Local time-stepping for the shallow water equations using CFL optimized forward-backward Runge-Kutta schemes
Not Available
Engineering topics
Publications and source records attributed to Capodaglio, Giacomo.
Not Available
Abstract We present the formulation and optimization of a Runge–Kutta-type time-stepping scheme for solving the shallow-water equations, aimed at substantially increasing the effective allowable time step over that of comparable methods. This scheme, called FB-RK(3,2), uses weighted forward–backward averaging of thickness data to advance the momentum equation. The weights for this averaging are chosen with an optimization process that employs a von Neumann–type analysis, ensuring that the weights maximize the admittable Courant number. Through a simplified local truncation error analysis and numerical experiments, we show that the method is at least second-order in time for any choice of weights and exhibits low dispersion and dissipation errors for well-resolved waves. Further, we show that an optimized FB-RK(3,2) can take time steps up to 2.8 times as large as a popular three-stage, third-order strong stability-preserving Runge–Kutta method in a quasi-linear test case. In fully nonlinear shallow-water test cases relevant to oceanic and atmospheric flows, FB-RK(3,2) outperforms SSPRK3 in admittable time step by factors roughly between 1.6 and 2.2, making the scheme approximately twice as computationally efficient with little to no effect on solution quality. Significance Statement The purpose of this work is to develop and optimize time-stepping schemes for models relevant to oceanic and atmospheric flows. Specifically, for the shallow-water equations we optimize for schemes that can take time steps as large as possible while retaining solution quality. We find that our optimized schemes can take time steps between 1.6 and 2.2 times larger than schemes that cost the same number of floating point operations, translating directly to a corresponding speedup. Our ultimate goal is to use these schemes in climate-scale simulations.
Nonlocal models allow for the description of phenomena which cannot be captured by classical partial differential equations. The availability of efficient solvers is one of the main concerns for the use of nonlocal models in real world engineering applications. Here, we present a domain decomposition solver that is inspired by substructuring methods for classical local equations. In numerical experiments involving finite element discretizations of scalar and vectorial nonlocal equations of integrable and fractional type, we observe improvements in solution time of up to 14.6x compared to commonly used solver strategies.
Linear kinematic features (LKFs) are found everywhere in the Arctic sea-ice cover. They are strongly localized deformations often associated with the formation of leads and pressure ridges. In viscous-plastic (VP) sea-ice models, the simulation of LKFs depends on several factors such as the grid resolution, the numerical solver convergence, and the placement of the variables on the mesh. In this study, we compare two recently proposed discretization with a CD-grid placement with respect to their ability to reproduce LKFs. The first (CD1) is based on a nonconforming finite element discretization, whereas the second (CD2) uses a conforming subgrid discretization. To analyze their resolution properties, we evaluate runs from different models (e.g., FESOM, MPAS) on a benchmark problem using quadrilateral, hexagonal and triangular meshes. Our findings show that the CD1 setup simulates more deformation structure than the CD2 setup. This highlights the importance of the type of spatial discretization for the simulation of LKFs. Due to the higher number of degrees of freedom, both CD-grids resolve more LKFs than traditional A, B, and C-grids at fixed mesh level. This is an advantage of the CD-grid approach, as high spatial mesh resolution is needed in VP sea-ice models to simulate LKFs.
The objective is to evaluate the efficiency and efficacy of local time-stepping (LTS) methods in the Model for Prediction Across Scales-Ocean (MPAS-Ocean).
This year, my team made a very productive use of LANL IC time. Four papers were published that used IC resources, and two more are under review. These publications fall into three categories: 1.) Improving tide modeling in the global ocean by adding self attraction and loading (Barton et. al. 2022), ice shelf cavities (Pal et al. 2023), and local time stepping (Lilly et al. 2023). 2.) A new sea ice numerical formulation (Capodaglio 2023). 3.) Performance comparisons, methods, and test cases for ocean model development, This includes a verification suite for ocean models (Bishnu, submitted) and a comparison between Julia and Fortran (Strauss, submitted). A new time-stepping method was introduced in Calandrini et al. (2022). These publications, and the use of LANL IC resources, go hand-in-hand with our mentoring efforts to train young scientists. Two of the lead authors are graduate students who are conducting their PhD research: Kristin Barton at the University of Michigan and Jeremy Lilly at Oregon State. These are both their first publications, and they both have DOE funding and DOE mentors. In addition, Bishnu is a post-doctoral researcher at LANL; Strauss conducted his research as a senior in high school; and Capodaglio, Calandrini and Pal are all early-career scientists who were converted to staff in 2021 or 2022. Here we highlight two publications on improvements in tidal modeling: Barton et al. 2022 and Lilly et al. 2023.
