Water clusters play a significant role in a variety of processes such as those pertaining to the atmosphere and biological systems, and studies of water clusters have suggested that they could help us learn more about hydrogen bonding. We first must understand the energetics and trends of isolated water molecules in order to better comprehend the spectroscopic properties of water clusters. Afterwards, we can look at water clusters and observe how the energetics and patterns change due to the interactions with other water molecules. We are studying the coupling among vibrations in water molecules and how they are reflected in the spectra. The discrete variable representation (DVR), a method used to solve the Schrödinger equation, was implemented to generate the water spectrum as well as energies and wave functions. The DVR results show that the theoretical intensities are consistent with the experimental results. These results contribute to our goal of analyzing the spectra of more complicated water cluster systems. Diffusion Monte Carlo (DMC) is a different method that allows us to explore larger systems and is used in the analysis of the coupling in assemblies that contain multiple water molecules.

 Alkenes are ubiquitous motifs in organic synthesis and are often found among pharmaceuticals and biologically active compounds. Moreover, a diastereoselective synthesis of the thermodynamically less stable Z-alkene isomer is a highly desirable reaction. However, classic methods of generating these targets remain limited. Because of this, our group is particularly interested in exploring transition metal-catalyzed hydrofunctionalization of a terminal alkyne to produce Z-alkenes. Hydrofunctionalization is an advantageous approach as it promotes the buildup of molecular complexity from simple terminal alkyne starting materials. We previously developed a method to access Z-alkene products through the direct reaction of a terminal alkyne, a primary alkylborane, and a silver triazole catalyst. Although we incorporated a wide variety of functional groups on both the alkyne and alkylborane substrates, the reaction was limited to primary alkylboranes. As a result, the goal of our current project was to overcome this limitation through rigorous screenings of reaction conditions that would incorporate secondary alkylboranes. To support screening efforts, my role is to synthesize various alkyne substrates to continue expanding substrate scope. We were able to accomplish our goal, and couple a secondary alkylborane with a terminal alkyne with moderate yields and selectivity. We are presently working to continue improving yield, selectivity, and expanding the functional group tolerance of this reaction which was otherwise inaccessible with our previous methodology. Through incorporating more sterically complex alkylboranes we provide access to a wide variety of structurally diverse Z-alkenes.

Polymers are commonly seen in our daily lives. Proteins and plastics are both familiar classes of polymeric materials whose utility is heavily relied upon. There are different architectures of polymers, for instance, linear and cyclic, each of which has unique properties. For example, cyclic polymers have a lower viscosity, smaller hydrodynamic volume, and a unique topology as an endless circle. In this project, we are investigating and improving a privileged method to approach cyclic polymers, Ring Expansion-Metathesis Polymerization (REMP), which grows the polymer chain while cyclizing it, using a Ruthenium-based (Ru) system. The goal of the research is to solve the recent major challenge of synthesizing cyclic polymers in a controlled fashion, through systematically modifying the structure of Ru-based initiators. My goal in the research project is to synthesize a precursor ligand, the subsequent initiator, and the monomers(commonly strained alkenes, norbornene). Eventually using the monomers to conduct polymerization reactions and analyze their properties and characteristics with spectroscopic instruments. Preliminary results of this research suggest we can make cyclic polymers that are more evenly distributed in size and weight. Since polymers have played an important role in people’s everyday life, improving the methodology through having better control on making cyclic polymers can make a big contribution to applications in the aspects of biomedicine and energy for our society. For example, cyclic polymers could generate biotherapeutics for the field of medicine; they could also serve as well-behaved and new types of conducting materials for the field of semiconductor.

The theory of quantum mechanics has been well-developed over the last hundred years. However, its application is limited by the computational power of modern computers. With the rise of Big Data and Artificial Intelligence, a new door is opening to us to untangle the fascinating world of quantum mechanics. In our lab, we use Diffusion Monte Carlo (DMC), a statistical simulation to solve molecular vibration and rotation problems. It is remarkably accurate and versatile, making it suited for notoriously difficult systems, like protonated methane (CH₅⁺). Yet, it requires millions of potential energy evaluations before quality results can be acquired, which often takes unrealistic amounts of time. In this work, we use TensorFlow, a neural network training framework developed by Google, with full Graphics processing unit (GPU)-acceleration support, to considerably speed up the evaluation of the potential energies needed for the DMC calculations. We started by running a small-scale conventional DMC simulation to collect a set of molecular configurations and their corresponding potential energies, which are then fed into a 3-layer deep neural network on Tensorflow with carefully-selected parameters. Once finished training, the neural network can replace the conventional potential energy evaluation method used in DMC to greatly expedite the process. We tested this model on water(H₂O), protonated methane(CH₅⁺) and water dimer((H₂O)₂), and was able to achieve a 15-fold acceleration, with less than 0.01% error compared to conventional methods. Our future goal is to further optimize the neural network to make it even faster and more accurate, then apply it to larger systems which were unsolvable before due to their computationally intractable time.