Automating the Development of High-Dimensional Reactive Potential Energy Surfaces with the robosurfer Program System

Dynamics of the Cl + CH3CN reaction on an automatically-developed full-dimensional ab initio potential energy surface

A full-dimensional analytical potential energy surface (PES) is developed for the Cl + CH3CN reaction following our previous work on the benchmark ab initio characterization of the stationary points. The spin-orbit-corrected PES is constructed using the Robosurfer program and a fifth-order permutationally invariant polynomial method for fitting the high-accuracy energy points determined by a ManyHF-based coupled-cluster/triple-zeta-quality composite method. Quasi-classical trajectory simulations are performed at six collision energies between 10 and 60 kcal mol-1. Multiple low-probability product channels are found, including isomerization to isonitrile (CH3NC), but out of the eight possible channels, only the H-abstraction has significant reaction probability; thus, detailed dynamics studies are carried out only for this reaction. The cross sections and opacity functions show that the probability of the H-abstraction reaction increases with increasing collision energy (Ecoll). Scattering angle, initial attack angle, and product relative translational energy distributions indicate that the mechanism changes with the collision energy from indirect/rebound to direct stripping. The distribution of initial attack angles shows a clear preference for methyl group attack but with different angles at different Ecoll values. Post-reaction energy distributions show that the energy transfer is biased toward the products' relative translational energy instead of their internal energy. Rotational and vibrational energy have about the same amount of contribution to the internal energy in the case of both products (HCl and CH2CN), i.e., both of them are formed with high rotational excitations. HCl is produced mostly in the ground vibrational state, while a notable fraction of CH2CN is formed with vibrational excitation.

High-level analytical potential-energy-surface-based dynamics of the OH- + CH3CH2Cl SN2 and E2 reactions in full (24) dimensions

We develop a coupled-cluster full-dimensional global potential energy surface (PES) for the OH- + CH3CH2Cl reactive system, using the Robosurfer program package, which automatically samples configurations along PES-based trajectories as well as performs ab initio computations with Molpro and fitting with the monomial symmetrization approach. The analytical PES accurately describes both the bimolecular nucleophilic substitution (SN2) and elimination (E2) channels leading to the Cl- + CH3CH2OH and Cl- + H2O + C2H4 products, respectively, and allows efficient quasi-classical trajectory (QCT) simulations. QCT computations on the new PES provide accurate statistically-converged integral and differential cross sections for the OH- + CH3CH2Cl reaction, revealing the competing dynamics and mechanisms of the SN2 and E2 (anti, syn, β-α transfer) channels as well as various additional pathways leading to induced inversion of the CH3CH2Cl reactant, H-exchange between the reactants, H2O⋯Cl- complex formation, and H2O + CH3CHCl- products via proton abstraction.

Revealing new pathways for the reaction of Criegee intermediate CH2OO with SO2

Criegee intermediates play an important role in the tropospheric oxidation models through their reactions with atmospheric trace chemicals. We develop a global full-dimensional potential energy surface for the CH2OO + SO2 system and reveal how the reaction happens step by step by quasi-classical trajectory simulations. A new pathway forming the main products (CH2O + SO3) and a new product channel (CO2 + H2 + SO2) are predicted in our simulations. The new pathway appears at collision energies greater than 10 kcal/mol whose behavior demonstrates a typical barrier-controlled reaction. This threshold is also consistent with the ab initio transition state barrier height. For the minor products, a loose complex OCH2O ∙ ∙ ∙ SO2 is formed first, and then in most cases it soon turns into HCOOH + SO2, in a few cases it decomposes into CO2 + H2 + SO2 which is a new product channel, and rarely it remains as ∙OCH2O ∙ + SO2.

Benchmark ab initio characterization of the complex potential energy surfaces of the HOO- + CH3Y [Y = F, Cl, Br, I] reactions

The α-effect is a well-known phenomenon in organic chemistry, and is related to the enhanced reactivity of nucleophiles involving one or more lone-pair electrons adjacent to the nucleophilic center. The gas-phase bimolecular nucleophilic substitution (SN2) reactions of α-nucleophile HOO- with methyl halides have been thoroughly investigated experimentally and theoretically; however, these investigations have mainly focused on identifying and characterizing the α-effect of HOO-. Here, we perform the first comprehensive high-level ab initio mapping for the HOO- + CH3Y [Y = F, Cl, Br and I] reactions utilizing the modern explicitly-correlated CCSD(T)-F12b method with the aug-cc-pVnZ [n = 2-4] basis sets. The present ab initio characterization considers five distinct product channels of SN2: (CH3OOH + Y-), proton abstraction (CH2Y- + H2O2), peroxide ion substitution (CH3OO- + HY), SN2-induced elimination (CH2O + HY + HO-) and SN2-induced rearrangement (CH2(OH)O- + HY). Moreover, besides the traditional back-side attack Walden inversion, the pathways of front-side attack, double inversion and halogen-bond complex formation have also been explored for SN2. With regard to the Walden inversion of HOO- + CH3Cl, the previously unaddressed discrepancies concerning the geometry of the corresponding transition state are clarified. For the HOO- + CH3F reaction, the recently identified SN2-induced elimination is found to be more exothermic than the SN2 channel, submerged by ∼36 kcal mol-1. The accuracy of our high-level ab initio calculations performed in the present study is validated by the fact that our new benchmark 0 K reaction enthalpies show excellent agreement with the experimental data in nearly all cases.

