The multi-level Monte Carlo finite element method for a stochastic Brinkman problem (Q373227): Difference between revisions

From MaRDI portal
Importer (talk | contribs)
Created a new Item
 
Importer (talk | contribs)
Changed an Item
Property / author
 
Property / author: Claude Jeffrey Gittelson / rank
 
Normal rank
Property / review text
 
The authors present a numerical analysis of the stochastic Brinkman problem on a bounded domain \(D \in \mathbb R^d\) with parameters \(t \geq 0\) and \(G\), defined by \[ \begin{cases} - t^2 Au + \exp(G) u + \nabla p = f, \\ \;\;\;\text{div} u = g, \end{cases} \] where \(Au = \text{div } \big((\nabla u + \nabla u^{t})/2 \big)\), \(G\) is a Gaussian field on \(D\) with bounded paths and mean field \(G_0\) and \(\exp(\cdot)\) is the matrix exponential. The problem of interest is to compute the pair \((u,p)\) of solutions of the Brinkman problem for given boundary conditions. Since in a general framework these solutions have no analytical expression they have to be approximated. The proposed approximating method is based on a mixed finite element and a multilevel Monte Carlo method. The first step consists in approximating the matrix \(\exp(G)\) using the Karhunen-Loève expansion of \(G\), with truncation size \(N\). Considering the induced truncated stochastic Brinkman problem (TSBP), the authors show that the solution \((u_N,p_N)\) of the TSBP converges (w.r.t. the \(L^q\)-norm, \(q \geq 1\)) towards \((u,p)\) and that the rates of convergence (for both solutions \(u_N\) and \(p_N\)) is of order \(N^{-s}\), for some positive \(s\). As a consequence, it is shown that the global truncated error w.r.t. the \(L^q\)-norm: \(\| u -u_N \|_q + \| p -p_N \|_q\), is bounded (up to a positive constant) by \((\| f \|_{V^{\star}} + \| g \|_{Q^{\star}}) N^{-s}\), for a positive \(s\) and where \(V^{\star}\) and \(Q^{\star}\) are the dual spaces of the domains where \(u\) and \(p\) belong to. Then, the authors consider a mixed finite element approximation of the TSBP for which they show the stability of the associated solutions \((u_{N,h},p_{N,h})\), where \(h\) is the global mesh width and \(N\) is the size of the Karhunen-Loève truncation. Furthermore, they estimate the global error coming from the approximation of the solution \((u_N,p_N)\) of the TSBP by \((u_{N,h},p_{N,h})\). The authors show in particular that for regular solutions \(u_N\) and \(p_N\) of the TSBP, this global error is bounded (up to a positive constant) by a power (depending on the order of the mixed finite element discretization) of the global mesh width \(h\). The next step of the work is devoted to the error analysis of the single and multilevel Monte Carlo (MC) method when estimating an expectation of \(U=(u,p)\) by a sample mean. For the single MC method for example, a sample mean of size \(M\) is \((1/M) \sum_{i=1}^M \hat U_{N,h}^i\), where \((\hat U_{N,h}^1, \dots, \hat U_{N,h}^M)\) is a sample of size \(M\) with the same distribution as the solution of the mixed finite element approximation of the TSBP, means \((u_{N,h},p_{N,h})\). The induced error bounds make appear the Monte Carlo approximation error, the truncation error and the finite element discretization error. The last section is devoted to numerical experiments in \(\mathbb R^2\), which aim to verify the performance of the proposed method.