This paper presents the first practical application of local time-stepping (LTS) schemes in the Model for Prediction Across Scales-Ocean (MPAS-O). We use LTS schemes in a single-layer, global ocean model that predicts the storm surge around the eastern coast of the United States during Hurricane Sandy. The variable-resolution meshes used are of unprecedentedly high resolution in MPAS-O, containing cells as small as 125 m wide in Delaware Bay. It is shown that a particular, third-order LTS scheme (LTS3) produces sea-surface height solutions that are of comparable quality to solutions produced by the classical four-stage, fourth-order Runge-Kutta method (RK4) with a uniform time step on the same meshes. Furthermore, LTS3 is up to 35% faster in the best cases considered, where the number of cells using the coarse time-step relative to those using the fine time-step is as low as 1:1. This shows that LTS schemes are viable for use in MPAS-O with the added benefit of substantially less computational cost. The results of these performance experiments inform us of the requirements for efficient mesh design and configuration of LTS regions for LTS schemes. In particular, we see that for LTS to be efficient on a given mesh, it is important to have enough cells using the coarse time-step relative to those using the fine time-step, typically at least 1:5 to see an increase in performance.
Here, we introduce a mathematically rigorous formulation for a nonlocal interface problem with jumps and propose an asymptotically compatible finite element discretization for the weak form of the interface problem. After proving the well-posedness of the weak form, we demonstrate that solutions to the nonlocal interface problem converge to the corresponding local counterpart when the nonlocal data are appropriately prescribed. Several numerical tests in one and two dimensions show the applicability of our technique, its numerical convergence to exact nonlocal solutions, its convergence to the local limit when the horizons vanish, and its robustness with respect to the patch test.
Sea ice plays an important role in the global climate system. The internal ice stress regulates the ice deformation and acts as a resistance to the motion. An accurate approximation of the internal ice stress is therefore crucial to determine the sea ice evolution.
Develop an efficient parallel implementation and assess the performance of local time-stepping (LTS) schemes for the shallow water equations in the Model for Prediction Across Scales (MPAS).
Time stepping algorithms are an important part of ocean models, and strongly influence both the accuracy of solution and performance. There have been a number of projects investigating various improvements for ocean time-stepping schemes in the Model for Prediction Across Scales-Ocean (MPAS-Ocean), a component of the DOE Energy Exascale Earth System Model. Ocean dynamics include fast surface gravity waves, which are two-dimensional, and slower internal waves, which are three-dimensional, so ocean models use a split time-stepping scheme that separates these barotropic and baroclinic modes for efficiency. MPAS-Ocean runs on variable-resolution horizontal meshes, and must scale to tens of thousands of cores and millions of horizontal gridcells. Ocean models require time stepping algorithms that are customized to these needs, and which are tuned for performance on various resolutions and architectures.
Historically, B-grid formulations of sea ice dynamics have been dominant because they have matched the grid type used by ocean models. The reason for the grid match is simple – it facilitates penetration of the curl of ice-ocean stress into the deep ocean with minimal numerical diffusivity because sea ice and ocean velocity are co-located. In recent years, as ocean models have increasingly progressed to C-grids, sea ice models have followed suit on quadrilateral meshes, but the implementation of an unstructured C-grid sea ice models is new. We present an unstructured C-grid discretization of the Elastic Viscous Plastic (EVP) rheology, where the velocity unknowns are discretized at the edges of the mesh cells with n-sides, where typically n is greater than or equal to four, rather than at the vertices, as in the B-grid. Our framework of choice is the Model for Prediction Across Scales (MPAS) within E3SM, the climate model of the U.S. Department of Energy, although our approach is general and could be applied to other models as well. While MPAS-Seaice is currently defined on a B-grid, MPAS-Ocean runs on a C-grid, hence interpolation operators are heavily used when coupled simulations are performed. In this work, we describe a mathematical formulation to transition the dynamics of MPAS-Seaice to a C-grid, in order to ultimately facilitate the coupling with MPAS-Ocean and reduce numerical errors associated with this communication. Numerical results are reported to highlight the features of the method.
