Graph convolutional neural network applied to the prediction of normal boiling point.

In this article, we describe training and validation of a machine learning model for the prediction of organic compound normal boiling points. Data are drawn from the experimental literature as captured in the NIST Thermodynamics Research Center (TRC) SOURCE Data Archival System. The machine learning model is based on a graph neural network approach, a methodology that has proven powerful when applied to a variety of chemical problems. Model input is extracted from a 2D sketch of the molecule, making the methodology suitable for rapid prediction of normal boiling points in a wide variety of scenarios. Our final model predicts normal boiling points within 6 K (corresponding to a mean absolute percent error of 1.32%) with sample standard deviation less than 8 K. Additionally, we found that our model robustly identifies errors in the input data set during the model training phase, thereby further motivating the utility of systematic data exploration approaches for data-related efforts.

Assessing Arsenic Species in Foods Using Regularized Linear Regression of the Arsenic K-edge X-ray Absorption Near Edge Structure.

The toxicity and bioavailability of arsenic is heavily dependent on its speciation. Therefore, robust and accurate methods are needed to determine arsenic speciation profiles for materials related to public health initiatives, such as food safety. Here, X-ray spectroscopies are attractive candidates as they provide in situ, nondestructive analyses of solid samples without perturbation to the arsenic species therein. This work provides a speciation analysis for three certified reference materials for the food chemistry community, whose assigned values may be used to assess the merit of the X-ray spectroscopy results. Furthermore, extracts of SRM 3232 Kelp Powder, which is value-assigned for arsenic species, are measured to provide further evidence of its efficacy. These analyses are performed on the results of As K-edge X-ray Absorption Near Edge Structure (XANES) measurements collected on each sample. Notably, such analyses have traditionally relied on linear combination fitting of a minimal subset of empirical standards selected by stepwise regression. This is known to be problematic for compounds with meaningfully collinear spectra and can yield overestimates of the accuracy of the analysis. Therefore, the least absolute shrinkage and selection operator (lasso) regression method is used to reduce the risk of overfitting and increase the interpretability of statistical inferences. As this is a biased statistical method, results and uncertainties are estimated using a bootstrap method accounting for the dominant sources of variability. Finally, this method does not separate model and data selection from regression analysis. Indeed, a survey of many spectral influences is presented including changes in the: state of methylation, state of protonation, oxidation state, coordination geometry, and sample phase. These compounds were all included in the model's training set, preventing model over-simplification and enabling high-throughput and robust analyses.

Predicting Kováts Retention Indices Using Graph Neural Networks.

The Kováts retention index is a dimensionless quantity that characterizes the rate at which a compound is processed through a gas chromatography column. This quantity is independent of many experimental variables and, as such, is considered a near-universal descriptor of retention time on a chromatography column. The Kováts retention indices of a large number of molecules have been determined experimentally. The "NIST 20: GC Method/Retention Index Library" database has collected and, more importantly, curated retention indices of a subset of these compounds resulting in a highly valued reference database. The experimental data in the library form an ideal data set for training machine learning models for the prediction of retention indices of unknown compounds. In this article, we describe the training of a graph neural network model to predict the Kováts retention index for compounds in the NIST library and compare this approach with previous work [1]. We predict the Kováts retention index with a mean unsigned error of 28 index units as compared to 44, the putative best result using a convolutional neural network [1]. The NIST library also incorporates an estimation scheme based on a group contribution approach that achieves a mean unsigned error of 114 compared to the experimental data. Our method uses the same input data source as the group contribution approach, making its application straightforward and convenient to apply to existing libraries. Our results convincingly demonstrate the predictive powers of systematic, data-driven approaches leveraging deep learning methodologies applied to chemical data and for the data in the NIST 20 library outperform previous models.

Site-Dependent Spin Delocalization and Evidence of Ferrimagnetism in Atomically Precise Au25(SR)180 Clusters as Seen by Solution 13C NMR Spectroscopy.

We report a simple but detailed solution 13C nuclear magnetic resonance spectroscopic study of atomically precise neutral Au25(SR)180 (SR = alkyl thiolate) clusters. The paramagnetic 13C Knight shift of alkyl chain carbons, which is proportional to the local electron spin density, exhibits an electron spin delocalization that exponentially decays along the alkyl chain. The magnitude and decay constant of the observed electron spin delocalization, although largely independent of alkyl chain length, depend on where, that is, "in" versus "out" (vide infra) position, the alkyl chain is bound, in agreement with density functional theory calculations. Notably, the determined position-dependent decay constants, 1.70/Å and 0.41/Å for "in" and "out" ligands, respectively, not only could have important ramifications in molecular spintronics but are also comparable to measured decay constants in molecular electrical conductance of alkyl chains, potentially offering an alternative, simple method for estimating the latter. Moreover, the negative intercept temperatures of linear fits of reciprocal 13C (as well its bound 1H) Knight shift versus temperature strongly suggest the existence of local ferrimagnetism in individual Au25(SR)180 clusters.

