arXiv subjects

Heather J. Kulik

Publications and source records attributed to Heather J. Kulik.

46 records · Page 3Linked to original sources

arXiv2017

Recovering the flat-plane condition in electronic structure theory at semi-local DFT cost

The flat plane condition is the union of two exact constraints in electronic structure theory: i) energetic piecewise linearity with fractional electron removal or addition and ii) invariant energetics with change in electron spin in a half filled orbital. Semi-local density functional theory (DFT) fails to recover the flat plane, exhibiting convex fractional charge errors (FCE) and concave fractional spin errors (FSE) that are related to delocalization and static correlation errors. We previously showed that DFT+U eliminates FCE but now demonstrate that, like other widely employed corrections (i.e., Hartree Fock exchange), it worsens FSE. To find an alternative strategy, we examine the shape of semi-local DFT deviations from the exact flat plane, and we find this shape to be remarkably consistent across ions and molecules. We introduce the jmDFT approach, wherein corrections are constructed from few-parameter, low- order, functional forms that fit the shape of semi-local DFT errors. We select one such physically intuitive form and incorporate it self-consistently to correct semi-local DFT. We demonstrate on model systems that this jmDFT approach represents the first easy-to-implement, no-overhead approach to recovering the flat plane from semi-local DFT.

cond-mat.mtrl-sci↗

arXiv2017

Resolving transition metal chemical space: feature selection for machine learning and structure-property relationships

Machine learning (ML) of quantum mechanical properties shows promise for accelerating chemical discovery. For transition metal chemistry where accurate calculations are computationally costly and available training data sets are small, the molecular representation becomes a critical ingredient in ML model predictive accuracy. We introduce a series of revised autocorrelation functions (RACs) that encode relationships between the heuristic atomic properties (e.g., size, connectivity, and electronegativity) on a molecular graph. We alter the starting point, scope, and nature of the quantities evaluated in standard ACs to make these RACs amenable to inorganic chemistry. On an organic molecule set, we first demonstrate superior standard AC performance to other presently-available topological descriptors for ML model training, with mean unsigned errors (MUEs) for atomization energies on set-aside test molecules as low as 6 kcal/mol. For inorganic chemistry, our RACs yield 1 kcal/mol ML MUEs on set-aside test molecules in spin-state splitting in comparison to 15-20x higher errors from feature sets that encode whole-molecule structural information. Systematic feature selection methods including univariate filtering, recursive feature elimination, and direct optimization (e.g., random forest and LASSO) are compared. Random-forest- or LASSO-selected subsets 4-5x smaller than RAC-155 produce sub- to 1-kcal/mol spin-splitting MUEs, with good transferability to metal-ligand bond length prediction (0.004-5 Å MUE) and redox potential on a smaller data set (0.2-0.3 eV MUE). Evaluation of feature selection results across property sets reveals the relative importance of local, electronic descriptors (e.g., electronegativity, atomic number) in spin-splitting and distal, steric effects in redox potential and bond lengths.

cond-mat.mtrl-sci↗

arXiv2017

Predicting Electronic Structure Properties of Transition Metal Complexes with Neural Networks

High-throughput computational screening has emerged as a critical component of materials discovery. Direct density functional theory (DFT) simulation of inorganic materials and molecular transition metal complexes is often used to describe subtle trends in inorganic bonding and spin-state ordering, but these calculations are computationally costly and properties are sensitive to the exchange-correlation functional employed. To begin to overcome these challenges, we trained artificial neural networks (ANNs) to predict quantum-mechanically-derived properties, including spin-state ordering, sensitivity to Hartree-Fock exchange, and spin- state specific bond lengths in transition metal complexes. Our ANN is trained on a small set of inorganic-chemistry-appropriate empirical inputs that are both maximally transferable and do not require precise three-dimensional structural information for prediction. Using these descriptors, our ANN predicts spin-state splittings of single-site transition metal complexes (i.e., Cr-Ni) at arbitrary amounts of Hartree-Fock exchange to within 3 kcal/mol accuracy of DFT calculations. Our exchange-sensitivity ANN enables improved predictions on a diverse test set of experimentally-characterized transition metal complexes by extrapolation from semi-local DFT to hybrid DFT. The ANN also outperforms other machine learning models (i.e., support vector regression and kernel ridge regression), demonstrating particularly improved performance in transferability, as measured by prediction errors on the diverse test set. We establish the value of new uncertainty quantification tools to estimate ANN prediction uncertainty in computational chemistry, and we provide additional heuristics for identification of when a compound of interest is likely to be poorly predicted by the ANN.

