Physical Sciences and Engineering

Advanced Electron Microscopy of Quantum Materials Across Extreme Temperature Range
Nishkarsh Agarwal, Materials Science and Engineering

The rich physics permeating the phase diagram of newly discovered quantum materials and their potential for transformative applications have driven extensive research into exotic phase transitions, including superconducting, ferroic, and charge-order transitions. Many of these transitions involve electronic, magnetic, or lattice order, which emerge on atomic to mesoscopic scales and remain incompletely understood. Moreover, stabilizing these fragile states requires precise control of temperature, magnetic fields, or electrical bias. In this context, in-situ experiments performed in the transmission electron microscope (TEM) under precisely controlled conditions, modern detectors, and resolution down to the atomic-level, offer unique insights into the nature and dynamics of these quantum phase transitions. My dissertation explores in-house development and usage of advanced in-situ TEM techniques to probe a range exotic quantum materials–including magnetic moiré lattices, charge density waves systems and intercalated layered materials, at temperatures ranging from cryogenic (23 K) to high-temperatures (>1173 K).

Microbial Mineralization for Subsurface Engineering
Eva Albalghiti, Environmental Engineering

Climate change mitigation and adaptation necessitates utilizing the energy production, energy storage, and carbon sequestration capabilities of the subsurface. Doing so safely and economically will, however, require tools for controlling subsurface fluid flow and sealing potential leakage pathways. Microbial mineralization is one promising candidate, but despite years of research, the factors influencing microbial mineralization in natural rocks remain poorly understood. This dissertation aims to close the knowledge gap through innovative approaches to experiments on natural rock samples.

In chapter 1, the effect of pore structure on the sealing capability of microbial mineralization is assessed, and a mechanism of flow-induced precipitate removal from pore walls is proposed to explain results. Chapter 2 introduces the additional variable of mineralogy, which is found to work in tandem with pore size to influence microbial attachment to solid surfaces. Finally, chapter 3 examines how microbes contribute to mineralization activity differently on surfaces versus in suspension.

Characterizing the Space of UV-Complete Amplitudes
Justin Berman, Physics

Scattering amplitudes in quantum field theory are the fundamental quantities physicists use to predict the behavior of particles at high energies. Dispersion relations for two-particle to two-particle amplitudes explicitly display how the particle content of a theory affects its physical observables and highlight how physical requirements like unitarity (probability preservation), causality (no action without cause), and locality (no action at a distance) place constraints on amplitudes. In my thesis, I describe how analyzing these dispersion relations allows us to find two-sided bounds on the masses of hadrons in large-N QCD, identify field-theoretic input which makes the open superstring the unique tree-level UV completion of maximally supersymmetric Yang-Mills, and completely map uncolored identical scalar theories with a one-parameter family of “extremal” amplitudes.

The Origin of Metals in Galaxy Clusters: Characterizing the Early Enrichment Population
Anne Blackwell, Astronomy and Astrophysics

Metals (anything heavier than helium) are made by stars, but when and where those stars existed in the universe is an outstanding problem. We know these stars must exist because we measure approximately eight times the amount of expected metals in massive structures called galaxy clusters. The leading, yet untested, theory is a stellar population existing early in the universe: the early enrichment population (EEP). For my dissertation, I am producing the first observable and testable EEP quantities. I first derive the contribution of the EEP to galaxy cluster metals, then use that value to calculate two observables: number of supernovae (death of high-mass stars) as a function of time, and remaining light in galaxy clusters from the low-mass stars that have longer lives. Comparing predicted to observed values defines the initial conditions of the EEP. My dissertation will lay the groundwork for theoretical predictions, finally testing the EEP theory.

Fast Summation for Geophysical Fluid Dynamics
Anthony Chen, Applied and Interdisciplinary Mathematics

Geophysical fluid dynamics is the study of fluids in which rotation and the Coriolis force play a role and is of great interest and importance, both theoretically and practically, as the foundation of modern weather and climate modeling. Many problems in geophysical fluid dynamics can be formulated in a way to take advantage of fast summation techniques, which are methods for approximating sums at reduced asymptotic complexity, such as the classical tree code or fast multipole method. This thesis describes a cubed sphere tree code (CSTC) suitable for the fast approximation of convolutions on the sphere, and the application of this to the problems of computing self attraction and loading and the time integration of the barotropic vorticity and shallow water equations. The CSTC is parallelized for a variety of architectures, including CPU only configurations as well as GPUs.

Revealing the Assembly of Galaxies by Probing the Life Cycle of Star Clusters
Yingtian Chen, Astronomy

Many stars in the night sky are bound together in compact groups known as star clusters, which are observed in all types of galaxies. The life cycle of star clusters involves three stages: 1) birth in clouds, 2) evolution and migration through galaxies, and 3) dissolution into stellar streams. The interactions between clusters and their host galaxies play important roles in shaping the cluster properties in all stages. In previous years, I built comprehensive models to uncover the mysteries encoded in these interactions, with particular emphasis on the second and third stage. In the final year of my Ph.D., I will employ cutting-edge hydrodynamical simulations to explore the first stage, when clusters directly record information about the earliest galaxies. By disentangling the complex link between cluster properties and the galactic environment, I aim to provide theoretical insights into how galaxies have assembled and evolved over time.

Exploring Cosmic Ray Feedback and Its Observational Signatures in Cosmic Ray Magneto-Hydrodynamical Simulations of Normal and Ram-Pressure Stripped Galaxies
HuaiHsuan Chiu, Astronomy and Astrophysics

Cosmic rays (CRs) are particles that are in equipartition with magnetic and thermal energy in the interstellar medium (ISM). As such, they play a key role in galactic feedback processes. In this dissertation, we first generate mock observations for a CR magneto hydrodynamical (MHD) simulation that integrates two-moment CR transport physics and a multi-phase ISM model using the steady-state assumption for CR loss processes. We then explore the local far infrared (FIR)-radio correlation and assess the validity of the equipartition assumption within the simulation. We generate mock observations for ram-pressure stripped galaxies in a wind tunnel CR-MHD simulation, evolving the CR electron spectrum by relaxing the steady-state assumption and evolving the spectrum of CRs on the fly in the simulation. By comparing these CR-MHD simulations with actual observations, this work provides essential constraints on the models of CR feedback and key processes shaping galaxy formation.