Effect of surface-bound sulfide on oxygen reduction reaction on Pt: Breaking the scaling relationship and mechanistic insights.

A combined in situ electrochemical attenuated total reflection-surface enhanced IR absorption spectroscopy, microkinetic simulation, and density functional theory calculation study shows that not only can the adsorbed sulfide disproportionally affect the surface binding of OOH* (EOOH* ) vs OH* (EOH* ), i.e., breaking the original scaling relationship of pure metals (Ir, Pd, Pt, Au), to enhance oxygen reduction reaction (ORR) activity but can also be used as a reaction pathway alternating species to help deepen our mechanistic understanding of ORR.

Semiempirical configuration interaction calculations for ru-centered dyes.

Computational investigation of the photochemical properties of transition-metal-centered dyes typically involves optimization of the molecular structure followed by calculation of the UV/visible spectrum. At present, these steps are usually carried out using density functional theory (DFT) and time-dependent DFT calculations. Recently, we demonstrated that semiempirical methods with appropriate parameterization could yield geometries that were in very good agreement with DFT calculations, allowing large sets of molecules to be screened quickly and efficiently. In this article, we modify a configuration interaction (CI) method based on a semiempirical PM6 Hamiltonian to determine the UV/visible absorption spectra of Ru-centered complexes. Our modification to the CI method is based on a scaling of the two-center, two-electron Coulomb integrals. This modified, PM6-based method shows a significantly better match to the experimental absorption spectra versus the default configuration interaction method (in MOPAC) on a training set of 13 molecules. In particular, the modified PM6 method blue-shifts the location of the metal-to-ligand charge-transfer (MLCT) peaks, in better agreement with experimental and DFT-based computational results, correcting a significant deficiency of the unmodified method. Published 2018. This article is a U.S. Government work and is in the public domain in the USA.

One-dimensional ionic self-assembly in a fluorous solution: the structure of tetra-n-butylammonium tetrakis[3,5-bis(perfluorohexyl)phenyl]borate in perfluoromethylcyclohexane by small-angle neutron scattering (SANS).

Fluorous liquids are the least polarizable condensed phases known, and their nonpolar members form solutions with conditions the closest to being in vacuo. A soluble salt consisting of a large fluorophilic anion, tetrakis[3,5-bis(perfluorohexyl)phenyl]borate, and its counterion, tetra-n-butylammonium, dissolved in perfluoromethylcyclohexane produces ionic solutions with extremely low conductivity. These solutions were subjected to small-angle neutron scattering (SANS) to ascertain the solute structure. At concentrations of 9% mass fraction, the fluorophilic electrolyte forms straight, long (>160 Å) self-assembled structures that are, in essence, long, homogeneous cylinders. Molecular models were made assuming a requirement for electroneutrality on the shortest length scale possible. This shows a structure formed from a stack of alternating anions and cations, and the structures fit the experimental scattering well. At the lower concentration of 1%, the stacks of ion pairs are shorter and eventually break up to form solitary ion pairs in the solution. These characteristics suggest such conditions provide an interesting new way to form long, self-assembling ionic nanostructures with single-molecule diameters in free solution onto which various moieties could be attached.

Predicting Structures of Ru-Centered Dyes: A Computational Screening Tool.

Dye-sensitized solar cells (DSCs) represent a means for harvesting solar energy to produce electrical power. Though a number of light harvesting dyes are in use, the search continues for more efficient and effective compounds to make commercially viable DSCs a reality. Computational methods have been increasingly applied to understand the dyes currently in use and to aid in the search for improved light harvesting compounds. Semiempirical quantum chemistry methods have a well-deserved reputation for giving good quality results in a very short amount of computer time. The most recent semiempirical models such as PM6 and PM7 are parametrized for a wide variety of molecule types, including organometallic complexes similar to DSC chromophores. In this article, the performance of PM6 is tested against a set of 20 molecules whose geometries were optimized using a density functional theory (DFT) method. It is found that PM6 gives geometries that are in good agreement with the optimized DFT structures. In order to reduce the differences between geometries optimized using PM6 and geometries optimized using DFT, the PM6 basis set parameters have been optimized for a subset of the molecules. It is found that it is sufficient to optimize the basis set for Ru alone to improve the agreement between the PM6 results and the DFT results. When this optimized Ru basis set is used, the mean unsigned error in Ru-ligand bond lengths is reduced from 0.043 to 0.017 Å in the set of 20 test molecules. Though the magnitude of these differences is small, the effect on the calculated UV/vis spectra is significant. These results clearly demonstrate the value of using PM6 to screen DSC chromophores as well as the value of optimizing PM6 basis set parameters for a specific set of molecules.

