- Mojtaba Abdolmaleki
- Harsh Agarwal
- Touheed Anwar Atif
- Hayley Beltz
- Ryan Cardman
- Seokhyun Chung
- Madelyn Cook
- Matthew De Furio
- Changyu Deng
- Bradley Dirks
- Nazanin Farjam
- Siying Feng
- Lloyd Fisher, Jr.
- Vishwas Goel
- Duncan Greeley
- Wen Guo
- Yaohui Guo
- Huanqi He
- Aidan Herderschee
- Kwanghwi Je
- Zhe Jian
- Minjun Jin
- Gurcharan Kaur
- Sayantan Khan
- Tanvir Ahmed Khan
- Virginia Larson
- Christina Lee
- Julia Lenef
- Kaiwen Liu
- Meichen Liu
- Xiangpeng Luo
- Peter MacDonald
- Larissa Markwardt
- Sean McSherry
- Cameron Pratt
- Alireza Ramyar
- Agnes Resto
- Gurmeet Singh
- Anna Stuhlmacher
- Suk Hyun Sung
- Tessa Swanson
- Michael Wadas
- Logan Walker
- Yueqiao Wu
- Ziping Xu
- Colleen Yancey
- Cheng Yang
- Troy Zehnder
- Jennifer Zupancic
Leveraging Connectivity and Automation to Revolutionize the Future of Mobility
Mojtaba Abdolmaleki, Civil Engineering
We investigate the impacts of connected and automated vehicle (CAV) technologies on improving three critical components of transportation systems. First, we exploit graph coloring techniques to solve the long-lasting question of devising polynomial-time algorithms for finding intersection controls that simultaneously optimize the throughput and delay in a CAV environment under longitudinal/lateral motion control. Second, we exploit connectivity to develop dynamic pricing and relocation policies for ride-sourcing systems that reduce data dependency and susceptibility to the system stochasticity. Also, it exponentially decreases demand loss probability compared to static policies. Lastly, we include a polynomial-time routing schedule with a tight optimality gap that minimizes the energy consumption of the truck fleet, standing for 24 percent of total energy consumption of the transportation sector, by considering the truck platooning strategy that decreases air drag force and fuel consumption.
Improving Charge Transfer in Metal Ions for Low-Cost Flow Batteries
Harsh Agarwal, Chemical Engineering
Integrating renewables like solar and wind to meet the growing electricity demand has led to a search for low-cost electricity storage technologies. Redox flow batteries (RFBs) are a promising technology to store electricity, but they suffer from high costs in part arising from the slow kinetics of charge transfer reactions at the electrodes’ surface, preventing their large-scale deployment. The lack of fundamental understanding of charge transfer has hindered the design of electrolytes and electrocatalysts to increase the reaction rate. In this dissertation, using V2+/V3+ as the probe reaction, we develop mechanistic understanding of these charge transfer reactions in different electrolytes and on different electrocatalyst surfaces. We relate the kinetics to the identified adsorbed intermediate’s energy and hydrogen binding strength, using kinetic measurements, spectroscopic techniques, and microkinetic modeling. Consequently, we identify promising materials that enhance the charge transfer and lower RFB’s capital cost, making them more competitive for grid-scale energy storage.
Distributed Measurement Compression and Capacity Results for Quantum Network Problems Using Algebraic Structured Approach
Touheed Anwar Atif, Electrical and Computer Engineering
Quantum algorithms require quantum computers performing logical operations on a sufficiently large number of entangled qubits. Unfortunately, state-of-the-art quantum computers can only operate on tens of qubits. A solution to this scalability challenge is to employ the distributed paradigm, where a network of small-scale quantum computers is used in a distributed manner. It is the aim of this thesis to study the fundamental limits of distributed quantum problems and enhance their performance using asymptotically-good algebraic codes. We provide a novel protocol to faithfully simulate a distributed quantum measurement and characterize a set of sufficient rates. We make further improvements by employing structured-POVMs built using random algebraic codes. Moreover, we study how these algebraic codes can benefit the classical-quantum network problems. We derive a new set of sufficient conditions to communicate (i) a generic function using a classical-quantum multiple access channel, and (ii) over a $3-$user classical-quantum interference channel.
Understanding Hot to Ultrahot Jupiters Through 3D Atmospheric Models and High Resolution Spectroscopy
Hayley Beltz, Astronomy and Astrophysics
One of the newest, most extreme categories of exoplanets are known as “Ultrahot Jupiters” (UHJs) which refer to gas giant planets that orbit their host stars so closely that their equilibrium temperature exceeds 2200K and their orbits become tidally synchronous—resulting in permanent daysides and nightsides of the planets with large (sometimes greater than 1000K) temperature differences. In this temperature regime, exotic processes including magnetic drag and hydrogen dissociation/recombination begin to play a critical role in shaping the atmospheric structure. These extreme physical conditions require modeling in three dimensions in order to accurately depict the atmospheric state of the planet. In this dissertation, I demonstrate that high resolution emission spectroscopy is sensitive to the 3D thermal structure of the model spectra. I then explore the effects of magnetic drag in the atmospheres of UHJs and how those effects can manifest themselves in phase curves and other observables.
Towards Fundamental Physics with Circular-State Rydberg Atoms in Space
Ryan Cardman, Physics
Major subfields of fundamental physics are concerned with the search for dark matter (DM) and the solution of the “proton-radius puzzle” (PRP). Circular-state (CS) Rydberg atoms serve well for both of these tasks, as they interact efficiently with weak microwave fields. Photons interacting with CS atoms are from a spectroscopic microwave field designed to measure the Rydberg constant, thus helping to solve the PRP, or from the Primakoff conversion of DM axions, allowing DM detection. A space-based experiment, where gravitational forces are negligible, will provide the required interaction times with the microwave photons (~0.1 s). I explore a scheme of preparing CS Rydberg atoms with radio-frequency (RF) modulated laser fields via a ponderomotive interaction, necessary for the space-based experiments. Using this method, I will study several cases of Rydberg-state transitions, as well the initialization of CS Rydberg atoms.
