50107 research outputs found
Sort by
Data for "Impacts of lithium brine mining on groundwater-dependent ecosystems in a multi-producer basin"
No full textThis dataset accompanies the manuscript "Impacts of lithium brine mining on groundwater-dependent ecosystems in a multi-producer basin".
Global lithium demand under the electromobility transition is driving the rapid expansion of lithium mining. Most new exploration will occur in closed-basin lithium brine deposits, where lithium mining consumes both lithium-rich brine and fresh water. However, closed-basin lithium brines primarily occur in water-scarce, endorheic basins, where new groundwater abstraction could strain water resources. Environmental impact assessments of mining operations in these basins typically evaluate only a single producer and overlook the cumulative impacts of multiple water users. Furthermore, simple water footprint calculations do not capture the complex dynamics between dense brine and fresh water, requiring variable-density modelling[sm1.1][DC1.2] to reasonably predict and evaluate the impacts of groundwater abstraction in these basins. This study is the first to utilize a three-dimensional, variable-density groundwater-flow model to simulate the effects of groundwater abstraction by multiple lithium producers on groundwater-dependent ecosystems, providing a framework for basin-scale environmental impact assessments of groundwater abstraction that is transferable to over 100 lithium brine resources identified globally. This model of the Salar del Hombre Muerto eastern subbasin simulates the individual and combined effects of three producers on surface and groundwater flows to groundwater-dependent ecosystems. The simulations demonstrate that distribution of abstraction within the subbasin and well placement control environmental flow reductions and that conversion from evaporative techniques to direct lithium extraction technology with brine reinjection may have the potential to mitigate environmental flow impacts. These results indicate that direct lithium extraction may be more effective than evaporative techniques at reducing abstraction impacts beneath groundwater-dependent ecosystems.BMW Group, BAS
Advancing AI Factuality via Comprehensive Evaluation
No full textThe rapid advancement of AI models in natural language generation often outpaces the development of reliable evaluation metrics, making it difficult to capture nuances in per- formance across tasks such as long-form question answering (LFQA), machine translation, and instruction following. This thesis develops scalable, accurate tools and benchmarks for factuality assessment that set higher standards for AI evaluation.
We begin by identifying key limitations in current evaluation practices through a focused analysis of long-form question answering. We first collect expert annotations on LFQA answers across seven domains (e.g., history, economics, biology). We carefully design the annotation setup and instructions, and conduct a rigorous screening process to recruit qualified domain experts. We then compare these expert annotations to crowd-sourced judgments and automatic metrics. We find that both crowd workers and existing metrics often fail to detect factual errors that experts reliably identify. On the human evaluation side, experts offer more accurate assessments of factuality and completeness, while crowd workers tend to favor superficial qualities like conciseness. On the automatic metric side, while no automatic metric consistently aligns with human judgments of overall quality, some show strengths in evaluating specific dimensions such as coherence.
Building on these insights, we introduce VERISCORE, a general-purpose factuality metric designed for long-form model generations. VERISCORE combines enhanced claim extraction, web-based evidence retrieval, and verification judgments to provide accurate and efficient assessments of factual consistency. Our results show that VERISCORE effectively distinguishes between verifiable and unverifiable claims and is preferred by human annota- tors in 93% of cases compared to popular alternatives FACTSCORE and SAFE. Additionally, it produces verification judgments that closely aligns with GPT-4 outputs, as confirmed through human evaluation.
While VERISCORE focuses on evaluating static model outputs, a natural next step is to assess factuality in more interactive settings. With the advent of versatile web agents, an important question arises: can these agents reliably retrieve factual information? To investigate this, we introduce BEARCUBS, a new benchmark designed to evaluate AI agents’ ability to identify factual information in open-ended, real-world, and multimodal settings, such as interacting with live web content and navigating complex visual tasks. BEARCUBS uncovers a significant performance gap between humans, who achieve 85% accuracy, and earlier state-of-the-art agents, which achieve only 24%. The ChatGPT Agent released in July 2025 marks substantial progress, reaching 66% accuracy, yet still falling short of human performance. These results highlight the need for new directions in agent evaluation, including improving the interpretability of agent trajectories and enhancing assessments of source credibility.
