Abstract
We prove the solvability of the optimal control problem for elliptic equation with nonlocal boundary conditions in a circular sector with terminal quadratic cost functional in the class of distributed controls.
Access provided by Autonomous University of Puebla. Download chapter PDF
Similar content being viewed by others
Keywords
- Elliptic Boundary Value Problems
- Circular Section
- Distributed Control
- Nonlocal Boundary Conditions
- Linear Quadratic Optimal Control Problem
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
1 Introduction
The theory of linear-quadratic optimal control problems for distributed systems is well researched [1, 2]. In many cases the original problem can be decomposed with the help of Fourier method [3–5]. In this chapter we consider the control problem for elliptic equation with non-local boundary conditions in circular sector [6] with terminal quadratic cost functional. This problem does not allow total decomposition and using of \(L^2\)-theory. To resolve this problem in the class of distributed controls we use apparatus of specially constructed biorthonormal basis systems of functions [7] and then we analyze the solutions of Fredholm matrix equations.
2 Setting of the Problem
In a circular sector \(Q=\{(r,\theta ) | r\in (0,1), \ \theta \in (0,\pi )\}\) we consider the optimal control problem
where \(p\in C^1([0,\pi ])\) is given function, \(\alpha \in (0,1)\) is fixed number, \(\Vert \cdot \Vert _D\) is a norm in \(L^2(0,\pi )\), which is equivalent to standard one and is given by the equality
where \(\forall n\ge 1\), \(v_n=\int \limits _0^\pi v(\theta )\psi _n(\theta )d\theta \), \(\psi _0(\theta )=\frac{2}{\pi ^2}\), \(\psi _{2n}(\theta )=\frac{4}{\pi ^2} (\pi -\theta )\sin 2n\theta \), \(\psi _{2n-1}(\theta )=\frac{4}{\pi ^2}\cos 2n\theta \).
The aim of this paper is to establish classical solvability of the problem (21.1)–(21.2), that is to find optimal one among admissible processes \(\{u,y\}\in C(\bar{Q})\times \left( C(\bar{Q})\bigcap C^2(Q)\right) \). For the application of the spectral method we use biorthonormal and complete in \(L^2(0,\pi )\) well-known Samarsky-Ionkin systems of functions [7]
Then \(\forall u\in L^2(Q)\)
where \(u_n(r)=\int _0^\pi u(r,\theta )\psi _n(\theta )d\theta \). So, we will seek for the solution of the problem (21.1) in form
where the functions \(\{y_k(r)\}_{k=0}^\infty \) are solutions of the system of ordinary differential equations
where \(p_k=\int _0^\pi p(\theta ) \cdot \psi _k(\theta )d\theta \).
Thus the original problem (21.1)–(21.2) is reduced to the following one: among admissible pairs \(\{u_n(r),y_n(r)\}_{n=0}^\infty \) of the problem (21.6)–(21.8) one should minimize the cost functional
Herewith the optimal process \(\{\tilde{u}_n(r),\tilde{y}_n(r)\}_{n=0}^\infty \) should be such that the formula (21.4) defines function from \(C(\bar{Q})\), and the formula (21.5) defines function from \(C(\bar{Q})\bigcap C^2(Q)\).
3 Main Results
A structure of the problem (21.6)–(21.8) allows to reduce it to sequence of the following problems:
On the solutions of (21.6) one should minimize the cost functional
on the solutions of (21.7), (21.8) one should minimize the cost functional
For fixed \(\{u_k(r)\}_{k=0}^\infty \subset C([0,1])\) solutions of the problem (21.6)–(21.8) have form
where
where
where
Lemma 21.1
For any \(k\ge 0\) the formulas (21.12)–(21.14) define the solutions of the problem (21.6)–(21.8) \(y_k\in C([0,1])\bigcap C^2(0,1)\).
Proof
Since \(y_k\) are the solutions of the problem (21.6)–(21.8), then it suffices to show that \(\forall k\ge 0\) \(y_k\in C([0,1])\). We denote \(\prod =[0,1]\times [0,1]\). Then \(G_0 \in C(\prod )\), \(\max \limits _{\prod }|G_0(r,s)|=e^{-1}\), so, \(y_0\in C([0,1])\).
