Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “approximate domain boundaries”

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 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
DOE OSTI2020

Enhanced thermoelectric efficiency in nanocrystalline bismuth telluride nanotubes

We report on the thermal and thermoelectric properties of individual nanocrystalline Bi 2 Te 3 nanotubes synthesized by the solution phase method using 3ω method and a microfabricated testbench. Measurements show that the nanotubes offer improved ZT compared to bulk Bi 2 Te 3 near room temperature due to an enhanced Seebeck coefficient and suppressed thermal conductivity. This improvement in ZT originates from the nanocrystalline nature and low dimensionality of the nanotubes. Domain boundary filtering of low-energy electrons provides an enhanced Seebeck coefficient. The scattering of phonons at the surface of the nanotube leads to suppressed thermal conductivity. These have been theoretically analyzed using the Boltzmann equation based on the relaxation time approximation and Landauer approach. This work clearly demonstrates the possibility of achieving enhancement in thermoelectric efficiency by combining nanocrystalline and low-dimensional systems.

77 NANOSCIENCE AND NANOTECHNOLOGY↗
DOE OSTI2020

Spectral Element Applications in Complex Nuclear Reactor Geometries: Tet-to-Hex Meshing

The spectral element code Nek5000 is an open-source, higher-order computational fluid dynamics code developed at Argonne National Laboratory. It is designed to solve incompressible Navier-Stokes equations, but it also has a low-Mach-number approximation feature available. Large eddy simulation is approached by explicit filtering of the velocity field (and other fields) to mimic the effect of dissipation due to the unresolved scale. The computational domain is decomposed into second-order hexahedral elements that conform to the boundaries. However, generating a high-quality pure-hexahedral mesh can be challenging for some problems. For simple geometries, traditional blocking methods can be used to decompose the domain into smaller blocks to generate a so-called structural mesh. A structural mesh can maintain good orthogonality but can have a highly skewed mesh to conform to the geometry, as well as unnecessary refinement in the far field. Moreover, for geometries with relative complexity, blocking the geometry becomes impossible. To address these issues, we adopted a let-to-hex strategy to generate a pure hexahedral mesh for Nek5000. First, we generate a pure tetrahedral mesh for the geometry; then we divide one tetrahedral element into four hexahedral elements. A pure tetrahedral mesh could be easily generated for complex geometries by using many current meshing codes. In this paper, we use the commercial codes ANSYS meshing and ANSYSICEM to generate the pure tetrahedral mesh and then convert it to a pure hexahedral mesh. Boundary layers are extruded in ANSYSICEM to maintain near-wall resolution.

ANSYS ICEM↗
DOE OSTI2025

Magnetic mesh generation and field line reconstruction for scrape-off layer and divertor modeling in stellarators

The design of divertor targets and baffles for optimal heat and particle exhaust from magnetically confined fusion plasmas requires a combination of fast, low-fidelity models (such as EMC3-Lite (Feng 2022 Plasma Phys. Control. Fusion 64 125012)) for scoping studies and high-fidelity ones (such as EMC3-EIRENE (Feng et al 2014 Contrib. Plasma Phys. 54 426–31)) for verification. Both of those approaches benefit from a magnetic flux tube mesh for fast interpolation and mapping of field line segments (Feng et al 2005 Phys. Plasmas 12 052505). A new automated mesh generator for unstructured quadrilateral flux tubes with adaptive refinement is presented and integrated into FLARE (Frerichs 2024 Nucl. Fusion 64 106034). For HSX with an extended first wall, it is found that several layers of flux tubes can span the entire half field period before splitting is required. This is an advantage over the traditional setup of the EMC3-EIRENE mesh where careful construction of several sub- domains is required already for the much tighter present first wall. In particular, there is no longer the need to manually construct a suitable outer boundary for the mesh. The divide and conquer paradigm with unstructured mesh layout offers a powerful alternative for fast head load approximation that is suitable for integration into optimization workflows. Further examples for W7-X and CTH demonstrate the versatile application range.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2023

