On GLM curl cleaning for a first order reduction of the CCZ4 formulation of the Einstein field equations
Authors:
Michael Dumbser,
Francesco Fambri,
Elena Gaburro,
Anne Reinarz
Abstract:
In this paper we propose an extension of the generalized Lagrangian multiplier method (GLM) of Munz et al. (JCP 2000, JCP 2002), which was originally conceived for the numerical solution of the Maxwell and MHD equations with divergence-type involutions, to the case of hyperbolic PDE systems with curl-type involutions. The key idea here is to solve an augmented PDE system, in which curl errors prop…
▽ More
In this paper we propose an extension of the generalized Lagrangian multiplier method (GLM) of Munz et al. (JCP 2000, JCP 2002), which was originally conceived for the numerical solution of the Maxwell and MHD equations with divergence-type involutions, to the case of hyperbolic PDE systems with curl-type involutions. The key idea here is to solve an augmented PDE system, in which curl errors propagate away via a Maxwell-type evolution system. The new approach is first presented on a simple model problem, in order to explain the basic ideas. Subsequently, we apply it to a strongly hyperbolic first order reduction of the CCZ4 formulation (FO-CCZ4) of the Einstein field equations of general relativity, which is endowed with 11 curl constraints. Several numerical examples, including the long-time evolution of a stable neutron star in anti-Cowling approximation, are presented in order to show the obtained improvements with respect to the standard formulation without special treatment of the curl involution constraints. The main advantages of the proposed GLM approach are its complete independence of the underlying numerical scheme and grid topology and its easy implementation into existing computer codes. However, this flexibility comes at the price of needing to add for each curl involution one additional 3 vector plus another scalar in the augmented system for homogeneous curl constraints, and even two additional scalars for non-homogeneous curl involutions. For the FO-CCZ4 system with 11 homogeneous curl involutions, this means that additional 44 evolution quantities need to be added.
△ Less
Submitted 20 November, 2019; v1 submitted 8 September, 2019;
originally announced September 2019.
Influence of A-Posteriori Subcell Limiting on Fault Frequency in Higher-Order DG Schemes
Authors:
Anne Reinarz,
Jean-Matthieu Gallard,
Michael Bader
Abstract:
Soft error rates are increasing as modern architectures require increasingly small features at low voltages. Due to the large number of components used in HPC architectures, these are particularly vulnerable to soft errors. Hence, when designing applications that run for long time periods on large machines, algorithmic resilience must be taken into account. In this paper we analyse the inherent re…
▽ More
Soft error rates are increasing as modern architectures require increasingly small features at low voltages. Due to the large number of components used in HPC architectures, these are particularly vulnerable to soft errors. Hence, when designing applications that run for long time periods on large machines, algorithmic resilience must be taken into account. In this paper we analyse the inherent resiliency of a-posteriori limiting procedures in the context of the explicit ADER DG hyperbolic PDE solver ExaHyPE. The a-posteriori limiter checks element-local high-order DG solutions for physical admissibility, and can thus be expected to also detect hardware-induced errors. Algorithmically, it can be interpreted as element-local checkpointing and restarting of the solver with a more robust finite volume scheme on a fine subgrid. We show that the limiter indeed increases the resilience of the DG algorithm, detecting and correcting particularly those faults which would otherwise lead to a fatal failure.
△ Less
Submitted 21 May, 2019; v1 submitted 16 October, 2018;
originally announced October 2018.
