Search NASA⌕ Search

SEARCH · Search NASA

Results for “error budget”

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 55 records · Page 3
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Constraining gravity with a new precision 𝐸 𝐺 estimator using Planck + SDSS BOSS data

The 𝐸 𝐺 statistic is a discriminating probe of gravity developed to test the prediction of general relativity (GR) for the relation between gravitational potential and clustering on the largest scales in the observable Universe. We present a novel high-precision estimator for the 𝐸 𝐺 statistic using CMB lensing and galaxy clustering correlations that carefully matches the effective redshifts across the different measurement components to minimize corrections. A suite of detailed tests is performed to characterize the estimator’s accuracy, its sensitivity to assumptions and analysis choices, and the non-Gaussianity of the estimator’s uncertainty is characterized. After finalization of the estimator, it is applied to Planck CMB lensing and SDSS CMASS and LOWZ galaxy data. We report the first harmonic space measurement of 𝐸 𝐺 using the LOWZ sample and CMB lensing and also updated constraints using the final CMASS sample and the latest Planck CMB lensing map. We find $\hat{𝐸}$$^{Planck+CMASS}_{𝐺}$ = 0.3⁢6$^{+0.06}_{−0.05}$⁢(68.27%) and $\hat{𝐸}$$^{Planck+LOWZ}_{𝐺}$ = 0.4⁢0$^{+0.11}_{−0.09}$⁢(68.27%), with additional subdominant systematic error budget estimates of 2% and 3%, respectively. Using Ω m,0 constraints from Planck and SDSS BAO observations, Λ⁢CDM-GR predicts 𝐸$^{GR}_ {𝐺}$⁡(𝑧 =0.555) = 0.401 ± 0.005 and 𝐸$^{GR}_{𝐺}$⁡(𝑧 =0.316) = 0.452 ± 0.005 at the effective redshifts of the CMASS and LOWZ based measurements. We report the measurement to be in good statistical agreement with the Λ⁢CDM-GR prediction and report that the measurement is also consistent with the more general GR prediction of scale independence for 𝐸 𝐺 . Furthermore, this work provides a carefully constructed and calibrated statistic with which 𝐸 𝐺 measurements can be confidently and accurately obtained with upcoming survey data.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Improving cosmological analyses of HI clustering by reducing stochastic noise

High-number-density tracers of large-scale structure, such as the HI-rich galaxies measured by 21 cm intensity mapping, have low sampling noise, making them particularly promising as cosmological probes. At large scales, this sampling noise can be subdominant to other scale-independent contributions to the power spectrum; such contributions arise from nonlinear bias, and exceed the sampling noise if at least one of the associated bias coefficients is sufficiently large. This has important consequences for cosmological constraints obtained from such tracers, since it indicates that using the power spectrum does not lead to optimal constraints even in the linear regime. In this paper, we provide a conservative estimate of the possible improvement in constraining power of a 21 cm survey if one were to use an optimal analysis strategy (such as field-level analysis), where only the true sampling noise enters the error budget. We find that improvements in uncertainties on some cosmological parameters can be as large as 50%, depending on redshift, foreground cleaning efficiency, scales used in the analysis, and instrumental noise. One byproduct of our work is measurements of bias parameters and stochasticity for neutral hydrogen in the IllustrisTNG simulation over a wide range of redshifts; we provide simple fitting formulas for these measurements. Furthermore, our results motivate further exploration of new optimal analysis techniques and provide important insights into the constraining power of current and future 21 cm surveys.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Analysis of DESI × DES using the Lagrangian effective theory of LSS

