Data Quality, Data Sampling and Data Fitting: A Tutorial Guide for Constructing Full-dimensional Accurate Potential Energy Surfaces (PESs) of Small Molecular Systems

Li, Jun; Liu, Yang

doi:10.26434/chemrxiv-2022-1jpvc

“…For the dynamic characteristics of the PES, the dynamic outcome and energy conservation of the trajectories, initialized with various sampling in the ro-vibrational and translational motions, are checked to explore and improve the PES. As has been discussed in our previous work, the PES was improved gradually with more and more points added …”

Section: Methods and Calculation Detailsmentioning

confidence: 83%

“…According to the strategy given above and our experience in developing PESs, ,, a total of 68234 configurations are sampled and calculated at the level of UCCSD(T)-F12a/AVTZ for the Cl + SiH 4 system. The sampling covers configurations with the separated distance between reactants or products ranging 1–15 Å.…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Deciphering Dynamics of the Cl + SiH₄ → H + SiH₃Cl Reaction on a Machine Learning Made Globally Accurate Full-Dimensional Potential Energy Surface

Xu

¹

,

Li

²

2022

Self Cite

View full text Add to dashboard Cite

Chemical reaction dynamics needs the joint effort from both experiment and theory, and theory is useful to rationalize the experimental results by offering intimate details of chemical reaction dynamics and to explore new reaction pathways. With the aid of machine learning, we develop here an accurate fulldimensional potential energy surface (PES) for the reaction between Cl + SiH 4 . This PES can describe well the hydrogen abstraction channel to HCl + SiH 3 . It can also give a good description for the hydrogen substitution channel to H + SiH 3 Cl, which is the focus of the current study and has never been reported by theory. The dynamics of this substitution channel is revealed in detail by calculating ample quasi-classical trajectories (QCTs) on the new PES. The computed product angular distributions are in good agreement with the only crossed molecular beam experiment. Both theory and experiment suggest that this channel takes place mainly via the typical S N 2 inversion mechanism. Theory reveals that there also exists a novel torsion mechanism for the substitution channel. Two dynamic mechanisms are analyzed in detail. The present detailed theoretical dynamics study sheds insightful and novel understanding for this fundamentally important chemical reaction.

show abstract

“…To develop a high-fidelity full-dimensional PES, the first key is to choose a suitable electronic structure method, which should be sufficiently accurate, time-consuming affordably, and provide reliable descriptions for all dynamically relevant regions, including reactants, products, transition states and intermediates, and regions far from the equilibrium of these stationary points. 80,83 The UCCSD(T)-F12a/AVTZ method has succeeded in developing a series of PESs for reactive and non-reactive systems, 64,66,67,84,85 and was suggested for effective construction of full-dimensional accurate PESs. 86 The energies, geometries and harmonic frequencies of the stationary points along R1 and R2 were determined at the level of UCCSD(T)-F12a/AVTZ as implemented in the Molpro program package.…”

Section: Electronic Structure Calculationmentioning

confidence: 99%

Neural Network Based ∆-Machine learning approach efficiently brings the DFT potential energy surface to the CCSD(T) quality: a case for the OH + CH3OH reaction

Song

¹

,

Li

²

2022

Preprint

View full text Add to dashboard Cite

The recently proposed permutationally invariant polynomial-neural network (PIP-NN) based ∆-machine learning (∆-ML) approach (PIP-NN ∆-ML, J. Phys. Chem. Lett. 2022, 13, 4729) is a flexible, general, and highly cost-efficient method to develop full dimensional accurate potential energy surface (PES). Only a small portion of points, which can be actively selected from the low-level (often DFT) points, with high-level energies are needed to bring a low-level PES to a high-level of quality. The hydrogen abstraction reaction between methanol and hydroxyl radical, OH + CH3OH, has been studied by theories and experiments for a long time due to its great importance in combustion, atmospheric and interstellar chemistry. However, it is not trivial to develop the full dimensional accurate PES for it. In this work, the PIP-NN ∆-ML method is successfully applied to the title reaction. The DFT PES was fitted by using 140192 points. Only 5% of the DFT dataset was needed to be calculated at the level of UCCSD(T)-F12a/AVTZ, and to improve the DFT PES to the target high-level, UCCSD(T)-F12a/AVTZ. More than 92% of the original unaffordable calculation cost were saved. The kinetics, including rate coefficients and branching ratios, were then studied by performing quasi-classical trajectory calculations on this newly fitted PES for the title reaction.

