150813 research outputs found
Sort by
Dataset Design for Building Models of Chemical Reactivity
No full textModels can codify our understanding of chemical reactivity and serve a useful purpose in the development of new synthetic processes via, for example, evaluating hypothetical reaction conditions or in silico substrate tolerance. Perhaps the most determining factor is the composition of the training data and whether it is sufficient to train a model that can make accurate predictions over the full domain of interest. Here, we discuss the design of reaction datasets in ways that are conducive to data-driven modeling, emphasizing the idea that training set diversity and model generalizability rely on the choice of molecular or reaction representation. We additionally discuss the experimental constraints associated with generating common types of chemistry datasets and how these considerations should influence dataset design and model building
Aggregating Funnels for Faster Fetch&Add and Queues
No full textPPoPP ’25, Las Vegas, NV, USAMany concurrent algorithms require processes to perform fetch-and-add operations on a single memory location, which can be a hot spot of contention. We present a novel algorithm called Aggregating Funnels that reduces this contention by spreading the fetch-and-add operations across multiple memory locations. It aggregates fetch-and-add operations into batches so that the batch can be performed by a single hardware fetch-and-add instruction on one location and all operations in the batch can efficiently compute their results by performing a fetch-and-add instruction on a different location. We show experimentally that this approach achieves higher throughput than previous combining techniques, such as Combining Funnels, and is substantially more scalable than applying hardware fetch-and-add instructions on a single memory location. We show that replacing the fetch-and-add instructions in the fastest state-of-the-art concurrent queue by our Aggregating Funnels eliminates a bottleneck and greatly improves the queue's overall throughput
A new family of high-current cyclotrons for isotope production
No full textWe are developing a high-current cyclotron as a driver for the IsoDAR neutrino experiment. It accelerates 5 mA of H2 + to 60 MeV/amu, after which the electron is removed to produce a 10 mA, 60 MeV proton beam. The enabling innovations that offset space-charge effects occur at injection and in the first few turns, allowing one to construct cyclotrons with energies ranging from below 5 MeV up to 60 MeV/amu, or possibly higher, with the same performance for accelerated ions with Q/A = 0.5 (H2+, D+, He++, …). In this paper, we discuss the possible uses of such cyclotrons for isotope production, including production of long-lived generator parents (68Ga, 44Ti, 82Sr,…), as well as intense fast neutron beams from deuteron breakup for (n,2n) production of isotopes like 225Ac
The cost of CO2 capture and storage
No full textThe objective of this paper is to assess the current costs of CO2 capture and storage (CCS) for new fossil fuel power plants and to compare those results to the costs reported a decade ago in the IPCC Special Report on Carbon Dioxide Capture and Storage (SRCCS). Toward that end, we employed a similar methodology based on review and analysis of recent cost studies for the major CCS options identified in the SRCCS, namely, post-combustion CO2 capture at supercritical pulverized coal (SCPC) and natural gas combined cycle (NGCC) power plants, plus pre-combustion capture at coal-based integrated gasification combined cycle (IGCC) power plants. We also report current costs for SCPC plants employing oxy-combustion for CO2 capture - an option that was still in the early stages of development at the time of the SRCCS. To compare current CCS cost estimates to those in the SRCCS, we adjust all costs to constant 2013 US dollars using cost indices for power plant capital costs, fuel costs and other O&M costs. On this basis, we report changes in capital cost, levelized cost of electricity, and mitigation costs for each power plant system with and without CCS. We also discuss the outlook for future CCS costs
Tracking carbon fluxes across ocean interfaces using dissolved gas observations