Semiclassical theory and the Koopman-van Hove equation *

Abstract The phase space Koopman-van Hove (KvH) equation can be derived from the asymptotic semiclassical analysis of partial differential equations. Semiclassical theory yields the Hamilton–Jacobi equation for the complex phase factor and the transport equation for the amplitude. These two equations can be combined to form a nonlinear semiclassical version of the KvH equation in configuration space. There is a natural injection of configuration space solutions into phase space and a natural projection of phase space solutions onto configuration space. Hence, every solution of the configuration space KvH equation satisfies both the semiclassical phase space KvH equation and the Hamilton–Jacobi constraint. For configuration space solutions, this constraint resolves the paradox that there are two different conserved densities in phase space. For integrable systems, the KvH spectrum is the Cartesian product of a classical and a semiclassical spectrum. If the classical spectrum is eliminated, then, with the correct choice of Jeffreys–Wentzel–Kramers–Brillouin (JWKB) matching conditions, the semiclassical spectrum satisfies the Einstein–Brillouin–Keller quantization conditions which include the correction due to the Maslov index. However, semiclassical analysis uses different choices for boundary conditions, continuity requirements, and the domain of definition. For example, use of the complex JWKB method allows for the treatment of tunneling through the complexification of phase space. Finally, although KvH wavefunctions include the possibility of interference effects, interference is not observable when all observables are approximated as local operators on phase space. Observing interference effects requires consideration of nonlocal operations, e.g. through higher orders in the asymptotic theory.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

Transfer learning for metamaterial design and simulation

Abstract We demonstrate transfer learning as a tool to improve the efficacy of training deep learning models based on residual neural networks (ResNets). Specifically, we examine its use for study of multi-scale electrically large metasurface arrays under open boundary conditions in electromagnetic metamaterials. Our aim is to assess the efficiency of transfer learning across a range of problem domains that vary in their resemblance to the original base problem for which the ResNet model was initially trained. We use a quasi-analytical discrete dipole approximation (DDA) method to simulate electrically large metasurface arrays to obtain ground truth data for training and testing of our deep neural network. Our approach can save significant time for examining novel metasurface designs by harnessing the power of transfer learning, as it effectively mitigates the pervasive data bottleneck issue commonly encountered in deep learning. We demonstrate that for the best case when the transfer task is sufficiently similar to the target task, a new task can be effectively trained using only a few data points yet still achieve a test mean absolute relative error of 3 % with a pre-trained neural network, realizing data reduction by a factor of 1000.

Peng, Rixi↗
DOE OSTI2020

Complexion dictated thermal resistance with interface density in reactive metal multilayers

Multilayers composed of aluminum (A1) and platinum (Pt) exhibit a non-monotonic trend in thermal resistance with bilayer thickness as measured by time domain thermoreflectance. The thermal resistance initially increases with reduced bilayer thickness only to reach a maximum and then decrease with further shrinking of the multilayer period. These observations are attributed to the evolving impact of an intermixed amorphous complexion approximately 10 nm in thickness, which forms at each boundary between Al- and Pt-rich layers. Scanning transmission electron microscopy combined with energy dispersive X-ray spectroscopy find that the elemental composition of the complexion varies based on bilayer periodicity as does the fraction of the multilayer composed of this interlayer. These variations in complexion mitigate boundary scattering within the multilayers as shown by electronic transport calculations employing density functional theory and non-equilibrium Green's functions on amorphous structures obtained via finite temperature molecular dynamics. The lessening of boundary scattering reduces the total resistance to thermal transport leading to the observed non-monotonic trend thereby highlighting the central role of complexion on thermal transport within reactive metal multilayers.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Atomic Mixing of Initially Separated Mixtures

We consider the atomic mixing of two materials, initially separated in space, at fixed temperature– pressure conditions [T, p], and zero mass-mean velocity. Each material is composed of a specified set of atomic species whose density corresponds to the specieswise real Equation of State (EOS) $\tilde{&#961}$ i [T, p]. Because there are no spatial gradients in the mean [T, p] state, the only mechanism for mixing is a specieswise nonequilibrium thermal velocity distribution. The recipe for nonequilibrium species mass fluxes is taken directly from Hirschfelder, Curtiss, and Bird, who express the diffusion velocities in terms of gradients in the molar concentrations and pairwise coefficients of diffusion. EOS data from the SESAME library is used for all species. Data for the specieswise self diffusion coefficients is adopted from the Orbital Free Molecular Dynamics (OFMD) simulations of Ticknor et al.; the so-called Darken approximation is used to express the mutual diffusion coefficients of a mixture in the manner used by White et al. A one-dimensional spatial domain is used; with the left half occupied by one multispecies material and the right half occupied by the other, and with zero–gradient boundary conditions at the ends of the domain. Hence the total mass in the domain is a constant. At time t = 0 the materials are separated. Accordingly, spatial gradients at the separation point are infinite and the species fluxes are also infinite. As t → $\textbf{∞}$ the spatial gradients approach zero and a uniform set of species concentrations is obtained; at this point the system is said to be at “thermodynamic equilibrium.” Analytic solution of the problem is not possible in general due to the nonlinear concentration dependence of the species fluxes. Hence we utilize the numerical solution apparatus in Mathematica for obtaining accurate time-dependent solutions for the mixing of various initial species masses for conditions of interest in laser-driven Inertial Confinement Fusion. Concentration profiles and the speed of propagation of the mixing fronts are discussed in light of the nonlinearity of the problem.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