From Dyadic to Non-dyadic Human-Autonomy Interaction, Exploring Trust Dynamics and Human Behaviors
Hyesun Chung, Industrial and Operations Engineering, Barbour Scholar

This dissertation examines human trust and behavior dynamics in human-autonomy interactions, addressing the gap in research on non-dyadic interactions. It aims to develop predictive models by incorporating personal characteristics, autonomy features, team dynamics, and system design to support the creation of user-aware autonomous systems that foster effective and trustworthy collaboration.

The research comprises three phases. The first phase examines trust dynamics in dyadic interactions. It identifies three distinct trust dynamics types and their associated personal characteristics, which are used to create a personalized real-time trust prediction model. The second phase extends the scope to nondyadic interactions, where perceptions and behaviors are influenced by peer actions and team dynamics. Empirical studies using novel testbeds identify key factors, such as transparent communication, that shape these interactions. The third phase integrates insights from earlier phases, enhancing the predictive model to account for trust and behavior dynamics across diverse interaction scenarios and team-level metrics.

Developing Methods for Environmental Microplastic Identification
Madeline Clough, Chemistry

Understanding the chemical identities, and therefore likely sources, of environmental microplastics (MPs) is essential to informing future regulation and remediation. However, common strategies to assign polymer identities to unknown particles lack statistical certainty, jeopardizing the validity of MP quantification. To enhance confidence in MP identification, we report a machine-learning model known as conformal prediction to provide a guaranteed probability that the correct MP identity will be returned in a set of predicted identities. We next use conformal prediction to address glove-based contamination, which uniquely impacts MP analysis due to high spectral similarity between polymer identities and small molecule contaminants. Lastly, we apply conformal prediction to atmospheric samples collected across Michigan to quantify MPs in breathing air. In total, these chapters aid researchers to accurately identify MPs, enhancing our ability to best protect the environment and human health from this emerging contaminant.

Teleost Fish Evolution in the Cenozoic: New Insights from Paleogene Ichthyofaunas from the Eastern Desert of Egypt
Sanaa Elsayed, Earth and Environmental Sciences

This dissertation examines the response of acanthomorph (spiny-rayed) fishes to the Cretaceous-Paleogene (K-Pg) mass extinction (ca. 66 Ma) based on a remarkable new Egyptian fossil locality dating to ca. 62 Ma. The site yields a diverse paleotropical marine fish assemblage representing 23 taxa across 13 orders and 20 families, including the earliest occurrences of seven major groups. These fossils indicate rapid transition to compositionally modern fish assemblages, highlighted by the emergence of new acanthomorph groups no later than four million years after the K-Pg extinction. Quantitative comparisons with other faunas from elsewhere in the world imply an important role for the paleotropics in the evolution of today’s acanthomorph diversity. Two key fishes from the site, Mene (Carangiformes), which exhibits evolutionary stasis, and a new scombrid (Scombriformes), which replaced fast-swimming Cretaceous taxa, are the focus of detailed anatomical and phylogenetic analyses. Clumped-isotope analyses provide constraints on the environment inhabited by this fish community.

Electrically-Small Space-Time Modulated Electromagnetic Structures for Improved Bandwidth and Efficiency
Zachary Fritts, Electrical and Computer Engineering

Amidst the rapid technological advances leading to high-speed electronic devices with small form factors, there is one technology that cannot easily be miniaturized: antennas. This is because antenna size scales with wavelength. Antennas that are small compared to the wavelength of operation can either be efficient or broadband, but they cannot be both–a tradeoff quantified by the “efficiency-bandwidth product.” This dissertation focuses on surpassing physical limits to the efficiency-bandwidth product of a system by making one or more parameters of the system vary periodically in time. Breaking the fundamental assumption of time-invariance allows the efficiency-bandwidth product of the antenna to be increased five to 10 times over conventional small antennas. Crucially, the time-variation takes the special form of spatially-discrete traveling-wave modulation (SDTWM). In this work, we showcase how the special space-time rotational symmetry of SDTWM allows performance improvements in small antennas.

Identifying the Structure Function Trends in Rieske Oxygenases for the Development of Biocatalysts
Alejandro Garcia, Chemistry

Rieske oxygenases are underexplored enzymes for selective C-H bond functionalization. These metalloenzymes rely on a Rieske cluster and a non-heme iron site. These entities transfer electrons and form an Fe-based oxidant that is used to initiate a divergent set of reactions. Within the breadth of reaction outcomes, Rieske oxygenases have garnered interest for use as biocatalysts for bioremediation and pharmaceutical production. However, the applicability of Rieske oxygenases is limited by a gap in knowledge regarding their structure-function relationships. To this end, we explored and probed the residues that bridge the metallocenters to understand the electron transfer pathway in the protein family. Furthermore, by identifying the influence of different residues and protein components on quaternary architecture, we promoted stability and enhanced activity in our model systems. By applying our findings to disparate Rieske oxygenases and by bolstering our findings with bioinformatic analyses, we provided a framework for future Rieske oxygenase engineering.

Illuminating the Environmental Drivers of Extinction in the Plio-Pleistocene West Atlantic
Lucas Gomes, Earth and Environmental Sciences

The study of past extinction events provides unique insight into how ecosystems function and respond to major environmental changes. This dissertation investigates the environmental drivers of an extinction event that occurred in the West Atlantic Ocean during the Plio-Pleistocene transition (ca. 3.5 to 1.5 million years ago). Here I characterize the Plio-Pleistocene stratigraphy of fossil-rich sediments in south Florida, refining our understanding of the environmental and sea-level history of the Florida Platform. Building on this work, I analyze the trace elemental and carbonate clumped isotope chemistry of fossil shells to quantitatively reconstruct (i) planktonic productivity and (ii) oceanic water temperatures through the extinction interval. These studies enable the first systematic evaluation of marine climate change and productivity shifts as potential extinction drivers in the Plio-Pleistocene West Atlantic, informing future predictions of marine biodiversity loss as modern marine ecosystems face steadily warming ocean temperatures and human-induced alterations to local nutrient cycling.

