Interpolating moving least-squares methods for fitting potential energy surfaces: applications to classical dynamics calculations

As a continuation of our efforts to develop efficient and accurate interpolating moving least-squares (IMLS) methods for generating potential energy surfaces, we carry out classical trajectories and compute kinetics properties on higher degree IMLS surfaces. In this study, we have investigated the choice of coordinate system, the range of points (i.e., the cutoff radius) used in fitting, and strategies for selections of data points and basis elements. We illustrate and test the method by applying it to hydrogen peroxide (HOOH). In particular, reaction rates for the O-O bond breaking in HOOH are calculated on fitted surfaces using the classical trajectory approach to test the accuracy of the IMLS method for providing potentials for dynamics calculations.     

Interpolating moving least-squares methods for fitting potential energy surfaces: a strategy for efficient automatic data point placement in high dimensions

An accurate and efficient method for automated molecular global potential energy surface (PES) construction and fitting is demonstrated. An interpolating moving least-squares (IMLS) method is developed with the flexibility to fit various ab initio data: (1) energies, (2) energies and gradients, or (3) energies, gradients, and Hessian data. The method is automated and flexible so that a PES can be optimally generated for trajectories, spectroscopy, or other applications. High efficiency is achieved by employing local IMLS in which fitting coefficients are stored at a limited number of expansion points, thus eliminating the need to perform weighted least-squares fits each time the potential is evaluated. An automatic point selection scheme based on the difference in two successive orders of IMLS fits is used to determine where new ab initio data need to be calculated for the most efficient fitting of the PES. A simple scan of the coordinate is shown to work well to identify these maxima in one dimension, but this search strategy scales poorly with dimension. We demonstrate the efficacy of using conjugate gradient minimizations on the difference surface to locate optimal data point placement in high dimensions. Results that are indicative of the accuracy, efficiency, and scalability are presented for a one-dimensional model potential (Morse) as well as for three-dimensional (HCN), six-dimensional (HOOH), and nine-dimensional (CH4) molecular PESs.     

Interpolating moving least-squares methods for fitting potential energy surfaces: computing high-density potential energy surface data from low-density ab initio data points

A highly accurate and efficient method for molecular global potential energy surface (PES) construction and fitting is demonstrated. An interpolating-moving-least-squares (IMLS)-based method is developed using low-density ab initio Hessian values to compute high-density PES parameters suitable for accurate and efficient PES representation. The method is automated and flexible so that a PES can be optimally generated for classical trajectories, spectroscopy, or other applications. Two important bottlenecks for fitting PESs are addressed. First, high accuracy is obtained using a minimal density of ab initio points, thus overcoming the bottleneck of ab initio point generation faced in applications of modified-Shepard-based methods. Second, high efficiency is also possible (suitable when a huge number of potential energy and gradient evaluations are required during a trajectory calculation). This overcomes the bottleneck in high-order IMLS-based methods, i.e., the high cost/accuracy ratio for potential energy evaluations. The result is a set of hybrid IMLS methods in which high-order IMLS is used with low-density ab initio Hessian data to compute a dense grid of points at which the energy, Hessian, or even high-order IMLS fitting parameters are stored. A series of hybrid methods is then possible as these data can be used for neural network fitting, modified-Shepard interpolation, or approximate IMLS. Results that are indicative of the accuracy, efficiency, and scalability are presented for one-dimensional model potentials as well as for three-dimensional (HCN) and six-dimensional (HOOH) molecular PESs.     

Interpolating moving least-squares methods for fitting potential energy surfaces: Analysis of an application to a six-dimensional system

The basic formal and numerical aspects of different degree interpolated moving least-squares (IMLS) methods are applied to a six-dimensional potential energy surface (PES) of the HOOH molecule, for which an analytic ("exact") potential is available in the literature. The results of systematic investigations of the effects of weight function parameters, the degree and partial degree of IMLS, the number of data points allowed, and the optimal automatic point selection of data points up to full third-degree IMLS fits are reported. With partial reduction of cross terms and automatic point selection the full six-dimensional HOOH PES can be fit over a range of 100 kcal/mol to an accuracy of less than 1 kcal/mol with approximately 1350 ab initio points.     

