Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Reduce order modeling”

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

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

At least 901 records · Page 50
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

HYPPO: A Surrogate-Based Multi-Level Parallelism Tool for Hyperparameter Optimization

We present a new software, HYPPO, that enables the automatic tuning of hyperparameters of various deep learning (DL) models. Unlike other hyperparameter optimization (HPO) methods, HYPPO uses adaptive surrogate models and directly accounts for uncertainty in model predictions to find accurate and reliable models that make robust predictions. Using asynchronous nested parallelism, we are able to significantly alleviate the computational burden of training complex architectures and quantifying the uncertainty. HYPPO is implemented in Python and can be used with both TensorFlow and PyTorch libraries. We demonstrate various software features on time-series prediction and image classification problems as well as a scientific application in computed tomography image reconstruction. Finally, we show that (1) we can reduce by an order of magnitude the number of evaluations necessary to find the most optimal region in the hyperparameter space and (2) we can reduce by two orders of magnitude the throughput for such HPO process to complete.

adaptation models↗
DOE OSTI2024

Membrane-based carbon capture process optimization using CFD modeling

Carbon capture is a promising option to mitigate CO2 emissions from existing coal-fired power plants, cement and steel industries, and petrochemical complexes. Among the available technologies, membrane-based carbon capture presents the lowest energy consumption, operating costs, and carbon footprint. In addition, membrane processes have important operational flexibility and response times. On the other hand, the major challenges to widespread application of this technology are related to reducing capital costs and improving membrane stability and durability. To upscale the technology into stacked flat sheet configurations, high fidelity computational fluid dynamics (CFD) that describes the separation process accurately are required. High fidelity simulations have been shown to be effective in studying the complex transport phenomena in membrane systems. In addition, obtaining high CO2 recovery percentages and product purity requires a multi-stage membrane process, where the optimal network configuration of the membrane modules must be studied in a systematic way. In order to address the design problem at process scale, we formulate a superstructure for the membrane-based carbon capture, including up to three separation stages. In the formulation of the optimization problem, we include reduced models, based on rigorous CFD simulations of the membrane modules. Numerical results indicate that the optimal design includes three membrane stages, and the capture cost is 45.4 $/t-CO2.

Pedrozo, Hector A.↗
DOE OSTI2024

Membrane-based Carbon Capture Process Optimization using CFD Modeling

Carbon capture is a promising option to mitigate CO2 emissions from existing coal-fired power plants, cement and steel industries, and petrochemical complexes. Among the available technologies, membrane-based carbon capture presents the lowest energy consumption, operating costs, and carbon footprint. In addition, membrane processes have important operational flexibility and response times. On the other hand, the major challenges to widespread application of this technology are related to reducing capital costs and improving membrane stability and durability. To upscale the technology into stacked flat sheet configurations, high fidelity computational fluid dynamics (CFD) that describes the separation process accurately are required. High fidelity simulations have been shown to be effective in studying the complex transport phenomena in membrane systems. In addition, obtaining high CO2 recovery percentages and product purity re-quires a multi-stage membrane process, where the optimal network configuration of the mem-brane modules must be studied in a systematic way. In order to address the design problem at process scale, we formulate a superstructure for the membrane-based carbon capture, including up to three separation stages. In the formulation of the optimization problem, we include reduced models, based on rigorous CFD simulations of the membrane modules.

