Theoretical and Data-driven Study for New Perspectives on Material Chemistry

Abstract

With the rapid advances in experimental techniques and continued expansion of material spaces, millions of super-functional materials have been created and developed in the field of material science. These newly created super-functional materials exhibit totally different catalytic, optical, electrical, and thermal properties recording a high-value to be utilized in a wide variety of applications. Especially, the multi-composed (more than tertiary systems) and extremely downscaled (e.g. single atom catalysts, and strongly correlated systems) materials show dramatic and distinct properties under the influence of their complexity and quantum effect. However, the interpretation of these particular materials is really hard to understand by human-intuitions and the experimental observations because their properties are different from the conventional materials. Therefore, special techniques should be required to gain a more fundamental understanding of the materials than is observed in the experiments. Theoretical chemistry has emerged with the development of quantum theory and high-end supercomputers to interpret and explain the roles of electronic structures on material properties. In this regard, the physicochemical phenomena of super-functional materials could be understood with an aid of a theoretical approach using the first-principles calculations (e.g. density functional theory, coupled cluster, and configuration interaction). In fact, until now, theoretical chemistry is one of the powerful tools to elucidate and investigate the electronic structure of organic compounds, catalytic activity of energy materials, optical properties of solids, and electrical properties of perovskites. Indeed, the theoretical studies are now allowed us to obtain new insight into the material. Moreover, with the advent of machine learning, which is opened a golden period of artificial intelligence with numerous applications in visual recognition, healthcare, natural language processing, and even first-principles calculations, the data-driven studies combined with theoretical chemistry and machine learning have become very attractive in the field of material chemistry. Here, I depict the theoretical and data-driven researches showing the finding of new perspectives on material chemistry. In Section 1, general introduction and key concepts for theoretical and data-driven researches are introduced for a better understanding of this dissertation. Next, detailed examples related to the subject are described as following contents; the theoretical approach to materials, finding chemical trends with massive case-study, accelerated computational screening strategy to find promising catalysts, computational-aided material design, and derivation of the empirical equation with machine learning. In Section 2, theoretical studies for band alignment with dimensional reduction and defect control of material are investigated using the combinations of theory and experiment. The new-type of sheet-like zinc orthogermanate (Zn2GeO4, denoted as S-ZGO representing dimensional reduction) is synthesized and further modified using heat treatment to prepare the defected sample. A comparison of photocatalytic activity for water splilting are reported to investigate the effects of dimensional reduction and defects. For proving effects of them on photocatalytic activity, the band alignment is demonstrated with density functional theory showing the increased density of states at the edge of conduction band (CB) and valence band (VB), and a new defect level between CB and VB. In Section 3, A large-scale case study identifies chemical trends in carbene chemistry. From the models of diverse carbenes including CX2, C(YHn)2, and cyclic systems of C(ZHm)2, we elucidate the relationships among electron configurations, electron accepting/donating strength of atoms attached to carben center(:C), π conjugations, singlet/triplet energy gap, anisotropic hard wall radii, anisotropic electrostatic potentials, and amphotericity in carbenes, which are vital to carbene chemistry. The three specific electronic configurations (σ2, π2 or σπ) associated with :C on the :CA2 plane (where A is an adjacent atom) in singlet and triplet carbenes largely governs the amphoteric behavior along the :C tip and :C face-on directions. The :C tip and :C face-on sites of σ2 singlet carbenes tend to show negative and positive EPs favoring nucleophiles and electrophiles, respectively, whereas those of π2 singlet carbenes such as very highly π-conjugated 5-membered cyclic C(NCH)2 tend to show the opposite behavior. The open-shell σπ singlet (such as highly π-conjugated 5-membered cyclic C(CHCH)2) and triplet carbenes show less anisotropic and amphoteric behaviors. In Section 4, we introduce universal computational screening strategy that can accelerate the prediction of theoretical overpotential (ηDFT) for Oxygen Evolution/Reduction Reaction (OER/ORR) using only reaction free energy of O*. Our accelerated screening strategy can effectively reduce the computing time by skipping the costly calculations of reaction free energies of OH* and OOH*. Besides, the efficiency of accelerated screening strategy was verified using 1,008 combinations of single-atom-anchored transition metal dichalcogenides. The given candidate materials are rapidly screened using our strategy and finally 23 promising catalysts are found out of 1,008 candidates. In Section 5, a new oxygen-ligand steered SAC (M-O-C) is synthesized with a computation-aided approach. Based on theoretical calculations, the stability of various oxygen-ligand steered SACs is tested and the theoretical model proves that single metals can be stable with acetylacetonate ligands found in unzipped carbon materials. Besides, the newly designed M-O-C catalysts are experimentally realized and show a distinct electrocatalytic activity for oxygen evolution reaction (OER). The Ni-O-C shows excellent activity, followed by Co/Zn-O-C and their theoretical predictions for OER agree well with the experimental results. In Section 6, the analytic equations between band gap and properties of constituent elements (valence-electron numbers, electron affinities, melting points, van der Waals radii, group numbers, heat of vaporization and nonmetallic character, etc.) are derived using machine learning techniques suitable for small datasets: alternating conditional expectation (ACE) and projection pursuit regression (PPR). We present several models that predict the band gap with the coefficient of determination, R2 ≈ 0.9. These equations are a convenient tool in obtaining further insights into the nature of band gap and will pave the way for finding new light-active materials for optoelectronic devices.