12023 research outputs found
Sort by
Sulfur Isotopic Insights into the Modern and Ancient Marine Sulfur Cycles
Get PDFThe marine sulfur cycle plays a key role in regulating Earth's surface oxygen (O₂) levels through its interactions with the carbon and iron cycles. Our understanding of the sulfur cycle has traditionally come from measurements of the sulfur isotopic compositions of marine sulfate (SO₄²⁻) and sulfur-bearing materials in marine sediments. Because the residence time of SO₄²⁻ in seawater is long (Myr) compared to the mixing time of Earth's oceans (kyr), the concentration and sulfur isotopic composition of marine SO₄²⁻ are homogeneous in modern seawater and are assumed to have been homogeneous throughout most of the Phanerozoic Eon (541 Ma to the present). This assumption of homogeneity, when combined with sulfur isotopic composition measurements, has enabled box model reconstructions of the relative fluxes of oxidized versus reduced sulfur leaving the oceans at times in Earth's past. Such reconstructions have informed our understanding of the interactions between Earth's tectonics, climate, and elemental cycles.
This thesis tests some of the key assumptions made in sulfur cycle box models and attempts to better understand sulfur isotopic variability in geologic archives using a combination of measurements and modeling. Measurements of the sulfur isotopic composition (i.e., δ³⁴S) of SO₄²⁻ in Permo-Carboniferous brachiopod shells demonstrate that more precise records of SO₄²⁻ δ³⁴S may be generated via careful sampling that avoids diagenetically altered phases (Chapter II). Furthermore, measurements of heterogeneous carbonate associated sulfate (CAS) δ³⁴S within carbonates deposited across the End-Permian mass extinction (EPME) in South China show that a lack of careful sampling can substantially alter our understanding of the marine sulfur cycle at times in Earth's past (Chapter III). Simple models constructed in each of these studies indicate that changes in the δ³⁴S of the sulfur input to the ocean, the δ³⁴S offset (i.e., Δδ³⁴S) between the oxidized and reduced sulfur output fluxes, and the amount of SO₄²⁻ incorporated during diagenetic alteration - all assumed to be negligible in many studies of the marine sulfur cycle - may viably explain these data. Development of a sediment diagenesis model that includes sulfur isotopic species demonstrates that variations in organic matter rain rate, ferric iron input, sedimentation rate, bottom water O₂ concentration, and bottom water SO₄²⁻ concentration may all affect Δδ³⁴S in a given sedimentary environment (Chapter IV). Application of this model to pore water SO₄²⁻ and hydrogen sulfide H₂S δ³⁴S data from International Ocean Discovery Program (IODP) Expedition 361, IODP Expedition 363, and R.V. Knorr cruise KN223 sites shows that Δδ³⁴S is ubiquitously large in these deep ocean sedimentary environments (Chapter V). Cluster analysis of pore water [SO₄²⁻] profiles collected during previous deep ocean cruises successfully extracts and groups profiles that are similar to those observed on these three cruises (Chapter VI). Comparison of cluster data to a compilation of recent marine pyrite (FeS₂) δ³⁴S data confirms that pyrite burial in shelf sediments constitutes the majority of pyrite burial occurring globally in the modern day. However, changes in sea level or in other variables that affect sediment deposition may plausibly force an increase in deep ocean pyrite burial and a corresponding change in the global Δδ³⁴S. Future studies of the modern and ancient marine sulfur cycles must carefully consider the geologic and geochemical context of sulfur isotopic measurements - including sea level changes, sedimentation rate changes, and measured or presumed concentrations of other redox-active species - if interpretations of such data are to be robust.</p
Numerical Investigation of Compressibility Effects in Reacting Subsonic Flows
Get PDFDirect numerical simulations (DNS) of reacting flows are routinely performed either by solving the fully compressible Navier-Stokes equations or using the low Mach number approximation. The latter is obtained by performing a Mach number expansion of the Navier-Stokes equations for small Mach numbers. These two frameworks differ by their ability to capture compressibility effects, which can be broadly defined as phenomena that are not captured by the low Mach number approximation. These phenomena include acoustics, compressible turbulence, and shocks. In this thesis, we systematically isolate compressibility effects in subsonic flows by performing two sets of DNS: one using the fully compressible framework, and one using the low Mach number approximation. We are specifically interested in the interactions between turbulence, acoustics, and flames.