Novel Quantum Phenomena in Kagome Metals and Strongly Correlated Electron Systems
Kaila Jenkins, Applied Physics

Quantum oscillations in strong magnetic fields give us direct insight into the band structure of solids via the quantization of electron motion. Metals with a Kagome lattice structure host both dissipationless “flat-band” physics, as well as the Dirac electronic dispersion, which can be detected by quantum oscillations. A crucial question remains about the fate of the small orbits when the magnetic field is strong enough to push the electronic state to the quantum limit. By utilizing proximity detector oscillators, torque magnetometry cantilevers, and capacitance spectroscopy in high magnetic fields, we resolved strong quantum oscillations corresponding to Fermi surface orbits and other novel magnetic features in several Kagome metals. Quantum oscillation results in both pulsed and DC fields will be compared to find the electron effective mass and Fermi surface orbits of several compounds. This thesis aims to understand flat-band and topological physics in the broad family of Kagome metals.

Integrating Formal Methods and Human Feedback for Safe and Personalized Control Synthesis
Ruya Karagulle, Electrical and Computer Engineering

The advancement of autonomous technologies is transforming everyday domains ranging from transportation to personal assistance. These application areas are often safety-critical, where failure can lead to severe consequences, including damages, injuries, and even fatalities. At the same time, because these systems are in close contact with humans, their adoption and trust heavily depend on their ability to align with user preferences. Therefore, designing autonomous systems that are both provably safe and customizable to user-specific preferences is a crucial part in advancing next-generation autonomy. This thesis addresses the dual objectives of safety and personalization. While preference learning has been widely studied in conjunction with reinforcement learning, existing methods lack the necessary safety guarantees for critical applications like autonomous driving. To bridge this gap, this research leverages formal methods—mathematically rigorous techniques for specifying and verifying system behavior—within preference learning frameworks to ensure that learned preferences adhere to strict safety constraints. By integrating formal methods with machine learning techniques, this research aims to develop a unified framework for synthesizing safe and personalized behaviors which represents a significant step toward bridging the gap between safety guarantees and personalization in autonomy.

Hodge Modules on Toric Varieties and Applications of the Canonical Bundle Formula in Birational Geometry
Hyunsuk Kim, Mathematics

We study the intersection cohomology and the trivial Hodge modules on toric varieties. More precisely, we describe the torus action on the cohomology of the graded de Rham complexes of the intersection cohomology. Moreover, by analyzing the structure of the trivial Hodge module QHX, we give some results regarding the depth of the Du Bois differentials and the local cohomological dimension. Along the way, we also give several combinatorial results whose central component of the proof uses a detailed understanding of QHX.

Next, we give some results regarding the canonical bundle formula. We verify the b-semi-ampleness conjecture by Prokhorov and Shokurov when the fibers are primitive symplectic varieties, thus answering a question of Fujino. We also give an application toward the equivalence of the non-vanishing and Campana–Peternell conjectures, improving Schnell’s condition on the pseudo-effectivity of the canonical bundle downstairs and providing an inductive set-up using rigid currents.

Engineered Encapsulins as Advanced Drug Delivery Platforms and Catalytic Nanoreactors
Seokmu Kwon, Chemical Engineering

Protein nanocages have recently gained attention as a versatile technology platform for therapeutics delivery and biocatalysis. Encapsulins—naturally occurring microbial protein nanocompartments—in particular, are attractive engineering targets due to their robust self-assembly and selective protein packaging capabilities. While traditional drug delivery has focused on either protein or RNA delivery, concurrent delivery has not been widely explored. I engineered encapsulins with RNA-binding peptides, capturing target RNAs without disrupting protein loading. The resulting selective RNA and protein co-encapsulation system can co-deliver bioactive RNAs and proteins, with the potential to enhance therapeutic effects.

Encapsulins have also been exploited to sequester non-native enzymes to create enzyme nanoreactors. However, their pore properties often limit the accessibility of enzyme substrates compromising nanoreactor performance. To address this, I engineered encapsulins with larger pores, optimizing their shell porosity and molecular flux behavior. Leveraging this advancement, I constructed proof-of-concept nanoreactors confirming improved catalytic performance.

Computational Tools for Rational Kinetics-Oriented Drug Design
Thanh Lai, Biophysics

For some protein targets, drug-target residence time—or the duration of time a drug molecule is bound to its target protein—is a stronger indicator of in vivo drug efficacy than conventional metrics such as binding affinity. Residence time is dictated by the unbinding free energy landscape that a ligand traverses during its unbinding process. However, rational optimization of a ligand’s residence time (i.e. ”kinetics-oriented” drug design) is difficult because its unbinding pathway and transition state is inaccessible experimentally. Thus, computational tools capable of providing atomistic resolution of this process are indispensable for designing drugs with long residence times. To this end, we present high-throughput computational tools to detect hidden small molecule unbinding pathways in protein structures, and a methodology called MSλD+US that can characterize the unbinding free energy landscapes for multiple ligands in order to clarify and accelerate the process of kinetics-oriented drug design.

Decision-Making for Autonomous Vehicles Considering Human Interactions and Perception Uncertainties
Xiao Li, Aerospace Engineering

Autonomous driving (AD) vehicles must safely navigate human traffic while adapting to diverse conditions. Current AD technologies struggle with safety and reliability, especially with uncertainties in deep neural network (DNN) perception and interactions with human drivers.

My dissertation develops a resilient and adaptive AD framework incorporating robust uncertainty quantification, human behavior modeling, and system-level safety guarantees. Utilizing an ensemble of DNNs and conformal prediction, it provides statistically sound uncertainty estimates to enhance robustness against out-of-distribution observations. A behavioral model with social psychological insights and Bayesian estimation predicts human drivers’ intentions for safer decision-making. These methods form a conformal tube model predictive controller for uncertainty-aware planning. Additionally, a system-level safety guard incorporates uncertainties in the algorithmic loop, ensuring robust performance. This framework significantly advances safe, reliable AD technologies, with wider implications for robotics and human-machine interaction.

