KineticNet: Deep learning a transferable kinetic energy functional for orbital-free density functional theory

Orbital-free density functional theory (OF-DFT) holds the promise to compute ground state molecular properties at minimal cost. However, it has been held back by our inability to compute the kinetic energy as a functional of the electron density only. We here set out to learn the kinetic energy functional from ground truth provided by the more expensive Kohn-Sham density functional theory. Such learning is confronted with two key challenges: Giving the model sufficient expressivity and spatial context while limiting the memory footprint to afford computations on a GPU; and creating a sufficiently broad distribution of training data to enable iterative density optimization even when starting from a poor initial guess. In response, we introduce KineticNet, an equivariant deep neural network architecture based on point convolutions adapted to the prediction of quantities on molecular quadrature grids. Important contributions include convolution filters with sufficient spatial resolution in the vicinity of the nuclear cusp, an atom-centric sparse but expressive architecture that relays information across multiple bond lengths; and a new strategy to generate varied training data by finding ground state densities in the face of perturbations by a random external potential. KineticNet achieves, for the first time, chemical accuracy of the learned functionals across input densities and geometries of tiny molecules. For two electron systems, we additionally demonstrate OF-DFT density optimization with chemical accuracy.

[1]  B. Kozinsky,et al.  CIDER: An Expressive, Nonlocal Feature Set for Machine Learning Density Functionals with Exact Constraints. , 2021, Journal of chemical theory and computation.

[2]  Isaac Tamblyn,et al.  Toward Orbital-Free Density Functional Theory with Small Data Sets and Deep Learning. , 2021, Journal of chemical theory and computation.

[3]  Alexander L. Gaunt,et al.  Pushing the frontiers of density functionals by solving the fractional electron problem , 2021, Science.

[4]  M. Imada,et al.  Order- N orbital-free density-functional calculations with machine learning of functional derivatives for semiconductors and metals , 2021, Physical Review Research.

[5]  S. Ghasemi,et al.  Artificial neural networks for the kinetic energy functional of non-interacting fermions. , 2021, Journal of Chemical Physics.

[6]  T. Helgaker,et al.  Robust All-Electron Optimization in Orbital-Free Density-Functional Theory Using the Trust-Region Image Method. , 2020, The journal of physical chemistry. A.

[7]  A. Hauser,et al.  Machine Learning Approaches toward Orbital-free Density Functional Theory: Simultaneous Training on the Kinetic Energy Density Functional and Its Functional Derivative , 2020, Journal of chemical theory and computation.

[8]  H. Nakai,et al.  Orbital-free density functional theory calculation applying semi-local machine-learned kinetic energy density functional and kinetic potential , 2020, Chemical Physics Letters.

[9]  Timothy C. Berkelbach,et al.  Recent developments in the PySCF program package. , 2020, The Journal of chemical physics.

[10]  Sebastian Dick,et al.  Machine learning accurate exchange and correlation functionals of the electronic density , 2019, Nature Communications.

[11]  Ryo Nagai,et al.  Completing density functional theory by machine learning hidden messages from molecules , 2019, npj Computational Materials.

[12]  S. Kais,et al.  Direct Scheme Calculation of the Kinetic Energy Functional Derivative Using Machine Learning , 2020 .

[13]  Jonathan Schmidt,et al.  Machine Learning the Physical Nonlocal Exchange-Correlation Functional of Density-Functional Theory. , 2019, The journal of physical chemistry letters.

[14]  Max Welling,et al.  3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data , 2018, NeurIPS.

[15]  Sergei Manzhos,et al.  Kinetic energy densities based on the fourth order gradient expansion: performance in different classes of materials and improvement via machine learning. , 2018, Physical chemistry chemical physics : PCCP.

[16]  Junji Seino,et al.  Semi-local machine-learned kinetic energy density functional with third-order gradients of electron density. , 2018, The Journal of chemical physics.

[17]  Li Li,et al.  Tensor Field Networks: Rotation- and Translation-Equivariant Neural Networks for 3D Point Clouds , 2018, ArXiv.

[18]  Andrea Grisafi,et al.  Symmetry-Adapted Machine Learning for Tensorial Properties of Atomistic Systems. , 2017, Physical review letters.

[19]  Sandeep Sharma,et al.  PySCF: the Python‐based simulations of chemistry framework , 2018 .

[20]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[21]  Benjamin Stamm,et al.  Quantum, classical, and hybrid QM/MM calculations in solution: general implementation of the ddCOSMO linear scaling strategy. , 2014, The Journal of chemical physics.

[22]  Narbe Mardirossian,et al.  x B 97 XV : A 10-parameter , range-separated hybrid , generalized gradient approximation density functional with nonlocal correlation , designed by a survival-ofthe-fittest strategy , 2014 .

[23]  Benjamin Stamm,et al.  Domain decomposition for implicit solvation models. , 2013, The Journal of chemical physics.

[24]  Benjamin Stamm,et al.  Fast Domain Decomposition Algorithm for Continuum Solvation Models: Energy and First Derivatives. , 2013, Journal of chemical theory and computation.

[25]  John C. Snyder,et al.  Orbital-free bond breaking via machine learning. , 2013, The Journal of chemical physics.

[26]  G. Chan,et al.  Thomas–Fermi–Dirac–von Weizsäcker models in finite systems , 2001 .

[27]  N. Handy,et al.  Kinetic energy functionals from the Kohn–Sham potential , 2000 .

[28]  R. Parr,et al.  Density-functional theory of the electronic structure of molecules. , 1995, Annual review of physical chemistry.

[29]  A. Becke,et al.  Density-functional exchange-energy approximation with correct asymptotic behavior. , 1988, Physical review. A, General physics.

[30]  Parr,et al.  Development of the Colle-Salvetti correlation-energy formula into a functional of the electron density. , 1988, Physical review. B, Condensed matter.

[31]  P. Hohenberg,et al.  Inhomogeneous Electron Gas , 1964 .