Pedrozo, Hector A.↗
DOE OSTI2023

Symmetric Kondo Lattice States in Doped Strained Twisted Bilayer Graphene

We use the topological heavy fermion (THF) model and its Kondo lattice (KL) formulation to study the possibility of a symmetric Kondo (SK) state in twisted bilayer graphene. Via a large-N approximation, we find a SK state in the KL model at fillings v = 0,±1,±2 where a KL model can be constructed. In the SK state, all symmetries are preserved and the local moments are Kondo screened by the conduction electrons. At the mean-field level of the THF model at v = 0,±1,±2,±3 we also find a similar symmetric state that is adiabatically connected to the symmetric Kondo state. We study the stability of the symmetric state by comparing its energy with the ordered (symmetry-breaking) states found in [H. Hu et al., Phys. Rev. Lett. 131, 026502 (2023)., Z.-D. Song and B. A. Bernevig, Phys. Rev. Lett. 129, 047601 (2022).] and find the ordered states to have lower energy at v = 0,±1,±2. However, moving away from integer fillings by doping the light bands, our mean-field calculations find the energy difference between the ordered state and the symmetric state to be reduced, which suggests the loss of ordering and a tendency toward Kondo screening. In order to include many-body effects beyond the mean-field approximation, we also performed dynamical mean-field theory calculations on the THF model in the nonordered phase. The spin susceptibility follows a Curie behavior at v = 0,±1,±2 down to ~2 K where the onset of screening of the local moment becomes visible. This hints to very low Kondo temperatures at these fillings, in agreement with the outcome of our mean-field calculations. At noninteger filling v = ±0.5,±0.8,±1.2 dynamical mean-field theory shows deviations from a 1/T susceptibility at much higher temperatures, suggesting a more effective screening of local moments with doping. Lastly, we study the effect of a C 3⁢z -rotational-symmetry-breaking strain via mean-field approaches and find that a symmetric phase (that only breaks C 3⁢z symmetry) can be stabilized at sufficiently large strain at v = 0,±1,±2. Our results suggest that a symmetric Kondo phase is strongly suppressed at integer fillings, but could be stabilized either at noninteger fillings or by applying strain.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2021

Methods for Estimating Dose from Galactic Cosmic Rays and Jovian Radiation Belts on Reduced-Order Monte Carlo Geometries

This document outlines the methods and tools developed to estimate dose rates from galactic cosmic rays and the trapped radiation belts surrounding Jupiter using Monte Carlo particle transport software. Furthermore, the techniques described here are specific for translating high-resolution MAVRIC models to analogous but lower resolution Monte Carlo N-Particle Transport (MCNP) code models to estimate charged particle effects on models originally developed to study neutral particles only. This capability was initially developed to predict doses to various components of dynamic radioisotope power systems for spaceflight applications, but the tools and methods can be easily applied to other problem geometries and expanded to other space radiation environments.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

An Efficient Reactive Force Field without Explicit Coordination Dependence for Studying Caustic Aluminum Chemistry

Reactive force fields (RFFs) are an expedient approach to sample chemical reaction paths in complex systems, relative to density functional theory. However, there is continued need to improve efficiencies, specifically in systems that have slow transverse degrees of freedom, as in highly viscous and superconcentrated solutions. Here, we present an RFF that is differentiated from current models (e.g., ReaxFF) by omitting explicit dependence on the atom coordination and employing a small parameter set based on Lennard-Jones, Gaussian, and Stillinger–Weber potentials. In this study, the model was parametrized from AIMD simulation data and is used to model aluminate reactivity in sodium hydroxide solutions with extensive validation against experimental radial distribution functions, computed free energy profiles for oligomerization, and formation energies. In conclusion, the model enables simulation of early stage Al(OH) 3 nucleation which has significant relevance to industrial processing of aluminum and has a computational cost that is reduced by 1 order of magnitude relative to ReaxFF.

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

Deciphering supramolecular and polymer-like behavior in metallogels: real-time insights into temperature-modulated gelation and rapid self-assembly dynamics

