During pancreatic development, insulin-producing beta cells, glucagon-producing alpha cells, and somatostatin-producing delta cells differentiate from a common endocrine progenitor. While these cells have been considered terminally differentiated, over the last several years it has become evident that pancreatic islet cellular identity is more plastic than previously appreciated. A cell?? ??an?c?i???me is the final outcome that stems from the interactions of both genomic and epigenomic components. Dynamic changes in chromatin accessibility alongside the recruitment of histone marks that either facilitate or repress gene transcription in tandem with transcription factor co-localization come together to define and maintain cellular identity through gene expression. Understanding of each of these components sheds further light on understanding cellular identity, with profiling the transcriptome providing signature patterns that are responsible for the functional behavior of these cells. Building on this through the interrogation of chromatin accessibility similarities and differences provides further insight into the mechanisms that govern cellular identity when coupled with the appropriate histone modification and transcription factor data. Ultimately, genomic and epigenomic characterizations of these major islet endocrine cell types can provide an avenue to further understanding the complex regulatory mechanisms that define and maintain cellular identity. Furthermore, these characterizations provide a resource for answering outstanding questions in the field.

Spontaneity is a pervasive feature of our mental lives. Many of our thoughts and actions seem to unfold in the absence of explicit, cognitive control. This has led some to argue that spontaneous cognitive processes are always experienced passively: spontaneous cognition is the sort of thinking that seems to just happen to us. That is, it doesn’t usually appear as if we are in control of the dreams we have at night, the drifting of our attention as our minds wander, the ruminations we have about our concerns, or the flashes of creative insight we might have as we go about our life projects. When we experience these sorts of cognitive processes, they often seem to happen to us in the absence of any deliberation or effort. Instead of willfully generating the thoughts that inhabit our minds in such cases, these thoughts seem to spontaneously strike us.This dissertation argues that spontaneity in cognition and action does not necessarily imply passivity. Although spontaneous thoughts and actions may be experienced passively in many cases, I think there is also a sense in which we can let spontaneous thoughts and actions unfold. For example, one can intentionally let one’s mind wander during an inherently boring situation such as sitting in standstill traffic. Similarly, one may intentionally engage in an episode of reflective rumination, repetitively engaging the same ideas to solve a problem. In the realm of spontaneous skillful action, expert improvisers often claim they perform best when they intentionally let the actions flow through them. And in habitual behavior, such as habitual smartphone use, one might intentionally let their absent-minded scrolling behavior continue. In all cases, I think intentionally letting spontaneous processes unfold qualifies as a sort of mental action, in contrast to those who think that spontaneity implies passivity. At first glance, this may seem like a contradiction: intentional spontaneity seems oxymoronic. This dissertation will argue that this is not the case. I think one can intentionally engage in spontaneous thinking and acting via metacontrol, a form of metacognition that is directed at other control processes. In this dissertation, I will argue metacontrol underlies individuals’ ability to intentionally engage in spontaneous cognition and action. In particular, I characterize metacontrol in the context of intentional mind-wandering, rumination, lucid dreaming, improvisation, and doomscrolling to illustrate how metacontrol influences a range of both adaptative and maladaptive patterns of thinking and behavior.

Green chemistry encompasses a broad range of topics that enable us to understand and address major environmental issues. This dissertation advances multiple green chemistry topics through three independent projects: 1) Development of boron cluster-based cathode materials for use in developing a new approach to lithium-ion based energy storage. 2) Application of a quantitative probing technique for kinetic isotope effects, using the dual-clumped isotope proxy, to evaluate dissolved inorganic carbon transformation at geological sites being considered for use as CO2 storage sinks. 3) Advancing dual clumped isotopes as a next-generation probing technique to determine if terrestrial cave carbonate samples can be used to study climate through palaeothermometry and study kinetic isotope effects in speleothems.

Novel boron-cluster based molecules were synthesized for use as cathodes in lithium-ion batteries. The B12(OC7H6COOH)12 cluster exhibited the greatest potential for use as a cathode material in lithium-ion batteries. This novel molecule served as a scaffold for post-modification to enhance cyclability in the solid state. B12(OC7H6COOLi)12, a derivative of B12(OC7H6COOH)12, was deemed compatible with cell design and reversibly cycled for one redox event at an E1/2 of 3.4 V vs Li/Li+ in an in situ swagelock cell setup. However, this cluster suffered from specific capacity issues associated with only one (out of 2) reversible redox event and a high molecular weight relative to traditional metal oxides used in current generation cells.