No full textThe cycling and exchange of carbon between Earth’s systems play a pivotal role in regulating climate, yet two major carbon fluxes remain poorly constrained: the biological carbon pump (BCP) and carbon release from Arctic permafrost. This thesis focuses on dissolved gases as tracers and drivers of these processes through both autonomous and field-based observations. It encompasses (i) improvements to sensor-based measurements of O₂, (ii) the use of these measurements to assess the strength of the BCP in two distinct export regimes, and (iii) isotopic approaches to carbon dioxide (CO₂) and methane (CH₄) dynamics at a coastal permafrost site. The first part of the thesis is centered around the NASA EXPORTS campaign and studies the BCP at two contrasting field sites. Using autonomous platforms, carbon export was evaluated at both sites and demonstrated that at the lower productivity site, a greater proportion of fixed carbon was routed to sinking particulate organic carbon (POC), while the higher productivity site resulted in near equal proportions of dissolved organic carbon production and sinking POC. These findings underscore the value of autonomous sensors in capturing spatial and temporal variability in oceanic carbon cycling. The second part of this thesis shifts focus to the Arctic, where rapid warming threatens to mobilize vast (~1,500 Pg) amounts of carbon currently stored in permafrost. This study presents observations from the spring thaw at a coastal Arctic site and demonstrated that even sites with high CH₄ and CO₂ concentrations drew less than 10% of their carbon source from ancient permafrost sources. The variability in CH₄ and CO₂ emissions reflects the complex interplay between hydrological changes, primary productivity, and microbial processes. The research highlights the need for regular monitoring of Arctic rivers, which integrate changes in the terrestrial system, as a potential early warning system for abrupt permafrost thaw. This thesis leverages the fundamentals of dissolved gas geochemistry to examine key climate-relevant biogeochemical cycles across diverse environments that are sensitive to global change. These insights contribute to refining Earth system models and emphasize the need for expanded monitoring to predict future shifts in global carbon cycling and climate dynamics.Ph.D
Near-Optimal Learning and Planning in Separated Latent MDPs
No full textWe study computational and statistical aspects of learning Latent Markov Decision Processes (LMDPs). In this model, the learner interacts with an MDP drawn at the beginning of each epoch from an unknown mixture of MDPs. To sidestep known impossibility results, we consider several notions of δ-separation of the constituent MDPs. The main thrust of this paper is in establishing a nearly-sharp statistical threshold for the horizon length necessary for efficient learning. On the computational side, we show that under a weaker assumption of separability under the optimal policy, there is a quasi-polynomial algorithm with time complexity scaling in terms of the statistical threshold. We further show a near-matching time complexity lower bound under the exponential time hypothesis.S.M
The Geometry of Concepts: Sparse Autoencoder Feature Structure
No full textSparse autoencoders have recently produced dictionaries of high-dimensional vectors corresponding to the universe of concepts represented by large language models. We find that this concept universe has interesting structure at three levels: (1) The “atomic” small-scale structure contains “crystals” whose faces are parallelograms or trapezoids, generalizing well-known examples such as (man:woman::king:queen). We find that the quality of such parallelograms and associated function vectors improves greatly when projecting out global distractor directions such as word length, which is efficiently performed with linear discriminant analysis. (2) The “brain” intermediate-scale structure has significant spatial modularity; for example, math and code features form a “lobe” akin to functional lobes seen in neural fMRI images. We quantify the spatial locality of these lobes with multiple metrics and find that clusters of co-occurring features, at coarse enough scale, also cluster together spatially far more than one would expect if feature geometry were random. (3) The “galaxy”-scale large-scale structure of the feature point cloud is not isotropic, but instead has a power law of eigenvalues with steepest slope in middle layers. We also quantify how the clustering entropy depends on the layer
The Archean origin of assimilatory sulfate metabolisms provides novel insight into redox conditions of early Earth environments
No full textDissimilatory sulfur metabolisms recording differing biological isotopic fractionation are well studied, important components of sulfur cycling (Mateos et al., 2023). Assimilatory sulfur metabolisms and genes across life provide a complementary window into sulfur biogeochemistry with individual pathways having specific isotopic fractionations acting on distinct redox states (e.g. sulfate, sulfide, sulfite) for anabolism (Liu et al., 2012). An assimilation pathway exists, which starts with sulfate adenylyltransferase (sat/ATP sulfurylase) catalyzing a reaction of adenosine triphosphate (ATP) and sulfate (SO42-) resulting in adenosine 5’-phosphosulfate (APS), and incorporation of more reduced sulfur into biomolecules. This sat/ATP sulfurylase enzyme represents the first step required by life to incorporate sulfate