The addition of detailed chemistry in the compressible flow solver required the development of a novel time integration scheme. This scheme combines an iterative semi-implicit method for the integration of the species transport equations, and the classical Runge-Kutta method for the integration of the other flow quantities. It is found to perform well, yielding time steps limited by the acoustic CFL only. Furthermore, the computational cost per iteration of this hybrid scheme is low, being comparable to the one for the classical Runge-Kutta method.
After extensive validation, the first application is the investigation of flame-acoustics interactions in laminar premixed flames. The thermodynamic fluctuations that accompany the acoustic wave are shown to significantly impact the flame response. Using the Rayleigh criterion, the flame-acoustics system is found to be thermo-acoustically unstable for various fuels, flow conditions, and acoustic frequencies. As expected, the low Mach number approximation and the fully compressible framework are in good agreement at low frequencies, since the flame is very thin compared to the acoustic wavelength. The two frameworks differ for very large acoustic frequencies only. In the high frequency limit, the gain reaches a plateau using the low Mach number approximation, while it goes to zero using the fully compressible framework. This is related to the spatial variations in the acoustic pressure field, which are not present in the low Mach number approximation. However, for practically relevant acoustic frequencies, the low Mach number framework is found to yield accurate results.
Next, a numerical methodology to simulate compressible flows in geometries that lack a natural turbulence generation mechanism is presented. It is found that, unlike in incompressible flows, special care must be taken regarding the energy equation and the presence of standing acoustic modes. When using periodic boundary conditions, forcing the dilatational velocity field promotes the growth of unstable modes. This is explained by extracting the eigenvalues of the linearized forced Navier-Stokes equations. Based on these observations, it is found necessary to force the solenoidal velocity field only. This methodology is applied first to simulations of subsonic homogeneous non-reacting turbulence. We present simulations results for turbulent Mach numbers varying from 0.02 to 0.65. The Mach number dependence of various quantities, such as the dilatational to solenoidal kinetic energy ratio, is extracted. The Mach number scaling of all quantities of interest is found to be readily explained by the low Mach number expansion, specifically the zeroth and first order sets of equations, for turbulent Mach numbers up to 0.1.
Finally, the interaction between subsonic compressible turbulence and premixed flames is investigated. Compressibility effects are isolated by comparing results obtained with the low Mach number approximation and the fully compressible framework, at the same flow conditions. Compressibility effects on chemistry are found to be limited for turbulent Mach numbers at least up to 0.4, especially when contrasted with the large impact of the Karlovitz number. Compressibility effects give rise to significant thermodynamic fluctuations away from the flame front, but these remain small compared to the large fluctuations due to the presence of the turbulent flame brush. The low Mach number approximation thus remains a valid framework for the Mach numbers considered, when the primary goal is to characterize the impact of turbulence on the chemical processes at play.</p
Online Learning from Human Feedback with Applications to Exoskeleton Gait Optimization
Get PDFSystems that intelligently interact with humans could improve people's lives in numerous ways and in numerous settings, such as households, hospitals, and workplaces. Yet, developing algorithms that reliably and efficiently personalize their interactions with people in real-world environments remains challenging. In particular, one major difficulty lies in adapting to human-in-the-loop feedback, in which an algorithm makes sequential decisions while receiving online feedback from humans; throughout this interaction, the algorithm seeks to optimize its decision-making quality, as measured by the utility of its performance to the human users. Such algorithms must balance between exploration and exploitation: on one hand, the algorithm must select uncertain strategies to fully explore the environment and the interacting human's preferences, while on the other hand, it must exploit the empirically-best-performing strategies to maximize its cumulative performance.
