SEARCH · Engineering Papers

Results for “Model error”

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.

SourceResearch connectionFrom yearThrough year

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

At least 1,333 records · Page 74

All sources NASA NTRS DOE OSTI

DOE OSTI2023

gLaSDI: Parametric physics-informed greedy latent space dynamics identification

A parametric adaptive physics-informed greedy Latent Space Dynamics Identification (gLaSDI) method is proposed for accurate, efficient, and robust data-driven reduced-order modeling of high-dimensional nonlinear dynamical systems. In the proposed gLaSDI framework, an autoencoder discovers intrinsic nonlinear latent representations of high-dimensional data, while dynamics identification (DI) models capture local latent-space dynamics. Here, an interactive training algorithm is adopted for the autoencoder and local DI models, which enables identification of simple latent-space dynamics and enhances accuracy and efficiency of data-driven reduced-order modeling. To maximize and accelerate the exploration of the parameter space for the optimal model performance, an adaptive greedy sampling algorithm integrated with a physics-informed residual-based error indicator and random-subset evaluation is introduced to search for the optimal training samples on the fly. Further, to exploit local latent-space dynamics captured by the local DI models for an improved modeling accuracy with a minimum number of local DI models in the parameter space, a -nearest neighbor convex interpolation scheme is employed. The effectiveness of the proposed framework is demonstrated by modeling various nonlinear dynamical problems, including Burgers equations, nonlinear heat conduction, and radial advection. The proposed adaptive greedy sampling outperforms the conventional predefined uniform sampling in terms of accuracy. Compared with the high-fidelity models, gLaSDI achieves 17 to 2,658× speed-up with 1 to 5% relative errors.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2022

Bosonic field digitization for quantum computers

Quantum simulation of quantum field theory is a flagship application of quantum computers that promises to deliver capabilities beyond classical computing. The realization of quantum advantage will require methods that can accurately predict error scaling as a function of the resolution and parameters of the model and that can be implemented efficiently on quantum hardware. In this paper, we address the representation of lattice bosonic fields in a discretized field amplitude basis, develop methods to predict error scaling, and present efficient qubit implementation strategies. A low-energy subspace of the bosonic Hilbert space, defined by a boson occupation number cutoff, can be represented with exponentially good accuracy by a low-energy subspace of a finite-size Hilbert space. The finite representation construction and the associated errors are directly related to the accuracy of the Nyquist-Shannon sampling and the finite Fourier transforms of the boson number states in the field and the conjugate-field bases. We analyze the relation between the boson mass, the discretization parameters used for wave function sampling, and the finite representation size. Numerical simulations of small size Φ 4 problems demonstrate that the boson mass optimizing the sampling of the ground state wave function is a good approximation to the optimal boson mass yielding the minimum low-energy subspace size. However, we find that accurate sampling of general wave functions does not necessarily result in accurate representation. Finally, we develop methods for validating and adjusting the discretization parameters to achieve more accurate simulations.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

DOE OSTI2025

Basic Research Needs for Inverse Methods for Complex Systems under Uncertainty

Inverse problems, which aim to infer unknown properties of a system using experimental and observational data, are central to addressing many of the U.S. Department of Energy’s (DOE) most critical scientific and engineering challenges. Accurate, computationally efficient, and data-efficient solutions to inverse problems are essential for advancing DOE mission-critical science drivers, including analyzing data from large-scale experimental facilities, optimizing fusion reactor performance, accelerating materials discovery, enhancing geophysical imaging, improving wildfire predictions, and enabling autonomous systems and digital twins. However, these problems are becoming increasingly complex, often involving nonlinear, highdimensional, and interconnected systems and models that span multiple physics and scales, while relying on data with varying quantity, quality, and information content. Compounding these challenges is the uncertainty inherent in DOE-relevant systems, where errors in inputs, noise in data, incompleteness of data, and discrepancies between models and reality constrain the accuracy and precision of solutions. At the same time, the convergence of recent scientific computing trends—scientific machine learning, artificial intelligence, and computing advances such as exascale computing—is creating unprecedented opportunities for tackling these challenges. The cross-cutting nature of inverse problems, combined with their growing complexity and rapidly evolving data and algorithmic demands, strongly motivates the formulation of a prioritized research agenda to maximize their capabilities and impact. In response to this need, DOE’s Advanced Scientific Computing Research (ASCR) program in the Office of Science convened the Workshop on Basic Research Needs for Inverse Problems for Complex Systems Under Uncertainty in June 2025. This workshop brought together experts across disciplines to identify grand challenges and major opportunities in the field. Through collaborative discussions, the workshop defined transformative research directions aimed at addressing the mathematical, statistical, and computational challenges posed by inverse problems under uncertainty. As a result of these efforts, four priority research directions (PRDs) were identified to guide future research and development in this area. These PRDs, summarized below, represent a roadmap for advancing the foundational science and mathematics of inverse problems, enabling robust, scalable, and uncertainty-aware solutions that are critical for DOE applications.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2025