Dynamics of the HCl + C2H5 Multichannel Reaction on a Full-Dimensional Ab Initio Potential Energy Surface

We report a full-dimensional ab initio analytical potential energy surface (PES), which accurately describes the HCl + C2H5 multichannel reaction. The new PES is developed by iteratively adding selected configurations along HCl + C2H5 quasi-classical trajectories (QCTs), thereby improving our previous Cl(2P3/2) + C2H6 PES using the Robosurfer program package. QCT simulations for the H'Cl + C2H5 reaction reveal hydrogen-abstraction, chlorine-abstraction, and hydrogen-exchange channels leading to Cl + C2H5H', H' + C2H5Cl, and HCl + C2H4H', respectively. Hydrogen abstraction dominates in the collision energy (Ecoll) range of 1-80 kcal/mol and proceeds with indirect isotropic scattering at low Ecoll and forward-scattered direct stripping at high Ecoll. Chlorine abstraction opens around 40 kcal/mol collision energy and becomes competitive with hydrogen abstraction at Ecoll = 80 kcal/mol. A restricted opening of the cone of acceptance in the Cl-abstraction reaction is found to result in the preference for a backward-scattering direct-rebound mechanism at all energies studied. Initial attack-angle distributions show mainly side-on collision preference of C2H5 for both abstraction reactions, and in the case of the HCl reactant, H/Cl-side preference for the H/Cl abstraction. For hydrogen abstraction, the collision energy transfer into the product translational and internal energy is almost equally significant, whereas in the case of chlorine abstraction, most of the available energy goes into the internal degrees of freedom. Hydrogen exchange is a minor channel with nearly constant reactivity in the Ecoll range of 10-80 kcal/mol.

First-principles mode-specific reaction dynamics

Controlling the outcome of chemical reactions by exciting specific vibrational and/or rotational modes of the reactants is one of the major goals of modern reaction dynamics studies. In the present Perspective, we focus on first-principles vibrational and rotational mode-specific dynamics computations on reactions of neutral and anionic systems beyond six atoms such as X + C2H6 [X = F, Cl, OH], HX + C2H5 [X = Br, I], OH- + CH3I, and F- + CH3CH2Cl. The dynamics simulations utilize high-level ab initio analytical potential energy surfaces and the quasi-classical trajectory method. Besides initial state specificity and the validity of the Polanyi rules, mode-specific vibrational-state assignment for polyatomic product species using normal-mode analysis and Gaussian binning is also discussed and compared with experiment.

"On-the-fly" Crystal : How to reliably and automatically characterize and construct potential energy surfaces

In this work, the Crystal  code, developed previously by the authors to find "holes" as well as legitimate transition states in existing potential energy surface (PES) functions [JPC Lett. 11, 6468 (2020)], is retooled to perform on-the-fly "direct dynamics"-type PES explorations, as well as automatic construction of new PES functions. In all of these contexts, the chief advantage of Crystal  over other methods is its ability to globally map the PES, thereby determining the most relevant regions of configuration space quickly and reliably-even when the dimensionality is rather large. Here, Crystal  is used to generate a uniformly spaced grid of density functional theory (DFT) or ab initio points, truncated over the relevant regions, which can then be used to either: (a) hone in precisely on PES features such as minima and transition states, or; (b) create a new PES function automatically, via interpolation. Proof of concept is demonstrated via application to three molecular systems: water (H2 O), (reduced-dimensional) methane (CH4 ), and methylene imine (CH2 NH).

Vibrational mode-specificity in the dynamics of the OH- + CH3I multi-channel reaction

We report a comprehensive characterization of the vibrational mode-specific dynamics of the OH- + CH3I reaction. Quasi-classical trajectory simulations are performed at four different collision energies on our previously-developed full-dimensional high-level ab initio potential energy surface in order to examine the impact of four different normal-mode excitations in the reactants. Considering the 11 possible pathways of OH- + CH3I, pronounced mode-specificity is observed in reactivity: In general, the excitations of the OH- stretching and CH stretching exert the greatest influence on the channels. For the SN2 and proton-abstraction products, the reactant initial attack angle and the product scattering angle distributions do not show major mode-specific features, except for SN2 at higher collision energies, where forward scattering is promoted by the CI stretching and CH stretching excitations. The post-reaction energy flow is also examined for SN2 and proton abstraction, and it is unveiled that the excess vibrational excitation energies rather transfer into the product vibrational energy because the translational and rotational energy distributions of the products do not represent significant mode-specificity. Moreover, in the course of proton abstraction, the surplus vibrational energy in the OH- reactant mostly remains in the H2O product owing to the prevailing dominance of the direct stripping mechanism.

Accurate fundamental invariant-neural network representation of ab initio potential energy surfaces

Highly accurate potential energy surfaces are critically important for chemical reaction dynamics. The large number of degrees of freedom and the intricate symmetry adaption pose a big challenge to accurately representing potential energy surfaces (PESs) for polyatomic reactions. Recently, our group has made substantial progress in this direction by developing the fundamental invariant-neural network (FI-NN) approach. Here, we review these advances, demonstrating that the FI-NN approach can represent highly accurate, global, full-dimensional PESs for reactive systems with even more than 10 atoms. These multi-channel reactions typically involve many intermediates, transition states, and products. The complexity and ruggedness of this potential energy landscape present even greater challenges for full-dimensional PES representation. These PESs exhibit a high level of complexity, molecular size, and accuracy of fit. Dynamics simulations based on these PESs have unveiled intriguing and novel reaction mechanisms, providing deep insights into the intricate dynamics involved in combustion, atmospheric, and organic chemistry.

A Gaussian Process Based Δ-Machine Learning Approach to Reactive Potential Energy Surfaces

The Gaussian process (GP) is an efficient nonparametric machine learning (ML) method. A distinct advantage of the GP is its ability to provide an estimate of statistical uncertainties. This is particularly useful in constructing high-dimensional potential energy surfaces (PESs) from ab initio data as it offers an optimal way to add new geometries to reduce the overall error. In this work, GP is employed in the context of Δ-machine learning (Δ-ML), in which a correction PES to an inaccurate low-level PES is constructed using a small number of high-level ab initio calculations. This new method is tested in three prototypical reactive systems, namely, the H + H2 → H + H2, OH + H2 → H2O + H, and H + CH4 → H2 + CH3 reactions. The results show that the GP-based Δ-ML approach is more efficient than its direct application in constructing high-level PESs. We also compare the new method to a previously proposed neural-network-based Δ-ML approach [Liu and Li J. Phys. Chem. Lett. 2022, 13, 4729-4738]. The results indicate that the two Δ-ML methods have comparable efficiencies in constructing accurate PESs.

