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.     

Iterative solutions with energy selected bases for highly excited vibrations of tetra-atomic molecules

The use of energy selected bases (ESB) with iterative diagonalization of the Hamiltonian matrix is described for vibrations of tetra-atomic systems. The performance of the method is tested by computing vibrational states of HOOH below 10,000 cm(-1) (1296 A+ symmetry states) and H(2)CO below 13,500 cm(-1) (729 A(1) symmetry states). For iterative solutions, we tested both the implicitly restarted Lanczos method (IRLM) and the standard (nonreorthogonalizing) Lanczos approach. Comparison with other contracted basis approach as well as direct product grid representation shows superior performance of the ESB/IRLM approach. Of the two systems, H(2)CO is found to be more challenging than HOOH since it has much stronger couplings among vibrational modes, which leads to a drastically larger primitive basis set. For H(2)CO we also discuss some interesting behavior of the molecule in the high internal energy regime.     

Modified feed-forward neural network structures and combined-function-derivative approximations incorporating exchange symmetry for potential energy surface fitting

The classical interchange (permutation) of atoms of similar identity does not have an effect on the overall potential energy. In this study, we present feed-forward neural network structures that provide permutation symmetry to the potential energy surfaces of molecules. The new feed-forward neural network structures are employed to fit the potential energy surfaces for two illustrative molecules, which are H(2)O and ClOOCl. Modifications are made to describe the symmetric interchange (permutation) of atoms of similar identity (or mathematically, the permutation of symmetric input parameters). The combined-function-derivative approximation algorithm (J. Chem. Phys. 2009, 130, 134101) is also implemented to fit the neural-network potential energy surfaces accurately. The combination of our symmetric neural networks and the function-derivative fitting effectively produces PES fits using fewer numbers of training data points. For H(2)O, only 282 configurations are employed as the training set; the testing root-mean-squared and mean-absolute energy errors are respectively reported as 0.0103 eV (0.236 kcal/mol) and 0.0078 eV (0.179 kcal/mol). In the ClOOCl case, 1693 configurations are required to construct the training set; the root-mean-squared and mean-absolute energy errors for the ClOOCl testing set are 0.0409 eV (0.943 kcal/mol) and 0.0269 eV (0.620 kcal/mol), respectively. Overall, we find good agreements between ab initio and NN prediction in term of energy and gradient errors, and conclude that the new feed-forward neural-network models advantageously describe the molecules with excellent accuracy.     

Interpolating moving least-squares methods for fitting potential energy surfaces: Improving efficiency via local approximants

The local interpolating moving least-squares (IMLS) method for constructing potential energy surfaces is investigated. The method retains the advantageous features of the IMLS approach in that the ab initio derivatives are not required and high degree polynomials can be used to provide accurate fits, while at the same time it is much more efficient than the standard IMLS approach because the least-squares solutions need to be calculated only once at the data points. Issues related to the implementation of the local IMLS method are investigated and the accuracy is assessed using HOOH as a test case. It is shown that the local IMLS method is at the same level of accuracy as the standard IMLS method. In addition, the scaling of the method is found to be a power law as a function of number of data points N, N(-q). The results suggest that when fitting only to the energy values for a d-dimensional system by using a Qth degree polynomial the power law exponent q approximately Qd when the energy range fitted is large (e.g., E<100 kcalmol for HOOH), and q>Qd when the energy range fitted is smaller (E<30 kcalmol) and the density of data points is higher. This study demonstrates that the local IMLS method provides an efficient and accurate means for constructing potential energy surfaces.     