Fault-tolerant operation and materials science with neutral atom logical qubits

We report on the fault-tolerant operation of logical qubits on a neutral atom quantum computer, with logical performance surpassing physical performance for multiple circuits including Bell state preparation (12x error reduction), random circuits (15x), and a prototype Anderson Impurity Model ground state solver for materials science applications (up to 6x, non-fault-tolerantly). The logical qubits are implemented via the [[4, 2, 2]] code (C 4 ). Our work constitutes the first complete realization of the benchmarking protocol proposed by Gottesman 2016 demonstrating results consistent with fault tolerance. In light of recent advances on applying concatenated C 4 /C 6 detection codes to achieve error correction with high code rates and thresholds, our work can be regarded as a building block towards a practical scheme for fault tolerant quantum computation. Our demonstration of a materials science application with logical qubits particularly demonstrates the immediate value of these techniques on current experiments.

36 MATERIALS SCIENCE↗

DOE OSTI2026

Comparison of DeePMD, MTP, GAP, ACE and MACE Machine‐Learned Potentials for Radiation‐Damage Simulations: A User Perspective

Accurate and efficient interatomic potentials are essential for molecular dynamics (MD) simulations of radiation damage, gas diffusion, and phase stability in complex ceramics such as LiAlO 2 , especially under extreme conditions relevant to tritium production. Here, we evaluate the performance of six machine-learned interatomic potentials (MLIPs), moment tensor potential (MTP), Gaussian approximation potential, deep potential (DeePMD), atomic cluster expansion (ACE), message-passing ACE (multilayer atomic cluster expansion (MACE) pretrained) and MACE (trained from-scratch), all trained on the same density functional theory dataset with inclusion of tritium. The MLIPs are benchmarked against traditional Buckingham and ReaxFF potentials in terms of energy accuracy, density predictions, thermal equilibration behavior, threshold displacement energy (E d ), tritium diffusivity, and computational cost. Among the models, MTP shows the best overall balance between efficiency and accuracy, with low force and energy errors and realistic E d values for Li and Al. The ACE and MACE (pretrained and trained from scratch) models exhibit high E d (>200 eV) and unphysical pair interactions. DeePMD underestimates Ed due to overly repulsive behavior even at equilibrium distances. All models over-estimate tritium diffusion but the pretrained MACE model behaves well during tritium-diffusion simulations up to 500 K, maintaining diffusivities in the physically consistent 10 −11 m 2 /s range. Finally, we quantify the computational cost of each potential in large-scale atomic/molecular massively parallel simulator, finding that only MTP is more efficient than traditional empirical potentials, while others are significantly more expensive. These findings explain the trade-offs between accuracy and computational cost in MLIP development and provide essential guidance for use in high-throughput radiation damage and gas diffusion simulations in nuclear ceramics.

74 ATOMIC AND MOLECULAR PHYSICS↗

DOE OSTI2025

Understanding and Estimating Error Propagation in Neural Networks for Scientific Data Analysis

Neural networks are increasingly integrated into scientific discovery, where input data reduction and model quantization play a key role in accelerating inference. However, understanding and mitigating the impact of these techniques on output error is critical for ensuring reliable results, particularly in tasks demanding high numerical precision. This paper introduces a comprehensive framework for optimizing neural network inference in scientific computing by combining data reduction and weight quantization while maintaining error-controlled outcomes. We develop theoretical analyses to bound error propagation under these reductions and propose a framework that balances computational performance with error constraints. Evaluation on real-world learning-based combustion simulations and satellite image classification demonstrates that our derived error bounds accurately predict observed errors while enabling significant computational speedup under our framework. This work highlights the potential for further leveraging advancements in modern lossy compression algorithms and hardware accelerators that support lower-precision formats.