As AI models grow more capable, VERISCORE and BEARCUBS are becoming less discriminative, as the former is most effective on Wikipedia-style claims and the latter requires only linear web operations. To keep pace with AI progress, the development of more challenging benchmarks is essential. As part of this thesis, a fact verification dataset is developed that moves beyond the Wikipedia-centric scope of prior efforts, using history and politics as a case study. Claims are derived from historical and political non-fiction books, requiring multi-step retrieval and synthesis of dispersed evidence. State- of-the-art systems, including OpenAI Deep Research (63.89% accuracy on a three-way classification), fall short of reliable fact-checking under these conditions. This highlights substantial gaps in current factuality capabilities. The dataset is scalable and serves as both a benchmark for complex, non-Wikipedia factuality and a resource for training more capable fact-checkers.
Finally, this thesis concludes by proposing directions for advancing factuality evaluation, including the development of scalable benchmarks, improved alignment between human and automatic judgments, and training strategies for models that can reliably operate in high-stakes, open-domain settings.Doctor of Philosophy (Ph.D.
Quantitative Multiplexed Imaging of Nanomaterial Biological Distributions Using Laser Ablation Inductively Coupled Plasma Mass Spectrometry
No full textNanomaterial-based delivery systems have been increasingly employed in nanomedicine to improve targeting and enhance therapeutic efficacy. Understanding how nanomaterials behave in vivo is essential for optimizing the delivery of therapeutic cargos or drugs. This dissertation utilizes laser ablation inductively coupled plasma mass spectrometry imaging (LA-ICP-MSI) to investigate the fate of nanomaterials in biological systems. LA-ICP-MSI offers high sensitivity, multiplexing capability, site-specific information, and absolute quantification of nanomaterials in vivo. However, the technique faces limitations such as the lack of matrix-matched standards, signal drift caused by instrumental fluctuations, and relatively low spatial resolution.
To overcome these limitations, a tissue-mimicking approach was developed to better replicate the properties of biological tissues. When combined with an internal standard spiked into gelatin, this method enables more accurate absolute quantification of nanomaterials in biological samples. For instance, it was used to analyze the distribution of gold nanoparticles (AuNPs) in spleen tissue at one and six days after intravenous administration, revealing their interaction with the immune system. The method was also applied to evaluate the colocalization of nanozyme components, including AuNPs and a palladium catalyst, in tumors. Spatial mapping confirmed that both components remained intact in vivo. This dissertation also demonstrates the multiplexed tracking of proteins delivered by polymeric nanocarriers. By integrating metal-coded mass tags (MMTs) with ICP-MS and LA-ICP-MSI, this approach enables the quantification and spatial localization of multiple proteins delivered by polymers in a multiplexed fashion. This capability allows for the identification and optimization of nano-delivery systems better suited for targeting specific cells or disease sites, while minimizing off-target effects. Furthermore, improving the spatial resolution of LA-ICP-MSI is shown to be critical for investigating nanomaterial distribution within sub-organ regions. Computational methods were employed to enhance spatial resolution to approximately 5 µm, allowing for detailed assessment of nanomaterials in small anatomical structures. This capability was demonstrated by tracking AuNPs in sub-organ regions of the spleen and liver.
In summary, this dissertation highlights the value of LA-ICP-MSI in studying the in vivo behavior of nanomaterials. The insights gained from this work are valuable for advancing drug delivery research and developing more effective nanomedicine strategies.Doctor of Philosophy (Ph.D.)2026-09-0
JUST DECARBONIZATION IN THE US POWER SECTOR: A SYSTEMS ANALYSIS OF CARBON CAPTURE, AIR POLLUTANTS, AND POLICY DESIGN
Get PDFThis dissertation critically evaluates carbon capture (CC) as a decarbonization strategy for the US power sector, with particular attention to its implications for air pollution and environmental justice. While CC is often promoted as a climate solution capable of reducing up to 90% of CO₂ emissions from fossil-based electricity generation, its broader environmental and equity impacts remain insufficiently understood. This research explores how CC deployment affects power system operations, alters co-pollutant emissions, and reshapes the distribution of health damages. Although the analysis focuses on the US, the modeling framework and policy evaluation methods are adaptable to other jurisdictions.This work was supported by the ELEVATE program at the UMass Amherst Energy Transition Institute, which is funded by the National Science Foundation through the Research Traineeship (NRT) program (Award #2021693) and the Growing Convergence Research (GCR) program (Award #2020888). Additional support was provided by the Alfred P. Sloan Foundation (Grant #G-2021-14150), the UMass Amherst Spaulding-Smith Fellowship, and the Edwin V. Sisson Doctoral Fellowship.Doctor of Philosophy (Ph.D.
