Abstract
This article discusses the problem of constructing an external estimate of the limit set of controllability for a linear discrete system with convex control constraints. We have proposed a decomposition method that allows us to reduce the problem for the initial system to subsystems of smaller dimension by switching to the normal Jordan basis of the matrix of the system. The statement about the structure of the reference hyperplane to the limit set of controllability is formulated and proved. A method for constructing an external estimate of the limit set of controllability with an arbitrary order of accuracy in the sense of the Hausdorff distance is proposed based on the principle of contraction mappings. The paper provides examples.
REFERENCES
Sirotin, A.N. and Formal’skii, A.M., Reachability and Controllability of Discrete-Time Systems under Control Actions Bounded in Magnitude and Norm, Autom. Remote Control, 2003, vol. 64, no. 12, pp. 1844–1857.
Fisher, M.E. and Gayek, J.E., Estimating Reachable Sets for Two-Dimensional Linear Discrete Systems, J. Optim. Theory Appl., 1988, vol. 56, no. 1, pp. 67–88.
Desoer, C.A. and Wing, J., The Minimal Time Regulator Problem for Linear Sampled-Data Systems: General Theory, J. Franklin Inst., 1961, vol. 272, no. 3, pp. 208–228.
Hamza, M.H. and Rasmy, M.E., A Simple Method for Determining the Reahable Set for Linear Discrete Systems, IEEE Trans. on Automat. Control, 1971, vol. 16, pp. 281–282.
Corradini, M.L., Cristofaro, A., Giannoni, F., and Orlando, G., Estimation of the Null Controllable Region: Discrete-Time Plants, Control Systems with Saturating Inputs, Lecture Notes in Control and Information Sciences, Springer, 2012, vol. 424, pp. 33–52.
Hu, T., Miller, D.E., and Qiu, L., Null Controllable Region of LTI Discrete-Time Systems with Input Saturation, Automatica, 2002, vol. 38, no. 11, pp. 2009–2013.
Kalman, R., On the General Theory of Control Systems, Proceedings of the I Intern. Congr. IFAC, 1961, vol. 2, pp. 521–547.
Boltyansky, V.G., Matematicheskie metody optimal’nogo upravleniya (Mathematical Methods of Optimal Control), Moscow: Nauka, 1969.
Pontryagin, L.S., Boltyansky, V.G., Gamkrelidze, R.V., and Mishchenko, B.F., Matematicheskaya teoriya optimal’nykh protsessov (Mathematical Theory of Optimal Processes), Moscow: Nauka, 1969.
Moiseev, N.N., Elementy teorii optimal’nykh sistem (Elements of the Theory of Optimal Systems), Moscow: Nauka, 1975.
Agrachev, A.A. and Sachkov, Yu.L., Geometricheskaya teoriya upravleniya (Geometric Control Theory), Moscow: Nauka, 2005.
Yevtushenko, Yu.G., Metody resheniya ekstremal’nykh zadach i ikh prilozheniya v sistemakh optimizatsii (Methods for Solving Extremal Problems and Their Applications in Optimization Systems), Moscow: Nauka, 1982.
Holtzman, J.M. and Halkin, H., Directional Convexity and the Maximum Principle for Discrete Systems, J. SIAM Control, 1966, vol. 4, no. 2, pp. 263–275.
Propoi, A.I., Elementy teorii optimil’nykh diskretnykh protsessov (Elements of the Theory of Optimal Discrete Processes), Moscow: Nauka, 1973.
Bellman, R., Dinamicheskoe programmirovanie (Dynamic Programming), Moscow: Inostrannaya Literatura, 1960.
Kurzhanskiy, A.F. and Varaiya, P., Theory and Computational Techniques for Analysis of Discrete-Time Control Systems with Disturbancens, Optim. Method Software, 2011, vol. 26, nos. 4–5, pp. 719–746.
Boltyansky, V.G., Optimal’noe upravlenie diskretnymi sistemami (Optimal Control of Discrete Systems), Moscow: Nauka, 1973.
Lin, W.-S., Time-Optimal Control Strategy for Saturating Linear Discrete Systems, Int. J. Control, 1986, vol. 43, no. 5, pp. 1343–1351.