He, Weiming [New Jersey Institute of Technology]↗

DOE OSTI2020

A Bayesian framework for adsorption energy prediction on bimetallic alloy catalysts

Abstract For high-throughput screening of materials for heterogeneous catalysis, scaling relations provides an efficient scheme to estimate the chemisorption energies of hydrogenated species. However, conditioning on a single descriptor ignores the model uncertainty and leads to suboptimal prediction of the chemisorption energy. In this article, we extend the single descriptor linear scaling relation to a multi-descriptor linear regression models to leverage the correlation between adsorption energy of any two pair of adsorbates. With a large dataset, we use Bayesian Information Criteria (BIC) as the model evidence to select the best linear regression model. Furthermore, Gaussian Process Regression (GPR) based on the meaningful convolution of physical properties of the metal-adsorbate complex can be used to predict the baseline residual of the selected model. This integrated Bayesian model selection and Gaussian process regression, dubbed as residual learning, can achieve performance comparable to standard DFT error (0.1 eV) for most adsorbate system. For sparse and small datasets, we propose an ad hoc Bayesian Model Averaging (BMA) approach to make a robust prediction. With this Bayesian framework, we significantly reduce the model uncertainty and improve the prediction accuracy. The possibilities of the framework for high-throughput catalytic materials exploration in a realistic setting is illustrated using large and small sets of both dense and sparse simulated dataset generated from a public database of bimetallic alloys available in Catalysis-Hub.org.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2024

Aboveground woody biomass estimation of young bioenergy plantations of Populus and its hybrids using mobile (backpack) LiDAR remote sensing

Woody aboveground biomass (AGB) including short-rotation Populus is used as a feedstock for renewable and carbon-neutral bioenergy. While woody AGB can be estimated with allometric equations requiring labor-intensive field data, remote sensing technologies like mobile terrestrial light detection and ranging (LiDAR) can estimate woody AGB quickly and accurately. Therefore, the goals of this study were to develop a model to predict woody AGB of 2-year-old Populus spp. from three taxa (P. deltoides, P. deltoides × P. maximowiczii and P. deltoides × P. trichocarpa) using allometric (height and diameter at breast height (DBH)) or LiDAR-derived metrics from a mobile terrestrial (backpack) system. Likewise, we sought to compare LiDAR-estimated tree height and DBH with field-measured values. We found that a taxa-specific model containing LiDAR-measured tree height, crown volume, and taxa interactions with the height of the 10 th percentile, and the density of the lowest interval (density metric 0) explained 84 % of the variation in woody AGB with a root mean square error (RMSE) of 28.7 % and performed slightly better than the allometric model. The best model excluding taxa had a slightly higher RMSE but lower bias than the allometric model. LiDAR-derived tree heights were highly correlated with field-measured heights, but DBH could not be estimated accurately. Therefore, terrestrial mobile LiDAR systems can accurately estimate woody AGB and tree height of Populus in short rotation systems to aid in the fast and efficient quantification of woody bioenergy production and renewable energy resources.

AGB↗

DOE OSTI2024

Transformer Neural Networks with Spatiotemporal Attention for Predictive Control and Optimization of Industrial Processes

In the context of real-time optimization and model predictive control of industrial systems, machine learning, and neural networks represent cutting-edge tools that hold promise for enhancing dynamic modeling. This work presents a novel transformer neural network architecture for real-time optimization and model predictive control. This network design includes a modified attention mechanism inspired by positional embedding attention from vision transformers and task-specific modifications to the input-output structure of the transformer’s decoder stack. Experiments were conducted using data from a 450 MW coal-fired power plant to evaluate this approach's effectiveness. The transformer neural network was compared with conventional recurrent models, including GRU and LSTM. The transformer exhibited a 6% increase in the R-squared (R2) value of predictions and an 83% reduction in mean squared error (MSE). Computation time was also reduced by 84% compared to conventional recurrent models.