Application of an Artificial Neural Network to the Prediction of OH Radical Reaction Rate Constants for Evaluating Global Warming Potential.

Rate constants for reactions of chemical compounds with hydroxyl radical are a key quantity used in evaluating the global warming potential of a substance. Experimental determination of these rate constants is essential, but it can also be difficult and time-consuming to produce. High-level quantum chemistry predictions of the rate constant can suffer from the same issues. Therefore, it is valuable to devise estimation schemes that can give reasonable results on a variety of chemical compounds. In this article, the construction and training of an artificial neural network (ANN) for the prediction of rate constants at 298 K for reactions of hydroxyl radical with a diverse set of molecules is described. Input to the ANN consists of counts of the chemical bonds and bends present in the target molecule. The ANN is trained using 792 (•)OH reaction rate constants taken from the NIST Chemical Kinetics Database. The mean unsigned percent error (MUPE) for the training set is 12%, and the MUPE of the testing set is 51%. It is shown that the present methodology yields rate constants of reasonable accuracy for a diverse set of inputs. The results are compared to high-quality literature values and to another estimation scheme. This ANN methodology is expected to be of use in a wide range of applications for which (•)OH reaction rate constants are required. The model uses only information that can be gathered from a 2D representation of the molecule, making the present approach particularly appealing, especially for screening applications.

Estimation of Radiative Efficiency of Chemicals with Potentially Significant Global Warming Potential.

Halogenated chemical substances are used in a broad array of applications, and new chemical substances are continually being developed and introduced into commerce. While recent research has considerably increased our understanding of the global warming potentials (GWPs) of multiple individual chemical substances, this research inevitably lags behind the development of new chemical substances. There are currently over 200 substances known to have high GWP. Evaluation of schemes to estimate radiative efficiency (RE) based on computational chemistry are useful where no measured IR spectrum is available. This study assesses the reliability of values of RE calculated using computational chemistry techniques for 235 chemical substances against the best available values. Computed vibrational frequency data is used to estimate RE values using several Pinnock-type models, and reasonable agreement with reported values is found. Significant improvement is obtained through scaling of both vibrational frequencies and intensities. The effect of varying the computational method and basis set used to calculate the frequency data is discussed. It is found that the vibrational intensities have a strong dependence on basis set and are largely responsible for differences in computed RE values.

First-Principles Prediction of Enthalpies of Formation for Polycyclic Aromatic Hydrocarbons and Derivatives.

In this article, the first-principles prediction of enthalpies of formation is demonstrated for 669 polycyclic aromatic hydrocarbon (PAH) compounds and a number of related functionalized molecules. It is shown that by extrapolating density functional theory calculations to a large basis set limit and then applying a group based correction scheme that good results may be obtained. Specifically, a mean unsigned deviation and root mean squared deviation from the experimental enthalpies of formation data of 5.0 and 6.4 kJ/mol, respectively, are obtained using this scheme. This computational scheme is economical to compute and straightforward to apply, while yielding results of reasonable reliability. The results are also compared for a smaller set of molecules to the predictions given by the G3B3 and G3MP2B3 variants of the Gaussian-3 model chemistry with a mean unsigned deviation and root mean squared deviation from the experimental enthalpies of formation of 4.5 and 4.8 kJ/mol, respectively.

Formic acid oxidation on platinum: a simple mechanistic study.