In this work we use Lagrangian perturbation theory to analyze the harmonic space galaxy clustering signal of the Bright Galaxy Survey (BGS) and luminous red galaxies (LRGs) targeted by the dark energy spectroscopic instrument (DESI), combined with the galaxy-galaxy lensing signal measured around these galaxies using Dark Energy Survey Year 3 source galaxies. The BGS and LRG galaxies are extremely well characterized by DESI spectroscopy and, as a result, lens galaxy redshift uncertainty and photometric systematics contribute negligibly to the error budget of our “2 × 2-point” analysis. On the modeling side, this work represents the first application of the SPINOSAURUS code, implementing an effective field theory model for galaxy intrinsic alignments, and we additionally introduce a new scheme ( MAIAR ) for marginalizing over the large uncertainties in the redshift evolution of the intrinsic alignment signal. Furthermore, this is the first application of a hybrid effective field theory model for galaxy bias based on the A emulus 𝜈 simulations. Our main result is a measurement of the amplitude of the lensing signal, 𝑆 8 = 𝜎 8 ⁢(Ω 𝑚 /0.3) 0.5 = 0.85⁢0$^{+0.042}_{−0.050}$, consistent with values of this parameter derived from the primary cosmic microwave background. This constraint is artificially improved by a factor of 51% if we assume a more standard, but restrictive parametrization for the redshift evolution and sample dependence of the intrinsic alignment signal, and 63% if we additionally assume the nonlinear alignment model. Furthermore, we show that when fixing the cosmological model to the best-fit values from Planck PR4 there is > 5⁢𝜎 evidence for a deviation of the evolution of the intrinsic alignment signal from the functional form that is usually assumed in cosmic shear and galaxy-galaxy lensing studies.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Laboratory demonstration of the prediction of wind-blown turbulence by adaptive optics at 8 kHz with use of LQG control

The low-latency adaptive optical mirror system (LLAMAS) is designed to push the limits on achievable latencies and frame rates. It has 21 subapertures across its pupil. Here, a reformulated version of the linear quadratic Gaussian (LQG) method predictive Fourier control is implemented in LLAMAS; for all modes, it takes just 30 µs to compute. In the testbed, a turbulator mixes hot and ambient air to produce wind-blown turbulence. Wind prediction clearly improves correction when compared to an integral controller. Closed-loop telemetry shows that wind-predictive LQG removes the characteristic “butterfly” and reduces temporal error power by up to a factor of three for mid-spatial frequency modes. Strehl changes seen in focal plane images are consistent with telemetry and the system error budget.

47 OTHER INSTRUMENTATION↗
DOE OSTI2023

Mitigation of the Brighter-Fatter Effect in the LSST Camera

Thick, fully depleted charge-coupled devices (CCDs) are known to exhibit non-linear behavior at high signal levels due to the dynamic behavior of charges collecting in the potential wells of pixels, called the brighter-fatter effect (BFE). This particularly impacts bright calibration stars, which appear larger than their intrinsic shape, creating a flux-dependent point-spread function (PSF) that if left unmitigated, could make up a large fraction of the error budget in Stage IV weak-lensing (WL) surveys such as the Legacy Survey of Space and Time (LSST). In this paper, we analyze image measurements of flat fields and artificial stars taken at different illumination levels with the LSST Camera (LSSTCam) at SLAC National Accelerator Laboratory in order to quantify this effect in the LSST Camera before and after a previously introduced correction technique. We observe that the BFE evolves anisotropically as a function of flux due to higher-order BFEs, which violates the fundamental assumption of this correction method. We then introduce a new sampling method based on a physically motivated model to account these higher-order terms in the correction, and then we test the modified correction on both datasets. We find that the new method corrects the effect in flat fields better than it corrects the effect in artificial stars which we conclude is the result of a unmodeled curl component of the deflection field by the correction. We use these results to define a new metric for the full-well capacity of our sensors and advise image processing strategies to further limit the impact of the effect on LSST WL science pathways.

47 OTHER INSTRUMENTATION↗
DOE OSTI2026

The FastrSHWFS Project Development Motivation, Analysis, & Test Results

The recent 2020 Decadal Survey of Astronomy and Astrophysics listed habitable exoplanet imaging with future extreme adaptive optics (AO) on 30m-class telescopes as a key priority in the coming decade. However, there is a current 100x contrast gap between the best systems today and what is needed to enable this goal. Astronomical AO is a required approach to enable ground-based diffraction-limited imaging of exoplanets on future extremely large telescopes. Time lag between the end of an exposure and the application of deformable mirror commands is a major contributor to the error budget in many AO systems, and detector read time is often a large component of this lag. We present two designs for a modified Shack Hartmann wavefront sensor (SHWFS), named Focal plane Actualized Shifted Technique Realized for a SHWFS (fastrSHWFS), to reduce the time lag component. This design steers the spot pattern at the focal plane into a rectangular or linear array with a custom aspect ratio, reducing readout time. The mask with focus yields aberrated results while the mask with tip/tilt only yields some defined spots. This essay outlines the current SHWFS concept, our fastrSHWFS theoretical solution to addressing time lag, reflection and quality analysis of printed mask designs, and results from testing both masks on the High Contrast Testbed at Lawrence Livermore National Lab. This work follows the test of a previous fastrSHWFS design.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