Gallup, Ethan R.↗

DOE OSTI2023

Sidelobe modeling and mitigation for a three mirror anastigmat cosmic microwave background telescope

Telescopes measuring cosmic microwave background (CMB) polarization on large angular scales require exquisite control of systematic errors to ensure the fidelity of the cosmological results. In particular, far-sidelobe contamination from wide angle scattering is a potentially prominent source of systematic error for large aperture microwave telescopes. Here we describe and demonstrate a ray-tracing-based modeling technique to predict far sidelobes for a three mirror anastigmat telescope designed to observe the CMB from the South Pole. Those sidelobes are produced by light scattered in the receiver optics subsequently interacting with the walls of the surrounding telescope enclosure. After comparing simulated sidelobe maps and angular power spectra for different enclosure wall treatments, we propose a highly scattering surface that would provide more than an order of magnitude reduction in the degree-scale far-sidelobe contrast compared to a typical reflective surface. Finally, we conclude by discussing the fabrication of a prototype scattering wall panel and presenting measurements of its angular scattering profile.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

DOE OSTI2020

Octane Modeling of Isobutanol Blending into Gasoline

Thirty-four gasoline blendstocks for oxygenate blending were used to create finished gasoline blends with isobutanol content of 12.5 volume percent (vol. %) and 16 vol. %. The gasoline blendstocks and finished fuels were analyzed for octane number (research [RON] and motor [MON]) to determine the effect of blending isobutanol. Volumetric and molar linear blending models were developed to predict finished fuel RON and MON, starting from the properties and composition of the gasoline blendstocks and isobutanol. Results show the molar blending model provided a better fit for the experimental data than the volumetric blending model. The volumetric model was further improved by adding nonlinear terms, improving the error to within ~1 ON. Gasoline blendstock properties impacted the finished fuel RON/MON, with paraffins having a synergistic effect with isobutanol and olefins and aromatics having an antagonistic effect.

10 SYNTHETIC FUELS↗

DOE OSTI2025

..delta..-Learning of High-Fidelity Electronic Structure Using Graph Neural Networks with Modified Node-Level Features

In this work, we present a ..delta..-learning approach for predicting the eigenvalues calculated with the hybrid functional HSE06 (..epsilon..nkHSE) for a set of metal and nitrogen doped graphene catalysts (MNCs) from Perdew-Burke-Ernzerhof (PBE) inputs. The model presented here incorporates electronic scalar features along with structural information in a graph neural network (GNN). In particular, the PBE eigenvalues for different bands and k-points and orbital-resolved projectors are combined with the applied potential as node-level features along with structural information within the Atomistic Line Graph Neural Network (ALIGNN) architecture. These features enable flexibility for systems with electrified interfaces, such as in electrocatalysts and achieves mean absolute error (MAE) of less than 0.1 eV. The machine learning model reported here achieves a strong generalization to left-out adsorbates (MAE = 0.074 eV) and leave-one-chemical-space-out (MAE = 0.08 eV) and completely left-out metals (MAE = 0.072 eV), confirming the robustness of the machine learning (ML) model in predicting ..epsilon..nkHSE.

36 MATERIALS SCIENCE↗

DOE OSTI2024

On the Representativity of Electrode Microstructure Parameters and Their Electrochemical Response for Lithium Ion Batteries