Low‐Temperature Thermal Rate Constants for the Water Formation Reaction H2+OH from Rigorous Quantum Dynamics Calculations

Thermal rate constants for the prototypical water‐forming reaction H₂+OH→H+H₂O were obtained for temperatures between 150 K and 600 K by rigorous quantum dynamics calculations including all degrees of freedom. Results are reported for a recent, highly accurate neural network potential (NN1) and compared to results obtained on a previous, semi‐empirical potential (SE). The rate constants computed on both potentials significantly differ in their temperature dependence, and differences of over one order of magnitude in the rates were found. The rate constants computed for the NN1 potential compare very well to experimental work. Furthermore, the influence of overall rotation is discussed for the title reaction. While previous close‐coupling simulations were limited to thermal rate constants above room temperature, we report rate constants for temperatures as low as 250 K. The high‐level results reported here provide an accurate benchmark for the development of approximate methods for the calculation of thermal as well as microcanonical rate constants.     

Accelerating the Construction of Neural Network Potential Energy Surfaces: A Fast Hybrid Training Algorithm

Machine learning approaches have been promising in constructing high-dimensional potential energy surfaces (PESs) for molecules and materials. Neural networks (NNs) are one of the most popular such tools because of its simplicity and efficiency. The training algorithm for NNs becomes essential to achieve a fast and accurate fit with numerous data. The Levenberg-Marquardt (LM) algorithm has been recognized as one of the fastest and robust algorithms to train medium sized NNs and widely applied in recent NN based high quality PESs. However, when the number of ab initio data becomes large, the efficiency of LM is limited, making the training time consuming. Extreme learning machine (ELM) is a recently proposed algorithm which determines the weights and biases of a single hidden layer NN by a linear solution and is thus extremely fast. It, however, does not produce sufficiently small fitting error because of its random nature. Taking advantages of both algorithms, we report a generalized hybrid algorithm in training multilayer NNs. Tests on H+H₂ and CH₄+Ni(111) systems demonstrate the much higher efficiency of this hybrid algorithm (ELM-LM) over the original LM. We expect that ELM-LM will find its widespread applications in building up high-dimensional NN based PESs.     

Ab initio potential energy and dipole moment surfaces for H5O2 +

Full-dimensional ab initio potential energy surface (PES) and dipole moment surface (DMS) are reported for H(5)O(2) (+). Tens of thousands of coupled-cluster [CCSD(T)] and second-order Moller-Plesset (MP2) calculations of electronic energies, using aug-cc-pVTZ basis, were done. The energies were fit very precisely in terms of all the internuclear distances, using standard least-square procedures, however, with a fitting basis that satisfies permutational symmetry with respect to like atoms. The H(5)O(2) (+) PES is a fit to 48 189 CCSD(T) energies, containing 7962 polynomial coefficients. The PES has a rms fitting error of 34.9 cm(-1) for the entire data set up to 110 000 cm(-1). This surface can describe various internal floppy motions, including the H atom exchanges, monomer inversions, and monomer torsions. First- and higher-order saddle points have been located on the surface and compared with available previous theoretical work. In addition, the PES dissociates correctly (and symmetrically) to H(2)O+H(3)O(+), with D(e)=11 923.8 cm(-1). Geometrical and vibrational properties of the monomer fragments are presented. The corresponding global DMS fit (MP2 based) involves 3844 polynomial coefficients and also dissociates correctly.     

Using neural networks to represent potential surfaces as sums of products

By using exponential activation functions with a neural network (NN) method we show that it is possible to fit potentials to a sum-of-products form. The sum-of-products form is desirable because it reduces the cost of doing the quadratures required for quantum dynamics calculations. It also greatly facilitates the use of the multiconfiguration time dependent Hartree method. Unlike potfit product representation algorithm, the new NN approach does not require using a grid of points. It also produces sum-of-products potentials with fewer terms. As the number of dimensions is increased, we expect the advantages of the exponential NN idea to become more significant.     