For \(k\ge 1\) \(G_k\in C(\prod \setminus \{0,0\})\), \(\max \limits _{\prod }|G_k(r,s)|\le 1\), so, \(y_{2k-1}\in C([0,1])\). Since \(x^k\ln x\in C([0,1])\), \(\max \limits _{x\in [0,1]}|x^k\ln x|=e^{-1}\cdot k^{-1}\), then for \(\bar{G}_k\in C(\prod \setminus \{0,0\})\) we have: \(\max \limits _{\prod }|\bar{G}_k(r,s)|\le \frac{1}{k}\), so, \(y_{2k}\in C([0,1])\). Lemma is proved.
Theorem 21.1
The problems (21.10), (21.11) have the unique solution \(\{\tilde{u}_k\}_{k=0}^\infty \), moreover \(\forall k\ge 0\) \(\tilde{u}_k\in C([0,1])\).
Proof
From the formulas (21.12)–(21.14) it follows that the functionals \(J_0 : L^2(0,1)\mapsto \mathbf {R}\), \(J_k : L^2(0,1)\times L^2(0,1)\mapsto \mathbf {R}\) are strictly convex, continuous and coercive, which means under [1] that the problems (21.10), (21.11) have the unique solution in the spaces \(L^2(0,1)\) and \(L^2(0,1)\times L^2(0,1)\) correspondingly.
Equating to zero Frechet derivatives of \(J_0, J_k\), we obtain the following Fredholm integral equations:
Since \(\max \limits _{(p,s)\in \prod }|G_0(\alpha ,p)G_0(\alpha ,s)|\le e^{-2}<1\), then the Eq. (21.15) has the unique solution \(\tilde{u}_0\in C([0,1])\).
Put
Then from the Eqs. (21.16), (21.17) we have that vector
satisfies the equation
Under estimates from Lemma 21.1 we obtain
Then \(\forall k\ge 1\) the equation (21.18) has the unique solution
herewith \(\forall r\in [0,1]\)
The theorem is proved.
From the estimates (21.19) it follows that the series \(\sum \limits _{n=0}^\infty \tilde{u}_n(r)\varphi _n(\theta )\) converges uniformly on \(\bar{Q}\) and it defines the function \(\tilde{u}(r,\theta )\in C(\bar{Q})\) by the formula (21.4).
Theorem 21.2
Series
defines the function \(\tilde{y}(r,\theta )\in C(\bar{Q})\bigcap C^2(Q)\) by the formula (21.5), where \(\{\tilde{y}_n\}_{n=0}^\infty \) are the solutions of the system (21.6)–(21.8) with controls \(\{\tilde{u}_n\}_{n=1}^\infty \).
Proof
By the formulas (21.12)–(21.14) desired series has the form
The functions \(r^{2n}\sin 2n\theta \) and \(r^{2n}(\ln r\cdot \sin 2n\theta + \theta \cos 2n\theta )\) are harmonic, \(p\in C^1([0,\pi ])\), \(p(0)=0\), so, from [6] the first series in 21.20) is the function from the class \(C(\bar{Q})\bigcap C^2(Q)\).
From Lemma 21.1 and the estimates (21.19) we have under Weierstrass theorem that \(\tilde{y}\in C(\bar{Q})\).
On \(\forall [a,b]\times [c,d]\subset (0,1)\times (0,\pi )\) it remains to investigate the uniform convergence of the series from the first and second-order derivatives on \(r,\theta \) of functions
From the estimates (21.19) we obtain that the series from derivatives \(\frac{\partial }{\partial \theta }\), \(\frac{\partial ^2}{\partial \theta ^2}\) converge on \(\bar{Q}\) uniformly under Weierstrass theorem.