Alternating Stereospecificity upon Central-Atom Change: Dynamics of the F- +PH2 Cl SN 2 Reaction Compared to its C- and N-Centered Analogues

Central-atom effects on bimolecular nucleophilic substitution (SN 2) reactions are well-known in chemistry, however, the atomic-level SN 2 dynamics at phosphorous (P) centers has never been studied. We investigate the dynamics of the F- +PH2 Cl reaction with the quasi-classical trajectory method on a novel full-dimensional analytical potential energy surface fitted on high-level ab initio data. Our computations reveal intermediate dynamics compared to the F- +CH3 Cl and the F- +NH2 Cl SN 2 reactions: phosphorus as central atom leads to a more indirect SN 2 reaction with extensive complex-formation with respect to the carbon-centered one, however, the title reaction is more direct than its N-centered pair. Stereospecificity, characteristic at C-center, does not appear here either, due to the submerged front-side-attack retention path and the repeated entrance-channel inversional motion, whereas the multi-inversion mechanism discovered at nitrogen center is also undermined by the deep Walden-well. At low collision energies, 6 % of the PH2 F products form with retained configuration, mostly through complex-mediated mechanisms, while this ratio reaches 24 % at the highest energy due to the increasing dominance of the direct front-side mechanism and the smaller chance for hitting the deep Walden-inversion minimum. Our results suggest pronounced central-atom effects in SN 2 reactions, which can fundamentally change their (stereo)dynamics.

Full-dimensional automated potential energy surface development and detailed dynamics for the CH2OO + NH3 reaction

With the help of the ROBOSURFER program package, a global full-dimensional potential energy surface (PES) for the reaction of the Criegee intermediate, CH2OO, with the NH3 molecule is developed iteratively using different ab initio methods and the monomial symmetrization fitting approach. The final permutationally-invariant analytical PES is constructed based on 23447 geometries and the corresponding ManyHF-based CCSD(T)-F12b/cc-pVTZ-F12 energies. The accuracy of the PES is confirmed by the excellent agreement of its stationary-point properties and one-dimensional potential energy curves compared with the corresponding ab initio data. The reaction probabilities and integral cross sections are calculated for the ground-state and several vibrationally excited-state reactions by quasi-classical trajectory simulations. Remarkable is that the maximum impact parameter b where reactivity vanishes is almost independent of collision energy ranging from 1 to 40 kcal mol-1, and the reaction probability increases with increasing collision energy for this negative-barrier reaction. At the same time, a slight mode-specificity effect is observed. In addition, the deuterium effect is investigated and the sudden vector projection is discussed.

ManyHF-based full-dimensional potential energy surface development and quasi-classical dynamics for the Cl + CH3NH2 reaction

A full-dimensional spin-orbit (SO)-corrected potential energy surface (PES) is developed for the Cl + CH3NH2 multi-channel system. Using the new PES, a comprehensive reaction dynamics investigation is performed for the most reactive hydrogen-abstraction reactions forming HCl + CH2NH2/CH3NH. Hartree-Fock (HF) convergence problems in the reactant region are handled by the ManyHF method, which finds the lowest-energy HF solution considering several different initial guess orbitals. The PES development is carried out with the Robosurfer program package, which iteratively improves the surface. Energy points are computed at the ManyHF-UCCSD(T)-F12a/cc-pVDZ-F12 level of theory combined with basis set (ManyHF-RMP2-F12/cc-pVTZ-F12 - ManyHF-RMP2-F12/cc-pVDZ-F12) and SO (MRCI+Q/aug-cc-pwCVDZ) corrections. Quasi-classical trajectory simulations show that the CH3-side hydrogen abstraction occurs more frequently in contrast to the NH2-side reaction. In both cases, the integral cross sections decrease with increasing collision energy (Ecoll). A reaction mechanism shifting from indirect to direct stripping can be observed from the opacity functions, scattering angle, and translation energy distributions as Ecoll increases. Initial attack angle distributions reveal that chlorine prefers to abstract hydrogen from the approached functional group. The collision-energy dependence of the product energy distributions shows that the initial translational energy mainly transfers to product recoil. The HCl vibrational and rotational energy values are comparable and nearly independent of collision energy, while the CH2NH2 and CH3NH co-products' vibrational energy values are higher than the rotational energy values with more significant Ecoll dependence. The HCl(v = 0) rotational distributions are compared with experiment, setting the direction for future investigations.

Vibrational Mode-Specific Dynamics of the OH + C2H6 Reaction

We investigate the effects of the initial vibrational excitations on the dynamics of the OH + C2H6 → H2O + C2H5 reaction using the quasi-classical trajectory method and a full-dimensional analytical ab initio potential energy surface. Excitation of the initial CH, CC, and OH stretching modes enhances, slightly inhibits, and does not affect the reactivity, respectively. Translational energy activates the early-barrier title reaction more efficiently than OH and CC stretching excitations, in accord with the Polanyi rules whereas CH stretching modes have similar or higher efficacy than translation, showing that these rules are not always valid in polyatomic processes. Scattering angle, initial attack angle, and product translational energy distributions show the dominance of direct stripping with increasing collision energy, side-on OH and isotropic C2H6 attack preferences, and substantial reactant-product translational energy transfer without any significant mode specificity. The reactant vibrational excitation energy of OH and C2H6 flows into the H2O and C2H5 product vibrations, respectively, whereas product rotations are not affected. The computed mode-specific H2O vibrational distributions show that initial OH excitation appears in the asymmetric stretching vibration of the H2O product and allow comparison with experiments.