Analysis of the Rarefied Flow at Micro-Step using a DeepONet Surrogate Model with a Physics-Guided Zonal Loss Function
No full textRarefied gas flow over a micro backward-facing step is a canonical non-equilibrium benchmark featuring separation, recirculation and strong Knudsen-layer effects that are highly sensitive to both the Knudsen number and the step-height ratio. High-fidelity Direct Simulation Monte Carlo (DSMC) simulations resolve these phenomena but are prohibitively expensive for parametric studies, uncertainty quantification and design exploration. In this work, we develop a Deep Operator Network (DeepONet) surrogate for rarefied step flows that maps the Knudsen number Kn and the geometric ratio ℎ/ to the full two-dimensional velocity field in a micro-step geometry. The architecture is augmented with a physics-guided zonal loss that assigns higher weights to errors in the recirculation region ( < 0), thereby enforcing accurate prediction of separation and reattachment as well as the associated wall-shear and pressure distributions. Systematic comparisons with DSMC data in the slip and early transition regimes show that the
surrogate reproduces key physical trends, including the shortening and eventual disappearance of the separation bubble with increasing Kn and the non-monotonic variation of the reattachment length with ℎ/. A low-data study demonstrates that the model attains more than 90% of its asymptotic accuracy using only about 40–50% of the available high-fidelity simulations, substantially mitigating the cost of data generation. Furthermore, stochastic weight averaging Gaussian (SWAG) provides epistemic uncertainty estimates that are naturally localized near the shear layer and separation point, i.e. in the most non-equilibrium regions of the flow. The resulting framework offers a fast, robust and data-efficient tool for exploring rarefied micro-step flows, enabling many-query analyses in regimes where direct DSMC sampling is computationally intractable
Impact of micro-nano bubbles on the efficiency of ozone-based washing processes for fresh produce
No full textMicrobial contamination of fresh produce has emerged as a critical food safety challenge, as global consumption of raw and minimally processed fruits and vegetables continues to rise while conventional sanitizers show limited antimicrobial efficacy. Ozone has gained attention as a promising sanitizer due to its strong oxidizing properties and rapid decomposition into oxygen, leaving minimal harmful residues. However, its application in produce washing processes remains limited due to low solubility in water, high reactivity with organic matter, and slow gas-to-liquid transfer rates. Moreover, ozone’s instability and the need for on-site generation add operational complexity, while its efficacy can be reduced in the presence of high organic loads.
To address these limitations, micro-nano bubble technology has been increasingly explored as a means to enhance ozone delivery and efficacy in washing processes. Micro-nano bubbles exhibit a high surface area-to-volume ratio, low buoyancy, and prolonged stability, which enhance mass transfer rates and gas utilization efficiency. While these characteristics facilitate more effective dispersion of ozone in water and prolong its oxidizing activity, the combination of micro-nano bubbles and ozone has not been systematically explored as a washing strategy for fresh produce. In particular, the effects of processing parameters on the physicochemical properties and antimicrobial efficacy of ozone micro-nano bubbles remain poorly understood. Additionally, the impact of ozone micro-nano bubble treatment on the quality and sensory attributes of fresh produce has not been well characterized.
