Optimizing the design and performance of photonic systems has been an active and growing area of research over the past decade with many practical applications such as image sensors, augmented reality and virtual reality, on-chip photonic systems, and more. Moreover, gradient-based methods, such as the adjoint variable method (AVM), have led to very distinctive and complex designs for on-chip multiplexers, tapers, dielectric laser accelerators, and more, while yielding much higher performance metrics which classical design approaches using first principles physics cannot match. However, less research has been dedicated in the photonics community to understanding and improving the underlying numerical methods which are critical for the success of these applications. In this thesis, we will demonstrate four key numerical advancements in the use of gradient-based design methods using frequency domain numerical solvers of Maxwell's equations, particularly finite difference frequency domain (FDFD) solvers. The first is the application of domain decomposition techniques to gradient-based optimization, allowing us to reduce the effective system size for a gain in efficiency. The second exploits the physics of perturbative series expansions to efficiently determine the optimal learning rate essential to gradient-based optimization. The third leverages the fundamental similarities of the previous two methods, allowing us to combine the two to achieve a further multiplicative acceleration. The fourth is exploiting the choice of boundary condition in the context of perfectly matched layers to minimize overhead and optimize the efficiency of the simulations required during the optimization procedure. Furthermore, we will demonstrate one novel practical application in designing a next generation replacement for traditional filter-based image sensors that we term a 'color router'. By using a gradient-based approach, we demonstrate not only can we overcome the traditional limitations of filter-based approaches, but we can approach the absolute physical limit of color separation efficiency. In the context of this problem as well, we also demonstrate one further novel method to accelerate optimization, using an L1-like penalty method inspired by L1-regularization popularized in machine learning to improve the robustness to manufacturing errors and other perturbations to the device design. Finally, as a contrast to the gradient-based technique of optimization, we also showcase two examples of more traditional device optimization using the theoretical principles of Maxwell's equations. The first is to exploit analytic continuation and the band-structure of insulator-metal-insulator waveguides to design a reflector with superior reflection properties to that of a uniform metal but with lower loss (essentially a nearly metal-less metallic metamaterial). The second is to engineer interesting radiative properties and extraordinarily high reflection in atomically-thin monolayer graphene nano-ribbon system

Unprecedented volumes of data from upcoming sky surveys will yield precise constraints on parameters governing the evolution history of the Universe. One that has received particular attention over the past decade is the Hubble constant (H0) describing the expansion rate of the Universe. This thesis focuses on measuring H0 from an astrophysical phenomenon called strong gravitational lensing. The Vera Rubin Observatory's Legacy Survey of Space and Time (LSST) will increase the sample size of strong lenses from ~100 to ~100,000. This creates an opportunity to obtain the most precise measurement of H0 to date. Fully realizing the potential of LSST data entails rapidly extracting cosmological information from the images, tables, and time series associated with these lenses. My research has focused on developing analysis techniques using Bayesian deep learning, which combines the efficiency of deep learning with principled uncertainty quantification. The techniques promise to automate the analysis of tens of thousands of strong lensing systems in a robust manner. They constitute core methodology that can combine information from all the LSST lenses -- with varying types and signal-to-noise ratios -- into a large-scale hierarchical inference of H0

At That Time (2021) is a twenty-minute audiovisual chamber opera composed for the Decoder Ensemble, directed by Heinrich Horwitz, and premiered at the Music Biennale Zagreb in July 2021. I will discuss my artistic priorities and working process as they appear in this piece and connect to my recent research and composition

