UC Berkeley

Accelerated fuel qualification for advanced nuclear systems requires mechanistic understanding coupled with rigorous uncertainty quantification to compress traditional timelines. This dissertation investigates whether mechanistic models grounded in first-principles physics, systematically validated against experimental data across multiple operating regimes, and calibrated through Bayesian inference can substantially reduce epistemic uncertainty in fuel performance predictions while establishing defensible operational boundaries. The work applies this hypothesis to uranium mononitride (UN) fuel for micro-reactor deployment, addressing the challenge that historical UN irradiation data exhibit significant scatter and gaps at high burnup and elevated temperatures relevant to modern micro-reactor designs.A mechanistic model for fission gas release and swelling model was investigated in the BISON fuel performance code. The model explicitly captures temperature-dependent diffusion mechanisms through decomposition of effective diffusivity into thermal equilibrium, defect-assisted, and thermal components. Multi-population bubble tracking differentiates nucleation pathways, and grain boundary coalescence provides mechanistic basis for understanding saturation transitions and release. Validation against various historical irradiation test cases revealed regime-dependent model performance with robust agreement in low-temperature, low-burnup conditions and interpretable limitations at regime boundaries.The dissertation establishes that mechanistic modeling with rigorous uncertainty quantification may successfully enable compressed accelerated fuel qualification timelines. Broader methodological contributions include regime-stratified validation identifying mechanistic regime boundaries based on physics rather than arbitrary thresholds, surrogate-accelerated Bayesian inference achieving significant computational speedup while maintaining statistical rigor through hyperparameter optimization and empirical calibration inflation factors, and mechanistically guided prior specification improving parameter convergence by 30 to 50% while maintaining Bayesian validity. These approaches are generalizable to other high-dimensional inverse uncertainty quantification problems for expensive mechanistic models.The surrogate-accelerated Bayesian Markov chain Monte Carlo (MCMC), using latin hypercube sampling (LHS)-sampled input parameters to train Gaussian process emulators, achieved robust parameter within the primary regime for low temperature and low burnup. Parameter uncertainty was reduced by 80 to 90% through Bayesian inference, with posterior distributions full convergence (Gelman-Rubin R < 1.01). Forward propagation of calibrated posteriors produced predictions with 86.4% experimental coverage, 0.30 vol% mean absolute error for swelling, and 0.05% mean absolute error for fission gas release. Comparison of prior (uninformed) versus posterior predictions quantified information gain, with swelling prediction accuracy improving 78.6% and fission gas release accuracy improving 92.6%.

This dissertation explores how low-energy electronic states in molecular and coordination networks can be deliberately engineered and probed at the atomic scale. Using a combination of bottom-up synthesis, on-surface assembly, and low-temperature scanning tunneling microscopy and spectroscopy (STM/STS), I investigate how symmetry, topology, and local chemical environment control band edges, metallic channels, and symmetry-breaking instabilities in low-dimensional quantum materials.I first introduce a one-dimensional π-conjugated polymer, poly(difluorenoheptalene-ethynyl-ene) (PDFHE), in which a pseudo–Jahn–Teller (PJT) mechanism drives a collective distortion of the backbone. By combining on-surface synthesis, bond-resolved STM (BRSTM), and density functional theory (DFT), I show that the combination of non-benzenoid difluoreno-heptalene units and ethynylene linkers produces frontier states of the appropriate symmetry to undergo strong PJT vibronic coupling. This coupling destabilizes the high-symmetry structure, driving a collective symmetry-lowering distortion that splits the frontier bands and reshapes the electronic band gap.I then turn to graphene nanoribbons (GNRs) as a platform for semiconducting and metallic one-dimensional (1D) channels. For N = 8 armchair GNRs (8-AGNRs), I use solution-phase A2B2 Suzuki polymerization combined with matrix-assisted direct (MAD) transfer and on-surface cyclodehydrogenation on Au(111) to realize structurally precise ribbons with an experimentally measured band gap of 0.74 ± 0.04 eV, in the technologically relevant range between silicon and germanium. Beyond this, I develop an engineered a 5-membered-ring-decorated 7-AGNR (H2-7-iGNR) that can be toggled from a gapped semiconductor to a robust two-channel metal (7-iGNR) by STM-induced edge dehydrogenation. Wannier analysis of the metallic state maps its frontier bands onto an extended Su–Schrieffer–Heeger zigzag ladder model, and local dehydrogenation within a single ribbon yields atomically sharp metal-semiconductor junctions with nearly barrierless p-type alignment.Moving beyond purely carbon-based systems, I introduce organometallic lattices based on N-heterocyclic carbene (NHC) ligands coordinated to Au. Linear NHC–Au–NHC junctions are assembled into 1D chains and two-dimensional (2D) Kagome-type lattices on Au-based substrates. STM/STS and DFT show that C–Au–C bonding states hybridize into dispersive bands that cross the Fermi level, imparting intrinsic metallicity and exceptionally low work functions comparable to alkali metals, but realized here in structurally well-defined molecular networks.Finally, I describe ongoing work on Kagome-type metal–organic frameworks (MOFs), exemplified by monolayer Ni3(HITP)2 grown on van der Waals substrates using a metal-salt-based chemical vapor epitaxy (CVE) approach. STM imaging confirms the Kagome coordination network and monolayer character on HOPG and WSe2, while STS combined with DFT and GW calculations reveals an isolated Kagome band manifold with identifiable flat bands on both the conduction and valence sides. When grown on graphene/hBN field-effect devices, the valence-side flat band can be shifted relative to the Fermi level by electrostatic gating, establishing a gate-addressable molecular Kagome flat band on a device-compatible platform.Across these systems, common design principles emerge: exploiting symmetry and its breaking, tuning backbone topology and ring chemistry, controlling edge and coordination environments, and carefully managing substrate coupling. Together, the results demonstrate a progression from 1D polymers and nanoribbons to organometallic lattices and 2D MOFs, all viewed through the common lens of low-energy electronic states engineered by molecular design and interrogated by local spectroscopy. The methods and concepts developed here lay a foundation for actively steering low-energy states in designer molecular networks toward targeted functionalities in future quantum devices.

