Search NASA⌕ Search

SEARCH · Search NASA

Results for “explicit and implicit methods”

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 271 records · Page 15

Graphical Gaussian Process Regression Model for Aqueous Solvation Free Energy Prediction of Organic Molecules in Redox Flow Battery

The solvation free energy of organic molecules is a critical parameter in determining emergent properties such as solubility, liquid-phase equilibrium constants, and pKa and redox potentials in an organic redox flow battery. In this work, we present a machine learning (ML) model that can learn and predict the aqueous solvation free energy of an organic molecule using Gaussian process regression method based on a new molecular graph kernel. To investigate the performance of the ML model on electrostatic interaction, the nonpolar interaction contribution of solvent and the conformational entropy of solute in solvation free energy, three data sets with implicit or explicit water solvent models, and contribution of conformational entropy of solute are tested. We demonstrate that our ML model can predict the solvation free energy of molecules at chemical accuracy with a mean absolute error of less than 1 kcal/mol for subsets of the QM9 dataset and the Freesolv database. To solve the general data scarcity problem for a graph-based ML model, we propose a dimension reduction algorithm based on the distance between molecular graphs, which can be used to examine the diversity of the molecular data set. It provides a promising way to build a minimum training set to improve prediction for certain test sets where the space of molecular structures is predetermined.

25 ENERGY STORAGE↗

SO(3)-invariant PCA with application to molecular data

Principal component analysis (PCA) is a fundamental technique for dimensionality reduction and denoising; however, its application to three-dimensional data with arbitrary orientations -- common in structural biology -- presents significant challenges. A naive approach requires augmenting the dataset with many rotated copies of each sample, incurring prohibitive computational costs. In this paper, we extend PCA to 3D volumetric datasets with unknown orientations by developing an efficient and principled framework for SO(3)-invariant PCA that implicitly accounts for all rotations without explicit data augmentation. By exploiting underlying algebraic structure, we demonstrate that the computation involves only the square root of the total number of covariance entries, resulting in a substantial reduction in complexity. We validate the method on real-world molecular datasets, demonstrating its effectiveness and opening up new possibilities for large-scale, high-dimensional reconstruction problems.

Fraiman, Michael [Tel Aviv Univ., Tel Aviv (Israel↗

Recovery Discontinuous Galerkin Jacobian-Free Newton-Krylov Method for All-Speed Flows

A novel numerical algorithm (rDG-JFNK) for all-speed fluid flows with heat conduction and viscosity is introduced. The rDG-JFNK combines the Discontinuous Galerkin spatial discretization with the implicit Runge-Kutta time integration under the Jacobian-free Newton-Krylov framework. We solve fully-compressible Navier-Stokes equations without operator-splitting of hyperbolic, diffusion and reaction terms, which enables fully-coupled high-order temporal discretization. The stability constraint is removed due to the L-stable Explicit, Singly Diagonal Implicit Runge-Kutta (ESDIRK) scheme. The governing equations are solved in the conservative form, which allows one to accurately compute shock dynamics, as well as low-speed flows. For spatial discretization, we develop a “recovery” family of DG, exhibiting nearly-spectral accuracy. To precondition the Krylov-based linear solver (GMRES), we developed an “Operator-Split”-(OS) Physics Based Preconditioner (PBP), in which we transform/simplify the fully-coupled system to a sequence of segregated scalar problems, each can be solved efficiently with Multigrid method. Each scalar problem is designed to target/cluster eigenvalues of the Jacobian matrix associated with a specific physics.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Explicit Monotone Stable Super-Time-stepping Methods for Finite Time Singularities

We explore a novel way to numerically resolve the scaling behavior of finite-time singularities in solutions of nonlinear parabolic PDEs. The Runge–Kutta–Legendre (RKL) and Runge–Kutta–Gegenbauer (RKG) super-time-stepping methods were originally developed for nonlinear complex physics problems with diffusion. These are multistage single step second-order, forward-in-time methods with no implicit solves. The advantage is that the time-step size for stability scales with stage number 𝑠 as $\mathcal{O}$⁡(𝑠 2 ). Many interesting nonlinear PDEs have finite-time singularities, and the presence of diffusion often limits one to using implicit or semi-implicit time-step methods for stability constraints. Finite-time singularities are particularly challenging due to the large range of scales that one desires to resolve, often with adaptive spatial grids and adaptive time steps. Here, in this study, we show two examples of nonlinear PDEs for which the self-similar singularity structure has time and space scales that are resolvable using the RKL and RKG methods, without forcing even smaller time steps. Compared to commonly used implicit numerical methods, we achieve a significantly smaller run time while maintaining comparable accuracy. We also prove numerical monotonicity for both the RKL and RKG methods under their linear stability conditions for the constant coefficient heat equation, in the case of infinite domain and periodic boundary condition, leading to a theoretical guarantee of the superiority of the RKL and RKG methods over traditional super-time-stepping methods, such as the Runge-Kutta-Chebyshev and the orthogonal Runge-Kutta-Chebyshev methods. Code can be found at https://github.com/ZT220501/SRK-Singularity.

