Pushing the frontiers of density functionals by solving the fractional electron problem

Description Improving DFT with deep learning In the past 30 years, density functional theory (DFT) has emerged as the most widely used electronic structure method to predict the properties of various systems in chemistry, biology, and materials science. Despite a long history of successes, state-of-the-art DFT functionals have crucial limitations. In particular, significant systematic errors are observed for charge densities involving mobile charges and spins. Kirkpatrick et al. developed a framework to train a deep neural network on accurate chemical data and fractional electron constraints (see the Perspective by Perdew). The resulting functional outperforms traditional functionals on thorough benchmarks for main-group atoms and molecules. The present work offers a solution to a long-standing critical problem in DFT and demonstrates the success of combining DFT with the modern machine-learning methodology. —YS Deep learning solves a long-standing critical limitation of density functional theory in describing fractional charges and spins. Density functional theory describes matter at the quantum level, but all popular approximations suffer from systematic errors that arise from the violation of mathematical properties of the exact functional. We overcame this fundamental limitation by training a neural network on molecular data and on fictitious systems with fractional charge and spin. The resulting functional, DM21 (DeepMind 21), correctly describes typical examples of artificial charge delocalization and strong correlation and performs better than traditional functionals on thorough benchmarks for main-group atoms and molecules. DM21 accurately models complex systems such as hydrogen chains, charged DNA base pairs, and diradical transition states. More crucially for the field, because our methodology relies on data and constraints, which are continually improving, it represents a viable pathway toward the exact universal functional.

[1]  Colin A. Grambow,et al.  Reactants, products, and transition states of elementary chemical reactions based on quantum chemistry , 2019, Scientific Data.

[2]  A. Becke.  Density-functional thermochemistry. III. The role of exact exchange , 1993 .

[3]  Donald G Truhlar,et al.  Design of density functionals that are broadly accurate for thermochemistry, thermochemical kinetics, and nonbonded interactions. , 2005, The journal of physical chemistry. A.

[4]  Piotr Piecuch,et al.  Combining active-space coupled-cluster methods with moment energy corrections via the CC(P;Q) methodology, with benchmark calculations for biradical transition states. , 2012, The Journal of chemical physics.

[5]  Jan M. L. Martin,et al.  Benchmark thermochemistry of the C(n)H(2n+2) alkane isomers (n = 2-8) and performance of DFT and composite ab initio methods for dispersion-driven isomeric equilibria. , 2009, The journal of physical chemistry. A.

[6]  K. Hirao,et al.  A long-range-corrected time-dependent density functional theory. , 2004, The Journal of chemical physics.

[7]  Manoj K. Kesharwani,et al.  The X40×10 Halogen Bonding Benchmark Revisited: Surprising Importance of (n-1)d Subvalence Correlation. , 2018, The journal of physical chemistry. A.

[8]  M. Kaupp,et al.  A local hybrid functional with wide applicability and good balance between (de)localization and left-right correlation. , 2020, Journal of chemical theory and computation.

[9]  M. G. Medvedev,et al.  Density functional theory is straying from the path toward the exact functional , 2017, Science.

[10]  Richard Van Noorden,et al.  The top 100 papers , 2014, Nature.

[11]  Amir Karton,et al.  Comment on: “Estimating the Hartree–Fock limit from finite basis set calculations” [Jensen F (2005) Theor Chem Acc 113:267] , 2005, physics/0509216.

[12]  Yingkai Zhang,et al.  Comment on “Generalized Gradient Approximation Made Simple” , 1998 .

[13]  J. Barton,et al.  Mechanisms for DNA charge transport. , 2010, Chemical reviews.

[14]  S. Grimme,et al.  "Mindless" DFT Benchmarking. , 2009, Journal of chemical theory and computation.

[15]  E. Weinan,et al.  DeePKS: a comprehensive data-driven approach towards chemically accurate density functional theory , 2020, Journal of chemical theory and computation.

[16]  Tim Gould.  'Diet GMTKN55' offers accelerated benchmarking through a representative subset approach. , 2018, Physical chemistry chemical physics : PCCP.

