Search NASA⌕ Search

SEARCH · Search NASA

Results for “parallel parameters optimization”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Prediction of DIII-D Pedestal Structure from Externally Controllable Parameters

The sharp increase of pressure at the edge of a high confinement mode (H-mode) plasma, the pedestal, strongly impacts overall plasma performance. Predicting the pedestal is a necessity to control and optimize tokamak operations. An experimental data-driven machine learning (ML) approach is presented that predicts the pedestal heights and widths of electron density (ne) and electron temperature (Te) profiles as well as the separatrix ne from externally controllable parameters such as the plasma shape, heating method and power, and gas puff rate and integrated gas puff. The OMFIT framework was used with DIII-D data to efficiently, robustly, and automatically build a database of pedestal parameters to train machine learning models. Database creation was enabled by the search engine tool for DIII-D data, TokSearch, which parallelizes data fetching, enabling fast searches through basic signals of thousands of DIII-D shots and selection of relevant time intervals. Principal Component Analysis (PCA) separated the database into three clusters that represent classes of plasma shapes that are regularly used in DIII-D. The most important parameters for setting the pedestal structure were plasma current (Ip), toroidal magnetic field (Bφ), neutral beam heating power (PNBI) and shaping quantities. The Deep Jointly Informed Neural Networks (DJINN) algorithm was applied to identify suitable neural network (NN) architectures that appropriately capture the features of the pedestal database. Separate NNs were implemented for each pedestal parameter, and ensembling methods were used to improve the prediction accuracy and allowed estimation of the prediction uncertainty. The pedestal predictions of the test dataset lie within the measurement uncertainties of the pedestal parameters. The NN outperformed simple Linear Regression (LR) analysis, indicating non-linear dependencies in the pedestal structure. The presented achievements illustrate a promising path for future research, using feature extraction to infer experimental trends and thereby improve pedestal models as well as deploying NN for a fast pedestal prediction in DIII-D scenario development.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2022

m-CUBES An efficient and portable implementation of multi-dimensional integration for gpus

The task of multi-dimensional numerical integration is frequently encountered in physics and other scientific fields, e.g., in modeling the effects of systematic uncertainties in physical systems and in Bayesian parameter estimation. Multi-dimensional integration is often time-prohibitive on CPUs. Efficient implementation on many-core architectures is challenging as the workload across the integration space cannot be predicted a priori. We propose m-Cubes, a novel implementation of the well-known Vegas algorithm for execution on GPUs. Vegas transforms integration variables followed by calculation of a Monte Carlo integral estimate using adaptive partitioning of the resulting space. m-Cubes improves performance on GPUs by maintaining relatively uniform workload across the processors. As a result, our optimized Cuda implementation for Nvidia GPUs outperforms parallelization approaches proposed in past literature. We further demonstrate the efficiency of m-Cubes by evaluating a six-dimensional integral from a cosmology application, achieving significant speedup and greater precision than the CUBA library's CPU implementation of VEGAS. We also evaluate m-Cubes on a standard integrand test suite. m-Cubes outperforms the serial implementations of the Cuba and GSL libraries by orders of magnitude speedup while maintaining comparable accuracy. Our approach yields a speedup of at least 10 when compared against publicly available Monte Carlo based GPU implementations. In summary, m-Cubes can solve integrals that are prohibitively expensive using standard libraries and custom implementations. A modern C++ interface header-only implementation makes m-Cubes portable, allowing its utilization in complicated pipelines with easy to define stateful integrals. Compatibility with non-Nvidia GPUs is achieved with our initial implementation of m-Cubes using the Kokkos framework.

Sakiotis, Ioannis↗
DOE OSTI2022

Toward fully automated UED operation using two-stage machine learning model

To demonstrate the feasibility of automating UED operation and diagnosing the machine performance in real time, a two-stage machine learning (ML) model based on self-consistent start-to-end simulations has been implemented. This model will not only provide the machine parameters with adequate precision, toward the full automation of the UED instrument, but also make real-time electron beam information available as single-shot nondestructive diagnostics. Furthermore, based on a deep understanding of the root connection between the electron beam properties and the features of Bragg-diffraction patterns, we have applied the hidden symmetry as model constraints, successfully improving the accuracy of energy spread prediction by a factor of five and making the beam divergence prediction two times faster. The capability enabled by the global optimization via ML provides us with better opportunities for discoveries using near-parallel, bright, and ultrafast electron beams for single-shot imaging. It also enables directly visualizing the dynamics of defects and nanostructured materials, which is impossible using present electron-beam technologies.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Prediction of DIII-D Pedestal Structure From Externally Controllable Parameters

