Session F1: Computational Discovery and Design of Novel Materials V

Fast and accurate covalent bond predictions using perturbation theory in chemical space

I will discuss the predictive accuracy of perturbation theory based estimates of changes in covalent bonding due to linear alchemical interpolations among systems of different chemical composition. We have investigated single, double, and triple bonds 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 (within 1 kcal/mol) if the alchemical interpolation is vertical (fixed geometry) among chemical elements from third and fourth row of the periodic table[1]. When applied to nonbonded systems of molecular dimers or solids such as III-V semiconductors, alanates, alkali halides, and transition metals, similar observations hold, enabling rapid predictions of van der Waals energies, defect energies, band-structures, crystal structures, and lattice constants [2].\newline\newline [1] K. Y. S. Chang {\it et al} J. Chem. Phys. (2016)\newline [2] K. Y. S. Chang, O. A. von Lilienfeld, in preparation (2017); M. to Baben {\it et al} J. Chem. Phys. (2016); A. Solovyeva, O. A. von Lilienfeld Phys. Chem. Chem. Phys. (2016)   

New discovery tools for molecular materials design

First-principles modeling has emerged as a critical component of materials screening and design, particularly for bulk systems with limited compositional degrees of freedom. However, strategies for molecular materials design have lagged behind the heterogeneous screening efforts, owing to the larger chemical space spanned by such molecular motifs. Aiming to overcome current limitations in molecular discovery, we present our recently introduced open-source molSimplify$^{\mathrm{1}}$ toolkit. This software enables rapid, automated structure generation and discovery by building accurate geometries to enable high-throughput screening through a unique divide-and-conquer approach; it interfaces to multi-million-molecule databases to enable discovery; and we present recent strategies$^{\mathrm{2}}$ adapted from the therapeutic drug design community to enable truly rational and iteratively-improved design strategies. Finally, we present our recent efforts on further acceleration of discovery both through high-throughput, high-quality guesses for transition states away from equilibrium and multi-resolution modeling using artificial neural networks with quantified uncertainty. $^{\mathrm{1}}$E. I. Ioannidis, T. Z. H. Gani, and H. J. Kulik "molSimplify: A toolkit for automating discovery in inorganic chemistry" \textit{Journal of Computational Chemistry}, \textbf{37}, 2106-2117 (2016). $^{\mathrm{2}}$T. Z. H. Gani, E. I. Ioannidis, and H. J. Kulik "Computational Discovery of Hydrogen Bond Design Rules for Electrochemical Ion Separation" \textit{Chemistry of Materials}, \textbf{28}, 6207-6218 (2016).   

Design of Beta Phase Mo$_{\mathrm{2}}$C Catalyst for Hydrogen Evolution Reaction via Nanoparticle Morphology Control: Insights from First Principles Methods and Kinetic Modeling

Transition metal carbides, in particular $\beta $-phase Mo$_{\mathrm{2}}$Cs, are garnering increased attention as hydrogen evolution reaction (HER) catalysts due to their favorable synthesis conditions, stability and high catalytic efficiency. We use a thermodynamic approach in conjunction with density functional theory and a kinetic model of exchange current density to systematically study the HER activity of $\beta $-Mo$_{\mathrm{2}}$C under different experimental conditions. We show that (011) surface has the highest HER activity because this surface does not expose strong Mo-based hydrogen adsorption sites. We give definite maps between nanoparticle morphologies and experimental synthesis conditions, and show that an increase of hydrogen partial pressure during synthesis can expose (011) surface up to 90 percent, which increase by extension the HER activity. The volcano plot shows that under these optimum conditions, the NP exchange current densities can be increased by more than one order to 10$^{\mathrm{-5}}$ A/cm$^{\mathrm{2}}$, that is only slightly smaller than that of Pt (111).   

Direct partial oxidation of methane via single-site chemistry

Methane (CH$_{4}$), the cheapest source of hydrocarbons, is a difficult-to-store and hard-to-convert chemical, due to strong and isotropic C-H bonds. Having a selective and efficient method to partially oxidize methane into more useful chemicals including methanol, formaldehyde and alkenes has long been an open challenge for catalysis community. The main challenge is selectivity: if a catalytic material interacts strongly with methane, sufficient to break one C-H bond, it breaks all other bonds of the derivative molecules, as well. This leads to over-oxidization to CO$_{2}$. Here using density functional theory (DFT) modelings, we discuss the possibility of using defective (vacancy rich) 2-d materials e.g., MoS$_{2}$ to effectively trap single transition metal atoms; thereby, creating a single-site chemistry to enhance the selectivity of methane oxidation process. A single-site chemistry leads to competition between intermediates and products for limited active sites. Our strategy is to use this as an effective means to block unwanted reaction pathways leading to over-oxidization.   

