arXiv subjects

O. Anatole von Lilienfeld

Publications and source records attributed to O. Anatole von Lilienfeld.

At least 91 records · Page 5Linked to original sources

arXiv2017

Exploring water adsorption on isoelectronically doped graphene using alchemical derivatives

The design and production of novel 2-dimensional materials has seen great progress in the last decade, prompting further exploration of the chemistry of such materials. Doping and hydrogenating graphene is an experimentally realised method of changing its surface chemistry, but there is still a great deal to be understood on how doping impacts on the adsorption of molecules. Developing this understanding is key to unlocking the potential applications of these materials. High throughput screening methods can provide particularly effective ways to explore vast chemical compositions of materials. Here, alchemical derivatives are used as a method to screen the dissociative adsorption energy of water molecules on various BN doped topologies of hydrogenated graphene. The predictions from alchemical derivatives are assessed by comparison to density functional theory. This screening method is found to predict dissociative adsorption energies that span a range of more than 2 eV, with a mean absolute error $<0.1$ eV. In addition, we show that the quality of such predictions can be readily assessed by examination of the Kohn-Sham highest occupied molecular orbital in the initial states. In this way, the root mean square error in the dissociative adsorption energies of water is reduced by almost an order of magnitude (down to $\sim0.02$ eV) after filtering out poor predictions. The findings point the way towards a reliable use of first order alchemical derivatives for efficient screening procedures.

physics.chem-ph↗

arXiv2017

Constant Size Molecular Descriptors For Use With Machine Learning

A set of molecular descriptors whose length is independent of molecular size is developed for machine learning models that target thermodynamic and electronic properties of molecules. These features are evaluated by monitoring performance of kernel ridge regression models on well-studied data sets of small organic molecules. The features include connectivity counts, which require only the bonding pattern of the molecule, and encoded distances, which summarize distances between both bonded and non-bonded atoms and so require the full molecular geometry. In addition to having constant size, these features summarize information regarding the local environment of atoms and bonds, such that models can take advantage of similarities resulting from the presence of similar chemical fragments across molecules. Combining these two types of features leads to models whose performance is comparable to or better than the current state of the art. The features introduced here have the advantage of leading to models that may be trained on smaller molecules and then used successfully on larger molecules.

physics.chem-ph↗

arXiv2016

Genetic optimization of training sets for improved machine learning models of molecular properties

The training of molecular models of quantum mechanical properties based on statistical machine learning requires large datasets which exemplify the map from chemical structure to molecular property. Intelligent a priori selection of training examples is often difficult or impossible to achieve as prior knowledge may be sparse or unavailable. Ordinarily representative selection of training molecules from such datasets is achieved through random sampling. We use genetic algorithms for the optimization of training set composition consisting of tens of thousands of small organic molecules. The resulting machine learning models are considerably more accurate with respect to small randomly selected training sets: mean absolute errors for out-of-sample predictions are reduced to ~25% for enthalpies, free energies, and zero-point vibrational energy, to ~50% for heat-capacity, electron-spread, and polarizability, and by more than ~20% for electronic properties such as frontier orbital eigenvalues or dipole-moments. We discuss and present optimized training sets consisting of 10 molecular classes for all molecular properties studied. We show that these classes can be used to design improved training sets for the generation of machine learning models of the same properties in similar but unrelated molecular sets.

physics.comp-ph↗

arXiv2016

Water on BN doped benzene: A hard test for exchange-correlation functionals and the impact of exact exchange on weak binding

Density functional theory (DFT) studies of weakly interacting complexes have recently focused on the importance of van der Waals dispersion forces whereas, the role of exchange has received far less attention. Here, by exploiting the subtle binding between water and a boron and nitrogen doped benzene derivative (1,2-azaborine) we show how exact exchange can alter the binding conformation within a complex. Benchmark values have been calculated for three orientations of the water monomer on 1,2-azaborine from explicitly correlated quantum chemical methods, and we have also used diffusion quantum Monte Carlo. For a host of popular DFT exchange-correlation functionals we show that the lack of exact exchange leads to the wrong lowest energy orientation of water on 1,2-azaborine. As such, we suggest that a high proportion of exact exchange and the associated improvement in the electronic structure could be needed for the accurate prediction of physisorption sites on doped surfaces and in complex organic molecules. Meanwhile to predict correct absolute interaction energies an accurate description of exchange needs to be augmented by dispersion inclusive functionals, and certain non-local van der Waals functionals (optB88- and optB86b-vdW) perform very well for absolute interaction energies. Through a comparison with water on benzene and borazine (B$_3$N$_3$H$_6$) we show that these results could have implications for the interaction of water with doped graphene surfaces, and suggest a possible way of tuning the interaction energy.