Bayesian Predictive Analytics for the Modern IoT system: Methodologies and Applications
Seokhyun Chung, Industrial and Operations Engineering
Revolutionary advances in Internet of Things (IoT) technologies have brought significant changes in modern complex systems. Consequently, many unsolved challenges arise, such as handling heterogeneity across devices, incorporating prior knowledge to models, and, most importantly, exploiting edge computing resources to parallelize model inference and circumvent the need to share data. This thesis studies Bayesian predictive analytics tools tackling such challenges and their applications in IoT-enabled systems. First, I focus on extrapolating sensory data collected from multiple multi-sensor systems generating heterogeneous data. The second part proposes a weakly-supervised multi-output regression approach to analyze grouped data where some observations’ labels are missing. The third part studies the integration of physical knowledge on the structure of complex processes into a data-driven prediction model, with application to Additive Manufacturing. Finally, the last part of the dissertation proposes a privacy-ensured framework for distributed learning that exploits edge computing resources while improving generalization to unseen data.
Modern and Paleo-Perspectives on the Iodine to Calcium Proxy in Foraminifera and Its Application as an Indicator of Environmental Change
Madelyn Cook, Earth and Environmental Sciences
Understanding the full extent of climate and ocean variability prior to the instrumental record is crucial for forecasting ocean response to anthropogenic climate change. Without the ability to directly measure seawater from the past, oceanographers must study past oceans through proxy, relying on firm understanding of relationships between important seawater properties (e.g., dissolved oxygen concentrations, pH, temperature, salinity, etc.) and other oceanographic variables that offer indirect clues towards past ocean and ecosystem health. However, paleo-reconstructions are only as strong our understanding of how different climate and ocean signatures are preserved through time, and their limitations in the modern ocean. My dissertation uses modern (computational and seawater-based) and paleo (marine sediment-based) methods to better understand the relationship between iodine, foraminifera (unicellular calcifying marine organisms), and dissolved oxygen—ultimately strengthening the application of measured iodine-to-calcium (I/Ca) ratios in foraminifera tests as an indicator of oceanic dissolved oxygen concentrations in the past.
Exploring the Role of Stellar Mass and Birth Environment on the Formation of Stellar Multiples
Matthew De Furio, Astronomy and Astrophysics
The formation and evolution of multiple star systems has a significant impact on the chemical composition of galaxies, our understanding of star formation processes, and the ability to form planets. Observations of nearby stars and star-forming regions have discovered that the properties of stellar multiple populations depend on the stellar mass and their birth environment. In my dissertation, I use data from the Hubble Space Telescope to explore the role of stellar density on the formation of low-mass multiples in two young star-forming regions, IC 348 and the Orion Nebula Cluster. I also investigate the influence of stellar mass on the formation of multiple systems by characterizing the small separation companion population of intermediate-mass stars using long-baseline interferometry. My results will demonstrate the effect of high stellar density on the production and survival of multiple systems and the extent to which stellar mass may enhance the formation of companions.
Modeling and Optimizing the Transport in Batteries
Changyu Deng, Mechanical Engineering
In the dissertation, we focus on modeling and optimizing the transport of these batteries, especially lithium-ion with both liquid and solid-state electrolyte. We introduce a continuum model on lithium-ion batteries and the corresponding methods to experimentally measure the parameters. Special effort is made to achieve consistency of different measurement techniques when obtaining the electrode particle diffusivity. Transport in solid-state electrolyte is different from liquid electrolyte, thus an additional continuum model is formulated to describe the transport of lithium ions and electrons in solid-state electrolyte. For optimization, two types of algorithms have been investigated to find the optimal parameters in the models. A self-direct online learning method is designed to tackle high dimensional optimization problems and reduces the number of calculations by over two orders of magnitude compared with directly applying heuristic algorithms. A battery cycling optimization framework, which considers early stopping and parallel computation, is implemented to show great performance.
Studying Singularities Using Mixed Hodge Modules
Bradley Dirks, Mathematics
We study singularities of complete intersections X inside a smooth variety Y. To do this, for hypersurfaces, we use the properties of mixed Hodge modules as developed by Saito, for example, the condition of being strictly specializable along a hypersurface and the Direct Image Theorem. For higher codimension, the analogue of the strict specializability property has not been established in the literature, so we formulate and prove this property. We also prove a restriction theorem, which allows us to restrict a Hodge module M on Y to X, in a way related to the compatibility property. As applications, we study the Hodge filtration of local cohomology, and the Fourier transform of monodromic mixed Hodge modules.
Additive Nano-Manufacturing for Customized Device Fabrication
Nazanin Farjam, Mechanical Engineering
Wearable sensors; patient driven drug delivery; personalized glucose monitoring. As the uses for customized sensing and actuation, especially within the biomedical world, become more prevalent, there is an increasing need for advanced manufacturing processes that are capable of achieving customizable nano/micro-patterning of functional materials and devices on a wide range of surfaces. However, there currently exists no universal technique that simultaneously achieves controlled high-resolution patterning in the x-y-z axes, low cost, and process efficiency with a wide range of materials. In this dissertation, I am providing new scientific knowledge and enabling tools to realize a high-resolution additive nano-manufacturing platform that combines atomic layer deposition (sub-nanometer resolution in z) with electrohydrodynamic jet printing (e-jet) (sub-micrometer resolution in x-y) by identifying the capabilities/limitations of these two technologies and leveraging them toward improved device functionality. In particular, my research will provide a new understanding of the e-jet process through the development and subsequent investigation of a high-fidelity model of the jetting dynamics. This model will provide a method for studying key process physics and identifying the process parameters that drive the printing performance. My research also provides new capabilities and knowledge that contributes to the technical advancements in this additive manufacturing platform by developing a subtractive e-jet process that enables controlled, etch patterning in the X-Y plane for the fabrication of electronics and biosensors.