Bis(pyridyl) urea-based gelators, namely L2 and its isomeric mixture ( L1 + L2 ), are known to self-assemble into 1D architectures capable of inducing supramolecular gelation. Coordination with metal ions such as Ag( I ), Cu( II ), and Fe( III ) introduces structural reinforcement, enabling the formation of distinct 3D networks governed by metal-specific coordination geometries. Here, we present a comprehensive investigation into the temperature-responsive behavior (20–60 °C) of L2 and L1 + L2 , both in the absence and presence of Ag( I ), Dy( III ), Fe( III ), Cu( II ), and Ho( III ), using real-time small-angle neutron scattering (SANS). To probe long-term structural evolution/kinetics of self-assembly, real-time small-angle X-ray scattering (SAXS) was employed on L2 + Ag gels, complemented by differential scanning calorimetry (DSC) to evaluate thermal transitions. Our results reveal strikingly divergent gelation behaviors: L2 forms a highly rigid, covalent polymer-like network, while L1 + L2 exhibits remarkable thermal adaptability. Upon metal coordination, the assemblies exhibit pronounced crystallinity and exceptional thermal stability, as evidenced by persistent Bragg reflections and invariant d-spacings. Intriguingly, L2 : Fe (2 : 1) and L1 : L2 : Fe (0.5 : 0.5 : 1) in acetonitrile-d 3 (ACN-d 3 ) deviate from this trend, forming thermally labile amorphous gels. These systems show a complete loss of crystalline order, reduced Porod exponents—indicative of collapsed or branched fiber morphologies—and prominent melting and glass transition events in DSC. Fitting SANS and SAXS data to the correlation length model unveiled insightful nanostructural features. While most systems displayed minimal temperature-induced variation in mesh size or surface morphology, L2 : Ag in dimethyl sulfoxide-d 6 (DMSO-d 6 )/D 2 O and L2 : Fe (1 : 1) in ACN-d 3 exhibited a rare combination of thermally stable correlation lengths and increasing high- q exponents—strongly suggesting progressive fiber densification or surface smoothing within a robust gel framework. These findings highlight the tunability and structural resilience of supramolecular gels through precise control of ligand architecture, metal coordination, and temperature, offering valuable design principles for functional soft materials.

Pajoubpong, Jinnipha [Univ. of Cincinnati, OH (Uni↗
DOE OSTI2024

Estimates of global recycling coefficients for LTX- β discharges

We report the first observation of global recycling coefficient R near 0.5 in the Lithium Tokamak eXperiment-β (LTX-β), significantly below the minimum R previously reported in other devices. In a series of experiments with varied Li wall conditioning, estimates of the recycling coefficient have been made using a Lyman-α array and DEGAS2 modeling. A progressive reduction in Lyman-α emission with increased lithium and an increase in edge electron temperature are observed. It is also observed that with increasing Li coating thickness, the effective particle confinement time τp* is reduced and approaches TRANSP calculated energy confinement time (τE), with τp* near τE,TRANSP for the lowest recycling coefficients. Edge temperatures approaching core plasma temperatures, first reported in LTX, can now be directly connected to estimates of the recycling coefficient and qualitatively agree with previous UEDGE simulations. The particle flux to the limiting surfaces appears to be significantly reduced in comparison with fluid scrape-off layer (SOL) models, indicating that a large fraction of the SOL ions are mirror trapped. SOL collisionality drops more than an order of magnitude below the banana regime boundary, indicating the importance of kinetic effects. Full-f 1x2v gyrokinetic simulations of SOL field lines with the GKEYLL code indicate that the fraction of ions trapped along field lines increases as collisionality drops, as a result of increased lithium evaporation.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2023

Large inflationary logarithms in a nontrivial nonlinear sigma model

Loops of inflationary gravitons are known to induce large temporal and spatial logarithms that can cause perturbation theory to break down. Nonlinear sigma models possess the same kind of derivative interactions and induce the same sorts of large logarithms, without the complicated index structure and potential gauge problem. Previous studies have examined models with zero field space curvature that can be reduced to free field theories by local, invertible field redefinitions. Here we study a model that cannot be so reduced and still shows the same sorts of large logarithms. Additionally, we compute the evolution of the background at 1-loop and 2-loop orders, and we find the 1-loop β and γ functions.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2020

