Abstract:Machine learning (ML) approximations to density functional theory (DFT) potential energy surfaces (PESs) are showing great promise for reducing the computational cost of accurate molecular simulations, but at present, they are not applicable to varying electronic states, and in particular, they are not well suited for molecular systems in which the local electronic structure is sensitive to the medium to long-range electronic environment. With this issue as the focal point, we present a new machine learning approach called "BpopNN" for obtaining efficient approximations to DFT PESs. Conceptually, the methodology is based on approaching the true DFT energy as a function of electron populations on atoms; in practice, this is realized with available density functionals and constrained DFT (CDFT). The new approach creates approximations to this function with neural networks. These approximations thereby incorporate electronic information naturally into a ML approach, and optimizing the model energy with respect to populations allows the electronic terms to self-consistently adapt to the environment, as in DFT. We confirm the effectiveness of this approach with a variety of calculations on LinHn clusters.The Supporting Information is available free of charge at <a class="ext-link" href="/doi/10.1021/acs.jctc.0c00217?goto=supporting-info">https://pubs.acs.org/doi/10.1021/acs.jctc.0c00217</a>.Test-set results for the other BpopNN variants; details for training the GAP models with QUIP; test-set results for the GAP model variants (<a class="ext-link" href="/doi/suppl/10.1021/acs.jctc.0c00217/suppl_file/ct0c00217_si_001.pdf">PDF</a>)All training-set data, including CDFT energies, charges, population gradients, and nuclear-position forces; Geometries used for all training and test data sets (<a class="ext-link" href="/doi/suppl/10.1021/acs.jctc.0c00217/suppl_file/ct0c00217_si_002.zip">ZIP</a>)This article has not yet been cited by other publications.

Improving the accuracy of low level quantum chemical calculation for absorption energies: the genetic algorithm and neural network approach.

Correcting the Systematic Error of the Density Functional Theory Calculation: the Alternate Combination Approach of Genetic Algorithm and Neural Network

Back-Propagation Network Improved by Conjugate Gradient Based on Genetic Algorithm in Qsar Study on Endocrine Disrupting Chemicals

Neural network correction for heats of formation with a larger experimental training set and new descriptors

Linear regression correction to first principle theoretical calculations – Improved descriptors and enlarged training set

Generalized Regression Neural Network Based Quantitative Structure-Property Relationship for the Prediction of Absorption Energy

Improving the Accuracy of Density Functional Theory (DFT) Calculation for Homolysis Bond Dissociation Energies of Y-NO Bond: Generalized Regression Neural Network Based on Grey Relational Analysis and Principal Component Analysis

Improving the accuracy of predicting protein-ligand binding-free energy with semiempirical quantum chemistry charge.

Chemical accuracy for ligand-receptor binding Gibbs energies through multi-level SQM/QM calculations

Energy refinement and analysis of structures in the QM9 database via a highly accurate quantum chemical method

Accurately estimating activation energies by leveraging neural network methods and a large dataset

Combining Ab Initio and Machine Learning Method to Improve the Prediction of Diatomic Vibrational Energies

Beyond independent error assumptions in large GNN atomistic models

Calculate electronic excited states using neural networks with effective core potential

Chemical accuracy for ligand-receptor binding Gibbs energies through multi-level SQM/QM calculations

Insights and Challenges in Correcting Force Field Based Solvation Free Energies Using A Neural Network Potential

Incorporating Electronic Information into Machine Learning Potential Energy Surfaces via Approaching the Ground-State Electronic Energy as a Function of Atom-Based Electronic Populations

Machine learning prediction of accurate atomization energies of organic molecules from low-fidelity quantum chemical calculations

Artificial intelligence-enhanced quantum chemical method with broad applicability

Improving the B 3 LYP bond energies by using the X 1 method

Improving the B3lyp Bond Energies by Using the X1 Method