Open access peer-reviewed chapter

Optimal Control of Evolution Differential Inclusions with Polynomial Linear Differential Operators

Written By

Elimhan N. Mahmudov

Submitted: September 6th, 2019 Reviewed: December 18th, 2019 Published: February 13th, 2020

DOI: 10.5772/intechopen.90888

From the Edited Volume

Functional Calculus

Edited by Kamal Shah and Baver Okutmuştur

Chapter metrics overview

722 Chapter Downloads

View Full Metrics

Abstract

In this chapter, we studied a new class of problems in the theory of optimal control defined by polynomial linear differential operators. As a result, an interesting Mayer problem arises with higher order differential inclusions. Thus, in terms of the Euler-Lagrange and Hamiltonian type inclusions, sufficient optimality conditions are formulated. In addition, the construction of transversality conditions at the endpoints of the considered time interval plays an important role in future studies. To this end, the apparatus of locally adjoint mappings is used, which plays a key role in the main results of this chapter. The presented method is demonstrated by the example of the linear optimal control problem, for which the Weierstrass-Pontryagin maximum principle is derived.

Keywords

  • Euler-Lagrange
  • differential inclusion
  • set-valued mapping
  • polynomial differential operators
  • linear problem
  • transversality
  • Weierstrass-Pontryagin maximum principle

1. Introduction

This chapter concerns with the special kind of optimal control problem with differential inclusions, where the left-hand side of the evolution inclusion is polynomial linear differential operators with variable coefficients; in fact, the main difficulty in the considered problems is to construct the Euler-Lagrange type higher order adjoint inclusions and the transversality conditions. That is why in the whole literature, only the qualitative properties of second-order differential inclusions are investigated (see [1, 2, 3] and references therein).

The paper [1] gives necessary and sufficient conditions ensuring the existence of a solution to the second-order differential inclusion with Cauchy initial value problem. Furthermore, second-order interior tangent sets are introduced and studied to obtain such conditions. The paper [2] studies, in the context of Banach spaces, the problem of three boundary conditions for both second-order differential inclusions and second-order ordinary differential equations. The results are obtained in several new settings of Sobolev type spaces involving Bochner and Pettis integrals. In the paper [3], the existence of viable solutions to the Cauchy problem xFxx,x0=x0,x0=y0 is proved, where F is a set-valued map defined on a locally compact set MR2n, contained in the Frechet subdifferential of a φ-convex function of order two.

Some qualitative properties and optimization of first-order discrete and continuous time processes with lumped and distributed parameters have been expanding in all directions at an astonishing rate during the last few decades (see [4, 5, 6, 7, 8, 9, 10, 11, 12, 13] and their references).

The optimization of higher order differential inclusions was first developed by Mahmudov in [14, 15, 16, 17, 18, 19, 20, 21]. Since then this problem has attracted many author’s attentions (see [22] and their references). The paper [14] studies a new class of problems of optimal control theory with Sturm-Liouville type differential inclusions involving second-order linear self-adjoint differential operators. By using the discretization method guaranteeing transition to continuous problem, the discrete and discrete-approximate inclusions are investigated. Necessary and sufficient conditions, containing both the Euler-Lagrange and Hamiltonian type inclusions, and “transversality” conditions are derived. The paper [15] deals with the optimization of the Bolza problem with third-order differential inclusions and arbitrary higher order discrete inclusions. The work [16] is devoted to the Bolza problem of optimal control theory given by second-order convex differential inclusions with second-order state variable inequality constraints. According to the proposed discretization method, problems with discrete-approximate inclusions and inequalities are investigated. Necessary and sufficient conditions of optimality including distinctive “transversality” condition are proved in the form of Euler-Lagrange inclusions. The paper [17] is concerned with the necessary and sufficient conditions of optimality for second-order polyhedral optimization described by polyhedral discrete and differential inclusions. The paper [18] is devoted to the study of optimal control theory with higher order differential inclusions and a varying time interval. Essentially, under a more general setting of problems and endpoint constraints, the main goal is to establish sufficient conditions of optimality for higher order differential inclusions. Thus with the use of Euler-Lagrange and Hamiltonian type of inclusions and transversal conditions on the “initial” sets, the sufficient conditions are formulated. The paper [21] studies a new class of problems of optimal control theory with state constraints and second-order delay discrete and delay differential inclusions. Under the “regularity” condition by using discrete approximations as a vehicle, in the forms of Euler-Lagrange and Hamiltonian type inclusions, the sufficient conditions of optimality for delay DFIs, including the peculiar transversality ones, are proved.

The present chapter is ordered in the following manner.