The sharp increase of pressure at the edge of a high confinement mode (H-mode) plasma, the pedestal, strongly impacts overall plasma performance. Predicting the pedestal is a necessity to control and optimize tokamak operations. Here, an experimental data-driven machine learning (ML) approach is presented that predicts the pedestal heights and widths of electron density (n e ) and electron temperature (T e ) profiles as well as the separatrix ne from externally controllable parameters such as the plasma shape, heating method and power, and gas puff rate and integrated gas puff. The OMFIT framework was used with DIII-D data to efficiently, robustly, and automatically build a database of pedestal parameters to train machine learning models. Database creation was enabled by the search engine tool for DIII-D data, TokSearch, which parallelizes data fetching, enabling fast searches through basic signals of thousands of DIII-D shots and selection of relevant time intervals. Principal Component Analysis (PCA) separated the database into three clusters that represent classes of plasma shapes that are regularly used in DIII-D. The most important parameters for setting the pedestal structure were plasma current (I p ), toroidal magnetic field (B Φ ), neutral beam heating power (P NBI ) and shaping quantities. The Deep Jointly Informed Neural Networks (DJINN) algorithm was applied to identify suitable neural network (NN) architectures that appropriately capture the features of the pedestal database. Separate NNs were implemented for each pedestal parameter, and ensembling methods were used to improve the prediction accuracy and allowed estimation of the prediction uncertainty. The pedestal predictions of the test dataset lie within the measurement uncertainties of the pedestal parameters. The NN outperformed simple Linear Regression (LR) analysis, indicating non-linear dependencies in the pedestal structure. The presented achievements illustrate a promising path for future research, using feature extraction to infer experimental trends and thereby improve pedestal models as well as deploying NN for a fast pedestal prediction in DIII-D scenario development.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2021

Instability Issue of Paralleled Dies in an SiC Power Module in Solid-State Circuit Breaker Applications

Paralleled dies in a power module could have instability issues during high current switching transients. Here, the instability is caused by the differential-mode oscillation among paralleled MOSFETs. Conventional analyses of paralleled MOSFETs’ stability are normally limited to a single operating point, which ignores the influences of the switching trajectory and nonlinear device parameters on stability. This article reveals that the switching trajectory can significantly influence parallel stability. The analysis is improved by solving eigenvalues of state-space modeling system matrices of all operating points that the switching trajectory goes through considering nonlinear device parameters. Higher voltage and current stresses result in greater real parts of complex eigenvalues, which explains why the paralleled MOSFETs are more unstable with higher voltage and current stresses. To improve stability in solid-state circuit breaker applications, we propose a method to manipulate the switching trajectory to avoid the unstable region where the conventional hard switching trajectory normally goes through. Experimental results show that the turn- off current capability can be increased from ~five times of rated current with the gate oscillation using the conventional turn- off trajectory to ~ten times of rated current without the gate oscillation using the optimal turn- off trajectory.

42 ENGINEERING↗
DOE OSTI2017

Effect of 3D magnetic perturbations on divertor conditions and detachment in tokamak and stellarator

Enhanced perpendicular heat and momentum transport induces parallel pressure loss leading to divertor detachment, which can be produced by the increase of density in 2D tokamaks. However, in the 3D configurations such as tokamaks with 3D fields and stellarators, the fraction of perpendicular transport can be higher even in a lower density regime, which could lead to the early transition to detachment without passing through the high-recycling regime. 3D fields applied to the limiter tokamak plasmas produce edge stochastic layers close to the last closed flux surface (LCFS), which can allow for enhanced perpendicular transport and indeed the absence of high recycling regime and early detachment have been observed in TEXTOR and Tore Supra. However, in the X-point divertor tokamaks with the applied 3D fields, the parallel transport is still dominant and the detachment facilitation has not been observed yet. Rather, 3D fields affected detachment adversely under certain conditions, either by preventing detachment onset as seen in DIII-D or by re-attaching the existing detached plasma as shown in NSTX. The possible way for strong 3D effects to induce access to the early detachment in divertor tokamaks appears to be via significant perpendicular loss of parallel momentum by frictional force for the counter-streaming flows between neighboring flow channels in the divertor. In principle, the adjacent lobes in the 3D divertor tokamak may generate the counter-streaming flow channels. However, an EMC3-EIRENE simulation for ITER H-mode plasmas demonstrated that screened RMP leads to significantly reduced counter-flows near the divertor target, therefore the momentum loss effect leading to detachment facilitation is expected to be small. This is consistent with the observation in LHD, which showed screening (amplification) of RMP fields in the attachment (stable detachment) case. In conclusion, work for optimal parameter window for best divertor operation scenario is needed particularly for the 3D divertor tokamak configuration.