Mathematical Frameworks for Curb Space Infrastructure Design, Operational Policies, and User Behaviors in Next Generation Mobility Systems
Jisoon Lim, Civil Engineering

Transportation, a cornerstone of societal development, has evolved from enabling mass movement to addressing individual travel needs. Today, it is undergoing a transformation driven by advances in connectivity, autonomy, the sharing economy, and electrification. Curb space exemplifies this evolution, where demand has grown more diverse and intense. As a vital interface for urban activities, curb space serves both traditional and emerging mobility needs, offering opportunities to foster efficient, equitable, and sustainable transportation systems. This dissertation proposes mathematical frameworks to analyze curb space use and management. These models inform the design of infrastructure and operational policies that integrate technological layers, optimize resource allocation, and address diverse user needs. By integrating behavioral insights and user interactions, the theoretical models are transformed into practical solutions, enabling user-responsive and adaptive curb space management strategies. The dissertation also tackles scalability challenges of mathematical models to support socially inclusive and functional next-generation urban mobility systems.

Inactivating Airborne Viruses: A Study on Non-Thermal Plasma Inactivation of Viral Aerosols and Its Characteristics
Zhenyu Ma, Environmental Engineering

Aerosols are responsible for the airborne transmission of many diseases, which have caused significant losses and continue to threaten public health. In order to stop the airborne transmission of pathogenic aerosols, one critical step is to achieve a sufficient extent of air disinfection. In this study, the non-thermal plasma (NTP) technology is tested for its effectiveness in inactivation of airborne viruses. NTP is found to cause virucidal effects on airborne MS2 bacteriophage viruses within an exposure time of less than one second. Several factors affecting the NTP inactivation efficiency are also tested, including the presence of chemically reductive gas pollutants, the plasma power source, and the plasma reactor configuration. Ultraviolet fluorescence is used to study the change in viral aerosol characteristics after exposure to NTP, from which a correlation is found between viral aerosol fluorescence intensity and virus inactivation level from NTP treatment.

Advances in Artificial Intelligence Evaluation
Felipe Maia Polo, Statistics

This dissertation develops novel methods to address key challenges in evaluating modern artificial intelligence (AI) systems like large language models (LLMs). Current AI evaluation methods are often inefficient, costly, and inadequate for assessing the complex capabilities of contemporary AI. Drawing on psychometrics and statistical modeling, this work introduces efficient, interpretable, and scalable approaches for AI evaluation. The first chapter presents a method for compressing LLM benchmarks and reducing evaluation costs while maintaining accuracy through item response theory (IRT). The second chapter tackles prompt sensitivity in LLMs by introducing PromptEval, an IRT-based technique for efficient multi-prompt evaluation, requiring minimal computational resources. The third chapter proposes skill-based scaling laws for LLMs, modeling latent abilities like reasoning and providing robust performance predictions. The final chapter advances weak supervision evaluation by estimating bounds for key metrics using scalable convex optimization, addressing the lack of ground truth labels.

Modular Control of Hydrogel Structure to Drive Vascular Assembly and Host Integration to Support Ovarian Tissue Grafts
Firaol Midekssa, Biomedical Engineering

Tissue engineering aims to develop constructs that restore, maintain, or improve tissue function. Capillary beds within 100 to 200 µm of cells supply essential nutrients and oxygen to the surrounding tissue. As such, engineering viable tissues larger than 1mm³ requires microvasculature. Traditional vascularized grafts utilize soft biomaterials that suffer from poor mechanical stiffness, rapid resorption, and inadequate surgical handleability. This thesis aims to develop a robust vascularization strategy in stiff, long-lasting grafts for cell replacement therapies. Ovarian autotransplantation and cryopreservation therapy (OTCT), a cell replacement hormone therapy, is promising for treating infertility and restoring ovarian function. However, issues such as delayed vascular anastomosis, hypoxia, and ischemia cause stromal cell damage, follicle loss, and inflammation, reducing ovarian tissue transplant lifespan. Using a combination of cell biology techniques, materials engineering, and in vivo models, this thesis work will promote the revascularization of ovarian tissue grafts in the efforts to enhance the efficiency of OTCT.

Evolutionary and Ecological Insights from Fruits of Extinct and Extant Araceae
Jeronimo Morales Toledo, Earth and Environmental Sciences

Fruits are indispensable dispersal units that drive plant evolution and distribution through ecological interactions. Structural diversity of fruits underpins the success of flowering plants, which are the most species rich group of land plants and dominate terrestrial ecosystems. While much is known about fruit types across plant families, the structural evolution of fruits and the ecological shifts they reflect over Earth’s history remain poorly understood. My dissertation addresses this gap by integrating data from present-day and fossil fruits to characterize fruit evolution in the diverse aroid family. Exceptional fossilized fruits from the USA and Northern Africa provide evidence for the first time of climbing growth form in the fossil record for aroids, while a novel large dataset of living fruit diversity helps contextualize evolutionary patterns in the family. This approach bridges extant and fossil data, offering deeper insights into the evolution of terrestrial ecosystems.

Social Health by Design: Varied Effects of the Built Environment on Social Health
Hannah Myers, Design Science

The built environment—places we live, work, learn, and play—is not health-neutral. Substantial evidence supports this claim, yet most built environment and health (BEH) research has focused on physical and mental health, neglecting social health. Meanwhile, our nation’s social health—the aspect of health pertaining to social relationships—is in crisis. Some BEH studies suggest that built environment interventions may improve social health, but more research is required. This dissertation contributes three original BEH studies to address this gap. The first two chapters examine built environment elements within economically-disadvantaged neighborhoods. Chapter one finds a negative relationship between blight (vacant or neglected properties) and social health, suggesting blight remediation could improve social health. Chapter two investigates how blight undermines the positive social health effects of neighborhood elements that afford social connection. Chapter three explores how occupants of a graduate residence hall perceive its design to impact their social health.

Rare Cell Enrichment and Bioinformatic Analysis Enabled via Microfluidic Platforms
Neha Nagpal, Chemical Engineering

Disease management is often convoluted by a lack of a reliable biomarker. This thesis probes the diagnostic and prognostic value of rare, diseased cells captured with microfluidic technologies as part of a liquid biopsy (blood draw). The focus of the thesis is on two main diseases: ductal carcinoma in situ (DCIS) and HIV. First, DCIS patients face a risk of progression to more advanced breast cancer, and this thesis investigates circulating tumor cells (CTCs) as a prognostic biomarker to ultimately determine appropriate treatments for each patient. Second, HIV patients carrying latent virus are not fully treated despite long-term therapies, and this thesis develops a microfluidic platform to isolate single, latent HIV-infected CD4+ T cells. The goal is to probe how these latent viruses affect the transcriptomic environment of the cells they infect and to determine better treatment strategies. Overall, the results of this thesis aim to improve patient outcomes.