In Section 2 the necessary facts and supplementary results from the book of Mahmudov are given [23]; Hamiltonian function and locally adjoint mapping are introduced, and the problems with initial point constraints for polynomial linear differential operators governed by time-dependent set-valued mapping are formulated. In Section 3, we present the main results; on the basis of “transversality” conditions at the endpoints of the considered time interval, the sufficient conditions of optimality for differential inclusions with polynomial linear differential operators and with initial point constraints are proved. In particular, it is shown that our problems involve optimization of the so-called Sturm-Liouville type differential inclusions. To the best of our knowledge, there is no paper which considers optimality conditions for these problems in the literature, and we aim to fill this gap. Therefore, the novelty of our formulation of the problem is justified. To establish the Euler-Lagrange and Hamiltonian inclusions and the transversality conditions, we use the construction of a suitable rewriting of the primal polynomial linear differential operator and the rearrangement of its integration. The case of variable coefficients of polynomial linear differential operators turns out to be more complicated, unless transversality assumptions at the endpoints of the considered time interval are applied. It should be noted that the main proof can be easily generalized to the nonconvex case. Then, using the new approach given in Section 4 of this chapter, we construct the Weierstrass-Pontryagin maximum condition [24] for the linear optimal control problem. Consequently, in the particular case, the maximum principle follows from the Euler-Lagrange inclusion.

In Section 5 the optimality conditions are given for convex problem with second-order differential inclusions and endpoint constraints. By using second-order suitable Euler-Lagrange type adjoint inclusions and transversality conditions, Theorem 5.1 is proved.

The main results in this section can be extended to the case of Hilbert spaces 2,L2n. We remind that a Hilbert space H is a real or complex inner product space that is also a complete metric space with respect to the distance function induced by the inner product [2]. By definition, every Hilbert space is also a Banach space. Furthermore, in every Hilbert space, the following parallelogram identityx+y2+xy2=2x2+y2 holds. Conversely, every Banach space in which the parallelogram identity holds is a Hilbert space. Remember that 2 is a space of numerical sequences, such that if x=xi, then i=1xi2<. In fact 2 is an infinity dimensional coordinate-wise Hilbert space with the corresponding inner product xy=i=1xiyi. Endowing a relevant norm, we have a Banach space. Obviously, optimization of problem with PLDOs can be reduced to problem with geometric constraints in such finite-dimensional Hilbert space. As is known with all the pairs of elements of this space, a certain finite number is associated, i.e., inner product, existence of which is guaranteed by applying the familiar Cauchy Schwarz-Bunyakovskii [25] inequality. We remark that in our case for x=x0x1x22 and x=x0x1x22, the inner product xx=i=1xixi is finite numbers since this series is convergent. Besides it is known [25] that 2 is a self-adjoint space, i.e., p=q, and 1/p+1/q=1, and so 2=2 for p=2. Thus a dual cone constructed can be defined. The set of square integrable functions L2n01 is a Hilbert space with inner product xtyt=01xtytdt.

Advertisement

2. Preliminaries and problem statements

The basic concepts given in this section can be found in the book [23]; let Rn be a n-dimensional Euclidean space, xv be an inner product of elements x,vRn, and xv be a pair of x,v. Let F:RnRn be a set-valued mapping from Rn into the set of subsets of Rn. Therefore F is a convex set-valued mapping, if its graph gphF=xv:vFx is a convex subset of R2n. A set-valued mapping F is called closed if its gphF is a closed subset in R2n. The domain of a set-valued mapping F is denoted by domF and is defined as domF=x:Fx. A set-valued mapping F is convex-valued if Fx is a convex set for each xdomF.

The Hamiltonian function and argmaximum set corresponding to a set-valued mapping F are defined by the relations correspondingly:

HFxv=supvvv:vFx,vRn,FAxv=vFx:vv=HF(xv)

We set HFxv= if Fx=. The interior and relative interior of a set MR2n are denoted by intM and riM, respectively.

A convex cone KMz0, z0M is a cone of tangent directions if from z¯=x¯v¯KMz0 it follows that z¯ is a tangent vector to the set M at a point z0M, i.e., there exists such function q:R1R2n that z0+αz¯+qαM for sufficiently small α>0 and α1qα0, as α0.

For a set-valued mapping F, the set-valued mapping F:RnRn is defined by

Fvxvx:xvKgphF(xv)
KgphFxv=conegphFxv,x1v1gphF.

It is called the LAM to F at a point xvgphF, where K=z:z¯z0z¯K denotes the dual cone to the cone K, as usual. Below by using the Hamiltonian function, associated to a set-valued mapping F, we will define another LAM. Thus, the LAM to “nonconvex” mapping F is defined as follows:

Fvxvx:HFx1vHF(xv)xx1x,x1Rn,xvgphF,vFAxv.

Clearly, for the convex mapping F, the Hamiltonian function HFv is concave, and the latter definition of LAM coincides with the previous definition of LAM ([23], p. 62). Note that prior to the LAM, the notion of coderivative has been introduced for set-valued mappings in terms of the basic normal cone to their graphs by Mordukhovich [26] and for the smooth convex maps, the two notions are equivalent.