show abstract

“…To develop a high-fidelity full-dimensional PES, the first key is to choose a suitable electronic structure method, which should be sufficiently accurate, time-consuming affordably, and provide reliable descriptions for all dynamically relevant regions, including reactants, products, transition states and intermediates, and regions far from the equilibrium of these stationary points. 79,82 The UCCSD(T)-F12a/AVTZ method has succeeded in developing a series of PESs for reactive and non-reactive systems, 63,65,66,83,84 and was suggested for effective construction of full-dimensional accurate PESs. 85 The energies, geometries and harmonic frequencies of the stationary points along R1 and R2 were determined at the level of UCCSD(T)-F12a/AVTZ as implemented in the Molpro program package.…”

Section: Electronic Structure Calculationmentioning

confidence: 99%

Neural Network Based ∆-Machine learning approach efficiently brings the DFT potential energy surface to the CCSD(T) quality: a case for the OH + CH3OH reaction

Song

¹

,

Li

²

2022

Preprint

0

View full text Add to dashboard Cite

The recently proposed permutationally invariant polynomial-neural network (PIP-NN) based ∆-machine learning (∆-ML) approach (PIP-NN ∆-ML, J. Phys. Chem. Lett. 2022, 13, 4729) is a flexible, general, and highly cost-efficient method to develop full dimensional accurate potential energy surface (PES). Only a small portion of points, which can be actively selected from the low-level (often DFT) points, with high-level energies are needed to bring a low-level PES to a high-level of quality. The hydrogen abstraction reaction between methanol and hydroxyl radical, OH + CH3OH, has been studied by theories and experiments for a long time due to its great importance in combustion, atmospheric and interstellar chemistry. However, it is not trivial to develop the full dimensional accurate PES for it. In this work, the PIP-NN ∆-ML method is successfully applied to the title reaction. The DFT PES was fitted by using 140192 points. Only 5% of the DFT dataset was needed to be calculated at the level of UCCSD(T)-F12a/AVTZ, and to improve the DFT PES to the target high-level, UCCSD(T)-F12a/AVTZ. More than 92% of the original unaffordable calculation cost were saved. The kinetics, including rate coefficients and branching ratios, were then studied by performing quasi-classical trajectory calculations on this newly fitted PES for the title reaction.

show abstract

Data Quality, Data Sampling and Data Fitting: A Tutorial Guide for Constructing Full-dimensional Accurate Potential Energy Surfaces (PESs) of Small Molecular Systems

Abstract: We provide careful discussions and considerations on data quality, data sampling and data fitting for developing full-dimensional accurate potential energy surfaces with ample interpretative examples, centering on accuracy, efficiency, and generality.

Cited by 7 publications

References 176 publications

Deciphering Dynamics of the Cl + SiH₄ → H + SiH₃Cl Reaction on a Machine Learning Made Globally Accurate Full-Dimensional Potential Energy Surface

Deciphering Dynamics of the Cl + SiH₄ → H + SiH₃Cl Reaction on a Machine Learning Made Globally Accurate Full-Dimensional Potential Energy Surface

Neural Network Based ∆-Machine learning approach efficiently brings the DFT potential energy surface to the CCSD(T) quality: a case for the OH + CH3OH reaction

Neural Network Based ∆-Machine learning approach efficiently brings the DFT potential energy surface to the CCSD(T) quality: a case for the OH + CH3OH reaction

Contact Info

Product

Resources

About

Data Quality, Data Sampling and Data Fitting: A Tutorial Guide for Constructing Full-dimensional Accurate Potential Energy Surfaces (PESs) of Small Molecular Systems

Abstract: We provide careful discussions and considerations on data quality, data sampling and data fitting for developing full-dimensional accurate potential energy surfaces with ample interpretative examples, centering on accuracy, efficiency, and generality.

Cited by 7 publications

References 176 publications

Deciphering Dynamics of the Cl + SiH4 → H + SiH3Cl Reaction on a Machine Learning Made Globally Accurate Full-Dimensional Potential Energy Surface

Deciphering Dynamics of the Cl + SiH4 → H + SiH3Cl Reaction on a Machine Learning Made Globally Accurate Full-Dimensional Potential Energy Surface

Neural Network Based ∆-Machine learning approach efficiently brings the DFT potential energy surface to the CCSD(T) quality: a case for the OH + CH3OH reaction

Neural Network Based ∆-Machine learning approach efficiently brings the DFT potential energy surface to the CCSD(T) quality: a case for the OH + CH3OH reaction

Contact Info

Product

Resources

About

Deciphering Dynamics of the Cl + SiH₄ → H + SiH₃Cl Reaction on a Machine Learning Made Globally Accurate Full-Dimensional Potential Energy Surface

Deciphering Dynamics of the Cl + SiH₄ → H + SiH₃Cl Reaction on a Machine Learning Made Globally Accurate Full-Dimensional Potential Energy Surface