Property / review text: The authors present a numerical analysis of the stochastic Brinkman problem on a bounded domain \(D \in \mathbb R^d\) with parameters \(t \geq 0\) and \(G\), defined by \[ \begin{cases} - t^2 Au + \exp(G) u + \nabla p = f, \\ \;\;\;\text{div} u = g, \end{cases} \] where \(Au = \text{div } \big((\nabla u + \nabla u^{t})/2 \big)\), \(G\) is a Gaussian field on \(D\) with bounded paths and mean field \(G_0\) and \(\exp(\cdot)\) is the matrix exponential. The problem of interest is to compute the pair \((u,p)\) of solutions of the Brinkman problem for given boundary conditions. Since in a general framework these solutions have no analytical expression they have to be approximated. The proposed approximating method is based on a mixed finite element and a multilevel Monte Carlo method. The first step consists in approximating the matrix \(\exp(G)\) using the Karhunen-Loève expansion of \(G\), with truncation size \(N\). Considering the induced truncated stochastic Brinkman problem (TSBP), the authors show that the solution \((u_N,p_N)\) of the TSBP converges (w.r.t. the \(L^q\)-norm, \(q \geq 1\)) towards \((u,p)\) and that the rates of convergence (for both solutions \(u_N\) and \(p_N\)) is of order \(N^{-s}\), for some positive \(s\). As a consequence, it is shown that the global truncated error w.r.t. the \(L^q\)-norm: \(\| u -u_N \|_q + \| p -p_N \|_q\), is bounded (up to a positive constant) by \((\| f \|_{V^{\star}} + \| g \|_{Q^{\star}}) N^{-s}\), for a positive \(s\) and where \(V^{\star}\) and \(Q^{\star}\) are the dual spaces of the domains where \(u\) and \(p\) belong to. Then, the authors consider a mixed finite element approximation of the TSBP for which they show the stability of the associated solutions \((u_{N,h},p_{N,h})\), where \(h\) is the global mesh width and \(N\) is the size of the Karhunen-Loève truncation. Furthermore, they estimate the global error coming from the approximation of the solution \((u_N,p_N)\) of the TSBP by \((u_{N,h},p_{N,h})\). The authors show in particular that for regular solutions \(u_N\) and \(p_N\) of the TSBP, this global error is bounded (up to a positive constant) by a power (depending on the order of the mixed finite element discretization) of the global mesh width \(h\). The next step of the work is devoted to the error analysis of the single and multilevel Monte Carlo (MC) method when estimating an expectation of \(U=(u,p)\) by a sample mean. For the single MC method for example, a sample mean of size \(M\) is \((1/M) \sum_{i=1}^M \hat U_{N,h}^i\), where \((\hat U_{N,h}^1, \dots, \hat U_{N,h}^M)\) is a sample of size \(M\) with the same distribution as the solution of the mixed finite element approximation of the TSBP, means \((u_{N,h},p_{N,h})\). The induced error bounds make appear the Monte Carlo approximation error, the truncation error and the finite element discretization error. The last section is devoted to numerical experiments in \(\mathbb R^2\), which aim to verify the performance of the proposed method. / rank
 
