Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “computational statistical mechanics”

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 451 records · Page 25
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

A machine learning approach for efficient multi-dimensional integration

Many physics problems involve integration in multi-dimensional space whose analytic solution is not available. The integrals can be evaluated using numerical integration methods, but it requires a large computational cost in some cases, so an efficient algorithm plays an important role in solving the physics problems. We propose a novel numerical multi-dimensional integration algorithm using machine learning (ML). After training a ML regression model to mimic a target integrand, the regression model is used to evaluate an approximation of the integral. Then, the difference between the approximation and the true answer is calculated to correct the bias in the approximation of the integral induced by ML prediction errors. Because of the bias correction, the final estimate of the integral is unbiased and has a statistically correct error estimation. Three ML models of multi-layer perceptron, gradient boosting decision tree, and Gaussian process regression algorithms are investigated. The performance of the proposed algorithm is demonstrated on six different families of integrands that typically appear in physics problems at various dimensions and integrand difficulties. The results show that, for the same total number of integrand evaluations, the new algorithm provides integral estimates with more than an order of magnitude smaller uncertainties than those of the VEGAS algorithm in most of the test cases.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

Stochastic modal velocity field in rough-wall turbulence

Stochastically generated instantaneous velocity profiles are used to reproduce the outer region of rough-wall turbulent boundary layers in a range of Reynolds numbers extending from the wind tunnel to field conditions. Each profile consists in a sequence of steps, defined by the modal velocities and representing uniform momentum zones (UMZs), separated by velocity jumps representing the internal shear layers. Height-dependent UMZ is described by a minimal set of attributes: thickness, mid-height elevation, and streamwise (modal) and vertical velocities. These are informed by experimental observations and reproducing the statistical behaviour of rough-wall turbulence and attached eddy scaling, consistent with the corresponding experimental datasets. Sets of independently generated profiles are reorganized in the streamwise direction to form a spatially consistent modal velocity field, starting from any randomly selected profile. The operation allows one to stretch or compress the velocity field in space, increases the size of the domain and adjusts the size of the largest emerging structures to the Reynolds number of the simulated flow. By imposing the autocorrelation function of the modal velocity field to be anchored on the experimental measurements, we obtain a physically based spatial resolution, which is employed in the computation of the velocity spectrum, and second-order structure functions. The results reproduce the Kolmogorov inertial range extending from the UMZ and their attached-eddy vertical organization to the very-large-scale motions (VLSMs) introduced with the reordering process. The dynamic role of VLSM is confirmed in the –u'w' co-spectra and in their vertical derivative, representing a scale-dependent pressure gradient contribution.

42 ENGINEERING↗
DOE OSTI2020

Measurement Uncertainty in One-Of-A-Kind Event Data Analysis

A golden standard in science is to repeat an experiment a statistically significant number of times, recording data using the same set of detectors and the same data analysis methodology. In such case experimental error includes both the range of true values generated by repetitions of the experiment, and measurement uncertainty caused by the detector. They are independent. It is a huge and too frequently used simplification, to assume that one can measure multiple repetitions of an identical experiment, resulting in identical true experimental value. Repetitions, as similar is it is experimentally achievable, have unavoidable built-in differences resulting in a range of the true values rather than in a single value. When modern, very sensitive and well calibrated measurement systems are used, this range is not negligible, and sometimes dominates over the measurement uncertainty. Range of true values depends on built-in differences in physics of the experiment. Stochastic physical processes result typically in a broader range of true values than non-stochastic processes do. Measurement uncertainty depends on a measurement method (properties of the detector not of the experiment). Modern measurement methods, including digital ones, frequently make the measurement uncertainty very small. When data from one–of –a kind experiment are analyzed, only the measurement uncertainty is reported. It provides no information about the range of true experimental values, neither about reliability of a reported data point. Reliability of a data point is in general independent from its measurement uncertainty. However, in practice reliable measurement methods frequently have high measurement uncertainty, while low reliability methods are applied to limit measurement uncertainty. Comparison of reliable data with high measurement uncertainty to not so reliable data measured with low uncertainty is discussed – in different scenarios different data analysis methods are applicable. Methods for data analysis from an experiment repeated statistically significant number of times are very well developed. They do not require a detailed expertise in physics of an experiment, nor in the properties of the measurement system used, and meaning of the reported uncertainty is well understood in any scientific community. It all changes when data from one-of-a-kind experiment is analyzed. Analyst’s expertise is required both in the physics of the experiment and in all aspects of the measurement system, all possible malfunctions. Data users must remember that only measurement uncertainty is reported from any one-of-a-kind experiment. Theory with simulations may provide estimation of expected built-in differences in the experiment, and by this of expected range of true values for a given experiment; yet measurement uncertainty can never be used in place of the range of true experimental values.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2022