physics.chem-ph↗

arXiv2016

Understanding molecular representations in machine learning: The role of uniqueness and target similarity

The predictive accuracy of Machine Learning (ML) models of molecular properties depends on the choice of the molecular representation. Based on the postulates of quantum mechanics, we introduce a hierarchy of representations which meet uniqueness and target similarity criteria. To systematically control target similarity, we rely on interatomic many body expansions, as implemented in universal force-fields, including Bonding, Angular, and higher order terms (BA). Addition of higher order contributions systematically increases similarity to the true potential energy and predictive accuracy of the resulting ML models. We report numerical evidence for the performance of BAML models trained on molecular properties pre-calculated at electron-correlated and density functional theory level of theory for thousands of small organic molecules. Properties studied include enthalpies and free energies of atomization, heatcapacity, zero-point vibrational energies, dipole-moment, polarizability, HOMO/LUMO energies and gap, ionization potential, electron affinity, and electronic excitations. After training, BAML predicts energies or electronic properties of out-of-sample molecules with unprecedented accuracy and speed.

physics.chem-ph↗

arXiv2016

Machine Learning Energies of 2 M Elpasolite (ABC$_2$D$_6$) Crystals

Elpasolite is the predominant quaternary crystal structure (AlNaK$_2$F$_6$ prototype) reported in the Inorganic Crystal Structure Database. We have developed a machine learning model to calculate density functional theory quality formation energies of all $\sim$2 M pristine ABC$_2$D$_6$ elpasolite crystals which can be made up from main-group elements (up to bismuth). Our model's accuracy can be improved systematically, reaching 0.1 eV/atom for a training set consisting of 10 k crystals. Important bonding trends are revealed, fluoride is best suited to fit the coordination of the D site which lowers the formation energy whereas the opposite is found for carbon. The bonding contribution of elements A and B is very small on average. Low formation energies result from A and B being late elements from group (II), C being a late (I) element, and D being fluoride. Out of 2 M crystals, 90 unique structures are predicted to be on the convex hull---among which NFAl$_2$Ca$_6$, with peculiar stoichiometry and a negative atomic oxidation state for Al.

physics.chem-ph↗

arXiv2016

Rapid yet accurate first principle based predictions of alkali halide crystal phases using alchemical perturbation

We assess the predictive power of alchemical perturbations for estimating fundamental properties in ionic crystals. Using density functional theory we have calculated formation energies, lattice constants, and bulk moduli for all sixteen iso-valence-electronic combinations of pure pristine alkali halides involving elements $A \in \{$Na, K, Rb, Cs$\}$ and $X \in \{$F, Cl, Br, I$\}$. For rock salt, zincblende and cesium chloride symmetry, alchemical Hellmann-Feynman derivatives, evaluated along lattice scans of sixteen reference crystals, have been obtained for all respective 16$\times$15 combinations of reference and predicted target crystals. Mean absolute errors (MAE) are on par with density functional theory level of accuracy for energies and bulk modulus. Predicted lattice constants are less accurate. NaCl is the best reference salt for alchemical estimates of relative energies (MAE $<$ 40 meV/atom) while alkali fluorides are the worst. By contrast, lattice constants are predicted best using NaF as a reference salt (MAE $<$ 0.5Å), yielding only semi-quantitative accuracy. The best reference salt for the prediction of bulk moduli is CsCl (MAE $<$ 0.4$\times$10$^{11}$ dynes/cm$^2$). Alchemical derivatives can also be used to predict competing rock salt and cesium chloride phases in binary and ternary solid mixtures with CsCl. Alchemical predictions based on dispersion corrected density functional theory with pure RbI as a reference salt reproduce reasonably well the reversal of the rock salt/cesium chloride stability trend for binary $(AX)_{1-x}$CsCl$_x$ as well as for ternary $(AX)_{0.5-0.5x}(BY)_{0.5-0.5x}$CsCl$_x$ mixtures.

physics.chem-ph↗

arXiv2016

Machine Learning, Quantum Mechanics, and Chemical Compound Space

We review recent studies dealing with the generation of machine learning models of molecular and solid properties. The models are trained and validated using standard quantum chemistry results obtained for organic molecules and materials selected from chemical space at random.

physics.chem-ph↗

arXiv2016

Tuning dissociation using isoelectronically doped graphene and hexagonal boron nitride: water and other small molecules