Acceleration Techniques of Sparse Linear Algebra on Emerging Computer Architectures
Siying Feng, Computer Science and Engineering
Recent years have witnessed a tremendous surge of research interests in sparse basic linear algebra subprograms (SpBLAS), a fundamental building block and usually the performance bottleneck of a wide range of applications, including machine learning, graph analytics, and scientific computing. SpBLAS is notoriously memory storage intensive due to the irregular data access pattern induced by the sparse nature, which renders today’s general-purpose processors under-performant. In this dissertation, chip architecture designs with emerging techniques are proposed to accelerate SpBLAS applications. First, we proposed a flexible, programmable, and efficient general-purpose architecture with massively parallel cores connected to a 2-level memory hierarchy consisting of reconfigurable storage and interconnect. Then, an intelligent framework was built upon the reconfigurable hardware substrate to judiciously determine the best software algorithm and hardware configuration during execution for sparse matrix vector multiplication and graph analytics. Finally, we designed a near-memory multi-way merge solution for sparse matrix transposition and dataflows.
Using Time-Resolved and Nonlinear Spectroscopic Techniques to Investigate the Influence of Molecular Design and Host Material Properties on the Excited State Dynamics of Carbazole-Based Thermally Activated Delayed Fluorescent Chromophores
Lloyd Fisher, Jr., Chemistry
Thermally activated delayed fluorescent (TADF) is the notion that non-emissive states can be converted into emissive states for theoretically achievable 100 percent emission efficiency. Researchers have synthesized numerous TADF dyes for use in organic light emitting diodes (OLED). However, few in-depth spectroscopic investigations have been conducted. This thesis has three goals. First, determining if incorporating a phenylene bridge between a carbazole-based donor and triazine acceptor or directly linking the carbazole-based donor and phthalonitrile acceptor achieves greater TADF activity. Second, determining if multiple carbazole donors in conjunction with a phenylene bridge induces TADF activity. Third, investigating how the excited state dynamics of TADF systems change when going from solution to the solid state (films). I utilize nanosecond transient absorption spectroscopy to observe the excited state dynamics of emitters in solution and film. Current results show that inclusion of the phenylene bridge significantly suppresses TADF activity. These results will inform OLED design principles.
A Materials-Genome-Initiative-Based Framework for Developing Li-Ion Batteries with High Energy Density and Fast Charging Capability
Vishwas Goel, Materials Science and Engineering
For widespread adoption of electric vehicles (EVs), Li-ion-batteries (LIBs) with high energy and power density are required. However, the existing LIBs do not offer these properties simultaneously due to an inherent trade-off that arises from their electrode architecture. Thus, a modification in the electrode architecture is necessary to overcome this trade-off. In this dissertation, I have developed a Materials-Genome-Initiative-based framework, which combines computational tools, experimental measurements, and digital data to accelerate the development of electrode architectures that can deliver the desired performance. The thesis is focused on building computational tools and data-driven methods that are informed by experiments and, in turn, guide them. In collaboration with experimentalists, I have jointly studied, developed, and optimized two electrode architectures that have enabled <15-minute charging in industrially relevant energy-dense LIBs while alleviating Li-plating.
Three-Dimensional Investigation of Cyclic Deformation Mechanisms in Magnesium-Rare-Earth Alloys Using High Energy Diffraction
Duncan Greeley, Materials Science and Engineering
Magnesium-Rare-Earth (RE) alloys are promising materials for lightweighting due to their formability and high strength-to-weight ratio. Integration of Mg-RE in structural applications provides a pathway for emission reductions in the transportation industries but increased use in structural parts requires a detailed understanding of deformation behavior during cyclic loading conditions experienced in-service. In this dissertation, the role of RE-modified texture, dislocation slip activity, and twinning-detwinning during cyclic deformation were investigated using in-situ mechanical testing and crystal plasticity finite element modelling. The cyclic evolution of grain-average elastic strain and orientation in three-dimensional volumes of pure magnesium, Mg-Nd, Mg-Y, and Mg-Al alloys were characterized using High Energy X-Ray Diffraction Microscopy, and a novel data reconstruction technique was developed to track grain-scale cyclic twin activity in small twin lattices. RE-alloying is found to reduce the tensile-compressive asymmetry of cyclic twinning-detwinning through a rotated basal extrusion texture and reduced prismatic-basal critical resolved shear stress ratio.
Investigations of Biological Molecules at Interfaces Using Nonlinear Optical Spectroscopic Techniques
Wen Guo, Chemistry
Functions of biomolecules such as proteins/peptides at interfaces play important roles in many applications ranging from biomedical implants, antifouling coatings, biosensors/biochips, to antibody drugs. It is important to determine interfacial structures of proteins/peptides because such structures determine interfacial functions. For the first time, this dissertation developed a systematic method to determine molecular structures of proteins/peptides at buried solid/liquid interfaces in situ nondestructively using a nonlinear optical spectroscopic technique, sum frequency generation (SFG) vibrational spectroscopy, supplemented by linear spectroscopic methods, computer simulation, isotope labeling, and Hamiltonian spectral calculation. With this systematic approach, detailed conformation and orientation information of a variety of proteins/peptides including surface immobilized peptides for bacteria capture, surface immobilized proteins for uranyl ion binding, widely studied protein Gb1, and antimicrobial peptides interacting with model cell membranes was obtained. The developed methodology in this thesis is widely applicable to study detailed structures and molecular interactions of peptides/proteins at interfaces.
Trust-Driven Multi-Agent Human-Robot Interaction
Yaohui Guo, Industrial and Operations Engineering
Establishing an appropriate level of trust between a human and a robot is essential for seamless human-robot interaction (HRI). Previous research focused on trust calibration before the interaction process while trust can be highly dynamic in HRI. In response to this problem, we propose the following three tasks to study trust-driven HRI: (1) predicting real-time human-robot trust in HRI; (2) developing trust-driven decision-making algorithms in dyadic human-robot teams; (3) developing trust-driven decision-making in multi-agent human-robot teams. The contributions of the proposed work include: first, this research will result in fundamental contributions to understand the trust formation and evolution process when human operators interact with robotic systems; second, the research will lead to computational models for estimating human operators’ trust in automation in real-time, filling a research gap between human factor and computational robotics; third, the proposed trust-driven decision-making method will provide new strategies for other cognition-based human-robot interaction algorithms.