Objective assessment of cancer imaging studies remains a gap in patient care. Although lesion assessment is a critical part of assessing cancer treatment response, in current clinical practice it is mostly done qualitatively through free-text reports. Quantitative lesion assessment is usually only done in clinical trials, representing a small fraction of all cancer cases. To address this need, this thesis presents an automated pipeline for longitudinal assessment of cancer lesions. Given an annotated baseline exam, the system automatically detects, segments and measures the corresponding target lesions in one or more follow-up exams. It also learns from prior time points by training on partial lesion annotations made by physicians. The core components of this system are neural networks for organ and lesion segmentation, image registration from 3D SIFT-like keypoints, and a method for combining all of these into an end-to-end system that learns from prior time points. This thesis presents several contributions to each of these technical areas. For organ segmentation, we present a new dataset labeling six organs in a variety of CT scans. Despite the relative ease of locating organs in the human body, automated organ segmentation has been hindered by the scarcity of labeled training data. Due to the tedium of labeling organ boundaries, most datasets are limited to either a small number of cases or a single organ. Furthermore, many are restricted to specific imaging conditions unrepresentative of clinical practice. To address this need, we developed a diverse dataset of 140 CT scans containing six organ classes: liver, lungs, bladder, kidney, bones and brain. For the lungs and bones, we expedited annotation using unsupervised morphological segmentation algorithms, which were accelerated by 3D Fourier transforms. We trained a deep neural network which requires only 4.3 s to simultaneously segment all the organs in a CT exam. We also show how to efficiently augment the data to improve model generalization, providing a GPU library for doing so. The dataset and code are available through The Cancer Imaging Archive. For lesion segmentation, we show how the same neural network architecture can be generalized to segment lesions using public datasets with CT scans of two different organs: the liver and the lungs. For our ultimate goal of longitudinal segmentation, we also show how to train these models on partially labeled CT scans from prior timepoints in the clinical record. For image registration, we present a method based on 3D scale- and rotation-invariant keypoints. The method extends the Scale Invariant Feature Transform (SIFT) to arbitrary dimensions by making key modifications to orientation assignment and gradient histograms. Rotation invariance is proven mathematically. Additional modifications are made to extrema detection and keypoint matching based on the demands of image registration. Our experiments suggest that the choice of neighborhood in discrete extrema detection has a strong impact on image registration accuracy. In head MR images, the brain is registered to a labeled atlas with an average Dice coefficient of 92%, outperforming registration from mutual information as well as an existing 3D SIFT implementation. In abdominal CT images, the spine is registered with an average error of 4.82 mm. Furthermore, keypoints are matched with high precision in simulated head MR images exhibiting lesions from multiple sclerosis. These results were achieved using only affine transforms, and with no change in parameters across a wide variety of medical images. This work is freely available as a cross-platform software library. Finally, we combine all these technologies into and end-to-end system for tracking cancer lesions over time in CT images. We tested this system on a retrospective cohort of clinical trial lesion annotations which was automatically curated from our institutional record. Our retrospective study involved 407 patients (51% male and 49% female) with exam dates ranging from 05/02/2006 -- 09/30/2020. Given an annotated baseline exam, we used an automatic pipeline to detect, segment and measure lesions in subsequent exams. We compared our lesion measurements to those made in clinical trials, reporting statistics on lesion detection and measurement errors, as well as the improvement from training on prior exams. When tracking a labeled a baseline lesion in a follow-up exam, our system detected the correct follow-up lesion for 55.5% of liver and 49.6% of lung annotations. In other cases, the system either identified a different lesion (37.9% liver, 49.6% lung), or found no corresponding lesions in the follow-up (6.6% liver, 0.8% lung). The median measurement error for correctly-detected lesions was 3.47 mm (95% CI: [2.22, 5.67]) for liver and 1.36 mm (95% CI: [0.97, 2.03]) for lung. This resulted in 83.9% of exams receiving a correct tumor response grade for liver and 90.4% for lung, using the Response Evaluation Criteria in Solid Tumors (RECIST) guidelines. Training on prior exams improved both detection (p = .000024) and segmentation (p = .018) in the liver, while degrading both detection (p = .00019) and segmentation (p = .018) in the lung. Overall, our system automates detection and measurement of roughly half of all target lesions across multiple organs. As the main contribution is heavily dependent on deep learning, this thesis also contains a pair of appendices presenting original results on neural network theory. The first discusses convexity properties of the training loss function, and the second discusses the effect of model architecture on neuron death, studied through random initialization of model parameters. The appendix proves some local convexity properties of the training loss and inference functions, as well as upper and lower bounds on the probability that a ReLU network is initialized to a trainable point, as a function of model hyperparameters. It also contains some proposals for initialization and local optimization strategies based on these results

The focusing of a seismic image is directly linked to the accuracy of the migration velocity model. Therefore, a critical step in a seismic imaging workflow is to perform a focusing analysis on a seismic image to determine errors within the migration velocity model. While the residual curvature information provided within the offset/angle axis is commonly used to assess velocity errors, the focusing within the physical (i.e., midpoint) axes of a seismic image is often neglected. This is due to the highly interpretative nature of the focusing of geological features within seismic images, therefore making focusing analysis within the physical space difficult to automate. In this dissertation, I present an automatic data-driven approach that makes use of a convolutional neural network (CNN) equipped with high-dimensional convolutional operators to automate seismic image-focusing analysis. Training the CNN on focused and unfocused geological faults within synthetic and real prestack seismic images, I demonstrate this approach can make use of both spatial and offset/angle focusing information to robustly estimate velocity errors within seismic images. To test effectiveness of the approach, I apply it to an unfocused 2D limited-aperture image from the Gulf of Mexico and an unfocused 3D image from the Dutch Sector of the North Sea and show that it can robustly estimate velocity errors in the presence of limited illumination and coherent noise. For both applications, I show that performing a refocusing correction using the estimated velocity errors leads to an improved focusing in the unfocused images and consequently, improves the automatic detection of the faults within the images

