报告题目1：A randomized point-block Schwarz preconditioner on GPUs for solving nonlinear problems

报告人：马文鹏 博士

报告时间：4月9号下午14：30开始

报告地点：数学与统计学院315报告厅

报告摘要：We propose a randomized point-block Schwarz preconditioner on multiple GPUs for solving nonlinear systems arising from multiphysics problems. To enhance the performance of the traditional Schwarz preconditioner on state-of-the-art GPU clusters, we investigate the potential of randomized computations that fit the fine-grained characteristics of GPUs. The discussions in this talk include a randomized, iterative concept for the two building blocks of the preconditioner construction, the point-block ILU factorization and the triangular systems of equations. Then the thread partition strategies for the efficient implementation on heterogeneous computing systems consisting of both CPUs and GPUs are introduced. Some representative cases are also studied to show the advantages of the randomized preconditioner over the traditional deterministic version.

马文鹏，博士，副教授，2015年获中国科学院大学博士学位，主要从事高性能计算，GPU计算、大规模并行数值计算软件的研究，主持完成国家自然科学基金青年项目1项，省部级项目2项。在学术期刊The International Journal of Supercomputing Applications, CCF Transactions on HPC等发表论文10余篇。

报告题目2：Residual-Based a Posteriori Error Estimates for the Time-Dependent Ginzburg–Landau Equations of Superconductivity

报告人： 张秋雨 博士

报告时间：4月9日15：30开始

报告地点：数学与统计学院315报告厅

报告摘要：In this paper, we propose and analyze residual-based a posteriori error estimator for a new finite element method for the time-dependent Ginzburg–Landau equations with the temporal gauge of superconductivity. The magnetic potential variable is approximated by H(curl)-conforming element and the scalar order parameter is approximated by the H1-conforming element. Using the dual problem of a linearization of the original problem, we prove the reliability of the a posteriori error estimator, and an adaptive algorithm with the temporal and spatial refining and coarsening steps is then proposed. Numerical results are presented for illustrating the a posteriori error estimator and the adaptive algorithm in convex and nonconvex domains.

张秋雨，博士，2023年获得武汉大学理学博士学位，主要从事有限元方法、自适应方法等方面的研究，在学术期刊Journal of Scientific Computing， Numerical Methods for Partial Differential Equations，Numerical Algorithms等发表SCI论文4篇。