The aim of Section 3 is to obtain the Euler-Lagrange type adjoint inclusion and sufficient optimality conditions for a problem with polynomial linear differential operators:

Minimizeφx1x1xs11,E1
(PV)LxtFxtt,a.e.t01,E2
x0Q0, x0Q1x0Q2,,xs10Qs1E3

where Lx=k=1spktDkx is a PLDO of degree s with variable coefficients pk:01R1 and Dk,k=1,,s is the operator of kth-order derivatives. In what follows for each k, a scalar function pk is kth-order continuously differentiable function, pst0 on 01 identically, Ft:RnRn is time-dependent set-valued mapping, φ:RnsR1 is continuous function, QjRn,j=0,1,,s1 are nonempty subsets of Rn, and ss2 is an arbitrary fixed natural number. It is required to find an arc x˜t of the problem Eqs. (1)(3) for the sth-order differential inclusions satisfying Eq. (2) almost everywhere (a.e.) on a considered time interval and minimizing the functional φx1xs11. An arc x is absolutely continuous and s1 order differentiable function, where xsdsxdtsL1n01. Obviously, such class of functions is a Banach space, endowed with the different equivalent norms.

Remark 2.1. Notice that to get sufficient condition of optimality for the Mayer problem (PV) described by ordinary evolution differential inclusions with PLDOs and with initial point constraints, using the discretized method, we consider the following sth-order discrete-approximate problem instead of the problem (PV):

minimizeφx1s1hΔx1s1hΔs1x1s1h,
k=1spktΔkxtFxtt,t=0,h,2h,,1sh;
Δkx0Qk,k=0,,s1.

Here kth-order difference operator is defined as follows:

Δkxt=1hks=0k1sCksxt+ksh,Cks=k!s!ks!, t=0,h,,1h.

Thus by using the method of approximation [23, 26, 27], we can establish necessary and sufficient conditions for the rather complicated sth-order discrete-approximate problem. Then by passing to the limit in necessary and sufficient conditions of this problem as h0, we can construct the optimality condition for the Mayer problem (PV) described by higher order differential inclusions with PLDOs and with initial point constraints. But in this chapter to avoid long calculations, derivations of these conditions are omitted.

Advertisement

3. Optimization of evolution differential inclusions with PLDOs

In the present section, we study sufficient optimality conditions for the problem (PV). Before all, we formulate the so-called sth-order Euler-Lagrange type differential inclusion and the transversality conditions:

  1. LxtFxtx˜tLx˜tt, a.e. t01,

    where Lxt=k=1s1kDkpktxt is the adjoint PLDO of the primal operator L.

  2. k=0s11skDsk1psk0x0KQ0x˜0;

    k=0s21sk1Dsk2psk0x0KQ1x˜0;

    Dps0x0ps10x0KQs2x˜s20;

    ps0x0KQs1x˜s10.

  3. k=0s11skDsk1psk1x1, k=0s21sk1Dsk2psk1x1,, Dps1x1ps11x1ps1x1φx˜1x˜1x˜s11.

    Later on we assume that xt,t01 is absolutely a continuous function with the higher order derivatives until s1 and xsL1n01. The following condition ensures that the LAM F is nonempty:

  4. Lx˜tFAx˜txtt, a.e. t01 or, equivalently,

    Lx˜txt=HFx˜txt, Lx˜tFx˜tt.

The following are sufficient optimality conditions for evolution differential inclusions with PLDOs.

Theorem 3.1. Let φ be a continuous and convex function and Ft a convex set-valued mapping. Moreover, let

Qj,j=0,,s1

be convex sets. Then for optimality of the trajectory x˜ in the problem (PV) with evolution differential inclusions and PLDOs, it is sufficient that there exists an absolutely continuous function x with the higher order derivatives until s1, satisfying a.e. the Euler-Lagrange type differential inclusion with PLDOs Eqs. (i) and (iv) and transversality conditions Eqs. (ii) and (iii) at the endpoints t=0 and t=1.

Proof. Using Theorem 2.1 ([23], p. 62), the definition of the Hamiltonian function and condition Eq. (i), we obtain

HFxtxttHFx˜txttLxtxtx˜t

which can be rewritten as follows

HFxtxttHFx˜txtt
k=1s1kpktxtkxtx˜t.E4

Further using the definition of the Hamiltonian function, Eq. (4) can be converted to the inequality

0LxtLx˜txtLxtxtx˜t

or

0k=1spktxtx˜tkxtk=1s1kpktxtkxtx˜t.E5

Integrating Eq. (5) over the interval 01, we have

001k=1spktxtx˜tkxt
k=1s1kpktxtkxtx˜tdtE6

Let us denote

B=k=1sxtx˜tkpktxtk=1s1kpktxtkxtx˜t

In what follows our approach lies in reducing B in a relationship consisting of s sums from kk=1s to s of suitable derivatives of scalar products; thus, after some transformations we can deduce an important representation for a first term of B as follows:

k=1sxtx˜tkpktxt=k=1sddtxk1tx˜k1tpktxt
k=2sddtxk2tx˜k2tpktxt
+k=3sddtxk3tx˜k3tpktxt
+k=s2sddtxks+2tx˜ks+2t1m3pktxts3E7
+k=s1sddtxks+1tx˜ks+1t1s2pktxts2
+ddtxtx˜t1m1pstxts1
+k=1sxtx˜t1kpktxtk.