Ab initio treatment of the chemical reaction precursor complex Br(2P)-HCN. 1. Adiabatic and diabatic potential surfaces

The three adiabatic potential surfaces of the Br(2P)-HCN complex that correlate to the 2P ground state of the Br atom were calculated ab initio. With the aid of a geometry-dependent diabatic mixing angle, also calculated ab initio, these adiabatic potential surfaces were transformed into a set of four diabatic potential surfaces required to define the full 3 x 3 matrix of diabatic potentials. Each of these diabatic potential surfaces was expanded in terms of the appropriate spherical harmonics in the atom-linear molecule Jacobi angle theta. The dependence of the expansion coefficients on the distance R between Br and the HCN center of mass and on the CH bond length was fit to an analytic form. For HCN in its equilibrium geometry, the global minimum with De = 800.4 cm(-1) and Re = 6.908a0 corresponds to a linear Br-NCH geometry, with an electronic ground state of Sigma symmetry. A local minimum with De = 415.1 cm-1, Re = 8.730a0, and a twofold degenerate Pi ground state is found for the linear Br-HCN geometry. The binding energy, De, depends strongly on the CH bond length for the Br-HCN complex and much less strongly for the Br-NCH complex, with a longer CH bond giving stronger binding for both complexes. Spin-orbit coupling was included and diabatic states were constructed that correlate to the ground 2P3/2 and excited 2P1/2 spin-orbit states of the Br atom. For the ground spin-orbit state with electronic angular momentum j = (3/2) the minimum in the potential for projection quantum number omega = +/-(3/2) coincides with the local minimum for linear Br-HCN of the spin-free case. The minimum in the potential for projection quantum number omega = +/-(1/2) occurs for linear Br-NCH but is considerably less deep than the global minimum of the spin-free case. According to the lowest spin-orbit coupling included adiabatic potential the two linear isomers, Br-NCH and Br-HCN, are about equally stable. In the subsequent paper, we use these potentials in calculations of the rovibronic states of the Br-HCN complex.     

Ultrafast vibrational spectroscopy of water and aqueous N-methylacetamide: Comparison of different electronic structure/molecular dynamics approaches

Kwac and Cho [J. Chem. Phys. 119, 2247 (2003)] have recently developed a combined electronic structure/molecular dynamics approach to vibrational spectroscopy in liquids. The method involves fitting ab initio vibrational frequencies for a solute in a cluster of solvent molecules to a linear combination of the electrostatic potentials on the solute atoms due to the charges on the solvent molecules. These authors applied their method to the N-methylacetamide-D/D(2)O system. We (S. A. Corcelli, C. P. Lawrence, and J. L. Skinner, [J. Chem. Phys. 120, 8107 (2004)]) have recently explored a closely related method, where instead of the electrostatic potential, the solute vibrational frequencies are fit to the components of the electric fields on the solute atoms due to the solvent molecules. We applied our method to the HOD/D(2)O and HOD/H(2)O systems. In order to make a direct comparison of these two approaches, in this paper we apply their method to the water system, and our method to the N-methylacetamide system. For the water system we find that the electric field method is superior to the potential approach, as judged by comparison with experiments for the absorption line shape. For the N-methylacetamide system the two methods are comparable.     

Calibration-quality adiabatic potential energy surfaces for H3(+) and its isotopologues

Calibration-quality ab initio adiabatic potential energy surfaces (PES) have been determined for all isotopologues of the molecular ion H(3)(+). The underlying Born-Oppenheimer electronic structure computations used optimized explicitly correlated shifted Gaussian functions. The surfaces include diagonal Born-Oppenheimer corrections computed from the accurate electronic wave functions. A fit to the 41,655 ab initio points is presented which gives a standard deviation better than 0.1 cm(-1) when restricted to the points up to 6000 cm(-1) above the first dissociation asymptote. Nuclear motion calculations utilizing this PES, called GLH3P, and an exact kinetic energy operator given in orthogonal internal coordinates are presented. The ro-vibrational transition frequencies for H(3)(+), H(2)D(+), and HD(2)(+) are compared with high resolution measurements. The most sophisticated and complete procedure employed to compute ro-vibrational energy levels, which makes explicit allowance for the inclusion of non-adiabatic effects, reproduces all the known ro-vibrational levels of the H(3)(+) isotopologues considered to better than 0.2 cm(-1). This represents a significant (order-of-magnitude) improvement compared to previous studies of transitions in the visible. Careful treatment of linear geometries is important for high frequency transitions and leads to new assignments for some of the previously observed lines. Prospects for further investigations of non-adiabatic effects in the H(3)(+) isotopologues are discussed. In short, the paper presents (a) an extremely accurate global potential energy surface of H(3)(+) resulting from high accuracy ab initio computations and global fit, (b) very accurate nuclear motion calculations of all available experimental line data up to 16,000 cm(-1), and (c) results suggest that we can predict accurately the lines of H(3)(+) towards dissociation and thus facilitate their experimental observation.     