Quantum Dynamical Investigation of Dihydrogen-Hydride Exchange in a Transition-Metal Polyhydride Complex

The ongoing shift toward clean, sustainable energy is a primary driving force behind hydrogen fuel research. Safe and effective storage of hydrogen is a major challenge (particularly for mobile applications) and requires a detailed understanding of the atomic level interactions of hydrogen with its host materials. The light mass of hydrogen, however, implies that quantum effects are important, so a quantum dynamical treatment is required to properly account for these effects in computational simulations. As one such example, we describe herein the hydrogen exchange dynamics between a hydride and a dihydrogen ligand in the [FeH(H2)(PH3)4]+ model complex. A global three-dimensional (3D) potential energy surface (PES) was constructed by fitting to and interpolating from a discrete set of grid points computed using density functional theory; exact quantum dynamical calculations were then carried out on the 3D PES using discrete variable representation basis sets. Energy levels and their quantum tunneling splittings were computed up to 3000 cm-1 above the ground state. Within that energy range, all three fundamentals have been identified using wave function plots, as well as the first three overtones of the exchange (reaction coordinate) motion and several of its combination bands. From the tunneling splittings, the Boltzmann-averaged tunneling rates were computed. The Arrhenius plot of the total exchange rate shows a clear transition around 150 K, below which the activation energy is essentially zero and above which it is less than half of the electronic structure barrier. This indicates that exchange rates are governed by quantum tunneling throughout the relevant temperature range with the low-temperature regime dominated by a single quantum (ground) state. This work is the first-ever fully quantum dynamical study to investigate the hydrogen exchange dynamics between hydride and dihydrogen ligands coordinated to a transition-metal complex.

Exploring the versatile reactivity of the F- + SiH3Cl system on a full-dimensional coupled-cluster potential energy surface

We develop a full-dimensional analytical potential energy surface (PES) for the F- + SiH3Cl reaction using Robosurfer for automatically sampling the configuration space, the robust [CCSD-F12b + BCCD(T) - BCCD]/aug-cc-pVTZ composite level of theory for computing the energy points, and the permutationally invariant polynomial method for fitting. Evolution of the fitting error and the percentage of the unphysical trajectories are monitored as a function of the iteration steps/number of energy points and polynomial order. Quasi-classical trajectory simulations on the new PES reveal rich dynamics resulting in high-probability SN2 (SiH3F + Cl-) and proton-transfer (SiH2Cl- + HF) products as well as several lower-probability channels, such as SiH2F- + HCl, SiH2FCl + H-, SiH2 + FHCl-, SiHFCl- + H2, SiHF + H2 + Cl-, and SiH2 + HF + Cl-. The Walden-inversion and front-side-attack-retention SN2 pathways are found to be competitive, producing nearly racemic products at high collision energies. The detailed atomic-level mechanisms of the various reaction pathways and channels as well as the accuracy of the analytical PES are analyzed along representative trajectories.

Quasiclassical Trajectory Simulation as a Protocol to Build Locally Accurate Machine Learning Potentials

Direct trajectory calculations have become increasingly popular in recent computational chemistry investigations. However, the exorbitant computational cost of ab initio trajectory calculations usually limits its application in mechanistic explorations. Recently, machine learning-based potential energy surface (ML-PES) provides a powerful strategy to circumvent the heavy computational cost and meanwhile maintain the required accuracy. Despite the appealing potential, constructing a robust ML-PES is still challenging since the training set of the PES should cover a broad enough configuration space. In this work, we demonstrate that when the concerned properties could be collected by the localized sampling of the configuration space, quasiclassical trajectory (QCT) calculations can be invoked to efficiently obtain locally accurate ML-PESs. We prove our concept with two model reactions: methyl migration of i-pentane cation and dimerization of cyclopentadiene. We found that the locally accurate ML-PESs are sufficiently robust for reproducing the static and dynamic features of the reactions, including the time-resolved free energy and entropy changes, and time gaps.

Quasi-classical trajectory study of the OH- + CH3I reaction: theory meets experiment

Regarding OH^- + CH₃I, several studies have focused on the dynamics of the reaction. Here, high-level quasi-classical trajectory simulations are carried out at four different collision energies on our recently developed potential energy surface. In all, more than half a million trajectories are performed, and for the first time, the detailed quasi-classical trajectory results are compared with the reanalysed crossed-beam ion imaging experiments. Concerning the previously reported direct dynamics study of OH^- + CH₃I, a better agreement can be obtained between the revised experiment and our novel theoretical results. Furthermore, in the present work, the benchmark geometries, frequencies and relative energies of the stationary points are also determined for the OH^- + CH₃I proton-abstraction channel along with the earlier characterized S_N2 channel.

Recent advances in quantum theory on ro-vibrationally inelastic scattering

Molecular collisions are of fundamental importance in understanding intermolecular interaction and dynamics. Its importance is accentuated in cold and ultra-cold collisions because of the dominant quantum mechanical nature of the scattering. We review recent advances in the time-independent approach to quantum mechanical characterization of non-reactive scattering in tetratomic systems, which is ideally suited for large collisional de Broglie wavelengths characteristic in cold and ultracold conditions. We discuss quantum scattering algorithms between two diatoms and between a triatom and an atom and their implementation, as well as various approximate schemes. They not only enable the characterization of collision dynamics in realistic systems but also serve as benchmarks for developing more approximate methods.

PESPIP: Software to fit complex molecular and many-body potential energy surfaces with permutationally invariant polynomials

We wish to describe a potential energy surface by using a basis of permutationally invariant polynomials whose coefficients will be determined by numerical regression so as to smoothly fit a dataset of electronic energies as well as, perhaps, gradients. The polynomials will be powers of transformed internuclear distances, usually either Morse variables, exp(-r_i,j/λ), where λ is a constant range hyperparameter, or reciprocals of the distances, 1/r_i,j. The question we address is how to create the most efficient basis, including (a) which polynomials to keep or discard, (b) how many polynomials will be needed, (c) how to make sure the polynomials correctly reproduce the zero interaction at a large distance, (d) how to ensure special symmetries, and (e) how to calculate gradients efficiently. This article discusses how these questions can be answered by using a set of programs to choose and manipulate the polynomials as well as to write efficient Fortran programs for the calculation of energies and gradients. A user-friendly interface for access to monomial symmetrization approach results is also described. The software for these programs is now publicly available.