Moroz, A.I., Synthesis of Time-Optimal Control for Linear Discrete Objects of the Third Order, Autom. Remote Control, 1965, vol. 25, no. 9, pp. 193–206.
Ibragimov, D.N. and Sirotin, A.N., On the Problem of Optimal Speed for the Discrete Linear System with Bounded Scalar Control on the Basis of 0-Controllability Sets, Autom. Remote Control, 2015, vol. 76, no. 9, pp. 1517–1540.
Ibragimov, D.N., Speed-Optimal Control of the Motion of a Balloon, Internet Magazine, Tr. MAI, 2015, no. 83. http://trudymai.ru/published.php
Ibragimov, D.N., Approximation of the Set of Admissible Controls in the Time-Optimal Problem by a Linear Discrete System, Internet Magazine, Tr. MAI, 2016, no. 87. http://trudymai.ru/published.php.
Ibragimov, D.N. and Portseva, E.Yu., Outer Approximation Algorithm for a Convex Set of Admissible Controls for a Discrete System with Bounded Control, Modeling and Data Analysis, 2019, no. 2, pp. 83–98.
Rockafellar, R., Vypuklyi analiz (Convex Analysis), Moscow: Mir, 1973.
Horn, R. and Johnson, C., Matrichnyi analiz (Matrix Analysis), Moscow: Mir, 1989.
Ibragimov, D.N. and Sirotin, A.N., On the Problem of Operation Speed for the Class of Linear Infinite-Dimensional Discrete-Time Systems with Bounded Control, Autom. Remote Control, 2017, vol. 78, no. 10, pp. 1731–1756.
Kronover, R.M., Fraktaly i khaos v dinamicheskikh sistemakh (Fractals and Chaos in Dynamical Systems), Moscow: Postmarket, 2000.
Kolmogorov, A.N. and Fomin, S.V., Elementy teorii funktsii i funktsional’nogo analiza (Elements of the Theory of Functions and Functional Analysis), Moscow: Fizmatlit, 2012.
Author information
Authors and Affiliations
Corresponding authors
Additional information
This paper was recommended for publication by M.V. Khlebnikov, a member of the Editorial Board
APPENDIX
APPENDIX
Proof of Lemma 1. Denote the initial states of the system (A1, \({{\mathcal{U}}_{1}}\)) and (A2, \({{\mathcal{U}}_{2}}\)) by x0,е1 ∈ \({{\mathbb{R}}^{{{{n}_{1}}}}}\) and x0, 2 ∈ \({{\mathbb{R}}^{{{{n}_{2}}}}}\), respectively. Then x0 = \(\left( \begin{gathered} {{x}_{{0,1}}} \\ {{x}_{{0,2}}} \\ \end{gathered} \right)\) is the initial state of the system (A, \(\mathcal{U}\)).
By (1) it is true that for all N ∈ \(\mathbb{N}\)
Then x(N) = 0 if and only if there exist u1(0), …, u1(N – 1) ∈ \({{\mathcal{U}}_{1}}\) and u2(0), …, u2(N – 1) ∈ \({{\mathcal{U}}_{2}}\) such that x1(N) = 0, x2(N) = 0. The equality data is, by virtue of (2), equivalent to including x0, 1 ∈ \({{\mathcal{X}}_{1}}\)(N), x0, 2 ∈ \({{\mathcal{X}}_{2}}\)(N). Hence,
Let x0 ∈ \({{\mathcal{X}}_{\infty }}\). Then according to (3) there exists \(\tilde {N} \in \mathbb{N}\) such that
Then \({{\mathcal{X}}_{\infty }}\) ⊂ \({{\mathcal{X}}_{{1,\infty }}}\) × \({{\mathcal{X}}_{{2,\infty }}}\).