In epithelial cancers, cells must invade through basement membranes (BMs) to metastasize. The BM, a thin layer of extracellular matrix (ECM) underlying epithelial and endothelial tissues, is primarily composed of laminin and collagen IV and serves as a structural barrier to cancer cell invasion, intravasation, and extravasation. Bm invasion has been thought to require protease degradation since cells, which are typically on the order of 10um in size, are too large to squeeze through the nanometer-scale pores of the BM. However, recent studies point toward a more complex picture, with physical forces generated by cancer cells facilitating protease-independent BM invasion. In this dissertation, I present work studying the impact of ECM mechanics on cancer cell invasion of the BM during breast cancer progression. Specifically, I will present work investigating the impact of ECM stiffness on the ability of cancer cells to form invadopodia protrusions in three-dimensional (3D) BM-like matrices. We show that increased stiffness of BM-like matrices diminishes invadopodia protrusion formation and cancer cell migration, contrasting the results from traditional 2D culture studies. Additionally, I will discuss my work on the development of a 3D in vitro model of collective invasion through endogenously produced BM. We find that cells utilize both proteases and forces to invade through the BM collectively. We also show that cells rely on volume expansion and contractility to exert pulling forces on the BM during invasion. These results reveal new mechanisms underlying breast cancer invasion and may lead to new strategies to inhibiting BM invasion. Ultimately, a comprehensive understanding of the diverse modes of BM invasion may yield new strategies for blocking cancer progression and metastasis

The act of prediction is ubiquitous in health care--be it via clinician gestalt, or through algorithms. Increasingly, predictive algorithms are beginning to displace clinician decision-making and to drive interventions of all types. Underscoring their importance, just one category of clinical predictive algorithms alone--those used for case management--is used to drive care for over 200 million individuals in the U.S. each year. However, evidence remains scant that algorithm deployments for this and other use cases actually improve clinical outcomes. This evidence gap is further compounded by the particular impracticality of conducting randomized controlled trials to study these algorithms' impacts, and by the lack of alternative design approaches. Moreover, there also exists a currently-unmet need for approaches to iteration on preexisting algorithmic deployments, so that an intervention can be re-prioritized to the patients most likely to benefit. This thesis attempts to fill this methodological vacuum using established tools from causal inference, with the problem of hospital readmission prediction and prevention as a test case

In 2009, the Linac Coherent Light Source (LCLS) generated Angstrom wavelength x-ray pulses with femtosecond pulse duration. This marked the advent of the first Angstrom wavelength x-ray laser in history. The ultrafast, ultrabright and highly coherent x-ray pulses opened up new opportunities to catch a glimpse of the bizarre and fascinating world of atom dynamics. To explore these dynamics at the natural spatial and temporal scale of atoms and molecules, new x-ray optics systems are required to precisely manipulate the x-ray laser pulse. This is the focus of my research. In this thesis, I summarize my effort of developing an efficient numerical simulation platform for the time-dependent behavior of Bragg crystal optics, and the design and experimental demonstration of a dispersion-free ultra-stable optical system to split, delay and recombine hard x-rays in a manner that will enable a new generation of high sensitivity ultrafast measurements. In the end, I present a future-oriented conceptual study that applies these techniques to the well-known chirped pulse amplification (CPA) process that has revolutionized high power optical lasers and show how this could be achieved at Angstrom wavelengths to deliver over a terawatt of x-ray power

High brightness electron beams have revolutionized accelerator science and enabled a large range of high impact applications such as X-ray free electron lasers(XFEL), inverse Compton-scattering, plasma wakefield accelerators and future linear colliders. In pursue of higher beam brightness, collective effects become key limiting factors. Those collective effects happen during the beam generation, acceleration, compression and transport, including coherent synchrotron radiation, space charge and resistive wall wakefields. As people are entering the new parameter regime — such as the generation of ultra-short electron bunches or ultra-high peak current — collective effects can play a dominant role and classical theories of wakefield can break down. In this work we describe our efforts in understanding and controlling the collective dynamics of high-brightness electron beams. In the first half of the thesis, we focus on one of the most challenging beam dynamics problems in accelerator physics — coherent synchrotron radiation (CSR) and introduce a novel computational model of 2D/3D CSR in relativistic beams. The model achieves both accurate and efficient simulations of CSR in the regime that conventional 1D models become invalid, which can be crucial for the design of next generation accelerators. In the second half, we switch gear to techniques that enable better control of the collective effects. Specifically we examine laser shaping techniques both in spatial and temporal domains as a powerful beam manipulation tool. We first discuss in detail the numerical and experimental studies using Laguerre-Gaussian mode laser in the laser heater to improve the microbunching instability suppression in XFEL drivers. Then we introduce temporal laser shaping to achieve flexible longitudinal phase space manipulation and apply it as an active out-coupling method for cavity-based free-electron lasers

Big data and pervasive digitization have transformed both the risks corporations face and the methods used to insure against them. What does this mean for the existence and efficiency of markets for risk-sharing and speculation? In my dissertation, Data, Digital Risks, and Financial Markets, I model algorithmic approaches, data dimensionality, and novel hazards to modernize our understanding of insurability, information aggregation, and optimal regulation in financial markets. The dissertation consists of three chapters. The first chapter (joint work with Yilin (David) Yang) studies the equilibrium coexistence, in financial markets, of traders using high-dimensional datasets and algorithmic strategies. The second chapter describes the extent to which insurance in markets for new or nonstationary risks exists and when government data policy can help. The third chapter argues how insurance availability might actually worsen ransomware, a pressing new cyberrisk, and presents sublimit caps as a practical regulatory measure