Early-Time Exponential Instabilities in Nonchaotic Quantum Systems

The majority of classical dynamical systems are chaotic and exhibit the butterfly effect: a minute change in initial conditions has exponentially large effects later on. But this phenomenon is difficult to reconcile with quantum mechanics. One of the main goals in the field of quantum chaos is to establish a correspondence between the dynamics of classical chaotic systems and their quantum counterparts. In isolated systems in the absence of decoherence, there is such a correspondence in dynamics, but it usually persists only over a short time window, after which quantum interference washes out classical chaos. In this work, we demonstrate that quantum mechanics can also play the opposite role and generate exponential instabilities in classically nonchaotic systems within this early-time window. Our calculations employ the out-of-time-ordered correlator (OTOC)—a diagnostic that reduces to the Lyapunov exponent in the classical limit but is well defined for general quantum systems. We show that certain classically nonchaotic models, such as polygonal billiards, demonstrate a Lyapunov-like exponential growth of the OTOC at early times with Planck’s-constant-dependent rates. This behavior is sharply contrasted with the slow early-time growth of the analog of the OTOC in the systems’ classical counterparts. These results suggest that classical-to-quantum correspondence in dynamics is violated in the OTOC even before quantum interference develops.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2022

srlife: A Fast Tool for High Temperature Receiver Design and Analysis

This report describes a tool for estimating the structural service life of tubular, panel solar receivers operating at high temperatures. A complete version of the tool is available as open source software at https://github.com/Argonne-National-Laboratory/srlife and can be installed through the PyPi (https://pypi.org) package manager. Given the basic receiver geometry and the thermal loads on the receiver, the tool provides 1D, 2D, or 3D thermal and structural (single tube and simplified system) analysis and creep-fatigue service life prediction for six metallic alloys – 316H, 800H, Alloy 617, Alloy 740H, and Alloy 230. With the exception of Alloy 230 where the material data is preliminary, the software included a detailed set of models for the materials, well-supported by high temperature experimental test data. The tool is designed for easy integration with a software stack, including solar field and thermohydraulic simulations, for optimizing receiver designs to meet service life and economic targets. The report describes several heuristics that can be applied in srlife to reduce the analysis time by several orders of magnitude but with fairly accurate life estimation when compared with full analysis. The report provides several examples demonstrating the utility of srlife in receiver design. Finally, the report discusses high temperature tests on Alloy 282, collected as part of this project, used to develop and support the material model for that alloy.

47 OTHER INSTRUMENTATION↗
DOE OSTI2020

On the character of turbulent-like flows in self-consistent models of core-collapse supernovae

Neutrino-driven convection plays a crucial role in the development of core-collapse supernova (CCSN) explosions. However, the complex mechanism that triggers the shock revival and the subsequent explosion has remained inscrutable for many decades. Multidimensional simulations suggest that the growth of fluid instabilities, especially development of convection, will determine the morphology of the explosion. We have performed 3D simulations using spherical-polar coordinates covering a reduced angular extent (90° computational domain), and with angular resolutions of 2°, 1°, 1/2°, and 1/4°, to study the development of turbulent-like flows in core-collapse supernova explosions on a time scale of order 100 ms. We have employed the multi-physics Chimera code that includes detailed nuclear physics and spectral neutrino transport. Coarse resolution models do not develop an inertial range, presumably due to the bottleneck effect, such that the energy is prevented from cascading down to small scales and tends to accumulate at large scales. High-resolution models instead, start to recover the k −5/3 scaling of Kolmogorov's theory. Stochasticity and few simulation samples limit our ability to predict the development of explosions. Over the simulated time period, our models show no clear trend in improving (or diminishing) conditions for explosion as the angular resolution is increased. However, we find that disordered flow provides an effective pressure, as characterized by the Reynolds stress, behind the shock. This contribution, ~40%–50% of the thermal pressure, aids shock revival and thus the development of the explosion. Finally, we show that the kinetic energy power spectrum of reduced angular extent and full 4π models are consistent, thus indicating that a 90° computational domain is an adequate configuration to study the character of turbulence in CCSNe.