Comparison of hydrated hydroperoxide anion (HOO-)(H2O)n clusters with alkaline hydrogen peroxide (HOOH)(OH-)(H2O)(n-1) clusters, n = 1-8, 20: an ab initio study

Hydroperoxide anion (HOO(-)), the conjugate base of hydrogen peroxide (HOOH), has been relatively little studied despite the importance of HOOH in commercial processes, atmospheric science, and biology. The anion has been shown to exist as a stable species in alkaline water. This project explored the structure of gas phase (HOO(-))(H(2)O)(n) clusters and identified the lowest energy configurations for n ≤ 8 at the B3LYP/6-311++G** level of theory and for n ≤ 6 at the MP2/aug-cc-pVTZ level of theory. As a start toward understanding equilibration between HOO(-) and HOOH in an alkaline environment, (HOOH)(OH(-))(H(2)O)(n-1) clusters were likewise examined, and the lowest energy configurations were determined for n ≤ 8 (B3LYP/6-311++G**) and n ≤ 6 (MP2/aug-cc-pVTZ). Some studies were also done for n = 20. The two species have very different solvation behaviors. In low energy (HOOH)(OH(-))(H(2)O)(n-1) clusters, HOOH sits on the surface of the cluster, is 4-coordinated (each O is donor once and acceptor once), and donates to the hydroxide ion. In contrast, in low energy (HOO(-))(H(2)O)(n) clusters, (HOO(-)) takes a position in the cluster center surrounded on all sides by water molecules, and its optimum coordination number appears to be 7 (one O is donor-acceptor-acceptor while the other is a 4-fold acceptor). For n ≤ 6 the lowest (HOOH)(OH(-))(H(2)O)(n-1) cluster lies 1.0-2.1 kcal/mol below the lowest (HOO(-))(H(2)O)(n) cluster, but the lowest clusters found for n = 20 favor (HOO(-))(H(2)O)(20). The results suggest that ambient water could act as a substantial kinetic brake that slows equilibration between (HOOH)(OH(-)) and (HOO(-))(H(2)O) because extensive rearrangement of solvation shells is necessary to restabilize either species after proton transfer.     

Molecular dynamics investigations of the dissociation of SiO2 on an ab initio potential energy surface obtained using neural network methods

The neural network (NN) procedure to interpolate ab initio data for the purpose of molecular dynamics (MD) simulations has been tested on the SiO(2) system. Unlike other similar NN studies, here, we studied the dissociation of SiO(2) without the initial use of any empirical potential. During the dissociation of SiO(2) into Si+O or Si+O(2), the spin multiplicity of the system changes from singlet to triplet in the first reaction and from singlet to pentet in the second. This paper employs four potential surfaces. The first is a NN fit [NN(STP)] to a database comprising the lowest of the singlet, triplet, and pentet energies obtained from density functional calculations in 6673 nuclear configurations. The other three potential surfaces are obtained from NN fits to the singlet, triplet, and pentet-state energies. The dissociation dynamics on the singlet-state and NN(STP) surfaces are reported. The results obtained using the singlet surface correspond to those expected if the reaction were to occur adiabatically. The dynamics on the NN(STP) surface represent those expected if the reaction follows a minimum-energy pathway. This study on a small system demonstrates the application of NNs for MD studies using ab initio data when the spin multiplicity of the system changes during the dissociation process.     

