Accurate prediction of higher-level electronic structure energies for large databases using neural networks, Hartree-Fock energies, and small subsets of the database.

A novel method is presented that significantly reduces the computational bottleneck of executing high-level, electronic structure calculations of the energies and their gradients for a large database that adequately samples the configuration space of importance for systems containing more than four atoms that are undergoing multiple, simultaneous reactions in several energetically open channels. The basis of the method is the high-degree of correlation that generally exists between the Hartree-Fock (HF) and higher-level electronic structure energies. It is shown that if the input vector to a neural network (NN) includes both the configuration coordinates and the HF energies of a small subset of the database, MP4(SDQ) energies with the same basis set can be predicted for the entire database using only the HF and MP4(SDQ) energies for the small subset and the HF energies for the remainder of the database. The predictive error is shown to be less than or equal to the NN fitting error if a NN is fitted to the entire database of higher-level electronic structure energies. The general method is applied to the computation of MP4(SDQ) energies of 68,308 configurations that comprise the database for the simultaneous, unimolecular decomposition of vinyl bromide into six different reaction channels. The predictive accuracy of the method is investigated by employing successively smaller subsets of the database to train the NN to predict the MP4(SDQ) energies of the remaining configurations of the database. The results indicate that for this system, the subset can be as small as 8% of the total number of configurations in the database without loss of accuracy beyond that expected if a NN is employed to fit the higher-level energies for the entire database. The utilization of this procedure is shown to save about 78% of the total computational time required for the execution of the MP4(SDQ) calculations. The sampling error involved with selection of the subset is shown to be about 10% of the predictive error for the higher-level energies. A practical procedure for utilization of the method is outlined. It is suggested that the method will be equally applicable to the prediction of electronic structure energies computed using even higher-level methods than MP4(SDQ).

[1]  Sergei Manzhos,et al.  Using neural networks to represent potential surfaces as sums of products. , 2006, The Journal of chemical physics.

[2]  Sergei Manzhos,et al.  A random-sampling high dimensional model representation neural network for building potential energy surfaces. , 2006, The Journal of chemical physics.

[3]  H. Rabitz,et al.  A global H2O potential energy surface for the reaction O(1D)+H2→OH+H , 1996 .

[4]  Mark N. Gibbs,et al.  Combining ab initio computations, neural networks, and diffusion Monte Carlo: An efficient method to treat weakly bound molecules , 1996 .

[5]  T. Carrington,et al.  A nested molecule-independent neural network approach for high-quality potential fits. , 2006, The journal of physical chemistry. A.

[6]  Kersti Hermansson,et al.  Representation of Intermolecular Potential Functions by Neural Networks , 1998 .

[7]  Roger M. Smith,et al.  Applications of genetic algorithms and neural networks to interatomic potentials , 1999 .

[8]  Michael A. Collins,et al.  Potential energy surfaces and dynamics for the reactions between C(3P) and H3+(1A1′) , 1998 .

[9]  R Komanduri,et al.  A self-starting method for obtaining analytic potential-energy surfaces from ab initio electronic structure calculations. , 2009, The journal of physical chemistry. A.

[10]  Donald L. Thompson,et al.  Interpolating moving least-squares methods for fitting potential energy surfaces: Detailed analysis of one-dimensional applications , 2003 .

[11]  J. Mason,et al.  Algorithms for approximation , 1987 .

[12]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[13]  Sergei Manzhos,et al.  Using neural networks, optimized coordinates, and high-dimensional model representations to obtain a vinyl bromide potential surface. , 2008, The Journal of chemical physics.

[14]  Steven D. Brown,et al.  Data processing using neural networks , 1993 .

[15]  R Komanduri,et al.  Ab initio potential-energy surfaces for complex, multichannel systems using modified novelty sampling and feedforward neural networks. , 2005, The Journal of chemical physics.

[16]  George C. Schatz,et al.  Automatic potential energy surface generation directly from ab initio calculations using Shepard interpolation: A test calculation for the H2+H system , 1997 .

[17]  Michael A. Collins,et al.  Molecular potential-energy surfaces by interpolation: Furtherrefinements , 1997 .

[18]  Donald L Thompson,et al.  Interpolating moving least-squares methods for fitting potential energy surfaces: an application to the H2CN unimolecular reaction. , 2007, The Journal of chemical physics.

[19]  Joel M Bowman,et al.  Ab initio potential energy and dipole moment surfaces for H5O2 +. , 2005, The Journal of chemical physics.