Quantifying local rearrangements in three-dimensional granular materials: Rearrangement measures, correlations, and relationship to stresses

Quantifying the ways in which local particle rearrangements contribute to macroscopic plasticity is one of the fundamental pursuits of granular mechanics and soft matter physics. Here we examine local rearrangements that occur naturally during the deformation of three samples of 3D granular materials subjected to distinct boundary conditions by employing in situ x-ray measurements of particle-resolved structure and stress. We focus on five distinct rearrangement measures, their statistics, interrelationships, contributions to macroscopic deformation, repeatability, and dependence on local structure and stress. Our most significant findings are that local rearrangements (1) are correlated on a scale of three to four particle diameters, (2) exhibit volumetric strain-shear strain and nonaffine displacement-rotation coupling, (3) exhibit correlations that suggest either rearrangement repeatability or that rearrangements span multiple steps of incremental sample strain, and (4) show little dependence on local stress but correlate with quantities describing local structure, such as porosity. Our results are presented in the context of relevant plasticity theories and are consistent with recent findings suggesting that local structure may play at least as important of a role as local stress in determining the nature of local rearrangements.

58 GEOSCIENCES↗
DOE OSTI2020

Limited-view Cone Beam CT reconstruction using 3D Patch-based Supervised and Adversarial Learning [Slides]

We present a novel machine learning CNN architecture that can learn from limited data combined appropriately with physics and statistical priors (e.g., forward models and noise models). To address the limited availability of training data we adopt a 3D patch-based approach for our models. Patch-based learning is central to several image reconstruction methods and demands fewer training data than DL approaches, as a single data volume can be broken into several millions of overlapping 3D sub-volumes or patches. This creates a very large number of training sub-volumes from a limited number of overall image volumes. A 3D Generative Adversarial Networks (GAN) is then trained to remove artifacts at the sub-volume level. The combination of a sub-volume-based approach with DL allows us to exploit the richness of the latter in extracting and representing image features, while avoiding risks associated with overfitting due to limited training data.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
NASA NTRS1995

Dynamics and structure of turbulent premixed flames

In earlier work (Mantel & Bilger, 1994) the structure of the turbulent premixed flame was investigated using statistics based on conditional averaging with the reaction progress variable as the conditioning variable. The DNS data base of Trouve and Poinsot (1994) was used in this investigation. Attention was focused on the conditional dissipation and conditional axial velocity in the flame with a view to modeling these quantities for use in the conditional moment closure (CMC) approach to analysis of kinetics in premixed flames (Bilger, 1993). Two remarkable findings were made: there was almost no acceleration of the axial velocity in the flame front itself; and the conditional scalar dissipation remained as high, or higher, than that found in laminar premixed flames. The first finding was surprising since in laminar flames all the fluid acceleration occurs through the flame front, and this could be expected also for turbulent premixed flames at the flamelet limit. The finding gave hope of inventing a new approach to the dynamics of turbulent premixed flames through use of rapid distortion theory or an unsteady Bernoulli equation. This could lead to a new second order closure for turbulent premixed flames. The second finding was contrary to our measurements with laser diagnostics in lean hydrocarbon flames where it is found that conditional scalar dissipation drops dramatically below that for laminar flamelets when the turbulence intensity becomes high. Such behavior was not explainable with a one-step kinetic model, even at non-unity Lewis number. It could be due to depletion of H2 from the reaction zone by preferential diffusion. The capacity of the flame to generate radicals is critically dependent on the levels of H2 present (Bilger, et al., 1991). It seemed that a DNS computation with a multistep reduced mechanism would be worthwhile if a way could be found to make this feasible. Truly innovative approaches to complex problems often come only when there is the opportunity to work close at hand with the (in this case numerical) experimental data. Not only can one spot patterns and relationships in the data which could be important, but one can also get to know the limitations of the technique being used, so that when the next experiment is being designed it will address resolvable questions. A three-year grant from the Australian Research Council has enabled us to develop a small capability at the University of Sydney to work on DNS of turbulent reacting flow, and to analyze data bases generated at CTR. Collaboration between the University of Sydney and CTR is essential to this project and finding a workable modus operandum for this collaboration, given the constraints involved, has been a major objective of the past year's effort. The overall objectives of the project are: (1) to obtain a quantitative understanding of the dynamics of turbulent premixed flames at high turbulence levels with a view to developing improved second order closure models; and (2) to carry out new DNS experiments on turbulent premixed flames using a carefully chosen multistep reduced mechanism for the chemical kinetics, with a view to elucidating the laser diagnostic findings that are contrary to the findings for DNS using one-step kinetics. In this first year the objectives have been to make the existing CTR data base more accessible to coworkers at the University of Sydney, to make progress on understanding the dynamics of the flame in this existing CTR data base, and to carefully construct a suitable multistep reduced mechanism for use in a new set of DNS experiments on turbulent premixed flames.