The Optical Corrector for the Dark Energy Spectroscopic Instrument

The Dark Energy Spectroscopic Instrument (DESI) is currently measuring the spectra of 40 million galaxies and quasars, the largest such survey ever made to probe the nature of cosmological dark energy. The 4 m Mayall telescope at Kitt Peak National Observatory has been adapted for DESI, including the construction of a 3$_{.}^{°}$2 diameter prime focus corrector that focuses astronomical light onto a 0.8 m diameter focal surface with excellent image quality over the DESI bandpass of 360–980 nm. The wide-field corrector includes six lenses, as large as 1.1 m in diameter and as heavy as 237 kilograms, including two counterrotating wedged lenses that correct for atmospheric dispersion over zenith angles from 0° to 60°. The lenses, cells, and barrel assembly all meet precise alignment tolerances on the order of tens of microns. The barrel alignment is maintained throughout a range of observing angles and temperature excursions in the Mayall dome by use of a hexapod, which is itself supported by a new cage, ring, and truss structure. In this paper we describe the design, fabrication, and performance of the new corrector and associated structure, focusing on how they meet DESI requirements. In particular, we describe the prescription and specifications of the lenses, design choices and error budgeting of the barrel assembly, stray light mitigations, and integration and test at the Mayall telescope. We conclude with some validation highlights that demonstrate the successful corrector on-sky performance, and we list some lessons learned during the multiyear fabrication phase.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2020

RELICS: A Very Large ( θ E ~ 40") Cluster Lens—RXC J0032.1+1808

Extensive surveys with the Hubble Space Telescope over the past decade, targeting some of the most massive clusters in the sky, have uncovered dozens of galaxy cluster strong lenses. The massive cluster strong-lens scale is typically θ E ~ 10'' to ~30''–35'', with only a handful of clusters known with Einstein radii θ E ~ 40'' or above (for z source = 2, nominally). Here we report another very large cluster lens, RXC J0032.1+1808 (z = 0.3956), the second-richest cluster in the redMapper cluster catalog and the 85th most massive cluster in the Planck Sunyaev–Zel'dovich catalog. With our light-traces-mass and fully parametric approaches, we construct strong-lensing models based on 18 multiple images of five background galaxies newly identified in the Hubble data, mainly from the Reionization Lensing Cluster Survey (RELICS), in addition to a known sextuply imaged system in this cluster. Furthermore, we compare these models to Lenstool and GLAFIC models that were produced independently as part of the RELICS program. All models reveal a large effective Einstein radius of θ E ≃ 40'' (z source = 2), owing to the obvious concentration of substructures near the cluster center. Although RXC J0032.1+1808 has a very large critical area and high lensing strength, only three magnified high-redshift candidates are found within the field targeted by RELICS. Nevertheless, we expect many more high-redshift candidates will be seen in wider and deeper observations with Hubble or the James Webb Space Telescope. Finally, the comparison between several algorithms demonstrates that the total error budget is largely dominated by systematic uncertainties.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Hierarchical Inference of the Lensing Convergence from Photometric Catalogs with Bayesian Graph Neural Networks

Abstract We present a Bayesian graph neural network (BGNN) that can estimate the weak lensing convergence ( κ ) from photometric measurements of galaxies along a given line of sight (LOS). The method is of particular interest in strong gravitational time-delay cosmography (TDC), where characterizing the “external convergence” ( κ ext ) from the lens environment and LOS is necessary for precise Hubble constant ( H 0 ) inference. Starting from a large-scale simulation with a κ resolution of ∼1′, we introduce fluctuations on galaxy–galaxy lensing scales of ∼1″ and extract random sight lines to train our BGNN. We then evaluate the model on test sets with varying degrees of overlap with the training distribution. For each test set of 1000 sight lines, the BGNN infers the individual κ posteriors, which we combine in a hierarchical Bayesian model to yield constraints on the hyperparameters governing the population. For a test field well sampled by the training set, the BGNN recovers the population mean of κ precisely and without bias (within the 2 σ credible interval), resulting in a contribution to the H 0 error budget well under 1%. In the tails of the training set with sparse samples, the BGNN, which can ingest all available information about each sight line, extracts a stronger κ signal compared to a simplified version of the traditional method based on matching galaxy number counts, which is limited by sample variance. Our hierarchical inference pipeline using BGNNs promises to improve the κ ext characterization for precision TDC. The code is available as a public Python package, Node to Joy ⏬ .

