Table of Contents Author Guidelines Submit a Manuscript
Journal of Control Science and Engineering
Volume 2009, Article ID 202094, 5 pages
http://dx.doi.org/10.1155/2009/202094
Research Article

Solution to an Optimal Control Problem via Canonical Dual Method

1Department of Applied Mathematics, Tongji University, Shanghai 200092, China
2Department of Mathematics, Tongji University, Shanghai 200092, China

Received 14 June 2009; Accepted 1 September 2009

Academic Editor: George Gang Yin

Copyright © 2009 Jinghao Zhu and Jiani Zhou. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

The analytic solution to an optimal control problem is investigated using the canonical dual method. By means of the Pontryagin principle and a transformation of the cost functional, the optimal control of a nonconvex problem is obtained. It turns out that the optimal control can be expressed by the costate via canonical dual variables. Some examples are illustrated.

1. Introduction

Consider the following optimal control problem (primal problem (𝒫) in short):

(𝒫)min𝑇0[]𝐹(𝑥)+𝑃(𝑢)𝑑𝑡,(1)s.t.̇𝑥=𝐴(𝑡)𝑥+𝐵(𝑡)𝑢,𝑥(0)=𝑥0𝑅𝑛,[],𝑡0,𝑇𝑢1,(2) where 𝐹() is continuous on 𝑅𝑛, and 𝑃() is twice continuously differentiable in 𝑅𝑚. An admissible control, taking values on the unit ball 𝐷={𝑢𝑅𝑚𝑢1}, is integrable or piecewise continuous on [0,𝑇]. In (2) we assume that 𝐴(𝑡),𝐵(𝑡) are continuous matrix functions in 𝐶([0,𝑇],𝑅𝑛×𝑛) and 𝐶([0,𝑇],𝑅𝑛×𝑚), respectively. This problem often comes up as a main objective in general optimal control theory [1].

By the classical control theory [2], we have the following Hamilton-Jacobi-Bellman function:

𝐻(𝑡,𝑥,𝑢,𝜆)=𝜆(𝐴(𝑡)𝑥+𝐵(𝑡)𝑢)+𝐹(𝑥)+𝑃(𝑢).(3) The state and costate systems are

̇𝑥=𝐻𝜆(𝑡,𝑥,𝑢,𝜆)=𝐴(𝑡)𝑥+𝐵(𝑡)𝑢,𝑥(0)=𝑥0,̇𝜆=𝐻𝑥(𝑡,𝑥,𝑢,𝜆)=𝐴𝜆𝐹(𝑥),𝜆(𝑇)=0.(4)

In general, it is difficult to obtain an analytic form of the optimal feedback control for the problem (1)-(2). It is well known that, in the case of unconstraint, if 𝑃(𝑢) is a positive definite quadratic form and 𝐹(𝑥) is a positive semidefinite quadratic form, then a perfect optimal feedback control is obtained by the solution of a Riccati matrix differential equation. The primal goal of this paper is to present an analytic solution to the optimal control problem (𝒫).

We know from the Pontryagin principle [1] that if the control ̂𝑢 is an optimal solution to the problem (𝒫), with ̂𝑥() and ̂𝜆() denoting the state and costate corresponding to ̂𝑢(), respectively, then ̂𝑢 is an extremal control, that is, we have

̇̂𝑥=𝐻𝜆̂𝜆𝑡,̂𝑥,̂𝑢,=𝐴(𝑡)̂𝑥+𝐵(𝑡)̂𝑢,̂𝑥(0)=𝑥0,(5)̇̂𝜆=𝐻𝑥̂𝜆𝑡,̂𝑥,̂𝑢,=𝐴̂̂𝜆𝐻̂𝜆𝐹(̂𝑥),(𝑇)=0,(6)𝑡,̂𝑥(𝑡),̂𝑢(𝑡),𝜆(𝑡)=min𝑢1𝐻̂,[].𝑡,̂𝑥(𝑡),𝑢,𝜆(𝑡)a.e.𝑡0,𝑇(7)