Learning from human feedback can be difficult, as people are often unreliable in specifying numerical scores. In contrast, humans can often more accurately provide various types of qualitative feedback, for instance pairwise preferences. Yet, sample efficiency is a significant concern in human-in-the-loop settings, as qualitative feedback is less informative than absolute metrics, and algorithms can typically pose only limited queries to human users. Thus, there is a need to create theoretically-grounded online learning algorithms that efficiently, reliably, and robustly optimize their interactions with humans while learning from online qualitative feedback.
This dissertation makes several contributions to algorithm design for human-in-the-loop learning. Firstly, this work develops the Dueling Posterior Sampling (DPS) algorithmic framework, a model-based, Bayesian approach for online learning in the settings of preference-based reinforcement learning and generalized linear dueling bandits. DPS is developed together with a theoretical regret analysis framework, and yields competitive empirical performance in a range of simulations. Additionally, this thesis presents the CoSpar and LineCoSpar algorithms for sample-efficient, mixed-initiative learning from pairwise preferences and coactive feedback. CoSpar and LineCoSpar are both deployed in human subject experiments with a lower-body exoskeleton to identify optimal, user-preferred exoskeleton walking gaits. This work presents the first demonstration of preference-based learning for optimizing dynamic crutchless exoskeleton walking for user comfort, and makes progress toward customizing exoskeletons and other assistive devices for individual users.</p
Annular Links with sl₂-Irreducible Annular Khovanov Homology
Get PDFWe prove that the rank of annular Khovanov homology of a braid in its next-to-top annular grading is always greater than 1, and as an immediate consequence prove that annular Khovanov homology of an annular link as a representation over the Lie algebra sl₂ is irreducible if and only if the annular link is isotopic to the core of the annulus. We also conjecture an analogue of Fox's trapezoid conjecture in the context of annular Khovanov homology with a computer-assisted supporting evidence
Mechanism of Action of a Therapeutic Peptide, Risuteganib, Suggests that Supporting Mitochondrial Function Underlies its Clinical Efficacy in Treating Leading Causes of Blindness
Get PDFAge-related macular degeneration (AMD) and diabetic retinopathy (DR) are the leading causes of blindness in the developed world and on the rise globally due to the growth of an aging population and an increasing number of diabetics. Antibodies of vascular endothelial growth factor (VEGF), which target neovascularization in the advanced stages of both diseases, have been the main treatment for the past decade. However, anti-VEGF therapies suffer from short half-life and high cost inherent to antibodies, limiting the medical availability to a broader population.
To fill the unmet medical need for treating retinal diseases, a novel therapeutic oligopeptide, risuteganib, is currently in Phase II clinical trials. Results from completed trials suggest that risuteganib has comparable drug efficacy to anti-VEGF therapies, long half-life, low cost, and absence of drug-related adverse events in several hundred patients enrolled in clinical trials for diabetic macular edema (DME) and dry AMD. Risuteganib was originally designed to target neovascularization, intending to inhibit integrin cell-surface receptors and thereby block adhesion and migration of abnormal blood vessel cells. Early in our study, we found experimental evidence contrary to this mechanism of action (MOA).
Our journey began with an unbiased search for the binding loci in retinal tissue, using peptide-directed fluorescent labeling. We found out that risuteganib specifically binds to a monolayer of cells, the retinal pigment epithelium (RPE), which has essential functions in maintaining the homeostasis of the retina, and its dysfunction is the hallmark for both blinding retinal diseases. In vitro study in an RPE cell model, ARPE19, showed that risuteganib protects cells against elevated oxidative stress that is associated with AMD and DR. This protective effect correlates with maintaining mitochondrial function. Further study of mitochondrial bioenergetics, in collaboration with Dr. Cris Kenney at UCI, revealed that risuteganib supports oxidative phosphorylation metabolism in the mitochondria.