In this work we assess the performance of a set of local time-stepping (LTS) schemes for the shallow water equations implemented in the Model for Prediction Across Scales (MPAS). The goal of LTS is to speed up the simulation by allowing different time-steps on different regions of the computational grid. The LTS schemes considered here were originally introduced by Hoang et al. (2019) [26], who laid out the mathematical foundation of the methods. Here, the authors take on the task of presenting a fast, efficient and scalable parallel implementation of these LTS methods on high performance computing machines, with the aim to provide a recipe for other climate modeling groups that may be interested in employing LTS algorithms in their codes. As a matter of fact, even if MPAS is our framework of choice, our approach is general enough and could be of interest to other groups beyond the MPAS community. Due to their nature, LTS methods possess an inherent load imbalance that needs to be carefully addressed in order to obtain efficient scalability. Even more important is the far from trivial task of computing the right-hand side terms only on specific LTS regions during the time-stepping procedure. An inefficient handling of this task causes a drastic decay of the CPU time performance, making the LTS algorithms practically of no use. The emphasis of the present work is therefore on the computational and parallel aspects of the LTS methods, whose proper treatment is crucial to make the methods run faster against existing strategies, such as for instance high-order explicit global time-stepping schemes. This is in fact the ultimate goal of using an LTS procedure and it is the one to which we direct all our optimization efforts.
Abstract A mathematical framework is provided for a substructuring‐based domain decomposition (DD) approach for nonlocal problems that features interactions between points separated by a finite distance. Here, by substructuring it is meant that a traditional geometric configuration for local partial differential equation (PDE) problems is used in which a computational domain is subdivided into non‐overlapping subdomains. In the nonlocal setting, this approach is substructuring‐based in the sense that those subdomains interact with neighboring domains over interface regions having finite volume, in contrast to the local PDE setting in which interfaces are lower dimensional manifolds separating abutting subdomains. Key results include the equivalence between the global, single‐domain nonlocal problem and its multi‐domain reformulation, both at the continuous and discrete levels. These results provide the rigorous foundation necessary for the development of efficient solution strategies for nonlocal DD methods.
Here, in this paper, we design efficient quadrature rules for finite element (FE) discretizations of nonlocal diffusion problems with compactly supported kernel functions. Two of the main challenges in nonlocal modeling and simulations are the prohibitive computational cost and the nontrivial implementation of discretization schemes, especially in three-dimensional settings. In this work, we circumvent both challenges by introducing a parametrized mollifying function that improves the regularity of the integrand, utilizing an adaptive integration technique, and exploiting parallelization. We first show that the “mollified” solution converges to the exact one as the mollifying parameter vanishes, then we illustrate the consistency and accuracy of the proposed method on several two- and three-dimensional test cases. Furthermore, we demonstrate the good scaling properties of the parallel implementation of the adaptive algorithm and we compare the proposed method with recently developed techniques for efficient FE assembly.
Multi-resolution global ocean simulations are possible with the Model for Prediction Across Scales-Ocean (MPAS-Ocean), although the model is currently constrained to a uniform time step. The time step is determined by the size of the smallest cell in the grid, according to the CFL condition. To overcome this issue, a local time stepping (LTS) scheme has been designed for MPAS-Ocean. We have made progress on implementing the LTS scheme, as well as quantitative results and performance tests.
Time stepping algorithms are an important part of ocean models, and strongly influence both the accuracy of solution and performance. There have been a number of projects investigating various improvements for ocean time-stepping schemes in the Model for Prediction Across Scales-Ocean (MPAS-Ocean), a component of the DOE Energy Exascale Earth System Model. Ocean dynamics include fast surface gravity waves, which are two-dimensional, and slower internal waves, which are three-dimensional, so ocean models use a split time-stepping scheme that separates these barotropic and baroclinic modes for efficiency. MPAS-Ocean runs on variable-resolution horizontal meshes, and must scale to tens of thousands of cores and millions of horizontal gridcells. Ocean models require time stepping algorithms that are customized to these needs, and which are tuned for performance on various resolutions and architectures.
In this paper we design efficient quadrature rules for finite element discretizations of nonlocal diffusion problems with compactly supported kernel functions. Two of the main challenges in nonlocal modeling and simulations are the prohibitive computational cost and the nontrivial implementation of discretization schemes, especially in three-dimensional settings. In this work we circumvent both challenges by introducing a parametrized mollifying function that improves the regularity of the integrand, utilizing an adaptive integration technique, and exploiting parallelization. We first showthat the “mollified” solution converges to the exact one as the mollifying parameter vanishes, then we illustrate the consistency and accuracy of the proposed method on several two- and three-dimensional test cases. Furthermore, we demonstrate the good scaling properties of the parallel implementation of the adaptive algorithm and we compare the proposed method with recently developed techniques for efficient finite element assembly.