By means of the Pontryagin principle and the dynamic programming theory, many numerical algorithms have been suggested to approximate the solution to the problem (𝒫) (see, [35]). This is due to the nonlinear integrand in the cost functional. It is even difficult for the case of 𝑃(𝑢) being nonconvex on the unit ball 𝐷 in 𝑅𝑚. We know that when 𝑃(𝑢) is nonconvex on the unit ball 𝐷, sometimes the optimal control of the problem (𝒫) may exist. Let us see the following simple example for 𝑛=𝑚=1:

(𝒫)min101𝑥2𝑢2[]𝑑𝑡,s.t.̇𝑥=𝑥+𝑢,𝑥(0)=0,𝑡0,𝑇,|𝑢|1.(8) In fact, it is easy to see that ̂𝑢(𝑡)1;𝑡[0,𝑇] is an optimal control.

In this paper, we consider 𝑃(𝑢) to be nonconvex. If the optimal control of the problem (𝒫) exists, we solve the problem (1) to find the optimal control which is an expression of the costate. We see that, with respect to 𝑢, the minimization in (7) is equivalent to the following global nonconvex optimization over a sphere:

min𝑢1̂𝑃(𝑢)+𝜆(𝑡)[]𝐵(𝑡)𝑢,a.e.𝑡0,𝑇,(9) when 𝑃(𝑢) is a nonconvex quadratic function, the problem (9) can be solved completely by the canonical dual transformation [68]. In [9], the global concave optimization over a sphere is solved by use of a differential system with the canonical dual function. Because the Pontryagin principle is a necessary condition for a control to be optimal, it is not sufficient for obtaining an optimal control to solve only the optimization (9). In this paper, combing the method given in [6, 9] with the Pontryagin principle, we solve problem (1)-(2) which has nonconvex integrand on the control variable in the cost functional and present the optimal control expressed by the costate via canonical dual variables.

2. Global Optimization over a Sphere

In this section we present a differential flow to deal with the global optimization, which is used to find the optimal control expressed by the costate in the next section. Here we use the method in our another paper (see [9]).

In what follows we consider the function 𝑃(𝑥) to be twice continuously differentiable and nonconvex on the unit ball in 𝑅𝑚. Define the set

𝐺=𝜌>02𝑃(𝑥)+𝜌𝐼>0,𝑥𝑥1.(10) Since 𝑃(𝑥) is nonconvex and 2𝑃(𝑥) is bounded on 𝐷={𝑥𝑅𝑚𝑥𝑥1}, 𝐺 is an open interval (𝜌,+) for the nonnegative real number 𝜌 depending on 𝑃(𝑥). Let 𝜌𝐺 and ̃𝑥{𝑥𝑥1} satisfy the following KKT equation:

𝑃(̃𝑥)+𝜌̃𝑥=0.(11) We focus on the flow ̂𝑥(𝜌) defined near 𝜌 by the following backward differential equation:

𝑑̂𝑥+𝑑𝜌2𝑃(̂𝑥)+𝜌𝐼1𝜌̂𝑥=0,𝜌𝛿,𝜌,𝜌(12)̂𝑥=̃𝑥.(13) The flow ̂𝑥(𝜌) can be extended to wherever 𝜌𝐺(0,𝜌] [10]. The dual function [6] with respect to a given flow ̂𝑥(𝜌) is defined as

𝑃𝑑𝜌(𝜌)=𝑃(̂𝑥(𝜌))+2̂𝑥𝜌(𝜌)̂𝑥(𝜌)2.(14) We have

𝑑̂𝑥(𝜌)𝑑𝜌2𝑃(̂𝑥(𝜌))+𝜌𝐼𝑑̂𝑥(𝜌)=𝑑𝜌12𝑑̂𝑥(𝜌)̂𝑥(𝜌)𝑑𝑑𝜌=2𝑃𝑑(̂𝜌)𝑑𝜌20.(15) Consequently

𝑑2𝑃𝑑(𝜌)𝑑𝜌20.(16)

It means that 𝑑𝑃𝑑(𝜌)/𝑑𝜌 decreases when 𝜌 increases in 𝐺. If, for a ̂𝜌𝐺, 𝑑𝑃𝑑(̂𝜌)/𝑑𝜌0, then 𝑑𝑃𝑑(𝜌)/𝑑𝜌0 for 𝜌𝐺[̂𝜌,). Therefore,

𝑃𝑑(̂𝜌)𝑃𝑑(𝜌),(17) as long as 𝜌̂𝜌.

