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.

[1]  Joel M. Bowman,et al.  Vibrational self-consistent field method for many-mode systems: A new approach and application to the vibrations of CO adsorbed on Cu(100) , 1997 .

[2]  Nicholas C. Handy,et al.  On the representation of potential energy surfaces of polyatomic molecules in normal coordinates , 2002 .

[3]  T. Carrington,et al.  Six-dimensional variational calculation of the bending energy levels of HF trimer and DF trimer , 2001 .

[4]  Hua Guo,et al.  Six-dimensional quantum calculations of highly excited vibrational energy levels of hydrogen peroxide and its deuterated isotopomers , 2001 .

[5]  H. Meyer,et al.  Reactive scattering using the multiconfiguration time‐dependent Hartree approximation: General aspects and application to the collinear H+H2→H2+H reaction , 1995 .

[6]  H. Meyer,et al.  Theoretical investigation of intramolecular vibrational energy redistribution in highly excited HFCO. , 2006, The Journal of chemical physics.

[7]  H. Rabitz,et al.  General foundations of high‐dimensional model representations , 1999 .

[8]  Didier Lemoine,et al.  Pseudospectral method for solving the time-dependent Schrödinger equation in spherical coordinates , 1992 .

[9]  G. Rauhut.  Efficient calculation of potential energy surfaces for the generation of vibrational wave functions. , 2004, The Journal of chemical physics.

[10]  Joel M Bowman,et al.  Quantum studies of the vibrations in H3O2- and D3O2-. , 2005, The Journal of chemical physics.

[11]  H. Rabitz,et al.  Efficient input-output model representations , 1999 .

[12]  Hans-Dieter Meyer,et al.  Rotational excitation cross sections of para-H2 + para-H2 collisions. A full-dimensional wave-packet propagation study using an exact form of the kinetic energy. , 2005, The Journal of chemical physics.

[13]  E. Schmidt.  Zur Theorie der linearen und nichtlinearen Integralgleichungen , 1907 .

[14]  Francesco Piazza,et al.  Learning and Approximation Capabilities of Adaptive Spline Activation Function Neural Networks , 1998, Neural Networks.

[15]  P. Schleyer.  Encyclopedia of computational chemistry , 1998 .

[16]  T. Carrington,et al.  A general discrete variable method to calculate vibrational energy levels of three‐ and four‐atom molecules , 1993 .

[17]  G. Kroes,et al.  Multiconfiguration time-dependent Hartree method applied to molecular dissociation on surfaces: H2 + Pt(111). , 2006, The Journal of chemical physics.

[18]  Ken-ichi Funahashi,et al.  On the approximate realization of continuous mappings by neural networks , 1989, Neural Networks.

[19]  J. Tennyson.  The calculation of the vibration-rotation energies of triatomic molecules using scattering coordinates , 1986 .

[20]  P. Kollman,et al.  Encyclopedia of computational chemistry , 1998 .

[21]  Joel M. Bowman,et al.  Variational calculations of rovibrational energies of CH4 and isotopomers in full dimensionality using an ab initio potential , 1999 .

[22]  Claude Leforestier,et al.  Grid representation of rotating triatomics , 1991 .

[23]  Hans-Dieter Meyer,et al.  Product representation of potential energy surfaces. II , 1996 .

[24]  R. Wyatt,et al.  Dynamics of molecules and chemical reactions , 1996 .

[25]  R. Kosloff.  Time-dependent quantum-mechanical methods for molecular dynamics , 1988 .

[26]  Francesco Piazza,et al.  Multilayer feedforward networks with adaptive spline activation function , 1999, IEEE Trans. Neural Networks.

[27]  N. Handy,et al.  Extensions and tests of “multimode”: a code to obtain accurate vibration/rotation energies of many-mode molecules , 1998 .

[28]  William H. Miller,et al.  Beyond transition-state theory: A rigorous quantum theory of chemical reaction rates , 1993 .

[29]  M. Beck,et al.  The multiconfiguration time-dependent Hartree (MCTDH) method: A highly efficient algorithm for propa , 1999 .

[30]  Allan Pinkus,et al.  Lower bounds for approximation by MLP neural networks , 1999, Neurocomputing.

[31]  Lorenz S. Cederbaum,et al.  Multiconfigurational time‐dependent Hartree study of complex dynamics: Photodissociation of NO2 , 1992 .

[32]  Donald G Truhlar,et al.  Calculation of converged rovibrational energies and partition function for methane using vibrational-rotational configuration interaction. , 2004, The Journal of chemical physics.

[33]  H. Rabitz,et al.  High Dimensional Model Representations , 2001 .

[34]  Tucker Carrington,et al.  Efficient calculation of highly excited vibrational energy levels of floppy molecules: The band origins of H+3 up to 35 000 cm−1 , 1994 .

[35]  William H. Press,et al.  Numerical Recipes: FORTRAN , 1988 .

[36]  Hans-Dieter Meyer,et al.  Intramolecular vibrational energy redistribution in toluene: a nine-dimensional quantum mechanical study using the MCTDH algorithm , 2004 .

[37]  N. Handy,et al.  The variational method for the calculation of RO-vibrational energy levels , 1986 .

[38]  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.

[39]  J. J. Soares Neto,et al.  The fitting of potential energy surfaces using neural networks: Application to the study of vibrational levels of H3+ , 1998 .

[40]  T. Carrington,et al.  Discrete‐Variable Representations and their Utilization , 2007 .

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

[42]  Aurelio Uncini,et al.  Regularising neural networks using flexible multivariate activation function , 2004, Neural Networks.

[43]  Joel M Bowman,et al.  Full-dimensional vibrational calculations for H5O2+ using an ab initio potential energy surface. , 2005, The Journal of chemical physics.

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

[45]  Alexander N. Gorban,et al.  Approximation of continuous functions of several variables by an arbitrary nonlinear continuous function of one variable, linear functions, and their superpositions , 1998 .

[46]  B. Kuhn,et al.  A new six-dimensional analytical potential up to chemically significant energies for the electronic ground state of hydrogen peroxide , 1999 .

[47]  Peter L. Bartlett,et al.  Neural Network Learning - Theoretical Foundations , 1999 .

[48]  Ranga Komanduri,et al.  Molecular dynamics investigations of the dissociation of SiO2 on an ab initio potential energy surface obtained using neural network methods. , 2006, The Journal of chemical physics.

[49]  R. Saykally,et al.  Fully coupled six-dimensional calculations of the water dimer vibration-rotation-tunneling states with a split Wigner pseudo spectral approach , 1997 .

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