The integration of autonomous vehicles into transportation infrastructure presents a fundamental challenge: these systems must operate within environments designed by and for human drivers while sharing road spaces with human-operated vehicles. This dissertation addresses the central question of how autonomous vehicles can function as cooperative agents in mixed-autonomy traffic systems, where behavioral coordination between automated and human-driven vehicles determines overall system performance.The research develops a four-dimensional framework for analyzing this interaction paradigm: coexistence in shared road spaces, coordination through hierarchical control architectures, comprehension of human behavioral patterns for risk assessment, and certification protocols enabling third-party evaluation without proprietary access. Each dimension represents a distinct facet of the operational challenges facing autonomous vehicle deployment.This work progresses from theoretical foundations in cooperative control theory to practical implementations validated through field experiments. Mathematical frameworks combining partial differential equation models for aggregate traffic dynamics with stochastic processes for individual vehicle behavior provide the foundation for control strategies that leverage autonomous vehicles as mobile actuators within mixed traffic flows.The empirical validation demonstrates quantifiable improvements across all dimensions. Reinforcement learning-based controllers achieve 15% improvement in minimum traffic flux through strategic speed harmonization. Field deployment of 100 connected vehicles reveals 52% reduction in bottleneck density, while exposing complex driver engagement patterns that influence system performance. Probabilistic risk assessment using spatiotemporal occupancy heatmaps reduces conflict rates by 49% in urban scenarios compared to deterministic metrics. The vendor-agnostic evaluation framework enables statistical safety assessment through Monte Carlo simulation without requiring access to proprietary control algorithms.These findings establish that successful autonomous vehicle deployment requires addressing technical optimization, behavioral adaptation, probabilistic safety assessment, and institutional trust mechanisms simultaneously. The research contributes both theoretical frameworks and practical methodologies for designing cooperative transportation systems where autonomous vehicles enhance overall traffic performance through strategic interaction with human drivers. This holistic understanding provides foundations for the transition toward mixed-autonomy transportation networks that leverage the complementary capabilities of human and automated agents.

The suppression of Vodou was waged heavily during the U.S. occupation of Haiti, from 1915 to 1934. Amidst its prohibition, a substantial flow of Vodou drums and other sacred objects circulated within and out of the country, and ultimately into museums abroad. Bringing together anthropologies of collecting, theoretical investigations of the ways objects move, and interdisciplinary Vodou scholarship, I trace the itineraries of these materials out of Haiti, from museums and archives in the United States and in Europe, tracking how, where, and why they circulated along the way. Part I investigates the adjacent forces of U.S. military occupation and U.S-American anthropology, as well as the enduring presence of Vodou, during the early- and mid-century period of collecting. Part II brings together object biography and itinerary methods to trace the circulation of two sacred classes of objects within and out of Haiti. I argue that, following Paul Mocombe’s concept of “the Vodou ethic,” ritual objects embodied not only spiritual power but the social values of collectivism, justice, accountability, and freedom. These principles were central to sévitè, or those who ‘serve the spirits,’ and were wielded in spaces where sacred objects were used — including ritual ceremonies, combité agricultural collectives, and even on the battlefield. As ideological apparatuses, these objects were deliberately targeted by arbiters of U.S capitalism and law enforcement. This research contributes to broader theories of materiality, anthropologies of war and diplomacy, and considerations for museum- based research. It also offers new perspectives on the enduring impacts of U.S-American interventions in Haiti.