Let x0 ∈ \({{\mathcal{X}}_{{1,\infty }}}\) × \({{\mathcal{X}}_{{2,\infty }}}\). Therefore, there are \({{\tilde {N}}_{1}}\), \({{\tilde {N}}_{2}} \in \mathbb{N}\) such that x0 ∈ \({{\mathcal{X}}_{1}}({{\tilde {N}}_{1}})\) × \({{\mathcal{X}}_{2}}({{\tilde {N}}_{2}})\) ⊂ \({{\mathcal{X}}_{1}}(\tilde {N})\) × \({{\mathcal{X}}_{2}}(\tilde {N})\), where \(\tilde {N}\) = max{\({{\tilde {N}}_{1}}\), \({{\tilde {N}}_{2}}\)}. Then, by point 1 of Lemma 1
It follows that \({{\mathcal{X}}_{{1,\infty }}}\) × \({{\mathcal{X}}_{{2,\infty }}} \subset {{\mathcal{X}}_{\infty }}\).
Finally, we get that \({{\mathcal{X}}_{\infty }}\) = \({{\mathcal{X}}_{{1,\infty }}}\) × \({{\mathcal{X}}_{{2,\infty }}}\). Lemma 1 is proved.
Proof of Lemma 2. Let \(\{ y(k)\} _{{k = 0}}^{N}\) be the trajectory of the system (\({{S}^{{ - 1}}}AS\), \({{S}^{{ - 1}}}\mathcal{U}\)), i.e., y(N) according to (1) for the initial state y0 ∈ \({{\mathbb{R}}^{n}}\) admits the following representation:
where \({v}\)(0), …, \({v}\)(N – 1) ∈ \({{S}^{{ - 1}}}\mathcal{U}\).
By virtue of (2) y0 ∈ \(\mathcal{Y}(N)\) if and only if y(N) = 0, i.e.,
which, by virtue of (2), is equivalent to including Sy0 ∈ \(\mathcal{X}(N)\), since by construction \(S{v}\)(0), …, \(S{v}\)(N – 1) ∈ \(\mathcal{U}\). Whence follows the equality \(\mathcal{X}(N)\) = \(S\mathcal{Y}(N)\).
Let x0 ∈ \({{\mathcal{X}}_{\infty }}\). By (3), there exists \(\tilde {N} \in \mathbb{N} \cup \{ 0\} \) such that x0 ∈ \(\tilde {\mathcal{X}}(\tilde {N})\), which is equivalent to including x0 ∈ \(S\mathcal{Y}(\tilde {N})\) according to point 1 of Lemma 2. Hence S–1x0 ∈ \(\mathcal{Y}(\tilde {N})\). Then S–1x0 ∈ \(\bigcup\nolimits_{N = 0}^\infty {\mathcal{Y}(N)} \) = \({{\mathcal{Y}}_{\infty }}\), i.e., x0 ∈ \(S{{\mathcal{Y}}_{\infty }}\). Then \({{\mathcal{X}}_{\infty }}\) ⊂ \(S{{\mathcal{Y}}_{\infty }}\).
Let x0 ∈ \(S{{\mathcal{Y}}_{\infty }}\), then S–1x0 ∈ \({{\mathcal{Y}}_{\infty }}\). By (3) there exists \(\tilde {N} \in \mathbb{N} \cup \{ 0\} \) such that S–1x0 ∈ \(\mathcal{Y}(\tilde {N})\). Then x0 ∈ \(S\mathcal{Y}(\tilde {N})\), which is equivalent to including x0 ⊂ \(\mathcal{X}(\tilde {N})\) by point 1 of Lemma 2. According to relations (3), the inclusion x0 ∈ \({{\mathcal{X}}_{\infty }}\) is also true. Then \(S{{\mathcal{Y}}_{\infty }}\) ⊂ \({{\mathcal{X}}_{\infty }}\).
Finally, we get that \(S{{\mathcal{Y}}_{\infty }}\) = \({{\mathcal{X}}_{\infty }}\). Lemma 2 is proved.