Bilger, R. W.↗
DOE OSTI2021

Data-Driven Learning for the Mori--Zwanzig Formalism: A Generalization of the Koopman Learning Framework

A theoretical framework which unifies the conventional Mori--Zwanzig formalism and the approximate Koopman learning of deterministic dynamical systems from noiseless observation is presented. In this framework, the Mori--Zwanzig formalism, developed in statistical mechanics to tackle the hard problem of construction of reduced-order dynamics for high-dimensional dynamical systems, can be considered as a natural generalization of the Koopman description of the dynamical system. We next show that, similar to the approximate Koopman learning methods, data-driven methods can be developed for the Mori--Zwanzig formalism with Mori's linear projection operator. We have developed two algorithms to extract the key operators, the Markov and the memory kernel, using time series of a reduced set of observables in a dynamical system. We have adopted the Lorenz `96 system as a test problem and solved for the above operators. These operators exhibit complex behaviors, which are unlikely to be captured by traditional modeling approaches in Mori--Zwanzig analysis. The nontrivial generalized fluctuation-dissipation relationship, which relates the memory kernel with the two-time correlation statistics of the orthogonal dynamics, was numerically verified as a validation of the solved operators. Here we present numerical evidence that the generalized Langevin equation, a key construct in the Mori--Zwanzig formalism, is more advantageous in predicting the evolution of the reduced set of observables than the conventional approximate Koopman operators.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Multifidelity validation of digital surrogates using variable-density turbulent mixing models

In this study, ensembles of experimental data are presented and utilized to compare and validate two models used in the simulation of variable-density (Atwood = 0.22), compressible turbulent mixing. Though models of this kind (Reynolds averaged NavierStokes and large-eddy simulations) have been validated extensively with more canonical flows in previous studies, here the present approach offers novelty in the complexity of the geometry, the ensemble-based validation, and the uniformity of the computational framework on which the models are tested. Moreover, all experimental and computational tasks were completed by the authors which has led to a tightly coupled experimental configuration with its “digital twin.” The experimental divergent-shock-tube facility and its data acquisition methods are described and replicated in simulation space. A 2D Euler model which neglects the turbulent mixing at the interface is optimized to experimental data using a Gaussian process. This model then serves as the basis for both the 2D RANS and 3D LES studies that make comparisons to the mixing-layer data from the experiment. A relatively simple RANS model is shown to produce good agreement with experimental data only at late flow development times. The LES ensembles generally show good agreement with experimental data but display sensitivity to the characterization of initial conditions. Resolution-dependent behavior is also observed for certain higher-order statistics of interest. Overall, the LES model successfully captures the effects of divergent geometry, compressibility, and combined nonlinear instabilities inherent to the problem. The successful prediction of mixing width and its growth rate highlight the existence of three distinct regimes in the development of the instability, each with similarities to previously studied instabilities.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2023