Normal rank
Property / reviewed by
 
Property / reviewed by: Abass Sagna / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 65C30 / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 65C05 / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 60H15 / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 35R60 / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 60H35 / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 35J60 / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 65N30 / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 65N12 / rank
 
Normal rank
Property / Mathematics Subject Classification ID
 
Property / Mathematics Subject Classification ID: 65N15 / rank
 
Normal rank
Property / zbMATH DE Number
 
Property / zbMATH DE Number: 6217654 / rank
 
Normal rank
Property / zbMATH Keywords
 
Karhunen-Loève expansion
Property / zbMATH Keywords: Karhunen-Loève expansion / rank
 
Normal rank
Property / zbMATH Keywords
 
mixed finite element
Property / zbMATH Keywords: mixed finite element / rank
 
Normal rank
Property / zbMATH Keywords
 
multilevel Monte Carlo
Property / zbMATH Keywords: multilevel Monte Carlo / rank
 
Normal rank
Property / zbMATH Keywords
 
error bound
Property / zbMATH Keywords: error bound / rank
 
Normal rank
Property / zbMATH Keywords
 
stochastic Brinkman problem
Property / zbMATH Keywords: stochastic Brinkman problem / rank
 
Normal rank
Property / zbMATH Keywords
 
convergence
Property / zbMATH Keywords: convergence / rank
 
Normal rank
Property / zbMATH Keywords
 
stability
Property / zbMATH Keywords: stability / rank
 
Normal rank
Property / zbMATH Keywords
 
numerical experiments
Property / zbMATH Keywords: numerical experiments / rank
 
Normal rank

Revision as of 09:58, 29 June 2023

scientific article
Language Label Description Also known as
English
The multi-level Monte Carlo finite element method for a stochastic Brinkman problem
scientific article

    Statements

    The multi-level Monte Carlo finite element method for a stochastic Brinkman problem (English)
    0 references
    0 references
    0 references
    0 references
    22 October 2013
    0 references
    The authors present a numerical analysis of the stochastic Brinkman problem on a bounded domain \(D \in \mathbb R^d\) with parameters \(t \geq 0\) and \(G\), defined by \[ \begin{cases} - t^2 Au + \exp(G) u + \nabla p = f, \\ \;\;\;\text{div} u = g, \end{cases} \] where \(Au = \text{div } \big((\nabla u + \nabla u^{t})/2 \big)\), \(G\) is a Gaussian field on \(D\) with bounded paths and mean field \(G_0\) and \(\exp(\cdot)\) is the matrix exponential. The problem of interest is to compute the pair \((u,p)\) of solutions of the Brinkman problem for given boundary conditions. Since in a general framework these solutions have no analytical expression they have to be approximated. The proposed approximating method is based on a mixed finite element and a multilevel Monte Carlo method. The first step consists in approximating the matrix \(\exp(G)\) using the Karhunen-Loève expansion of \(G\), with truncation size \(N\). Considering the induced truncated stochastic Brinkman problem (TSBP), the authors show that the solution \((u_N,p_N)\) of the TSBP converges (w.r.t. the \(L^q\)-norm, \(q \geq 1\)) towards \((u,p)\) and that the rates of convergence (for both solutions \(u_N\) and \(p_N\)) is of order \(N^{-s}\), for some positive \(s\). As a consequence, it is shown that the global truncated error w.r.t. the \(L^q\)-norm: \(\| u -u_N \|_q + \| p -p_N \|_q\), is bounded (up to a positive constant) by \((\| f \|_{V^{\star}} + \| g \|_{Q^{\star}}) N^{-s}\), for a positive \(s\) and where \(V^{\star}\) and \(Q^{\star}\) are the dual spaces of the domains where \(u\) and \(p\) belong to. Then, the authors consider a mixed finite element approximation of the TSBP for which they show the stability of the associated solutions \((u_{N,h},p_{N,h})\), where \(h\) is the global mesh width and \(N\) is the size of the Karhunen-Loève truncation. Furthermore, they estimate the global error coming from the approximation of the solution \((u_N,p_N)\) of the TSBP by \((u_{N,h},p_{N,h})\). The authors show in particular that for regular solutions \(u_N\) and \(p_N\) of the TSBP, this global error is bounded (up to a positive constant) by a power (depending on the order of the mixed finite element discretization) of the global mesh width \(h\). The next step of the work is devoted to the error analysis of the single and multilevel Monte Carlo (MC) method when estimating an expectation of \(U=(u,p)\) by a sample mean. For the single MC method for example, a sample mean of size \(M\) is \((1/M) \sum_{i=1}^M \hat U_{N,h}^i\), where \((\hat U_{N,h}^1, \dots, \hat U_{N,h}^M)\) is a sample of size \(M\) with the same distribution as the solution of the mixed finite element approximation of the TSBP, means \((u_{N,h},p_{N,h})\). The induced error bounds make appear the Monte Carlo approximation error, the truncation error and the finite element discretization error. The last section is devoted to numerical experiments in \(\mathbb R^2\), which aim to verify the performance of the proposed method.
    0 references
    0 references
    Karhunen-Loève expansion
    0 references
    mixed finite element
    0 references
    multilevel Monte Carlo
    0 references
    error bound
    0 references
    stochastic Brinkman problem
    0 references
    convergence
    0 references
    stability
    0 references
    numerical experiments
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references