Lithium-ion battery electrochemical models require an accurate description of the electrodes microstructures to be predictive that can be achieved through nanoscale imaging. Such observations are however limited by their field of view (FOV), as they provide only a subset of the whole electrode volume that does not necessarily represent the whole electrode microstructure heterogeneity, and therefore can bias the microstructure analysis. A microstructure scale electrochemical model was used to investigate lithium plating onset, material non-uniform utilization, and in-plane heterogeneities for an NMC-graphite full cell. To evaluate the representativeness, and thus relevance, of these model predictions, a coupled representativity analysis has been performed on the microstructure parameters and, in a novel way, on the full cell electrochemical response. Electrode microstructure parameters representativeness has been first quantified using the representative volume element (RVE) methodology. The RVE major flaw is that ultimately it can only conclude if a FOV contains representative subvolumes of the FOV, but not if the FOV itself is representative of the electrode volume. Analysis can conclude negatively ('FOV is not representative'), but not positively ('FOV is representative'). One major contribution of this work was to quantify the convergence of the RVE size with the FOV, to actually investigate the FOV representativeness and thus partly remedy this intrinsic limitation. The analysis determined that performing a standard RVE calculation, without exploring its FOV convergence, is likely to strongly underestimate the actual RVE size. The new RVE methodology has been automated in the NREL open-source Microstructure Analysis Toolbox (MATBOX) and is available to the battery community. Representativeness of microstructure parameters is however only an intermediate step, as the end-results of an electrochemical model are performances predictions. Indeed, what is the practical consequence of a given deviation for a microstructure parameter? The microstructure parameter deviation propagations to the 3D microstructure scale electrochemical response have been then quantified for different charge rates. This defines a threshold for the microstructure parameters FOV for a desired maximum deviation of the electrochemical response. Such deviation propagation analysis is analogous to error propagation analysis and is necessary to determine the relevance of microstructure scale model predictions for macroscale predictions. Electrochemical model shows cell representative section areas are increasing with C-rate, due to higher in-plane heterogeneities, indicating larger FOVs are required specifically for fast charge modeling. Therefore, we introduced the novel concept of electrochemical RVE (eRVE) that is a function of the operating conditions (thus defined as a dynamic RVE), with an increasing dependence with the C-rate. Representativity analysis of the investigated cell determined a FOV of 144.4 x 54.4 m2 is large enough to establish a convergence on the representative section areas for low to intermediate C-rate (=2.5C), but not large enough to conclude for higher rates. This work aims to emphasize the importance of representativity analysis for LIB electrode microstructures, as it is required to estimate the error, and thus the relevance, of microstructure parameters intended to be used in macroscale models. The methodology and results can help researchers to select the relevant imaging and associated FOV required to provide accurate enough microstructure parameters.

ADVANCED PROPULSION SYSTEMS↗

DOE OSTI2022

Physics-constrained deep learning of nonlinear normal modes of spatiotemporal fluid flow dynamics

In this study, we present a physics-constrained deep learning method to discover and visualize from data the invariant nonlinear normal modes (NNMs) which contain the spatiotemporal dynamics of the fluid flow potentially containing strong nonlinearity. Specifically, we develop a NNM-physics-constrained convolutional autoencoder (NNM-CNN-AE) integrated with a multi-temporal-step dynamics prediction block to learn the nonlinear modal transformation, the NNMs containing the spatiotemporal dynamics of the flow, and reduced-order reconstruction and long-time future-state prediction of the flow fields, simultaneously. In test cases, we apply the developed method to analyze different flow regimes past a cylinder, including laminar flows with low Reynolds number in transient and steady states (RD = 100) and high Reynolds number flow (RD = 1000), respectively. The results indicate that the identified NNMs are able to reveal the nonlinear spatiotemporal dynamics of these flows, and the NNMs-based reduced-order modeling consistently achieves better accuracy with orders of magnitudes smaller errors in construction and prediction of the nonlinear velocity and vorticity fields, compared to the linear proper orthogonal decomposition (POD) method and the Koopman-constrained-CNN-AE using the same number or dimension of modes. We perform an analysis of the modal energy distribution of NNMs and find that compared to POD modes, the few fundamental NNMs capture a very high level of total energy of the flow, which is advantageous for reduced-order modeling and representation of the complex flows. Finally, we discuss the potentials and limitations of the presented method.

Mechanics↗

DOE OSTI2024

Designing ITER motional Stark effect line shift (MSE-LS) spectrometers

As a part of ITER beam aided diagnostics, the design of Motional Stark Effect (MSE) diagnostic observing the emission from the Balmer-α line is underway. The physics of Stark splitting shows that the Stark manifold is polarization dependent, and the energy splitting results in a line shift proportional to the electric field. Due to the challenges of maintaining the calibration of the plasma facing mirrors in ITER, the conventional MSE polarimetry measurement technique is replaced with a spectral approach that is deemed more favorable in the ITER environment. The MSE line shift (LS) diagnostic is designed to quantify the Lorentz electric field magnitude by measuring the Stark manifold using visible spectroscopy. In the presence of large magnetic fields and high energy heating beams of 1 MeV, the expected Stark splitting is much larger than in typical devices. The MSE-LS design has unique challenges requiring careful consideration and modeling of its viewing geometry and photon budget. The MSE-LS approach on ITER is promising but has stringent demands on the allowable errors for the statistical and systematic fitting uncertainties. In this study, a full system model and numerical simulations of data for each sightline are completed. For a range of optical transmission fractions, photon noise analysis is conducted to determine the statistical uncertainties. This provides guidance on the spectrometer throughput, dispersion at the detector, optics, and other design choices. A conceptual design of a high throughput spectrometer with a volume phase transmission grating is presented.