Interpolating moving least-squares methods for fitting potential energy surfaces: an application to the H2CN unimolecular reaction

Classical trajectories have been used to compute rates for the unimolecular reaction H2CN-->H+HCN on a fitted ab initio potential energy surface (PES). The ab initio energies were obtained from CCSD(T)/aug-cc-pvtz electronic structure calculations. The ab initio energies were fitted by the interpolating moving least-squares (IMLS) method. This work continues the development of the IMLS method for producing ab initio PESs for use in molecular dynamics simulations of many-atom systems. A dual-level scheme was used in which the preliminary selection of data points was done using a low-level theory and the points used for fitting the final PES were obtained at the desired higher level of theory. Classical trajectories were used on various low-level IMLS fits to tune the fit to the unimolecular reaction under study. Procedures for efficiently picking data points, selecting basis functions, and defining cutoff limits to exclude distant points were investigated. The accuracy of the fitted PES was assessed by comparing interpolated values of quantities to the corresponding ab initio values. With as little as 330 ab initio points classical trajectory rate constants were converged to 5%-10% and the rms error over the six-dimensional region sampled by the trajectories was a few tenths of a kcal/mol.     

Haptic quantum chemistry

We present an implementation designed to physically experience quantum mechanical forces between reactants in chemical reactions. This allows one to screen the profile of potential energy surfaces for the study of reaction mechanisms. For this, we have developed a interface between the user and a virtual laboratory by means of a force‐feedback haptic device. Potential energy surfaces of chemical reactions can be explored efficiently by rendering in the haptic device the gradients calculated with first‐principles methods. The underlying potential energy surface is accurately fitted on the fly by the interpolating moving least‐squares (IMLS) scheme to a grid of quantum chemical electronic energies (and geometric gradients). In addition, we introduce a new IMLS‐based method to locate minimum‐energy paths between two points on a potential energy surface. © 2009 Wiley Periodicals, Inc. J Comput Chem 2009     

Interpolating moving least-squares methods for fitting potential-energy surfaces: further improvement of efficiency via cutoff strategies

In standard applications of interpolating moving least squares (IMLS) for fitting a potential-energy surface (PES), all available ab initio points are used. Because remote ab initio points negligibly influence IMLS accuracy and increase IMLS time-to-solution, we present two methods to locally restrict the number of points included in a particular fit. The fixed radius cutoff (FRC) method includes ab initio points within a hypersphere of fixed radius. The density adaptive cutoff (DAC) method includes points within a hypersphere of variable radius depending on the point density. We test these methods by fitting a six-dimensional analytical PES for hydrogen peroxide. Both methods reduce the IMLS time-to-solution by about an order of magnitude relative to that when no cutoff method is used. The DAC method is more robust and efficient than the FRC method.     

A hierarchical construction scheme for accurate potential energy surface generation: an application to the F+H2 reaction

We present a hierarchical construction scheme for accurate ab initio potential energy surface generation. The scheme is based on the observation that when molecular configuration changes, the variation in the potential energy difference between different ab initio methods is much smaller than the variation for potential energy itself. This means that it is easier to numerically represent energy difference to achieve a desired accuracy. Because the computational cost for ab initio calculations increases very rapidly with the accuracy, one can gain substantial saving in computational time by constructing a high accurate potential energy surface as a sum of a low accurate surface based on extensive ab initio data points and an energy difference surface for high and low accuracy ab initio methods based on much fewer data points. The new scheme was applied to construct an accurate ground potential energy surface for the FH(2) system using the coupled-cluster method and a very large basis set. The constructed potential energy surface is found to be more accurate on describing the resonance states in the FH(2) and FHD systems than the existing surfaces.     