Casanova Bustamante, Jordi↗
DOE OSTI2021

Predicted success of prophylactic antiviral therapy to block or delay SARS-CoV-2 infection depends on the drug's mechanism of action

Repurposed drugs that are safe and immediately available constitute a first line of defense against new viral infections. Despite limited antiviral activity against SARS-CoV-2, several drugs are being tested as medication or as prophylaxis to prevent infection. Using a stochastic model of early phase infection, we evaluate the success of prophylactic treatment with different drug types to prevent viral infection. We find that there exists a critical efficacy that a treatment must reach in order to block viral establishment. Treatment by a combination of drugs reduces the critical efficacy, most effectively by the combination of a drug blocking viral entry into cells and a drug increasing viral clearance. Below the critical efficacy, the risk of infection can nonetheless be reduced. Drugs blocking viral entry into cells or enhancing viral clearance reduce the risk of infection more than drugs that reduce viral production in infected cells. The larger the initial inoculum of infectious virus, the less likely is prevention of an infection. In our model, we find that as long as the viral inoculum is smaller than 10 infectious virus particles, viral infection can be prevented almost certainly with drugs of 90% efficacy (or more). Even when a viral infection cannot be prevented, antivirals delay the time to detectable viral loads. The largest delay of viral infection is achieved by drugs reducing viral production in infected cells. A delay of virus infection flattens the within-host viral dynamic curve, possibly reducing transmission and symptom severity. Thus, antiviral prophylaxis, even with reduced efficacy, could be efficiently used to prevent or alleviate infection in people at high risk.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2022

Integrating a ponderomotive guiding center algorithm into a quasi-static particle-in-cell code based on azimuthal mode decomposition

High fidelity modeling of plasma based acceleration (PBA) requires the use of three dimensional, fully nonlinear, and kinetic descriptions based on the particle-in-cell (PIC) method. In PBA an intense particle beam or laser (driver) propagates through a tenuous plasma whereby it excites a plasma wave wake. Three-dimensional PIC algorithms based on the quasi-static approximation (QSA) have been successfully applied to efficiently model the interaction between relativistic charged particle beams and plasma. In a QSA PIC algorithm, the plasma response to a charged particle beam or laser driver is calculated based on forces from the driver and self-consistent forces from the QSA form of Maxwell's equations. These fields are then used to advance the charged particle beam or laser forward by a large time step. Since the time step is not limited by the regular Courant-Friedrichs-Lewy (CFL) condition that constrains a standard 3D fully electromagnetic PIC code, a 3D QSA PIC code can achieve orders of magnitude speedup in performance. Recently, a new hybrid QSA PIC algorithm that combines another speedup technique known as an azimuthal Fourier decomposition has been proposed and implemented. This hybrid algorithm decomposes the electromagnetic fields, charge and current density into azimuthal harmonics and only the Fourier coefficients need to be updated, which can reduce the algorithmic complexity of a 3D code to that of a 2D code. Modeling the laser-plasma interaction in a full 3D electromagnetic PIC algorithm is very computationally expensive due the enormous disparity of physical scales to be resolved. In the QSA the laser is modeled using the ponderomotive guiding center (PGC) approach. We describe how to implement a PGC algorithm compatible for the QSA PIC algorithms based on the azimuthal mode expansion. Here this algorithm permits time steps orders of magnitude larger than the cell size and it can be asynchronously parallelized. Details on how this is implemented into the QSA PIC code that utilizes an azimuthal mode expansion, QPAD, are also described. Benchmarks and comparisons between a fully 3D explicit PIC code (OSIRIS), as well as a few examples related to laser wakefield acceleration, are presented.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Multilevel Monte Carlo methods for the Grad-Shafranov free boundary problem

The equilibrium configuration of a plasma in an axially symmetric reactor is described mathematically by a free boundary problem associated with the celebrated Grad-Shafranov equation. The presence of uncertainty in the model parameters introduces the need to quantify the variability in the predictions. This is often done by computing a large number of model solutions on a computational grid for an ensemble of parameter values and then obtaining estimates for the statistical properties of solutions. In this study, we explore the savings that can be obtained using multilevel Monte Carlo methods, which reduce costs by performing the bulk of the computations on a sequence of spatial grids that are coarser than the one that would typically be used for a simple Monte Carlo simulation. We examine this approach using both a set of uniformly refined grids and a set of adaptively refined grids guided by a discrete error estimator. Numerical experiments show that multilevel methods dramatically reduce the cost of simulation, with cost reductions typically on the order of 60 or more and possibly as large as 200. Furthermore, adaptive griding results in more accurate computation of geometric quantities such as x-points associated with the model.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2021