Proof of Lemma 3. Let x0 ∈ \({{\mathbb{R}}^{{{{n}_{1}}}}} \times {{\mathcal{X}}_{{2,\infty }}}\). Then x0 = \(\left( \begin{gathered} {{x}_{{0,1}}} \\ {{x}_{{0,2}}} \\ \end{gathered} \right)\), where x0, 1 ∈ \({{\mathbb{R}}^{{{{n}_{1}}}}}\), x0, 2 ∈ \({{\mathcal{X}}_{{2,\infty }}}\), whence according to (3) there exists \(\tilde {N} \in \mathbb{N} \cup \{ 0\} \) such that x0, 2 ∈ \({{\mathcal{X}}_{2}}(\tilde {N})\), which according to (2) is equivalent to the existence of \(u_{2}^{*}\)(0), …, \(u_{2}^{*}\)(N – 1) ∈ \({{\mathcal{U}}_{2}}\) such that x2(\(\tilde {N}\)) = 0. Then for the system (A, \(\mathcal{U}\)) there are u(0), …, u(\(\tilde {N}\) – 1) ∈ \(\mathcal{U}\) such that u(k) = \(\left( \begin{gathered} {{u}_{1}}(k) \\ u_{2}^{*}(k) \\ \end{gathered} \right)\), k = \(\overline {0,\tilde {N} - 1} \). By (1), x(\(\tilde {N}\)) has the representation
According to Lemma 1, it suffices to show that there exists \({{\mathcal{U}}_{1}} \subset {{\mathbb{R}}^{{{{n}_{1}}}}}\) such that \({{\mathcal{U}}_{1}}\) × {0} ⊂ \(\mathcal{U}\) and \({{\mathcal{X}}_{{1,\infty }}}\) = \({{\mathbb{R}}^{{{{n}_{1}}}}}\), where \({{\mathcal{X}}_{{1,\infty }}}\) is the limit set of 0-controllability of the system (A1, \({{\mathcal{U}}_{1}}\)).
Denote by S ∈ \({{\mathbb{R}}^{{{{n}_{1}} \times {{n}_{1}}}}}\) the transition matrix to the normal Jordan basis of the matrix A1. Since 0 ∈ int\(\mathcal{U}\), there exists umax > 0 such that S[–umax; \({{u}_{{\max }}}{{]}^{{{{n}_{1}}}}}\) × {0} ⊂ \(\mathcal{U}\). Moreover, due to the non-degeneracy of the matrix S and Lemma 2, the equality \({{\mathcal{X}}_{{1,\infty }}}\) = \({{\mathbb{R}}^{{{{n}_{1}}}}}\) is true for the case \({{\mathcal{U}}_{1}}\) = S[–umax; \({{u}_{{\max }}}{{]}^{{{{n}_{1}}}}}\) if and only if \({{S}^{{ - 1}}}{{\mathcal{X}}_{{1,\infty }}}\) = \({{\mathbb{R}}^{{{{n}_{1}}}}}\), where \({{S}^{{ - 1}}}{{\mathcal{X}}_{{1,\infty }}}\) is the limit set of 0-controllability of the system (S–1A1S, [–umax; \({{u}_{{\max }}}{{]}^{{{{n}_{1}}}}}\)). Moreover, according to the normal Jordan form theorem [25], the equality
where the Jordan cells Ji corresponding to the real eigenvalues λi ∈ \(\mathbb{R}\) of the matrix A1 have the form
and the Jordan cells Ji corresponding to the complex eigenvalues λi ∈ \(\mathbb{C}\) of the matrix A1 have the form
where ri = |λi|, φi = arg(λi).
Hence, by virtue of Lemma 1, it suffices to show that for |λi| \(\leqslant \) 1, the limit sets of null-controllability of the system are (Ji, [–umax; \({{u}_{{\max }}}{{]}^{{{{m}_{i}}}}}\)) for the case of (A.1) and the systems (Ji, [–umax; \({{u}_{{\max }}}{{]}^{{2{{m}_{i}}}}}\)) for the case of (A.2) coincide with \({{\mathbb{R}}^{{{{m}_{i}}}}}\) and \({{\mathbb{R}}^{{2{{m}_{i}}}}}\) correspondingly, for all i = \(\overline {1,{{{\tilde {n}}}_{1}}} \).
