Machine Learning Configuration Interaction.

We propose the concept of machine learning configuration interaction (MLCI) whereby an artificial neural network is trained on-the-fly to predict important new configurations in an iterative selected configuration interaction procedure. We demonstrate that the neural network can discriminate between important and unimportant configurations, that it has not been trained on, much better than by chance. MLCI is then used to find compact wave functions for carbon monoxide at both stretched and equilibrium geometries. We also consider the multireference problem of the water molecule with elongated bonds. Results are contrasted with those from other ways of selecting configurations: first-order perturbation, random selection, and Monte Carlo configuration interaction. Compared with these other serial calculations, this prototype MLCI is competitive in its accuracy, converges in significantly fewer iterations than the stochastic approaches, and requires less time for the higher-accuracy computations.

[1]  Alexandre Tkatchenko,et al.  Quantum-chemical insights from deep tensor neural networks , 2016, Nature Communications.

[2]  Andrew G. Taube,et al.  Improving the accuracy of Møller-Plesset perturbation theory with neural networks. , 2017, The Journal of chemical physics.

[3]  G. Lewicki,et al.  Approximation by Superpositions of a Sigmoidal Function , 2003 .

[4]  J. Greer.  Monte Carlo Configuration Interaction , 1998 .

[5]  R. Bartlett,et al.  Monte Carlo configuration interaction with perturbation corrections for dissociation energies of first row diatomic molecules: C2, N2, O2, CO, and NO. , 2014, The Journal of chemical physics.

[6]  J. P. Coe,et al.  State-averaged Monte Carlo configuration interaction applied to electronically excited states. , 2013, The Journal of chemical physics.

[7]  M. Nolan,et al.  A Monte Carlo configuration generation computer program for the calculation of electronic states of atoms, molecules, and quantum dots , 2000 .

[8]  Heather J Kulik,et al.  Predicting electronic structure properties of transition metal complexes with neural networks† †Electronic supplementary information (ESI) available. See DOI: 10.1039/c7sc01247k , 2017, Chemical science.

[9]  Kyle Mills,et al.  Deep learning and the Schrödinger equation , 2017, ArXiv.

[10]  J. P. Coe,et al.  Multireference X-ray emission and absorption spectroscopy calculations from Monte Carlo configuration interaction , 2015, Theoretical Chemistry Accounts.

[11]  J P Coe,et al.  Development of Monte Carlo configuration interaction: natural orbitals and second-order perturbation theory. , 2012, The Journal of chemical physics.

[12]  Kun Yao,et al.  Kinetic Energy of Hydrocarbons as a Function of Electron Density and Convolutional Neural Networks. , 2015, Journal of chemical theory and computation.

[13]  J. P. Coe,et al.  Applying Monte Carlo configuration interaction to transition metal dimers: Exploring the balance between static and dynamic correlation , 2014, 1405.5709.

[14]  Robert J. Harrison,et al.  Approximating full configuration interaction with selected configuration interaction and perturbation theory , 1991 .

[15]  J. Greer,et al.  Quasiparticle energies and lifetimes in a metallic chain model of a tunnel junction. , 2013, The Journal of chemical physics.

[16]  Martin Head-Gordon,et al.  A deterministic alternative to the full configuration interaction quantum Monte Carlo method. , 2016, The Journal of chemical physics.

[17]  Shigeru Nagase,et al.  Projector Monte Carlo method based on configuration state functions. Test applications to the H4 system and dissociation of LiH , 2008 .

[18]  Francesco A Evangelista,et al.  Adaptive multiconfigurational wave functions. , 2014, The Journal of chemical physics.

[19]  Jeffrey B Schriber,et al.  Communication: An adaptive configuration interaction approach for strongly correlated electrons with tunable accuracy. , 2016, The Journal of chemical physics.

[20]  Ali Alavi,et al.  Fermion Monte Carlo without fixed nodes: a game of life, death, and annihilation in Slater determinant space. , 2009, The Journal of chemical physics.

[21]  J. Greer.  Consistent treatment of correlation effects in molecular dissociation studies using randomly chosen configurations , 1995 .

[22]  J. Muenter,et al.  Electric dipole moment of carbon monoxide , 1975 .

[23]  Yann Garniron,et al.  Deterministic Construction of Nodal Surfaces within Quantum Monte Carlo: The Case of FeS. , 2017, Journal of chemical theory and computation.

[24]  J. Hasegawa,et al.  Selected configuration interaction method using sampled first-order corrections to wave functions. , 2017, The Journal of chemical physics.

[25]  C J Umrigar,et al.  Heat-Bath Configuration Interaction: An Efficient Selected Configuration Interaction Algorithm Inspired by Heat-Bath Sampling. , 2016, Journal of chemical theory and computation.

[26]  R. Bartlett,et al.  Monte Carlo configuration interaction predictions for the electronic spectra of Ne, CH2, C2, N2, and H2O compared to full configuration interaction calculations. , 2008, The Journal of chemical physics.

[27]  Paul Murphy,et al.  Development of spin‐orbit coupling for stochastic configuration interaction techniques , 2018, J. Comput. Chem..

[28]  P. Knowles,et al.  An efficient internally contracted multiconfiguration–reference configuration interaction method , 1988 .

[29]  Kerstin Andersson,et al.  Second-order perturbation theory with a CASSCF reference function , 1990 .

[30]  Li Li,et al.  Bypassing the Kohn-Sham equations with machine learning , 2016, Nature Communications.

[31]  J. P. Malrieu,et al.  Iterative perturbation calculations of ground and excited state energies from multiconfigurational zeroth‐order wavefunctions , 1973 .

[32]  Francesco A. Evangelista,et al.  Adaptive Configuration Interaction for Computing Challenging Electronic Excited States with Tunable Accuracy. , 2017, Journal of chemical theory and computation.

[33]  J. P. Coe,et al.  Approaching exact hyperpolarizabilities via sum-over-states Monte Carlo configuration interaction. , 2014, The Journal of chemical physics.

[34]  M. Gastegger,et al.  High-Dimensional Neural Network Potentials for Organic Reactions and an Improved Training Algorithm. , 2015, Journal of chemical theory and computation.

[35]  Matthias Rupp,et al.  Big Data Meets Quantum Chemistry Approximations: The Δ-Machine Learning Approach. , 2015, Journal of chemical theory and computation.

[36]  M. Head‐Gordon,et al.  A fifth-order perturbation comparison of electron correlation theories , 1989 .

[37]  J. P. Coe,et al.  Investigating Multireference Character and Correlation in Quantum Chemistry. , 2015, Journal of chemical theory and computation.

[38]  B. Roos,et al.  The complete active space SCF (CASSCF) method in a Newton–Raphson formulation with application to the HNO molecule , 1981 .

[39]  Rich Caruana,et al.  Do Deep Nets Really Need to be Deep? , 2013, NIPS.

[40]  R. Bartlett,et al.  A full coupled‐cluster singles and doubles model: The inclusion of disconnected triples , 1982 .