Numerical simulations are an important tool for prediction of turbulent flows. Today, most simulations in real-world applications are Reynolds-averaged Navier-Stokes (RANS) simulations, which average the governing equations to solve for the mean flow quantities. RANS simulations require modeling of an unknown quantity, the Reynolds stress tensor, using turbulence models. These models are limited in their accuracy for many complex flows, such as those involving strong stream-line curvature or adverse pressure gradients, making RANS predictions less reliable for design decisions. For RANS predictions to be useful in engineering design practice, it is therefore important to quantify the uncertainty in the predictions. More specifically, in this dissertation the focus is on quantifying the model-form uncertainty associated with the turbulence model. A data-free eigenperturbation framework introduced in the past few years, allows to make quantitative uncertainty estimates for all quantities of interest. It relies on a linear mapping from the eigenvalues of the Reynolds stress into the barycentric domain. In this framework, perturbations are added to the eigenvalues in that barycentric domain by perturbing them towards limiting states of 1 component, 2 component, and 3 component turbulence. Eigenvectors are permuted to find the extreme states of the turbulence kinetic energy production term. These eigenperturbations allow to explore a range of shapes and alignments of the Reynolds stress tensor within constraints of physical realizability of the resulting Reynolds stresses. However, this framework is limited by the introduction of a uniform amount of perturbation throughout the domain and by the need to specify a parameter governing the amount of perturbation. Data-driven eigenvalue perturbations are therefore introduced in this work to address those limitations. They are built on the eigenperturbation framework, but use a data-driven approach to determine how much perturbation to impose locally at every cell. The target amount of perturbation is the expected distance between the RANS prediction and the true solution in the barycentric domain. A general set of features is introduced, computed from the RANS mean flow quantities. The periodic flow over a wavy wall (for which also a detailed high-fidelity simulation dataset is available) serves as training case. A random forest machine learning model is trained to predict the target distance from the features. A hyperparameter study is carried out to find the most appropriate hyperparameters for the random forest. Random forest feature importance estimates confirm general expectations from physical intuition. The framework is applied to two test cases, the flow over a backward-facing step and the flow in an asymmetric diffuser. Both test cases and the training case exhibit a flow separation where the cross sectional area increases. The distribution of key features is studied for these cases and compared against the one from the training case. It is found that the random forest is not extrapolating. The results on the two test cases show uncertainty estimates that are characteristic of the true error in the predictions and give more representative bounds than the data-free framework does. The sets of eigenvectors from the RANS prediction and the true solution can be connected through a rotation. The idea of data-driven eigenvector rotations as a data-driven extension to the eigenvectors is studied. However, continuousness of the prediction targets is not generally achievable because of the ambiguity of the eigenvector direction. The lack of smoothness prevents the machine learning models from learning the relationship between the features and the targets, making data-driven eigenvector rotations in the discussed setup not practical. The last chapter of this dissertation introduces a data-driven baseline simulation, which corresponds to the expected value in the data-driven eigenvalue perturbation framework. The Reynolds stress is a weighted sum of the Reynolds stresses from the extreme states. A random classification forest trained to predict which extreme state is closest to the true Reynolds stress is used to compute these weights. It does so by giving a probabilistic meaning to the raw predictions of the constituent decision trees. On the test cases, the data-driven baseline predictions are similar but not equal to the data-free baseline. They complement the uncertainty estimates from the data-driven eigenvalue perturbations

Forces from small earthquakes, moving cars, or even footsteps during a short period can cause deformation of the earth's crust to varying degrees, but these actions do not cause large-scale permanent damage to the crust. In contrast, forces from geological processes over millions of years cause rocks to bend and break permanently. The work presented in this dissertation presents two geomechanical modeling techniques, finite element method, and machine learning, to improve how we model the second kind of deformation: large-scale deformation. Specifically, the proposed modeling techniques enhance our understanding of the evolution of pore pressure and stress (part one of the dissertation) and the kinematic efficiency of the strike-slip fault system (part two of the dissertation). In terms of the evolution of pore pressure, predictions based on basin and petroleum system modeling (BPSM) still have limitations, especially those models with underlying assumptions that exclude inelasticity and non-vertical deformation. Hence, we offer an alternative modeling approach by incorporating a fully-coupled hydromechanical simulator that includes inelasticity, enabling us to track the dynamic properties of rock over time. Regarding the strike-slip fault system, the current understanding of off-fault deformation behavior is limited because quantifying individual control (i.e., fault geometry, roughness, and connectivity) neglects the effects of the inter-relationship of these three controls on fault behaviors. Hence, we offer an alternative solution by harnessing a machine learning algorithm that can relate all relevant parameters in higher dimensions to estimate off-fault deformation. The first part of this dissertation explains how this work improves predictions of the evolutionary pore pressure and stress. It consists of two chapters and addresses the following research objectives: (1) integrating advanced geomechanical concepts into basin and petroleum system modeling by incorporating more realistic assumptions, e.g., inelastic constitutive relation and non-vertical stress; (2) constructing a model that captures evolving properties of shale rocks as a function of stress and pore pressure for different levels of model complexity, such as fracturing, geometry, and boundary conditions; and (3) investigating stress, pore pressure generation, and pore pressure dissipation in a tectonically complex region by using a nonlinear stress-induced upscaled permeability. The second part of this dissertation explains the improvement in predicting the kinematic efficiency of strike-slip fault systems. It consists of one chapter and addresses the following research objectives: (1) utilizing a comprehensive labeled dataset under various loading conditions of simulated strike-slip faults to build a predictive model of off-fault deformation; (2) searching for the most appropriate architecture, loss functions, and hyperparameters that maximize the performance of the convolutional neural network (CNN) model on an unseen fault dataset; and (3) testing the hypothesis that the trained CNN model can estimate off-fault deformation that is consistent with geologic observations

