Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “accelerator modeling”

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 1,045 records · Page 58
All sourcesNASA NTRSDOE OSTI
NASA NTRS1999

Horizontal vestibuloocular reflex evoked by high-acceleration rotations in the squirrel monkey. I. Normal responses

The horizontal angular vestibuloocular reflex (VOR) evoked by high-frequency, high-acceleration rotations was studied in five squirrel monkeys with intact vestibular function. The VOR evoked by steps of acceleration in darkness (3,000 degrees /s(2) reaching a velocity of 150 degrees /s) began after a latency of 7.3 +/- 1.5 ms (mean +/- SD). Gain of the reflex during the acceleration was 14.2 +/- 5.2% greater than that measured once the plateau head velocity had been reached. A polynomial regression was used to analyze the trajectory of the responses to steps of acceleration. A better representation of the data was obtained from a polynomial that included a cubic term in contrast to an exclusively linear fit. For sinusoidal rotations of 0.5-15 Hz with a peak velocity of 20 degrees /s, the VOR gain measured 0.83 +/- 0.06 and did not vary across frequencies or animals. The phase of these responses was close to compensatory except at 15 Hz where a lag of 5.0 +/- 0.9 degrees was noted. The VOR gain did not vary with head velocity at 0.5 Hz but increased with velocity for rotations at frequencies of >/=4 Hz (0. 85 +/- 0.04 at 4 Hz, 20 degrees /s; 1.01 +/- 0.05 at 100 degrees /s, P < 0.0001). No responses to these rotations were noted in two animals that had undergone bilateral labyrinthectomy indicating that inertia of the eye had a negligible effect for these stimuli. We developed a mathematical model of VOR dynamics to account for these findings. The inputs to the reflex come from linear and nonlinear pathways. The linear pathway is responsible for the constant gain across frequencies at peak head velocity of 20 degrees /s and also for the phase lag at higher frequencies being less than that expected based on the reflex delay. The frequency- and velocity-dependent nonlinearity in VOR gain is accounted for by the dynamics of the nonlinear pathway. A transfer function that increases the gain of this pathway with frequency and a term related to the third power of head velocity are used to represent the dynamics of this pathway. This model accounts for the experimental findings and provides a method for interpreting responses to these stimuli after vestibular lesions.

NASA Program Biomedical Research and Countermeasur↗
DOE OSTI2024

An active learning framework for the rapid assessment of galvanic corrosion

The current present in a galvanic couple can define its resistance or susceptibility to corrosion. However, as the current is dependent upon environmental, material, and geometrical parameters it is experimentally costly to measure. To reduce these costs, Finite Element (FE) simulations can be used to assess the cathodic current but also require experimental inputs to define boundary conditions. Due to these challenges, it is crucial to accelerate predictions and accurately predict the current output for different environments and geometries representative of in-service conditions. Machine learned surrogate models provides a means to accelerate corrosion predictions. However, a one-time cost is incurred in procuring the simulation and experimental dataset necessary to calibrate the surrogate model. Therefore, an active learning protocol is developed through calibration of a low-cost surrogate model for the cathodic current of an exemplar galvanic couple (AA7075-SS304) as a function of environmental and geometric parameters. The surrogate model is calibrated on a dataset of FE simulations, and calculates an acquisition function that identifies specific additional inputs with the maximum potential to improve the current predictions. This is accomplished through a staggered workflow that not only improves and refines prediction, but identifies the points at which the most information is gained, thus enabling expansion to a larger parameter space. The protocols developed and demonstrated in this work provide a powerful tool for screening various forms of corrosion under in-service conditions.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Vibration of a cryomodule’s coupler during road transportation: Multibody analysis and experimental validation

The road transportation of critical components for cryogenic cryomodules is a challenging task. Indeed, the shipping and the handling of fundamental sub-assemblies can expose these components to harmful dynamic loads that can jeopardize their structural integrity. This article has the purpose of developing a finite element multibody model capable of assessing the acceleration on a test coupler for a cavity of the Single Spoke Resonator 1 subjected to a typical road transportation trip. A multibody model consisting of rigid bodies and finite element–imported flexible bodies has been created allowing a comparison of the simulated and experimental acceleration. Afterwards, an experimental on-road test, whose set-up is described in this article, has been carried out at Fermilab with instrumented components. Being able to effectively simulate and validate the dynamic effects on such sub-assemblies makes it possible, for the future, to simulate more complex structures subject to dynamic loadings, such as the entire cryomodules during shipping. Furthermore, the results from the simulations can be used to guide the design of new suspension systems suitable for the reduction mitigation of structural vibration during the road transportation of cryomodules.