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

Binning is Sinning (Supernova Version): The Impact of Self-calibration in Cosmological Analyses with Type Ia Supernovae

Recent cosmological analyses (e.g., JLA, Pantheon) of Type Ia supernovae (SNe Ia) have propagated systematic uncertainties into a covariance matrix and either binned or smoothed the systematic uncertainty vectors in redshift space. We demonstrate that systematic error budgets of these analyses can be improved by a factor of ~ 1.5 × with the use of unbinned and unsmoothed covariance matrices. To understand this, we employ a separate approach that simultaneously fits for cosmological parameters and additional self-calibrating scale parameters that constrain the size of each systematic. We show that the covariance-matrix approach and scale-parameter approach indeed yield equivalent results, implying that in both cases the data can self-calibrate certain systematic uncertainties, but that this ability is hindered when information is binned or smoothed in redshift space. We review the top systematic uncertainties in current analyses and find that the reduction of systematic uncertainties in the unbinned case depends on whether a systematic is solely degenerate with the cosmological model in redshift space or whether it can be described by additional correlations between supernova properties and luminosity. Furthermore, we show that the power of self-calibration increases with the size of the data set, which presents a tremendous opportunity for upcoming analyses of photometrically classified samples, like those of Legacy Survey of Space and Time (LSST) and the Nancy Grace Roman Telescope (NGRST). However, to take advantage of self-calibration in large, photometrically classified samples, we must first address the issue that binning is required in currently used photometric analysis methods.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

A Systems Approach to Estimating the Uncertainty Limits of X-Ray Radiographic Metrology

Micro- and nanomanufacturing capabilities have rapidly expanded over the past decade to include complex three-dimensional (3D) structure fabrication; however, the metrology required to accurately assess these processes via part inspection and characterization has struggled to keep pace. X-ray computed tomography (CT) is considered an ideal candidate for providing the critically needed metrology on the smallest scales, especially internal features, or inaccessible regions. X-ray CT supporting micro- and nanomanufacturing often push against the poorly understood resolution and variation limits inherent to the machines, which can distort or hide fine structures. In this study, we have developed and experimentally verify a comprehensive analytical uncertainty propagation signal variation flow graph (SVFG) model for X-ray radiography in this work to better understand resolution and image variability limits on the small scale. The SVFG approach captures, quantifies, and predicts variations occurring in the system that limit metrology capabilities, particularly in the micro/nanodomain. This work is the first step to achieving full uncertainty modeling of CT reconstructions and provides insight into improving X-ray attenuation imaging systems. The SVFG methodology framework is applied to generate a complete basis set of functions describing the major sources of variation in radiographs. Five models are identified, covering variation in energy, intensity, length, blur, and position. Radiographic system experiments are defined to measure the parameters required by the SVFGs. Best practices are identified for these measurements. The SVFG models are confirmed via direct measurement of variation to predict variation within 30% on average.

47 OTHER INSTRUMENTATION↗
DOE OSTI2021

Subtropical Eastern North Pacific SST bias in earth system models

Karthik Balaguru (PNNL), Luke Van Roekel (LLNL), L. Ruby Leung (PNNL) and Milena Veneziani (LLNL) Abstract: This study systematically evaluates the warm SST bias in the Subtropical Eastern North Pacific (SENP), a problem plaguing most CMIP6 models, using the Energy Exascale Earth System Model version 1 (E3SM). Compared to the model at its standard resolution (1o atmosphere, 30-60 km ocean), the SST bias improves in several regions at high resolution (0.25o atmosphere, 18-6 km ocean) except in subtropical eastern boundary regions, such as the SENP. A mixed layer heat budget analysis reveals that errors in the surface radiative fluxes and horizontal heat advection are primarily responsible for the bias. Analysis of HighResMIP models indicates that the lack of improvement in the SENP SST bias with resolution is not native to E3SM alone. A larger CMIP6 inter-model correlation of SST bias with biases in surface winds than radiation highlights the important role of large-scale circulation biases in the SENP SST bias in coupled models.