Accelerating Multivariate Functional Approximation Computation with Domain Decomposition Techniques⋆

Modeling large datasets through Multivariate Functional Approximations (MFA) provide an elegant way to handle many visualization and scientific analysis workflows. The process necessitates scalable data partitioning methods to compute MFA representations efficiently without compromising the accuracy or continuity of the reconstructed solution. We propose a domain -decomposed method for computing the MFA with B -spline bases, which reduces the total work per task and uses a restricted Additive Schwarz (RAS) method to converge the control point data degrees -of -freedom along subdomain boundaries. We provide an in-depth analysis of the parallel approach with domain decomposition solvers, aiming to minimize local subdomain error residuals and recover high -order continuity at subdomain interfaces with appropriate choices of knot overlaps. The communication cost, determined by the overlap regions in the RAS implementation, is optimized to recover the numerical error profile of the single subdomain case. Our proposed method stands in contrast to previous methods, which typically only recover either C 0 or at best C 1 continuity for arbitrary B -spline degree expansions, or those that require post -processing to blend discontinuities in the reconstructed data. We demonstrate the effectiveness of our approach using analytical and real -world datasets in 1D, 2D, and 3D through both strong and weak scaling studies. The performance results indicate that the overall cost of computing the approximation is directly proportional to the underlying nearest -neighbor communication implementation, and is only weakly dependent on the overlap region size that determines the size of the messages. This finding underscores the efficiency and scalability of our proposed method, making it a promising solution for handling large datasets in scientific workflows.

additive Schwarz solvers↗
DOE OSTI2025

Boundary Corrections for Kernel Approximation to Differential Operators

The kernel-based approach to operator approximation for partial differential equations has been shown to be unconditionally stable for linear PDEs and numerically exhibit unconditional stability for non-linear PDEs. These methods have the same computational cost as an explicit finite difference scheme but can exhibit order reduction at boundaries. In previous work on periodic domains, order reduction was addressed, yielding high-order accuracy. The issue addressed in this work is the elimination of order reduction of the kernel-based approach for a more general set of boundary conditions. Further, we consider the case of both first and second order operators. To demonstrate the theory, we provide not only the mathematical proofs but also experimental results by applying various boundary conditions to different types of equations. The results agree with the theory, demonstrating a systematic path to high order for kernel-based methods on bounded domains.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Physics‐Informed Machine Learning Method for Large‐Scale Data Assimilation Problems

Abstract We develop a physics‐informed machine learning approach for large‐scale data assimilation and parameter estimation and apply it for estimating transmissivity and hydraulic head in the two‐dimensional steady‐state subsurface flow model of the Hanford Site given synthetic measurements of said variables. In our approach, we extend the physics‐informed conditional Karhunen‐Loéve expansion (PICKLE) method to modeling subsurface flow with unknown flux (Neumann) and varying head (time‐dependent Dirichlet) boundary conditions. We demonstrate that the PICKLE method is comparable in accuracy with the standard maximum a posteriori (MAP) method, but is significantly faster than MAP for large‐scale problems. Both methods use a mesh to discretize the computational domain. In MAP, the parameters and states are discretized on the mesh; therefore, the size of the MAP parameter estimation problem directly depends on the mesh size. In PICKLE, the mesh is used to evaluate the residuals of the governing equation, while the parameters and states are approximated by the truncated conditional Karhunen‐Loéve expansions with the number of parameters controlled by the smoothness of the parameter and state fields, and not by the mesh size. For a considered example, we demonstrate that the computational cost of PICKLE increases near linearly (as N 1.15 ) with the number of grid nodes N , while that of MAP increases much faster (as N 3.28 ). We also show that once trained for one set of Dirichlet boundary conditions (i.e., one river stage), the PICKLE method provides accurate estimates of the hydraulic head for any value of the Dirichlet boundary conditions (i.e., for any river stage).