Novel uses for 2-dimensional materials like graphene and hexagonal boron nitride (h-BN) are being frequently discovered especially for membrane and catalysis applications. Still however, a great deal remains to be understood about the interaction of environmentally and industrially elevant molecules such as water with these materials. Taking inspiration from advances in hybridising graphene and h-BN, we explore using density functional theory, the dissociation of water, hydrogen, methane, and methanol on graphene, h-BN, and their isoelectronic doped counterparts: BN doped graphene and C doped h-BN. We find that doped surfaces are considerably more reactive than their pristine counterparts and by comparing the reactivity of several small molecules we develop a general framework for dissociative adsorption. From this a particularly attractive consequence of isoelectronic doping emerges: substrates can be doped to enhance their reactivity specifically towards either polar or non-polar adsorbates. As such, these substrates are potentially viable candidates for selective catalysts and membranes, with the implication that a range of tuneable materials can be designed.

cond-mat.mtrl-sci↗

arXiv2016

Fast and accurate predictions of covalent bonds in chemical space

We assess the predictive accuracy of perturbation theory based estimates of changes in covalent bonding due to linear alchemical interpolations among molecules. We have investigated $σ$ bonding to hydrogen, as well as $σ$ and $π$ bonding between main-group elements, occurring in small sets of iso-valence-electronic molecular species with elements drawn from second to fourth rows in the $p$-block of the periodic table. Numerical evidence suggests that first order estimates of covalent bonding potentials can achieve chemical accuracy if (i) the alchemical interpolation is vertical (fixed geometry), (ii) involves molecules containing elements in the third and fourth row of the periodic table, and (iii) a reference geometry is optimized. In this case, changes in the bonding potential become near-linear in coupling parameter, resulting in analytical predictions with very high accuracy ($\sim$1 kcal/mol). Second order estimates deteriorate the prediction. If initial and final molecules differ not only in composition but also in geometry, all estimates become substantially worse, with second order being slightly more accurate than first order. The independent particle approximation to the second order perturbation performs poorly when compared to the coupled perturbed or finite difference approach. Taylor series expansions up to fourth order of the potential energy curve of highly symmetric systems indicate a finite radius of convergence, as illustrated for the alchemical stretching of H$_2^+$. Numerical results are presented for covalent bonds to hydrogen in 12 molecules with 8 valence electrons; (ii) main-group single bonds in 9 molecules with 14 valence electrons; (iii) main-group double bonds in 9 molecules with 12 valence electrons; (iv) main-group triple bonds in 9 molecules with 10 valence electrons; (v) H$_2^+$ single bond with 1 electron.

physics.chem-ph↗

arXiv2015

Machine Learning for Quantum Mechanical Properties of Atoms in Molecules

We introduce machine learning models of quantum mechanical observables of atoms in molecules. Instant out-of-sample predictions for proton and carbon nuclear chemical shifts, atomic core level excitations, and forces on atoms reach accuracies on par with density functional theory reference. Locality is exploited within non-linear regression via local atom-centered coordinate systems. The approach is validated on a diverse set of 9k small organic molecules. Linear scaling of computational cost in system size is demonstrated for saturated polymers with up to sub-mesoscale lengths.

physics.chem-ph↗

arXiv2015

Electronic Spectra from TDDFT and Machine Learning in Chemical Space

Due to its favorable computational efficiency time-dependent (TD) density functional theory (DFT) enables the prediction of electronic spectra in a high-throughput manner across chemical space. Its predictions, however, can be quite inaccurate. We resolve this issue with machine learning models trained on deviations of reference second-order approximate coupled-cluster singles and doubles (CC2) spectra from TDDFT counterparts, or even from DFT gap. We applied this approach to low-lying singlet-singlet vertical electronic spectra of over 20 thousand synthetically feasible small organic molecules with up to eight CONF atoms. The prediction errors decay monotonously as a function of training set size. For a training set of 10 thousand molecules, CC2 excitation energies can be reproduced to within $\pm$0.1 eV for the remaining molecules. Analysis of our spectral database via chromophore counting suggests that even higher accuracies can be achieved. Based on the evidence collected, we discuss open challenges associated with data-driven modeling of high-lying spectra, and transition intensities.

physics.chem-ph↗

arXiv2015

Machine learning for many-body physics: efficient solution of dynamical mean-field theory

Machine learning methods for solving the equations of dynamical mean-field theory are developed. The method is demonstrated on the three dimensional Hubbard model. The key technical issues are defining a mapping of an input function to an output function, and distinguishing metallic from insulating solutions. Both metallic and Mott insulator solutions can be predicted. The validity of the machine learning scheme is assessed by comparing predictions of full correlation functions, of quasi-particle weight and particle density to values directly computed. The results indicate that with modest further development, machine learning approach may be an attractive computational efficient option for real materials predictions for strongly correlated systems.