and informs our understanding of biological processes performing this fundamental chemical reaction. A phylogenetic and molecular clock analysis of the sat/ATP sulfurylase protein family (E.C. 2.7.7.4) was performed to determine the age of sulfate assimilation proteins. Extant diversity of sat proteins was estimated to have a last common ancestor ~3.24 Ga (95% CI 3.52–3.06 Ga) using relaxed molecular clocks calibrated with eukaryotic and cyanobacteria age ranges from previously published fossil calibrated investigations. These results suggest sulfate cycling in Paleoarchean environments, despite extensive evidence of low marine sulfate concentrations (Crowe & Canfield et al., 2014). Archean sulfate biogeochemical cycling could result from microbial sulfur oxidation and sources could include abiotic oxidation of volcanic sulfur, hydrothermal processes or pyrite (Canfield, 2001, Lyons et al., 2024). This phylogenomic evidence of sulfate during Archean times provides an independent complement to geochemical records and indicates that sulfur redox chemistry during the Archean was likely more complex than previously described.S.M
Decoupling Economic Growth and Carbon Emissions
No full textAll economic activity requires energy; to the extent this energy comes from fossil fuels, the energy use results in emissions of carbon dioxide, CO2. The nature of this link between the growth in economic activity and carbon emissions is a critical question for climate change.1 Linkage implies that deep emission reductions will constrain economic growth; decoupling implies that deep emission reductions are possible with little or no effect on growth. An answer to this question is important for the United States, but more crucial for rapidly growing emerging economies such as China and India that seek to improve their citizens' access to low-cost energy while respecting the need to protect the global environment
High Order Immersed Finite Difference Methods for Complex Domains with Moving Boundaries and Interfaces
No full textMoving domain boundaries and material interfaces are a hallmark of multiphysics systems such as fluid-structure interaction, alloy solidification, and multiphase flows. Simulating moving interfaces with traditional techniques requires a moving mesh that continuously adapts to the interface, which is costly and places restrictions on the interface motion. Immersed methods avoid these challenges by simulating moving geometries on a stationary Cartesian grid, locally altering the numerical method to account for boundaries and interfaces that are not grid-aligned. Most existing immersed methods have low-order spatial accuracy, requiring fine grids to generate accurate results. High order immersed methods can produce more accurate results at lower resolution, making them a promising tool for 3D simulations with tight error tolerances. However, the majority of available high order immersed methods have been numerical experiments developed for stationary 2D geometries and simple PDEs. In this thesis we demonstrate that high order immersed methods can be extended to complex nonlinear PDEs and moving 3D geometries, both of which are necessary to simulate practical engineering problems. We begin by introducing a boundary treatment that locally approximates PDE solutions with high order accuracy using a weighted least-squares fit, and show that the procedure remains valid for smooth 2D or 3D geometries satisfying a local curvature constraint. This boundary treatment is combined with a high order finite difference method to discretize the Poisson equation with up to sixth order accuracy. We then expand the scope of the method to include PDEs with immersed material interfaces, spatially-variable coefficients, vector-valued unknowns, cross-derivative terms, and nonlinearities. These techniques are applied to generate a sixth-order discretization of 2D nonlinear elasticity, demonstrating the applicability of high order immersed methods to complex PDE systems relevant in mechanical engineering. In the second half, we focus on large-scale 3D simulations with moving boundaries. We construct a third order immersed advection discretization with provable stability in one dimension, and show experimentally that the scheme remains stable in 2D and 3D domains. To treat moving boundaries, we introduce a general framework that allows high order immersed methods to maintain their accuracy in both space and time when paired with any explicit Runge-Kutta time integrator. We conclude by presenting results from massively-parallel high order simulations of the 3D advection-diffusion equation with moving boundaries on a multiresolution grid. Taken together, these results demonstrate that high order immersed methods can achieve the scale and complexity necessary to enable practical simulations that are difficult or impossible with traditional mesh-based techniques.Ph.D