cond-mat.mtrl-sci↗

arXiv2017

Systematic Quantum Mechanical Region Determination in QM/MM Simulation

Hybrid quantum mechanical-molecular mechanical (QM/MM) simulations are widely used in enzyme simulation. Over ten convergence studies of QM/MM methods have revealed over the past several years that key energetic and structural properties approach asymptotic limits with only very large (ca. 500-1000 atom) QM regions. This slow convergence has been observed to be due in part to significant charge transfer between the core active site and surrounding protein environment, which cannot be addressed by improvement of MM force fields or the embedding method employed within QM/MM. Given this slow convergence, it becomes essential to identify strategies for the most atom-economical determination of optimal QM regions and to gain insight into the crucial interactions captured only in large QM regions. Here, we extend and develop two methods for quantitative determination of QM regions. First, in the charge shift analysis (CSA) method, we probe the reorganization of electron density when core active site residues are removed completely, as determined by large-QM region QM/MM calculations. Second, we introduce the highly-parallelizable Fukui shift analysis (FSA), which identifies how core/substrate frontier states are altered by the presence of an additional QM residue on smaller initial QM regions. We demonstrate that the FSA and CSA approaches are complementary and consistent on three test case enzymes: catechol O-methyltransferase, cytochrome P450cam, and hen eggwhite lysozyme. We also introduce validation strategies and test sensitivities of the two methods to geometric structure, basis set size, and electronic structure methodology. Both methods represent promising approaches for the systematic, unbiased determination of quantum mechanical effects in enzymes and large systems that necessitate multi-scale modeling.

physics.chem-ph↗

arXiv2016

Where Does the Density Localize? Convergent Behavior for Global Hybrids, Range Separation, and DFT+U

Approximate density functional theory (DFT) suffers from many-electron self- interaction error, otherwise known as delocalization error, that may be diagnosed and then corrected through elimination of the deviation from exact piecewise linear behavior between integer electron numbers. Although paths to correction of energetic delocalization error are well- established, the impact of these corrections on the electron density is less well-studied. Here, we compare the effect on density delocalization of DFT+U, global hybrid tuning, and range- separated hybrid tuning on a diverse test set of 32 transition metal complexes and observe the three methods to have qualitatively equivalent effects on the ground state density. Regardless of valence orbital diffuseness (i.e., from 2p to 5p), ligand electronegativity (i.e., from Al to O), basis set (i.e., plane wave versus localized basis set), metal (i.e., Ti, Fe, Ni) and spin state, or tuning method, we consistently observe substantial charge loss at the metal and gain at ligand atoms (ca. 0.3-0.5 e or more). This charge loss at the metal is preferentially from the minority spin, leading to increasing magnetic moment as well. Using accurate wavefunction theory references, we observe that a minimum error in partial charges and magnetic moments occur at higher tuning parameters than typically employed to eliminate energetic delocalization error. These observations motivate the need to develop multi-faceted approximate-DFT error correction approaches that separately treat density delocalization and energetic errors in order to recover both correct density and magnetization properties.

cond-mat.mtrl-sci↗

arXiv2016

How large should the QM region be in QM/MM calculations? The case of catechol O-methyltransferase

Hybrid quantum mechanical-molecular mechanical (QM/MM) simulations are widely used in studies of enzymatic catalysis. Until recently, it has been cost prohibitive to determine the asymptotic limit of key energetic and structural properties with respect to increasingly large QM regions. Leveraging recent advances in electronic structure efficiency and accuracy, we investigate catalytic properties in catechol O-methyltransferase, a representative example of a methyltransferase critical to human health. Using QM regions ranging in size from reactants-only (64 atoms) to nearly one-third of the entire protein (940 atoms), we show that properties such as the activation energy approach within chemical accuracy of the large-QM asymptotic limits rather slowly, requiring approximately 500-600 atoms if the QM residues are chosen simply by distance from the substrate. This slow approach to asymptotic limit is due to charge transfer from protein residues to the reacting substrates. Our large QM/MM calculations enable identification of charge separation for fragments in the transition state as a key component of enzymatic methyl transfer rate enhancement. We introduce charge shift analysis that reveals the minimum number of protein residues (ca. 11-16 residues or 200-300 atoms for COMT) needed for quantitative agreement with large-QM simulations. The identified residues are not those that would be typically selected using criteria such as chemical intuition or proximity. These results provide a recipe for a more careful determination of QM region sizes in future QM/MM studies of enzymes.

q-bio.BM↗

arXiv2015

Direct Observation of Early-stage Quantum Dot Growth Mechanisms with High-temperature Ab Initio Molecular Dynamics