[17]  Jianmin Tao,et al.  Density functional with full exact exchange, balanced nonlocality of correlation, and constraint satisfaction , 2008, 0808.2523.

[18]  David M. Ceperley,et al.  Towards the solution of the many-electron problem in real materials: equation of state of the hydrogen chain with state-of-the-art many-body methods , 2017, 1705.01608.

[19]  W. H. Li,et al.  Concerted vs stepwise mechanisms in dehydro-Diels-Alder reactions. , 2011, The Journal of organic chemistry.

[20]  K. Burke,et al.  Describing static correlation in bond dissociation by Kohn-Sham density functional theory. , 2004, The Journal of chemical physics.

[21]  Stefan Grimme,et al.  Effect of the damping function in dispersion corrected density functional theory , 2011, J. Comput. Chem..

[22]  G. Scuseria,et al.  Comparative assessment of a new nonempirical density functional: Molecules and hydrogen-bonded complexes , 2003 .

[23]  Qin Wu,et al.  A direct optimization method for calculating density functionals and exchange–correlation potentials from electron densities , 2003 .

[24]  Hyungjun Kim,et al.  Energy refinement and analysis of structures in the QM9 database via a highly accurate quantum chemical method , 2019, Scientific Data.

[25]  Ekin D. Cubuk,et al.  Kohn-Sham equations as regularizer: building prior knowledge into machine-learned physics , 2020, Physical review letters.

[26]  Arne Lüchow,et al.  Isomerization of bicyclo[1.1.0]butane by means of the diffusion quantum Monte Carlo method. , 2010, The journal of physical chemistry. A.

[27]  David Pfau,et al.  Ab-Initio Solution of the Many-Electron Schrödinger Equation with Deep Neural Networks , 2019, Physical Review Research.

[28]  M. Head‐Gordon,et al.  How Accurate Is Density Functional Theory at Predicting Dipole Moments? An Assessment Using a New Database of 200 Benchmark Values. , 2017, Journal of chemical theory and computation.

[29]  Michael Gastegger,et al.  Machine Learning Force Fields , 2020, Chemical reviews.

[30]  J. Perdew,et al.  Competing stripe and magnetic phases in the cuprates from first principles , 2019, Proceedings of the National Academy of Sciences.

[31]  Burke,et al.  Generalized Gradient Approximation Made Simple. , 1996, Physical review letters.

[32]  David Pfau,et al.  Better, Faster Fermionic Neural Networks , 2020, ArXiv.

[33]  Weitao Yang,et al.  Describing strong correlation with fractional-spin correction in density functional theory , 2018, Proceedings of the National Academy of Sciences.

[34]  Klaus-Robert Müller,et al.  Finding Density Functionals with Machine Learning , 2011, Physical review letters.

[35]  Jan M. L. Martin,et al.  DSD-PBEP86: in search of the best double-hybrid DFT with spin-component scaled MP2 and dispersion corrections. , 2011, Physical chemistry chemical physics : PCCP.

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

[37]  L. Curtiss,et al.  Assessment of Gaussian-3 and density functional theories for a larger experimental test set , 2000 .

[38]  Hilke Bahmann,et al.  Efficient Self-Consistent Implementation of Local Hybrid Functionals. , 2015, Journal of chemical theory and computation.

[39]  K. Houk,et al.  Diels–Alder Reactions of Allene with Benzene and Butadiene: Concerted, Stepwise, and Ambimodal Transition States , 2014, The Journal of organic chemistry.

[40]  Weitao Yang,et al.  Localized Orbital Scaling Correction for Systematic Elimination of Delocalization Error in Density Functional Approximations , 2017, 1707.00856.

[41]  Stefan Grimme,et al.  A General Database for Main Group Thermochemistry, Kinetics, and Noncovalent Interactions - Assessment of Common and Reparameterized (meta-)GGA Density Functionals. , 2010, Journal of chemical theory and computation.

[42]  Nicholas P. Bauman,et al.  Combining active-space coupled-cluster approaches with moment energy corrections via the CC(P;Q) methodology: connected quadruple excitations , 2017 .

[43]  Pavlo O. Dral,et al.  Quantum chemistry structures and properties of 134 kilo molecules , 2014, Scientific Data.