5-fold increase of hydrogen uptake in MOF74 through linker decorations

We present \emph{ab initio} results for linker decorations in Mg-MOF74---i.e.\ attaching various metals $\mathcal{M}$ = Li, Na, K, Sc, Cr, Mn, Fe, Ni, Cu, Zn, Rb, Pd, Ag, and Pt near the ring of the linker---creating new strong adsorption sites and thus maximizing small molecule uptake.\footnote{C. Arter, S. Zuluaga, D. Harrison, E. Welchman, and T. Thonhauser, Phys. Rev. B {\bf 94}, 144105 (2016).} We find that in most cases these decorations influence the overall form and structure of Mg-MOF74 only marginally. After the initial screening we chose metals that bind favorably to the linker and further investigate adsorption of H$_{2}$, CO$_{2}$, and H$_{2}$O for $\mathcal{M}$ = Li, Na, K, and Sc. For the case of H$_2$ we show that up to 24 additional guest molecules can be adsorbed in the MOF unit cell, with binding energies comparable to the original open-metal sites at the six corners of the channel. This leads to a 5-fold increase of the molecule uptake in Mg-MOF74, with tremendous impact on many applications in general and hydrogen storage in particular---where the gravimetric hydrogen density increases from $1.63$~mass\% to $7.28$~mass\% and the volumetric density from 15.10 g H$_{2}$ L$^{-1}$ to 75.50~g~H$_2$~L$^{-1}$.   

Macrocycles inserted in graphene: from coordination chemistry on graphene to graphitic carbon oxide.

Tuning the electronic structure and the chemical properties of graphene by binding with metals has become a focus in the area of two dimension materials. Despite many interesting results and promising potentials, the approach suffers from weak binding and the high reactivity of the metal atoms. On the other hand, many macrocyclic molecules such as crown ether show strong and selective binding with metal atoms. The alliance of the two substances will largely benefit the two parallel fields: it will provide a scaffold for coordination chemistry as well as a controllable method for tuning the electronic structure of graphene through strong binding with metals. Here, using crown ether as an example, we demonstrate by first principles calculations that the embedment of macrocyclic molecules into graphene honeycomb lattice can be very thermochemically favored. The embedment of crown ether on graphene can form a family of new two-dimensional materials that possess varying band gaps and band edges. The one with highest O composition (C$_{2}$O), with similar structure features as graphilic C$_{3}$N$_{4}$, shows strong potentials for photolysis and as true two-dimensional superconductor while binding with alkali metals.   

Adsorbate phases of H on ZnO $(10 \overline{1}0)$ surface as a function of temperature and pressure from first principles

Zinc oxide (ZnO) is a highly multifunctional material with unique properties and a wide range of applications. To understand atomic hydrogen adsorption on the thermodynamically stable $(10 \overline{1}0)$ surface at realistic H chemical potentials, we combine a first-principles cluster-expansion approach with {\em ab initio} atomistic thermodynamics. Our study reveals that at coverages below 6\%, H atoms adsorb exclusively on surface O. At higher coverages, H adsorbs also on Zn, but there is an excess of O-H over Zn-H at all coverages, except 50\% and 100\%. Due to an interplay of long- and short-range electrostatic interactions, neighboring O-H/Zn-H pairs form chains along surface -O-Zn- rows, with each chain anchored at the excess O-H, in a wide range of $(T,p_{\rm H_2})$. Our results offer a ``road map'' for H adsorption on the ZnO $(10 \overline{1}0)$ surface at various conditions, consolidating findings from previous experiments [1-4]. ---[1] Y. Wang {\em et al.}, PRL 95, (2005); [2] K. Ozawa and K. Mase, Phys. Stat. Sol. App. Mat. Sci. 207, (2010); [3] K. Ozawa and K. Mase, PRB 83, (2011); [4] J. Deinert {\em et al.}, PRB 91, (2015).   

Molecules coating magnetic nanoparticles for oil-field applications

Magnetic nanoparticles have recently attracted significant attention in scientific and industrial communities due to their use in the fields of catalysis, spintronics, biomedical applications, and oil recovery and reservoir characterization. However, these nanoparticles have to be protected with a coating layer of molecules that prevents the nanoparticles from oxidation, which is known to occur in air, and from agglomeration into larger nanoparticles. Therefore, the binding of the molecules to the nanoparticles is critical before a large scale implementation can be done. Here we report results of density functional theory calculations on several molecules (methylamine, acetic acid, boronic acid, ethyl phosphate, and ethyl trihydroxysilane) and magnetic nanoparticles (Fe$_3$O$_4$, NiFe$_2$O$_4$, and Fe$_3$C). We focus on two main points: 1) the bond strength between the organic molecule and the nano particle, and 2) how, H$_2$O and H$^+$ in the oil well may facilitate the desorption of the molecules. The results show that H$^+$ and H$_2$O molecules facilitate the desorption of molecules reducing the bond strength by several eV. On the other hand, the results allow us to identify and design molecules that exhibit the best performance in protecting each nanoparticle.   