Projection pursuit adaptation on polynomial chaos expansions

Here, the present work addresses the issue of accurate stochastic approximations in high-dimensional parametric space using tools from uncertainty quantification (UQ). The basis adaptation method and its accelerated algorithm in polynomial chaos expansions (PCE) were recently proposed to construct low-dimensional approximations adapted to specific quantities of interest (QoI). The present paper addresses one difficulty with these adaptations, namely their reliance on quadrature point sampling, which limits the reusability of potentially expensive samples. Projection pursuit (PP) is a statistical tool to find the “interesting” projections in high-dimensional data and thus bypass the curse-of-dimensionality. In the present work, we combine the fundamental ideas of basis adaptation and projection pursuit regression (PPR) to propose a novel method to simultaneously learn the optimal low-dimensional spaces and PCE representation from given data. While this projection pursuit adaptation (PPA) can be entirely data-driven, the constructed approximation exhibits mean-square convergence to the solution of an underlying governing equation and thus captures the supports and probability distributions associated with the physics constraints. The proposed approach is demonstrated on a borehole problem and a structural dynamics problem, demonstrating the versatility of the method and its ability to discover low-dimensional manifolds with high accuracy with limited data. In addition, the method can learn surrogate models for different quantities of interest while reusing the same data set.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Efficient simulation of low-temperature physics in one-dimensional gapless systems

Here, we discuss the computational efficiency of the finite-temperature simulation with minimally entangled typical thermal states (METTS). To argue that METTS can be efficiently represented as matrix product states, we present an analytic upper bound for the average entanglement Rényi entropy of METTS for a Rényi index 0 < q ≤ 1. In particular, for one-dimensional (1D) gapless systems described by conformal field theories, the upper bound scales as O⁡(cN 0 ⁢log⁡β) where c is the central charge and N is the system size. Furthermore, we numerically find that the average Rényi entropy exhibits a universal behavior characterized by the central charge and is roughly given by half of the analytic upper bound. Based on these results, we show that METTS can provide a speedup compared to employing the purification method to analyze thermal equilibrium states at low temperatures in 1D gapless systems.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2022

A better understanding of the mechanics of borehole breakout utilizing a finite strain gradient-enhanced micropolar continuum model

Borehole breakout denotes the failure in rock mass subjected to drilling, caused by stress concentrations exceeding the material strength. Depending on the material properties, the preexisting in situ stress state, and the borehole dimensions, different types of borehole breakout, such as spiral-shaped breakout or v-shaped breakout, are distinguished in the literature. In the present work, we address the influence of the material properties on the predicted borehole breakout mode in a comprehensive finite element study. To this end, herein we employ a gradient-enhanced micropolar damage-plasticity model based on the Mohr–Coulomb strength criterion, formulated in the finite strain regime, which is calibrated based on experimental results from plane strain compression tests on Red Wildmoor sandstone. In the numerical study, the influence of the in situ stress state, the material friction angle, plastic dilation, post peak residual strength, and the inherent material length scale parameters are investigated. Thereby, we demonstrate that depending on the material parameters, substantially different failure modes, characterized by strongly localized shear bands or diffuse failure zones, are predicted. It is shown that in particular the brittleness of the material in the post peak regime has a major influence on the predicted breakout type. Moreover, a statistical validation of the results is obtained by considering different random field distributions of the initial material strength.

58 GEOSCIENCES↗
DOE OSTI2020

A generalized approximate control variate framework for multifidelity uncertainty quantification

We describe and analyze a variance reduction approach for Monte Carlo (MC) sampling that accelerates the estimation of statistics of computationally expensive simulation models using an ensemble of models with lower cost. These lower cost models — which are typically lower fidelity with unknown statistics — are used to reduce the variance in statistical estimators relative to a MC estimator with equivalent cost. We derive the conditions under which our proposed approximate control variate framework recovers existing multifidelity variance reduction schemes as special cases. We demonstrate that existing recursive/nested strategies are suboptimal because they use the additional low-fidelity models only to efficiently estimate the unknown mean of the first low-fidelity model. As a result, they cannot achieve variance reduction beyond that of a control variate estimator that uses a single low-fidelity model with known mean. However, there often exists about an order-of-magnitude gap between the maximum achievable variance reduction using all low-fidelity models and that achieved by a single low-fidelity model with known mean. We show that our proposed approach can exploit this gap to achieve greater variance reduction by using non-recursive sampling schemes. The proposed strategy reduces the total cost of accurately estimating statistics, especially in cases where only low-fidelity simulation models are accessible for additional evaluations. Several analytic examples and an example with a hyperbolic PDE describing elastic wave propagation in heterogeneous media are used to illustrate the main features of the methodology.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