Theoretical Study on Mechanism of Reaction of OH with HO2NO2

The reaction of HO2NO2 (peroxynitric acid, PNA) with OH was studied by the hybrid density functional B3LYP and CBS-QB3 methods. Based on the calculated potential energy surface, five reaction channels, H2O+NO2+O2, HOOH+NO3, NO2+HO3H, HO2+HONO2 and HO2+HOONO, were examined in detail. The major reaction channel is PNA+OH→M1→TS1→H2O+NO2+O2. Taking a pre-equilibrium approximation and using the CBS-QB3 energies, the theoretical rate constant of this channel was calculated as 1.13×10-12 cm3/(molecule s) at 300 K, in agreement with the experimental result. Comparison between reactions of HOONO2+OH and HONO2+OH was carried out. For HOR+OH reactions, the total rate constants increase from R=NO2 to R=ONO2, which is consistent with experimental measurements.     

Potential energy surface and rovibrational calculations for the Mg+-H2 and Mg+-D2 complexes

A three-dimensional potential energy surface is developed to describe the structure and dynamical behavior of the Mg(+)-H(2) and Mg(+)-D(2) complexes. Ab initio points calculated using the RCCSD(T) method and aug-cc-pVQZ basis set (augmented by bond functions) are fitted using a reproducing kernel Hilbert space method [Ho and Rabitz, J. Chem. Phys. 104, 2584 (1996)] to generate an analytical representation of the potential energy surface. The calculations confirm that Mg(+)-H(2) and Mg(+)-D(2) essentially consist of a Mg(+) atomic cation attached, respectively, to a moderately perturbed H(2) or D(2) molecule in a T-shaped configuration with an intermolecular separation of 2.62 A? and a well depth of D(e) = 842 cm(-1). The barrier for internal rotation through the linear configuration is 689 cm(-1). Interaction with the Mg(+) ion is predicted to increase the H(2) molecule's bond-length by 0.008 A?. Variational rovibrational energy level calculations using the new potential energy surface predict a dissociation energy of 614 cm(-1) for Mg(+)-H(2) and 716 cm(-1) for Mg(+)-D(2). The H-H and D-D stretch band centers are predicted to occur at 4059.4 and 2929.2 cm(-1), respectively, overestimating measured values by 3.9 and 2.6 cm(-1). For Mg(+)-H(2) and Mg(+)-D(2), the experimental B and C rotational constants exceed the calculated values by ～1.3%, suggesting that the calculated potential energy surface slightly overestimates the intermolecular separation. An ab initio dipole moment function is used to simulate the infrared spectra of both complexes.     

Reaction dynamics simulations of the identity S(N)2 reaction H(2)O + HOOH(2)(+)--> H(2)OOH(+)+ H(2)O. Requirements for reaction and competition with proton transfer

Despite the fact that the transition structure of the gas phase S(N)2 reaction H(2)O + HOOH(2)(+)--> HOOH(2)(+)+ H(2)O is well below the reactants in potential energy, the reaction has not yet been observed by experiment. Variational transition state RRKM theory reveals a strong preference for the competing proton transfer reaction H(2)O + HOOH(2)(+)--> H(3)O(+)+ HOOH due to entropy factors. Born-Oppenheimer reaction dynamics simulations confirm these results. However, by increasing the collision energy to around 7.5 eV the probability for nucleophilic substitution increases relative to proton transfer. These observations are explained by the presence of the key common intermediate HOO(H)[dot dot dot]H-OH(2)(+) which leads to effective proton transfer, but can be avoided with increasing collision energy. However, the S(N)2 probability remains below 0.2 since successful passage through the TS requires optimum initial orientation of the reactants, excitation of the relative translational motion and good phase correlation between the O-O vibration and the motion of the incoming water.     