For \(\forall r\in [a,b]\), \(\forall n>1\)
(summands which do not contain integrals are mutually canceled)
Since \(\int \limits _0^r s^{2n+1}ds=\frac{r^{2n+1}}{2n+2}\),
then \(\exists C_1>0\) such that
so, the series \(\sum \limits _{n=2}^\infty \frac{\partial }{\partial r}B_n(r,\theta )\), \(\sum \limits _{n=2}^\infty \frac{\partial }{\partial r} C_n(r,\theta )\), \(\sum \limits _{n=2}^\infty \frac{\partial ^2}{\partial r \partial \theta } B_n(r,\theta )\), \(\sum \limits _{n=2}^\infty \frac{\partial ^2}{\partial r \partial \theta }C_n(r,\theta )\) converge uniformly on \([a,b]\times [c,d]\).
From the same estimates \(|b''_n(r)|\le C_2\cdot \frac{\alpha ^{2n-1}}{n}(|p_{2n-1}| + |p_{2n}|)\) and, thereby the series \(\sum \limits _{n=2}^\infty \frac{\partial ^2}{\partial r^2} B_n(r,\theta )\), \(\sum \limits _{n=2}^\infty \frac{\partial ^2}{\partial r^2} C_n(r,\theta )\) converge uniformly on \([a,b]\times [c,d]\).
For the function \(d_n(r)\) we have \(\forall r\in [a,b]\):
Since
then \(\exists C_2 >0\) such that
so, the series \(\sum \limits _{n=2}^\infty \frac{\partial }{\partial r} D_n(r,\theta )\), \(\sum \limits _{n=2}^\infty \frac{\partial ^2}{\partial r \partial \theta }D_n(r,\theta )\) converge uniformly on \([a,b]\times [c,d]\).
It is easy to see that \(\exists C_3 >0\) such that
Hence, the series \(\sum \limits _{n=2}^\infty \frac{\partial ^2}{\partial r^2} D_n(r,\theta )\) converges uniformly on \([a,b]\times [c,d]\).
Thereby, \(\tilde{y}\in C(\bar{Q})\bigcap C^2(Q)\) and Theorem is proved.
Remark 21.1
If \(u(r,\theta )\in C(\bar{Q})\) and for some constant \(C>0\) \(\forall n\ge 1\) \(|u_n(r)|\le \frac{C}{n^2}\), then the control \(u\) is admissible in the problem (21.1)–(21.2), that is the corresponding function \(y(r,\theta )\) from (21.5) defines classical solution of (21.1).
4 Conclusions
In this paper we proved a solvability of the optimal control problem on the classical solutions of elliptic boundary value problem in a circular sector with equality of flows on radiuses and equality of the solution on the one from radiuses to zero in distributed control class for quadratic cost functional.
References
Lions, J.-L.: Optimal Problem in PDE Systems. Mir, Moscow (1972)
Egorov, A.I.: Optimal Control in Heat and Diffusion Processes. Nauka, Moscow (1978)
Belozerov, V.E., Kapustyan, V.E.: Geometrical Methods of Modal Control. Naukova Dumka, Kyiv (1999)
Kapustyan, V.E.: Optimal stabilization of the solutions of a parabolic boundary-value problem using bounded lumped control. J. Autom. Inf. Sci. 31(12), 45–52 (1999)
Kapustyan, E.A., Nakonechny, A.G.: Optimal bounded control synthesis for a parabolic boundary-value problem with fast oscillatory coefficients. J. Autom. Inf. Sci. 31(12), 33–44 (1999)
Moiseev, E.I., Ambarzumyan, V.E.: About resolvability of non-local boundary-value problem with equality of fluxes. Differ. Equ. 46(5), 718–725 (2010)
Ionkin, N.I.: Solution of boundary-value problem from heat theory with non-classical boundary conditions. Differ. Equ. 13(2), 294–304 (1977)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2014 Springer International Publishing Switzerland
About this chapter
Cite this chapter
Kapustyan, V.O., Kapustian, O.A., Mazur, O.K. (2014). Distributed Optimal Control in One Non-Self-Adjoint Boundary Value Problem. In: Zgurovsky, M., Sadovnichiy, V. (eds) Continuous and Distributed Systems. Solid Mechanics and Its Applications, vol 211. Springer, Cham. https://doi.org/10.1007/978-3-319-03146-0_21
Download citation
DOI: https://doi.org/10.1007/978-3-319-03146-0_21
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-03145-3
Online ISBN: 978-3-319-03146-0
eBook Packages: EngineeringEngineering (R0)