Mechanistic Study of Organometallic Intermediates in 3-D Transition Metal-Mediated C–H Bond Functionalization
Emily Nolan, Chemistry

With the majority of current chemical processes involving catalytic transformations, catalysis is vital to the production of commodities including polymers, fuels, pharmaceuticals, and agrochemicals. Transition metal catalysis has traditionally relied on precious metals such as iridium, palladium, rhodium, and ruthenium, which are subject to supply fluctuations due to their limited abundance in the earth’s crust. Thus, ongoing research efforts are aimed at decreasing reliance on precious metals by replacing them with abundant first-row transition metals such as nickel, iron, cobalt, and copper. This dissertation aims to expand the ubiquity of late first-row transition metals in C–H bond functionalization by increasing fundamental mechanistic understanding of these processes. More specifically, this work contains three projects that focus on isolation and reactivity studies of organometallic intermediates relevant to nickel-, cobalt-, and iron-catalyzed C–H functionalization reactions.

(Non)Vanishing in the Cohomology of Symplectic Groups
Urshita Pal, Mathematics

In this thesis we establish a strong algebraic relationship between the rational cohomology of the symplectic groups Sp_{2n}Z and general linear groups GL_nZ, and use this to construct previously unknown non-zero cohomology classes for Sp_{2n}Z. Work of Borel-Serre and Bieri-Eckmann shows how one can study these group cohomologies via certain simplicial complexes, namely the Solomon-Tits buildings of Types A and B. Recently, Avner Ash, Jeremy Miller, and Peter Patzt used these buildings to show the existence of a free graded Hopf algebra structure on H∗(GL_nZ; Q). In this thesis, we use these complexes to construct a module and comodule structure on H∗(Sp_{2n}Z; Q), and show that H∗(Sp_{2n}Z; Q) forms a Hopf module over a certain quotient of H∗(GL_nZ; Q). This, combined with the work of Ash-Miller-Patzt, allows us to ‘multiply’ together cohomology classes of GL_nZ and Sp_{2n}Z to get new cohomology classes of Sp_{2n}Z.

Understanding and Controlling the Selectivity of Ion-Exchange Membranes Toward Like-Charged Ions
Harsh Hemantkumar Patel, Chemical Engineering

Ion exchange membranes (IEMs) with the capability to selectively differentiate between counter-ions or like-charged ions could revolutionize separation processes such as lithium extraction, scalant removal, and nitrate recovery. Advancing the development of IEMs for such applications requires a deeper understanding of the relationship between membrane structure and counter-ion selectivity. While significant progress has been made in understanding single-electrolyte transport phenomena in IEMs, there is a notable gap in research addressing counter-ion selectivity in IEMs under mixed salt conditions. My dissertation systematically explores mixed ion partitioning and diffusion in IEMs to bridge this critical knowledge gap. My work emphasizes achieving selectivity between ions of differing valences (e.g., monovalent/divalent, divalent/trivalent) and between ions of the same valence (e.g., monovalent/monovalent). The insights gained from this work will provide a foundation for the rational design of membranes with enhanced and tunable ion-selectivity, enabling applications in resource recovery, water treatment, and contaminant removal applications.

Density-Controlled Ion Transport in Amorphous Metal Oxides: Implications for Next-Generation Energy and Microelectronic Technologies
Dongjae Shin, Materials Science and Engineering

Ion transport in amorphous materials is critical for emerging energy and microelectronic technologies, yet our fundamental understanding remains limited. This dissertation investigates the relationship between density and ion transport in amorphous metal oxides, focusing on hafnium oxide (HfO2) thin films. We demonstrate that density variations of over 20 percent can be achieved through vapor deposition techniques. Using a new tracer diffusion method with isotopically enriched multilayer structures, our results reveal that atomic layer deposited HfO2 exhibits significantly slower oxygen diffusion compared to sputtered films, attributed to differences in material density. This work establishes density as a key design parameter for controlling ion transport in amorphous materials, contrasting sharply with traditional crystalline material principles. The findings have direct implications for the performance and reliability of resistive memory devices, providing new design strategies for next-generation energy and microelectronic applications.

Modeling, Design, and Demonstration of Electrochemical Reactors to Enable Circular Use of Carbon-Based Energy Carriers
Rachel Silcox, Mechanical Engineering

This work develops and validates a novel electrochemical method for carbon dioxide (CO2) removal from ocean water, while also functioning as energy storage. By leveraging ocean water, a key CO2 sink affected by ocean acidification, the method has the potential to operate continuously despite renewable energy intermittency. Using a cyclic pH-swing process with proton-coupled reactions, energy use and cost are minimized through modeling and testing a lab-scale reactor. The reactor’s performance is optimized to enhance mass transport and reduce costs via design and operational parameters. Results indicate improvements in fouling management, cost savings, and integration with renewable energy systems. The research addresses scale-up challenges by consulting industry experts and exploring uses for the extracted CO2, such as converting it to methanol. This comprehensive approach combines modeling, lab experiments, and scalability analysis to accelerate the uptake of carbon removal technologies and safeguard the environment from the impacts from climate change.

Multimodal Fusion and Temporal Reasoning for Intelligent Robot Perception
Jingyu Song, Robotics

Enabling robust, generalizable robot perception in challenging environments remains an open problem. This thesis proposes advanced methods in multimodal fusion and temporal reasoning to address challenges in robotics such as low-texture conditions, sensor failures, and environmental disturbances. I first design TURTLMap, which enables marine autonomy with a multimodal framework fusing acoustic and visual sensors for localization and mapping, validated in real-world tests. Then, I design LiRaFusion and CRKD to achieve adaptive and scalable sensor fusion for autonomous driving. I also develop MemFusionMap, an advanced technique to enhance temporal scene understanding for robots operating in highly complex environments. Finally, I propose a unified framework to learn a generalizable foundation backbone via self-supervised learning with temporal fusion of multimodal data. This framework will improve various downstream perception tasks for autonomous vehicles and general-purpose robots. Collectively, these contributions offer a comprehensive perception architecture that fuses diverse data sources across time, empowering robots with advanced perceptual capabilities.