42 ENGINEERING↗
NASA NTRSDate not supplied

The Additive Manufacturing Moment Measure (AM3) Approach to Predictions of Solid Cooling Rate and Time Above Melt

Qualification of a laser powder bed fusion additive manufacturing (LPBF-AM) process requires knowledge of the multi-scale material physics during the process, per part. As the LPBF-AM build occurs, each moment is influenced by the process history. Knowledge of the build sequence can be used to generate a discretized time-space-condition point field that when coupled with a nearest neighbors’ calculation results in a generalized and fully parallel process model computation. This GPU accelerated approach was developed for part-scale analysis of build files along with in-situ process monitoring sensor data and is termed the “Additive Manufacturing Moment Measure” (AM3). The AM3 approach will be presented and then used to evaluate an AM Bench relevant geometry with synchronized in-situ process data, ex-situ nondestructive evaluation, and optical microscopy observations. These comparisons permit a better understanding of how the process actions can affect the LPBF-AM build quality and the signals generated during in-situ process monitoring.

Additive Manufacturing↗
DOE OSTI2021

Quantum dynamics of a nucleon in the Fermi accelerator

Highlights: • In nuclear physics, the Fermi accelerator is a coherent open quantum system. • There is not a true equilibrium for a nucleon‘s mean-energy in the Fermi accelerator. • Occupation of higher energy eigen-states remains even after the mean-energy plateaus. • This work paves the way for work on quantum tunnelling in the Fermi accelerator. The quantum dynamics of a particle in a one-dimensional box with an oscillating wall (the Fermi accelerator) is investigated. The model is applied to the motion of a single nucleon in the mean-field potential of a heavy atomic nucleus whose surface vibrates. By directly solving the time-dependent Schrödinger equation, both the state of the particle and its mean-energy are studied. The effects of the frequency of the wall oscillation on the nucleon’s energy are addressed. Its energy oscillates in phase with the moving wall for all frequencies, showing no chaotic behaviour. There is a large initial peak of the nucleon’s energy as the particle adjusts to the sudden change in the size of the box and a varying relaxation time as it plateaus towards lower energy and a partial equilibrium. Small oscillations in energy continue, since there cannot be a true equilibrium while the wall is moving. The quantum coherence between the different parts of the nucleon’s wave-function in real space is very much preserved. This research lays the foundation for future investigations into quantum tunnelling in the Fermi accelerator.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2026

Surrogates for Valve-Controlled Pipe Flow: Accelerating Nuclear Reactor Design

Neural surrogate models are developed to replace expensive steady-state RANS CFD simulations for valve-controlled pipe flow in nuclear reactor design. Using parametric CFD data generated with MOOSE Pronghorn across a range of valve geometry and flow conditions, three approaches are compared: a POD-based reduced-order model, a structured UNet on a cylindrical grid, and unstructured models (DeepONet and BiStride MeshGraphNet) on nondimensionalized point clouds. POD achieves the highest accuracy (99%) with fast inference but requires storing all solution snapshots, while the DeepONet and BSMS-GNN both achieve ~89% accuracy at sub-second inference, with the BSMS-GNN offering superior geometric generalizability. These surrogates enable rapid ranking of candidate valve designs and can warm-start CFD solvers to accelerate convergence, supporting agentic design iteration on the Prometheus platform.