Then in view of Eq. (7) in the definition of B, we have an efficient formula:

B=k=1sddtxk1tx˜k1tpktxt
k=2sddtxk2tx˜k2tpktxtE8
+k=3sddtxk3tx˜k3tpktxt
+k=s2sddtxks+2tx˜ks+2t1s3pktxts3
+k=s1sddtxks+1tx˜ks+1t1s2pktxts2
+ddtxtx˜t1s1pstxts1.

Then taking into account the structure of B in Eq. (8), we can compute the integral on the right-hand side of Eq. (6) as follows:

01Bdt=k=1s01dxk1tx˜k1tpktxt
k=2s01dxk2tx˜k2tpktxt
+k=3s01dxk3tx˜k3tpktxt
+k=s2s01dxks+2tx˜ks+2t1s3pktxts3
+k=s1s01dxks+1tx˜ks+1t1s2pktxts2
+01dxtx˜t1s1pstxts1.

Thus, integrating B, we can obtain

01Bdt=k=1sxk11x˜k11pk1x1
xk10x˜k10pk0x0
k=2sxk21x˜k21pk1x1xk20x˜k20pk0x0
+k=3sxk31x˜k31pk1x1
xk30x˜k30pk0x0
+k=s2sxks+21x˜ks+211s3pk1x1s3
xks+20x˜ks+201s3pk0x0s3
+k=s1sxks+11x˜ks+111s2pk1x1s2
xks+10x˜ks+101s2pk0x0s2
+x1x˜11s1ps1x1s1
x0x˜01s1ps0x0s1.

Here by suitable rearrangement and necessary simplification, we have

01Bdt=k=1sxk11x˜k11pk1x1
k=2sxk21x˜k21pk1x1
+k=3sxk31x˜k31pk1x1
+k=s2sxks+21x˜ks+211s3pk1x1s3
+k=s1sxks+11x˜ks+111s2pk1x1s2E9
+x1x˜11s1ps1x1s1+x0x˜0k=0s11skDsk1psk0x0
+x0x˜0k=0s21sk1Dsk2psk0x0
+x0x˜0k=0s31sk2Dsk3psk0x0
+xs20x˜s20Dps0x0+ps10x0xs10x˜s10ps0x0.

In order to make use of the transversality condition Eq. (ii), we rewrite it in a more relevant form:

x0x˜0k=0s11skDsk1psk0x0

+x0x˜0k=0s21sk1Dsk2psk0x0+x0x˜0k=0s31sk2Dsk3psk0x0

+xs20x˜s20Dps0x0+ps10x0xs10x˜s10ps0x00;xk0KQkx˜k0,

k=0,,s1. Thus, from Eqs. (6) and (9), we have

0k=1sxk11x˜k11pk1x1
k=2sxk21x˜k21dpk1x1dt
+k=3sxk31x˜k31d2pk1x1dt2+k=s2sxks+21x˜ks+211s3ds3pk1x1dts3
+k=s1sxks+11x˜ks+111s2ds2pk1x1dts2
+x1x˜11s1ds1ps1x1dts1.

Using the derivative operator D, it is not hard to see that the relation described above can be expressed in a more compact form:

0k=0s11sk1Dsk1psk1x1x1x˜1
+k=0s21sk2Dsk2psk1x1x1x˜1+E10
Dps1x1ps11x1xs21x˜s21
+ps1x1xs11x˜s11.

Furthermore, applying the definition of the transversality condition Eq. (iii) for all feasible arc x, we have

φx1x1xs11φx˜1x˜1x˜s11
k=0s11skDsk1psk1x1x1x˜1
+k=0s21sk1Dsk2psk1x1x1x˜1+
+Dps1x1ps11x1xs21x˜s21
ps1x1xs11x˜s11E11

Then from the last two inequalities Eqs. (10) and (11) for all feasible arc x, we have immediately φx1x1xs11φx˜1x˜1x˜s11, that is, x˜ is an optimal trajectory. □

Remark 3.1. It can be noted that in the particular case, if p2t=pt,p1t=pt, where p:010, the second-order linear differential operator Lx=p2tx+p1tx is a well-known self-adjoint Sturm-Liouville operator Lxpx.

Corollary 3.1. Let Ft be a closed set-valued mapping. Then under the assumptions of Theorem 3.1, the conditions Eqs. (i) and (iii) can be rewritten in terms of Hamiltonian function as follows:

LxtxHFx˜txtt; .Lx˜tvHx˜txtt,a.e.t01

Proof. Indeed, by Theorem 2.1 ([23], p. 62) and Lemma 5.1 [14], we can write.

Fvxvt=xHFxvt,andFAxvt=vHFxvt

respectively. Then it is easy to see that the result of corollary are equivalent with the conditions Eqs. (i) and (iv) of Theorem 3.1. □

Below nonconvexity of a set-valued mapping Ft means that its Hamilton function in general is a nonconcave function satisfying the condition Eq. (a).

Theorem 3.2. Suppose that we have the “nonconvex” problem (PV), that is, φ:RnsR1 and Ft:RnRn in general are nonconvex function and set-valued mapping, respectively. Moreover, suppose that KQjx˜j0,x˜j0Qj is the cones of tangent directions to Qj,j=0,,s1.