97 MATHEMATICS AND COMPUTING↗

Expedited Real Time Processing for the NETL HYPER Cyber-Physical System (Final Report)

The primary objective of the proposed project was to provide the National Energy Technology Laboratory’s Hybrid Performance (Hyper) Facility the needed numerical methods algorithm(s), software development and implementation support to enact real time cyber-physical systems (CPSs) that simulate process dynamics on the order of five milliseconds or smaller. The proposed paths forward comprised three distinct approaches to hasten transient simulations. They fell under the numerical methods categorical options of: 1) optimizing key parameters within the facility’s pre-existing real time processing scheme; 2) introducing an “informed” processing approach wherein a priori computations expedite real time attempts; and 3) implementing alternatives to the employed explicit-implicit blended finite difference (spatio-temporal) approach. Each of these three classes were proposed independently as options for improvement, yet in some cases one class of approach could have complemented another.

20 FOSSIL-FUELED POWER PLANTS↗

A Concise Method for Storing and Communicating the Data Covariance Matrix

The covariance matrix associated with experimental cross section or transmission data consists of several components. Statistical uncertainties on the measured quantity (counts) provide a diagonal contribution. Off-diagonal components arise from uncertainties on the parameters (such as normalization or background) that figure into the data reduction process; these are denoted systematic or common uncertainties, since they affect all data points. The full off-diagonal data covariance matrix (DCM) can be extremely large, since the size is the square of the number of data points. Fortunately, it is not necessary to explicitly calculate, store, or invert the DCM. Likewise, it is not necessary to explicitly calculate, store, or use the inverse of the DCM. Instead, it is more efficient to accomplish the same results using only the various component matrices that appear in the definition of the DCM. Those component matrices are either diagonal or small (the number of data points times the number of data-reduction parameters); hence, this implicit data covariance method requires far less array storage and far fewer computations while producing more accurate results.

97 MATHEMATICS AND COMPUTING↗

Linear and Nonlinear Solvers for Simulating Multiphase Flow within Large-Scale Engineered Subsurface Systems

Simulation of multiphase flow in the subsurface is well-known to be computationally challenging. While there have been many studies that have explored approaches to overcoming these challenges, they often utilize relatively simple case studies. In this paper, we focus on the unique numerical challenges posed by modeling large-scale engineered subsurface systems, characterized by discrete features embedded in a heterogeneous natural subsurface setting. The man-made features such as shafts, tunnels, and barriers often cause multiple challenges in modeling the domain for multiphase porous media flow. This flow scenario can have a wide range of applications such as nuclear waste repositories, enhanced recovery of a petroleum reservoir, geothermal engineering, and carbon sequestration. An example of these severe numerical challenges is the case of performance assessment (PA) for Waste Isolation Pilot Plant (WIPP), the only operating deep geological repository in the US, which simulates extreme material properties of bedded salt rock formation and extreme contrast due to open excavation next to the formation. The models have extremes not only of permeability and porosity but also of the constitutive models needed for multiphase flow; additionally, they have process models like salt creep closure reducing porosity over time, fracturing in clay and anhydrite interbeds of the bedded salt, gas generation from the waste materials, and unintentional human borehole intrusions in some scenarios. Numerical simulations require the solution of coupled systems of nonlinear PDEs; in our work, we use the open-source simulator PFLOTRAN which is based on Finite Volume discretization. The solution of the nonlinear equations requires use of the Newton-Raphson iteration at each time step, which entails the solution of the linearized Jacobian system at each iteration. The effects of all the processes (i.e., large number of unknowns, highly nonlinear constitutive relations, large contrasts in material properties in short distances) lead to an ill-conditioned Jacobian matrix that severely challenges traditional linear solver, i.e., stabilized biconjugate gradient with block Jacobi incomplete LU preconditioner (BCGS-ILU) leading to non-convergence for traditional Newton-Raphson nonlinear solver causing unacceptably long computation time for each model. This paper presents linear solvers such as constrained pressure residual (CPR) two-stage preconditioner with alternate-block-factorization (ABF) and quasi- implicit pressure and explicit saturation (QIMPES) decouplers and flexible generalized residual solver (FGMRES). The new general-purpose nonlinear solver, Newton trust-region dogleg Cauchy (NTRDC), is also introduced to resolve extreme nonlinearities in the models. We demonstrate the effectiveness of each method relative to the default BCGS-Newton solver. The two best cases had nearly 50 times speed-up and achieved completion of a simulation in 14 hours that never completed due to non-convergence with the default solver. We also investigate the strong scalability of each method and discuss some of the deficiencies found for Block Jacobi preconditioner using parallel domain decomposition, and node packing effects of modern processor architecture.