Let J ∈ \({{\mathbb{R}}^{{m \times m}}}\) satisfy (A.1). Then for all N \( \geqslant \) m the following relations hold
where we denote the number of combinations of N choose m by \(C_{N}^{m}\):
Denote by {y(k), \({v}\)(k – 1), \({{y}_{0}}\} _{{k = 1}}^{N}\) the process of controlling the system (J, [–umax; umax]m). Hence
If we denote z0 = JNy0, then by (A.3) it is right for each ith coordinate of z0, that
Let us assume, that |λ| < 1. Then for all N \( \geqslant \) 2m the following relations hold
Then there exists \(\tilde {N} \in \mathbb{N}\) such, that for all i = \(\overline {1,m} \)
Let us take \({v}\)(0) = … = \({v}(\tilde {N} - 2)\) = 0 and \({v}(\tilde {N} - 1)\) = –z0 ∈ [–umax; umax]m. Then we obtain y(\(\tilde {N}\)) = 0, i.e., y0 ∈ \(\mathcal{Y}(\tilde {N})\). Therefore, by choosing arbitrary y0 ∈ \({{\mathbb{R}}^{m}}\) and Eq. (3) we obtain the result \({{\mathcal{Y}}_{\infty }}\) = \({{\mathbb{R}}^{m}}\).
Let us assume, that |λ| = 1. Then by (A.3) for some Nm ∈ \(\mathbb{N}\) and mth coordinate of y(Nm) it is right that
Here we choose Nm ∈ \(\mathbb{N}\), requiring |ym(0)| \(\leqslant \) Nmumax. Then we take \({v}\)(0), …, \({v}\)(Nm – 1) ∈ [–umax; umax]m in accordance with the following condition:
We obtain
Let us assume that for some N ∈ \(\mathbb{N}\) and i = \(\overline {1,m - 1} \) it is right that ym(N) = … = \({{y}_{{m - i + 1}}}(N)\) = 0. Then if \({{{v}}_{m}}\)(N + Nm – i – k – 1) = … = \({{{v}}_{{m - i + 1}}}\)(N + Nm – i – k – 1) = 0, k = \(\overline {0,{{N}_{{m - i}}} - 1} \) is right, then by (A.3) the following result is straightforward
where \({{N}_{{m - i}}} \in \mathbb{N}\) is chosen from the condition |ym – i(N)| \(\leqslant \) \({{N}_{{m - i}}}{{u}_{{\max }}}\). Then we determine \({v}\)(N), …, \({v}\)(N + \({{N}_{{m - i}}}\) – 1) ∈ [–umax; umax]m in order that
We obtain
Then there exists N ∈ \(\mathbb{N}\) such, that y(N) = 0, i.e., y0 ∈ \(\mathcal{Y}(N)\) by the method of mathematical induction. Therefore, we obtain the result \({{\mathcal{Y}}_{\infty }}\) = \({{\mathbb{R}}^{m}}\) by choosing arbitrary y0 ∈ \({{\mathbb{R}}^{m}}\) and Eq. (3).
Let J ∈ \({{\mathbb{R}}^{{2m \times 2m}}}\) satisfy the case (A.2). Then for all N \( \geqslant \) m the following relations hold
Denote the process of controlling the system (J, [–umax; umax]2m) by {y(k), \({v}\)(k – 1), \({{y}_{0}}\} _{{k = 1}}^{N}\). Hence
If we denote z0 = JNy0, then by (A.4) for each ith two-dimensional subvector z0 it is right, that
where z0 = \((z_{{0,1}}^{{\text{T}}}\), …, \(z_{{0,m}}^{{\text{T}}}{{)}^{{\text{T}}}}\), y0 = \((y_{{0,1}}^{{\text{T}}}\), …, \(y_{{0,m}}^{{\text{T}}}{{)}^{{\text{T}}}}\).
Let us assume, that r < 1. Then for all N \( \geqslant \) 2m the following relations hold
Then there exists \(\tilde {N} \in \mathbb{N}\) such, that for all i = \(\overline {1,m} \) it is right that
Let us determine \({v}\)(0) = … = \({v}\)(\(\tilde {N}\) – 2) = 0 and \({v}\)(\(\tilde {N}\) – 1) = –z0 ∈ [–umax; umax]2m. We obtain the result y(\(\tilde {N}\)) = 0, i.e., y0 ∈ \(\mathcal{Y}(\tilde {N})\). Therefore, we obtain the result \({{\mathcal{Y}}_{\infty }}\) = \({{\mathbb{R}}^{{2m}}}\) by choosing arbitrary y0 ∈ \({{\mathbb{R}}^{{2m}}}\) and Eq. (3).
