Algebraic Multigrid Method for Solving Finite Element Equations Combined with Algorithm-Based Error Detection

摘要

A low-overhead, error-detecting version of large Finite Element (FE) sparse equations Algebraic Multigrid (AMG) solver is presented for parallel and distributed computations. Invariants are created to implement checking in the relaxing, restriction and interpolation operations of AMG The float-point roundoff error's bound expression is used to act as error detection criterion. The efficient AMG solver combined with ABED is used to solve a electrode grounding system as a example. The numerical results show our algorithm posses low overhand and high error coverage.