Then for optimality of the trajectory x˜, it is sufficient that there exists an absolutely continuous function x, satisfying the following conditions:

  1. LxtFxtx˜tLx˜tt, a.e. t01,

  2. k=0s1j1skjDsk1jpsk0x0KQjx˜j0,j=0,,s1,

  3. φν0v1vs1φx˜1x˜1x˜s11

    j=0s1k=0s1j1skjDsk1jpsk1x1νjx˜j1,

    vjRn,j=0,,s1,

  4. Lx˜txt=HFx˜txtt, a.e. t01.

Proof. In the proof of Theorem 3.1, we have used the following inequality:

HFxtxttHFx˜txtt
k=1s1kpktxtkxtx˜t.E12

Hence, from the inequality Eq. (12), immediately we have the inequality Eq. (10). Moreover, setting νj=x˜j1j=0s1 for all feasible trajectories x, it is not hard to see that for nonconvex φ the following inequality holds:

φx1x1xs11φx˜1x˜1x˜s11j=0s1k=0s1j1skjDsk1jpsk1x1xj1x˜j1,j=0,1,,s1,

Then for the furthest proof, we proceed by analogy with the preceding derivation of Theorem 3.1. □

Advertisement

4. Some applications to optimal control problems with PLDOs

In this section we give two applications of our results. The first one is the particular Mayer problem for differential inclusions involving PLDOs with constant coefficients, and the second one concerns optimization of “linear” differential inclusions with PLDOs and constant coefficients. Thus, suppose now we have the following optimization problem (for simplicity we consider a convex problem) with sth-order PLDO with constant coefficients:

Minimizeφ0x1,
PCLxtFxtt,a.e.t01,Lx=Dsx+p1Ds1x++ps1Dx
x0=α0,x0=α1,x0=α2,,xs10=αs1,.E13

where L is the sth-order polynomial operator, pk,k=1,,s1 are some real constants, Ft:RnRn is a convex set-valued mapping, φ0:RnR1 is a continuous convex function, and αjRn,j=0,,s1 are fixed n-dimensional vectors. It is known that the multiplication operation is commutative for polynomial linear differential operators with constant coefficients. On the other hand, the sth-order adjoint operator is defined as follows:

Lx=1sDsx+1s1p1Ds1x+ps1Dx.

Corollary 4.1. Let φ0 and Ft be a convex function and a set-valued mapping, respectively. Then, for the trajectory x˜ to be optimal in the problem (PC), it is sufficient that there exists an absolutely continuous function x satisfying the Euler-Lagrange type differential inclusion.

LxtFxt(x˜tLx˜t)t;Lx˜txt=HFx˜txt,
a.e.t01,Lx=1sxs+1s1p1xs1+ps1x

and transversality condition at the endpoint t=1.

1sxs11φ0x˜1, xj1=0,j=0,,m2.

Proof . We conclude this proof by returning to the conditions Eqs. (i)–(iii) of Theorem 3.1. Clearly, a problem (PC) can be reduced to the problem of form (PV), where

φx1x1xs11φ0x1.

It follows that φx1x1xs11=xφ0x1×(0,,0s1). On the other hand, since pst1,pjt=psj, and j=1,,s1 are constants, by sequentially substitution in the transversality condition Eq. (iii), we derive that

k=0s1j1skjDsk1jpsk1x1=xs1j1=0,j=1,,s1,

and therefore for j=0.

k=0s11skDsk1psk1x1=1sxs11φ0x˜1.□

Suppose now that we have the so-called linear Mayer problem with PLDOs:

Minimizeφ0x1,E14
LxtFxtt,a.e.t01,
xj0=αj,j=0,,s1, Fxt=Atx+BtUE15

where φ0 differentiable convex function; At and Bt are n×n and n×r continuous matrices, respectively; U is a convex compact of Rr; αj, j=0,,s1 are constant vectors. It is required to find a control function u˜ such that the corresponding trajectory x˜ minimizes the Mayer functional φ0x1.

In fact, this is optimization of Cauchy problem for “linear” differential inclusions with PLDO. The controlling parameter u is called admissible if it only takes values in the given control set U which is a nonempty, convex compact.

Theorem 4.1. The arc x˜t corresponding to the controlling parameter u˜t is a solution to Eqs. (14) and (15) if there exists an absolutely continuous function x, satisfying the Euler-Lagrange type differential equation, the transversality condition, and Weierstrass-Pontryagin maximum principle:

Lxt=Atxt,a.e.t01,
1sxs11=φ0x˜1, .xj1=0,j=0,,s2
Btu˜txt=maxuUBtuxt.

Proof. Obviously, the Hamiltonian is

HFxvt=Atxv+maxuUBtuv.

Hence,

Fvxv˜t=xHFxvt=Atv,v˜FAxvtv˜=Atx+Btu˜

where the argmaximum inclusion v˜FAxvt implies that Btu˜v=maxuUBtuv and Fvxv˜t. Thus, applying Theorem 3.1, we obtain

Lxt=Atxt,Lx˜tFAx˜txtt,
Btu˜txt=maxuUBtuxt.