Ab initio potential energy and dipole moment surfaces of (H2O)2

A full-dimensional ab initio potential energy surface (PES) and dipole moment surface (DMS) are reported for the water dimer, (H2O)2. The CCSD(T)-PES is a very precise fit to 19,805 ab initio energies obtained with the coupled-cluster (CCSD(T)) method, using an aug-cc-pVTZ basis. The standard counterpoise correction was applied to approximately eliminate basis set superposition errors. The fit is based on an approach that incorporates the permutational symmetry of identical atoms [Huang, X.; Braams, B.; Bowman, J. M. J. Chem.Phys. 2005, 122, 044308]. The DMS is a fit to the dipole moment obtained with M?ller-Plesset (MP2) theory, using an aug-cc-pVTZ basis. The PES has an RMS fitting error of 31 cm(-1) for energies below 20,000 cm(-1), relative to the global minimum. This surface can describe various internal floppy motions, including various monomer inversions, and isomerization pathways. Ten characteristic stationary points have been located on the surface, four of which are transition structures and the rest are higher order saddle points. Their geometrical and vibrational properties are presented and compared with available previous theoretical work. The CCSD(T)-PES and MP2-DMS dissociate correctly (and symmetrically) to two H2O monomers, with D(e) = 1665.7 cm(-1) (19.93 kJ/mol). Accurate quantum calculations of the zero-point energy of the dimer (using diffusion Monte Carlo) and the monomers (using a vibrational configuration interaction approach) are reported, and these together with D(e) give a value of D0 of 1042 cm(-1) (12.44 kJ/mol). A best estimated value is 1130 cm(-1) (13.5 kJ/mol).     

Cis-->trans, trans-->cis isomerizations and N-O bond dissociation of nitrous acid (HONO) on an ab initio potential surface obtained by novelty sampling and feed-forward neural network fitting

The isomerization and dissociation dynamics of HONO are investigated on an ab initio potential surface obtained by fitting the results of electronic structure calculations at 21 584 configurations by using previously described novelty sampling and feed-forward neural network (NN) methods. The electronic structure calculations are executed by using GAUSSIAN 98 with a 6-311G(d) basis set at the MP4(SDQ) level of accuracy. The average absolute error of the NN fits varies from 0.012 eV (1.22 kJ mol(-1)) to 0.017 eV (1.64 kJ mol(-1)). The average computation time for a HONO trajectory using a single NN surface is approximately 4.8 s. These computation times compare very favorably with those required by other methods primarily because the NN fitting needs to be executed only one time rather than at every integration point. If the average result obtained from a committee of NNs is employed at each point rather than a single NN, increased fitting accuracy can be achieved at the expense of increased computational requirements. In the present investigation, we find that a committee comprising five NN potentials reduces the average absolute interpolation error to 0.0111 eV (1.07 kJ mol(-1)). Cis-trans isomerization rates with total energy of 1.70 eV (including zero point energy) have been computed for a variety of different initial distributions of the internal energy. In contrast to results previously reported by using an empirical potential, where cis-->trans to trans-->cis rate coefficient ratios at 1.70 eV total energy were found to lie in the range of 2.0-12.9 depending on the vibration mode excited, these ratios on the ab initio NN potential lie in the range of 0.63-1.94. It is suggested that this result is a reflection of much larger intramode coupling terms present in the ab initio potential surface. A direct consequence of this increased coupling is a significant decrease in the mode specific rate enhancement when compared to results obtained by using empirical surfaces. All isomerizations are found to be first order in accordance with the results reported by using empirical potentials. The dissociation rate to NO+OH has been investigated at internal HONO energies of 3.10 and 3.30 eV for different distributions of this energy among the six vibrational modes of HONO. These dissociations are also found to be first order. The computed dissociation rate coefficients exhibit only modest mode specific rate enhancement that is significantly smaller than that obtained on an empirical surface because of the much larger mode couplings present on the ab initio surface.