Theorem 1. If the flow ̂𝑥(𝜌),𝜌𝐺(0,𝜌], defined by (11)–(13), passes through a boundary point of the ball 𝐷={𝑥𝑅𝑚𝑥1} at ̂𝜌𝐺, that is, []̂𝑥(̂𝜌)̂𝑥(̂𝜌)=1,̂𝜌𝐺0,𝜌,(18) then ̂𝑥 is a global minimizer of 𝑃(𝑥) over the ball 𝐷. Further one has min𝐷𝑃(𝑥)=𝑃(̂𝑥)=𝑃𝑑(̂𝜌)=max𝜌𝑃̂𝜌𝑑(𝜌).(19)

Proof. Since ̂𝜌𝐺, ̂𝜌>0. For each 𝑥𝐷 and whenever 𝜌̂𝜌 we have 𝑃𝜌(𝑥)𝑃(𝑥)+2𝑥𝑥1inf𝐷𝜌𝑃(𝑥)+2𝑥𝑥1=𝑃𝑑(𝜌).(20) By (17), (18), we have 𝑃(𝑥)max𝜌𝑃̂𝜌𝑑(𝜌)=𝑃𝑑𝜌(̂𝜌)=𝑃(̂𝑥(̂𝜌))+2(̂𝑥(̂𝜌))̂𝑥(̂𝜌)1=𝑃(̂𝑥(̂𝜌)).(21) Thus min𝐷𝑃(𝑥)=max𝜌𝑃̂𝜌𝑑(𝜌).(22) This concludes the proof of Theorem 1.

To illustrate the canonical dual method, let us present several examples as follows.

Example 2. Let us consider the following one-dimensional concave minimization problem: 𝑝=min𝑃(𝑥)=1𝑥124𝑥2+𝑥,s.t.𝑥21.(23) We have 𝑃(𝑥)=13𝑥32𝑥+1,𝑃(𝑥)=𝑥22<0,𝑥21.(24) By choosing 𝜌=10, we solve the following equation in {𝑥2<1}: 13𝑥32𝑥+1+10𝑥=0(25) to get a solution ̃𝑥=0.1251. Next we solve the following boundary value problem of the ordinary differential equation: 𝑑𝑥(𝜌)=𝑑𝜌𝑥(𝜌)𝑥2𝜌(𝜌)+2𝜌,𝑥=0.1251,𝜌10.(26) To find a parameter such that 𝑥2(𝜌)=1,(27) we get ̂𝜌=103,(28) which satisfies 𝑃(𝑥)+̂𝜌=𝑃(𝑥)+103>0,𝑥21.(29) Let 𝑥(10/3) be denoted by ̂𝑥. To find the value of ̂𝑥, we compute the solution of the following algebra equation: 13𝑥32𝑥+1+103𝑥=0,𝑥2=1(30) and get ̂𝑥=1. It follows from Theorem 1 that ̂𝑥=1 is the global minimizer of 𝑃(𝑥) over [1,1].

Example 3. We now consider the nonconvex minimization problem: 𝑝1=min𝑃(𝑥)=3𝑥3+2𝑥,s.t.𝑥21.(31) By choosing 𝜌=72, we solve the following equation in {𝑥2<1}: 𝑥2+2+72𝑥=0(32) to get a solution ̃𝑥=2/(4+32). Next we solve the following boundary value problem of the ordinary differential equation: ̇𝑥=𝑥2𝑥+𝑡,𝑡𝑥72,=7224+32.(33) To find a parameter such that 𝑥2(𝜌)=1,(34) we get ̂𝜌=3,(35) which satisfies 𝑃(𝑥)+̂𝜌=𝑃(𝑥)+3=2𝑥+3>0,𝑥21.(36) Let 𝑥(3) be denoted by ̂𝑥. To find the value of ̂𝑥, we compute the solution of the following algebra equation: 𝑥2+2+3𝑥=0,𝑥2=1(37) and get ̂𝑥=1. It follows from Theorem 1 that ̂𝑥=1 is the global minimizer of 𝑃(𝑥) over [1,1].