Consequently, the transversality condition Eq. (ii) of Theorem 3.1 is unnecessary and by Corollary 4.1 1sDs1x1=φ0x˜1, Djx1=0,j=0,,s2. □

Remark 4.1. Suppose that in the definition of sth-order PLDO (see Eq. (13)) s=1, pi=0,i=1,..,s1 and U is a convex closed polyhedron. Then we have linear equations with variable coefficients x=Atx+Btu,uU in the finite time interval t01. Obviously, for such problems an adjoint Euler-Lagrange type differential equation and transversality condition at a point t=1 consist of the following: xt=Atxt,x1=φ0x˜1. We remind that along with Pontryagin’s maximum principle (see, e.g., [24]) under the condition for generality of position for time-optimal problem, the existence results of optimal control are proved.

Example 4.1. Let us consider the following Mayer problem with second-order PLDO Lx=D2x=x:

Infimumφx1x1is subject tox=u,u11,x0Q0,x0Q1.E16

Here φx1x1=x21x1 and Q0=0,Q1=1.

It should be noted that substituting Ft=ut,xt=at,m=1 into Newton’s second law Ft=mat, we have x=u.

Obviously, in this problem FxtFx=u:u1, s=2.

Then Eq. (16) has the form:

Infimumφx1x1is subject toLxFx,t01,x0=0,x0=1.E17

It can be easily seen that in the adjoint inclusion Eq. (i).

Dp1txt+D2p2txtFxtx˜tLx˜t

of Corollary 3.1 p2t0 and p2t1, and so we have

xtFxtx˜tx˜t.

Now, it is not hard to see that

HFxv=maxuuv:u1=vE18

and

Fvxv=xHFxv0,vFAxv=1+1.E19

Then taking into account Lx=d2x/dt2, as a result of Theorem 3.1 (see also Corollary 3.1) from Eq. (19), we deduce that

x=0,t01,

for which the solution is a linear function of the form xt=C1t+C2, where C1,C2 are arbitrary constants. Then Eq. (18) implies that u˜txt=xt or

u˜t=sgnxt,ifxt0,u011,ifxt=0.E20

Further, from the linearity of x and from Eq. (20), we insure that each optimal control function is a piecewise constant function.

In addition, by the transversality condition Eq. (iii) of Corollary 3.1, we can write.

x1x1φx˜1x˜1. On the other hand, it is not hard to see that φxy=y2x is a convex function; in fact, the 2×2 Hessian matrix

φxy=φxxxyφxyxyφyxxyφyyxy=0002

is a positive semidefinite, that is, all eigenvalues of φxy are nonnegative. Indeed, denoting this matrix by A, we see that the characteristic equation AλE=λ22λ=0(E is a 2×2 unique square matrix) has two real nonnegative eigenvalues λ1=0,λ2=2. Consequently, φxy is convex and φxy=12y. It follows that φx˜1x˜1=12x˜1. Comparing this relation with x1x1φx˜1x˜1, we immediately have x1=1,x1=2x˜1. Then from a general solution of the adjoint Euler-Lagrange type inclusion (equation) xt=C1t+C2, we have 2x˜1=x1=C1+C2,1=x1=C1 (C1,C2 are arbitrary constants), and so xt=1t2x˜1, whence xt0, if tτ=12x˜1. Therefore, Eq. (20) implies that for optimal control u˜, there are four possibilities:

u˜t=1,xt>0,t01.E21
u˜t=1,xt<0,t01.E22
u˜t=1,if0t<τ,1,ifτ<t1.E23
u˜t=1,if0t<τ,1,ifτ<t1.E24

(observe that τ is a point of discontinuity of u˜ and the values of the control functions u˜ at a point of discontinuity τ are unessential). As a consequence, it follows that either the sign of the linear function xt does not change for the whole interval 01 or xt>0,0t<τ;xt<0,τ<t1 for a some τ in the interval 0<τ<1 (the case Eq. (24) is excluded). Therefore, since u˜t is a piecewise constant function, having not more than two intervals of constancy, we have either the cases Eqs. (21) and (22) or the case Eq. (23). In general, using Eqs. (21)(23), by solving the Cauchy problem

xt=ut,x0=0,x0=1E25

we have a unique solution of the initial value problem Eq. (25). Thus for the time interval on which u=1, we have xt=t+c1;xt=t2/2+c1t+c2(c1,c2 are constants). From Eq. (25) we obtain.

xt=t+1;xt=t2/2+t.E26

By analogy, for u=1 we have.

xt=1t;xt=t2/2+t.E27

Now, let x1t and x2t be parabolas of Eqs. (26) and (27), respectively. Here, in the case Eq. (21), u˜t=1,t01, and so from Eq. (26), we have x˜11=0.5+1=1.5; x˜11=2. Consequently, the value of problem Eq. (16) is φx˜11x˜11=x˜121x˜11=221.5=2.5, if u˜t=1,t01. By a similar way, for a control function u˜t=1,t01 from Eq. (27), we obtain that x˜21=12/2+1=0.5,x˜21=0 and φx˜21x˜21=020.5=0.5.