3D effects on divertor↗
DOE OSTI2025

MoE-Inference-Bench: Performance Evaluation of Mixture of Expert Large Language and Vision Models

Mixture of Experts (MoE) models have enabled the scaling of Large Language Models (LLMs) and Vision Language Models (VLMs) by achieving massive parameter counts while maintaining computational efficiency. However, MoEs introduce several inference-time challenges, including load imbalance across experts and the additional routing computational overhead. To address these challenges and fully harness the benefits of MoE, a systematic evaluation of hardware acceleration techniques is essential. We present MoE-Inference-Bench, a comprehensive study to evaluate MoE performance across diverse scenarios. We analyze the impact of batch size, sequence length, and critical MoE hyperparameters such as FFN dimensions and number of experts on throughput. We evaluate several optimization techniques on Nvidia H100 GPUs, including pruning, Fused MoE operations, speculative decoding, quantization, and various parallelization strategies. Our evaluation includes MoEs from the Mixtral, DeepSeek, OLMoE and Qwen families. The results reveal performance differences across configurations and provide insights for the efficient deployment of MoEs.

Chitty-Venkata, Krishna Teja↗
DOE OSTI2021

Understanding the Design Space of Sparse/Dense Multiphase Dataflows for Mapping Graph Neural Networks on Spatial Accelerators

Graph Neural Networks (GNNs) have garnered a lot of recent interest because of their success in learning representations from graph-structured data across several critical applications in cloud and HPC. Owing to their unique compute and memory characteristics that come from an interplay between dense and sparse phases of computations, the emergence of reconfigurable dataflow (aka spatial) accelerators offers promise for acceleration by mapping optimized dataflows (i.e., computation order and parallelism) for both phases. The goal of this work is to characterize and understand the design-space of dataflow choices for running GNNs on spatial accelerators in order for the compilers to optimize the dataflow based on the workload. Specifically, we propose a taxonomy to describe all possible choices for mapping the dense and sparse phases of GNNs spatially and temporally over a spatial accelerator, capturing both the intra-phase dataflow and the inter-phase (pipelined) dataflow. Using this taxonomy, we do deep-dives into the cost and benefits of several dataflows and perform case studies on implications of hardware parameters for dataflows and value of flexibility to support pipelined execution.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2018

AN INNOVATIVE Nb3Sn FILM APPROACH AND ITS POTENTIAL FOR SRF APPLICATIONS

A novel electro-chemical technique to produce Nb3Sn films on Nb substrates was developed and optimized at Fermilab. The Nb3Sn phase is obtained in a two-electrode cell, by electrodeposition from aqueous solutions of Sn lay-ers and Cu intermediate layers onto Nb substrates. Subse-quent thermal treatments in inert atmosphere are realized at a maximum temperature of 700°C to obtain the Nb3Sn superconducting phase. Several superconducting Nb3Sn films were obtained on Nb substrates by studying and op-timizing most parameters of the electro-plating process. Samples were characterized at Fermilab, NIMS, KEK and JLab, including SEM/EPMA analyses, DC and inductive tests of critical temperature Tc0, and lower critical field Hc1(4.2 K) by SQUID. In parallel to sample development and fabrication at FNAL, at JLab and KEK effort was put into etching and electro-polishing techniques adequate to remove the Cu and bronze phases from the samples’ outer surface. This is necessary prior to measurements at JLab of the surface impedance of flat samples in a setup that make use of an RF host cavity.

43 PARTICLE ACCELERATORS↗
DOE OSTI2020

Parallel-in-Time Simulation of Lindblad's Equation

Constructing fast quantum logic gates is critical to building a scalable quantum computer. We consider a qudit, a quantum version of a bit that can take an arbitrary number of states, coupled with a cavity. In this project, we wish to force the qudit to reach the 0-state, for any possible initial state. The coupled system changes in time according to Lindblad’s equation, an ordinary differential equation on the density matrix of the quantum system. Lindblad’s equation contains some parameters that we can control, so-called control functions. We seek control functions which force the qudit to the 0-state within 2 microseconds, which is much faster than what is currently done in practice. The search method is gradient descent, a numerical optimization method that uses gradient information to iteratively improve the control parameters. My contribution to this project is an attempt to speed up the computation of the gradient. It currently takes about 40 seconds to compute the gradient which involves solving a set of ODEs sequentially. Current supercomputers have thousands of cores, but sequential computations can only make use of 1 core at a time. We wish to divide up the work better, so that we can use many more cores at once. To this end, we have implemented the Multigrid Reduction in Time (MGRIT) algorithm. We perform a systematic parameter search on how to best apply this algorithm. Results indicate a 25 percent speed up for solving Lindblad’s equation and determining how close the final state is the 0-state.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2019