Based on the chemical similarity of risuteganib with a natural product, we hypothesized that risuteganib may act through a mitochondrial enzyme, pyruvate dehydrogenase kinase (PDK), specifically PDK1 that is responsive to disease-related hypoxia-inducible factor 1 alpha (HIF-1α). Protein phosphorylation assay and enzymatic assay confirmed that risuteganib inhibits PDK1, as a result, reducing phosphorylation of an essential enzyme, pyruvate dehydrogenase (PDH). Leaving PDH in its unphosphorylated form allows its continued activity in oxidative phosphorylation metabolism, which offers a molecular explanation of the ability to support mitochondrial activity. This leads to our current hypothesis that risuteganib’s mechanism of action (MOA) is through inhibition of PDK and protection of mitochondrial functions in RPE cells for treating retinal diseases.
Protecting mitochondrial functions may be beneficial to other cell types and in other diseases that subject cells to oxidative stress. As the mitochondria targeting is a potential therapy for diverse life-threatening diseases, including inflammatory disease, cardiovascular disease, and cancer, the present hypothesis invites us to expand our scope of view for this study to broader applications.</p
Context-Dependent, Combinatorial Logic of BMP Signaling
Get PDFEvolution generated diverse signaling proteins for the control of multicellular patterns and organ- isms. These include the proteins of the Bone Morphogenetic Protein (BMP) pathway. Nearly a dozen BMPs activate the BMP pathway to promote the formation of tissues as diverse as bone, cartilage, blood vessels, and the kidney, making them attractive therapeutics for regenerating those tissues in adults. During development, the response to a given BMP depends heavily on context, such as which other BMPs are present and which BMP receptors are expressed on the cell being ac- tivated. However, despite knowing that context matters, the overall logic of this context-dependent signal processing, including the roles of specific ligands and receptors in shaping context and how this logic arises from biochemical features of specific pathway components, remains unclear. Inspired by maps of gene epistasis and drug interactions that functionally classify members of complex biological systems, we comprehensively measured responses to all pairs of ten BMP homodimers (BMP2, BMP4, BMP5, BMP6, BMP7, BMP9, BMP10, GDF5, GDF6, and GDF7), combining robotic liquid handling with a high-throughput fluorescent reporter of pathway activa- tion. These data functionally classify ligands into "equivalence groups," or ligands that combine in the same way with all other ligands across combinations. Surprisingly, the functional groupings do not correlate with similarity of ligand sequence and can change with cell context. Together, the context-dependent equivalence groups summarize the diverse responses to combinations of BMP ligands and their dependence on specific BMP receptors. The experimentally observed pairwise responses are also consistent with a mathematical model where BMP ligands compete for limited BMP receptors with different affinities and then produce outputs with different ligand-specific activ- ities. Ultimately, these results provide a useful reference for explaining the unique effects of BMP combinations in different tissues or time points in development, as well as highlighting counter- intuitive mechanisms for this complex signal processing. Chapter 1 provides an introduction to how and why we study cell-cell signaling. Chapter 2 provides a summary of the determination of equivalence groups, their dependence on receptor context, and fitting the mathematical model of receptor competition. Chapter 3 provides suggestions for future work, including recommendations for improved model fitting as well as crucial extensions to the definitions of BMP "combinations" and "context" to deepen our understanding and control of this critical pathway
Understanding the Lifetime and Rate of Protein Production in Cell-Free Reactions While Maximizing Energy Use
Get PDFLiposomes, or vesicles, offer promising applications in fields including biofuel synthesis, drug delivery, and toxin removal. Programmable liposomes can be used for optimal protein synthesis and to prototype genetic technologies. However, one of the major challenges is the short lifetime of protein production. Here, we add metabolites and molecules to cell-free reactions at different times to interrogate their importance. Through this testing, we find that ATP only slightly enhances protein synthesis, and ADP can help a reaction reach steady state faster. We also note that the excess of particular molecules, such as NAD and 3PGA, can halt protein production. With this data, we developed a more accurate chemical reaction network-based model for cell-free reactions. We also begin to study an unexplained discrepancy in protein production between bulk and vesicle dynamics. To quantify protein synthesis, we use E. coli extract and energy buffer, often called cell-free transcription and translation (TXTL), with a chosen DNA template both within vesicles (encapsulated) and without (bulk). We have also been able to uncover fundamental properties of transcription/translation systems. We supplement this data with computational models utilizing chemical reaction networks. We established a vesicle setup with membrane pores and supplemental energy buffer on the outside which increased the efficiency of protein synthesis. By using chemical reaction network models, we have highlighted differences and similarities between models and experiments. With this setup, vesicles can be used for more complicated applications, such as drug delivery or genetic construct testing
Multi-Resolution Lattice Green's Function Method for High Reynolds Number External Flows
Get PDFThis work expands the state-of-the-art computational fluid dynamics (CFD) methods for simulating three-dimensional, turbulent, external flows by further developing the immersed boundary (IB) Lattice Green's function (LGF) method.