Preconditioner, Nonlinear, Porous media, Multiphas↗

Entropy–Preserving and Entropy–Stable Relaxation IMEX and Multirate Time–Stepping Methods

In this work, we propose entropy-preserving and entropy-stable partitioned Runge–Kutta (RK) methods. In particular, we extend the explicit relaxation Runge–Kutta methods to IMEX–RK methods and a class of explicit second-order multirate methods for stiff problems arising from scale-separable or grid-induced stiffness in a system. The proposed approaches not only mitigate system stiffness but also fully support entropy-preserving and entropy-stability properties at a discrete level. The key idea of the relaxation approach is to adjust the step completion with a relaxation parameter so that the time-adjusted solution satisfies the entropy condition at a discrete level. The relaxation parameter is computed by solving a scalar nonlinear equation at each timestep in general; however, as for a quadratic entropy function, we theoretically derive the explicit form of the relaxation parameter and numerically confirm that the relaxation parameter works the Burgers equation. Several numerical results for ordinary differential equations and the Burgers equation are presented to demonstrate the entropy-conserving/stable behavior of these methods. We also compare the relaxation approach and the incremental direction technique for the Burgers equation with and without a limiter in the presence of shocks.

97 MATHEMATICS AND COMPUTING↗

ORMATEX

The Oak Ridge Matrix Exponential (ORMATEX) software library contains methods to compute the matrix exponential and the action of the matrix exponential on a vector. Additionally, this package contains the related methods for the phi-functions which commonly appear in a wide class of exponential time integration methods. Krylov methods are provided to evaluate the matrix exponential-vector and phi-vector products for cases where the matrix is large and sparse. Utilizing these methods, ORMATEX implements performant exponential integrators for large systems of coupled ordinary differential equations (ODEs). The exponential time integration routines in ORMATEX are particularly suitable to large, stiff systems of equations. These routines may be utilized as a competitive alternative to classical implicit and explicit time integration schemes for certain classes of differential equations where the problem stiffness can be predominately explained by the linear terms.

Gurecky, William [Oak Ridge National Laboratory (O↗

Impact of Integration Scheme on Performance of Anisotropic Plasticity Models

Given the prevalent role of metals in a variety of industries, schemes to integrate corresponding constitutive models in finite element applications have long been studied. A number of formulations have been developed to accomplish this task; each with their own advantages and costs. Often the focus has been on ensuring the accuracy and numerical stability of these algorithms to enable robust integration. While important, emphasis on these performance metrics may often come at the cost of computational expense potentially neglecting the needs of individual problems. In the current work, the performance of two of the most common integration methods for anisotropic plasticity -- the convex cutting plane (CCP) and closest point projection (CPP) -- across a variety of metrics is assessed; including accuracy and cost. A variety of problems are considered ranging from single elements to large representative simulations including both implicit quasistatic and explicit transient dynamic type responses. The relative performance of each scheme in the different instances is presented with an eye towards guidance on when the different algorithms may be beneficial.

36 MATERIALS SCIENCE↗

A tightly-coupled domain-decomposition approach for highly nonlinear stochastic multiphysics systems

Multiphysics simulations often involve nonlinear components that are driven by internally generated or externally imposed random fluctuations. When used with a domain-decomposition (DD) algorithm, such components have to be coupled in a way that both accurately propagates the noise between the subdomains and lends itself to a stable and cost-effective temporal integration. We develop a conservative DD approach in which tight coupling is obtained by using a Jacobian-free Newton–Krylov (JfNK) method with a generalized minimum residual iterative linear solver. This strategy is tested on a coupled nonlinear diffusion system forced by a truncated Gaussian noise at the boundary. Enforcement of path-wise continuity of the state variable and its flux, as opposed to continuity in the mean, at interfaces between subdomains enables the DD algorithm to correctly propagate boundary fluctuations throughout the computational domain. Reliance on a single Newton iteration (explicit coupling), rather than on the fully converged JfNK (implicit) coupling, may increase the solution error by an order of magnitude. Increase in communication frequency between the DD components reduces the explicit coupling's error, but makes it less efficient than the implicit coupling at comparable error levels for all noise strengths considered. Finally, the DD algorithm with the implicit JfNK coupling resolves temporally-correlated fluctuations of the boundary noise when the correlation time of the latter exceeds some multiple of an appropriately defined characteristic diffusion time.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Coarse-grained simulations of vortex dynamics and transition in complex high-Re flows

Turbulent flow complexity in applications in engineering, geophysics and astrophysics typically requires achieving accurate and dependable large scale predictions of highly nonlinear processes with under-resolved computer simulation models. Laboratory observations typically demonstrate the end outcome of complex non-linear three-dimensional physical processes with many unexplained details and mechanisms. Carefully controlled computational experiments based on the numerical solution of the conservation equations for mass, momentum, and energy, provide insights into the underlying flow dynamics. Relevant computational fluid dynamics issues to be addressed relate to the modeling of the unresolved tlow conditions at the subgrid scale (SGS) level - within a computational cell, and at the supergrid (SPG) scale - at initialization and beyond computational boundaries. SGS and SPG information must be prescribed for closure of the equations solved numerically. SGS models appear explicitly or implicitly as additional source tenns in the modified flow equations solved by the numerical solutions being calculated, while SPG models provide the necessary set of initial and boundary conditions that must be prescribed to ensure unique well-posed solutions. From this perspective, it is clear that the simulation process is inherently determined by the SGS and SPG information prescription process. On the other hand, observables in laboratory experiments are always characterized by the finite scales of the instrumental resolution of measuring/visualizing devices, and subject as well to SPG issues. It is thus important to recognize the inherently intrusive nature of observations based on numerical or laboratory experiments. Ultimately, verification and validation (V & V) frameworks and appropriate metrics for the specific problems at hand are needed to establish predictability of the simulation model. Direct numerical simulation (DNS) - resolving all relevant space/time scales, is prohibitively expensive in the foreseeable future for most practical flows of interest at moderate-to-high Reynolds number (Re). On the other end of the simulation spectrum are the Reynolds-Averaged Navier-Stokes (RANS) approaches - which model the turbulent effects. In the coarsegrained large eddy simulation (LES) strategies, the large energy containing structures are resolved, the smaller structures are filtered out, and unresolved SGS effects are modeled. By necessity - rather than choice, LES effectively becomes the intermediate approach between DNS and RANS. Extensive work has demonstrated that predictive simulations of turbulent velocity fields are possible using a particular LES denoted implicit LES (ILES), using the class of nonoscillatory finite-volume (NFV) numerical algorithms. Use of the modified equation as framework for theoretical analysis, demonstrates that leading truncation tenns associated with NFV methods provide implicit SGS models of mixed anisotropic type and regularized motion of discrete observables. Tests in fundamental applications ranging from canonical to very complex flows indicate that ILES is competitive with conventional LES in the LES realm proper - flows driven by large scale features. High-Re flows are vortex dominated and governed by short convective timescales compared to those of diffusion, and kinematically characterized at the smallest scales by slender worm vortices with insignificant internal structure. This motivates nominally inviscid ILES methods capable of capturing the high-Re dissipation dynamics and of handling vortices as shocks in shock capturing schemes. Depending on flow regimes, initial conditions, and resolution, additional modeling may be needed to emulate SGS driven physics, such as backscatter, chemical reaction, material mixing, and near-wall flow-dynamics - where typically-intertwined SGS/SPG issues need to be addressed. A major research focus is recognizing when additional explicit models and/or numerical treatments are needed and ensuring that mixed explicit and implicit SGS models can effectively act in collaborative rather than interfering fashion. We survey our present understanding of the theoretical basis of lLES, including connections with the classical LES and finite-scale dynamics perspectives. Examples from recent lLES studies are presented, including canonical turbulence test cases and shock driven turbulence; relevant V & V issues are demonstrated in this context.