CaCO3 samples recovered from The Cedars, peridotite-associated springs located in Northern California, were analyzed using the novel dual clumped isotope proxy to evaluate CO2 transformation processes and kinetic isotope effects and biases. Surface floe samples begin to precipitate out of solution within the first hour of CO2 absorption, and the DIC pool requires a residence time of approximately 100 hours to achieve isotopic equilibria. The D47/D48 slope of samples from the Cedars (-2.803�0.500; 1.s.e.) agrees with published values from lab experiments designed to constrain CO2 hydrolysis-related kinetic fractionation (-2.911�0.200). Ancient travertine samples that are close to equilibrium in the D47-D48 space yield an average temperature of 8.3�1.1 �C, suggesting that the conditions during that time were approximately 10�C colder than present day temperature. The consistency of slopes in the multi-isotope space suggests the D47-D48 dual carbonate clumped isotope framework can be applied to study CO2-absorption processes in applied systems including sites of interest for geological sequestration.

CaCO3 samples from terrestrial cave sites at 13 localities from around the world were analyzed using dual clumped isotope (D47-D48) system. Data showed that 27 samples exhibited isotopic equilibrium with the majority having D47-T predicted temperatures matching measured temperatures. 17 samples exhibited isotopic disequilibria in the D47-D48 space. Of the latter suite of samples, 15 fell within quadrant 3, 1 within quadrant 2, and 1 within quadrant 1 in the DD47-DD48 space. Samples that fell within quadrant 2 exhibited traditional degassing disequilibria consistent with previous speleothem studies. Samples which fell within quadrants 1 and 3 were identified to have oxygen and carbon isotope exchange within the Rayleigh distillation model approach, however the specific type of exchange could only be resolved in the DD47-DD48 space. Samples within quadrant 1 were identified to have oxygen isotope exchange while samples within quadrant 3 had carbon isotope exchange. Cave raft samples collected from Mexico exhibited the largest carbon isotope exchange phenomenon with magnitudes of disequilibria into quadrant 3 correlating to the relative ventilation from caves that they were recovered from. This study suggests the D47-D48 dual carbonate clumped isotope framework can be applied to varying types of cave carbonate samples enabling quick identification of (dis)equilibria for use in paleoclimate reconstruction applications.

Online learning and convex optimization algorithms have become essential tools for solving problems in modern machine learning, statistics and engineering. Many algorithms for online learning and convex optimization can be interpreted as a discretization of a continuous time process, and studying the continuous time dynamics offers many advantages: the analysis is often simpler and more elegant in continuous time, it provides insights and leads to new interpretations of the discrete process, and streamlines the design of new algorithms, obtained by deriving the dynamics in continuous time, then discretizing. In this thesis, we apply this paradigm to two problems: the study of decision dynamics for online learning in games, and the design and analysis of accelerated methods for convex optimization.

In the first part of the thesis, we study online learning dynamics for a class of games called non-atomic convex potential games, which are used for example to model congestion in transportation and communication networks. We make a connection between the discrete Hedge algorithm for online learning, and an ODE on the simplex, known as the replicator dynamics. We study the asymptotic properties of the ODE, then by discretizing the ODE and using results from stochastic approximation theory, we derive a new class of online learning algorithms with asymptotic convergence guarantees. We further give a more refined analysis of these dynamics and their convergence rates. Then, using the Hedge algorithm as a model of decision dynamics, we pose and study two related problems: the problem of estimating the learning rates of the Hedge algorithm given observations on its sequence of decisions, and the problem of optimal control under Hedge dynamics.

In the second part, we study first-order accelerated dynamics for constrained convex optimization. We develop a method to design an ODE for the problem using an inverse Lyapunov argument: we start from an energy function that encodes the constraints of the problem and the desired convergence rate, then design an ODE tailored to that energy function. Then, by carefully discretizing the ODE, we obtain a family of accelerated algorithms with optimal rate of convergence. This results in a unified framework to derive and analyze most known first-order methods, from gradient descent and mirror descent to their accelerated versions. We give different interpretations of the ODE, inspired from physics and statistics. In particular, we give an averaging interpretation of accelerated dynamics, and derive simple sufficient conditions on the averaging scheme to guarantee a given rate of convergence. We also develop an adaptive averaging heuristic that empirically speeds up the convergence, and in many cases performs significantly better than popular heuristics such as restarting.