Parallel beamlet dose calculation via beamlet contexts in a distributed multi‐GPU framework

Purpose Dose calculation is one of the most computationally intensive, yet essential tasks in the treatment planning process. With the recent interest in automatic beam orientation and arc trajectory optimization techniques, there is a great need for more efficient model‐based dose calculation algorithms that can accommodate hundreds to thousands of beam candidates at once. Foundational work has shown the translation of dose calculation algorithms to graphical processing units ( GPU s), lending to remarkable gains in processing efficiency. But these methods provide parallelization of dose for only a single beamlet, serializing the calculation of multiple beamlets and under‐utilizing the potential of modern GPU s. In this paper, the authors propose a framework enabling parallel computation of many beamlet doses using a novel beamlet context transformation and further embed this approach in a scalable network of multi‐ GPU computational nodes. Methods The proposed context‐based transformation separates beamlet‐local density and TERMA into distinct beamlet contexts that independently provide sufficient data for beamlet dose calculation. Beamlet contexts are arranged in a composite context array with dosimetric isolation, and the context array is subjected to a GPU collapsed‐cone convolution superposition procedure, producing the set of beamlet‐specific dose distributions in a single pass. Dose from each context is converted to a sparse representation for efficient storage and retrieval during treatment plan optimization. The context radius is a new parameter permitting flexibility between the speed and fidelity of the dose calculation process. A distributed manager‐worker architecture is constructed around the context‐based GPU dose calculation approach supporting an arbitrary number of worker nodes and resident GPU s. Phantom experiments were executed to verify the accuracy of the context‐based approach compared to Monte Carlo and a reference CPU ‐ CCCS implementation for single beamlets and broad beams composed by addition of beamlets. Dose for representative 4 π beam sets was calculated in lung and prostate cases to compare its efficiency with that of an existing beamlet‐sequential GPU ‐ CCCS implementation. Code profiling was also performed to evaluate the scalability of the framework across many networked GPU s. Results The dosimetric accuracy of the context‐based method displays <1.35% and 2.35% average error from the existing serialized CPU ‐ CCCS algorithm and Monte Carlo simulation for beamlet‐specific PDD s in water and slab phantoms, respectively. The context‐based method demonstrates substantial speedup of up to two orders of magnitude over the beamlet‐sequential GPU ‐ CCCS method in the tested configurations. The context‐based framework demonstrates near linear scaling in the number of distributed compute nodes and GPU s employed, indicating that it is flexible enough to meet the performance requirements of most users by simply increasing the hardware utilization. Conclusions The context‐based approach demonstrates a new expectation of performance for beamlet‐based dose calculation methods. This approach has been successful in accelerating the dose calculation process for very large‐scale treatment planning problems ‐ such as automatic 4 π IMRT beam orientation and VMAT arc trajectory selection, with hundreds of thousands of beamlets ‐ in clinically feasible timeframes. The flexibility of this framework makes it as a strong candidate for use in a variety of other very large‐scale treatment planning tasks and clinical workflows.

Neph, Ryan↗
DOE OSTI2020

Tracking Radioactive Isotopes in HVAC and Application for Hot Cell Analyses - 20155

