Abstract
In this paper, elliptic PDE-constrained optimization problems with box constraints on the control are considered. To numerically solve the problems, we apply the ‘optimize-discretize-optimize’ strategy. Specifically, the alternating direction method of multipliers (ADMM) algorithm is applied in function space first, and then, the standard piecewise linear finite-element approach is employed to discretize the subproblems in each iteration. Finally, some efficient numerical methods are applied to solve the discretized subproblems based on their structures. Motivated by the idea of the multi-level strategy, instead of fixing the mesh size before the computation process, we propose the strategy of gradually refining the grid. Moreover, the subproblems in each iteration are solved by appropriate inexact methods. Based on the strategies above, an efficient convergent multi-level ADMM (mADMM) algorithm is proposed. We present the convergence analysis and the iteration complexity results o(1/k) for the mADMM algorithm. Some numerical experiments are done and the numerical results show the high efficiency of the mADMM algorithm.
Similar content being viewed by others
References
Bai Z, Benzi M, Chen F, Wang Z (2013) Preconditioned MHSS iteration methods for a class of block two-by-two linear systems with applications to distributed control problems. IMA J Numer Anal 33(1):343–369
Brandt A (1977) Multi-level adaptive solutions to boundary-value problems. Math Comp 31(138):333–390
Casas E, Tröltzsch F (2002) Error estimates for linear-quadratic elliptic control problems. International Worksing Conference on Analysis and Optimization of Differential Systems. DBLP
Chen L (2009) iFEM: An integrated finite element methods package in MATLAB. Technical Report. University of California at Irvine, Irvine
Chen Z, Song X, Zhang X, Yu B (2019) A FE-ADMM algorithm for Lavrentiev-regularized state-constrained elliptic control problem. ESAIM Control Optim Calc Var 25:5
Chen L, Sun D, Toh KC (2017) An efficient inexact symmetric Gauss-Seidel based majorized ADMM for high-dimensional convex composite conic programming. Math Program 161(1–2):237–270
Ciarlet PG (2002) The Finite Element Method for Elliptic Problems. Volume 40 of Classics in Applied Mathematics. Society for Industrial and Applied Mathematics, Philadelphia
Deuflhard P (2011) Newton methods for nonlinear problems: affine invariance and adaptive algorithms. Springer, Berlin
Ern A, Guermond JL (2004) Theory and practice of finite elements. Springer, New York
Fazel M, Pong TK, Sun D, Tseng P (2013) Hankel matrix rank minimization with applications to system identification and realization. SIAM J Matrix Anal Appl 34(3):946–977
Fortin M, Glowinski R (1983) On decomposition-coordination methods using an augmented Lagrangian. Augmented Lagrangian Methods: Applications to the Solution of Boundary Problems. Elsevier, Amsterdam
Gabay D, Mercier B (1976) A dual algorithm for the solution of nonlinear variational problems via finite element approximation. Comput Math Appl 2:17–40
Glowinski R (1980) Lectures on numerical methods for nonlinear variational problems. Springer, Berlin
Glowinski R, Marroco A (1975) Sur l’approximation, par éléments finis d’ordre un, et la résolution, par pénalisation-dualité d’une classe de problèmes de Dirichlet non linéaires. Analyse Numérique 9(R2):41–76
Hackbusch W (1978) On the multi-grid method applied to difference equations. Computing 20(4):291–306
Hackbusch W (1985) Multi-grid methods and applications. Springer, Berlin
Han D, Sun D, Zhang L (2017) Linear rate convergence of the alternating direction method of multipliers for convex composite programming. Math Oper Res 43(2):622–637
Hinze M, Meyer C (2010) Variational discretization of Lavrentiev-regularized state constrained elliptic optimal control problems. Comput Optim Appl 46(3):487–510
Hinze M, Vierling M (2012) The semi-smooth Newton method for variationally discretized control constrained elliptic optimal control problems; implementation, convergence and globalization. Optim Methods Softw 27(6):933–950
Hinze M, Pinnau R, Ulbrich M, Ulbrich S (2009) Optimization with PDE constraints. Springer, Berlin
Kinderlehrer D, Stampacchia G (1980) an introduction to variational inequalities and their applications. Academic Press, New York
