Portal de Periódicos da CAPES

A-Posteriori Error Estimates for the Localized Reduced Basis Multi-Scale Method

2014; Springer International Publishing; Linguagem: Inglês

10.1007/978-3-319-05684-5_41

2194-1017

Mario Ohlberger, Felix Schindler,

Model Reduction and Neural Networks

We present a localized a-posteriori error estimate for the Localized Reduced Basis Multi-Scale (LRBMS) method [1]. The LRBMS is a combination of numerical multi-scale methods and model reduction using reduced basis methods to efficiently reduce the computational complexity of parametric multi-scale problems with respect to the multi-scale parameter $$\varepsilon $$ and the online parameter $$\mu $$ simultaneously. We formulate the LRBMS based on a generalization of the SWIPDG discretization presented in [2] on a coarse partition of the domain that allows for any suitable discretization on the fine triangulation inside each coarse grid element. The estimator is based on the idea of a conforming reconstruction of the discrete diffusive flux, presented in [2], that can be computed using local information only. It is offline/online decomposable and can thus be efficiently used in the context of model reduction.