We also discuss AIMNet-ME expands applicability domain from neutral molecules toward open-shell systems. encompasses a dependence potential charge spin. brings ML physical models one step closer, ensuring correct behavior over total charge.We finally perhaps simplest possible physics-aware model, which combines extended Hückel method. ML-EHM, "Hierarchically Interacting Particle Network," HIP-NN generates molecule- environment-dependent Hamiltonian elements αμμ K‡. test example, show how contrast traditional theory, ML-EHM correctly describes orbital crossing bond rotations. Hence it learns underlying physics, highlighting inclusion constraints symmetries significantly improve generalization.

Language: Английский

---

Choosing the right molecular machine learning potential

Chemical Science, Journal Year: 2021, Volume and Issue: 12(43), P. 14396 - 14413

Published: Jan. 1, 2021

Quantum-chemistry simulations based on potential energy surfaces of molecules provide invaluable insight into the physicochemical processes at atomistic level and yield such important observables as reaction rates spectra. Machine learning potentials promise to significantly reduce computational cost hence enable otherwise unfeasible simulations. However, surging number begs question which one choose or whether we still need develop yet another one. Here, address this by evaluating performance popular machine in terms accuracy cost. In addition, deliver structured information for non-specialists guide them through maze acronyms, recognize each potential's main features, judge what they could expect from

Language: Английский

---