This dissertation aims to address these knowledge gaps through a series of interlinked studies. First, a meta-analysis was conducted to identify key processing parameters influencing the antimicrobial efficacy of ozone washing treatments. The results highlighted that sparging is the most effective method for microbial reduction, but also revealed gaps in the literature, including limited data on pH conditions and a lack of evaluations under dynamic and continuous systems that mimic industrial operations. Next, the influence of micro-nano bubbles on the antimicrobial efficacy of ozone-based washing was evaluated through comparison with conventional ozone water. The results demonstrated that micro-nano bubbles achieved higher microbial reduction compared to conventional ozone water, particularly after 10 minutes of treatment. The enhanced antimicrobial efficacy is attributed to the increased residual ozone concentration in micro-nano bubble water, resulting from reduced ozone decomposition and improved ozone stability. Given that fresh produce washing in industrial settings typically occurs within a short time frame of 1 to 5 minutes, the next study investigated strategies to improve microbial reduction within a 1-minute treatment time. In particular, the effect of solution pH was evaluated, as it can alter the ionic environment and thereby influence interactions between ozone micro-nano bubbles and bacterial cells. The findings showed that acidic conditions enhanced electrostatic interactions, leading to improved microbial inactivation within a 1-minute treatment time. Additionally, the study evaluated the effects of ozone micro-nano bubble treatment on post-wash quality and sensory attributes of romaine lettuce during storage to ensure the preservation of product quality. Finally, a novel bench-scale flume washer was developed to simulate industrial side-stream ozonation under laboratory conditions. The system incorporated rotating impellers and custom-designed 3D-printed cages to replicate produce movement through turbulent sanitizer flow while minimizing shear stress and maintaining consistent relative velocity between produce and fluid. Results demonstrated that the batch system achieved significantly higher microbial reduction than the continuous system without a side stream at an ozone concentration of 3 ppm. This was because ozone decomposed more rapidly in the continuous system due to enhanced mixing and turbulence caused by the rod rotation. Notably, by introducing a side-stream flow corresponding to 10% of the main flow rate, the continuous system achieved equivalent antimicrobial efficacy with only 1 ppm ozone concentration, compared to 3 ppm required in the batch system. This finding highlights the potential of ozone micro-nano bubble technology to improve antimicrobial efficacy while reducing ozone usage, thereby enhancing the sustainability of produce sanitation processes.
Overall, this dissertation evaluated the impact of micro-nano bubbles on the efficiency of ozone-based washing processes for fresh produce to enhance microbial safety and maintain product quality during storage. The effects of processing conditions on antimicrobial efficacy were investigated in a batch system. To reflect industrial conditions, a bench-scale flume washer was developed to simulate side-stream ozonation under dynamic flow. Ozone micro-nano bubbles in this system achieved similar microbial reduction with lower ozone input, improving both process sustainability and safety by reducing ozone exposure. This research demonstrates the application of ozone micro-nano bubble technology as a practical, effective, and safer solution for fresh produce washing. These findings will advance the development of ozone-based washing processes and help bridge the gap between laboratory research and industrial application.This work is supported by the United States Department of Agriculture National Institute of Food and Agriculture, AFRI project 2020-03324.Doctor of Philosophy (Ph.D.)2026-09-0
Immunomodulatory Lipid Nanoparticles for “Prime-Pull” Cancer Vaccination
No full textDespite major advances in cancer immunotherapy, many solid tumors remain refractory to treatment due to inadequate immune priming and poor immune cell infiltration into the tumor microenvironment. This dissertation presents a “prime-pull” cancer vaccination strategy leveraging multifunctional lipid nanoparticles (NPs) to address these critical barriers. The prime phase targets lymph nodes via subcutaneous administration to deliver tumor antigens and synergistic innate immune agonists, resulting in robust activation of tumor-specific T and B cell responses. The pull phase systemically reprograms the tumor microenvironment to enhance recruitment, infiltration, and activation of effector immune cells at the tumor site.NIH, UMass Institute for Applied Life ScienceDoctor of Philosophy (Ph.D.)2026-09-0
Iran’s Great Scientific Divergence: Counterfactual Evidence for the Long-Term Shock of the 1979 Revolution
No full textThis study quantifies the long-term impact of the 1979 Iranian Revolution on Iran’s trajectory of scientific publications, a critical examination for science and public policy. Using comprehensive data from 1960 to 2024, we benchmark Iran against pre-revolutionary peers (e.g., South Korea) and employ multidimensional scientometrics, including the rigorous Synthetic Control Method (SCM). Results demonstrate a significant divergence: Iran, which led its peers in 1978, experienced collapse and stagnation (1980–1999) while peers grew exponentially. SCM, which optimally matches pre-1979 Iran with South Korea, quantifies a cumulative knowledge deficit of approximately 551,000 publications by 2024. Furthermore, despite a post-2000 volume recovery, a persistent quality gap exists; research impact (measured by Field-Weighted Citation Impact) consistently lags behind the global average. This analysis provides a robust, data-driven quantification of the generational opportunity cost of the 1979 disruption on national scientific development
Resource allocation in quantum networks
No full textQuantum computing promises computational capabilities beyond those of classical systems by exploiting superposition, entanglement, and quantum interference. However, the scalability of current noisy intermediate-scale quantum (NISQ) devices is limited by physical and engineering constraints, restricting the number of available qubits. Distributed quantum computing (DQC) mitigates this limitation by interconnecting multiple quantum processing units (QPUs) via quantum networks capable of distributing high-fidelity entanglement. Realizing such networks requires overcoming probabilistic entanglement generation, finite memory coherence, photon loss, and architectural constraints.