cond-mat.str-el↗

arXiv2015

Water on hexagonal boron nitride from diffusion Monte Carlo

Despite a recent flurry of experimental and simulation studies, an accurate estimate of the interaction strength of water molecules with hexagonal boron nitride is lacking. Here we report quantum Monte Carlo results for the adsorption of a water monomer on a periodic hexagonal boron nitride sheet, which yield a water monomer interaction energy of -84 +/- 5 meV. We use the results to evaluate the performance of several widely used density functional theory (DFT) exchange correlation functionals, and find that they all deviate substantially. Differences in interaction energies between different adsorption sites are however better reproduced by DFT.

cond-mat.mtrl-sci↗

arXiv2015

Transferable atomic multipole machine learning models for small organic molecules

Accurate representation of the molecular electrostatic potential, which is often expanded in distributed multipole moments, is crucial for an efficient evaluation of intermolecular interactions. Here we introduce a machine learning model for multipole coefficients of atom types H, C, O, N, S, F, and Cl in any molecular conformation. The model is trained on quantum chemical results for atoms in varying chemical environments drawn from thousands of organic molecules. Multipoles in systems with neutral, cationic, and anionic molecular charge states are treated with individual models. The models' predictive accuracy and applicability are illustrated by evaluating intermolecular interaction energies of nearly 1,000 dimers and the cohesive energy of the benzene crystal.

physics.chem-ph↗

arXiv2015

Crystal Structure Representations for Machine Learning Models of Formation Energies

We introduce and evaluate a set of feature vector representations of crystal structures for machine learning (ML) models of formation energies of solids. ML models of atomization energies of organic molecules have been successful using a Coulomb matrix representation of the molecule. We consider three ways to generalize such representations to periodic systems: (i) a matrix where each element is related to the Ewald sum of the electrostatic interaction between two different atoms in the unit cell repeated over the lattice; (ii) an extended Coulomb-like matrix that takes into account a number of neighboring unit cells; and (iii) an Ansatz that mimics the periodicity and the basic features of the elements in the Ewald sum matrix by using a sine function of the crystal coordinates of the atoms. The representations are compared for a Laplacian kernel with Manhattan norm, trained to reproduce formation energies using a data set of 3938 crystal structures obtained from the Materials Project. For training sets consisting of 3000 crystals, the generalization error in predicting formation energies of new structures corresponds to (i) 0.49, (ii) 0.64, and (iii) 0.37 eV/atom for the respective representations.

physics.chem-ph↗

arXiv2015

Quantum Mechanical Treatment of Variable Molecular Composition: From "Alchemical" Changes of State Functions to Rational Compound Design

"Alchemical" interpolation paths, i.e.~coupling systems along fictitious paths that without realistic correspondence, are frequently used within materials and molecular modeling and simulation protocols for the estimation of relative changes in state functions such as free energies. We discuss alchemical changes in the context of quantum chemistry, and present illustrative numerical results for the changes of HOMO eigenvalues of the He atom due to a linear alchemical teleportation---the simultaneous annihilation and creation of nuclear charges at different locations. To demonstrate the predictive power of alchemical first order derivatives (Hellmann-Feynman) the covalent bond potential of hydrogen fluoride and hydrogen chloride is investigated, as well as the van-der-Waals binding in the water-water and water-hydrogen fluoride dimer, respectively. Based on converged electron densities for one configuration, the versatility of alchemical derivatives is exemplified for the screening of entire binding potentials with reasonable accuracy. Finally, we discuss constraints for the identification of non-linear coupling potentials for which the energy's Hellmann-Feynman derivative will yield accurate predictions.

physics.chem-ph↗

arXiv2015

Fourier series of atomic radial distribution functions: A molecular fingerprint for machine learning models of quantum chemical properties

We introduce a fingerprint representation of molecules based on a Fourier series of atomic radial distribution functions. This fingerprint is unique (except for chirality), continuous, and differentiable with respect to atomic coordinates and nuclear charges. It is invariant with respect to translation, rotation, and nuclear permutation, and requires no pre-conceived knowledge about chemical bonding, topology, or electronic orbitals. As such it meets many important criteria for a good molecular representation, suggesting its usefulness for machine learning models of molecular properties trained across chemical compound space. To assess the performance of this new descriptor we have trained machine learning models of molecular enthalpies of atomization for training sets with up to 10k organic molecules, drawn at random from a published set of 134k organic molecules. We validate the descriptor on all remaining molecules of the 134k set. For a training set of 5k molecules the fingerprint descriptor achieves a mean absolute error of 8.0 kcal/mol, respectively. This is slightly worse than the performance attained using the Coulomb matrix, another popular alternative, reaching 6.2 kcal/mol for the same training and test sets.

physics.chem-ph↗