Balaguru, Karthik↗
DOE OSTI2024

A Case Study Investigating the Low Summertime CAPE Behavior in the Global Forecast System

Convective available potential energy (CAPE) is an important index for storm forecasting. Recent versions (v15.2 and v16) of the Global Forecast System (GFS) predict lower values of CAPE during summertime in the continental United States than analysis and observation. We conducted an evaluation of the GFS in simulating summertime CAPE using an example from the Unified Forecast System Case Study collection to investigate the factors that lead to the low CAPE bias in GFS. Specifically, we investigated the surface energy budget, soil properties, and near-surface and upper-level meteorological fields. Results show that the GFS simulates smaller surface latent heat flux and larger surface sensible heat flux than the observations. This can be attributed to the slightly drier-than-observed soil moisture in the GFS that comes from an offline global land data assimilation system. The lower simulated CAPE in GFS v16 is related to the early drop of surface net radiation with excessive boundary layer cloud after midday when compared with GFS v15.2. A moisture-budget analysis indicates that errors in the large-scale advection of water vapor does not contribute to the dry bias in the GFS at low levels. Common Community Physics Package single-column model (SCM) experiments suggest that with realistic initial vertical profiles, SCM simulations generate a larger CAPE than runs with GFS IC. SCM runs with an active LSM tend to produce smaller CAPE than that with prescribed surface fluxes. Note that the findings are only applicable to this case study. Including more warm-season cases would enhance the generalizability of our findings.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

Energy-conserving physics for nonhydrostatic dynamics in mass coordinate models

Abstract. Motivated by reducing errors in the energy budget related to enthalpy fluxes within the Energy Exascale Earth System Model (E3SM), we study several physics–dynamics coupling approaches. Using idealized physics, a moist rising bubble test case, and the E3SM's nonhydrostatic dynamical core, we consider unapproximated and approximated thermodynamics applied at constant pressure or constant volume. With the standard dynamics and physics time-split implementation, we describe how the constant-pressure and constant-volume approaches use different mechanisms to transform physics tendencies into dynamical motion and show that only the constant-volume approach is consistent with the underlying equations. Using time step convergence studies, we show that the two approaches both converge but to slightly different solutions. We reproduce the large inconsistencies between the energy flux internal to the model and the energy flux of precipitation when using approximate thermodynamics, which can only be removed by considering variable latent heats, both when computing the latent heating from phase change and when applying this heating to update the temperature. Finally, we show that in the nonhydrostatic case, for physics applied at constant pressure, the general relation that enthalpy is locally conserved no longer holds. In this case, the conserved quantity is enthalpy plus an additional term proportional to the difference between hydrostatic pressure and full pressure.

58 GEOSCIENCES↗
DOE OSTI2020

Conserving Land–Atmosphere Synthesis Suite (CLASS)

Accurate estimates of terrestrial water and energy cycle components are needed to better understand climate processes and improve models’ ability to simulate future change. Various observational estimates are available for the individual budget terms; however, these typically show inconsistencies when combined in a budget. In this work, a Conserving Land–Atmosphere Synthesis Suite (CLASS) of estimates of simultaneously balanced surface water and energy budget components is developed. Individual CLASS variable datasets, where possible, 1) combine a range of existing variable product estimates, and hence overcome the limitations of estimates from a single source; 2) are observationally constrained with in situ measurements; 3) have uncertainty estimates that are consistent with their agreement with in situ observations; and 4) are consistent with each other by being able to solve the water and energy budgets simultaneously. First, available datasets of a budget variable are merged by implementing a weighting method that accounts both for the ability of datasets to match in situ measurements and the error covariance between datasets. Then, the budget terms are adjusted by applying an objective variational data assimilation technique (DAT) that enforces the simultaneous closure of the surface water and energy budgets linked through the equivalence of evapotranspiration and latent heat. Comparing component estimates before and after applying the DAT against in situ measurements of energy fluxes and streamflow showed that modified estimates agree better with in situ observations across various metrics, but also revealed some inconsistencies between water budget terms in June over the higher latitudes.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

SALT2 versus SALT3: updated model surfaces and their impacts on type Ia supernova cosmology