Operator Learning: Statistical Foundations and Scalable Methods
Unique Subedi, Statistics

Operator learning, a novel framework for learning maps between infinite-dimensional function spaces, has emerged as a powerful tool for scientific computing. A primary application is the development of fast and accurate surrogate models for the solution operators of partial differential equations (PDEs), which describe system dynamics across various scientific domains. It can also develop simulators for systems where the underlying mathematical models are unknown. While promising, operator learning faces two significant challenges: underdeveloped statistical foundations and scalability to real-world systems.

This dissertation addresses these limitations in two parts. The first part establishes statistical guarantees, focusing on sample complexity and error quantification unique to operator learning. The second part advances its scalability by introducing an active data collection strategy that reduces sample requirements and using mathematical structures of PDEs to design efficient learning architectures. These methods are empirically validated by learning solution operators for PDEs like Poisson’s, heat, and time-dependent Schrödinger equations.

Stem Cell-Based Embryo and Organ Models for Studying Early Human Development
Shiyu Sun, Mechanical Engineering, Barbour Scholar

The human embryo undergoes complex cellular and morphogenetic transformations during gastrulation and organogenesis. Yet, these critical stages of human development remain poorly understood and difficult to study. Human pluripotent stem cell (hPSC)-based embryo and organ models (e.g., embryoids and organoids) are powerful tools for studying human gastrulation and organogenesis. This thesis presents new human peri-gastrulation embryoids and gastrointestinal and cardiac organoids generated using bioengineering approaches. Human peri-gastrulation embryoids can mimic key aspects of human peri-gastrulation development, including trilaminar embryonic disc formation, amnion and yolk sac development, and primary hematopoiesis. Bioengineered gastrointestinal organoids can faithfully recapitulate morphogenetic dynamics during gastrointestinal development. Cardiac organoids, developed using microfluidic systems, can mimic heart tube formation and patterning, providing insights into early human cardiovascular development. Together, these models offer exciting new tools and insights into early human development and demonstrate the potential of bioengineering to create advanced, controllable models of human development.

Engineering an ECM Sequestering Biomimetic Platform to Improve In Vitro Ovarian Follicle Culture Outcomes
Emily Thomas, Biomedical Engineering

First-line cancer treatments pose significant long-term risks to survivors’ health, notably impacting fertility due to gonadotoxic effects. Current fertility preservation methods, such as egg and embryo cryopreservation, are not viable for all patients, particularly those with hormone-responsive cancers and pediatric patients. This dissertation addresses these limitations by developing a novel biomimetic platform to culture ovarian follicles, preserved prior to cancer treatment, in vitro. By engineering a hyaluronic acid-based hydrogel modified with extracellular matrix (ECM)-sequestering peptides, critical ECM proteins are retained, promoting follicle growth and development. Additionally, the incorporation of electrospun dextran-vinyl sulfone fibers enhances the structural mimicry of native ovarian tissue. This dual approach supports ovarian follicle maturation and provides a controlled environment to study follicle-ECM interactions. The platform not only advances fertility preservation options but also holds potential applications in other areas of women’s health research, such as screening suspected environmental toxins and developing treatments for ovarian diseases.

Fundamental Understanding of Far-From-Equilibrium Effects on Microstructural Evolution and Mechanical Behavior of Multi-Component Metal Alloys
Mustafa Tobah, Materials Science and Engineering

Far-from-equilibrium thermal processing of metals through additive manufacturing (AM) allows for highly unique microstructures with potentially superior mechanical behavior compared to metals made through traditional manufacturing. This thesis will combine multiple AM techniques, such as laser powder bed fusion, electron-beam powder bed fusion, and laser directed energy deposition to develop a comprehensive understanding of melt-pool microstructure evolution in multi-phase alloys. The work will be focused on duplex stainless steel with the primary constituent phases of austenite and ferrite, and a novel Ni-Cr binary eutectic alloy. Additionally, microstructure characterization via electron microscopy and electron backscatter diffraction, nanomechanical testing, including in situ tensile testing under scanning electron microscopy, and calculation of phase formation using AM computational modules will be performed. In addition, novel microstructures using the unique capabilities of AM techniques will be developed to demonstrate the ability of AM to surpass the capabilities of traditional manufacturing.

AI-powered Intelligent Interactive Systems to Enhance Visual Understanding in Medical Education
Jingying Wang, Computer Science and Engineering, Barbour Scholar

Learning medical procedures (surgeries/diagnostic imaging) is a highly visual process requiring trainees to analyze anatomical structures, make decisions, and handle tools with precision [1, 2]. While real-time operating room access and instructor guidance are limited, learning tools like video [3], 3D simulations [4, 5], and visual guidance like pointers [12] hold immense potential. Existing tools often focus on procedural steps[6, 7, 8, 9] and passive visual pointing [13], neglecting mechanisms to explain complex visual tasks, motivate tool operation, and support mutual visual understanding. My dissertation addresses these gaps by: 1) Surgment [11], web-based system for creating visual questions and feedback, which enhances video-based surgical training through interactive visual learning; 2) XSonoTutor [10], a Mixed Reality system for ultrasonography training with automated textual and visual feedback to improve cognitive and physical skills; and 3) a four-thrust study exploring gaze data, behavior modeling, AI-assisted debriefing of collaborative dynamics, and real-time gaze-sharing tools.

Coordination Dynamics and Thermodynamics of Porous Coordination Polymers
Hochul Woo, Macromolecular Science and Engineering

Coordinatively unsaturated sites (CUS) within metal-organic frameworks (MOFs) are crucial for determining selectivity of this over 100,000 member family of sorbents. CUS are the strongest binding sites for guest molecules and achieving theoretical surface area of a MOF is prevented by CUS-bound solvent. The exchange kinetics in a CUS-MOF were investigated by monitoring guest exchange in situ, revealing that kinetics differ depending on whether exchange occurs within pores or at CUS. Examining solvent exchange in isostructural CUS-MOFs reveals that exchange rates in isostructural MOFs are driven by kinetic factors that do not simply stem from guest binding thermodynamics. For the first time, the thermodynamics of solvent binding to CUS was experimentally elucidated. Furthermore, the relationship between kinetic diameter and pore architecture is being examined by observing size-exclusive diffusion through MOF single crystals. This dissertation provides a fundamental understanding of guest exchange dynamics and their implications for MOF functionality.