The original IB-LGF method applies an exact far-field boundary condition using fundamental solutions on regular Cartesian grids and allows active computational cells to be restricted to vortical flow regions in an adaptive fashion as the flow evolves. The combination of spatial adaptivity and regular Cartesian structure leads to superior efficiency, scalability, and robustness, but necessitates uniform grid spacing. However, the scale separation associated with thin boundary layers and turbulence at higher Reynolds numbers favors a more flexible distribution of elements/cells, which is achieved in this thesis by developing a multi-resolution LGF approach that permits block-wise grid refinement while maintaining the important properties of the original scheme. We further show that the multi-resolution LGF method can be fruitfully combined with the IB method to simulate external flows around complex geometries at high Reynolds numbers. This novel multi-resolution IB-LGF scheme retains good efficiency, parallel scaling as well as robustness (conservation and stability properties). DNS of bluff and streamlined bodies at Reynolds numbers O(104) are conducted and the new multi-resolution scheme is shown to reduce the total number of computational cells up to 99.87%.
We also extended this method to large-eddy simulation (LES) with the stretched-vortex sub-grid-scale model. In validating the LES implementation, we considered an isolated spherical region of turbulence in free space. The initial condition is spherically windowed, isotropic homogeneous incompressible turbulence. We study the spectrum and statistics of the decaying turbulence and compare the results with decaying isotropic turbulence, including cases representing different low wavenumber behavior of the energy spectrum (i.e. k2 versus k4). At late times the turbulent sphere expands with both mean radius and integral scale showing similar time-wise growth exponents. The low wavenumber behavior has little effect on the inertial scales, and we find that decay rates follow Saffman's predictions in both cases, at least until about 400 initial eddy turnover times. The boundary of the spherical region develops intermittency and features ejections of vortex rings. These are shown to occur at the integral scale of the initial turbulence field and are hypothesized to occur due to a local imbalance of impulse on this scale.</p
Numerical Simulation of Performance and Solar-to-Fuel Conversion Efficiency for Photoelectrochemical Devices
Get PDFThe Industrial Revolution was energized by coal, petroleum, and natural gas. It is clear that fossil fuels, which drive steam and electrical engines, made possible a monumental increase in the amount of productive energy available to humans. But in the meantime, the constant burning of fossil fuels has changed the natural greenhouse, intensified global warming, deteriorated air quality, and eventually caused irreversible environmental damage on our planet. Renewable energy especially solar energy offers a desirable approach toward meeting our growing energy needs while largely reducing fossil fuel burning. The major problems in terms of harvesting energy directly from sunlight turn out to be low energy concentration and intermittency. Building solar-fuel generators, which stores solar energy in chemical bonds, similar to photosynthesis in nature, provides a possible solution to these two problems. Carbon-free chemicals, such as hydrogen gas, which are produced by solar-driven water-splitting, or carbon-neutral chemicals, such as methane and ethylene, which are produced by solar-driven CO₂ reduction, are all promising clean fuels for solar storage.