Instruments & Instrumentation↗

DOE OSTI2026

Design of a Molten Salt Flow Cell for Combined Absorbance and Laser-Induced Breakdown Spectroscopy for Online Measurements

A novel flow cell allowing for multiple optical spectroscopy measurements on flowing molten salts was designed, and demonstrative calibrations of impurities in aqueous samples were performed. Online compositional measurements of molten salts are of high interest to monitor the state of relevant solar and nuclear systems. Here, the Spectroscopic Configuration for Optical Real-Time Characterization of High-Temperature (SCORCH) fluids cell was designed to meet this need by providing optical access to a high-temperature molten salt sample stream without physical contact between the sample and window materials. Laser-induced breakdown spectroscopy (LIBS) was utilized to quantify Li, Cr, Fe, Ni, Sr, and Pr at concentrations ranging nominally from 0 to 315 mmol L −1 . Laser power, frequency, and plasma position were optimized to mitigate challenges associated with sample splashing. Univariate calibration models were built with R 2 > 0.98, percent root mean square error of cross-validation (%RMSECV) as low as 2.7%, and limits of quantification (LOQs) down to 4.1 mmol L −1 . Simultaneously, absorbance calibrations were developed for the applicable analytes (Cr, Ni, and Pr) using Beer’s law with a pathlength of 4.41 ± 0.10 mm. These models provide excellent quantification performance with R 2 > 0.999, %RMSECV as low as 0.6%, and LODs down to 0.08 mmol L −1 . Although these calibrations were performed for each spectroscopic technique separately, the two methods may be combined in the future through multivariate modeling and sensor fusion to provide more robust models with the benefits of both techniques (e.g., absorbance: oxidation state concentrations, LIBS: elemental concentration). Additionally, optimized spectrometers may be deployed to enhance sensitivity.

absorbance spectroscopy↗

DOE OSTI2025

Perturbative Stability and Error-Correction Thresholds of Quantum Codes

Topologically ordered phases are stable to local perturbations, and topological quantum error-correcting codes enjoy thresholds to local errors. We connect the two notions of stability by constructing classical statistical mechanics models for decoding general Calderbank-Shor-Steane codes and classical linear codes. Our construction encodes correction success probabilities under uncorrelated bit-flip and phase-flip errors, and simultaneously describes a generalized ℤ 2 lattice-gauge theory with quenched disorder. We observe that the clean limit of the latter is precisely the discretized imaginary-time path integral of the corresponding quantum code Hamiltonian when the errors are turned into a perturbative 𝑋 or 𝑍 magnetic field. Motivated by error-correction considerations, we define general order parameters for all such generalized ℤ 2 lattice-gauge theories, and show that they are generally lower bounded by success probabilities of error correction. For CSS codes satisfying the low-density parity-check condition and with a sufficiently large code distance, we prove the existence of a low-temperature ordered phase of the corresponding lattice-gauge theories, particularly for those lacking Euclidean spatial locality and/or when there is a nonzero code rate. We further argue that these results provide evidence for stable phases in the corresponding perturbed quantum Hamiltonians, obtained in the limit of continuous imaginary time. To do so, we distinguish space- and timelike defects in the lattice-gauge theory. A high free-energy cost of spacelike defects corresponds to a successful “memory experiment” and suppresses the energy splitting among the ground states, while a high free-energy cost of timelike defects corresponds to a successful “stability experiment” and points to a nonzero gap to local excitations.

quantum error correction↗

DOE OSTI2021

GANpiler

Focal Area(s): Rather than augment or replace physical models with machine learning, we instead preserve the existing models and augment the underlying code with faster surrogate models created by generative adversarial networks (GAN). To leverage the performance optimization, we also propose a runtime system and user interfaces that allow prediction and tracking of accumulated error as well as dynamic, per-process decision making as to which model (if any) to use for each iteration. Science Challenge: This proposal sits at the nexus of two hard problems. First, climate models based on machine learning will be, by their nature, difficult to trust once their predictions begin diverging from the consensus. Second, compilers and hardware have been making only incremental performance gains for decades. GPGPUs have provided a welcome performance boost for codes that can take advantage of them, but there is no similar technology on the horizon to provide the next performance leap.

54 ENVIRONMENTAL SCIENCES↗