This dissertation conducts an in-depth and comprehensive study of resource allocation in quantum networks, aiming to support high-throughput and high-fidelity entanglement distribution over lossy infrastructures. Our work spans the quantum networking stack, from physical-layer entanglement distribution protocols to network planning, quantum storage architectures, and distributed quantum computing.
We begin by introducing two asynchronous entanglement distribution protocols-Sequential and Parallel-designed for early-stage quantum networks without global synchronization. Unlike many prior models, our protocols account for classical communication delays and quantum memory decoherence, reflecting practical constraints. Through simulations on real-world topologies such as SURFnet, we show that simple mechanisms such as cutoff-based memory management and sequential asynchronous entanglement distribution protocols can deliver competitive performance while significantly simplifying coordination, narrowing the gap between theoretical proposals and deployable protocols.
Next, we formulate quantum network planning as a utility maximization problem that jointly considers entanglement distribution rate and fidelity, treating repeater and memory placement as optimization variables. We propose scalable heuristics and mixed-integer formulations, evaluating them under realistic conditions such as finite coherence time of quantum memories, limited number of memories per repeaters and end nodes, and real network topologies such as ESNet. Our findings offer practical design guidance and highlight how optimal placements depend on underlying physical constraints.
To support dynamic workloads and respond to traffic bursts, we propose the Quantum Storage Network (QSN) architecture, where storage nodes buffer EPR pairs in advance of demand as an overlay network. This temporal decoupling improves responsiveness and throughput. Simulations across multiple topologies show that QSNs reduce entanglement distribution service latency by up to 40% and provide improved robustness compared to non-overlay designs, underscoring the value of storage-aware planning in future quantum networks.
Finally, we focus on DQC, where large-scale quantum circuits are executed across networked QPUs. We propose a window-based circuit partitioning algorithm that minimizes nonlocal gate execution while balancing load across QPUs. Building on this partitioning, we design an adaptive, dependency-aware entanglement generation scheduler that opportunistically initiates remote entanglement as soon as dependent gates are ready, while accounting for realistic constraints such as coherence limits, probabilistic link generation, and Bell-state measurement (BSM) contention. Applied to benchmark circuits such as QAOA and QFT, this joint approach reduces execution delay by up to 35%. We further perform a two-qubit gate fidelity-sensitivity analysis under to show when the distributed quantum computing can be beneficial. Results show that the number of QPUs significantly impacts sensitivity to non-local gate fidelity: with fewer QPUs, distributed execution can outperform monolithic execution under looser fidelity constraints, whereas with more QPUs, achieving the same advantage requires substantially higher non-local gate fidelities relative to local gate fidelities. These findings identify the parameter regimes in which distributed quantum computing is beneficial, mapping break-even regions (as a function of number of QPUs) and quantifying how much cross-QPU (non-local) gate fidelity must improve—or, equivalently, how much per-QPU local gate fidelity must improve under modularization—for distributed execution to outperform a monolithic baseline.
Together, these contributions form a cohesive framework for resource allocation in quantum networks in order to build scalable, delay-tolerant quantum networks and distributed computing systems. By grounding architectural design in realistic physical assumptions, this work advances the state of the art from theoretical modeling to practical, deployable quantum infrastructure.Doctor of Philosophy (Ph.D.
Moving from bepress to DSpace - A Migration Story
No full textThe University of Massachusetts Amherst Libraries migrated ScholarWorks@UMassAmherst from bepress’ Digital Commons to Atmire-hosted DSpace (7.x) over the course of the 2023-2024 academic year.
This presentation will cover the development and implementation of our migration plan as well as our process for fully implementing our new DSpace instance.
From blocking Google requests, identifying items that shouldn’t have migrated, lifting preemptively placed embargoes, to cleaning up bad date formats, discovering that previously suppressed metadata had been exposed, and managing expectations of eager submitters and users, we’ll present to you some lessons learned from the migration of ScholarWorks@UMassAmherst from Digital Commons to DSpace