Δ-Machine Learned Potential Energy Surfaces and Force Fields

There has been great progress in developing machine-learned potential energy surfaces (PESs) for molecules and clusters with more than 10 atoms. Unfortunately, this number of atoms generally limits the level of electronic structure theory to less than the "gold standard" CCSD(T) level. Indeed, for the well-known MD17 dataset for molecules with 9-20 atoms, all of the energies and forces were obtained with DFT calculations (PBE). This Perspective is focused on a Δ-machine learning method that we recently proposed and applied to bring DFT-based PESs to close to CCSD(T) accuracy. This is demonstrated for hydronium, N-methylacetamide, acetyl acetone, and ethanol. For 15-atom tropolone, it appears that special approaches (e.g., molecular tailoring, local CCSD(T)) are needed to obtain the CCSD(T) energies. A new aspect of this approach is the extension of Δ-machine learning to force fields. The approach is based on many-body corrections to polarizable force field potentials. This is examined in detail using the TTM2.1 water potential. The corrections make use of our recent CCSD(T) datasets for 2-b, 3-b, and 4-b interactions for water. These datasets were used to develop a new fully ab initio potential for water, termed q-AQUA.

Full-dimensional automated potential energy surface development and dynamics for the OH + C2H6 reaction

We develop a full-dimensional analytical potential energy surface (PES) for the OH + C2H6 reaction using the Robosurfer program system, which automatically (1) selects geometries from quasi-classical trajectories, (2) performs ab initio computations using a CCSD(T)-F12/triple-zeta-quality composite method, (3) fits the energies utilizing the permutationally-invariant monomial symmetrization approach, and iteratively improves the PES via steps (1)−(3). Quasi-classical trajectory simulations on the new PES reveal that hydrogen abstraction leading to H2O + C2H5 dominates in the collision energy range of 10−50 kcal/mol. The abstraction cross sections increase and the dominant mechanism shifts from rebound (small impact parameters and backward scattering) to stripping (larger impact parameters and forward scattering) with increasing collision energy as opacity functions and scattering angle distributions indicate. The abstraction reaction clearly favors side-on OH attack over O-side and the least-preferred H-side approach, whereas C2H6 behaves like a spherical object with only slight C−C-perpendicular side-on preference. Collision energy efficiently flows into the relative translation of the products, whereas product internal energy distributions show only little collision energy dependence. H2O/C2H5 vibrational distributions slightly/significantly violate zero-point energy and are nearly independent of collision energy, whereas the rotational distributions clearly blue-shift as collision energy increases.

The MD17 datasets from the perspective of datasets for gas-phase “small” molecule potentials

There has been great progress in developing methods for machine-learned potential energy surfaces. There have also been important assessments of these methods by comparing so-called learning curves on datasets of electronic energies and forces, notably the MD17 database. The dataset for each molecule in this database generally consists of tens of thousands of energies and forces obtained from DFT direct dynamics at 500 K. We contrast the datasets from this database for three “small” molecules, ethanol, malonaldehyde, and glycine, with datasets we have generated with specific targets for the potential energy surfaces (PESs) in mind: a rigorous calculation of the zero-point energy and wavefunction, the tunneling splitting in malonaldehyde, and, in the case of glycine, a description of all eight low-lying conformers. We found that the MD17 datasets are too limited for these targets. We also examine recent datasets for several PESs that describe small-molecule but complex chemical reactions. Finally, we introduce a new database, “QM-22,” which contains datasets of molecules ranging from 4 to 15 atoms that extend to high energies and a large span of configurations.

Unconventional SN2 retention pathways induced by complex formation: High-level dynamics investigation of the NH2 - + CH3I polyatomic reaction

Investigations on the dynamics of chemical reactions have been a hot topic for experimental and theoretical studies over the last few decades. Here, we carry out the first high-level dynamical characterization for the polyatom-polyatom reaction between NH₂ ^- and CH₃I. A global analytical potential energy surface is developed to describe the possible pathways with the quasi-classical trajectory method at several collision energies. In addition to S_N2 and proton abstraction, a significant iodine abstraction is identified, leading to the CH₃ + [NH₂⋯I]^- products. For S_N2, our computations reveal an indirect character as well, promoting the formation of [CH₃⋯NH₂] complexes. Two novel dominant S_N2 retention pathways are uncovered induced by the rotation of the CH₃ fragment in these latter [CH₃⋯NH₂] complexes. Moreover, these uncommon routes turn out to be the most dominant retention paths for the NH₂ ^- + CH₃I S_N2 reaction.

Rotational Mode-Specificity in the Cl + C2H6 → HCl + C2H5 Reaction

We perform rotational mode-specific quasi-classical trajectory simulations using a high-quality ab initio analytical potential energy surface for the Cl(²P_3/2) + C₂H₆ → HCl + C₂H₅ reaction. As ethane, being a prolate-type symmetric top, can be characterized by the J and K rotational quantum numbers, the excitation of two rotational modes, the tumbling (J, K = 0) and spinning (J, K = J) rotations of the reactant is carried out with J = 10, 20, 30, and 40 at a wide range of collision energies. The impacts of rotational excitation on the reactivity, the mechanism, and the post-reaction distribution of energy are investigated: (1) exciting both rotational modes enhances the reactivity with the spinning rotation being more effective due to its coupling to the C-H stretching vibrational normal modes (C-H bond elongating effect) and larger rotational energies, (2) rotational excitation increases the dominance of direct rebound over the stripping mechanism, while collision energy favors the latter, (3) investing energy in tumbling rotation excites the translational motion of the products, while the excess spinning rotational energy readily flows into the internal degrees of freedom of the ethyl radical or, less significantly, into the HCl vibration, probably due to the pronounced rovibrational coupling in this case. We also study the relative efficiency of vibrational and rotational excitation on the reactivity of the barrierless and thus translationally hindered title reaction.