Let us assume, that r = 1. Then by (A.4) for some Nm ∈ \(\mathbb{N}\) and mth two-dimensional subvector y(Nm) it is true, that
Let us take Nm ∈ \(\mathbb{N}\) which hold the inequality ||ym(0)|| \(\leqslant \) Nmumax. Then we choose \({v}\)(0), …, \({v}\)(Nm – 1) ∈ [‒umax; umax]2m in accordance with the equality
We obtain
Let us assume that for some N ∈ \(\mathbb{N}\) and i = \(\overline {1,m - 1} \) the relation ym(N) = … = \({{y}_{{m - i + 1}}}(N)\) = 0 is correct. Then if \({{{v}}_{m}}\)(N + \({{N}_{{m - i}}}\) – k – 1) = … = \({{{v}}_{{m - i + 1}}}\)(N + \({{N}_{{m - i}}}\) – k – 1) = 0, k = \(\overline {0,{{N}_{{m - i}}} - 1} \), then according to the (A.4)
where \({{N}_{{m - i}}}\) ∈ \(\mathbb{N}\) is selected from the condition ||\({{y}_{{m - i}}}\)(N)|| \(\leqslant \) \({{N}_{{m - i}}}{{u}_{{\max }}}\). Then we define \({v}\)(N), …, \({v}\)(N + \({{N}_{{m - i}}}\) – 1) ∈ [–umax; umax]2m in order that
We obtain
Then there exists N ∈ \(\mathbb{N}\) such, that y(N) = 0, i.e., y0 ∈ \(\mathcal{Y}(N)\) by the method of mathematical induction. Therefore, we obtain the result \({{\mathcal{Y}}_{\infty }}\) = \({{\mathbb{R}}^{{2m}}}\) by choosing arbitrary y0 ∈ \({{\mathbb{R}}^{{2m}}}\) and Eq. (3).
Hence Lemma 3 is proved.
Proof of Theorem 1. Let x0 ∈ \({{\mathcal{X}}_{\infty }}\), which by (3) is equivalent to the existence of N ∈ \(\mathbb{N}\) ∪ {0} such, that x0 ∈ \(\mathcal{X}\)(N). By (2) there exist u(0), …, u(N – 1) ∈ \(\mathcal{U}\) such, that x(N) = 0. Based on (1), for all h ∈ ∂B1(0), ε > 0 the following relations hold
where u(N) ∈ \(\mathcal{U}\). Considering 0 ∈ int \(\mathcal{U}\), there exists δ > 0 such, that Oδ(0) ⊂ \(\mathcal{U}\), and ε > 0 such, that \(\varepsilon {{A}^{{N + 1}}}{{B}_{1}}(0)\) ⊂ Oδ(0). Let us take
Then \(\tilde {x}\)(N + 1) = 0, i.e., for all h ∈ B1(0) it is true that x0 + εh ∈ \(\mathcal{X}\)(N + 1). As a result, Bε(x0) ⊂ \(\mathcal{X}\)(N + 1) ⊂ \({{\mathcal{X}}_{\infty }}\), i.e., x0 ∈ int \({{\mathcal{X}}_{\infty }}\). Hence, \({{\mathcal{X}}_{\infty }}\), is open.
Let x0, 1, x0, 2 ∈ \({{\mathcal{X}}_{\infty }}\), α ∈ [0; 1]. Then there exists N ∈ \(\mathbb{N}\) ∪ {0} such, that x0, 1, x0, 2 ∈ \(\mathcal{X}\)(N), i.e., there exist u1(0), u1(1), …, u1(N – 1), u2(0), u2(1), …, u2(N – 1) ∈ \(\mathcal{U}\) such, that x1(N) = 0, x2(N) = 0. According to (1) it is true that
According to the convexity of \(\mathcal{U}\) the relations \({v}\)(N – k – 1) = αu1(N – k – 1) + (1 – α)u2(N – k – 1) ∈ \(\mathcal{U}\), k = \(\overline {0,N - 1} \) are correct. Then αx0, 1 + (1 – α)x0, 2 ∈ \(\mathcal{X}\)(N) ⊂ \({{\mathcal{X}}_{\infty }}\), from which it follows that \({{\mathcal{X}}_{\infty }}\) is convex.
Theorem 1 is proved.