ABSTRACT For the past decade, SALT2 has been the most common model used to fit Type Ia supernova (SN Ia) light curves for dark energy analyses. Recently, the SALT3 model was released, which upgraded a number of model features but has not yet been used for measurements of dark energy. Here, we evaluate the impact of switching from SALT2 to SALT3 for a SN cosmology analysis. We train SALT2 and SALT3 on an identical training sample of 1083 well-calibrated Type Ia supernovae, ensuring that any differences found come from the underlying model framework. We publicly release the results of this training (the SALT ‘surfaces’). We then run a cosmology analysis on the public Dark Energy Survey 3-Yr Supernova data sample (DES-SN3YR), and on realistic simulations of those data. We provide the first estimate of the SN + CMB systematic uncertainty arising from the choice of SALT model framework (i.e. SALT2 versus SALT3), Δw = + 0.001 ± 0.005 – a negligible effect at the current level of dark energy analyses. We also find that the updated surfaces are less sensitive to photometric calibration uncertainties than previous SALT2 surfaces, with the average spectral energy density dispersion reduced by a factor of two over optical wavelengths. This offers an opportunity to reduce the contribution of calibration errors to SN cosmology uncertainty budgets.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Surface temperature comparison of the Arctic winter MOSAiC observations, ERA5 reanalysis, and MODIS satellite retrieval

Atmospheric model systems, such as those used for weather forecast and reanalysis production, often have significant and systematic errors in their representation of the Arctic surface energy budget and its components. The newly available observation data of the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) expedition (2019/2020) enable a range of model analyses and validation in order to advance our understanding of potential model deficiencies. In the present study, we analyze deficiencies in the surface radiative energy budget over Arctic sea ice in the ERA5 global atmospheric reanalysis by comparing against the winter MOSAiC campaign data, as well as, a pan-Arctic level-2 MODIS ice surface temperature remote sensing product. We find that ERA5 can simulate the timing of radiatively clear periods, though it is not able to distinguish the two observed radiative Arctic winter states, radiatively clear and opaquely cloudy, in the distribution of the net surface radiative budget. The ERA5 surface temperature over Arctic sea ice has a conditional error with a positive bias in radiatively clear conditions and a negative bias in opaquely cloudy conditions. The mean surface temperature error is 4°C for radiatively clear situations at MOSAiC and up to 15°C in some parts of the Arctic. The spatial variability of the surface temperature, given by 4 observation sites at MOSAiC, is not captured by ERA5 due to its spatial resolution but represented in the level-2 satellite product. The sensitivity analysis of possible error sources, using satellite products of snow depth and sea ice thickness, shows that the positive surface temperature errors during radiatively clear events are, to a large extent, caused by insufficient sea ice thickness and snow depth representation in the reanalysis system. A positive bias characterizes regions with ice thickness greater than 1.5 m, while the negative bias for thinner ice is partly compensated by the effect of snow.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Context-aware learning of hierarchies of low-fidelity models for multi-fidelity uncertainty quantification

Multi-fidelity Monte Carlo methods leverage low-fidelity and surrogate models for variance reduction to make tractable uncertainty quantification even when numerically simulating the physical systems of interest with high-fidelity models is computationally expensive. This work proposes a context-aware multi-fidelity Monte Carlo method that optimally balances the costs of training low-fidelity models with the costs of Monte Carlo sampling. It generalizes the previously developed context-aware bi-fidelity Monte Carlo method to hierarchies of multiple models and to more general types of low-fidelity models. When training low-fidelity models, the proposed approach takes into account the context in which the learned low-fidelity models will be used, namely for variance reduction in Monte Carlo estimation, which allows it to find optimal trade-offs between training and sampling to minimize upper bounds of the mean-squared errors of the estimators for given computational budgets. This is in stark contrast to traditional surrogate modeling and model reduction techniques that construct low-fidelity models with the primary goal of approximating well the high-fidelity model outputs and typically ignore the context in which the learned models will be used in upstream tasks. Further, the proposed context-aware multi-fidelity Monte Carlo method applies to hierarchies of a wide range of types of low-fidelity models such as sparse-grid and deep-network models. Numerical experiments with the gyrokinetic simulation code Gene show speedups of up to two orders of magnitude compared to standard estimators when quantifying uncertainties in small-scale fluctuations in confined plasma in fusion reactors. This corresponds to a runtime reduction from 72 days to four hours on one node of the Lonestar6 supercomputer at the Texas Advanced Computing Center.

42 ENGINEERING↗