Nitrous oxide dimer: a new potential energy surface and rovibrational spectrum of the nonpolar isomer

The spectrum of nitrous oxide dimer was investigated by constructing new potential energy surfaces using coupled-cluster theory and solving the rovibrational Schro?dinger equation with a Lanczos algorithm. Two four-dimensional (rigid monomer) global ab initio potential energy surfaces (PESs) were made using an interpolating moving least-squares (IMLS) fitting procedure specialized to describe the interaction of two linear fragments. The first exploratory fit was made from 1646 CCSD(T)/3ZaP energies. Isomeric minima and connecting transition structures were located on the fitted surface, and the energies of those geometries were benchmarked using complete basis set (CBS) extrapolations, counterpoise (CP) corrections, and explicitly correlated (F12b) methods. At the geometries tested, the explicitly correlated F12b method produced energies in close agreement with the estimated CBS limit. A second fit to 1757 data at the CCSD(T)-F12b/VTZ-F12 level was constructed with an estimated fitting error of less than 1.5?cm(-1). The second surface has a global nonpolar O-in minimum, two T-shaped N-in minima, and two polar minima. Barriers between these minima are small and some wave functions have amplitudes in several wells. Low-lying rovibrational wave functions and energy levels up to about 150?cm(-1) were computed on the F12b PES using a discrete variable representation/finite basis representation method. Calculated rotational constants and intermolecular frequencies are in very close agreement with experiment.     

Gradient incorporation in one-dimensional applications of interpolating moving least-squares methods for fitting potential energy surfaces

We present several approaches to use gradients in higher degree interpolating moving least squares (IMLS) methods for representing a potential energy surface (PES). General procedures are developed to obtain smooth approximations of the PES and its derivatives from quasi-uniform sets of energy and gradient data points. These methods are illustrated and analyzed for the Morse oscillator and a 1-D slice of the ground-state PES for the HCO radical computed using density functional theory. Variations in the IMLS fits with the number and distribution of points and the degree of the polynomial fitting basis set are examined. We determine the effects of gradient inclusion on the accuracy of the IMLS values of the energy, first and second derivatives for two 1-D test cases. Gradient inclusion reduces the number of data points required by up to 40%.     

A local interpolation scheme using no derivatives in potential sampling: application to O(1D) + H2 system

We recently proposed a local interpolation scheme, in which interpolant moving least squares (IMLS) and Shepard interpolation are employed to describe potential energy surfaces. This IMLS/Shepard scheme is used to interpolate quantum chemical potential energy surfaces for which analytical derivatives are not available. In this study, we apply the scheme to the highly exothermic O((1)D) + H(2) --> H + OH reaction and compare it with results based on Shepard interpolation using second-order Taylor expansions. An analytical surface is used to define the potential function so that errors in the interpolation function may accurately be determined. We find that the present scheme reproduces the correct reactive cross-sections more accurately than the Shepard scheme, and with rms errors for energy and gradients that are significantly smaller than those from Shepard interpolation. This occurs even though the present scheme does not utilize derivative and Hessian information, whereas the Shepard interpolation does. The Bayesian approach proposed by Bettens and Collins does not improve the IMLS/Shepard results significantly, although it does the Shepard-only approach. The accuracy of the IMLS/Shepard scheme is surprising, but can be explained by the more global nature of the interpolation.     