This thesis presents progress on a demonstration experiment of two-color ionization injection (TCII) in laser driven plasma accelerators (LPAs). The method has been theorized to potentially produce ultra-low, tens of nm rad, emittance beams, but has yet to be demonstrated in the laboratory. Were LPAs to reliably produce ultra-low emittance beams, they would become compact sources for a number of applications that require high brightness beams enabled by low emittance.Circularly polarized drive pulses were used for the first time to turn off ionization injection and prepare a dark-current-free wake for further injection. A regime was found in which wakes generated by circularly polarized pulses were proven to be of equivalent amplitude as those generated by linearly polarized pulses, but unlike the linearly polarized case, ionized electrons were not injected into the trapped region of the wake.High intensity, broadband, ultrashort third harmonic pulses were generated for ionization injection of electrons into a preformed wakefield accelerator for the first time. A low ponderomotive potential but high electric field injector was formed by harmonic conversion of 800 nm fundamental wavelength to 267 nm. Harmonic conversion efficiency as a function of intensity, bandwidth, and crystal thicknesses were measured. A UV pulse diagnostic was built to resolve pulse length and higher order nonlinearities.Planar laser-induced fluorescence (PLIF) was used to customize LPA targets for the first time. PLIF, combined with fluid simulations, were used to iteratively design gas density profiles for desired features. Manufactured nozzles were subsequently qualified via PLIF.Finally, methods of spatial and temporal coupling of the third harmonic injector into the preformed wake are presented and show achievement of femtosecond and micrometer precision required for controlled injection of low emittance beams.

This dissertation addresses the challenge of scalable post-training for large language models along two axes: data generation and algorithmic robustness. On the data side, we demonstrate that AI-generated preference data can achieve state-of-the-art alignment results through the Nectar dataset and Starling reward models, then advance to fully self-generated supervision via Meta-Rewarding, where models act as actor, judge, and meta-judge simultaneously. This progression---from expensive human annotation to external AI teachers to autonomous self-play---eliminates traditional data bottlenecks while maintaining or exceeding performance. The key insight is that comparative feedback (K-wise rankings, pairwise judgments) is more reliable than absolute scoring. On the algorithm side, we identify and resolve fundamental inconsistencies in standard RLHF through P3O (Pairwise Proximal Policy Optimization), which performs comparative reinforcement learning rather than optimizing absolute rewards. We formalize this through reward equivalence: reward models trained with Bradley-Terry loss are invariant to constant shifts, but PPO is not, leading to training instabilities. P3O extracts the comparative signal correctly, achieving superior KL-reward frontiers. We then extend preference optimization beyond outputs to internal reasoning via TPO (Thought Preference Optimization), demonstrating that thinking benefits general instruction following across diverse categories including creative writing, health advice, and marketing---not just mathematical tasks. Together, these contributions establish that scalable LLM post-training requires coordinated advances in both data generation and algorithmic robustness. The four contributions are tightly interconnected: Starling's K-wise ranking techniques inform Meta-Rewarding's judge design; Meta-Rewarding's self-play principle extends to TPO's thought generation; P3O formalizes the comparative feedback underlying all methods; and all four leverage iterative training with preference-based optimization. This framework enables autonomous model improvement without human annotation or algorithmic instability, pointing toward increasingly autonomous AI systems.

Isolated from the marine fungus Fusarium heterosporum, mangicols represent a new class of C25 terpenoids featuring an unprecedented spirotetracyclic core and notable cytotoxicity against human tumor cell lines. The intricate carbon framework of these molecules presents a formidable synthetic challenge that has remained unaddressed. This dissertation describes the development of a synthetic strategy toward mangicol D, during which an unexpected discrepancy in NMR data prompted a critical structural reassessment, ultimately leading to the stereochemical reassignment of the natural product.In Chapter 1, we introduce the neomangicol and mangicol sesterterpenes, covering their isolation, structural elucidation, and biosynthetic pathways, and provide an overview of previous synthetic efforts toward these natural products. Our discussion highlights the challenges associated with structural assignments, explores the possible biosynthetic relationships between neomangicol and mangicol, and examines the synthetic obstacles encountered by prior research groups.In Chapter 2, we detail the design of a divergent synthetic route toward the mangicol family and describe the development of a core-first strategy for mangicol D. Initial efforts led to an unexpected outcome, as none of the synthetic samples matched the reported NMR data for mangicol D. Through careful analysis, we identified the structural misassignments, revised the stereochemistry, and accomplished the first asymmetric synthesis of mangicol D in 16 steps.