On the other hand, in the case Eq. (23), the control function u˜t first is equal to +1 and then equal to 1, and the trajectory x˜t consists of two pieces of parabolas x˜1t and x˜2t (x˜t is continuous and piecewise smooth on the interval 0t1). Then the solution of the equation Eq. (25) on the interval 0tτ is given by Eq. (26); at a point τ are satisfied x1τ=τ2/2+τ,x1τ=1+τ. Consider now the initial value problem:

x2t=1,x2τ=τ2/2+τ,x2τ=1+τ, tτ1.E28

It is clear that τ2/2+τ=x2τ=τ2/2+c1τ+c2 and 1+τ=x2τ=τ+c1 from which we obtain that the solution of the initial value problem Eq. (28) is x˜2t=t2/2+1+2τtτ2. Substituting the value τ=12x˜1 into equation x˜2t=1t+2τ, we have 5x˜21=2,x˜21=x˜1=0.4 (it follows that τ=0.2). Moreover, x˜21=2ττ2+0.5 and x˜21=x˜1=3/24x˜21=0.86. Thus, the value of our Mayer problem is φx˜1x˜1=x˜21x˜1=2/5243/50=0.7, where u˜t is defined as in Eq. (23). Comparing the values 2.5,0.5,0.7, we believe that the value of Mayer problem is 0.7.

Advertisement

5. Sufficient conditions of optimality for second-order evolution differential inclusions with endpoint constraints

Note that in this section the optimality conditions are given for second-order convex differential inclusions (PM) with convex endpoint constraints. These conditions are more precise than any previously published ones since they involve useful forms of the Weierstrass-Pontryagin condition and second-order Euler-Lagrange type adjoint inclusions. In the reviewed results, this effort culminates in Theorem 5.1:

Minimizegx1x1,
PMxtFxtxtt,a.e.t01,
x0=x0,x0=x1;x1M0,x1M1,

where g is a convex continuous function, Ft:R2nRn is convex set-valued mapping, and M0,M1Rn are convex sets.

The following adjoint inclusion is the second-order Euler-Lagrange type inclusion for the problem (PM):

a1. xt+vtvtFxtx˜tx˜tx˜tt,a.e.t01,

where

b1. x˜tFAx˜tx˜txtt, a.e. t01.

In what follows we assume that xt, t01 is absolutely continuous function together with the first-order derivatives for which xL1n01. Besides the auxiliary function vt,t01 is absolutely continuous and vL1n01.

The transversality conditions at the endpoint t=1 consist of the following:

c1. v1+x1x1xugx˜1x˜1KM0x˜1×KM1x˜1.

Now we are ready to formulate the following theorem of optimality.

Theorem 5.1. Suppose that g is a continuous and convex function, Ft is a convex set-valued mapping, and M0,M1 are convex sets. Then for optimality of the feasible trajectory x˜t in the problem (PM), it is sufficient that there exists a pair of absolutely continuous functions:

xtvt,t01

satisfying a.e. the second-order Euler-Lagrange type inclusions Eqs. (a1) and (b1) and the transversality condition Eq. (c1) at the endpoint t=1.

Proof. By the proof idea of Theorem 3.1 from Eqs. (a1) and (b1), we obtain the adjoint differential inclusion of second order:

xt+vtvtxuHFx˜tx˜txtt,t01.

On the definition of subdifferential set of the Hamiltonian function HFt for all feasible trajectory xt,t01, we rewrite the last relation in the equivalent form:

HFxtxtxttHFx˜tx˜txtt
xt+vtxtx˜t+vtxtx˜t.E29

Now by using definition of the Hamiltonian function, the inequality Eq. (29) can be reduced to the inequality

0xtx˜txtxtxtx˜tddtvtxtx˜t.E30

Integrating of the inequality Eq. (30) over the interval 01, we derive that

001xtx˜txtxtxtx˜tdt
+v0x0x˜0v1x1x˜1.E31

For convenience we transform the expression in the square parentheses on the right-hand side of Eq. (31) as follows

xtx˜txtxtxtx˜t
=ddtxtx˜txtddtxtxtx˜t.

Thus by elementary property of the definite integrals, we can compute the integral on the right-hand side of Eq. (31):

01xtx˜txtxtxtx˜tdt
=x1x˜1x1x0x˜0x0x1x1x˜1+x0x0x˜0.E32

Then substituting Eq. (32) into Eq. (31), we have

0x1x˜1x1x0x˜0x0v1+x1x1x˜1+v0+x0x0x˜0.E33

Now, remember that x,x˜ are feasible trajectories and x0=x˜0=x0 and x0=x˜0.