In recent years, the surge in the number of isotope production facilities under design has increased the need to analyze isotope migration in facility Heating, Ventilation, and Air Conditioning (HVAC) during normal operations and accident scenarios. These facilities are used to produce isotopes for medical, security, and industrial applications and as such are subject to license and regulatory requirements 10CFR20, 10CFR30, 10CFR50, and 10CFR70. GOTHIC, a general-purpose thermal-hydraulics software package, includes the ability to model isotopic tracers, radioactive decay and isotope migration as well as HEPA and charcoal filters for isotope retention. A GOTHIC model was developed by Zachry Nuclear Engineering (ZNE) to examine the effect of negative room pressure, HEPA filtration, and HVAC fluctuations on radiation areas, hot cells and gloveboxes. Radioactive tracers were used to simulate the concentration of spills within contaminated areas, track the migration of isotopes of interest, and determine the isotopic retention and buildup on facility HEPA filters. A variety of isotopes with concern to dose (e.g., Kr-85, Sr-90, I-131, etc.), including their decay and progeny, are included in the analysis. Negative pressures are maintained in the regions of interest by a representative central HVAC system equipped with a volumetric fan that exhausts to the environment after a series of isolation valves and HEPA filters. GOTHIC is an industry trusted tool for providing engineering solutions for a variety of applications, including fission product tracking, aerosol and particulate transport and ventilation assessments. The software provides an integrated analysis environment that includes a graphical user interface (GUI) for constructing analysis models, a numerical solver that includes parallel processing capabilities and a post-processor for evaluating simulation results. It solves the conservation equations for mass, momentum and energy for multicomponent, multi-phase flow in lumped parameter and multi-dimensional geometries (1, 2, or full 3D), including the effects of turbulence, diffusion and buoyancy. It has been developed and maintained under a Quality Assurance program in compliance with the requirements of 10CFR50 Appendix B and applicable portions of ASME NQA-1 since 1995. GOTHIC has been used for assessing both forced and natural convection conditions for a wide range of applications, including: - Tracking concentration of hazardous gases and chemicals for habitability and safety assessments - Determining ventilation and filtration requirements and optimizing location and arrangement of these systems - Room heat-up, including diverse and Flexible coping strategies for Extended Loss of AC Power (FLEX/ELAP) - Equipment Qualification (EQ) A distinctive feature of GOTHIC is the ability to track many different fields/substances in a simulation, including user defined tracer elements, in the liquid, vapor and droplet fields as well as surfaces and filters. This capability allows GOTHIC to model fission product transport and release or the removal of particulates or harmful toxins from exhaust gases using a spray scrubber or other types of filtration systems. GOTHIC also includes models for engineered equipment, such as fans, filters, charcoal filters, dryers/demisters, dampers, etc. The aerosols and other filtered material are removed or accumulated in these components. The range of aerosol and radiological applications that GOTHIC has been used for includes: - Source Term: Primary Coolant (Equilibrium) Activity; Non-Water Coolant Source/Leakage. - Conditions for Iodine Re-evolution: Sump/Suppression Pool Conditions and pH; - RWST Conditions and pH. - Isotope Removal Mechanisms: Containment Sprayed and Unsprayed Region Mixing; Charcoal Filter Heating due to Iodine decay. - Radionuclide Transport and Decay: Post-LOCA Release in containment; Transport between connected Compartments and vent systems; Groundwater transport of radionuclides. - Non-Newtonian Fluid modeling for sludge, waste tanks, etc. ADAMS ML071581053 (titled 'Best Practice Guidelines for the use of CFD in Nuclear Reactor Safety Applications') poses guidelines for applying single phase CFD codes in nuclear reactor safety problems and GOTHIC is listed as a 'tool for 3D flows' and 'dispersal and deposition of radionuclides.' The Nuclear Quality Assurance (NQA) pedigree of GOTHIC is an important aspect for applications in the nuclear industry. The fundamental tracer models (convective transport, molecular and turbulent diffusion, removal mechanisms, etc.) have been verified using analytical solutions and validated against applicable separate effects tests. Also, GOTHIC has been benchmarked to many integrated effects tests, including Phebus FP (Fission Product). GOTHIC gives good agreement for the buildup and decay of fission products in Phebus Test 3. The model developed by ZNE demonstrates GOTHIC's applicability and acceptability for use in analyzing the migration and retention of radioactive isotopes and their progeny in normal operation and accident scenario analyses for isotope production facilities, hot cells, and gloveboxes. The tracer activities calculated by GOTHIC can then be used in downstream radiation transport and shielding codes like RADTRAD-NAI{sup C}, MCNP{sup R}, and MicroShield{sup R} to determine on-site and of-site doses. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2021

Flow Redirection and Induction in Steady State (FLORIS) Wind Plant Power Production Data Sets