Bipartite mutual information in classical many-body dynamics

Information theoretic measures have helped to sharpen our understanding of many-body quantum states. As perhaps the most well-known example, the entanglement entropy (or more generally, the bipartite mutual information) has become a powerful tool for characterizing the dynamical growth of quantum correlations. By contrast, although computable, the bipartite mutual information (MI) is almost never explored in classical many particle systems; this owes in part to the fact that computing the MI requires keeping track of the evolution of the full probability distribution, a feat which is rarely done (or thought to be needed) in classical many-body simulations. Here, we utilize the MI to analyze the spreading of information in 1D elementary cellular automata (CA). Broadly speaking, we find that the behavior of the MI in these dynamical systems exhibits a few different types of scaling that roughly correspond to known CA universality classes. Of particular note is that we observe a set of automata for which the MI converges parametrically slowly to its thermodynamic value. We develop a microscopic understanding of this behavior by analyzing a two-species model of annihilating particles moving in opposite directions. Furthermore, our work suggests the possibility that information theoretic tools such as the MI might enable a more fine-grained characterization of classical many-body states and dynamics.

Cellular automata↗
DOE OSTI2026

Transient hydroperoxyalkyl intermediates (•QOOH) in isopentane oxidation. II. Isomer-resolved unimolecular dynamics

Transient carbon-centered hydroperoxyalkyl intermediates (•QOOH) in isopentane oxidation are characterized by their time- and energy-resolved unimolecular dissociation dynamics to hydroxyl (OH) and cyclic ether products. Two distinct •QOOH isomers are examined with radical sites at a primary carbon of one of the methyl groups (β-Me) or a secondary carbon (β-Et) of the ethyl group. Energy-dependent unimolecular rates are obtained from the time-dependent appearance of OH products for the two isomers and compared with statistical microcanonical rates computed using RRKM theory, including heavy-atom tunneling, based on high-level theoretical calculations. A benchmark-corrected approach is utilized to compute high-accuracy stationary-point energies, most importantly, transition-state barriers, for the •QOOH_Me and •QOOH_Et isomers in isopentane oxidation, building on higher-level reference calculations for the oxidation of ethane (C 2 H 5 O 2 ) and propane (C 3 H 7 O 2 ), respectively. Here, the measured rates are compared with RRKM calculations incorporating the benchmark-corrected transition-state parameters, a vibrationally adiabatic multidimensional hindered-rotor treatment of key torsions, and quantum tunneling. Agreement between experiment and theory validates the statistical description and shows faster decay for •QOOH_Et due to its lower barrier. Both β-QOOH isomers decay almost exclusively to OH + cyclic ether products under the conditions studied.

Oxidation processes↗
DOE OSTI2021

Quadrupole-mediated dielectric response and the charge-asymmetric solvation of ions in water

Treating water as a linearly responding dielectric continuum on molecular length scales allows very simple estimates of the solvation structure and thermodynamics for charged and polar solutes. While this approach can successfully account for basic length and energy scales of ion solvation, computer simulations indicate not only its quantitative inaccuracies but also its inability to capture some basic and important aspects of microscopic polarization response. Here, we consider one such shortcoming, a failure to distinguish the solvation thermodynamics of cations from that of otherwise-identical anions, and we pursue a simple, physically inspired modification of the dielectric continuum model to address it. The adaptation is motivated by analyzing the orientational response of an isolated water molecule whose dipole is rigidly constrained. Its free energy suggests a Hamiltonian for dipole fluctuations that accounts implicitly for the influence of higher-order multipole moments while respecting constraints of molecular geometry. Finally, we propose a field theory with the suggested form, whose nonlinear response breaks the charge symmetry of ion solvation. An approximate variational solution of this theory, with a single adjustable parameter, yields solvation free energies that agree closely with simulation results over a considerable range of solute size and charge.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