This dissertation examines public discourse in two semi-authoritarian contexts, the Chinese territories of Macau and Hong Kong, through two empirical studies. It concludes that while it is possible to satisfy the high standards of a successful deliberative mini-public in these environments, under current political conditions, significant barriers remain to the development of deliberative authoritarianism

Bacterial infection of mammalian hosts involves competition between the two parties on various fronts. Listeria monocytogenes is an intracellular bacterial pathogen that can cause spontaneous abortions in pregnant women and diseases such as meningitis in immunocompromised individuals. A clever way in which L. monocytogenes goes up against its host is by spreading from one cell to the other without exposure to the extracellular space. In this way, the bacterium evades detection by the host's humoral immune response, even as it travels from epithelial cells in the intestine, the initial site of infection, to distal organs such as the brain, spleen and placenta. In this work, we focus on three aspects of Listeria monocytogenes's life within epithelial monolayers: cell-to-cell spread, intracellular replication and host cell extrusion from the monolayer. First, we show that the host cell junction protein, E-cadherin, promotes intercellular spread at the recipient side of cell contacts. We find that expression of an E-cadherin mutant that cannot be ubiquitinated and is expected to undergo reduced internalization, limits bacterial cell-to-cell spread. Based on these observations, we hypothesize that L. monocytogenes is transferred from donor to recipient cell concurrent with E-cadherin internalization at the recipient cell and produce results that are consistent with caveolin-mediated E-cadherin internalization and macropinocytosis as mechanisms that promote L. monocytogenes cell-to-cell spread. Overall, cell-to-cell spread appears to be a domain in which the bacterium prevails over the host cell by exploiting a host cell process (transendocytosis) to disseminate through the epithelial monolayer. Next, we use live-cell microscopy and quantitative image analysis to characterize intracellular growth of L. monocytogenes within epithelial cells. We show that early in infection, bacteria grow exponentially at a constant rate, but that growth rate begins to slow after a certain number of bacteria per host cell volume is reached. Our results suggest that bacterial growth is limited by nutrient availability and similar to bacterial growth in culture, accumulation of bacteria within an enclosed space exhausts the environment of nutrients over time. While our results suggest that L. monocytogenes delays this reduction of growth rate by accessing additional nutrients through cell-to-cell spread, bacterial growth rate for the focus as a whole does decline over time in our experimental paradigm. This might represent a scenario in which L. monocytogenes attempts to overcome a limitation of inhabiting its intracellular niche but is not entirely successful in doing so. Lastly, we describe a phenomenon in which the host cell emerges as the victor and effectively limits bacterial infection. Here, at late time points in infection, innate immune signaling causes the collective extrusion of infected cells from the epithelial monolayer. Once infected cells are expelled from the monolayer, bacteria are no longer able to effectively disseminate to cells in the intact monolayer and face nutrient deprivation as their extruded host cell dies over time

Urban areas cover a small portion of the earth, yet they accommodate a majority of the world population and are associated with a majority of the global energy use and greenhouse gas emissions. These facts make cities the forefront of combating climate change, with transportation, buildings (residential and commercial), and electricity generation sectors being the foci of this effort due to their significant environmental footprints. Thus, improving the efficiency of these sectors can play an essential role in reducing the environmental footprint of urban districts. Coordinated design of infrastructure has shown to be an effective approach to create more sustainable urban systems. Such coordination leads to integrated municipal supply systems (e.g., water and energy supply, and EV charging infrastructure) with improved overall performance when compared to the simple superposition of individual supply systems. This coordination can be most feasibly achieved at a neighborhood scale, i.e., 100-1000 adjacent buildings, where the supply and demand of the integrated infrastructure can be designed and optimized simultaneously. Since building mix largely determines the demand profile of urban neighborhoods, simultaneously optimizing the building mix and integrating infrastructure represents a potential opportunity for designing more sustainable urban districts. To investigate and understand this opportunity, this dissertation focuses on the design and optimization of the energy system (including EV charging infrastructure), wastewater treatment system, and building mix of urban neighborhoods. Typically, urban energy and water infrastructure systems are designed after the demands of the community are determined. At that late stage, improving the environmental and economic performance of the system is difficult. However, if the supply and demand of municipal services are optimized simultaneously, especially at the early stages of the design when such coordination is more feasible, the resulting urban system has proven to outperform the isolated design using several sustainability metrics. This dissertation proposes a novel method to design and optimize the hourly demand and supply of integrated energy and water system in an urban district for environmental and economic sustainability. The increasing electricity demand from the growing number of EVs on the road will impose pressure on local and regional electricity grids. Hence, considering the influence of EVs will be essential for designing future urban energy systems. This dissertation proposes a new method for integrating the design of energy system and building mix with the EV charging infrastructure in urban neighborhoods, and for optimizing these infrastructure systems for economic and environmental targets. The spatial configuration of urban infrastructure systems is essential in modeling the performance and minimizing the impacts of these systems, especially for the urban water and energy systems. However, integrating the design and optimization of the supply, demand, and network layout across multiple infrastructure systems at the neighborhood scale gives rise to highly complex optimization problems. Traditional optimization algorithms cannot properly explore the optimal solutions in the vast solution space of these problems in a reasonable amount of time. This dissertation develops a new method that facilitates concurrent optimization of the supply, demand, and spatial configuration of integrated infrastructure at a neighborhood scale for multiple objective functions. Keywords: Sustainable Urban Systems, Integrated Infrastructure, Water-Energy Nexus, EV Charging Infrastructure, Multiobjective Optimization, Mixed Integer Nonlinear Programming, Conditional Generative Adversarial Networks