The oxidation of small organic acids on noble metal surfaces under electrocatalytic conditions is important for the operation of fuel cells and is of scientific interest, but the basic reaction mechanisms continue to be a matter of debate. Formic acid oxidation on platinum is one of the simplest of these reactions, yet even this model system remains poorly understood. Historically, proposed mechanisms for the oxidation of formic acid involve the acid molecule as a reactant, but recent studies suggest that the formate anion is the reactant. Ab initio studies of this reaction do not address formate as a possible reactant, likely because of the difficulty of calculating a charged species near a charged solvated surface under potential control. Using the recently-developed joint density functional theory (JDFT) framework for electrochemistry, we perform ab initio calculations on a Pt(111) surface to explore this reaction and help resolve the debate. We find that when a formate anion approaches the platinum surface at typical operating voltages, with H pointing towards the surface, it reacts to form CO2 and adsorbed H with no barrier on a clean Pt surface. This mechanism leads to a reaction rate proportional to formate concentration and number of available platinum sites. Additionally, high coverages of adsorbates lead to large reaction barriers, and consequently, we expect the availability of metal sites to limit the experimentally observed reaction rate.

Measuring level alignment at the metal-molecule interface by in situ electrochemical (13) C NMR.

A new technique to measure energy-level alignment at a metal-molecule interface between the Fermi level of the metal and the frontier orbitals of the molecule is proposed and experimentally demonstrated. The method, which combines the electrochemistry of organo-ligand-stabilized Au nanoparticles with (13) C NMR spectroscopy (i.e. in situ electrochemical NMR), enables measuring both occupied and unoccupied states.

Inverse-micelle-encapsulated water-enabled bond breaking of dialkyl diselenide/disulfide: a critical step for synthesizing high-quality gold nanoparticles.

Inverse-micelle-encapsulated water formed in the two-phase Brust-Schiffrin method (BSM) synthesis of Au nanoparticles (NPs) is identified as essential for dialkyl diselenide/disulfide to react with the Au(III) complex in which the Se-Se/S-S bond is broken, leading to formation of higher-quality Au NPs.

Different mechanisms govern the two-phase Brust-Schiffrin dialkylditelluride syntheses of Ag and Au nanoparticles.

Here we report the first unambiguous identification of the chemical structures of the precursor species involving metal (Au and Ag) ions and Te-containing ligands in the Brust-Schiffrin syntheses of the respective metal nanoparticles, through which the different reaction pathways involved are delineated.

Mechanistic insights on one-phase vs. two-phase Brust-Schiffrin method synthesis of Au nanoparticles with dioctyl-diselenides.

Metal precursors in the one-phase (1p) and two-phase (2p) Brust-Schiffrin method (BSM) synthesis of Au nanoparticles (NPs) using dioctyl-diselenides were identified. A single dominant type of metal precursor was found in the 1p synthesis as compared to multiple ones in the 2p synthesis, which was proposed as the key reason why the former is better than the latter.

Evaluation of methods to predict reactivity of gold nanoparticles.

Several methods have appeared in the literature for predicting reactivity on metallic surfaces and on the surface of metallic nanoparticles. All of these methods have some relationship to the concept of frontier molecular orbital theory. The d-band theory of Hammer and Nørskov is perhaps the most widely used predictor of reactivity on metallic surfaces, and it has been successfully applied in many cases. Use of the Fukui function and the condensed Fukui function is well established in organic chemistry, but has not been so widely applied in predicting the reactivity of metallic nanoparticles. In this article, we will evaluate the usefulness of the condensed Fukui function in predicting the reactivity of a family of cubo-octahedral gold nanoparticles and make comparison with the d-band method.

Silicon carbide nanostructures: a tight binding approach.

A tight-binding model Hamiltonian is newly parametrized for silicon carbide based on fits to a database of energy points calculated within the density functional theory approach of the electronic energy surfaces of nanoclusters and the total energy of bulk 3C and 2H polytypes at different densities. This TB model includes s and p angular momentum symmetries with nonorthogonal atomic basis functions. With the aid of the new TB model, minima of silicon carbide cagelike clusters, nanotubes, ring-shaped ribbons, and nanowires are predicted. Energetics, structure, growth sequences, and stability patterns are reported for the nanoclusters and nanotubes. The band structure of SiC nanotubes and nanowires indicates that the band gap of the nanotubes ranges from 0.57 to 2.38 eV depending on the chirality, demonstrating that these nanotubes are semiconductors or insulators. One type of nanowire is metallic, another type is semiconductor, and the rest are insulators.

Dynamic and structural properties of aqueous arsenic solutions.

Aqueous arsenic solutions are important in geochemistry, environmental chemistry, and biochemistry. Thus, studies of arsenic solutions that probe the structure of the hydrated arsenic ion and the behavior of water around As ions have wide-ranging implications. The mechanisms for water dissociation around three arsenic ions (see picture) and structural details of the resulting aqueous complexes are presented.