42 - ENGINEERING↗
DOE OSTI2023

Comparing the Performance of Julia on CPUs versus GPUs and Julia-MPI versus Fortran-MPI: a case study with MPAS-Ocean (Version 7.1)

Abstract. Some programming languages are easy to develop at the cost of slow execution, while others are fast at runtime but much more difficult to write. Julia is a programming language that aims to be the best of both worlds – a development and production language at the same time. To test Julia's utility in scientific high-performance computing (HPC), we built an unstructured-mesh shallow water model in Julia and compared it against an established Fortran-MPI ocean model, the Model for Prediction Across Scales–Ocean (MPAS-Ocean), as well as a Python shallow water code. Three versions of the Julia shallow water code were created: for single-core CPU, graphics processing unit (GPU), and Message Passing Interface (MPI) CPU clusters. Comparing identical simulations revealed that our first version of the Julia model was 13 times faster than Python using NumPy, where both used an unthreaded single-core CPU. Further Julia optimizations, including static typing and removing implicit memory allocations, provided an additional 10–20× speed-up of the single-core CPU Julia model. The GPU-accelerated Julia code was almost identical in terms of performance to the MPI parallelized code on 64 processes, an unexpected result for such different architectures. Parallelized Julia-MPI performance was identical to Fortran-MPI MPAS-Ocean for low processor counts and ranges from 2× faster to 2× slower for higher processor counts. Our experience is that Julia development is fast and convenient for prototyping but that Julia requires further investment and expertise to be competitive with compiled codes. We provide advice on Julia code optimization for HPC systems.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Transferring a Molecular Foundation Model for Polymer Property Predictions

Transformer-based large language models have remarkable potential to accelerate design optimization for applications such as drug development and material discovery. Self-supervised pretraining of transformer models requires large-scale data sets, which are often sparsely populated in topical areas such as polymer science. Further, state-of-the-art approaches for polymers conduct data augmentation to generate additional samples but unavoidably incur extra computational costs. In contrast, large-scale open-source data sets are available for small molecules and provide a potential solution to data scarcity through transfer learning. In this work, we show that using transformers pretrained on small molecules and fine-tuned on polymer properties achieves comparable accuracy to those trained on augmented polymer data sets for a series of benchmark prediction tasks.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
NASA NTRS2010

Radiation Beamline Testbeds for the Simulation of Planetary and Spacecraft Environments for Human and Robotic Mission Risk Assessment

The Center for Radiation Engineering and Science for Space Exploration (CRESSE) at Prairie View A&M University, Prairie View, Texas, USA, is establishing an integrated, multi-disciplinary research program on the scientific and engineering challenges faced by NASA and the international space community caused by space radiation. CRESSE focuses on space radiation research directly applicable to astronaut health and safety during future long term, deep space missions, including Martian, lunar, and other planetary body missions beyond low earth orbit. The research approach will consist of experimental and theoretical radiation modeling studies utilizing particle accelerator facilities including: 1. NASA Space Radiation Laboratory (NSRL) at Brookhaven National Laboratory; 2. Proton Synchrotron at Loma Linda University Medical Center; and 3. Los Alamos Neutron Science Center (LANSCE) at Los Alamos National Laboratory. Specifically, CRESSE investigators are designing, developing, and building experimental test beds that simulate the lunar and Martian radiation environments for experiments focused on risk assessment for astronauts and instrumentation. The testbeds have been designated the Bioastronautics Experimental Research Testbeds for Environmental Radiation Nostrum Investigations and Education (BERT and ERNIE). The designs of BERT and ERNIE will allow for a high degree of flexibility and adaptability to modify experimental configurations to simulate planetary surface environments, planetary habitats, and spacecraft interiors. In the nominal configuration, BERT and ERIE will consist of a set of experimental zones that will simulate the planetary atmosphere (Solid CO2 in the case of the Martian surface.), the planetary surface, and sub-surface regions. These experimental zones can be used for dosimetry, shielding, biological, and electronic effects radiation studies in support of space exploration missions. BERT and ERNIE are designed to be compatible with the experimental areas associated with the above facilities. CRESSE has broad expertise in space radiation in the areas of space radiation environment modeling, Monte-Carlo radiation transport modeling, space radiation instrumentation and dosimetry, radiation effects on electronics, and multi-functional composite shielding materials. The BERT and ERNIE testbeds will be utilized in individual and collaborative research incorporating this expertise. The research goal is to maximize the technical readiness level (TRL) of radiation instrumentation for human and robotic missions, optimizing the return value of CRESSE for NASA exploration and international co-operative missions. Outcomes and knowledge from research utilizing BERT and ERNIE will be applied to a variety of scientific and engineering disciplines vital for safe and reliable execution of future space exploration missions, which can be negatively impacted by the space radiation environment. The testbeds will be central to a variety of university educational activities and educational goals of NASA. Specifically, BERT and ERNIE will enhance educational opportunities in science, technology, engineering and mathematics (STEM) disciplines for engineering and science students at PVAMU, a historically black college/university. Preliminary data on prototype testbed configurations, including simulated lunar regolith (JSC-1A stimulant based on Apollo 11 samples), regolith/polyethylene composites, and dry ice, will be presented to demonstrate the usefulness of BERT and ERNIE in radiation beam line experiments.