The accelerating need to mitigate global emissions of CO2 has driven significant advancements in designing systems and strategies that transition towards a sustainable future. However, a growing global population continues to rely on fossil-derived sources for power, chemicals, agriculture, and transportation sectors. Several strategies and technologies are being developed to mitigate greenhouse gas emissions that exacerbate global climate change. Of these, electrochemical carbon dioxide reduction (CO2R) exhibits promise for utilizing captured carbon to produce industrially relevant products, especially if coupled to point sources. The performance of CO2 electrolyzers is tied to a combination of coexisting phenomena with varying weighted contributions, which change under different operating regimes. This dissertation aims to elucidate governing factors and provide understanding for high-rate CO2 reduction in vapor-fed systems operating in bulk-neutral pH electrolyte. The adoption of gas diffusion electrodes (GDEs) brings forth new formulations and environments to study, providing increased reaction rates due to enhanced mass transport. Most strategies for targeting multi-carbon product formation rely on utilizing Cu within the electrode, derived from early work from Hori and co-workers. Several advancements in materials discovery, component analysis and configuration, and system design have led to higher reaction rates and selectivity towards these products. However, a lot of the current strategies for high-rate CO2R require highly alkaline pH systems that are expensive to maintain and lead to large anodic CO2 crossover. Conversely, the use of bulk-neutral pH electrolyte leads to lower selectivity due to increased rates of hydrogen evolution reaction. To explore this, we design electrolyzers using advanced manufacturing paired with a thin catalyst-GDE for improved understanding and performance in bulk-neutral pH systems. We observe high selectivity towards multi-carbon products (> 85%) at > 200 mA cm−2 due to a high local pH induced by the catalyst environment. Upon increased current density, mass transport greatly affects product formation based on potential and inlet CO2 flow rate within the mixed-control regime. Both experimental and modeling approaches help uncover underlying mass transport effects including temperature, hydroxide formation, and partial penetration into the diffusion media. To further uncover effects on the local reaction environment in both foil and gas diffusion electrodes, we tune the electrode surface area of catalysts by synthesizing a rough Cu nanoflower morphology shown to produce multi-carbon products when performing carbon monoxide reduction. Starting with foil, these catalyst nanostructures are translated to the GDE architecture to understand the interplay of catalyst surface and surrounding microenvironment on current-potential relationships. Like on Cu foil electrodes, the intrinsic activity of Cu does not appear to change based on nanostructuring GDEs. Furthermore, we observe lower overpotentials with increased catalyst area at a constant current density, uncovering a semi-linear relationship between the ratio of roughness factors and potential difference when activation-controlled. We then create a foil-GDE hybrid reaction environment by synthesizing a deterministic electrode using a two-photon printing technique. This electrode consists of a perfectly distributed grid, representing an ordered structure with uniform pore size. This results in geometric current densities operating between previous foil and GDEs in comparison, with probable product activity zones based on increased rates of HER and slow CO2 diffusion in aqueous phase. These approaches towards understanding changing reaction environments in combination with materials design can aid in the development of systems with increased energy efficiency without sacrificing selectivity. Our efforts alongside others investigating CO2R have provided compelling results towards commercialization. Several techno-economic assessments suggest selectivity, cell potential, electricity price, and anodic CO2 cross-over are among the most influential on production price; this is often compared to market price. However, these analyses do not consider differences in technology readiness level or cost class. In the final part of this dissertation, we present an energetics-first approach for analyzing these systems. Our results show a small threshold for cell potential is permitted in order to compare with ethylene production from conventional cracking technologies, thus requiring new formulations for operating in bulk-neutral pH. We assess the prospects of various anode reaction pairs for sustainable ethylene production. Finally, we conclude and provide outlook and considerations for the future development of these systems. With both experimental and modeling approaches, this dissertation evaluates the underlying effects and prospects for high-rate electrochemical CO2 reduction operating in bulk-neutral pH