Characterizing the Operation and Performance of Biological Nutrient Removal in the Hybrid Membrane Aerated Biofilm Reactor (MABR) Process
Huanqi He, Environmental Engineering
Biological nutrient removal (BNR), which removes total nitrogen (TN) and total phosphorus (TP) from wastewater, is critical to prevent eutrophication. Membrane aerated biofilm reactors (MABRs) represent an aerobic biofilm technology which uses a gas-permeable membrane to deliver oxygen into the biofilm growing on the membrane. The hybrid MABR process, which combines MABRs and suspended growth, is increasingly applied in BNR facilities to meet more stringent effluent limits. However, the hybrid MABR process is still a young technology. To fill the fundamental knowledge gaps, this dissertation uses a bench-scale system in Ann Arbor, MI and a pilot demonstration in Nanjing, China to characterize the operation and performance of BNR in the hybrid MABR process. Experimental and modeling techniques reveal the impacts of wastewater characteristics and process microbiology. High-quality effluent is expected to be achieved at a fraction of the energy demand and cost of existing alternatives.
Modern Techniques for Computing Scattering Amplitudes
Aidan Herderschee, Physics
In this dissertation, we discuss some novel techniques for studying boundary correlators in flat space and anti-de Sitter space (AdS). First, we generalize connections between positive geometry and scattering amplitudes to a novel class of scalar field theories. Second, we leverage positive geometry and the wall-crossing formalism to conjecture a complete list of all algebraic branch cuts that could appear in an 8-point amplitude of N=4 super Yang-Mills. Third, we introduce a new formalism for studying the double copy of theories with higher dimension operators and show how locality and unitarity constrain the appearance of higher dimension operators in this framework. Finally, we discuss a novel formalism for computing boundary correlators in AdS called the differential representation. In addition to simplifying tree-level computations in AdS, we show that the differential representation is applicable to one-loop Witten diagrams.
Assembly Behavior of Complex Colloidal Superstructures in Systems of Polyhedral Nanoparticles
Kwanghwi Je, Chemical Engineering
Synthesis of complex superstructures through self-assembly of nano-scale colloids provides opportunities for realization of novel functional materials, including photonic crystals, metamaterials, and biosensors. Yet, there have been relatively few demonstrations of colloidal superstructures with complex morphologies observed in atomic or molecular systems. Assembly of complex colloidal superstructures requires understanding of local nanoscale ordering of colloids, and experiments have great difficulty in resolving the length and time scales of the ordering process. Computer simulations enable direct observations of the ordering process and provide insights into physical factors governing the local ordering that leads to complex structure. This thesis presents computational studies on the assembly of polyhedral nanoparticles into complex colloidal superstructures, including quasicrystals and binary crystals. I discuss computational approaches used to simulate and analyze complex ordering process. I anticipate that simulations and computation approaches demonstrated in the thesis offer new paradigms in understanding and predicting assembly of complex colloidal materials.
β-Ga2O3 Based Devices for High Power Switching Applications
Zhe Jian, Electrical and Computer Engineering
The need for efficient power generation, distribution, and delivery is quickly expanding in different sectors of industry. Power electronics is at the heart of this industrial revolution, which can be found in various applications ranging from servers, solar inverters, electric vehicles, and motor servos in industrial robotics. However, more than 10 percent of generated electricity today is wasted through conversion losses which indicates an urgent need for efficient management and distribution of electrical power. β-Ga2O3 is emerging as an attractive semiconductor for high power devices, which will enable efficient high-power switches in the 2 kV-20 kV voltage range. This thesis focuses on addressing three critical issues in Ga2O3-based devices: (i) Thermal stability of Schottky contacts to enable high power applications in harsh environments. (ii) The impact of electron mobility and device geometry on the switching characteristics of Fin field effect transistors. (iii) A robust dielectric is required to take full advantage of high breakdown field of Ga2O3.
Characterization of the Robustness of Cell-Cycle Network and Its Control in Early Embryogenesis
Minjun Jin, Biophysics
Due to the importance of cell density for cell function, it is usually tightly regulated with very little variation within a given cell type. Recently developed methods revealed that cytoplasm density changes when cells undergo growth, division, differentiation, etc. Little is known about how cellular processes cope with these cytoplasmic density variations. Here we systematically diluted or concentrated Xenopus cytoplasmic extracts and found that the cell cycle maintains robust oscillations in the range of 0.2x to 1.22x of their endogenous density. A further dilution or concentration from these values results in a cell cycle arrest. Meanwhile, the robustness of cell cycle oscillators brings insights into the onset of Mid-blastula Transition (MBT). As the first major developmental transition in Xenopus embryos, the timing of MBT is critical for embryogenesis. Here, we plan to define how the Nuclear-to-cytoplasmic (N/C) volume ratio and DNA-to-cytoplasm (DNA/cytoplasm) ratio work together to ensure proper MBT timing.
Utilizing Auto-Regulated RUNX2 Suppression in Mesenchymal Stem Cell Derived Chondrocytes to Improve Cartilage Repair and Attenuate Joint Inflammation
Gurcharan Kaur, Biomedical Engineering
Articular cartilage at the ends of long bones in joints provides a low friction, load bearing surface to facilitate smooth joint movement. When traumatically injured, cartilage tissue has limited healing capabilities due to lack of blood vessels, causing progression of an injury into post-traumatic osteoarthritis (PTOA). Mesenchymal stem cells (MSCs) provide an appealing solution for tissue engineering based cartilage repair due to their ability to transform into cartilage cells called chondrocytes and produce cartilage tissue macromolecules. However, the inflammatory environment in the injured joint poses significant challenges towards their clinical application for cartilage regeneration. Inflammation inhibits the ability of MSCs to transform into chondrocytes and secrete the matrix macromolecules required to restore healthy function of the repaired tissue. My thesis aims to utilize synthetic biology approaches to create genetically engineered MSCs that can regenerate cartilage under inflammation while also attenuating whole joint inflammation to create a conducive healing environment.