Investigating Molecular Behaviors at Buried Interfaces in Polymer Adhesives, Vaccines, and Lithium-Ion Batteries In Situ
Yuchen Wu, Macromolecular Science and Engineering

Interfaces determine the performance of many materials and devices, from adhesives to energy storage devices and biomedical materials. Molecular-level interfacial structures and interactions govern interfacial properties, thus it is important to investigate such structures/interactions. It is difficult to do so because of the lack of tools and methods to probe buried interfaces. To overcome this challenge, this dissertation applied sum frequency generation (SFG) vibrational spectroscopy, a non-destructive technique sensitive to monolayer structures, to study buried interfaces in collaboration with BASF, Dow, Merck, and Ford. The research elucidated the mechanisms of improving polyurethane adhesion (via silane coating and plasma treatment) and silicone adhesion (via silane promoters), and the environmental effects on RTV silicone adhesives. It also explored adjuvant-protein interactions for vaccine development and studied the electrode surface changes in lithium-ion batteries. The findings highlighted the significance of interfacial chemistry and demonstrated the power of SFG in advancing material design and performance.

General Latent Embedding Approaches for Modeling High-Dimensional Hypergraphs in the Modern Data Era
Shihao Wu, Statistics

Large-scale high-dimensional hypergraph data, capturing multi-way interactions such as co-occurrence, co-functioning, and collaboration, are increasingly common in fields including electronic health record analysis, language model pre-training, and biology research in the modern data era. Their growing scale and complexity pose new challenges in the modeling and utilizing of these data. This dissertation addresses three important problems.

The first part of the dissertation develops a general latent embedding approach for non-uniform, high-dimensional sparse hypergraphs with multiplicity. This approach overcomes limitations of existing methods including their reliance on uniform hyperlink orders and inability to handle multiplicity, enabling analysis on general high-dimensional hypergraphs.

The second part proposes a generative model framework for efficiently generating high-fidelity synthetic hyperlinks from denoising diffused latent embeddings, which outperforms existing methods on this task. These synthetic hyperlinks enable medical centers to share patient profiles (encoded as hyperlinks) without compromising patient privacy.

The third part focuses on healthcare applications of hypergraph embeddings. We introduce an embedding-focused architecture for International Classification of Diseases (ICD) codes. The embeddings capture latent relationships and hierarchical structures among the ICD codes, and demonstrate strong performance in downstream tasks such as disease clustering, patient mortality prediction, and medical insurance planning.

On the Integration of Connected and Automated Vehicles: System Analysis and Control Strategies
Minghui Wu, Civil Engineering

Recent advancements in communication and vehicle technologies are revolutionizing the way people travel. Tools like navigation apps help travelers to make smarter, more informed decisions. In the coming decades, these decision-support tools are expected to grow even more powerful with the integration of driving automation, allowing connected and automated vehicle (CAV) users to make strategic, algorithm-driven choices. My dissertation examines how these advanced capabilities transform individual travel behaviors, impact traffic patterns, and offer new opportunities for traffic control that benefit society.

The first phase investigates how CAV users make strategic decisions and how these behaviors collectively affect traffic flow across a network. An online learning algorithm is proposed to guide CAV users in selecting optimal routes based on past experiences and real-time predictions. A theoretical regret bound is established for the worst-case performance guarantees, providing a foundation for modeling CAV learning behavior. As the proportion of CAVs increases, individual decisions become interdependent, influencing overall traffic dynamics. To capture this interplay, a game-theoretical framework is introduced, modeling interactions among strategic commuters making sequential decisions. A multiday user equilibrium concept is defined to describe the system’s steady-state behavior.

Building on these insights, the dissertation shifts from understanding behaviors to actively shaping them. We introduce participatory traffic control, a novel paradigm where CAVs act as control actuators to influence the behavior of other drivers, redistributing demand spatially and temporally to enhance system efficiency. Using optimal control theory and mean-field approximations, the theoretical foundation for participatory traffic control is established. Numerical experiments demonstrate that this approach can achieve up to 25 percent of the maximum theoretical efficiency gains, even with a CAV penetration rate as low as 10 percent, highlighting its potential for broad impact.

Finally, the dissertation bridges the gap between the theoretical control framework and practical deployment. We calibrate the proposed models using real-world vehicle trajectory data from General Motors. This ensures that the control strategies account for the actual human behavior, making them practical to implement. Additionally, an economic incentive model is developed to promote active participation by CAV users, ensuring fairness and cost-effectiveness in deployment.

Innovative Methods for the Degradation and Control of Disinfection Byproducts: Applications in Water Treatment and Daily Disinfection Scenarios
Yuhao Xian, Environmental Engineering

This dissertation investigates innovative approaches to minimizing the formation of carcinogenic disinfection byproducts (DBPs) during water treatment and examines the risks associated with DBP exposure during surface disinfection. The first chapter evaluates the degradation of DBPs in potable water reuse through non-thermal air plasma, offering an alternative to traditional processes that rely on chemical additives. The second chapter applies emerging advanced oxidation processes at pilot-scale, focusing on the simultaneous removal of 1,4-dioxane and DBP control, bridging laboratory-scale research with real-world challenges in municipal water treatment. The third chapter explores DBP formation during routine surface disinfection in high-use environments such as hospitals and strategies for DBP mitigation. Collectively, these chapters contribute to a deeper understanding of the health risks associated with DBPs and propose recommendations for more sustainable treatment and disinfection practices.