The following dissertation details the development of methods for the construction of bonds in a selective fashion by transition metal-mediated catalysis, enzymatic catalysis, and a combination of both transition metal-mediated and enzymatic catalysts.Chapter 1 provides an overview of the recent advances and applications of merging homogeneous transition-metal mediated catalysis with proteins in the past five years. This chapter discusses the development of one-pot, two-step processes and tandem processes in which the complementary reactivity of homogeneous transition-metal catalysts and biocatalysts enables the formation of products. Furthermore, this chapter discusses the recent advances in the development of artificial metalloenzymes and the growing potential of de novo protein design for the encapsulation of homogeneous transition-metal catalysts. This chapter is subdivided into sections highlighting the reactivity of the first-third row transition metals.Chapter 2 describes the study and development of a new class of catalysts for the borylation of alkyl C–H bonds. New catalysts for the undirected borylation of alkyl C–H bonds can lead to new strategies for the synthesis or derivatization of organic molecules. Catalysts composed of phenanthroline ligands, in combination with iridium precursors, have led to the borylation of aryl and some alkyl C–H bonds. However, catalysts that are more active or that react with distinct selectivity toward alkyl C–H bonds, or both, are needed. A new class of catalyst ligated by N-heterocyclic carbenes were predicted by computation to be more active for the borylation of alkyl C–H bonds than those with phenanthroline ligands. We show that these ligands generate catalysts that react in ways that are different from the pathways predicted. Specifically, we show that they rearrange to LX-type ligands by cyclometallation and then catalyze the borylation of THF at the position α-to oxygen in the presence of alkoxide and that they do not react with alkyl C–H bonds in the absence of alkoxide. From computational and experimental studies, we provide strong evidence that the reactions in the presence of alkoxide occur by a catalytic cycle involving an anionic [iridium(III)trisboryl]–Na+ complex that gives rise to the unique selectivity for the borylation of this saturated oxygen heterocycle and substantiate this conclusion by showing that the borylation of arylpyridines occurs by distinct pathways in the presence and absence of alkoxide.Chapter 3 describes the development of a sequential chemoenzymatic process to achieve the selective synthesis of linear primary amines. Methods that form amines from feedstock chemicals are an important component of industrial chemistry. Many methods for the synthesis of amines yield mixtures of primary, secondary, and tertiary amines, necessitating costly downstream separations. We present a chemoenzymatic approach to hydroaminomethylation that addresses these challenges by combining hydroformylation catalysed by a phosphine-bound rhodium complex with enzymatic transamination catalysed by an ω-transaminase from Vibrio fluvialis (VfTA). This sequential chemoenzymatic hydroaminomethylation reaction converts olefins to linear primary amines with high regioselectivity and chemoselectivity for the linear primary amine, and we demonstrate that this process occurs with series of olefins with varying structure. Furthermore, we report progress towards the incorporation of a biocatalytic cascade for recycling an intermediate amine donor, which is required for the transaminase enzyme to use an ammonium salt as the terminal nitrogen source. This study illustrates the potential to combine chemo- and biocatalytic reactions to produce valuable materials from readily available feedstocks in a one-pot, two-step sequential process, with selectivities that have not been achieved by transition-metal catalysts.Chapter 4 builds on the work of Chapter 3 and describes the development of a tandem chemoenzymatic process for the synthesis of linear primary amines from feedstock chemicals and under mild conditions. In this chapter, we report a tandem chemoenzymatic hydroaminomethylation strategy that converts terminal olefins into linear primary amines under mild conditions by combining a Rh/DPPon catalyst for hydroformylation with a wild‐type ω‐transaminase from Vibrio fluvialis (WT-VfTA) and an amine donor recycling system composed of an alanine dehydrogenase (WT-AlaDH) and a glucose dehydrogenase (WT-GDH), both from Bacillus subtilis. The inputs for this tandem hydroaminomethylation system are inexpensive feedstock reagents, including ammonium salts, alanine, carbon monoxide, hydrogen, and glucose and high selectivity for primary amines without the formation of secondary and tertiary amine byproducts was achieved. Moreover, the versatility of this system is demonstrated by the synthesis of 15N‐labeled primary amines.Chapter 5 describes the development of a substrate masquerade strategy to achieve the site-, stereo-, and chemoselective chlorination of alkyl C–H bonds with a non-heme halogenase. Enzymatic halogenation of C–H bonds is a promising approach to synthesize chlorine-containing compounds. However, few halogenases chlorinate C(sp3)–H bonds of molecules lacking a carrier protein, and only a small subset accommodate non-native substrates. Competitive oxygenation of non-native substrates makes halogenation of such substrates a challenge to achieve. Herein, we report a strategy for the halogenation of unnatural substrates by which an anchoring group leads them to masquerade as the native substrate. By this approach, a series of terpenoids connected to an indole moiety, undergo enzymatic halogenation catalyzed by WelO5*, a non-heme, a-ketoglutarate-dependent halogenase. We generated WelO5* variants that catalyze the chlorination of C(sp3)–H bonds in a series of non-native substrates with high selectivity for chlorination over oxygenation and with excellent stereoselectivity and variants that catalyze bromination and azidation. Studies that vary the anchoring group showed that a series of heteroaromatic and aromatic groups can enhance reactivity and can influence the degree of chlorination of the anchored menthol substrate. Cleavage of the ester tethering the indole anchoring group to the terpenoid gives the free halogenated compound.

