Acceleration Techniques for Adjoint-Based Error Estimation and Mesh Adaptation
In this paper we introduce two ideas for reducing the cost of output-based error estimation and mesh adaptation in steady and unsteady simulations.The first of these is the use of sub-iterations during adaptations,where a sub-iteration is an adaptation iteration in which the most expensive solves,the primal and fine-space adjoint,are done only approximately.The subiterations are interspersed with standard full-solve iterations during which accurate error estimates are available.The use of sub-iterations reduces the computational cost without much effect on the performance per iteration.The second strategy is the use of coarser spaces in the context of an adjoint-weighted residual for creating an adaptive indicator.While the resulting error estimate is not accurate,the adaptive indicator still contains useful information,at a much-reduced computational overhead compared to standard fine-space error estimation.We demonstrate these methods for steady,compressible Euler simulations discretized with the discontinuous Galerkin (DG) finite-element method,and for unsteady scalar advection simulations discretized with the active ux method.
Adjoint Solution Verification Numerical Error Estimation Mesh Adaptation
Kaihua Ding KaKrzysztof J. Fidkowski Philip L. Roe
University of Michigan, Ann Arbor, MI 48105, USA
国际会议
The 8th International Conference of Computational Fluid Dynamics, (ICCFD8)(第八届国际计算流体力学会议)
成都
英文
1-22
2014-07-25（万方平台首次上网日期，不代表论文的发表时间）