Wilkins, Richard↗
DOE OSTI2023

Effect of CLIQ on Training of HL-LHC Quadrupole Magnets

The high-luminosity LHC upgrade requires stronger than LHC low-beta quadrupole magnets to reach the luminosity goals of the project. The project is well advanced and HL-LHC quadrupole magnets are currently being commissioned in US Labs (MQXFA magnets) and CERN (MQXFB magnets). Those are the first Nb 3 Sn magnets to be used in any large particle accelerator. At development stages, many Nb 3 Sn accelerator sub-scale models showed relatively slow training and MQXFA magnets were projected to have low tens of quenches before reaching operational field. Recently it was shown that dedicated capacitor-based de-vices can affect Nb 3 Sn magnet training, and it was suggested that CLIQ, a capacitor-based device intended for quench protection, can do too. Here, the present paper investigates effects on training likely induced by CLIQ, using the base fact that only half the coils in a quadrupole experience upward current modulation at quench because of capacitor discharge. The study encompasses all MQXFA production magnets trained at BNL to date. No other high-statistics data from identical magnets (series) with CLIQ protection exist so far. Implications and opportunities stemming from data analysis are discussed and conclusions drawn.

43 PARTICLE ACCELERATORS↗
NASA NTRSDate not supplied

Alternate Methods of Model Reduction to Avoid Dynamic Modal Truncation Error

Loads analysis is traditionally performed using dynamically reduced models, which provides the benefit to reduce run time. If the reduced model frequency cutoff is not chosen appropriately, the model will lack the dynamic content required to fully represent the response of the non-reduced model. Standard guidance for reduced model frequency content, provided in NASA-STD-5002, is to solve fixed base modes up to a minimum of 1.5x the model frequency content of interest and to employ static modal truncation methods such as residual vectors, the mode acceleration method, and the residual flexibility method to account for the truncated flexibility of the missing modes. Fixed base modes require the selection of a set of degrees of freedom to be constrained which, if not properly selected, may affect the accuracy of the reduced model by excluding some of the dynamic characteristic of the full model. In this case, the standard NASA guidance would be insufficient, but it may not be readily apparent that a portion of the reduced model response is missing. This error was encountered during an independent verification and validation (IV&V) effort, where it was observed that the resulting dynamic response was lower than the inline analysis. In this specific case, despite following the standard NASA model reduction guidelines in the selection of the frequency cutoff, the inline model still did not fully capture the necessary dynamic content. As part of the IV&V, an alternate reduction methodology was employed using an unconstrained mode acceleration method. The original model initially performed a constrained reduction to twice the frequency content of interest before doing a free-free run, employing the mode acceleration method to account for the truncated modes. In contrast in the IV&V, the unconstrained model reduced directly to the needed frequency content of the free-free run, avoiding any interactions between constraints and dynamic content. To verify the model, the original reduction methodology was used to generate a series of Hurty-Craig-Bampton reductions, each with a higher frequency cutoff than the previous. The results were shown to converge once the frequency cutoff increased past eight times the frequency content of interest. At the same frequency cutoff, the results of the Hurty-Craig-Bampton model converged with the results of the unconstrained mode acceleration model. This comparative study provided confidence that the results of the unconstrained modal acceleration reduced model were correct and that the Hurty-Craig-Bampton needed to increase its frequency cutoff to fully capture the dynamic response.