Scalable and Reliable Coordination in Embodied Intelligent Networks: A Submodular Optimization and Online Learning Approach
Zirui Xu, Aerospace Engineering

Embodied intelligent networks are collections of distributed autonomous agents that can sense, reason, communicate, and act. Scalable and reliable coordination among such agents can benefit society in tasks ranging from environmental monitoring to transportation to surveillance. But achieving scalability is challenging due to the agents’ limited resources versus their resource-demanding tasks, often combinatorial and NP-hard. Achieving reliability is challenging due to environmental unpredictability, limited environmental observability, and untrustworthiness of commands externally suggested by human operators or machine learning algorithms. This thesis lays the theoretical and algorithmic foundation to overcome these challenges by introducing discrete optimization and online learning capabilities that enable multi-agent networks to self-configure their communication topology to balance the trade-off of scalability versus coordination performance, adapt online to unpredictable environments, and robustly benefit from untrustworthy external commands. The provided methods are evaluated in information-gathering tasks of mapping, target tracking, and surveillance via both physics-based simulations and field experiments.

Principled Measurement, Characterization, and Circumvention of Connection Interference on the Internet
Diwen Xue, Computer Science and Engineering

The same rights that people have offline must also be protected online. Protecting the freedom of expression on the internet, however, requires understanding and countering the technical mechanisms restricting these freedoms. With advancements in networking technologies, users increasingly face disruptions, throttling, or tampering of their internet connections by adversaries like nation-state censors along the network path. These connection interference practices remain deliberately opaque and constantly evolving, challenging civil society’s ability to monitor, understand, and develop lasting circumvention solutions.

My dissertation studies these scenarios through three thrusts: measuring network filtering devices to bring accountability to both authorities deploying censorship and the technology enabling it; advancing principled design of network traffic obfuscation schemes to ensure long-term sustainability of censorship circumvention; and examining the ecosystem surrounding circumvention tools, focusing on their security and privacy efficacy for vulnerable populations.

Physics-Consistent Multiscale Modeling of All Solid State Batteries
Mingze Yao, Mechanical Engineering

Despite the achievements so far, all-solid-state batteries (ASSBs) still face various challenges including but not limited to Li-ion transportation, interfacial contact, etc. Physics-based modeling helps gain mechanistic insights to the key phenomenon in ASSBs, therefore potentially providing solutions to resolve the issues. Constrained by the physics that dominates the key phenomenon in ASSBs, on the cathode side I resolved a long-standing problem of open-circuit voltage models in literature violating the second law of thermodynamics and proposed one that is second-law-consistent. Additionally, I pointed out the limitation of effective medium theory for tortuosity modeling in ASSB cathodes and modeled cathode tortuosity using flux-based simulation. For anode-separator interface, a contact area model was developed consistent with the Mullins-Sekerka approach. Integrating these contributions, I constructed a pseudo-2-D model for ASSBs, which provided a roadmap for achieving 350 Wh/kg specific energy at practical discharge rates.

Low-Dimensional Structures of Learning and Computation in Deep Learning
Can Yaras, Electrical and Computer Engineering

Over the past decade, deep learning has formed the backbone of intelligent systems that can generate highly realistic content and even perform complex reasoning. Yet, there is little understanding behind the representations of the real world learned by these systems, as well as the process by which they learn. Moreover, training and deploying these systems in a non-trivial task, requiring immense computational resources. In this dissertation, we simultaneously tackle these challenges through the lens of low-dimensional structure. First, we uncover low-dimensional geometries formed in both high- and low-level representations of data, gaining insight into the mechanisms that drive successes and failures of intelligent systems, while achieving greater scalability through more compact representations. Second, we reveal low-dimensional properties of the learning process, which we leverage to design more robust and computationally tractable algorithms for training and deploying these systems.

Photocatalytic N2 and CO2 Reduction: Investigating Optimal Reaction Parameters and the Impact of Additives on Charge-Carrier Dynamics in TiO2
Carissa Yim, Chemical Engineering

Photocatalysis is a promising method to store solar energy in chemical bonds. In a photocatalytic process, a semiconductor absorbs photons that generate electron-hole pairs (charge-carriers) to drive reactions. However, rapid charge-carrier recombination limits high efficiency. We use in situ photoluminescence to determine how the addition of transition and alkali metals, which can increase catalytic activity at surface sites, effects charge-carrier dynamics within the semiconductor under relevant reaction conditions. Additionally, we evaluate how temperature, pressure, and illumination influence the photocatalytic conversion of CO2 and N2 to hydrogen-dense fuels, methane (CH4) and ammonia (NH3), respectively. We find that incorporating additives into the semiconductor does not compromise charge-carrier mobility and demonstrate that light accelerates the transformation of feedstocks to useful products by enhancing the rate-limiting step and mediating the distribution of surface intermediates.

Online Learning for Robot Control Under Uncertainty with Regret Guarantees
Hongyu Zhou, Aerospace Engineering

In the future, mobile robots will automate fundamental tasks such as package delivery, inspection and maintenance, and target tracking. Such tasks require accurate and efficient tracking control. But this is challenging since they require robots to operate under uncertain conditions, particularly under unknown dynamics and disturbances. For example, they require quadrotors to deliver packages of unknown weight, inspect and maintain outdoor infrastructure under wind gusts, and chase mobile targets with unpredictable motion behavior. This thesis focuses on online learning algorithms for simultaneous system identification and model predictive control. We provide algorithms that have finite-time near-optimality guarantees and asymptotically converge to the optimal (non-causal) controller. We validate our algorithms in hardware experiments and physics-based simulations, with applications to high-speed trajectory tracking of quadrotors despite wind/ground effects, payload transportation of quadrupeds, and target tracking despite unknown target motions.

Transport and Magnetic Properties of Topological Quantum Materials
Yuan Zhu, Physics, Barbour Scholar

Topological quantum materials, characterized by the non-trivial topological band structures, display a broad range of interesting phenomena, including exotic surface states and anomalous transport properties, with various potential applications in electronics. My study focuses on two types of materials: Kagome lattice compounds and topological Kondo insulators. Kagome lattice compounds exhibit a unique shared-corner triangular structure that yields flat-band, Dirac points, and van Hove singularities, generating distinct novel quantum states. Topological Kondo insulators combine strong electron correlations with topological order, showing metallic surface states and unconventional quantum oscillations arising from charge-neutral fermions. We employ a variety of experimental techniques, including thermoelectric and magnetometry measurements, to study the transport and magnetic properties of these materials.