This dataset contains turbine- and plant-level power outputs for 252,500 cases of diverse wind plant layouts operating under a wide range of yawing and atmospheric conditions. The power outputs were computed using the Gaussian wake model in NREL's FLOw Redirection and Induction in Steady State (FLORIS) model, version 2.3.0. The 252,500 cases include 500 unique wind plants generated randomly by a specialized Plant Layout Generator (PLayGen) that samples randomized realizations of wind plant layouts from one of four canonical configurations: (i) cluster, (ii) single string, (iii) multiple string, (iv) parallel string. Other wind plant layout parameters were also randomly sampled, including the number of turbines (25-200) and the mean turbine spacing (3D-10D, where D denotes the turbine rotor diameter). For each layout, 500 different sets of atmospheric conditions were randomly sampled. These include wind speed in 0-25 m/s, wind direction in 0 deg.-360 deg., and turbulence intensity chosen from low (6%), medium (8%), and high (10%). For each atmospheric inflow scenario, the individual turbine yaw angles were randomly sampled from a one-sided truncated Gaussian on the interval 0 deg.-30 deg. oriented relative to wind inflow direction. This random data is supplemented with a collection of yaw-optimized samples where FLORIS was used to determine turbine yaw angles that maximize power production for the entire plant. To generate this data, a subset of cases were selected (50 atmospheric conditions from 50 layouts each for a total of additional 2,500 cases) for which FLORIS was re-run with wake steering control optimization. The IEA onshore reference turbine, which has a 130 m rotor diameter, a 110 m hub height, and a rated power capacity of 3.4 MW was used as the turbine for all simulations. The simulations were performed using NREL's Eagle high performance computing system in February 2021 as part of the Spatial Analysis for Wind Technology Development project funded by the U.S. Department of Energy Wind Energy Technologies Office. The data was collected, reformatted, and preprocessed for this OEDI submission in May 2023 under the Foundational AI for Wind Energy project funded by the U.S. Department of Energy Wind Energy Technologies Office. This dataset is intended to serve as a benchmark against which new artificial intelligence (AI) or machine learning (ML) tools may be tested. Baseline AI/ML methods for analyzing this dataset have been implemented, and a link to their repository containing those models has been provided. The .h5 data file structure can be found in the GitHub repository under explore_wind_plant_data_h5.ipynb.

AI↗
DOE OSTI2024

Variation of the Passive Film on Compositionally Concentrated Dual-Phase Al 0.3 Cr 0.5 Fe 2 Mn 0.25 Mo 0.15 Ni 1.5 Ti 0.3 and Implications for Corrosion

The passive film on a dual-phase Al 0.3 Cr 0.5 Fe 2 Mn 0.25 Mo 0.15 Ni 1.5 Ti 0.3 FCC + Heusler (L2 1 ) compositionally concentrated alloy formed during extended exposure to an applied potential in the passive range in dilute chloride solution was characterized. Each phase, with its own distinct composition of passivating elements, formed unique passive films separated by a heterophase interface. High-resolution, surface sensitive characterization enabled chemical analysis of the passive film formed over individual phases. The film formed over the L2 1 phase had a higher concentration of Al, Ni, and Ti, while the film formed over FCC phase was of similar thickness but contained comparatively higher Cr, Fe, and Mo concentrations, consistent with the differences in bulk microstructure composition. The passive film was continuous across phase boundaries and the distribution of passivating elements (Al, Cr, and Ti) indicated both phases were independently passivated. Spatially resolved analysis of the surface chemistry of the dual-phase CCA revealed that the cation with the highest composition in passive film formed on the FCC phase was Cr (52.4 at. pct) and for the L2 1 phase was Ti (53.1 at. pct) despite the bulk concentration of each element being below 20 at. pct in their respective phases. Al, Cr, and Ti were enriched in both phases within the passive film relative to their respective bulk compositions. In parallel studies, single-phase alloys with compositions representative of the FCC and L2 1 phases were synthesized to evaluate the corrosion behavior of each phase in isolation. The corrosion behavior of the dual-phase alloy showed passivity evidenced by a pitting potential of 0.615 V SCE in 0.01 M NaCl. The pitting potential and other electrochemical parameters suggested a combination of behaviors of both single-phase samples, suggesting that the global corrosion behavior may be represented by a composite theory applied to phases, their area fractions, and interphase length. However, the interphase in the dual-phase CCA was a local corrosion initiation site and may limit localized corrosion protectiveness. The alloy design implications for optimization of second phase structure and morphology are discussed.

36 MATERIALS SCIENCE↗
DOE OSTI2019

Exploring small-scale chemostats to scale up microbial processes: 3-hydroxypropionic acid production in S. cerevisiae

The physiological characterization of microorganisms provides valuable information for bioprocess development. Chemostat cultivations are a powerful tool for this purpose, as they allow defined changes to one single parameter at a time, which is most commonly the growth rate. The subsequent establishment of a steady state then permits constant variables enabling the acquisition of reproducible data sets for comparing microbial performance under different conditions. We performed physiological characterizations of a 3-hydroxypropionic acid (3-HP) producing Saccharomyces cerevisiae strain in a miniaturized and parallelized chemostat cultivation system. The physiological conditions under investigation were various growth rates controlled by different nutrient limitations (C, N, P). Based on the cultivation parameters obtained subsequent fed-batch cultivations were designed.We report technical advancements of a small-scale chemostat cultivation system and its applicability for reliable strain screening under different physiological conditions, i.e. varying dilution rates and different substrate limitations (C, N, P). Exploring the performance of an engineered 3-HP producing S. cerevisiae strain under carbon-limiting conditions revealed the highest 3-HP yields per substrate and biomass of 16.6 %C-mol and 0.43 g gCDW -1 , respectively, at the lowest set dilution rate of 0.04 h -1 . 3-HP production was further optimized by applying N- and P-limiting conditions, which resulted in a further increase in 3-HP yields revealing values of 21.1 %C-mol and 0.50 g gCDW -1 under phosphate-limiting conditions. The corresponding parameters favoring an increased 3-HP production, i.e. dilution rate as well as C- and P-limiting conditions, were transferred from the small-scale chemostat cultivation system to 1-L bench-top fermenters operating in fed-batch conditions, revealing 3-HP yields of 15.9 %C-mol and 0.45 g gCDW -1 under C-limiting, as well as 25.6 %C-mol and 0.50 g gCDW -1 under phosphate-limiting conditions. Small-scale chemostat cultures are well suited for the physiological characterization of microorganisms, particularly for investigating the effect of changing cultivation parameters on microbial performance. In our study, optimal conditions for 3-HP production comprised (i) a low dilution rate of 0.04 h -1 under carbon-limiting conditions and (ii) the use of phosphate-limiting conditions. Similar 3-HP yields were achieved in chemostat and fed-batch cultures under both C- and P-limiting conditions proving the growth rate as robust parameter for process transfer and thus the small-scale chemostat system as powerful tool for process optimization.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2020

Collaborative Research: Advancing Arctic Climate Projection Capability at Seasonal to Decadal Scales (Final Technical Report)

The Regional Arctic System Model (RASM) at process resolving configurations has been used to (i) advance understanding of physical processes and feedbacks involved in Arctic amplification and (ii) understand and potentially reduce uncertainty in prediction of arctic climate change at seasonal to decadal scales. RASM consists the atmosphere (Weather and Research Forecasting model, WRF), ocean (Parallel Ocean Program, POP), sea ice (CICE), land hydrology (Variable Infiltration Capacity model, VIC), river routing scheme (RVIC), marine biogeochemistry components and the coupling framework (CPL7). Its domain is pan-Arctic, with the atmosphere and land components configured on a 50-km or 25-km grid and four configurations of the ocean and sea ice components: 1/12°(~9.3km) or 1/48°(~2.4km) and 45 or 60 vertical layers. These RASM configurations have been motivated by the emerging exascale capability for high performance computing to improve model fidelity. The dynamical downscaling of reanalysis allows comparison of RASM results with observations in place and time to: (i) advance system level understanding of physical processes and coupling involved in an event, (ii) optimize model parameter space, (iii) diagnose and reduce model biases and (iv) produce realistic and consistent across all the components initial conditions for predictions and predictability studies, which are all unique capabilities not available in global Earth System Models (ESMs). An evaluation of RASM 1.0 (Cassano et al. 2017) revealed that it had a cold bias over the oceans and a warm bias over land areas due largely to cloud and radiation biases in the model, with too little cloud cover simulated over land and too much cloud cover simulated over sub-polar oceans. This study has motivated an upgrade to WRF version 3.7.1 in RASM and allowed for the inclusion of the radiative impact of convective clouds. A variety of atmospheric physics parameterizations were evaluated against observations (e.g. data from the Arctic Clouds in Summer Experiment (ACSE); Sedlar et al. 2020) to identify an optimal suite of WRF physics options in RASM. The RASM with the optimized WRF physics were used to study the impact of strong mesoscale winds over the ocean around the southern tip of Greenland (DuVivier and Cassano 2016) and their impact on oceanic convection (DuVivier et al. 2017a). Data from the PolarWinds field campaign were used to evaluate WRF boundary layer physics and resolution impacts on the simulation of a Greenland barrier wind event (DuVivier et al. 2017b). The RVIC streamflow routing model has been implemented in RASM to realistically represent high-resolution streamflow processes (Hamman et al. 2017) and to couple the land buoyancy fluxes to the ocean. The RASM-RVIC high-resolution data set of all coastal freshwater fluxes for the Arctic drainage basin and surrounding areas for 1979-2014 was published as a separate product (https://doi.org/10.5281/zenodo.293037). The fidelity of atmospheric momentum transfer to and the response of polar marine Ekman layer in RASM and Community Earth System Model (CESM) was investigated by Roberts et al. (2015). The increased frequency of oceanic flux exchange in CESM, following the RASM guidance, caused a considerable increase in the median inertial ice speed across the Southern Ocean and parts of the Arctic. A comprehensive evaluation of the RASM1.0 atmosphere-ocean-sea ice-land interface was completed by Brunke et al. (2018). RASM was also demonstrated for its capability to simulate extreme events in agreement with observations in space and time (Lee et al. submitted). In particular, the development of three open water events, known as polynyas, have been simulated north of Greenland in February of 2011, 2017 and 2018, in agreement with satellite observations for the past four decades. The optimized RASM sea ice results have been favorably evaluated against satellite observations and a subset of eleven CMIP6 models (Watts et al. submitted). In a complementary project, Jin et al. (2018) have shown that RASM with higher-resolution and new sea-ice processes contributed to lower model errors in sea-ice conditions, concentrations of nutrients and ice algae, in comparison to results from the coarse-resolution (1°) CESM. In two other complementary studies, RASM results were used (i) to explain areas of concentrated use by bowhead whales, the seasonal progression in the use, and the physical environment within those areas (Citta et al. 2015) and (ii) for a synthesis of fall bowhead whales distribution and migration in the Bering-Chukchi-Beaufort (BCB) Sea to investigate whale movements and feeding to the local ocean hydrography and currents (Citta et al. 2018). However, the multi-decadal output from the CESM Large Ensemble yielded unrealistic forcing. Instead, the shorter NCEP CFSv2 9-month forecasts were successfully tested and afforded an increased ensemble size (~30) to demonstrate gains of dynamical downscaling at sub-seasonal to intra-annual time scales. The improved model physics and coupling among RASM model components have yielded more realistic representation of the sea ice cover and consistent across all model components initial conditions. Consequently, RASM demonstrates significant gains compared to simulation of sea ice in the NCEP reanalysis. In addition, RASM 6-month ensemble forecasts yield very realistic sea ice distribution, which demonstrates both significant gains of dynamical downscaling and the continued impact of the initial conditions on forecasts out to 6 months (https://nps.edu/web/rasm/predictions). A follow up study using RASM for dynamical downscaling of the more realistic CESM initialized Decadal Prediction Large Ensemble output is currently ongoing as part of the DOE RGMA HiLAT-RASM project.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

HARMONY: Large-Scale Architecture Search for Efficient Hybrid Language Models

As large language models scale to trillions of parameters, their computational and memory requirements present critical challenges for efficient training and deployment. While Mixture of Experts (MoE) architectures enable efficient scaling through sparse parameter activation, and state-space models like Mamba offer linear-time complexity, principled methods for combining these paradigms remain undeveloped. We introduce HARMONY (Hybrid Architecture Research for Mamba, Optimized with Neural efficiencY), a multi-objective evolutionary neural architecture search framework for discovering efficient hybrid language models that integrate Transformer attention mechanisms, Mixture-of-Experts routing, and Mamba state-space components. Through large-scale distributed search using 16,384 MI250X GPUs on the Frontier supercomputer, HARMONY explores a comprehensive design space encompassing six attention variants (MHA, MQA, GQA, MLA, SWA, and Mamba-2), variable MoE configurations with both routed and shared experts, and extensive Mamba hyperparameters. Our framework discovers heterogeneous architectures that balance training performance with computational efficiency through multi-objective optimization incorporating latency penalties and fitness-based selection. Analysis of discovered architectures reveals that optimal hybrid designs favor heterogeneous component mixing rather than homogeneous patterns, with Mamba-2 and Multi-Head Latent Attention (MLA) emerging as preferred mechanisms. Discovered architectures demonstrate superior training efficiency: our best configuration achieves a final perplexity of 1.0874 with 2.38B parameters while processing 4,320 tokens/second, outperforming significantly larger manually designed models. Full-scale evaluation shows HARMONY's top architectures achieve better loss trajectories than equivalently-sized models using state-of-the-art configurations including Mixtral, Jamba, and Samba. Additionally, we demonstrate 91% weak scaling efficiency when training discovered 36B-parameter models across 1,024 GPUs. HARMONY is released as an open framework with comprehensive tools for building and training hybrid models using expert-data-pipeline parallelism, democratizing access to automated architecture design for next-generation language models.

Herron, Emily [ORNL] (ORCID:0000000273008172)↗