Li X, Sun D, Toh KC (2016) A Schur complement based semi-proximal ADMM for convex quadratic conic programming and extensions. Math Program 155(1–2):333–373
Li X, Sun D, Toh KC (2018) QSDPNAL: A two-phase Newton-CG proximal augmented Lagrangian method for convex quadratic semidefinite programming problems. Math Program Comput 10:703–743
Li J, Wang X, Zhang K (2018) An efficient alternating direction method of multipliers for optimal control problems constrained by random Helmholtz equations. Numer Algorith 78(1):161–191
Saad Y, Schultz MH (1986) GMRES: Ageneralized minimum residual algorithm for solving nonsymmetric linear systems. SIAM J Sci Stat Comput 7(3):856–869
Song X (2018) Some alternating direction iteration methods for solving PDE-constrained optimization problems. PhD thesis, Dalian University of Technology, Dalian
Song X, Yu B (2018) A two phase strategy for control constrained elliptic optimal control problems. Numer Linear Algebra Appl 25:e2138
Song X, Yu B, Zhang X, Wang Y (2017) A FE-inexact heterogeneous ADMM algorithm for elliptic optimal control problems with \(L^{1}\)-control cost. J Syst Sci Complex 31(6):1659–1697
Sun D, Toh KC, Yang L (2015) A convergent 3-block semiproximal alternating direction method of multipliers for conic programming with 4-type constraints. SIAM J Optim 25(2):882–915
Yang L, Li J, Sun D, Toh KC (2018) A fast globally linearly convergent algorithm for the computation of Wasserstein Barycenters. arXiv preprint arXiv:1809.04249
Zhang K, Li J, Song Y, Wang X (2017) An alternating direction method of multipliers for elliptic equation constrained optimization problem. Sci China Math 60(2):361–378
Acknowledgements
We would like to thank Prof. Long Chen very much for his FEM package iFEM Chen (2009) in Matlab. This work is supported by the National Natural Science Foundation of China (No. 11971092).
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Jinyun Yuan.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Proof of Lemma 2
We employ the mathematical induction to prove the conclusion. While \(k=1\), for u-subproblems, \(\bar{u}^{1}\) and \(\bar{u}_{h_{1}}^{1}\) satisfy the following optimality conditions, respectively:
By subtracting two equalities above, we have:
where we define:
For the term \(E_{1}\), we make use of the decomposition:
Notice that \(\Vert S-S_{h_{1}}\Vert \le ch_{1}^{2}\) and \(S,S^{*},S_{h_{1}},S^{*}_{h_{1}}\) are bounded linear operators. From Lemma 1 and the triangle inequality, we have \(\Vert S_{h_{1}}\Vert _{\mathcal L(L^{2}, L^{2})}\le ch_{1}^{2}+\Vert S\Vert _{\mathcal L(L^{2}, L^{2})}\). With the fact that \(h_{1}\) can be bounded above by a non-negative constant, \(\Vert S_{h_{1}}\Vert _{\mathcal L(L^{2}, L^{2})}\) can be bounded above by a non-negative constant independent of \(h_{1}\). Thus, we have:
where \(c_{1}\) is a constant independent of \(h_{1}\).
For the term \(E_{2}\):
where \(c_{2}\) is a constant independent of \(h_{1}\). In the last equality, we used the property that \(h_{1}\) can be bounded above by a non-negative constant.
For the term \(E_{3}\), similarly:
where \(c_{3}\) is a constant independent of \(h_{1}\).
For the term \(E_{4}\):
where \(C_{\lambda , 1}:=c_{I}\Vert \lambda ^{0}\Vert _{H^{1}(\varOmega _{h_{1}})}\) is a constant independent of \(h_{1}\).
For the term \(E_{5}\), similarly:
Then, we know from (A.1) and the estimations of \(L^{2}\) norms of \(\lbrace E_{i}\rbrace _{i=1}^{5}\) that there exist a constant \(\hat{C}_{u,1}\) independent of \(h_{1}\), such that:
Notice that:
Thus, we have:
where \(C_{u,1}:=\frac{1}{(\alpha +\sigma )^{2}}\hat{C}_{u,1}\) is a constant independent of \(h_{1}\).
For z-subproblems, \(\bar{z}^{1}\) and \(\bar{z}^{1}_{h_{1}}\) satisfy:
then, we know from the projection operator \(\Pi \) is nonexpansive that:
where \(C_{z, 1}:=C_{u,1}+\frac{1}{\sigma }C_{\lambda , 1}\).
Hence, we have completed the proof for the case \(k=1\).
While \(k>1\), we assume that, for \(\forall j\le k,\) we have \(\Vert \bar{u}^{j}-\bar{u}^{j}_{h_{j}}\Vert _{L^{2}(\varOmega _{h_{j}})} \le C_{u,j}h_{j}\), \(\Vert \bar{z}^{j}-\bar{z}^{j}_{h_{j}}\Vert _{L^{2}(\varOmega _{h_{j}})} \le C_{z,j}h_{j}\) and \(\Vert \bar{\lambda }^{j-1}-\bar{\lambda }^{j-1}_{h_{j}}\Vert _{L^{2}(\varOmega _{h_{j}})} \le C_{\lambda ,j}h_{j}\), where \(C_{u,j}, C_{z,j}, C_{\lambda ,j}\) are constants independent of \(h_{j}\).