Dynamics of the Mapping Class Group of Non-Orientable Surfaces
Sayantan Khan, Mathematics
In this thesis, we study the action of the mapping class group of a non-orientable surface of demigenus d on the Teichmüller space. In the first part, we analyze the limit set of this action, and show that it is the set of projective measured foliations without any one-sided leaves, resolving a conjecture of Gendulphe. In the second part, we develop Patterson-Sullivan theory for the action of the mapping class group on the Thurston boundary, and use that to asymptotically compute the number of lattice points in a ball of radius R.
Rescuing Data Center Processors
Tanvir Ahmed Khan, Computer Science and Engineering
To serve billions of users around the world, modern web applications that run across data centers access huge datasets and perform complex application logic. As a result, data center applications face two major challenges: (1) poor data access behavior and (2) poor instruction access behavior. In my research, I demonstrate that novel hardware-software codesign effectively solves both challenges. Specifically, I observe that both data and instruction accesses in data center applications follow a deeply repetitive pattern that can be efficiently optimized by profiling the application’s program flow behavior. Consequently, I propose multiple profile-guided data and instruction access optimizations to improve the performance of data center applications. My research on data center applications’ performance optimizations has appeared in top computer architecture and systems venues like ISCA, MICRO, FAST, EuroSys, PLDI, and OSDI. Recently, I helped ARM adopt one of such performance optimization methodologies for their state-of-the-art Neoverse N1 data center processors that power 49 percent of all Amazon Web Service machines, along with Alibaba, and Microsoft data centers. My proposed performance optimization techniques are already adopted by real-world Intel processors. My profile-guided mechanism to optimize Linux Kernel is also deployed at Facebook.
Earth-Abundant Transition Metal Molecular Complexes for Water Splitting
Virginia Larson, Chemistry
Further developing and understanding the reactions of earth-abundant transition metal molecular complexes in water splitting reactions is necessary to develop a sustainable hydrogen economy. Cobalt bis(benzenedithiolate) species are adept at hydrogen evolution. I have synthesized an amine-modified catalyst and am evaluating its hydrogen evolution activity in thin films on electrode surfaces. We will also investigate photoelectrocatalytic and photocatalytic HER with the same complex. On the anodic side of water splitting, high valent metal-oxygen species are implicated as key intermediates in O-O bond formation. Similarly, they are also implicated in the reverse reaction, oxygen reduction in fuel cells. However, these intermediates are often short-lived, and their geometric and electronic structures are often not well understood. Various spectroscopic methods and density functional theory calculations are used to further understand such reactive short-lived species synthesized by our international collaborators.
Assessing Movement Control, Performance, and Independence in Upper Limb Prosthesis Users
Christina Lee, Biomedical Engineering
Upper limb amputation impacts more than 41,000 Americans. Many choose to abandon their prescribed prosthesis due to a lack of function. Therefore, my dissertation focuses on evaluating movement control and performance in upper limb prosthesis users. Specifically, I explore fundamental characteristics of prosthetic movements during goal-directed reaching and compare inter-joint coordination patterns between activities of daily living completed with prosthetic and anatomical limbs. Acknowledging the recent emergence of multi-grasp prostheses and their controllers, I also develop a novel tool to quantify the performance of different control approaches for these new prostheses. Using this tool and a virtual platform, I evaluate myoelectric grasp selection performance using regenerative peripheral nerve interfaces. I hope that my dissertation can lead to a better understanding of challenges associated with prosthetic control, with the goal of eliminating disability and physical limitations associated with amputation.
Atomic Layer Deposition for Electrochemical CO2 Recycling and Next-Generation Electronics
Julia Lenef, Materials Science and Engineering
Electrochemical conversion of CO2 into multi-carbon products using catalysts presents a unique opportunity to directly recycle CO2, while simultaneously reducing demand for petroleum inputs into manufactured products. However, the majority of research on CO2 electrocatalysis has explored planar architectures, and the relationship between the 3D architecture and catalytic performance has rarely been studied. As a result, it is not well understood how spatial inhomogeneities of the catalyst and the 3-D support affect the resulting product distribution. In my research, we employ atomic layer deposition (ALD), a gas-phase deposition technique to achieve sub-nanometer thickness and compositional control while maintaining high conformality within high-aspect ratio and 3-D supports. So far, I have developed a novel plasma-enhanced ALD process to tune catalyst phase, oxidation state, and film morphology, and demonstrated p-type thin-film transistors. Next, we are working towards applying this synthetic framework to the ALD electrocatalysts for CO2 recycling on 3-D architectures.
Control and Learning Methods for Safe Autonomous Driving
Kaiwen Liu, Aerospace Engineering
This dissertation develops game-theoretic based decision frameworks to handle interactions between human-driven vehicle and autonomous vehicle (AV) and learning frameworks to operate AV in unknown circumstances, with emphasis on ensuring safety during the interaction and learning process. Firstly, predictive models for human interactions are developed based on game theory and a decision-making framework is proposed based on hybrid estimation and optimal control with explicit safety characterization. The effectiveness of the proposed approach is validated based on a comprehensive set of simulation-based case studies. Secondly, two learning schemes are developed for AV to autonomously explore its limits. The learning schemes rely on control theory and set theory, ensure the safety of AV when operating in unknown circumstances, and are verified based on several autonomous driving applications. The main contributions of this dissertation are developments of autonomous driving control and learning technologies that explicitly accounted for and ensured safety during their operations.