Vibrational mode-specific dynamics of the F- + CH3CH2Cl multi-channel reaction

We investigate the mode-specific dynamics of the ground-state, C-Cl stretching (v₁₀), CH₂ wagging (v₇), sym-CH₂ stretching (v₁), and sym-CH₃ stretching (v₃) excited F^- + CH₃CH₂Cl(v_k = 0, 1) [k = 10, 7, 1, 3] → Cl^- + CH₃CH₂F (S_N2), HF + CH₃CHCl^-, FH⋯Cl^- + C₂H₄, and Cl^- + HF + C₂H₄ (E2) reactions using a full-dimensional high-level analytical global potential energy surface and the quasi-classical trajectory method. Excitation of the C-Cl stretching, CH₂ stretching, and CH₂/CH₃ stretching modes enhances the S_N2, proton abstraction, and FH⋯Cl^- and E2 channels, respectively. Anti-E2 dominates over syn-E2 (kinetic anti-E2 preference) and the thermodynamically-favored S_N2 (wider reactive anti-E2 attack angle range). The direct (a) S_N2, (b) proton abstraction, (c) FH⋯Cl^- + C₂H₄, (d) syn-E2, and (e) anti-E2 channels proceed with (a) back-side/backward, (b) isotropic/forward, (c) side-on/forward, (d) front-side/forward, and (e) back-side/forward attack/scattering, respectively. The HF products are vibrationally cold, especially for proton abstraction, and their rotational excitation increases for proton abstraction, anti-E2, and syn-E2, in order. Product internal-energy and mode-specific vibrational distributions show that CH₃CH₂F is internally hot with significant C-F stretching and CH₂ wagging excitations, whereas C₂H₄ is colder. One-dimensional Gaussian binning technique is proved to solve the normal mode analysis failure caused by methyl internal rotation.

REANN: A PyTorch-based end-to-end multi-functional deep neural network package for molecular, reactive, and periodic systems

In this work, we present a general purpose deep neural network package for representing energies, forces, dipole moments, and polarizabilities of atomistic systems. This so-called recursively embedded atom neural network model takes advantages of both the physically inspired atomic descriptor based neural networks and the message-passing based neural networks. Implemented in the PyTorch framework, the training process is parallelized on both the central processing unit and the graphics processing unit with high efficiency and low memory in which all hyperparameters can be optimized automatically. We demonstrate the state-of-the-art accuracy, high efficiency, scalability, and universality of this package by learning not only energies (with or without forces) but also dipole moment vectors and polarizability tensors in various molecular, reactive, and periodic systems. An interface between a trained model and LAMMPs is provided for large scale molecular dynamics simulations. We hope that this open-source toolbox will allow for future method development and applications of machine learned potential energy surfaces and quantum-chemical properties of molecules, reactions, and materials.

Hitting the Trifecta: How to Simultaneously Push the Limits of Schrödinger Solution with Respect to System Size, Convergence Accuracy, and Number of Computed States

Methods for solving the Schrödinger equation without approximation are in high demand but are notoriously computationally expensive. In practical terms, there are just three primary factors that currently limit what can be achieved: 1) system size/dimensionality; 2) energy level excitation; and 3) numerical convergence accuracy. Broadly speaking, current methods can deliver on any two of these three goals, but achieving all three at once remains an enormous challenge. In this paper, we shall demonstrate how to "hit the trifecta" in the context of molecular vibrational spectroscopy calculations. In particular, we compute the lowest 1000 vibrational states for the six-atom acetonitrile molecule (CH₃CN), to a numerical convergence of accuracy 10^-2 cm^-1 or better. These calculations encompass all vibrational states throughout most of the dynamically relevant range (i.e., up to ∼4250 cm^-1 above the ground state), computed in full quantum dimensionality (12 dimensions), to near spectroscopic accuracy. To our knowledge, no such vibrational spectroscopy calculation has ever previously been performed.

Uncovering an oxide ion substitution for the OH- + CH3F reaction

Theoretical investigations on chemical reactions allow us to understand the dynamics of the possible pathways and identify new unexpected routes. Here, we develop a global analytical potential energy surface (PES) for the OH⁻ + CH₃F reaction in order to perform high-level dynamics simulations. Besides bimolecular nucleophilic substitution (S_N2) and proton abstraction, our quasi-classical trajectory computations reveal a novel oxide ion substitution leading to the HF + CH₃O⁻ products. This exothermic reaction pathway occurs via the CH₃OH⋯F⁻ deep potential well of the S_N2 product channel as a result of a proton abstraction from the hydroxyl group by the fluoride ion. The present detailed dynamics study of the OH⁻ + CH₃F reaction focusing on the surprising oxide ion substitution demonstrates how incomplete our knowledge is of fundamental chemical reactions.

Reaction dynamics simulations on a high-level ab initio analytical potential energy surface reveal a novel oxide ion substitution channel for the OH⁻ + CH₃F reaction.

Vibrational mode-specific dynamics of the F(2P3/2) + C2H6 → HF + C2H5 reaction

We investigate the competing effect of vibrational and translational excitation and the validity of the Polanyi rules in the early- and negative-barrier F(²P_3/2) + C₂H₆ → HF + C₂H₅ reaction by performing quasi-classical dynamics simulations on a recently developed full-dimensional multi-reference analytical potential energy surface. The effect of five normal-mode excitations of ethane on the reactivity, the mechanism, and the post-reaction energy flow is followed through a wide range of collision energies. Promoting effects of vibrational excitations and interaction time, related to the slightly submerged barrier, are found to be suppressed by the early-barrier-induced translational enhancement, in contrast to the slightly late-barrier Cl + C₂H₆ reaction. The excess vibrational energy mostly converts into ethyl internal excitation while collision energy is transformed into product separation. The substantial reaction energy excites the HF vibration, which tends to show mode-specificity and translational energy dependence as well. With increasing collision energy, direct stripping becomes dominant over the direct rebound and indirect mechanisms, being basically independent of reactant excitation.