For u-subproblems, \(\bar{u}^{k+1}\) and \(\bar{u}^{k+1}_{h_{k+1}}\) satisfy the following optimality conditions, respectively:
By subtracting two equalities above, we have:
where we define:
For the term \(\hat{E}_{1}\), we make use of the decomposition:
Notice that \(\Vert S-S_{h_{k+1}}\Vert \le ch_{k+1}^{2}\) and \(S,S^{*},S_{h_{k+1}},S^{*}_{h_{k+1}}\) are bounded linear operators. Moreover, we know from Lemma 1 and the triangle inequality that \(\Vert S_{h_{k+1}}\Vert _{\mathcal L(L^{2}, L^{2})}\le ch_{k+1}^{2}+\Vert S\Vert _{\mathcal L(L^{2}, L^{2})}\). With the fact that \(h_{k+1}\) can be bounded above by a non-negative constant, \(\Vert S_{h_{k+1}}\Vert _{\mathcal L(L^{2}, L^{2})}\) can be bounded above by a non-negative constant independent of \(h_{k+1}\). Thus, we have:
where \(\hat{c}_{1}\) is a constant independent of \(h_{k+1}\) and it can be bounded above by a non-negative constant.
Similarly, for the term \(\hat{E}_{2}\) and \(\hat{E}_{3}\), we have:
and
where \(\hat{c}_{2}\) and \( \hat{c}_{3}\) are constants independent of \(h_{k+1}\), and both of them can be bounded above by a non-negative constant.
For the term \(\hat{E}_{4}\):
where \(C_{\lambda ,k+1}\) is a constant independent of \(h_{k+1}\), and it can be bounded above by a non-negative constant. In the last equality, we use the property \(\sum _{k=0}^{\infty }h_{k}< \infty \), and thus, there exists a constant \(C_{k+1}\), such that \(h_{k}<C_{k+1}h_{k+1}\).
For the term \(\hat{E}_{5}\), similarly:
where \(\hat{c}_{5}\) is a constant independent of \(h_{k+1}\) and it can be bounded above by a non-negative constant.
Then, we know from (A.2) and the estimations of \(L^{2}\) norms of \(\lbrace \hat{E}_{i}\rbrace _{i=1}^{5}\) that there exists a constant \(\hat{C}_{u, k+1}\) independent of \(h_{k+1}\), such that:
Notice that:
Thus, we have:
where \(C_{u,k+1}:=\frac{1}{(\alpha +\sigma )^{2}}\hat{C}_{u,k+1}\) is a constant independent of \(h_{k+1}\).
Hence, the conclusion holds for the case \(k+1\), and we can get the assertion:
Moreover, for z-subproblems, we have:
where \(C_{z,k+1}:=C_{u,k+1}+\dfrac{1}{\sigma }C_{\lambda ,k+1}\) is a constant independent of \(h_{k+1}\), and it can be bounded above by a non-negative constant. Hence, we have completed the proof.
Proof of Proposition 1
First, for any \(f_{1}, f_{1}^{\prime }, f_{2}, f_{2}^{\prime }\in L^{2}(\varOmega )\), we have the following two important equalities:
The proof of the above two equalities can be easily obtained by the definition of \(L^{2}-\)norm.
By the optimality condition at point \((u^{k}_{h_{k}},z^{k}_{h_{k}})\), we have:
Moreover, the KKT point \(\left( u^{*}_{h_{k}}, z^{*}_{h_{k}}, \lambda ^{*}_{h_{k}}\right) \) of (DRP) satisfies:
Then, we have:
Moreover, from the definition of the subdifferential operator, we have the following inequality:
For the convenience of analyzing, we define \(r_{h_{k}}^{k}=u_{h_{k}}^{k}-z_{h_{k}}^{k}\). By substituting (B.3)–(B.7) into (B.8) and (B.9), we have:
By adding the above two equalities, we obtain:
Next, we estimate the last two terms on the left side separately:
where we used (B.1).
Moreover, by employing (B.2) and using \(u^{*}_{h_{k}}=z^{*}_{h_{k}}\), we have:
Then, substituting (B.11) and (B.12) into (B.10), we can get the assertion of Proposition 1.
Rights and permissions
About this article
Cite this article
Chen, X., Song, X., Chen, Z. et al. A multi-level ADMM algorithm for elliptic PDE-constrained optimization problems. Comp. Appl. Math. 39, 331 (2020). https://doi.org/10.1007/s40314-020-01379-1
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s40314-020-01379-1