[20]  Richard Dawes,et al.  Interpolating moving least-squares methods for fitting potential energy surfaces: computing high-density potential energy surface data from low-density ab initio data points. , 2007, The Journal of chemical physics.

[21]  M. Malshe,et al.  Development of generalized potential-energy surfaces using many-body expansions, neural networks, and moiety energy approximations. , 2009, The Journal of chemical physics.

[22]  Michael A. Collins,et al.  Molecular potential-energy surfaces for chemical reaction dynamics , 2002 .

[23]  B. Braams,et al.  Quasiclassical trajectory calculations of the OH+NO2 association reaction on a global potential energy surface. , 2007, The Journal of chemical physics.

[24]  L. Raff,et al.  Molecular dissociation of hydrogen peroxide (HOOH) on a neural network ab initio potential surface with a new configuration sampling method involving gradient fitting. , 2009, The Journal of chemical physics.

[25]  Alex Brown,et al.  Classical and quasiclassical spectral analysis of CH5+ using an ab initio potential energy surface , 2003 .

[26]  R. P. Bettens,et al.  Potential energy surface for the reactions BeH2 + H BeH + H2 , 1999 .

[27]  A. Gross,et al.  Representing high-dimensional potential-energy surfaces for reactions at surfaces by neural networks , 2004 .

[28]  Quantum calculations of vibrational energies of H3O2- on an ab initio potential. , 2004, Journal of the American Chemical Society.

[29]  A. Varandas,et al.  Intermolecular and Intramolecular Potentials: Topographical Aspects, Calculation, and Functional Representation via A Double Many‐Body Expansion Method , 2007 .

[30]  Akio Kawano,et al.  Improving the accuracy of interpolated potential energy surfaces by using an analytical zeroth-order potential function. , 2004, The Journal of chemical physics.

[31]  B. Braams,et al.  Ab initio global potential-energy surface for H5(+) --> H3(+) + H2. , 2005, The Journal of chemical physics.

[32]  Donald L Thompson,et al.  Interpolating moving least-squares methods for fitting potential energy surfaces: Improving efficiency via local approximants. , 2007, The Journal of chemical physics.

[33]  M. A. Collins,et al.  Molecular potential energy surfaces by interpolation , 1994 .

[34]  Akio Kawano,et al.  Interpolating moving least-squares methods for fitting potential energy surfaces: Analysis of an application to a six-dimensional system. , 2004, The Journal of chemical physics.

[35]  Richard Dawes,et al.  Interpolating moving least-squares methods for fitting potential energy surfaces: a strategy for efficient automatic data point placement in high dimensions. , 2008, The Journal of chemical physics.

[36]  Michael A. Collins,et al.  Interpolated potential energy surface and dynamics for the reactions between N(4S) and H3+(1A1′) , 1998 .

[37]  G Reibnegger,et al.  Neural networks as a tool for compact representation of ab initio molecular potential energy surfaces. , 1996, Journal of Molecular Graphics.

[38]  A. Pukrittayakamee,et al.  Simultaneous fitting of a potential-energy surface and its corresponding force fields using feedforward neural networks. , 2009, The Journal of chemical physics.

[39]  L. Raff,et al.  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. , 2008, The Journal of chemical physics.

[40]  Michael A. Collins,et al.  Interpolated potential energy surface and reaction dynamics for O(3P)+H3+(1A1′) and OH+(3Σ−)+H2(1Σg+) , 1999 .

[41]  Michael A. Collins,et al.  Learning to interpolate molecular potential energy surfaces with confidence: A Bayesian approach , 1999 .

[42]  Martin T. Hagan,et al.  Neural network design , 1995 .

[43]  M. Malshe,et al.  Theoretical investigation of the dissociation dynamics of vibrationally excited vinyl bromide on an ab initio potential-energy surface obtained using modified novelty sampling and feedforward neural networks. II. Numerical application of the method. , 2007, The Journal of chemical physics.

[44]  Akio Kawano,et al.  Interpolating moving least-squares methods for fitting potential energy surfaces: applications to classical dynamics calculations. , 2004, The Journal of chemical physics.

[45]  A. Varandas,et al.  Accurate double many-body expansion potential energy surface for triplet H[sub 3][sup +]. I. The lowest adiabatic sheet (a[sup 3]Σ[sub u][sup +]) , 2003 .