Analysing and Rationalising Molecular and Materials Databases Using Machine-Learning

Computational materials design promises to greatly accelerate the process of discovering new or more performant materials. Several collaborative efforts are contributing to this goal by building databases of structures, containing between thousands and millions of distinct hypothetical compounds, whose properties are computed by high-throughput electronic-structure calculations. The complexity and sheer amount of information has made manual exploration, interpretation and maintenance of these databases a formidable challenge, making it necessary to resort to automatic analysis tools. Here we will demonstrate how, starting from a measure of (dis)similarity between database items built from a combination of local environment descriptors, it is possible to apply hierarchical clustering algorithms, as well as dimensionality reduction methods such as sketchmap, to analyse, classify and interpret trends in molecular and materials databases, as well as to detect inconsistencies and errors. Thanks to the agnostic and flexible nature of the underlying metric, we will show how our framework can be applied transparently to different kinds of systems ranging from organic molecules and oligopeptides to inorganic crystal structures as well as molecular crystals.   

Effect of Crystal Packing on the Electronic Properties of Molecular Crystals

Large scale quantum mechanical simulations are performed to study the effect of crystal packing on the electronic and optical properties of molecular crystals, which are essential for applications in organic electronics and photovoltaics. For structure prediction, we use the massively parallel genetic algorithm (GA) package, GAtor, which relies on the evolutionary principle of survival of the fittest to find low-energy crystal structures of a given molecule. Dispersion-inclusive DFT, implemented in the FHI-aims code, is used for structural relaxation and energy evaluations. The structure generation package, Genarris, performs fast screening of randomly generated structures with a Harris approximation, whereby the molecular crystal density is constructed by replicating the single molecule density, which is calculated only once. Many-body perturbation theory, within the GW approximation and the Bethe-Salpeter equation (BSE), as implemented in the BerkeleyGW code, is employed to describe properties derived from charged and neutral excitations. We show that transport, optical absorption, and singlet fission efficiency may be enhanced by modifying the crystal packing of TCS3 and rubrene.   

ChemHTPS - A virtual high-throughput screening program suite for the chemical and materials sciences

The discovery of new compounds, materials, and chemical reactions with exceptional properties is the key for the grand challenges in innovation, energy and sustainability. This process can be dramatically accelerated by means of the virtual high-throughput screening (HTPS) of large-scale candidate libraries. The resulting data can further be used to study the underlying structure-property relationships and thus facilitate rational design capability. This approach has been extensively used for many years in the drug discovery community. However, the lack of openly available virtual HTPS tools is limiting the use of these techniques in various other applications such as photovoltaics, optoelectronics, and catalysis. Thus, we developed ChemHTPS, a general-purpose, comprehensive and user-friendly suite, that will allow users to efficiently perform large in silico modeling studies and high-throughput analyses in these applications. ChemHTPS also includes a massively parallel molecular library generator which offers a multitude of options to customize and restrict the scope of the enumerated chemical space and thus tailor it for the demands of specific applications. To streamline the non-combinatorial exploration of chemical space, we incorporate genetic algorithms into the framework. In addition to implementing smarter algorithms, we also focus on the ease of use, workflow, and code integration to make this technology more accessible to the community.   

Machine learning and pattern recognition from surface molecular architectures.

The ability to utilize molecular assemblies as data storage devices requires capability to identify individual molecular states on a scale of thousands of molecules. We present a novel method of applying machine learning techniques for extraction of positional and rotational information from ultra-high vacuum scanning tunneling microscopy (STM) images and apply it to self-assembled monolayer of $\pi $-bowl sumanene molecules on gold. From density functional theory (DFT) simulations, we assume existence of distinct polar and multiple azimuthal rotational states. We use DFT-generated templates in conjunction with Markov Chain Monte Carlo (MCMC) sampler and noise modeling to create synthetic images representative of our model. We extract positional information of each molecule and use nearest neighbor criteria to construct a graph input to Markov Random Field (MRF) model to identify polar rotational states. We train a convolutional Neural Network (cNN) on a synthetic dataset and combine it with MRF model to classify molecules based on their azimuthal rotational state. We demonstrate effectiveness of such approach compared to other methods. Finally, we apply our approach to experimental images and achieve complete rotational class information extraction.   

Machine Learning of Accurate Energy-Conserving Molecular Force Fields

Efficient and accurate access to the Born-Oppenheimer potential energy surface (PES) is essential for long time scale molecular dynamics (MD) simulations. Using conservation of energy -- a fundamental property of closed classical and quantum mechanical systems -- we develop an efficient gradient-domain machine learning (GDML) approach to construct accurate molecular force fields using a restricted number of samples from ab initio MD trajectories (AIMD). The GDML implementation is able to reproduce global potential-energy surfaces of intermediate-size molecules with an accuracy of 0.3 kcal/mol for energies and 1 kcal/mol/{\AA} for atomic forces using only 1000 conformational geometries for training. We demonstrate this accuracy for AIMD trajectories of molecules, including benzene, toluene, naphthalene, malonaldehyde, ethanol, uracil, and aspirin. The challenge of constructing conservative force fields is accomplished in our work by learning in a Hilbert space of vector-valued functions that obey the law of energy conservation. The GDML approach enables quantitative MD simulations for molecules at a fraction of cost of explicit AIMD calculations, thereby allowing the construction of efficient force fields with the accuracy and transferability of high-level ab initio methods.