多原子反应体系的高精度拟合势能面

本文介绍了近几年来我们组构建多原子反应体系的高精度拟合势能面的进展。我们基于神经网络(NN)方法,成功构建了多原子气相分子体系和气相分子在金属表面解离的一系列势能面。这些势能面的拟合精度相当高,并且经过了严格的量子动力学测试,能广泛应用到动力学研究中。我们还提出了一种新的置换不变势能面的拟合方法,即基本不变量神经网络方法(FI-NN)。基本不变量的使用极大地减少了神经网络输入层多项式的个数,有效提高了势能面的计算速度。     

Generation of potential energy surfaces in high dimensions and their haptic exploration

A method is proposed for the automated generation of potential energy surfaces in high dimensions. It combines the existing algorithm for the definition of new energy data points, based on the interpolating moving least-squares algorithm with a simulated annealing procedure. This method is then studied in a haptic quantum chemistry environment that requires a fast evaluation of gradients on a potential energy surface with automatic improvement of its accuracy. As an example we investigate the nitrogen binding pathway in the Schrock dinitrogen fixation complex with this set-up.     

Gaussian-4 theory using reduced order perturbation theory

Two modifications of Gaussian-4 (G4) theory [L. A. Curtiss et al., J. Chem. Phys. 126, 084108 (2007)] are presented in which second- and third-order perturbation theories are used in place of fourth-order perturbation theory. These two new methods are referred to as G4(MP2) and G4(MP3), respectively. Both methods have been assessed on the G3/05 test set of accurate experimental data. The average absolute deviation from experiment for the 454 energies in this test set is 1.04 kcalmol for G4(MP2) theory and 1.03 kcalmol for G4(MP3) theory compared to 0.83 kcalmol for G4 theory. G4(MP2) is slightly more accurate for enthalpies of formation than G4(MP3) (0.99 versus 1.04 kcalmol), while G4(MP3) is more accurate for ionization potentials and electron affinities. Overall, the G4(MP2) method provides an accurate and economical method for thermochemical predictions. It has an overall accuracy for the G3/05 test set that is much better than G3(MP2) theory (1.04 versus 1.39 kcalmol) and even better than G3 theory (1.04 versus 1.13 kcalmol). In addition, G4(MP2) does better for challenging hypervalent systems such as H(2)SO(4) and for nonhydrogen species than G3(MP2) theory.     

Least-squares numerical Rayleigh-Ritz and minimum-variance methods for molecular calculations

A general method is presented to find in a least-squares sense a set of orthogonal eigenfunctions and their eigenvalues from local energy and numerical integration methods or by any other dissymmetric approach to solve the eigenvalue problem of a Hermitian operator. By this method a generalization of the minimum variance method to more than one eigenfunction is obtained, which is a variant of Scott's method. Also a new method is derived—called the minimum-overlap method—that is a least-squares numerical version of the standard Rayleigh-Ritz method. Test calculations on the atoms Be and Tm and the molecules H₂ and CO have been performed with both numerical Hartree-Fock and Hartree-Fock-Slater methods. The least-squares solutions are an improvement over other methods in the case of accurate basis sets. Numerical Hartree-Fock calculations of moderate accuracy are found to be considerably faster than the analytic method.     

The Diels-Alder reaction of ethene and 1,3-butadiene: an extended multireference ab initio investigation.

The concerted and stepwise mechanisms of the Diels-Alder reaction between 1,3-butadiene and ethene have been investigated using highly correlated multireference methods (MRAQCC) and extended basis sets. Full MRAQCC geometry optimizations have been performed in all cases. The best estimate for the energy barrier of the Diels-Alder reaction is 22 kcalmol(-1). Anti- and gauche-out minima for the biradical structures and corresponding fragmentation saddle points have been determined. The biradical anti fragmentation saddle point is located 6.5 kcalmol(-1) above the concerted saddle point. The gauche-in structure does not correspond to a local minimum, but leads on geometry optimization directly to cyclohexene.     