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Magnetotransport of Sm 2 Ir 2 O 7 across the pressure-induced quantum-critical phase boundary

Rare-earth pyrochlore iridates host two interlocking magnetic sublattices of corner-sharing tetrahedra and can harbour a unique combination of frustrated moments, exotic excitations and highly correlated electrons. They are also the first systems predicted to display both topological Weyl semimetal and axion insulator phases. We have measured the transport and magnetotransport properties of single-crystal Sm 2 Ir 2 O 7 up to and beyond the pressure-induced quantum critical point for all-in-all-out (AIAO) Ir order at p c = 63 kbar previously identified by resonant X-ray scattering and close to which Weyl semimetallic behavior has been previously predicted. Our findings overturn the accepted expectation that the suppression of AIAO order should lead to metallic conduction persisting down to zero temperature. Instead, the resistivity-minimum temperature, which tracks the decrease in the AIAO ordering temperature for pressures up to 30 kbar, begins to increase under further application of pressure, pointing to the presence of a second as-yet unidentified mechanism leading to non-metallic behavior. The magnetotransport does track the suppression of Ir magnetism, however, with a strong hysteresis observed only within the AIAO phase boundary, similar to that found for Ho 2 Ir 2 O 7 and attributed to plastic deformation of Ir domains. Around pc we find the emergence of a new type of electronic phase, characterized by a negative magnetoresistance with small hysteresis at the lowest temperatures, and hysteresis-free positive magnetoresistance above approximately 5K. The temperature dependence of our low-temperature transport data are found to be best described by a model consistent with a Weyl semimetal across the entire pressure range.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Magnetotransport of Sm 2 Ir 2 O 7 across the pressure-induced quantum-critical phase boundary

Rare-earth pyrochlore iridates host two interlocking magnetic sublattices of corner-sharing tetrahedra and can harbour a unique combination of frustrated moments, exotic excitations and highly correlated electrons. They are also the first systems predicted to display both topological Weyl semimetal and axion insulator phases. We have measured the transport and magnetotransport properties of single-crystal Sm 2 Ir 2 O 7 up to and beyond the pressure-induced quantum critical point for all-in-all-out (AIAO) Ir order at $p_c$ = 63 kbar previously identified by resonant X-ray scattering and close to which Weyl semimetallic behavior has been previously predicted. Our findings overturn the accepted expectation that the suppression of AIAO order should lead to metallic conduction persisting down to zero temperature. Instead, the resistivity-minimum temperature, which tracks the decrease in the AIAO ordering temperature for pressures up to 30 kbar, begins to increase under further application of pressure, pointing to the presence of a second as-yet unidentified mechanism leading to non-metallic behavior. The magnetotransport does track the suppression of Ir magnetism, however, with a strong hysteresis observed only within the AIAO phase boundary, similar to that found for Ho 2 Ir 2 O 7 and attributed to plastic deformation of Ir domains. Around $p_c$ we find the emergence of a new type of electronic phase, characterized by a negative magnetoresistance with small hysteresis at the lowest temperatures, and hysteresis-free positive magnetoresistance above approximately 5 K. The temperature dependence of our low-temperature transport data are found to be best described by a model consistent with a Weyl semimetal across the entire pressure range.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Phase-field simulations of fission gas bubble growth and interconnection in U-(Pu)-Zr nuclear fuel

Abstract The growth and interconnection of fission gas bubbles in the hotter central regions of U-(Pu)-Zr nuclear fuel has been simulated with a phase-field model. The Cahn-Hilliard equation was used to represent the two-phase microstructure, with a single defect species. The volume fraction of the bubble phase and surface area of the bubble-matrix interface were determined during growth and interconnection. Surface area increased rapidly during the initial stages of growth, then slowed and finally decreased as bubble interconnection began and coarsening acted to reduce surface area. The fraction of the bubbles vented to a simulation domain boundary, f V , was quantified as a measure of the microstructure’s interconnectivity and plotted as a function of porosity p . The defect species diffusivity was varied; although changes in diffusivity significantly affected the microstructure, the plots of f V vs. p did not change significantly. The percolation threshold p c was calculated to be approximately 0.26, depending on the assumed diffusivity and using an initial bubble number density based on experimental observations. This is slightly smaller than the percolation threshold for continuum percolation of overlapping 3D spheres. The simulation results were used to parameterize two different engineering-scale swelling models for U-(Pu)-Zr in the nuclear fuel performance code BISON.