This thesis is focused on studying the performance and solar to fuel conversion efficiency of existing and hypothetical test-bed photoelectrochemical prototypes using multi-physics modeling and simulation to lay a foundation for future implementation and scale-up of the integrated, solar-driven systems. For water-splitting systems, a sensitivity analysis has been made to assess the relative importance of improvements in electrocatalysts, light absorbers, and system geometry on the efficiency of solar-to-hydrogen generators. Besides, an integrated photoelectrolysis system sustained by water vapor is designed and modeled. Under concentrated sunlight, the performance of the photoelectrochemical system with 10× solar concentrators was simulated and the impact of hydrogen bubbles that are generated inside the cathodic chamber on the performance of the photoelectrolysis system was evaluated. For CO₂ reduction systems, operational constraints and strategies for systems to effect the sustainable, solar-driven reduction of atmospheric CO₂ were investigated. The spatial and light-intensity dependence of product distributions in an integrated photoelectrochemical CO₂ reduction system was modeled and simulated. Finally, the performance a flow-through gas diffusion electrode for electrochemical reduction of CO or CO₂ was evaluated.
This thesis can be divided into three parts. The first part discusses the importance of solar energy. The second part includes Chapter II, Chapter III, Chapter IV, and Chapter V, which deals with solar-driven water-splitting cells, and the third part includes Chapter VI, Chapter VII, and Chapter VIII, which deals with solar-driven CO₂ reduction cells.</p
Multiscale, Data-Driven and Nonlocal Modeling of Granular Materials
Get PDFGranular materials are ubiquitous in both nature and technology. They play a key role in many applications ranging from storing food and energy to building reusable habitats and soft robots. Yet, predicting the continuum mechanical response of granular materials continues to present extraordinary challenges, despite the apparently simple laws that govern particle-scale interactions. This is largely due to the complex history dependence arising from the continuous rearrangement of their internal structure, and the nonlocality emerging from their self-organization. There is clearly an urge to develop methods that adequately address these two aspects, while bridging the long-standing divide between the grain- and the continuum scale.
This dissertation introduces novel theoretical and computational approaches for behavior prediction in granular solids. To begin with, we develop a framework for investigating their incremental behavior from the perspective of plasticity theory. It relies on systematically probing, through level-set discrete element calculations, the response of granular assemblies from the same initial state to multiple directions is stress space. We then extract the state- and history-dependent elasticity and plastic flow, and investigate the evolution of pertinent internal variables. We specifically study assemblies of sand particles characterized by X-ray computed tomography, as well as morphologically simpler counterparts of the same systems. Naturally arising from this investigation is the concept of a granular genome. Next, inspired by the abundance of generated high-fidelity micromechanical data, we develop an alternative data-driven approach for behavior prediction. This new multiscale modeling paradigm completely bypasses the need to define a constitutive law. Instead, the problem is directly formulated on a material data set, generated by grain-scale calculations, while pertinent constraints and conservation laws are enforced. We particularly focus on the sampling of the mechanical phase space, and develop two methods for parametrizing material history, one thermodynamically motivated and one statistically inspired. In the remainder of the thesis, we direct our attention to the understanding and modeling of nonlocality. We base our investigation on data derived from a discrete element simulation of a sample of sand subjected to triaxial compression and undergoing shear banding. By representing the granular system as a complex network, we study the self-organized and cooperative evolution of topology, kinematics and kinetics within the shear band. We specifically characterize the evolution of fundamental topological structures called force cycles, and propose a novel order parameter for the system, the minimal cycle coefficient. We find that this coefficient governs the stability of force chains, which succumb to buckling as they grow beyond a characteristic maximum length. We also analyze the statistics of nonaffine kinematics, which involve rotational and vortical particle motion. Finally, inspired by these findings, we extend the previously introduced data-driven paradigm to include nonaffine kinematics within a weakly nonlocal micropolar continuum description. By formulating the problem on a phase space augmented by higher-order kinematics and their conjugate kinetics, we bypass for the first time the need to define an internal length scale, which is instead discovered from the data. By carrying out a data-driven prediction of shear banding, we find that this nonlocal extension of the framework resolves the ill-posedness inherent to the classical continuum description. Finally, by comparing with available experimental data on the same problem, we are able to validate our theoretical developments.</p