Colloidal quantum dots (QDs) exhibit highly desirable size- and shape-dependent properties for applications from electronic devices to imaging. Indium phosphide QDs have emerged as a primary candidate to replace the more toxic CdSe QDs, but production of InP QDs with the desired properties lags behind other QD materials due to a poor understanding of how to tune the growth process. Using high-temperature ab initio molecular dynamics (AIMD) simulations, we report the first direct observation of the early stage intermediates and subsequent formation of an InP cluster from separated indium and phosphorus precursors. In our simulations, indium agglomeration precedes formation of In-P bonds. We observe a predominantly intercomplex pathway in which In-P bonds form between one set of precursor copies while the carboxylate ligand of a second indium precursor in the agglomerated indium abstracts a ligand from the phosphorus precursor. This process produces an indium-rich cluster with structural properties comparable to those in bulk zinc-blende InP crystals. Minimum energy pathway characterization of the AIMD-sampled reaction events confirms these observations and identifies that In-carboxylate dissociation energetics solely determine the barrier along the In-P bond formation pathway, which is lower for intercomplex (13 kcal/mol) than intracomplex (21 kcal/mol) mechanisms. The phosphorus precursor chemistry, on the other hand, controls the thermodynamics of the reaction. Our observations of the differing roles of precursors in controlling QD formation strongly suggests that the challenges thus far encountered in InP QD synthesis optimization may be attributed to an overlooked need for a cooperative tuning strategy that simultaneously addresses the chemistry of both indium and phosphorus precursors.

cond-mat.mtrl-sci↗

arXiv2015

Towards quantifying the role of exact exchange in predictions of transition metal complex properties

We estimate the prediction sensitivity with respect to Hartree-Fock exchange in approximate density functionals for representative Fe(II) and Fe(III) octahedral complexes. Based on the observation that the range of parameters spanned by the most widely-employed functionals is relatively narrow, we compute electronic structure property and spin-state orderings across a relatively broad range of Hartree-Fock exchange (0-50%) ratios. For the entire range considered, we consistently observe linear relationships between spin-state ordering that differ only based on the element of the direct ligand and thus may be broadly employed as measures of functional sensitivity in predictions of organometallic compounds. The role Hartree-Fock exchange in hybrid functionals is often assumed to play is to correct self-interaction error-driven electron delocalization (e.g. from transition metal centers to neighboring ligands). Surprisingly, we instead observe that increasing Hartree-Fock exchange reduces charge on iron centers, corresponding to effective delocalization of charge to ligands, thus challenging notions of the role of Hartree-Fock exchange in shifting predictions of spin-state ordering.

cond-mat.mtrl-sci↗

arXiv2015

Quantum Chemistry for Solvated Molecules on Graphical Processing Units (GPUs)using Polarizable Continuum Models

The conductor-like polarization model (C-PCM) with switching/Gaussian smooth discretization is a widely used implicit solvation model in chemical simulations. However, its application in quantum mechanical calculations of large-scale biomolecular systems can be limited by computational expense of both the gas phase electronic structure and the solvation interaction. We have previously used graphical processing units (GPUs) to accelerate the first of these steps. Here, we extend the use of GPUs to accelerate electronic structure calculations including C-PCM solvation. Implementation on the GPU leads to significant acceleration of the generation of the required integrals for C-PCM. We further propose two strategies to improve the solution of the required linear equations: a dynamic convergence threshold and a randomized block-Jacobi preconditioner. These strategies are not specific to GPUs and are expected to be beneficial for both CPU and GPU implementations. We benchmark the performance of the new implementation using over 20 small proteins in solvent environment. Using a single GPU, our method evaluates the C-PCM related integrals and their derivatives more than 10X faster than a conventional CPU based implementation. Our improvements to the linear solver provide a further 3X acceleration. The overall calculations including C-PCM solvation require typically 20-40% more effort than their gas phase counterparts for moderate basis set and molecule surface discretization level. The relative cost of the C-PCM solvation correction decreases as the basis sets and/or cavity radii increase. Therefore description of solvation with this model should be routine. We also discuss applications to the study of the conformational landscape of an amyloid fibril.

physics.chem-ph↗

arXiv2006

Density functional theory in transition-metal chemistry: a self-consistent Hubbard U approach

Transition-metal centers are the active sites for many biological and inorganic chemical reactions. Notwithstanding this central importance, density-functional theory calculations based on generalized-gradient approximations often fail to describe energetics, multiplet structures, reaction barriers, and geometries around the active sites. We suggest here an alternative approach, derived from the Hubbard U correction to solid-state problems, that provides an excellent agreement with correlated-electron quantum chemistry calculations in test cases that range from the ground state of Fe$_2$ and Fe$_2^-$ to the addition-elimination of molecular hydrogen on FeO$^+$. The Hubbard U is determined with a novel self-consistent procedure based on a linear-response approach.

cond-mat.soft↗