Aagesen, Larry K. (ORCID:000000034936676X)↗
DOE OSTI2020

Anharmonicity and Phase Diagram of Magnesium Oxide in the Megabar Regime

With density functional molecular dynamics simulations, we computed the phase diagram of MgO from 50 to 2000 GPa up to 20000 K. Via thermodynamic integration (TDI), we derive the Gibbs free energies of the B1, B2, and liquid phases and determine their phase boundaries. With TDI and a pseudo-quasi-harmonic approach, we show that anharmonic effects are important and stabilize the B1 phase in particular. As a result, the B1-B2 transition boundary in the pressure-temperature plane exhibits a steep slope. We predict the B1-B2-liquid triple point to occur at approximately T= 10000 K and P = 370 GPa, which is higher in pressure than was inferred with quasi-harmonic methods alone. We predict the principal shock Hugoniot curve to enter the B2 phase stability domain but only over a very small range of parameters. Here, this may render it difficult to observe this phase with shock experiments because of kinetic effects.

3-dimensional systems↗
DOE OSTI2020

Efficient Generalized Boundary Detection Using a Sliding Information Distance

In this work, we present a general machine learning algorithm for boundary detection within general signals based on an efficient, accurate, and robust approximation of the universal normalized information distance. Our approach uses an adaptive sliding information distance (SLID) combined with a wavelet-based approach for peak identification to locate the boundaries. Special emphasis is placed on developing an adaptive formulation of SLID to handle general signals with multiple unknown and/or drifting section lengths. Although specialized algorithms may outperform SLID when domain knowledge is available, these algorithms are limited to specific applications and do not generalize. SLID excels in these cases. We demonstrate the versatility and efficacy of SLID on a variety of signal types, including synthetically generated sequences of tokens, binary executables for reverse engineering applications, and time series of seismic events.

42 ENGINEERING↗
DOE OSTI2020

Efficacy of the Cell Perturbation Method in Large-Eddy Simulations of Boundary Layer Flow over Complex Terrain

A challenge to simulating turbulent flow in multiscale atmospheric applications is the efficient generation of resolved turbulence motions over an area of interest. One approach is to apply small perturbations to flow variables near the inflow planes of turbulence-resolving simulation domains nested within larger mesoscale domains. While this approach has been examined in numerous idealized and simple terrain cases, its efficacy in complex terrain environments has not yet been fully explored. Here, we examine the benefits of the stochastic cell perturbation method (CPM) over real complex terrain using data from the 2017 Perdigão field campaign, conducted in an approximately 2-km wide valley situated between two nearly parallel ridges. Following a typical configuration for multiscale simulation using nested domains within the Weather Research and Forecasting (WRF) model to downscale from the mesoscale to a large-eddy simulation (LES), we apply the CPM on a domain with horizontal grid spacing of 150 m. At this resolution, spurious coherent structures are often observed under unstable atmospheric conditions with moderate mean wind speeds. Results from such an intermediate resolution grid are often nested down for finer, more detailed LES, where these spurious structures adversely affect the development of turbulence on the subsequent finer grid nest. We therefore examine the impacts of the CPM on the representation of turbulence within the nested LES domain under moderate mean flow conditions in three different stability regimes: weakly convective, strongly convective, and weakly stable. In addition, two different resolutions of the underlying terrain are used to explore the role of the complex topography itself in generating turbulent structures. We demonstrate that the CPM improves the representation of turbulence within the LES domain, relative to the use of high-resolution complex terrain alone. During the convective conditions, the CPM improves the rate at which smaller-scales of turbulence form, while also accelerating the attenuation of the spurious numerically generated roll structures near the inflow boundary. During stable conditions, the coarse mesh spacing of the intermediate LES domain used herein was insufficient to maintain resolved turbulence using CPM as the flow develops downstream, highlighting the need for yet higher resolution under even weakly stable conditions, and the importance of accurate representation of flow on intermediate LES grids.