Proof of Lemma 4. Let x0 ∈ \({{\mathcal{X}}_{\infty }}\). Then by (3) there exists N ∈ \(\mathbb{N}\) ∪ {0} such, that x0 ∈ \(\mathcal{X}\)(N). According to (2) there exist u(0), u(1), …, u(N – 1) ∈ \(\mathcal{U}\) such, that x(N) = 0. Based on (1), it is right that
Let p belong to \({{\mathbb{R}}^{n}}\)\{0}. Then
Since 0 ∈ \(\mathcal{U}\), then for all k ∈ \(\mathbb{N}\)
Then
i.e., x0 ∈ \({{\mathcal{H}}_{p}}\). It follows that \({{\mathcal{X}}_{\infty }} \subset {{\mathcal{H}}_{p}}\).
Let us consider the following quantity for some p ∈ \({{\mathbb{R}}^{n}}\)\{0}:
Then x* ∈ \(\partial {{\mathcal{H}}_{p}}\).
Since all eigenvalues of the matrix A are strictly greater than 1 in absolute value, then the series \(\sum\nolimits_{k = 1}^\infty {{{A}^{{ - k}}}u_{k}^{*}} \) converges. Then xN = \( - \sum\limits_{k = 1}^N {{{A}^{{ - k}}}u_{k}^{*}} \xrightarrow{{N \to \infty }}x\text{*}\). Demonstrate that xN ∈ \(\mathcal{X}\)(N) ⊂ \({{\mathcal{X}}_{\infty }}\). Let us take u(k) = \(u_{{k + 1}}^{*}\) ∈ \(\mathcal{U}\). Then
Then xN ∈ \(\mathcal{X}\)(N) ⊂ \({{\mathcal{X}}_{\infty }}\) for all N ∈ \(\mathbb{N}\). It follows that x* = \(\mathop {\lim }\limits_{N \to \infty } {{x}_{N}} \in \overline {{{\mathcal{X}}_{\infty }}} \).
Lemma 4 is proved.
Corollary 2. Let all eigenvalues of the matrix A ∈ \({{\mathbb{R}}^{{n \times n}}}\) be strictly greater than 1 in absolute value, \({{\mathcal{X}}_{\infty }}\) is defined by (3).
Then for all p ∈ \({{\mathbb{R}}^{n}}\)\{0} it is right that:
(1) \({{\mathcal{X}}_{\infty }} \subset {{\mathcal{H}}_{{ - p}}} = \left\{ {x \in {{\mathbb{R}}^{n}}{\kern 1pt} {\kern 1pt} :\;(p,x)\; \geqslant \;\sum\limits_{k = 1}^\infty {\mathop {\min }\limits_{{{u}_{k}} \in \mathcal{U}} \left( { - {{{\left( {{{A}^{{ - k}}}} \right)}}^{{\text{T}}}}p,\,\,{{u}_{k}}} \right)} } \right\}\);
(2) \(x\text{*} = - \sum\limits_{k = 1}^\infty {{{A}^{{ - k}}}u_{k}^{*}} \in \overline {{{\mathcal{X}}_{\infty }}} \cap \partial {{\mathcal{H}}_{{ - p}}}\), where
Proof of Corollary 2. For proving that it is sufficient to consider the conditions of Lemma 4 for the vector –p.
Proof of Lemma 5. Let p = (0 … 0 1 0 … 0)T ∈ \({{\mathbb{R}}^{n}}\), where 1 corresponds to the ith coordinate of the vector p. Then for arbitrary k ∈ \(\mathbb{N}\)
Let us consider the case λ > 1. Since ui ∈ [ui, min; ui, max], then for all k ∈ \(\mathbb{N}\) the following inequalities hold
Additionally, the following equalities hold
By Lemma 4 and Corollary 2, these equalities imply, that for all x ∈ \({{\mathcal{X}}_{\infty }}\) inequalities hold
Since \({{\mathcal{X}}_{\infty }}\) is open according to Theorem 1, these inequalities strictly hold, i.e.,
Let us consider the case λ < –1. For all k ∈ \(\mathbb{N}\) it is true that
Then
Then by Lemma 4 for all x ∈ \({{\mathcal{X}}_{\infty }}\) it is right, that
Similarly
Then by Corollary 2 for all x ∈ \({{\mathcal{X}}_{\infty }}\) it is right that
Since \({{\mathcal{X}}_{\infty }}\) is open according to Theorem 1
Lemma 5 is proved.