Oxidation Dynamics of Supported Catalytic Cu Clusters: Coupling to Fluxionality

Copper oxide nanoclusters have a wide range of catalytic applications, such as the selective oxidation of hydrocarbons. O 2 binding to the catalyst, activation, and release upon reagent oxidation are key events in these catalytic chemistries. Furthermore, these events are expected to be accompanied by significant structural changes of the Cu clusters, because O atoms integrate into the cluster, rather than bind to its surface. Topping the complexity of the problem, partially oxidized Cu clusters are known to exhibit strong fluxionality and feature diverse and interconverting structures and oxygen contents in conditions of oxidative dehydrogenation (ODH). Hence, a significant dynamic coupling between the “hot” O 2 molecule impacting the cluster at reaction temperatures and the cluster fluxionality can be expected. In this work, we focus on the dynamics of dioxygen integration into a partially oxidized Cu cluster supported on hydroxylated amorphous alumina–a system recently reported to be an exceptionally selective catalyst for cyclohexane ODH with very little CO produced, whose mechanistic underpinnings are of utmost interest. The statistics over a swarm of adsorption and scattering trajectories where O 2 hits various sites on the cluster at reaction temperature shows that the O 2 binding does not only follow the minimal energy paths. O 2 also rarely integrates into the cluster in a single step and instead first binds to a single Cu atom via either an η 1 -O 2 or an η 2 -O 2 mode. Surprisingly, this step often has a higher barrier than the subsequent O 2 integration and dissociation, which in turn take multiple steps and complete the oxidation process. Dynamic trajectories starting from the key transition state of integration of the adsorbed O 2 can also lead to different intermediate structures during or right after the dissociation, due to the energy released from the transition state and the thermal intracluster effects. From these activated O 2 chemisorbed structures, O 2 dissociation occurs with moderate barriers (~0.5 eV), producing multiple final oxidized Cu 4 O 4 states. Hence, a diversity of reaction profiles for the attack of supported Cu cluster by O 2 emerges due to the dynamic effects, with implications for mechanisms, kinetic models, and catalyst design principles.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

An Open Benchmark of One Million High-Fidelity Cislunar Trajectories

Cislunar space spans from geosynchronous altitudes to beyond the Moon and will underpin future exploration, science, and security operations. We describe and release an open dataset of one million numerically propagated cislunar trajectories generated with the open-source Space Situational Awareness Python package (SSAPy). The model includes high-degree Earth/Moon gravity, solar gravity, and Earth/Sun radiation pressure; other planetary gravities are omitted by design for computational efficiency. Initial conditions uniformly sample commonly used osculating-element ranges, and each trajectory is propagated for up to six years under a single, fixed start epoch. The dataset is intended as a reusable benchmark for method development (e.g., space domain awareness, navigation, and machine-learning pipelines), a reference library for statistical studies of orbit families, and a starting point for community-driven extensions (e.g., alternative epochs). We report empirically observed stability trends (e.g., a band near ~5 GEO and persistence of some co-orbital classes including L4/L5 librators) as dataset descriptors rather than new dynamical results. The chief contribution is the scale, fidelity, organization (CSV/HDF5 with full state time series and metadata), and open availability, which together lower the barrier to comparative and data-driven studies in the cislunar regime.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

A hybrid deep neural operator/finite element method for ice-sheet modeling

One of the most challenging and consequential problems in climate modeling is to provide probabilistic projections of sea level rise. A large part of the uncertainty of sea level projections is due to uncertainty in ice sheet dynamics. At the moment, accurate quantification of the uncertainty is hindered by the cost of ice sheet computational models. In this work we develop a hybrid approach to approximate existing ice sheet models at a fraction of their cost. Our approach consists of replacing the finite element model for the momentum equations for the ice velocity, the most expensive part of an ice sheet model, with a Deep Operator Network, while we retain a classic finite element discretization for the evolution of the ice thickness. We show that the resulting hybrid model is very accurate and it is an order of magnitude faster than the traditional finite element model. Further, a distinctive feature of the proposed model, compared to other neural network approaches, is that it can handle high-dimensional parameter spaces (parameter fields) such as the basal friction at the bed of the glacier and can therefore be used for generating samples for uncertainty quantification. Further, we study the impact of hyper-parameters, number of unknowns and correlation length of the parameter distribution on the training and accuracy of the Deep Operator Network on a synthetic ice sheet model. We then target the evolution of the Humboldt glacier in Greenland and show that our hybrid model can provide accurate statistics of the glacier mass loss and can be effectively used to accelerate the quantification of uncertainty.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