“…There has been many existing work concerning the design of novel numerical methods for multiscale problems and the mathematics to foresee and assess their performance in engineering and scientific applications, such as homogenization [25,41], numerical homogenization [2,13,44], heterogeneous multi-scale methods [1,14,27,30], multiscale network approximations [6], multi-scale finite element methods [3,8,15,17], variational multi-scale methods [5,24], flux norm homogenization [7,38], rough polyharmonic splines (RPS) [40], generalized multi-scale finite element methods [10,11,16], localized orthogonal decomposition [21,22,29,42], etc. Fundamental questions for numerical homogenization are: how to approximate the high dimensional solution space by a low dimensional approximation space with optimal error control, and furthermore, how to construct the approximation space efficiently, for example, whether its basis can be localized on a coarse patch.…”