Reduced Basis Greedy Selection Using Random Training Sets

Reduced Basis Greedy Selection Using Random Training Sets
复制标题

使用随机训练集的减少基贪婪选择

DOI:
10.1051/m2an/2020004
复制
发表时间:
2020
期刊:
ESAIM: Mathematical Modelling and Numerical Analysis
影响因子:
--
通讯作者:
Nichols, James
Nichols, James
中科院分区:
--
文献类型:
--
作者:
Cohen, Albert;Dahmen, Wolfgang;DeVore, Ronald;Nichols, James

文献摘要

参考文献

被引文献

相似文献

在需要许多在线查询的应用程序中,为逼近参数化偏微分方程引入了简化基。Binevet等人(SIAM J. Math)已经从理论上证实了它们对这类问题的数值效率。al.43(2011) 1457-1472)和DeVoreet al.(建设性逼近37(2013)455-466),其中表明,由某种贪婪策略构造的降维基空间evn具有与解流形的kolmogorov宽度相关的最优空间相似的逼近误差。简化基空间的贪心构造是在离线阶段进行的,每一步都要求在参数空间上实现当前误差的最大化。为了数值计算的目的,这种最大化是在通过参数域离散化得到的有限训练集上进行的。为了保证贪心算法生成空间的最终逼近误差ε,原则上要求与该训练集相关联的快照构成一个精度为ε阶的解流形的逼近网络。因此,训练集的大小是ε覆盖数form,当解流形具有宽度衰减o (n−s)时,该覆盖数通常表现为exp(Cε−1/s)。因此,当ε较小时,训练集的剪切大小阻碍了算法的实现。本文的主要结果表明,如果我们愿意接受具有高概率而不是确定性的结果,那么对于大量的相关问题,我们可以用ε−1的大小多项式的随机训练集来代替精细离散化。我们对这个事实的证明是用高维多项式的逆不等式来建立的。
Reduced bases have been introduced for the approximation of parametrized PDEs in applications where many online queries are required. Their numerical efficiency for such problems has been theoretically confirmed in Binevet al.(SIAM J. Math. Anal.43(2011) 1457–1472) and DeVoreet al.(Constructive Approximation37(2013) 455–466), where it is shown that the reduced basis spaceVnof dimensionn, constructed by a certain greedy strategy, has approximation error similar to that of the optimal space associated to the Kolmogorovn-width of the solution manifold. The greedy construction of the reduced basis space is performed in an offline stage which requires at each step a maximization of the current error over the parameter space. For the purpose of numerical computation, this maximization is performed over a finitetraining setobtained through a discretization of the parameter domain. To guarantee a final approximation errorεfor the space generated by the greedy algorithm requires in principle that the snapshots associated to this training set constitute an approximation net for the solution manifold with accuracy of orderε. Hence, the size of the training set is theεcovering number forMand this covering number typically behaves like exp(Cε−1/s) for someC> 0 when the solution manifold hasn-width decayO(n−s). Thus, the shear size of the training set prohibits implementation of the algorithm whenεis small. The main result of this paper shows that, if one is willing to accept results which hold with high probability, rather than with certainty, then for a large class of relevant problems one may replace the fine discretization by a random training set of size polynomial in ε−1. Our proof of this fact is established by using inverse inequalities for polynomials in high dimensions.
DOI: 10.1080/10407790802424204
发表时间: 2008
期刊: Numerical Heat Transfer, Part B: Fundamentals
影响因子: --
作者:
S. Sen
通讯作者: S. Sen
DOI: --
发表时间: 2015
期刊:
影响因子: --
作者:
M. Bachmayr;A. Cohen;G. Migliorati
通讯作者: G. Migliorati
DOI: 10.1007/s00365-013-9186-2
发表时间: 2013-06-01
影响因子: 2.7
作者:
DeVore, Ronald;Petrova, Guergana;Wojtaszczyk, Przemyslaw
通讯作者: Wojtaszczyk, Przemyslaw
DOI: --
发表时间: 2019
期刊: Making Sense of the ECG
影响因子: --
作者:
A. Houghton
通讯作者: A. Houghton
Marcinkiewicz 型离散化定理
DOI: --
发表时间: 2017
影响因子: 2.7
作者:
V. Temlyakov;V. Temlyakov
通讯作者: V. Temlyakov