58 GEOSCIENCES↗

CTF-R: A novel residual-based thermal hydraulic solver

The traditional scientific process has been revolutionized by the advent of computational modeling, but the nuclear industry generally uses "legacy codes," which were developed early in the evolution of computers. One example of a legacy code, the thermal hydraulic subchannel code CTF, is modernized in this work through the development of a novel residual-based version, CTF-R. Unlike its predecessor, CTF-R is not limited by the strict computational limitations of the early 1980's, and can therefore be designed such that it is inherently flexible and easy to use. A case study is examined to demonstrate how the flexibility of the code can be used to improve simulation results. In this example, the coupling between the solid and liquid fields is examined. Traditionally, this coupling is modeled explicitly, which imposes numerical stability limits on the time step size. These limits are derived and it is shown that they are removed when the coupling is made implicit. Further, the development of CTF-R will enable future improvements in next generation reactor modeling, numerical methods, and coupling to other codes. Through the further development of CTF-R and other residual-based codes, state-of-the-art simulation is possible.

42 ENGINEERING↗

Scalable smoothing strategies for a geometric multigrid method for the immersed boundary equations

The immersed boundary (IB) method is a widely used approach to simulating fluid-structure interaction (FSI). Although explicit versions of the IB method can suffer from severe time step size restrictions, these methods remain popular because of their simplicity and generality. In prior work (Guy et al., Adv Comput Math, 2015), some of us developed a geometric multigrid preconditioner for a stable semi-implicit IB method under Stokes flow conditions; however, this solver methodology used a Vanka-type smoother that presented limited opportunities for parallelization. This work extends this Stokes-IB solver methodology by developing smoothing techniques that are suitable for parallel implementation. Specifically, we demonstrate that an additive version of the Vanka smoother can yield an effective multigrid preconditioner for the Stokes-IB equations, and we introduce an efficient Schur complement-based smoother that is also shown to be effective for the Stokes-IB equations. We investigate the performance of these solvers for a broad range of material stiffnesses, both for Stokes flows and flows at nonzero Reynolds numbers, and for thick and thin structural models. We show here that linear solver performance degrades with increasing Reynolds number and material stiffness, especially for thin interface cases. Nonetheless, the proposed approaches promise to yield effective solution algorithms, especially at lower Reynolds numbers and at modest-to-high elastic stiffnesses.

97 MATHEMATICS AND COMPUTING↗

The NonConforming Virtual Element Method for the Stokes Equations

In this paper, we present the nonconforming virtual element method (VEM) for the numerical approximation of velocity and pressure in the steady Stokes problem. The pressure is approximated using discontinuous piecewise polynomials, while each component of the velocity is approximated using the nonconforming virtual element space. On each mesh element the local virtual space contains the space of polynomials of up to a given degree, plus suitable nonpolynomial functions. The virtual element functions are implicitly defined as the solution of local Poisson problems with polynomial Neumann boundary conditions. As typical in VEM approaches, the explicit evaluation of the non-polynomial functions is not required. This approach makes it possible to construct nonconforming (virtual) spaces for any polynomial degree regardless of the parity, for two- and three-dimensional problems, and for meshes with very general polygonal and polyhedral elements. We show that the nonconforming VEM is inf-sup stable and establish optimal a priori error estimates for the velocity and pressure approximations. Finally, numerical examples confirm the convergence analysis and the effectiveness of the method in providing high-order accurate approximations.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Asymmetric kinetic equilibria: Generalization of the BAS model for rotating magnetic profile and non-zero electric field