Example 4. Given a symmetric matrix 𝐺𝑅𝑚×𝑚 and a nonzero vector 𝑓𝑅𝑚, let 𝑃(𝑥)=(1/2)𝑥𝐺𝑥𝑓𝑥 be a nonconvex quadratic function. Consider the following global optimization problem over a sphere: 1min𝑃(𝑥)=2𝑥𝐺𝑥𝑓𝑥,s.t.𝑥𝑥1.(38) Suppose that 𝐺 has 𝑝𝑚 distinct eigenvalues 𝑎1<𝑎2<<𝑎𝑝. Since 𝑃(𝑥)=(1/2)𝑥𝐺𝑥𝑓𝑥 is nonconvex, 𝑎1<0. Let us choose a large 𝜌>𝑎1 such that 0<𝐺+𝜌𝐼1𝑓<1.(39) By solving the boundary value problem of ordinary differential equation 𝑑𝑥𝑑𝜌=(𝐺+𝜌𝐼)1𝜌𝑥,=𝐺+𝜌𝐼1𝑓,𝜌𝜌,(40) we get the unique solution 𝑥(𝜌)=(𝐺+𝜌𝐼)1𝑓,𝜌𝜌.(41) Since 𝐺 is symmetric, there exists an orthogonal matrix 𝑅 such that 𝑅𝐺𝑅=𝐷=(𝑎𝑖𝛿𝑖𝑗) (a diagonal matrix) and correspondingly 𝑅𝑓=𝑔=(𝑔𝑖) (a vector). By (41), we have 𝑥(𝜌)𝑥(𝜌)=𝑓(𝐺+𝜌𝐼)2𝑓=𝑝𝑖=1𝑔2𝑖𝑎𝑖+𝜌2.(42) Since 𝑓(𝐺+𝜌𝐼)2𝑓<1 and lim𝜌>𝑎1,𝜌𝑎1𝑝𝑖=1𝑔2𝑖𝑎𝑖+𝜌2=+,(43) there exists ̂𝜌(𝑎1,𝜌) uniquely such that 𝑥(̂𝜌)𝑥(̂𝜌)=𝑓(𝐺+̂𝜌𝐼)2𝑓=𝑝𝑖=1𝑔2𝑖𝑎𝑖+̂𝜌2=1.(44) By Theorem 1, we see that 𝑥(̂𝜌)=(𝐺+̂𝜌𝐼)1𝑓 is a global minimizer of the problem.

3. Find an Analytic Solution to the OptimalControl Problem

In this section, we consider 𝐴(𝑡),𝐵(𝑡) in problem (1)-(2) to be constant matrices, 𝐹(𝑥)=𝑐𝑥 and

1𝑃(𝑢)=2𝑢𝐺𝑢𝑏𝑢,(45)

where 𝑐𝑅𝑛×1,𝑏𝑅𝑚×1, and 𝐺(𝑅𝑚×𝑚) is a symmetric matrix. Suppose that 𝐺 has 𝑝𝑚 distinct eigenvalues 𝑎1<𝑎2<<𝑎𝑝 and 𝑎1<0. Moreover, we need the following basic assumption:

𝐵rank𝐵,𝑏>rank.()

We consider the following optimal control problem:

(𝒫)min𝐽(𝑢)=𝑇0𝑐1𝑥+2𝑢𝐺𝑢𝑏𝑢𝑑𝑡,(46)s.t.̇𝑥=𝐴𝑥+𝐵𝑢,𝑥(0)=𝑥0[],𝑡0,𝑇,𝑢1.(47)

To solve the above problem, we define the function 𝜙(𝑡,𝑥)=𝜓(𝑡)𝑥, where 𝜓(𝑡) is the solution to the following Cauchy boundary value problem of the ordinary differential equation:

̇𝜓(𝑡)=𝐴𝜓(𝑡)+𝑐,(48)𝜓(𝑇)=0.(49) By comparing (48)-(49) with (6) in terms of this special problem (46)-(47), we see that

[]𝜓(𝑡)=𝜆(𝑡),a.e.𝑡0,𝑇.(50)

Noting that 𝜓(𝑇)=0 and 𝑥(0)=𝑥0, we have

𝐽(𝑢)=𝑇0𝑐1𝑥+2𝑢𝐺𝑢𝑏𝑢=𝑑𝑡𝑇0̇𝜓(𝑡)+𝐴𝜓(𝑡)1𝑥+2𝑢𝐺𝑢𝑏𝑢=𝑑𝑡𝑇0̇𝜓(𝑡)𝑥+𝜓(𝑡)1𝐴𝑥+2𝑢𝐺𝑢𝑏𝑢=𝑑𝑡𝑇0̇𝜓(𝑡)𝑥+𝜓(𝑡)(𝐴𝑥+𝐵𝑢)𝜓(𝑡)1𝐵𝑢+2𝑢𝐺𝑢𝑏𝑢=𝑑𝑡𝑇0̇𝜓(𝑡)𝑥(𝑡)+𝜓(𝑡)̇𝑥(𝑡)𝜓(𝑡)1𝐵𝑢+2𝑢𝐺𝑢𝑏𝑢=𝑑𝑡𝑇0̇𝜙(𝑡,𝑥(𝑡))𝜓(𝑡)1𝐵𝑢+2𝑢𝐺𝑢𝑏𝑢+𝑑𝑡=𝜙(𝑇,𝑥(𝑇))𝜙(0,𝑥(0))𝑇012𝑢𝐺𝑢𝑏𝑢𝜓(𝑡)𝐵𝑢𝑑𝑡=𝜙(0,𝑥(0))+𝑇012𝑢𝐺𝑢𝑏𝑢𝜓(𝑡)𝐵𝑢𝑑𝑡.(51) Thus,

min𝐽(𝑢)=𝜙(0,𝑥(0))+min𝑇012𝑢𝐺𝑢𝑏𝑢𝜓(𝑡)𝐵𝑢𝑑𝑡.(52)

Consequently, we deduce that, for almost every 𝑡 in [0,𝑇], the optimal control is

̂𝑢(𝑡)=argmin𝑢𝑢112𝑢𝐺𝑢𝑏𝑢𝜓(𝑡)𝐵𝑢.(53) By the relation between 𝜓(𝑡) and the costate in (50), for given 𝑡[0,𝑇], we need to solve the following nonconvex optimization:

1min2𝑢𝐺𝑢𝑏𝐵𝜆(𝑡)𝑢,s.t.𝑢𝑢1.(54) It follows from the basic assumption () that 𝑏𝐵𝑇𝜆(𝑡)0 for each 𝑡[0,𝑇]. By Example 4 and (53), for almost every 𝑡 in [0,𝑇], we have

̂𝑢(𝑡)=𝐺+𝜌𝑡𝐼1𝑏𝐵𝜆(𝑡),(55) with the dual variable 𝜌𝑡>𝑎1 satisfying

𝑏𝐵𝜆(𝑡)𝐺+𝜌𝑡𝐼2𝑏𝐵𝜆(𝑡)=1.(56) We define the function 𝜌(𝜆) with respect to 𝜆 by the following equation:

𝑏𝐵𝜆(𝐺+𝜌(𝜆)𝐼)2𝑏𝐵𝜆=1,𝜌(𝜆)>𝑎1(57) and obtain an analytic solution to the optimal control problem via a costate expression

̂𝑢=(𝐺+𝜌(𝜆)𝐼)1𝑏𝐵𝜆.(58)

On the other hand, by the solution of the Cauchy boundary value problem of the ordinary differential equation (48)-(49), we have