[44]  Amir Karton,et al.  The S66 Non-Covalent Interactions Benchmark Reconsidered Using Explicitly Correlated Methods Near the Basis Set Limit , 2017, 1712.09395.

[45]  Weitao Yang,et al.  Fractional spins and static correlation error in density functional theory. , 2008, The Journal of chemical physics.

[46]  Li Li,et al.  Understanding Machine-learned Density Functionals , 2014, ArXiv.

[47]  Anton V. Sinitskiy,et al.  Physical machine learning outperforms "human learning" in Quantum Chemistry , 2019, 1908.00971.

[48]  Andreas Hansen,et al.  A look at the density functional theory zoo with the advanced GMTKN55 database for general main group thermochemistry, kinetics and noncovalent interactions. , 2017, Physical chemistry chemical physics : PCCP.

[49]  W. Kohn,et al.  Self-Consistent Equations Including Exchange and Correlation Effects , 1965 .

[50]  M. Head‐Gordon,et al.  Systematic optimization of long-range corrected hybrid density functionals. , 2008, The Journal of chemical physics.

[51]  G. Scuseria,et al.  Climbing the density functional ladder: nonempirical meta-generalized gradient approximation designed for molecules and solids. , 2003, Physical review letters.

[52]  Narbe Mardirossian,et al.  ωB97X-V: a 10-parameter, range-separated hybrid, generalized gradient approximation density functional with nonlocal correlation, designed by a survival-of-the-fittest strategy. , 2014, Physical chemistry chemical physics : PCCP.

[53]  A. Tarnopolsky,et al.  Atomization energies of the carbon clusters C n (n = 2−10) revisited by means of W4 theory as well as density functional, Gn, and CBS methods , 2008, 0812.3852.

[54]  Jan M. L. Martin,et al.  W4-11: A high-confidence benchmark dataset for computational thermochemistry derived from first-principles W4 data , 2011 .

[55]  C. Corminboeuf,et al.  Machine learning models of the energy curvature vs particle number for optimal tuning of long-range corrected functionals. , 2020, The Journal of chemical physics.

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

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

[58]  Weitao Yang,et al.  Challenges for density functional theory. , 2012, Chemical reviews.

[59]  Roger B. Grosse,et al.  Optimizing Neural Networks with Kronecker-factored Approximate Curvature , 2015, ICML.

[60]  Amir Karton,et al.  W4‐17: A diverse and high‐confidence dataset of atomization energies for benchmarking high‐level electronic structure methods , 2017, J. Comput. Chem..

[61]  Namita Srivastava,et al.  The Machine‐Learning Approach , 2020, Machine Learning for iOS Developers.

[62]  Wotao Yin,et al.  A feasible method for optimization with orthogonality constraints , 2013, Math. Program..

[63]  Axel D. Becke,et al.  Density functionals for static, dynamical, and strong correlation. , 2013, The Journal of chemical physics.

[64]  Sebastian Dick,et al.  Machine learning accurate exchange and correlation functionals of the electronic density , 2020, Nature Communications.

[65]  Fractional Kohn–Sham Occupancies from a Strong-Correlation Density Functional , 2014 .

[66]  Weitao Yang,et al.  Insights into Current Limitations of Density Functional Theory , 2008, Science.

[67]  J. Perdew,et al.  Density-Functional Theory for Fractional Particle Number: Derivative Discontinuities of the Energy , 1982 .

[68]  H. G. Petersen,et al.  Error estimates on averages of correlated data , 1989 .

[69]  D. Truhlar,et al.  The M06 suite of density functionals for main group thermochemistry, thermochemical kinetics, noncovalent interactions, excited states, and transition elements: two new functionals and systematic testing of four M06-class functionals and 12 other functionals , 2008 .

[70]  Adrienn Ruzsinszky,et al.  Strongly Constrained and Appropriately Normed Semilocal Density Functional. , 2015, Physical review letters.

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

[72]  David A. Strubbe,et al.  Deep learning and density-functional theory , 2018, Physical Review A.

[73]  Sandeep Sharma,et al.  PySCF: the Python‐based simulations of chemistry framework , 2017, 1701.08223.