Atomistic dynamics of elimination and nucleophilic substitution disentangled for the F- + CH3CH2Cl reaction

Chemical reaction dynamics are studied to monitor and understand the concerted motion of several atoms while they rearrange from reactants to products. When the number of atoms involved increases, the number of pathways, transition states and product channels also increases and rapidly presents a challenge to experiment and theory. Here we disentangle the dynamics of the competition between bimolecular nucleophilic substitution (S_N2) and base-induced elimination (E2) in the polyatomic reaction F^- + CH₃CH₂Cl. We find quantitative agreement for the energy- and angle-differential reactive scattering cross-sections between ion-imaging experiments and quasi-classical trajectory simulations on a 21-dimensional potential energy hypersurface. The anti-E2 pathway is most important, but the S_N2 pathway becomes more relevant as the collision energy is increased. In both cases the reaction is dominated by direct dynamics. Our study presents atomic-level dynamics of a major benchmark reaction in physical organic chemistry, thereby pushing the number of atoms for detailed reaction dynamics studies to a size that allows applications in many areas of complex chemical networks and environments.

Detailed quasiclassical dynamics of the F- + CH3Br reaction on an ab initio analytical potential energy surface

Dynamics and mechanisms of the F^- + CH₃Br(v = 0) → Br^- + CH₃F (S_N2 via Walden inversion, front-side attack, and double inversion), F^- + inverted-CH₃Br (induced inversion), HF + CH₂Br^- (proton abstraction), and FH⋯Br^- + ¹CH₂ reactions are investigated using a high-level global ab initio potential energy surface, the quasiclassical trajectory method, as well as non-standard configuration- and mode-specific analysis techniques. A vector-projection method is used to identify inversion and retention trajectories; then, a transition-state-attack-angle-based approach unambiguously separates the front-side attack and the double-inversion retention pathways. The Walden-inversion S_N2 channel becomes direct rebound dominated with increasing collision energy as indicated by backward scattering, initial back-side attack preference, and the redshifting of product internal energy peaks in accord with CF stretching populations. In the minor retention and induced-inversion pathways, almost the entire available energy transfers into product rotation-vibration, and retention mainly proceeds with indirect, slow double inversion following induced inversion with about 50% probability. Proton abstraction is dominated by direct stripping (evidenced by forward scattering) with CH₃-side initial attack preference, providing mainly vibrationally ground state products with significant zero-point energy violation.

Parsimonious Potential Energy Surface Expansions Using Dictionary Learning with Multipass Greedy Selection

Potential energy surfaces fit with basis set expansions have been shown to provide accurate representations of electronic energies and have enabled a variety of high-accuracy dynamics, kinetics, and spectroscopy applications. The number of terms in these expansions scales poorly with system size, a drawback that challenges their use for systems with more than ∼10 atoms. A solution is presented here using dictionary learning. Subsets of the full set of conventional basis functions are optimized using a newly developed multipass greedy regression method inspired by forward and backward selection methods from the statistics, signal processing, and machine learning literatures. The optimized representations have accuracies comparable to the full set but are 1 or more orders of magnitude smaller, and notably, the number of terms in the optimized multipass greedy expansions scales approximately linearly with the number of atoms.

Permutationally Invariant Polynomial Expansions with Unrestricted Complexity

A general strategy is presented for constructing and validating permutationally invariant polynomial (PIP) expansions for chemical systems of any stoichiometry. Demonstrations are made for three categories of gas-phase dynamics and kinetics: collisional energy-transfer trajectories for predicting pressure-dependent kinetics, three-body collisions for describing transient van der Waals adducts relevant to atmospheric chemistry, and nonthermal reactivity via quasiclassical trajectories. In total, 30 systems are considered with up to 15 atoms and 39 degrees of freedom. Permutational invariance is enforced in PIP expansions with as many as 13 million terms and 13 permutationally distinct atom types by taking advantage of petascale computational resources. The quality of the PIP expansions is demonstrated through the systematic convergence of in-sample and out-of-sample errors with respect to both the number of training data and the order of the expansion, and these errors are shown to predict errors in the dynamics for both reactive and nonreactive applications. The parallelized code distributed as part of this work enables the automation of PIP generation for complex systems with multiple channels and flexible user-defined symmetry constraints and for automatically removing unphysical unconnected terms from the basis set expansions, all of which are required for simulating complex reactive systems.

Searching Configurations in Uncertainty Space: Active Learning of High-Dimensional Neural Network Reactive Potentials

Neural network (NN) potential energy surfaces (PESs) have been widely used in atomistic simulations with ab initio accuracy. While constructing NN PESs, their training data points are often sampled by molecular dynamics trajectories. This strategy can be however inefficient for reactive systems involving rare events. Here, we develop an uncertainty-driven active learning strategy to automatically and efficiently generate high-dimensional NN-based reactive potentials, taking a gas-surface reaction as an example. The difference between two independent NN models is used as a simple and differentiable uncertainty metric, allowing us to quickly search in the uncertainty space and place new samples at which the PES is less reliable. By interfacing this algorithm with the first-principles simulation package, we demonstrate that a globally accurate NN potential of the H₂ + Ag(111) system can be constructed with merely ∼150 data points. This PES can be further refined to describe H₂ dissociation on Ag(100) by adding ∼130 more configurations on this facet. The entire process is completely automatic and self-terminated once the relative error criterion is fulfilled. Impressively, data points sampled by this uncertainty-driven strategy are substantially fewer than by the traditional trajectory-based sampling. The final NN PES not only converges well the quantum dissociation probability of the molecule but also well-reproduces the phonon properties of the substrate and is capable of describing surface temperature effects. These results show the potential of this active learning approach in developing high-dimensional NN reactive potentials in gas and condensed phases.

First-Principles Reaction Dynamics beyond Six-Atom Systems

Moving beyond the six-atomic benchmark systems, we discuss the new age and future of first-principles reaction dynamics, which investigates complex, multichannel chemical reactions. We describe the methodology starting from the benchmark ab initio characterization of the stationary points, followed by full-dimensional potential energy surface (PES) developments and reaction dynamics computations. We highlight our composite ab initio approach providing benchmark stationary-point properties with subchemical accuracy, the Robosurfer program system enabling automatic PES development, and applications for the Cl + C2H6, F + C2H6, and OH- + CH3I post-six-atom reactions focusing on ab initio issues and their solutions as well as showing the excellent agreement between theory and experiment.

