Search

We present a novel algorithmic approach and an error analysis leveraging Quasi-Monte Carlo (QMC) points for training deep neural network (DNN) surrogates of holomorphic Data-to-Observable (DtO) maps in engineering design. Our analysis reveals higher-order consistent, deterministic choices of training points in the input parameter space for both deep and shallow neural networks with holomorphic activation functions such as tanh. We prove that higher-order QMC training points facilitate higher-order decay (in terms of the number of training samples) of the underlying generalization error, with consistency error bounds that are free from the curse of dimensionality in terms of the number of input parameters, provided that DNN weights in hidden layers satisfy certain summability conditions. We present numerical experiments for DtO maps from elliptic and parabolic PDEs with uncertain inputs that confirm the theoretical analysis. [ABSTRACT FROM AUTHOR]

We propose a supervised deep learning algorithm based on low-discrepancy sequences as the training set. By a combination of theoretical arguments and extensive numerical experiments we demonstrate that the proposed algorithm significantly outperforms standard deep learning algorithms that are based on randomly chosen training data for problems in moderately high dimensions. The proposed algorithm provides an efficient method for building inexpensive surrogates for many underlying maps in the context of scientific computing. [ABSTRACT FROM AUTHOR]

We design schemes for the approximation of entropy solutions to nonlinear hyperbolic conservation laws in the regime where small-scale effects drive the dynamics of nonclassical shock waves in these solutions. Typically, the small-scale effects are modeled by adding higher-order dissipation terms taking into account the viscosity, capillarity, the Hall coefficient, etc., of the material under consideration. We analyze a strategy for the design of numerical methods adapted to these problems, referred to as schemes with well-controlled dissipation (WCD). Nonclassical entropy solutions are approximated by (small-scale) consistent and converging schemes with high accuracy--the main challenge being to capture physically relevant shocks. Following earlier works by LeFloch and collaborators, we rely on the equivalent equation, which provides a suitable tool in order to guarantee that small-scale dependent shocks are computed in a consistent and accurate way. WCD schemes are also intended to capture (nonclassical) shocks with arbitrary large strength. Our strategy is exemplified with examples and numerical experiments encompassing the cubic conservation law, the nonlinear elasticity system, and a reduced magnetohydrodynamics model. We compute the kinetic functions associated with the schemes and, as the order of the WCD schemes is increased, we observe convergence toward the exact kinetic function--even for strong shocks. [ABSTRACT FROM AUTHOR]

The vanishing viscosity limit of nonconservative hyperbolic systems depends heavily on the specific form of the viscosity. Numerical approximations, such as the path consistent schemes of [C. Parés, SIAM J. Numer. Anal., 41 (2007), pp. 169-185], may not converge to the physically relevant solutions of the system. We construct entropy stable path consistent (ESPC) schemes to approximate nonconservative hyperbolic systems by combining entropy conservative discretizations with numerical diffusion operators that are based on the underlying viscous operator. Numerical experiments for the coupled Burgers system and the two-layer shallow water equations demonstrating the robustness of ESPC schemes are presented. [ABSTRACT FROM AUTHOR]

We introduce a class of numerical schemes that preserve a discrete version of vorticity in conservation laws which involve grad advection. These schemes are based on reformulating finite volume schemes in terms of vertex centered numerical potentials. The resulting potential-based schemes have a genuinely multidimensional structure. A suitable choice of potentials leads to discrete vorticity preserving schemes that are simple to code, computationally inexpensive, and proven to be stable. We extend our discussion to other classes of genuinely multidimensional schemes. Numerical examples for linear grad advection equations, linear and nonlinear wave equation systems, and the Euler equations of gas dynamics are presented. [ABSTRACT FROM AUTHOR]

We propose a finite volume method for the shallow water equations that accurately approximates the transport of vorticity. The algorithm is based on a predictor-corrector-type projection method. Any consistent finite volume scheme may be used in the prediction step of the algorithm. An elliptic equation is solved and the momentum field is corrected to obtain the correct evolution of vorticity. We describe this projection algorithm for the wave equation and the shallow water equations. The crucial role played by the pseudovorticity transport is highlighted. Numerical experiments demonstrating a considerable gain in computational efficiency with the vorticity projection algorithm are presented. [ABSTRACT FROM AUTHOR]

We consider the scalar conservation law with flux function discontinuous in the space variable, i.e., \begin{eqnarray} \label{eq1} u_t+(H(x)f(u)+(1-H(x))g(u))_{x} &=& 0 \quad \mbox{in } \R \times \R_{+}, \nonumber \\ u(0, x) &=& u_{0}(x) \quad \mbox{in } \R, \label{0.1} \end{eqnarray} where $H$ is the Heaviside function and $f$ and $g$ are smooth with the assumptions that either $f$ is convex and $g$ is concave or $f$ is concave and $g$ is convex. The existence of a weak solution of (\ref{eq1}) is proved by showing that upwind finite difference schemes of Godunov and Enquist--Osher type converge to a weak solution. Uniqueness follows from a Kruzkhov-type entropy condition. We also provide explicit solutions to the Riemann problem for (\ref{eq1}). At the level of numerics, we give easy-to-implement numerical schemes of Godunov and Enquist--Osher type. The central feature of this paper is the modification of the singular mapping technique (the main analytical tool for these types of equations) which allows us to show that the numerical schemes converge. Equations of type (\ref{eq1}) with the above hypothesis on the flux may occur when considering the following scalar conservation law with discontinuous flux: \begin{equation} \label{eq2} \begin{array}{r@{\;}l} u_t + (k (x) f (u))_x &= 0, \\ u (0, x) &= u_0 (x), \end{array} \end{equation} with $f$ convex and $k$ of indefinite sign. [ABSTRACT FROM AUTHOR]