Conjugated polymers (CPs), polymers with backbones of alternating single and double bonds, can conduct electricity yet remain lightweight, flexible, optically transparent, and biocompatible. With this attractive combination of properties, CPs have the potential to transform devices for clean energy generation and storage, computing, and personalized health. However, CPs cannot currently meet the full range of performance requirements in many of these demanding applications, while the intrinsic disorder in CP morphology frustrates experimental efforts to characterize film microstructure and to elucidate the structure-function relationships necessary for rational device design. Computational and theoretical techniques have the resolution to probe these microstructural details, but the reliance on generic models, which do not capture important chemical details underlying CP flexibility, has limited their ability to investigate CP morphology. I seek to extend generic models to capture these details, creating a reliable foundation to study the influence of CP structure on device morphology and transport. Generic molecular dynamics (MD) models commonly used to study CP film and solution microstructure were not trained on CPs, yet their extrapolation to CP systems has not been validated. Density functional theory (DFT) calculations can reliably model CP flexibility, but the plethora of density functionals (DFs) are empirically formulated and are not all accurate for every system. Thus, I completed two projects to improve MD models of a popular CP system, poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS). First, I benchmarked several DFs against high-accuracy ab initio predictions of CP flexibility. I found that electron localization induced by Hartree-Fock (HF) exchange systematically influences these predictions: DFs which balance HF with other contributions to the DFT Hamiltonian, such as ωB97x-D, provide better predictions. Then, I used ωB97x-D to parameterize an MD force field (FF) for PEDOT in undoped and highly doped states. These FFs are more accurate than the generic models and capture the effect of doping on CP flexibility. Accurately representing PEDOT:PSS microstructure proves to be foundational to understanding charge transport (CT) mechanisms in the material. In experiment, organic electrochemical transistors (OECTs) with a PEDOT:PSS channel exhibited an exponential decay of the mobility with increasing PSS loading, which existing percolation-based and effective-medium CT models could not explain. In a collaborative effort, I helped develop a model proposing that holes tunnel between isolated PEDOT chains in the PSS-rich phase, giving rise to macroscopic CT. Our multiscale CT model, using the optimized MD FF to represent film microstructure, replicates the observed mobility scaling. Thus, the ion-conductive phase in OECTs, traditionally understood as electrically insulating, is in fact a barrier to tunneling and therefore plays a critical role in CT. We demonstrate that the choice of ion-conducting material influences tunneling rates, suggesting a new relationship to guide future design of OECTs and other CP devices. Statistical polymer models, such as the wormlike chain (WLC), permit the study of polymer behavior over longer length and time scales than atomistic simulation, but existing models fail to capture the anisotropic bending present in CP chains and CP melt morphology. To introduce this anisotropy, I created a generalized WLC model called the ribbon-like chain (RLC) and solved the RLC free-chain Green's function. The Green's function solutions can calculate single-chain statistics such as tangent autocorrelations, the radius of gyration, and chain shape, and clearly demonstrate the influence of anisotropic bending on these properties. When the RLC model is applied to MD simulations of CP melts, the chain statistics also highlight anisotropic bending and are reflective of CP structure. Thus, the RLC model can better examine the relationship between CP structure and melt morphology. Altogether, this dissertation advances models of CP morphology and transport by incorporating chemical details influential on the conformational behavior of CPs. These models lay a foundation for more robust computational and theoretical studies of CPs, supporting experimental improvements of these promising materials

How we move and how much we move profoundly affect our health and wellbeing. In biomechanics, we analyze movement to help identify, monitor, and treat diseases and disorders that impact movement. However, biomechanical assessments and interventions have historically been restricted to expensive laboratory settings. My thesis work includes three studies that leverage advancements in computer science, biomechanics, and psychology to translate biomechanical interventions to a clinical or home setting. First, we developed a machine learning model to predict knee loading from inputs that could be extracted from 2D video and demonstrate the feasibility of prescribing personalized biomechanical interventions with a smartphone camera. Second, we evaluated the psychological construct of mindset in people with knee osteoarthritis and found that mindset relates to physical activity levels and the use of exercise for symptom management. Finally, we developed a platform to deploy a nationwide at-home biomechanics study that included an order of magnitude more participants than traditional laboratory studies. With this large dataset, we explored new relationships between biomechanical parameters and measures of health and wellbeing. Together, these studies contribute to a future where simple, scalable movement assessments are used to evaluate health and treat musculoskeletal diseases

