1: \begin{abstract}

2: In this paper, we propose oversampling strategies in the Generalized 

3: Multiscale Finite Element Method (GMsFEM) framework. 

4: The GMsFEM, which has been recently introduced in \cite{egh12}, allows solving

5: multiscale parameter-dependent problems at a reduced computational cost

6: by constructing a reduced-order representation of the solution on a coarse

7: grid. The main idea of the method consists of (1) the construction

8: of snapshot space, (2) the construction of the offline space, and (3)

9: construction of the online space (the latter for parameter-dependent problems).

10:  In \cite{egh12}, it was shown that the GMsFEM

11: provides a flexible tool to solve multiscale problems 

12: with a complex input space by generating

13: appropriate snapshot, offline, and online spaces. In this paper,

14: we develop oversampling techniques to be used in this context (see \cite{hw97} 

15: where

16: oversampling is introduced for multiscale finite element methods).

17: It is known (see \cite{hw97})

18: that the oversampling can improve the accuracy of

19: multiscale methods. In particular, the oversampling technique uses

20: larger regions (larger than the target coarse block) in constructing

21: local basis functions.

22: Our motivation stems from the analysis presented in this paper which show

23: that when using oversampling techniques in the construction of the snapshot 

24: space

25: and offline space, GMsFEM will converge independent of

26: small scales and high-contrast under certain assumptions. 

27: We consider the use

28: of multiple eigenvalue problem to improve the convergence

29: and discuss their relation to single spectral problems that use oversampled

30: regions.

31:  The oversampling procedures

32: proposed in this paper differ from those in \cite{hw97}. In particular,

33: the oversampling domains are partially used in constructing local

34: spectral problems. We present numerical results and compare various

35: oversampling techniques in order to complement the proposed technique and analysis. 

36: 

37: 

38: \end{abstract}

39: