Randomized Local Model Order Reduction

Abstract

In this paper we propose local approximation spaces for localized model order reduction procedures such as domain decomposition and multiscale methods. Those spaces are constructed from local solutions of the partial differential equation (PDE) with random boundary conditions, yield an approximation that converges provably at a nearly optimal rate, and can be generated at close to optimal computational complexity. In many localized model order reduction approaches like the generalized finite element method, static condensation procedures, and the multiscale finite element method local approximation spaces can be constructed by approximating the range of a suitably defined transfer operator that acts on the space of local solutions of the PDE. Optimal local approximation spaces that yield in general an exponentially convergent approximation are given by the left singular vectors of this transfer operator [I. Babuška and R. Lipton 2011, K. Smetana and A. T. Patera 2016]. However, the direct calculation of these singular vectors is computationally very expensive. In this paper, we propose an adaptive randomized algorithm based on methods from randomized linear algebra [N. Halko et al. 2011], which constructs a local reduced space approximating the range of the transfer operator and thus the optimal local approximation spaces. Moreover, the adaptive algorithm relies on a probabilistic a posteriori error estimator for which we prove that it is both efficient and reliable with high probability. Several numerical experiments confirm the theoretical findings.Key words. localized model order reduction, randomized linear algebra, domain decomposition methods, multiscale methods, a priori error bound, a posteriori error estimation AMS subject classifications. 65N15, 65N12, 65N55, 65N30, 65C20, 65N25

Randomized residual‐based error estimators for the proper generalized decomposition approximation of parametrized problems

“…Next, we describe how to build u M (µ) and Y L (µ) in an intertwined manner. Within each iteration of Algorithm 1 we first compute the primal rank-one correction using (9) Compute the primal rank-one correction using (9) or (10) 10:…”

Static Condensation Optimal Port/Interface Reduction and Error Estimation for Structural Health Monitoring

“…As in [2,6,24] we may then introduce a transfer operator T : S → R that takes arbitrary data ζ on Γ out as an input, solves the PDE Au = 0 on Ω with that data ζ as Dirichlet boundary conditions on Γ out , and finally restricts the local solution to Γ in . Introducing the source and range spaces S := {w| Γout : w ∈ H} and R := {(w − P ker(A) (w))| Γin : w ∈ H} the transfer operator is thus defined as T (w| Γout ) = w − P ker(A) (w) | Γin for w ∈ H.…”

“…In [6] it has been shown that by employing methods from randomized numerical linear algebra an extremely accurate approximation of those optimal port spaces can be computed in close to optimal computational complexity. To account for variations in a material or geometric parameter in [24] a parameter-independent port space is generated from the optimal parameter-dependent port spaces via a spectral greedy algorithm.…”

Fixed‐precision randomized low‐rank approximation methods for nonlinear model order reduction of large systems

“…Bistrian and Navon proposed an adaptive randomized DMD method to compute a reduced basis for nonintrusive MOR, which makes use of a fixed‐rank randomized SVD. Buhr and Smetana have used a rank‐adaptive approach for local MOR. It is based on randomized low‐rank approximation and uses methods and error estimators described in the work of Halko et al to construct a reduced space from solutions of the full system for randomized boundary conditions.…”