Erin Simmons↗
NASA NTRSDate not supplied

Alternate Methods of Model Reduction to Avoid Dynamic Modal Truncation Error

Loads analysis is traditionally performed using dynamically reduced models, which provides the benefit to reduce run time. If the reduced model frequency cutoff is not chosen appropriately, the model will lack the dynamic content required to fully represent the response of the non-reduced model. Standard guidance for reduced model frequency content, provided in NASA-STD-5002, is to solve fixed base modes up to a minimum of 1.5x the model frequency content of interest and to employ static modal truncation methods such as residual vectors, the mode acceleration method, and the residual flexibility method to account for the truncated flexibility of the missing modes. Fixed base modes require the selection of a set of degrees of freedom to be constrained which, if not properly selected, may affect the accuracy of the reduced model by excluding some of the dynamic characteristic of the full model. In this case, the standard NASA guidance would be insufficient, but it may not be readily apparent that a portion of the reduced model response is missing. This error was encountered during an independent verification and validation (IV&V) effort, where it was observed that the resulting dynamic response was lower than the inline analysis. In this specific case, despite following the standard NASA model reduction guidelines in the selection of the frequency cutoff, the inline model still did not fully capture the necessary dynamic content. As part of the IV&V, an alternate reduction methodology was employed using an unconstrained mode acceleration method. The original model initially performed a constrained reduction to twice the frequency content of interest before doing a free-free run, employing the mode acceleration method to account for the truncated modes. In contrast in the IV&V, the unconstrained model reduced directly to the needed frequency content of the free-free run, avoiding any interactions between constraints and dynamic content. To verify the model, the original reduction methodology was used to generate a series of Hurty-Craig-Bampton reductions, each with a higher frequency cutoff than the previous. The results were shown to converge once the frequency cutoff increased past eight times the frequency content of interest. At the same frequency cutoff, the results of the Hurty-Craig-Bampton model converged with the results of the unconstrained mode acceleration model. This comparative study provided confidence that the results of the unconstrained modal acceleration reduced model were correct and that the Hurty-Craig-Bampton needed to increase its frequency cutoff to fully capture the dynamic response.

Erin Simmons↗
DOE OSTI2021

Transport of Protostellar Cosmic Rays in Turbulent Dense Cores

We report recent studies have suggested that low-energy cosmic rays (CRs) may be accelerated inside molecular clouds by the shocks associated with star formation. We use a Monte Carlo transport code to model the propagation of CRs accelerated by protostellar accretion shocks through protostellar cores. We calculate the CR attenuation and energy losses and compute the resulting flux and ionization rate as a function of both radial distance from the protostar and angular position. We show that protostellar cores have nonuniform CR fluxes that produce a broad range of CR ionization rates, with the maximum value being up to two orders of magnitude higher than the radial average at a given distance. In particular, the CR flux is focused in the direction of the outflow cavity, creating a "flashlight" effect and allowing CRs to leak out of the core. The radially averaged ionization rates are less than the measured value for the Milky Way of ζ ≈ 10 -16 s -1 ; however, within r ≈ 0.03 pc from the protostar, the maximum ionization rates exceed this value. We show that variation in the protostellar parameters, particularly in the accretion rate, may produce ionization rates that are a couple of orders of magnitude higher or lower than our fiducial values. Finally, we use a statistical method to model unresolved subgrid magnetic turbulence in the core. We show that turbulence modifies the CR spectrum and increases the uniformity of the CR distribution but does not significantly affect the resulting ionization rates.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Assessment of the Pan-Arctic Accelerated Rate of Sea Ice Decline in CMIP6 Historical Simulations

Abstract Arctic sea ice loss in response to a warming climate is assessed in 42 models participating in phase 6 of the Coupled Model Intercomparison Project (CMIP6). Sea ice observations show a significant acceleration in the rate of decline commencing near the turn of the twenty-first century. It is our assertion that state-of-the-art climate models should qualitatively reflect this accelerated trend within the limitations of internal variability and observational uncertainty. Our analysis shows that individual CMIP6 simulations of sea ice depict a wide range of model spread on biases and anomaly trends both across models and among their ensemble members. While the CMIP6 multimodel mean captures the observed sea ice area (SIA) decline relatively well, an individual model’s ability to represent the acceleration in sea ice decline remains a challenge. Seventeen (40%) out of 42 CMIP6 models and 37 (13%) out of the total 286 ensemble members reasonably capture the observed trends and acceleration in SIA decline. In addition, a larger ensemble size appears to increase the odds for a model to include at least one ensemble member skillfully representing the accelerated SIA trends. Simulations of sea ice volume (SIV) show much larger spread and uncertainty than SIA; however, due to limited availability of sea ice thickness data, these are not as well constrained by observations. Finally, we find that models with more ocean heat transport simulate larger sea ice declines, which suggests an emergent constraint in CMIP6 ensembles. This relationship points to the need for better understanding and modeling of ice–ocean interactions, especially with respect to frazil ice growth.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

A MeerKAT-meets-LOFAR study of Abell 1413: a moderately disturbed non-cool-core cluster hosting a ~500 kpc ‘mini’-halo

Many relaxed cool-core clusters host diffuse radio emission on scales of hundreds of kiloparsecs: mini-haloes. However, the mechanism responsible for generating them, as well as their connection with central active galactic nuclei, is elusive and many questions related to their physical properties and origins remain unanswered. This paper presents new radio observations of the galaxy cluster Abell 1413 performed with MeerKAT (L-band; 872 to 1712 MHz) and LOFAR HBA (120 to 168 MHz) as part of a statistical and homogeneous census of mini-haloes. Abell 1413 is unique among mini-halo clusters as it is a moderately disturbed non-cool-core cluster. Our study reveals an asymmetric mini-halo up to 584 kpc in size at 1283 MHz, twice as large as first reported at similar frequencies. The spectral index is flatter than previously reported, with an integrated value of α = −1.01 ± 0.06, shows significant spatial variation, and a tentative radial steepening. We studied the point-to-point X-ray/radio surface brightness correlation to investigate the thermal/non-thermal connection: our results show a strong connection between these components, with a superlinear slope of $b = 1.63^{+0.10}_{-0.10}$ at 1283 MHz and $b = 1.20^{+0.13}_{-0.11}$ at 145 MHz. We also explore the X-ray surface brightness/radio spectral index correlation, finding a slope of $b = 0.59^{+0.11}_{-0.11}$. Both investigations support the evidence of spectral steepening. Finally, in the context of understanding the particle acceleration mechanism, we present a simple theoretical model which demonstrates that hybrid scenarios – secondary electrons (re-)accelerated by turbulence – reproduce a superlinear correlation slope.

79 ASTRONOMY AND ASTROPHYSICS↗
NASA NTRS2019

Accelerated Simulation of Air Pollution Using NVIDIA RAPIDS

Atmospheric chemistry models are a central tool to study and forecast the impact of air pollution on the environment, vegetation, and human health. However, the numerical simulation of chemical kinetics is computationally expensive due to the stiffness of the system of ordinary differential equations that describes atmospheric chemistry. Here we present an alternative approach to the computation of atmospheric chemistry based on machine learning. Our training data set is produced using the NASA Goddard Earth Observing System (GEOS) model with GEOS-Chem chemistry, run on the NASA Center for Climate Simulation (NCCS) Discover supercomputing cluster on 384 Intel Xeon Haswell cores. This model spends more than 50% of total run time on solving atmospheric chemistry. The data set contains as input features the air pollution concentrations before solving the differential equations, together with some key physical parameters such as temperature and sun intensity. As target variables we define the air pollution concentrations after solving the differential equations. Using Dask-cuDF and Dask-XGBoost on the NVIDIA RAPIDS platform on 8 Tesla V100 GPUs, we generate from this training set gradient boosted decision tree models that can reproduce the simulation of chemical kinetics. We do this on the NCCS Advanced Data Analytics Platform (ADAPT) science cloud environment. Our application takes full advantage of recent advances in Dask-XGBoost, such as multi-node and multi-GPU scaling for distributed training with large data sets. The increase in training data size enabled by this is critical to capture the full range of chemical environments encountered across the globe and all annual seasons.The boosted tree models offer good predictability and show many of the features of the full chemistry reference simulation. Further improvements can be achieved through mass balance considerations and by accounting for error correlations. We incorporate the boosted tree models into the GEOS reference model using XGBoost's C API. This enables a seamless integration of the GPU trained models into GEOS-Chem, which is written in Fortran and optimized for use in a massively parallel CPU environment. We show the benefits of this approach and discuss the potential speedup of this machine learning accelerated atmospheric chemistry model.

Keller, Christoph A.↗