Ab initio potential energy surface and quantum dynamics for the H + CH4 → H2 + CH3 reaction

A new full-dimensional potential energy surface for the title reaction has been constructed using the modified Shepard interpolation scheme. Energies and derivatives were calculated using the UCCSD(T) method with aug-cc-pVTZ and 6-311++G(3df,2pd) basis sets, respectively. A total number of 30,000 data points were selected from a huge number of molecular configurations sampled by trajectory method. Quantum dynamical calculations showed that the potential energy surface is well converged for the number of data points for collision energy up to 2.5 eV. Total reaction probabilities and integral cross sections were calculated on the present surface, as well as on the ZBB3 and EG-2008 surfaces for the title reaction. Satisfactory agreements were achieved between the present and the ZBB3 potential energy surfaces, indicating we are approaching the final stage to obtain a global potential energy surface of quantitative accuracy for this benchmark polyatomic system. Our calculations also showed that the EG-2008 surface is less accurate than the present and ZBB3 surfaces, particularly in high energy region.     

The X1 method for accurate and efficient prediction of heats of formation

We propose the X1 method which combines the density functional theory method with a neural network (NN) correction for an accurate yet efficient prediction of heats of formation. It calculates the final energy by using B3LYP6-311+G(3df,2p) at the B3LYP6-311+G(d,p) optimized geometry to obtain the B3LYP standard heats of formation at 298 K with the unscaled zero-point energy and thermal corrections at the latter basis set. The NN parameters cover 15 elements of H, Li, Be, B, C, N, O, F, Na, Mg, Al, Si, P, S, and Cl. The performance of X1 is close to the Gn theories, giving a mean absolute deviation of 1.43 kcalmol for the G399 set of 223 molecules up to 10 nonhydrogen atoms and 1.48 kcal/mol for the X107 set of 393 molecules up to 32 nonhydrogen atoms.     

Multireference configuration interaction calculations for the F(2P)+HCl-->HF+Cl(2P) reaction: a correlation scaled ground state (1 2A') potential energy surface

This paper presents a new ground state (1 (2)A(')) electronic potential energy surface for the F((2)P)+HCl-->HF+Cl((2)P) reaction. The ab initio calculations are done at the multireference configuration interaction+Davidson correction (MRCI+Q) level of theory by complete basis set extrapolation of the aug-cc-pVnZ (n=2,3,4) energies. Due to low-lying charge transfer states in the transition state region, the molecular orbitals are obtained by six-state dynamically weighted multichannel self-consistent field methods. Additional perturbative refinement of the energies is achieved by implementing simple one-parameter correlation energy scaling to reproduce the experimental exothermicity (DeltaE=-33.06 kcalmol) for the reaction. Ab initio points are fitted to an analytical function based on sum of two- and three-body contributions, yielding a rms deviation of <0.3 kcalmol for all geometries below 10 kcalmol above the barrier. Of particular relevance to nonadiabatic dynamics, the calculations show significant multireference character in the transition state region, which is located 3.8 kcalmol with respect to F+HCl reactants and features a strongly bent F-H-Cl transition state geometry (theta approximately 123.5 degrees ). Finally, the surface also exhibits two conical intersection seams that are energetically accessible at low collision energies. These seams arise naturally from allowed crossings in the C(infinityv) linear configuration that become avoided in C(s) bent configurations of both the reactant and product, and should be a hallmark of all X-H-Y atom transfer reaction dynamics between ((2)P) halogen atoms.     

Highly correlated particle densities and idempotent one-densities

The problem of determining idempotent one-densities which integrate to the exact or to a highly correlated particle density is considered. A method for obtaining the minimum energy idempotent one-density integrating to a given correlated particle density within a finite basis is described. The implications of this are twofold. First, Hartree–Fock accuracy can be exceeded in describing the electron density with an idempotent one-density; this is particularly relevant to the problem of constructing orbitals from experimental x-ray scattering data. Second, electron densities from analytic CI or MCSCF wave functions can be made available in a form as compact as the Hartree–Fock density by reporting the orbitals which define the correlated density via an idempotent one-density. A numerical example of the new method is given in which an accurate correlated density for He is “fitted” by an idempotent one-density represented in a finite (near Hartree–Fock) basis. Considering the deficiencies of the basis for this purpose, a technique is suggested for constructing basis sets optimized for prediction of one-electron properties rather than for energy.