A multifidelity method for a nonlocal diffusion model

Nonlocal models feature a finite length scale, referred to as the horizon, such that points separated by a distance smaller than the horizon interact with each other. Such models have proven to be useful in a variety of settings. However, due to the reduced sparsity of discretizations, they are also generally computationally more expensive compared to their local differential equation counterparts. In this work, we introduce a multifidelity Monte Carlo method that combines the high-fidelity nonlocal model of interest with surrogate models that use coarser grids and/or smaller horizons and thus have lower fidelities and lower costs. Using the multifidelity method, the overall computational cost of uncertainty quantification is reduced without compromising accuracy. It is shown for a one-dimensional nonlocal diffusion example that speedups of up to two orders of magnitude can be achieved using the multifidelity method to estimate the expectation of an output of interest.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

On reduced modelling of the modulational dynamics in magnetohydrodynamics

This paper explores structure formation in two-dimensional magnetohydrodynamic (MHD) turbulence as a modulational instability (MI) of turbulent fluctuations. We focus on the early stages of structure formation and consider simple backgrounds that allow for a tractable model of the MI while retaining the full chain of modulational harmonics. This approach allows us to systematically examine the validity of popular closures such as the quasilinear approximation and other low-order truncations. We find that, although such simple closures can provide quantitatively accurate approximations of the MI growth rates in some regimes, they can fail to capture the modulational dynamics in adjacent regimes even qualitatively, falsely predicting MI when the system is actually stable. We find that this discrepancy is due to the excitation of propagating spectral waves (PSWs) which can ballistically transport energy along the modulational spectrum, unimpeded until dissipative scales, thereby breaking the feedback loops that would otherwise sustain MIs. The PSWs can be self-maintained as global modes with real frequencies and drain energy from the primary structure at a constant rate until the primary structure is depleted. To describe these waves within a reduced model, we propose an approximate spectral closure that captures them and MIs on the same footing. We also find that introducing corrections to ideal MHD, conservative or dissipative, can suppress PSWs and reinstate the accuracy of the quasilinear approximation. In this sense, ideal MHD is a ‘singular’ system that is particularly sensitive to the accuracy of the closure within mean-field models.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

Separable physics-informed DeepONet: Breaking the curse of dimensionality in physics-informed machine learning

The deep operator network (DeepONet) has shown remarkable potential in solving partial differential equations (PDEs) by mapping between infinite-dimensional function spaces using labeled datasets. However, in scenarios lacking labeled data, the physics-informed DeepONet (PI-DeepONet) approach, which utilizes the residual loss of the governing PDE to optimize the network parameters, faces significant computational challenges, particularly due to the curse of dimensionality. This limitation has hindered its application to high-dimensional problems, making even standard 3D spatial with 1D temporal problems computationally prohibitive. Additionally, the computational requirement increases exponentially with the discretization density of the domain. Here, to address these challenges and enhance scalability for high-dimensional PDEs, we introduce the Separable physics-informed DeepONet (Sep-PI-DeepONet). This framework employs a factorization technique, utilizing sub-networks for individual one-dimensional coordinates, thereby reducing the number of forward passes and the size of the Jacobian matrix required for gradient computations. By incorporating forward-mode automatic differentiation (AD), we further optimize computational efficiency, achieving linear scaling of computational cost with discretization density and dimensionality, making our approach highly suitable for high-dimensional PDEs. We demonstrate the effectiveness of Sep-PI-DeepONet through three benchmark PDE models: the viscous Burgers’ equation, Biot’s consolidation theory, and a parameterized heat equation. Our framework maintains accuracy comparable to the conventional PI-DeepONet while reducing training time by two orders of magnitude. Notably, for the heat equation solved as a 4D problem, the conventional PI-DeepONet was computationally infeasible (estimated 289.35 h), while the Sep-PI-DeepONet completed training in just 2.5 h. These results underscore the potential of Sep-PI-DeepONet in efficiently solving complex, high-dimensional PDEs, marking a significant advancement in physics-informed machine learning.

Neural operator↗