Keywords: parallel adaptive mesh refinement, domain decomposition methods, AMPFETI, Krylov subspace recycling.

In a previous work, we developed a new domain decomposition method: Adaptive Multipreconditioned FETI (AMPFETI). AMPFETI is the combination of the dual domain decomposition method FETI and the adaptive multipreconditioned conjugate gradient. AMPFETI is robust enough to solve ill conditioned finite element systems arising from large scale non linear industrial problems.

Adaptive mesh refinement is the next step to fulfil parallel efficiency. Classical adaptive mesh refinement processes applied in a massively distributed context mainly follow three goals: to adapt the mesh size to the involved physics; to optimize the load balancing; and to minimize overall volume of communication.

The specificities of the employed domain decomposition method and of the underlying linear solver are rarely taken into account. For instance, the shape of the interfaces between subdomains may evolve during the process. The modification of the subdomains shape has two drawbacks. In the case of the non linear problem, the transfer of state variables (such as the plastic strain tensor), from the old subdomain mesh to the new one is no longer embarrassingly parallel. Also, the modification of the subdomains shape hinders the reuse of Krylov subspaces across time steps.

In this contribution, we present a new parallel mesh refinement process that preserves the subdomains interface shape. The developed algorithm and some numerical results demonstrating the scalability of the procedure will be shown. The advantage of preserving the interface geometry for Krylov subspace reuse will be presented on an industrial example

purchase the full-text of this paper (price £22)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £25 +P&P)