𝜆(𝑡)=𝜓(𝑡)=𝑒𝐴𝑇0𝑇𝑡𝑒𝐴𝑡𝑒𝐴𝑠𝑑𝑠𝑐=𝑒𝐴(𝑇𝑡)0𝑇𝑡𝑒𝐴𝑠𝑑𝑠𝑐.(59)

Example 5. Consider the following optimal control problem: (𝒫)min101𝑥2𝑢2𝑥[],𝑑𝑡s.t.̇𝑥=𝑥+𝑢,(0)=0,𝑡0,1|𝑢|1.(60) This is a simple case of (46),(47). We have 𝐺=1,𝑐=1,𝑏=0,𝐴=1,𝐵=1,𝑇=1. By (59), we have 𝜆(𝑡)=𝑒1𝑡01𝑡𝑒𝑠𝑑𝑠=𝑒1𝑡10(𝑡1).(61) To find an analytic solution of the optimal control problem, we solve the equation (𝜌1)2𝜆2(𝑡)=1,𝜌>1(62) to get ||||𝑒𝜌=1+𝜆(𝑡)=1+1𝑡1.(63) By (58), we obtain an analytic solution of the optimal control problem which can be expressed as ̂𝑢(𝑡)=(𝜌1)1[]=𝑒𝜆(𝑡)1𝑡111𝑒1𝑡=1,(𝑡1).(64)

4. Concluding Remarks

In this paper, a new approach to optimal control problems has been investigated using the canonical dual method. Some nonlinear and nonconvex problems can be solved by global optimizations, and therefore, the differential flow defined by the KKT equation (see (11)) can produce an analytic solution of the optimal control problem. Meanwhile, by means of the canonical dual function, an optimality condition is proved (see Theorem 1). The global optimization problem is solved by a backward differential equation with an equality condition (see (12), (18)). More research needs to be done for the development of applicable canonical dual theory.

Acknowledgment

This research was partly supported by the National Science Foundation of China under grants No. 10671145.

References

  1. L. S. Pontryagin, The Mathematical Theory of Optimal Processes, Pergamon Press, Oxford, UK, 1964.
  2. D. L. Russell, Mathematics of Finite-Dimensional Control Systems, vol. 43 of Lecture Notes in Pure and Applied Mathematics, Marcel Dekker, New York, NY, USA, 1979.
  3. C. A. Floudas and V. Visweswaran, “Quadratic optimization,” in Handbook of Global Optimization, R. Horst and P. M. Pardalos, Eds., vol. 2, pp. 217–269, Kluwer Academic Publishers, Dordrecht, The Netherlands, 1995. View at Google Scholar
  4. Y. Ye, “On affine scaling algorithms for nonconvex quadratic programming,” Mathematical Programming, vol. 56, no. 3, pp. 285–300, 1992. View at Publisher · View at Google Scholar
  5. J. Zhu and Z. Zou, “An approximation to discrete optimal feedback controls,” International Journal of Mathematics and Mathematical Sciences, vol. 2003, no. 47, pp. 2989–3001, 2003. View at Publisher · View at Google Scholar
  6. D. Y. Gao, “Canonical duality theory and solutions to constrained nonconvex quadratic programming,” Journal of Global Optimization, vol. 29, no. 4, pp. 377–399, 2004. View at Google Scholar
  7. D. Y. Gao, Duality Principles in Nonconvex Systems, vol. 39 of Nonconvex Optimization and Its Applications, Kluwer Academic Publishers, Dordrecht, The Netherlands, 2000.
  8. D. Y. Gao, “Solutions and optimality criteria to box constrained nonconvex minimization problems,” Journal of Industrial and Management Optimization, vol. 3, no. 2, pp. 293–304, 2007. View at Google Scholar
  9. J. Zhu, S. Tao, and D. Y. Gao, “A study on concave optimization via canonical dual function,” Journal of Computational and Applied Mathematics, vol. 224, no. 2, pp. 459–464, 2009. View at Publisher · View at Google Scholar
  10. C. Robinson, Dynamical Systems, Studies in Advanced Mathematics, CRC Press, Boca Raton, Fla, USA, 2nd edition, 1999.