Insights on Earthquakes and the Composition of Earth’s Mantle from an Analysis of Seismic Waves
Meichen Liu, Earth and Environmental Science
Seismic waves deliver valuable messages from earthquakes and the mantle. Seismographs deployed around the world record earthquake faulting processes and the interaction of waves with Earth’s layers. My thesis includes seismological studies that address two frontier research questions: how do earthquake faults rupture and how can seismic waves be used to map the composition of the mantle. Using a new spectra analysis of seismic waves, I investigate the change of stress on shallow and deep earthquake faults to address the physics of earthquakes in the crust and mantle (Chapters 1, 2, and 3). Additionally, I use seismic imaging techniques to map the phase transitions of rock-forming minerals in the upper mantle (Chapter 4) and to estimate the attenuation of seismic waves in the lower mantle (Chapter 5). These studies help us constrain how convection in Earth’s mantle affects Earth’s structure.
Detection and Manipulation of Complex Electric and Magnetic Dipole Textures in Three- and Two-Dimensional Crystals
Xiangpeng Luo, Physics
In this thesis, I present the research of multipolar ordered states—complex textures of electric or magnetic dipoles—studied by static and ultrafast nonlinear optical spectroscopy and inelastic light scattering. In the first part, I will describe our improving the sensitivity of second harmonic generation (SHG) and the incorporated pump-probe scheme. I show with SHG the first detection of a ferro-rotational quadrupolar charge density wave in 1T-TaS2, then with added time-resolution the detection of complex collective excitations of this order and the ultrafast manipulation of them through strong optical pumping. In the second part, I will discuss the magnetism in two-dimension (2D) and introduce the magnetic properties in the layered magnet CrI3. Magneto-Raman signatures of spin waves and magnetism-coupled phonons, dynamic fingerprints of layered antiferromagnetic orders are identified. In twisted double bilayers I further show nontrivial spin textures and demonstrate the unprecedented morié engineering of 2D magnetism.
Structured Latent Space Models for Multiplex Networks
Peter MacDonald, Statistics
Network data are often collected through the observation of highly complex and interconnected systems, leading to multiplex network data. Methods in statistical network analysis are traditionally designed for a single network, but applying these methods to an aggregated multiplex network can miss important layer-heterogeneous structure in the data. In this dissertation, we provide a comprehensive framework for modeling, estimation, inference, and visualization of multiplex networks accompanied by layer-specific auxiliary information. This auxiliary information could consist of, for instance, a grouping or time-ordering of the network layers. We parameterize the multiplex network expectation through structured low-dimensional positions, which govern the activity of each node on each layer. We provide adaptive estimation approaches, establish theoretical results for recovery of unknown parameters, and apply our methods to simulated and real data, including networks of international trade and political interactions.
Characterizing Trojan Asteroid Populations Throughout the Solar System
Larissa Markwardt, Astronomy and Astrophysics
For my thesis, I have developed a general “Shift and Stack” tool which can be used to detect faint solar system objects. I have focused on using this tool to search for new Trojan asteroids, as these objects are an ideal use case for this tool. Discoveries of such Trojan asteroids will improve our understanding of the primordial Solar System and its dynamical history. I have conducted two surveys of ETs with ground-based telescopes to try to make these discoveries. I will also conduct follow-up observations to study the surfaces of NTs in order to constrain the formation and evolution of our outer Solar System.
Spectral Control of Thermal Emission at High Temperatures and in the Near Field
Sean McSherry, Chemical Engineering
Spectral control of thermal emission is an enabling capability for thermal energy storage and molecular sensing technologies. However, there remain two key challenges for achieving spectral control of thermal emission at high temperatures and for on-chip devices. Here, I demonstrate two breakthrough technologies that address these challenges. Specifically, I demonstrate a BaZr0.5Hf0.5O3/MgO heterostructure, which offers tunable spectral control over high-temperature thermal emission, while mitigating common high-temperature failure modes in nanophotonics. This new material system, which represents the most thermally stable nanophotonic to date, can improve the operating efficiency of thermal energy storage technologies by ~20 percent. Next, I refocus my efforts to develop an on-chip infrared spectrometer. For the first time, this technology implements tunable and narrowband near-field thermal radiation as a mechanism to sense infrared vibrational modes from an analyte medium. This fundamental achievement, made possible through recent nanofabrication breakthroughs, has the potential to transform portable, infrared sensors.
Optimizing the Extraction of the Sunyaev—Zel’dovich Effect
Cameron Pratt, Astronomy and Astrophysics
Galaxies grow via the accumulation of ambient gas that pervades the space between them. Evidently, the infall of gas is extremely complex, and there exist many factors that facilitate/impede the flow of gas. The Sunyaev–Zel’dovich (SZ) effect is a relatively new technique that has the potential to shed a lot of light on such processes. Unfortunately, there is currently a lack of reliable, publicly available SZ data to conduct such studies. Our goal is to make major improvements in the quality of SZ data and make them available to the public.
Power Processing Architectures for Sustainable Power and Energy
Alireza Ramyar, Electrical and Computer Engineering
Power processing transforms energy so it can be used for work, extracted from clean power generation, or stored effectively and sustainably. I work on architectures and methods for efficiently extracting the power in solar photovoltaic systems, optimally storing energy in second-use batteries, and developing a canonical framework to design, optimize, and control heterogeneous energy networks. I use the intrinsic energy storage (diffusion capacitance) in solar cells to balance the power among mismatched cells using Diffusion Charge Redistribution (DCR). Intermittent energy generation like solar requires intermediate energy storage that can be produced sustainably. I optimize the power processing within second-use battery energy storage systems using Lite-Sparse Hierarchical Partial Power Processing (LS-HiPPP) to minimize cost yet maximize performance. I construct a framework for heterogeneous energy networks using pseudograph representations with operations for transformation and calculation of metrics. This energy network framework can further extend the capability and performance of DCR and LS-HiPPP.