Finding kinetic equilibria for non-collisional/collisionless tangential current layers is a key issue as well for their theoretical modeling as for our understanding of the processes that disturb them, such as tearing or Kelvin Helmholtz instabilities. The famous Harris equilibrium [E. Harris, Il Nuovo Cimento Ser. 10 23, 115–121 (1962)] assumes drifting Maxwellian distributions for ions and electrons, with constant temperatures and flow velocities; these assumptions lead to symmetric layers surrounded by vacuum. This strongly particular kind of layer is not suited for the general case: asymmetric boundaries between two media with different plasmas and different magnetic fields. The standard method for constructing more general kinetic equilibria consists in using Jeans theorem, which says that any function depending only on the Hamiltonian constants of motion is a solution to the steady Vlasov equation [P. J. Channell, Phys. Fluids (1958–1988) 19, 1541 (1976); M. Roth et al., Space Sci. Rev. 76, 251–317 (1996); and F. Mottez, Phys. Plasmas 10, 1541–1545 (2003)]. The inverse implication is however not true: when using the motion invariants as variables instead of the velocity components, the general stationary particle distributions keep on depending explicitly of the position, in addition to the implicit dependence introduced by these invariants. The standard approach therefore strongly restricts the class of solutions to the problem and probably does not select the most physically reasonable. The BAS (Belmont-Aunai-Smets) model [G. Belmont et al., Phys. Plasmas 19, 022108 (2012)] used for the first time the concept of particle accessibility to find new solutions: considering the case of a coplanar-antiparallel magnetic field configuration without electric field, asymmetric solutions could be found while the standard method can only lead to symmetric ones. These solutions were validated in a hybrid simulation [N. Aunai et al., Phys. Plasmas (1994-present) 20, 110702 (2013)], and more recently in a fully kinetic simulation as well [J. Dargent and N. Aunai, Phys. Plasmas (submitted)]. Nevertheless, in most asymmetric layers like the terrestrial magnetopause, one would indeed expect a magnetic field rotation from one direction to another without going through zero [J. Berchem and C. T. Russell, J. Geophys. Res. 87, 8139–8148 (1982)], and a non-zero normal electric field. In this paper, we propose the corresponding generalization: in the model presented, the profiles can be freely imposed for the magnetic field rotation (although restricted to a 180 rotation hitherto) and for the normal electric field. As it was done previously, the equilibrium is tested with a hybrid simulation.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Adjoint-Based Implicit Uncertainty Analysis for Figures of Merit in a Laser Inertial Fusion Engine

A primary purpose of computational models is to inform design decisions and, in order to make those decisions reliably, the confidence in the results of such models must be estimated. Monte Carlo neutron transport models are common tools for reactor designers. These types of models contain several sources of uncertainty that propagate onto the model predictions. Two uncertainties worthy of note are (1) experimental and evaluation uncertainties of nuclear data that inform all neutron transport models and (2) statistical counting precision, which all results of a Monte Carlo codes contain. Adjoint-based implicit uncertainty analyses allow for the consideration of any number of uncertain input quantities and their effects upon the confidence of figures of merit with only a handful of forward and adjoint transport calculations. When considering a rich set of uncertain inputs, adjoint-based methods remain hundreds of times more computationally efficient than Direct Monte-Carlo methods. The LIFE (Laser Inertial Fusion Energy) engine is a concept being developed at Lawrence Livermore National Laboratory. Various options exist for the LIFE blanket, depending on the mission of the design. The depleted uranium hybrid LIFE blanket design strives to close the fission fuel cycle without enrichment or reprocessing, while simultaneously achieving high discharge burnups with reduced proliferation concerns. Neutron transport results that are central to the operation of the design are tritium production for fusion fuel, fission of fissile isotopes for energy multiplication, and production of fissile isotopes for sustained power. In previous work, explicit cross-sectional uncertainty analyses were performed for reaction rates related to the figures of merit for the depleted uranium hybrid LIFE blanket. Counting precision was also quantified for both the figures of merit themselves and the cross-sectional uncertainty estimates to gauge the validity of the analysis. All cross-sectional uncertainties were small (0.1-0.8%), bounded counting uncertainties, and were precise with regard to counting precision. Adjoint/importance distributions were generated for the same reaction rates. The current work leverages those adjoint distributions to transition from explicit sensitivities, in which the neutron flux is constrained, to implicit sensitivities, in which the neutron flux responds to input perturbations. This treatment vastly expands the set of data that contribute to uncertainties to produce larger, more physically accurate uncertainty estimates.

42 ENGINEERING↗

Physics constrained learning for data-driven inverse modeling from sparse observations

Deep neural networks (DNN) have been used to model nonlinear relations between physical quantities. Those DNNs are embedded in physical systems described by partial differential equations (PDE) and trained by minimizing a loss function that measures the discrepancy between predictions and observations in some chosen norm. This loss function often includes the PDE constraints as a penalty term when only sparse observations are available. As a result, the PDE is only satisfied approximately by the solution. However, the penalty term typically slows down the convergence of the optimizer for stiff problems. We present a new approach that trains the embedded DNNs while numerically satisfying the PDE constraints. We develop an algorithm that enables differentiating both explicit and implicit numerical solvers in reverse-mode automatic differentiation. This allows the gradients of the DNNs and the PDE solvers to be computed in a unified framework. We demonstrate that our approach enjoys faster convergence and better stability in relatively stiff problems compared to the penalty method. Furthermore, our approach allows for the potential to solve and accelerate a wide range of data-driven inverse modeling, where the physical constraints are described by PDEs and need to be satisfied accurately.

97 MATHEMATICS AND COMPUTING↗