On the Convergence of System-AMG in Reservoir Simulation

Summary System-algebraic multigrid (AMG) provides a flexible framework for linear systems in simulation applications that involve various types of physical unknowns. Reservoir-simulation applications, with their driving elliptic pressure unknown, are principally well-suited to exploit System-AMG as a robust and efficient solver method. However, the coarse grid correction must be physically meaningful to speed up the overall convergence. It has been common practice in constrained-pressure-residual (CPR) -type applications to use an approximate pressure/saturation decoupling to fulfill this requirement. Unfortunately, this can have significant effects on the AMG applicability and, thus, is not performed by the dynamic row-sum (DRS) method. This work shows that the pressure/saturation decoupling is not necessary for ensuring an efficient interplay between the coarse grid correction process and the fine-level problem, demonstrating that a comparable influence of the pressure on the different involved partial-differential equations (PDEs) is much more crucial. As an extreme case with respect to the outlined requirement, linear systems from compositional simulations under the volume-balance formulation will be discussed. In these systems, the pressure typically is associated with a volume balance rather than a diffusion process. It will be shown how System-AMG can still be used in such cases.

On acceleration technologies of parallel decomposition methods

Одним из главных препятствий масштабированному распараллеливанию алгебраических методов декомпозиции для решения сверхбольших разреженных систем линейных алгебраических уравнений (СЛАУ) является замедление скорости сходимости аддитивного итерационного алгоритма Шварца в подпространствах Крылова при увеличении количества подобластей. Целью настоящей статьи является сравнительный экспериментальный анализ различных приeмов ускорения итераций: параметризованное пересечение подобластей, использование специальных интерфейсных условий на границах смежных подобластей, а также применение грубосеточной коррекции (агрегации, или редукции) исходной СЛАУ для построения дополнительного предобусловливателя. Распараллеливание алгоритмов осуществляется на двух уровнях программными средствами для распределeнной и общей памяти. Тестовые СЛАУ получаются при помощи конечно-разностных аппроксимаций задачи Дирихле для диффузионно-конвективного уравнения с различными значениями конвективных коэффициентов на последовательности сгущающихся сеток. One of the main obstacles to the scalable parallelization of the algebraic decomposition methods for solving large sparse systems of linear algebraic equations consists in slowing the convergence rate of the additive iterative Schwarz algorithm in the Krylov subspaces when the number of subdomains increases. The aim of this paper is a comparative experimental analysis of various ways to accelerate the iterations: a parametrized intersection of subdomains, the usage of interface conditions at the boundaries of adjacent subdomains, and the application of a coarse grid correction (aggregation, or reduction) for the original linear system to build an additional preconditioner. The parallelization of algorithms is performed on two levels by programming tools for the distributed and shared memory. The benchmark linear systems under study are formed using the finite difference approximations of the Dirichlet problem for the diffusion-convection equation with various values of the convection coefficients and on a sequence of condensing grids.

A Coarse Grid Correction to Domain Decomposition Based Preconditioners for Meso-Scale Simulation of Concrete

Meso-scale simulation is one of the important ways to study dynamic behaviors of concrete materials, while most of the simulation time is used to solve the sparse linear systems. Because the discrete grid is three dimensional and is of large scale, iterations are the best solutions. But the convergence depends on the distribution of the eigenvalues of the coefficient matrix, to make the eigenvalues distributed more closely each other, it is required to adopt preconditioning techniques. In this paper, with the characteristics of the sparse linear systems considered, there provides a coarse grid correction algorithm, which is based on domain decomposition preconditioners and aggregation of sub-domains, with each aggregated into a single super-node. A linear system with small scale size is formed, which contains the global information and the solution is used to correct the solution components of the original auxiliary linear system. For incomplete factorization preconditioner parallelized with block Jacobi, classic additive Schwarz, and factors combination techniques, the experiments show that the presented algorithm can improve the convergence rate and the efficiency.

Analysis of a Two-level Schwarz Method with Coarse Spaces Based on Local Dirichlet-to-Neumann Maps

Coarse grid correction is a key ingredient in order to have scalable domain decomposition methods. For smooth problems, the theory and practice of such two-level methods is well established, but this is not the case for problems with complicated variation and high contrasts in the coefficients. In a previous study, two of the authors introduced a coarse space adapted to highly heterogeneous coefficients using the low frequency modes of the subdomain DtN maps. In this work, we present a rigorous analysis of a two-level overlapping additive Schwarz method with this coarse space, which provides an automatic criterion for the number of modes that need to be added per subdomain to obtain a convergence rate of the order of the constant coefficient case. Our method is suitable for parallel implementation and its efficiency is demonstrated by numerical examples on some challenging problems with high heterogeneities for automatic partitionings.