Modeling Human Epiblast Cyst Morphogenesis
Agnes Resto, Mechanical Engineering
Human development is arguably the most complex process we’d want to study. Understanding how it takes place from the bottom up has strong implications for understanding infertility and reproductive success as well as advancing the field of regenerative medicine. An event of particular importance in early human development is the formation of the epiblast cyst and its development into the amniotic sac. In my work, I have combined an in vitro human embryonic stem cell (hESC)-based model of epiblast cyst formation with image processing tools and in silico modeling to begin to parse the mechanisms at work during early human embryogenesis. The aims of this work include: (1) creation of a computational platform for in silico experimentation, (2) creation of a live cell imaging analysis pipeline, and (3) validation of the in silico model and coupling with in vitro experimentation for studying epiblast-like cyst morphogenesis.
Multiscale Modeling of Vitrimers and Additively Manufactured Materials
Gurmeet Singh, Aerospace Engineering
Epoxies, used as binders in aircraft composites, are inherently non-recyclable. There has been tremendous interest in developing alternative epoxies that can be recycled. Vitrimers are one such promising system that combines recyclability with self-healing behavior, allowing sustainable use of composite materials. Despite the clear advantages, the mechanical response of vitrimers is not well characterized and this has limited its widespread use in aerospace applications. The thesis aims to address this gap by developing mechanical response models for vitrimers while solving a major challenge in vitrimer modeling which is accounting for dynamic cross-linking during thermal cycling. A topological reaction model was developed in which reaction templates are used along with a reaction rate model to initiate and equilibrate chemical reactions in molecular dynamics simulations. Self-healing and its effect on mechanical properties are modeled for the first time in literature, providing insights into the design of processing methods to improve mechanical properties.
Optimal Scheduling and Control of Uncertain Coupled Power-Water Distribution Networks
Anna Stuhlmacher, Electrical and Computer Engineering
Drinking water distribution networks can be treated as flexible, controllable assets to the power grid by leveraging the power consumption of water pumps and storage capabilities of water tanks. Therefore, water distribution networks can be operated as flexible loads to provide multiple local and grid level services—such as voltage and frequency regulation—to the power grid. To achieve this, an integrated water-power optimization problem is developed subject to the water and power network constraints and multiple sources of uncertainty. The framework solves for the scheduled water distribution network operation based on forecasted water and power demands and real-time water pump power adjustments to respond to a frequency regulation signal or a local constraint violation. Convex reformulations and emerging stochastic and robust solution approaches are identified and evaluated to ensure that the framework is accurate, scalable to larger networks and scheduling horizons, and computationally tractable.
Exploring 2D Materials in 3D Via Advanced Electron Microscopy
Suk Hyun Sung, Materials Science and Engineering
Two dimensional (2D) materials often exhibit unexpected, emergent physical properties—often due to their confined dimensionality and unique symmetry breaking. However, 2D materials still exist in the 3D world. They often consist of multiple atomic layers, not perfectly flat, and interact with environment above and below them. Therefore, understanding of full 3D structure of 2D materials is paramount to harnessing the true potential of 2D materials. In this dissertation, we will discuss advanced electron microscopy techniques to probe out-of-plane information of various 2D materials. Furthermore, we manipulate out-of-plane interactions to stabilize latent quantum ground state at well above room temperatures. Lastly, we engineer periodically modulated out-of-plane interaction forming Moiré patterns of 2D materials. This work spans theoretical, computational, and experimental results.
Utilizing Big Data for Evaluating Access and Resilience
Tessa Swanson, Industrial and Operations Engineering
Access to essential services determines individuals’ ability to meet health, safety, and social needs that enable them to thrive in their daily lives. But such access is not equitable and this inequity can be exacerbated when a community is faced with a disruption. Opportunistically collected location-based services (LBS) data available from cell phones offers new opportunities to evaluate access to essential services by revealing regular mobility patterns as well as deviations from those patterns following a disruptive event. Using LBS data, I develop large-scale data-driven methods for detecting changes in individuals’ mobility following disruptions as well as changes in availability of essential service facilities. I then use outputs from these methods to evaluate the impact of access to essential services on recovery through statistical modeling. I contribute to quantifying the impacts of equity and access on community resilience to motivate infrastructure solutions that facilitate more equitable and resilient communities.
Hydrodynamic Phenomena and Mixing at Interfaces Subjected to Finite-Amplitude Pressure Waves
Michael Wadas, Mechanical Engineering
Finite-amplitude pressure waves occur in compressible flows following large and sudden disturbances, such as the detonation of mining explosives or the rapid transit of a high-velocity aircraft. When these waves interact with material interfaces, they may become attenuated or strengthened, and the mixing they induce has critical implications in many scientific and engineering contexts. For example, shock-induced mixing of astrophysical material naturally follows supernova explosions and may govern the structure of an ensuing nebula. Furthermore, in inertial confinement fusion, mixing leads to a reduction in compression that severely penalizes fusion yields. The focus of this dissertation is improving our understanding of these complex mixing processes to advance astrophysics, fusion research, and many other applications. This is accomplished through a synergy of hydrodynamic theory, simulations, and experiments that explores and characterizes novel hydrodynamic phenomena and regimes of mixing.
A Multimodal Investigation of Mouse Cortical Inhibitory Circuitry Convergence
Logan Walker, Biophysics
It has long been known that the mammalian brain is constructed of thousands of distinct cell types which are individually defined by their gene expression, morphology, connectivity to other neurons, and electrophysical response to stimulus. Inhibitory neurons (INs) are one class of these neurons, which have broad disease implications. In this work, I propose to use a broad toolkit of biochemical and bioinformatic tools to investigate the role of INs in the neural circuitry of the mouse cortex, including multiround expansion microscopy and novel neuron morphometric programs. Through this multimodal approach, we will gain the most complete understanding of the morphology and connectivity of inhibitory networks in the mouse cortex, and how different molecular cell subtypes are involved in the network. The tools that I create will be robustly validated and we commit to share these tools with other researchers in the neuroscience community under the FAIR principles.