Facilitated inversion complicates the stereodynamics of an SN2 reaction at nitrogen center

Bimolecular nucleophilic substitution (S_N2) reactions at carbon center are well known to proceed with the stereospecific Walden-inversion mechanism. Reaction dynamics simulations on a newly developed high-level ab initio analytical potential energy surface for the F⁻ + NH₂Cl nitrogen-centered S_N2 and proton-transfer reactions reveal a hydrogen-bond-formation-induced multiple-inversion mechanism undermining the stereospecificity of the N-centered S_N2 channel. Unlike the analogous F⁻ + CH₃Cl S_N2 reaction, F⁻ + NH₂Cl → Cl⁻ + NH₂F is indirect, producing a significant amount of NH₂F with retention, as well as inverted NH₂Cl during the timescale within the unperturbed NH₂Cl molecule gets inverted with only low probability, showing the important role of facilitated inversions via an FH…NHCl⁻-like transition state. Proton transfer leading to HF + NHCl⁻ is more direct and becomes the dominant product channel at higher collision energies.

Multiple-inversion, the analogue of the double-inversion pathway recently revealed for S_N2@C, is the key mechanism in S_N2 at N center undermining stereospecificity.

Spectroscopy and Scattering Studies Using Interpolated Ab Initio Potentials

The Born-Oppenheimer potential energy surface (PES) has come a long way since its introduction in the 1920s, both conceptually and in predictive power for practical applications. Nevertheless, nearly 100 years later-despite astonishing advances in computational power-the state-of-the-art first-principles prediction of observables related to spectroscopy and scattering dynamics is surprisingly limited. For example, the water dimer, (H₂O)₂, with only six nuclei and 20 electrons, still presents a formidable challenge for full-dimensional variational calculations of bound states and is considered out of reach for rigorous scattering calculations. The extremely poor scaling of the most rigorous quantum methods is fundamental; however, recent progress in development of approximate methodologies has opened the door to fairly routine high-quality predictions, unthinkable 20 years ago. In this review, in relation to the workflow of spectroscopy and/or scattering studies, we summarize progress and challenges in the component areas of electronic structure calculations, PES fitting, and quantum dynamical calculations.

Neural Network Potential Energy Surfaces for Small Molecules and Reactions

We review progress in neural network (NN)-based methods for the construction of interatomic potentials from discrete samples (such as ab initio energies) for applications in classical and quantum dynamics including reaction dynamics and computational spectroscopy. The main focus is on methods for building molecular potential energy surfaces (PES) in internal coordinates that explicitly include all many-body contributions, even though some of the methods we review limit the degree of coupling, due either to a desire to limit computational cost or to limited data. Explicit and direct treatment of all many-body contributions is only practical for sufficiently small molecules, which are therefore our primary focus. This includes small molecules on surfaces. We consider direct, single NN PES fitting as well as more complex methods that impose structure (such as a multibody representation) on the PES function, either through the architecture of one NN or by using multiple NNs. We show how NNs are effective in building representations with low-dimensional functions including dimensionality reduction. We consider NN-based approaches to build PESs in the sums-of-product form important for quantum dynamics, ways to treat symmetry, and issues related to sampling data distributions and the relation between PES errors and errors in observables. We highlight combinations of NNs with other ideas such as permutationally invariant polynomials or sums of environment-dependent atomic contributions, which have recently emerged as powerful tools for building highly accurate PESs for relatively large molecular and reactive systems.

Plumbing Potentials for Molecules with Up To Tens of Atoms: How to Find Saddle Points and Fix Leaky Holes

Potential energy surfaces (PESs) play an indispensable role in molecular dynamics but are notoriously difficult to flesh out properly in large-dimensional spaces. In particular, the undetected presence of PES holes, i.e., unphysical saddle points beyond which the potential energy drops arbitrarily, can have devastating effects on both classical and quantum dynamics calculations. In this study, the Crystal algorithm is developed as a tool for efficiently and accurately finding PES holes, as well as legitimate saddle points, even in very large-dimensional configuration spaces. The approach is applied to three large-dimensional PESs for molecular systems of current interest: uracil, naphthalene, and formic acid dimer. Low-lying PES holes are discovered and located for the first two systems-including naphthalene, for which no holes were previously suspected, to the best of our knowledge. Likewise, the double-well, double-proton-transfer isomerization saddle point for formic acid dimer is also located.

High-Fidelity Potential Energy Surfaces for Gas-Phase and Gas-Surface Scattering Processes from Machine Learning

In this Perspective, we review recent advances in constructing high-fidelity potential energy surfaces (PESs) from discrete ab initio points, using machine learning tools. Such PESs, albeit with substantial initial investments, provide significantly higher efficiency than direct dynamics methods and/or high accuracy at a level that is not affordable by on-the-fly approaches. These PESs not only are a necessity for quantum dynamical studies because of delocalization of wave packets but also enable the study of low-probability and long-time events in (quasi-)classical treatments. Our focus here is on inelastic and reactive scattering processes, which are more challenging than bound systems because of the involvement of continua. Relevant applications and developments for dynamical processes in both the gas phase and at gas-surface interfaces are discussed.

Theory Finally Agrees with Experiment for the Dynamics of the Cl + C2H6 Reaction

Since the pioneering reaction dynamics studies of H + H2 in the 1970s, theory increased the system size by one atom in every decade arriving to six-atom reactions in the early 2010s. Here, we take a significant step forward by reporting accurate dynamics simulations for the nine-atom Cl + ethane (C2H6) reaction using a new high-quality spin-orbit-ground-state ab initio potential energy surface. Quasi-classical trajectory simulations on this surface cool the rotational distribution of the HCl product molecules, thereby providing unprecedented agreement with experiment after several previous failed attempts of theory. Unlike Cl + CH4, the Cl + C2H6 reaction is exothermic with an adiabatically submerged transition state, allowing testing of the validity of the Polanyi rules for a negative-barrier reaction.