Metal-catalyzed C--H functionalization has had an immense impact on streamlining synthesis by enabling previously inaccessible synthetic disconnections in organic synthesis and late-stage diversification of complex molecules. Dirhodium tetracarboxylate catalysts represent the state-of-the-art for intermolecular C(sp3)--H amination, but struggle to functionalize allylic sites due to the competitive reactivity of alkenes. Further advances in dirhodium-mediated C--H amination will require developing catalysts with enhanced stability and different ligand architectures to control selectivity in reactions with substrates bearing reactive functional groups and numerous C--H bonds. However, carboxylate ligands are inherently limited in their electronic, steric, and geometric tunability, which has severely hampered efforts to expand the scope of dirhodium-catalyzed C--H amination. To move beyond dirhodium tetracarboxylate catalysts, we developed a new, modular family of bridging ligands based on 2-pyridones. This work has advanced by capitalizing on a unique combination of computing tools, including molecular dynamics, transition state analysis, and statistical modeling, in a rationally connected feedback loop. These novel dirhodium catalysts mediate intermolecular allylic C−H amination with unprecedented selectivity. The most effective catalyst, Rh2(btz)3(O2CtBu), features one carboxylate and three substituted 2-pyridone ligands that create a spatially complex, conformationally dynamic reactive site. In addition to mediating high yielding amination reactions, these catalysts challenge the paradigm that rigid, well-defined ligand frameworks are optimal for controlling reaction outcomes. Our results demonstrate the versatility of 2-pyridone-derived ligands for optimizing catalyst performance. In addition, this work represents the first successful divergence from dirhodium tetracarboxylate complexes for catalyzing intermolecular C--H amination reactions. More broadly, our design blueprint showcases how data science and computational techniques can be implemented synergistically to help advance reaction design. By capitalizing on the efficiency of dirhodium-catalyzed intermolecular C--H amination reactions, we have developed a novel strategy for constructing saturated azacycles from readily available precursors. The preparation of substituted azetidines and larger ring, nitrogen-containing saturated heterocycles is enabled through efficient and selective intermolecular sp3-C--H amination of alkyl bromide derivatives. A range of substrates are demonstrated to undergo C--H amination and subsequent sulfamate alkylation in good to excellent yield. N-Phenoxysulfonyl-protected products can be unmasked under neutral or mild basic conditions to yield the corresponding cyclic secondary amines. The preparative convenience of this protocol is demonstrated through gram-scale and telescoped multistep procedures. Application of this technology is highlighted in a nine-step total synthesis of an unusual azetidine-containing natural product, penaresidin B.

Quantitative predictions of fluid flow and transport in porous media are often compromised by multi-scale heterogeneity and insufficient site characterization. These factors introduce uncertainty on the input and output of physical systems which are generally expressed as partial differential equations (PDEs). The characterization of this predictive uncertainty is typically done with forward propagation of input uncertainty as well as inverse modeling for the dynamic data integration. The main challenges of forward uncertainty propagation arise from the slow convergence of Monte Carlo Simulations (MCS) especially when the goal is to compute the probability distribution which is necessary for risk assessment and decision making under uncertainty. On the other hand, reliable inverse modeling is often hampered by the ill-posedness of the problem, thus the incorporation of geological constraints becomes increasingly important. In the thesis, four significant contributions are made to alleviate these outstanding issues on forward and inverse problems. First, the method of distributions for the steady-state flow problem is developed to yield a full probabilistic description of outputs via probability distribution function (PDF) or cumulative distribution (CDF). The derivation of deterministic equation for CDF relies on stochastic averaging techniques and self-consistent closure approximation which ensures the resulting CDF has the same mean and variance as those computed with moment equations or MCS. We conduct a series of numerical experiments dealing with steady-state two-dimensional flow driven by either a natural hydraulic head gradient or a pumping well. These experiments reveal that the proposed method remains accurate and robust for highly heterogeneous formations with the variance of log conductivity as large as five. For the same accuracy, it is also up to four orders of magnitude faster than MCS with a required degree of confidence. The second contribution of this work is the extension of the distribution-based method to account for uncertainty in the geologic makeup of a subsurface environment and non-stationary cases. Our CDF-RDD framework provides a probabilistic assessment of uncertainty in highly heterogeneous subsurface formations by combining the method of distributions and the random domain decomposition (RDD). Our numerical experiments reveal that the CDF-RDD remains accurate for two-dimensional flow in a porous material composed of two heterogeneous geo-facies, a setting in which the original distribution method fails. For the same accuracy, the CDF-RDD is an order of magnitude faster than MCS. Next, we develop a complete distribution-based method for the probabilistic forecast of two-phase flow in porous media. The CDF equation for travel time is derived within the efficient streamline-based framework to replace the MCS in the previous FROST method. For getting fast and stable results, we employ numerical techniques including pseudo-time integration, flux-limited scheme, and exponential grid spacing. Our CDF-FROST framework uses the results of the method of distributions for travel time as an input of FROST method. The proposed method provides a probability distribution of saturation without using any sampling-based methods. The numerical tests demonstrate that the CDF-FROST shows good accuracy in estimating the probability distributions of both saturation and travel time. For the same accuracy, it is about 5 and 10 times faster than the previous FROST method and naive MCS, respectively. Lastly, we propose a consensus equilibrium (CE) framework to reconstruct the realistic geological model by the inverse modeling of sparse dynamic data. The optimization-based inversion techniques are integrated with recent machine learning-based methods (e.g., variational auto-encoder and convolutional neural network) by the proposed CE algorithm to capture the complicated geological features. The numerical examples verify that the proposed method well preserves the geological realism, and it efficiently quantifies the uncertainty conditioned on dynamic information