17 WIND ENERGY↗
DOE OSTI2022

Riverbed Temperature and 4D ERT Monitoring Reveals Heterogenous Horizontal and Vertical Groundwater-Surface Water Exchange Flows Under Dynamic Stage Conditions

Groundwater surface water exchange plays a critical role in physical, biological, and geochemical function of coastal and riverine systems. Observing exchange flow behavior in heterogeneous systems is a primary challenge, particularly when flows are governed by dynamic river stage or tidal variations. In this paper we demonstrate a novel application of time-lapse 3D electrical resistivity tomography and temperature monitoring where an array of thermistors installed beneath a riverbed double as resistivity electrodes. We use the array to monitor stage driven exchange flows over a 6-day period in a dynamic, stage-driven high order stream. We present a method for addressing the otherwise confounding effects of the moving river-surface boundary on the raw resistivity data, thereby enabling successful tomographic imaging. Temperature time-series at each thermistor location and time-lapse 3D images of changes in bulk electrical conductivity together provide a detailed description of exchange dynamics over a 10-meter by 45-meter section of the riverbed, to a depth of approximately 5 m. Results reveal highly variable flux behavior throughout the monitoring domain including both horizontal and vertical exchange flows.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations

In the last 50 years there has been a tremendous progress in solving numerically the Navier-Stokes equations using finite differences, finite elements, spectral, and even meshless methods. Yet, in many real cases, we still cannot incorporate seamlessly (multi-fidelity) data into existing algorithms, and for industrial-complexity applications the mesh generation is time consuming and still an art. Moreover, solving ill-posed problems (e.g., lacking boundary conditions) or inverse problems is often prohibitively expensive and requires different formulations and new computer codes. Here, we employ physics-informed neural networks (PINNs), encoding the governing equations directly into the deep neural network via automatic differentiation, to overcome some of the aforementioned limitations for simulating incompressible laminar and turbulent flows. We develop the Navier-Stokes flow nets (NSFnets) by considering two different mathematical formulations of the Navier-Stokes equations: the velocity-pressure (VP) formulation and the vorticity-velocity (VV) formulation. Since this is a new approach, we first select some standard benchmark problems to assess the accuracy, convergence rate, computational cost and flexibility of NSFnets; analytical solutions and direct numerical simulation (DNS) databases provide proper initial and boundary conditions for the NSFnet simulations. The spatial and temporal coordinates are the inputs of the NSFnets, while the instantaneous velocity and pressure fields are the outputs for the VP-NSFnet, and the instantaneous velocity and vorticity fields are the outputs for the VV-NSFnet. This is unsupervised learning and, hence, no labeled data are required beyond boundary and initial conditions and the fluid properties. The residuals of the VP or VV governing equations, together with the initial and boundary conditions, are embedded into the loss function of the NSFnets. No data is provided for the pressure to the VP-NSFnet, which is a hidden state and is obtained via the incompressibility constraint without extra computational cost. Unlike the traditional numerical methods, NSFnets inherit the properties of neural networks (NNs), hence the total error is composed of the approximation, the optimization, and the generalization errors. Here, we empirically attempt to quantify these errors by varying the sampling (“residual”) points, the iterative solvers, and the size of the NN architecture. For the laminar flow solutions, we show that both the VP and the VV formulations are comparable in accuracy but their best performance corresponds to different NN architectures. The initial convergence rate is fast but the error eventually saturates to a plateau due to the dominance of the optimization error. For the turbulent channel flow, we show that NSFnets can sustain turbulence at , but due to expensive training we only consider part of the channel domain and enforce velocity boundary conditions on the subdomain boundaries provided by the DNS data base. We also perform a systematic study on the weights used in the loss function for balancing the data and physics components, and investigate a new way of computing the weights dynamically to accelerate training and enhance accuracy. In the last part, we demonstrate how NSFnets should be used in practice, namely for ill-posed problems with incomplete or noisy boundary conditions as well as for inverse problems. We obtain reasonably accurate solutions for such cases as well without the need to change the NSFnets and at the same computational cost as in the forward well-posed problems. As a result, we also present a simple example of transfer learning that will aid in accelerating the training of NSFnets for different parameter settings.

97 MATHEMATICS AND COMPUTING↗