A Non-Archimedean Approach to Sasaki-Einstein Metrics and K-Stability
Yueqiao Wu, Mathematics
Sasaki-Einstein metrics are Einstein metrics on Sasaki manifolds, which are odd-dimensional analogues of Kähler manifolds. As predicted and proved by Collins and Sezékelyhdi, the existence of such metrics is equivalent to the cone over the Sasaki manifold being equivariant K-stable. In this thesis, we explore this equivalence using a non-Archimedean approach from a variational point of view. We prove that the Monge-Ampère energy has limit slope equal to a non-Archimedean Monge-Ampère energy. We also study, along the way, some non-Archimedean pluripotential theory on polarized affine varieties. We show that we can extend the non-Archimedean Monge-Ampère energy to be defined on a broader class of functions on the Berkovich analytification of the variety. The main ingredient in the proof is the theory of integration on Berkovich spaces developed by Chambert-Loir and Ducros.
Benefits of Transfer Learning in Reinforcement Learning
Ziping Xu, Statistics
One of the hallmarks of natural intelligence, especially human intelligence, is that it can solve a wide variety of tasks. Moreover, humans often require only a small amount of experience to become competent at new tasks that are similar to previously encountered ones. However most artificial intelligence and machine learning (ML) research is still focused on learning a single task and display poor generalization to new tasks. To bridge this gap between natural and artificial intelligence, this dissertation focuses on the intersection between two important ML problems: Reinforcement Learning (RL) and Transfer Learning (TL). We theoretically studied the learning benefits of TL on various important ML problems: we showed the statistical benefits of TL in supervised learning setting and exploration benefits of TL in RL. Moreover, we designed practical TL algorithms that significantly improve the performances of RL in real-world problems including email marketing and chemistry.
The Biosynthetic Repertoire of Microcystis Aeruginosa in Western Lake Erie Harmful Algal Blooms: Insights into Synthesis of Known and Novel Compounds in a Natural Population Using a “Multi-Omic” Approach
Colleen Yancey, Earth and Environmental Science
Cyanobacterial harmful algal blooms are a persistent, annual hazard in the western basin of Lake Erie. These blooms, which are dominated by the cyanobacterium Microcystis aeruginosa, degrade freshwater ecosystems and pose risks to human and animal health as a result of toxin production encoded in complex genomes. In this dissertation, I sequence DNA and RNA obtained directly from lake water to examine the secondary metabolite repertoire of Microcystis in western Lake Erie as it relates to known and novel compound production, the functionality of these molecules in situ, and targets for future public health and biotechnical research. Results reveal Microcystis produces several, unmonitored toxins in Lake Erie, as well as novel compounds whose synthesis may be influenced by the abundance of competing/predatory organisms. Finally, this dissertation provides a framework for integrating multiple “data types” (DNA, RNA, chemical profiles) to better understand the chemical microbiology of complex environmental samples.
The Development of Electrocatalytic Strategies for Lignin Degradation: From Mechanistic Investigation to Catalyst Design
Cheng Yang, Chemistry
One of the greatest challenges facing modern society today concerns improving environmental sustainability, specifically by limiting anthropogenic climate change from fossil fuel usage. The work detailed herein aims to develop selective and efficient transformations for lignin degradation, a renewable chemical feedstock isolated from tree bark, using electricity as a more sustainable energy source. Specially, the research focuses on electrocatalytic lignin oxidation mediated by hydrogen atom transfer (HAT) catalysts, Phthalimide N-oxyl (PINO) and its derivatives. A comprehensive mechanistic study has provided new insight to advance PINO-catalyzed electrochemical oxidations, inspiring the design of more efficient HAT mediators. Furthermore, electro-reductive fragmentation of oxidized lignin has also been explored. We anticipate the incorporation of electrochemistry and principles for lignin valorization to be of great value to a greener chemical production process and a sustainable future.
Olefination of Functionalized Imines Mediated Ruthenium Alkylidenes
Troy Zehnder, Chemistry
Access to highly functionalized alkenes found in chemical feedstocks, pharmaceuticals, and natural products provides a desirable focus for the development of new carbon-carbon bond forming reactions. Previous approaches to imine-olefin metathesis have shown great potential for functionalized alkene synthesis, but remain limited by stochiometric formation of catalytically inert metal imide products. In my dissertation, I will demonstrate that functionalized imines—hydrazones and oximes—can undergo a productive metathesis reaction with olefins mediated by Ru-alkylidenes to access functionalized alkenes. Utilization of functionalized imines as reactive partners unlocks catalysis via formation of a reducible Ru-nitride intermediate. Specifically, I report development of the first method for intramolecular olefination of oximes and hydrazones, investigation into the mechanism of this transformation, and proof of principle for catalysis in the presence of suitable carbene precursors. Future work will focus on improving catalytic turnover and demonstrating the utility of functionalized imine-olefination for intramolecular reactions.
Engineered Nanobodies for Penetrating the Blood-Brain Barrier and Targeting Pathological Species in Alzheimer’s Disease
Jennifer Zupancic, Chemical Engineering
Neurodegenerative diseases, such as Alzheimer’s and Parkinson’s diseases, remain largely untreatable. One of the key problems in treating these disorders is the inability to efficiently deliver biologics to the brain due to the blood-brain barrier (BBB). A second challenge is developing biologics with strict conformational specificity for only targeting misfolded pathological proteins and ignoring the same proteins in their normal conformations. To address these challenges, we have developed a novel approach for generating high-affinity nanobodies in a surprisingly simple manner, and we applied this methodology to generate nanobodies specific for tau aggregates associated with Alzheimer’s disease and a novel BBB protein (CD98hc) associated with efficient transcellular transport. These nanobodies are currently being combined into bispecific nanobody shuttles to optimize delivery of tau nanobodies to the brain parenchyma in wild-type and tau transgenic mice. This research holds great potential to improve brain delivery of diverse off-the-shelf nanobodies and antibodies.