Proof of Lemma 6. Let p = (0 0 … \({{\tilde {p}}^{{\text{T}}}}\) … 0)T ∈ \({{\mathbb{R}}^{{2n}}}\), \(\tilde {p}\) = (p1 p2)T ∈ \({{\mathbb{R}}^{2}}\), \(p_{1}^{2}\) + \(p_{2}^{2}\) = 1, where \(\tilde {p}\) corresponds to the (2i – 1)th and 2ith coordinates of the vector p. Then for arbitrary k ∈ \(\mathbb{N}\) it is true that
Let ui ∈ \({{\mathbb{R}}^{2}}\), i = \(\overline {1,n} \), u = (u1T, ..., unT)T ∈ \({{\mathbb{R}}^{{2n}}}\). Then
Then by Lemma 4 for arbitrary x ∈ \({{\mathcal{X}}_{\infty }}\) it is right that
Since \({{\mathcal{X}}_{\infty }}\) according to Theorem 1 is open, then
Lemma 6 is proved.
Proof of Theorem 2. Let us consider for some B ∈ \({{\mathbb{R}}^{{n \times n}}}\) and \(\mathcal{C} \in {{\mathbb{K}}_{n}}\) the mapping
Let us demonstrate, that if B : \({{\mathbb{R}}^{n}} \to {{\mathbb{R}}^{n}}\) is a contraction mapping with the compression ratio β ∈ [0; 1), then \(\tilde {T}\) : \({{\mathbb{K}}_{n}} \to {{\mathbb{K}}_{n}}\) is also a contraction mapping.
Then \(\tilde {T}\) is a contraction mapping with the compression ratio β.
According to (4) the following equality holds
where B = A–M, \(\mathcal{C}\) = \(\sum\nolimits_{k = 1}^M {( - {{A}^{{ - k}}}\mathcal{U})} \).
Since all eigenvalues of the matrix A are strictly greater than 1 in absolute value, then all eigenvalues of the matrix A–1 are strictly less than 1 in absolute value. Then according to the [24, Theorem 5.6.12] ||A–k|| \(\xrightarrow{{k \to \infty }}\) 0. Then by definition of the limit for α ∈ [0; 1) there exists M ∈ \(\mathbb{N}\) such, that ||A–M|| < α. As the following inequality
is true A–M is a contraction mapping with the compression ratio α ∈ [0; 1). Then TM: \({{\mathbb{K}}_{n}} \to {{\mathbb{K}}_{n}}\) is also a contraction mapping with the compression ratio α.
By virtue of Lemma 7 for all N ∈ \(\mathbb{N}\) it is right, that \(\mathcal{X}\)(N) ⊂ \(\mathcal{X}\)(N + 1), in addition, \(\mathcal{X}\)(N) is a compact set. Then according to the ([27], Corollary A.3.4)
In contrast, by virtue of ([27], Theorem A.3.9) the metric space (\({{\mathbb{K}}_{n}}\), ρH) is complete. Then the contraction mapping \(\tilde {T}\) has a unique fixed point \(\mathcal{X}\text{*} \in {{\mathbb{K}}_{n}}\), which can be computed by fixed point iteration method:
where \(\mathcal{X} \in {{\mathbb{K}}_{n}}\) is arbitrary. Let us take \(\mathcal{X}\) = {0}. Then by virtue of Lemma 7
According to the uniqueness of the limit point and formulae (A.5) and (A.6)
The error in the fixed point iteration method can be estimated by the following formula [28]:
Theorem 2 is proved.
Proof of Theorem 3. By virtue of the point 3 of Theorem 2
Then according to the definition of the Hausdorff distance
where
Theorem 3 is proved.
Rights and permissions
About this article
Cite this article
Berendakova, A.V., Ibragimov, D.N. About the Method for Constructing External Estimates of the Limit Controllability Set for the Linear Discrete-Time System with Bounded Control. Autom Remote Control 84, 83–104 (2023). https://doi.org/10.1134/S0005117923020030
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1134/S0005117923020030