The interiors of Uranus, Neptune, and ice giant–like exoplanets remain poorly constrained, particularly regarding the high-pressure chemical processes that govern their structure, evolution, and energy balance. To address this gap, we conducted high-pressure experiments on a range “Synthetic Uranus” (water, ammonium hydroxide, and isopropanol) compositions formulated to approximate cosmic carbon, hydrogen, oxygen, and nitrogen ratios relevant to ice-giant mantles. In particular, we studied carbon-poor compositions with <10% carbon and correspondingly larger abundances of oxygen or nitrogen.Using laser-heated diamond anvil cells with in situ X-ray diffraction and Raman spectroscopy, we observed diamond precipitation between 14 and 55 GPa at temperatures as low as 1,500 K, and for compositions with as little as ~2% carbon. These pressures, temperatures, and carbon abundances are significantly lower than reported for simpler hydrocarbon mixtures. Our results indicate that oxygen- and nitrogen-bearing species can reduce the barrier for carbon dissociation, enabling diamond formation at comparatively moderate conditions. The presence of these volatiles therefore suggests that “diamond hail” may occur at shallower depths and lower C concentrations than previously estimated, leading to consequences for internal heat transport, stratification, convective stability, and magnetic field generation. More broadly, these findings provide new constraints on the chemical pathways expected in water–ammonia–methanol–rich planetary interiors and contribute to ongoing efforts to refine evolution models for Uranus, Neptune, and analogous exoplanets.As a secondary study, we investigated cyclohexane under dynamic compression to examine how molecular structure and hydrogen content influence hydrocarbon dissociation, carbon clustering, and metallization pathways. Cyclohexane, while not a direct compositional analog to Synthetic Uranus, serves as a clean model for isolating fundamental high-pressure reaction mechanisms that are otherwise obscured in multicomponent mixtures. We see evidence of changes in molecular structure and possible dissociation as expressed in the measured equation of state. Together, the static and dynamic results provide a more comprehensive picture of hydrocarbon and mixed-volatile chemistry at planetary interior pressures and temperatures, informing models of material behavior across a diverse range of icy and carbon-rich worlds.

This dissertation focuses on the design, synthesis and application of fluorogenic enzymatically cleavable fluorophores for biological imaging in genetically targetable cells.Chapter 1 provides an overview of chemigenetic approaches for studying membrane potential in electrically excitable cells. The mechanisms of action, along with the spectral and physical properties of catalytically cleavable and covalently linked chemigenetic indicators, are discussed.Chapter 2 examines improved esterase-cleavable chemigenetic voltage-sensitive fluorophores (VFs) for targeted imaging in cells. Introducing a benzyl linker between the xanthene alcohol of fluorescein and the ester caging group increases in vivo enzyme activity substantially. In cellular systems, switching from mono-sulfonated to di-sulfonated caged VFs had a greater effect on enzyme activity than the benzyl linker.Chapter 3 explores additional enzymatically cleavable chemigenetic indicators for targeted voltage imaging. VFs for β-galactosidase, cellulase, and protease were successfully synthesized. Their physical and spectral properties were characterized.Appendix I presents additional caged fluorophores for assessing enzyme cleavage in mitochondria.Appendix II provides supplementary information on the loading conditions for di-sulfonated VFs in both HEK293T cells and primary hippocampal neurons. Appendix III provides a protocol for imaging catalytically cleavable chemigenetic VFs in mouse brain slice.