=x1 . Then it follows from Eq. (33) that

0x1x˜1x1v1+x1x1x˜1.E34

Now, thanking to the transversality conditions Eq. (c1) at the endpoint t=1, we can rewrite

gx1x1gx˜1x˜1v1+x1+x1x1x˜1+x1x1x1x˜1,x1KM0x˜1,x1KM1x˜1

or, in other words

gx1x1gx˜1x˜1
v1+x1x1x˜1x1x1x˜1.E35

Thus, summing the inequalities Eqs. (34) and (35) for all feasible trajectories x, satisfying the initial conditions x0=x0 and x0=x1 and endpoint constraints x1M0,x1M1, we have the needed inequality:

gx1x1gx˜1x˜10 or gx1x1gx˜1x˜1. □

Advertisement

6. Conclusion

According to proposed method, the problem with the differential inclusions described by polynomial linear differential operators is investigated. Obviously, this problem is an important generalization of problems with first-order differential inclusions. Thus, sufficient conditions of optimality for such problems are deduced. Here the existence of nonfunctional initial point or endpoint constraints generates different kinds of transversality conditions. Besides, there can be no doubt that investigations of optimality conditions of problems with second- and fourth-order Sturm-Liouville type differential inclusions can play an important role in the development of modern optimization and there is every reason to believe that this role will be even more significant in the future. Thus, the suggested problem with linear differential operators and variable coefficients can be used in various forms in applied problems.

References

  1. 1. Auslender A, Mechler J. Second order viability problems for differential inclusions. Journal of Mathematical Analysis and Applications. 1994;181:205-218
  2. 2. Azzam-Laouir D, Castaing C, Thibault L. Three boundary value problems for second order differential inclusion in Banach spaces. Control and Cybernetics. 2002;31:659-693
  3. 3. Cernea A. On the existence of viable solutions for a class of second order differential inclusions. Discussiones Mathematicae. Differential Inclusions, Control and Optimization. 2002;22:67-78
  4. 4. Barbu V, Precupanu T. Convex Control Problems in Banach Spaces. 4th ed. Netherlands: Springer; 2012. 368 p
  5. 5. Blagodatskikh VI, Filippov AF. Differential inclusions and optimal control. Trudy Matematicheskogo Instituta Imeni VA Steklova. 1985;169:194-252
  6. 6. Buttazzo G, Drakhlin ME, Freddi L, Stepanov E. Homogenization of optimal control problems for functional differential equations. Journal of Optimization Theory and Applications. 1997;93:103-119
  7. 7. Cannarsa P, Sinestrari C. Semiconcave Functions, Hamilton-Jacobi Equations and Optimal Control. Boston: Birkhäuser; 2004. 304p
  8. 8. Clarke FH. Functional analysis, calculus of variations and optimal control. Graduate Texts in Mathematics. London: Springer; 2013. 264p
  9. 9. Kolmogorov AN, Fomin SV. Elements of the Theory of Functions and Functional Analysis. (Dover Books on Mathematics) Paperback. New York: Dover Publications; 1999. 128p
  10. 10. Lasiecka I, Triggiani R. Control Theory for Partial Differential Equations: Vol. 1, Abstract Parabolic Systems: Continuous and Approximation Theories. Cambridge: Cambridge University Press; 2000. 672p
  11. 11. Papageorgiou NS, Rădulescu VD. Periodic solutions for time-dependent subdifferential evolution inclusions. Evolution Equations & Control Theory. 2017;6:277-297
  12. 12. Long DQ, Dang Quang A, Luan VT. Iterative method for solving a fourth order differential equation with nonlinear boundary condition. Applied Mathematical Sciences. 2010;4:3467-3481
  13. 13. Zhou Y, Vijayakumar V, Murugesu R. Controllability for fractional evolution inclusions without compactness. Evolution Equations and Control Theory. 2015;4:507-524
  14. 14. Mahmudov EN. Optimization of Mayer problem with Sturm–Liouville-type differential inclusions. Journal of Optimization Theory and Applications. 2018;177:345-375
  15. 15. Mahmudov EN. Approximation and optimization of higher order discrete and differential inclusions. Nonlinear Differential Equations and Applications NoDEA. 2014;21:1-26
  16. 16. Mahmudov EN. Convex optimization of second order discrete and differential inclusions with inequality constraints. Journal of Convex Analysis. 2018;25:1-26
  17. 17. Mahmudov EN. Mathematical programming and polyhedral optimization of second order discrete and differential inclusions. Pacific Journal of Optimization. 2015;11:495-525
  18. 18. Mahmudov EN. Free time optimization of higher order differential inclusions with endpoint constraints. Applicable Analysis. 2018;97:2071-2084
  19. 19. Mahmudov EN. Optimization of boundary value problems for certain higher-order differential inclusions. Journal of Dynamical and Control Systems. 2019;25:17-27
  20. 20. Mahmudov EN. Optimization of second order differential inclusions with boundary value conditions. Journal of Nonlinear and Convex Analysis. 2017;18:1653-1664
  21. 21. Mahmudov EN. Optimal control of second order delay-discrete and delay differential inclusions with state constraints. Evolution Equations & Control Theory. 2018;7:501-529. DOI: 10.3934/eect.2018024
  22. 22. Bors D, Majewski M. On Mayer problem for systems governed by second-order ODE. Optimization. 2014;63:239-254
  23. 23. Mahmudov EN. Approximation and Optimization of Discrete and Differential Inclusions. Waltham, USA: Elsevier; 2011. 396p
  24. 24. Pontryagin LS, Boltyanskii VG, Gamkrelidze RV, Mishchenko EF. The Mathematical Theory of Optimal Processes. New York/London/Sydney: John Wiley & Sons, Inc.; 1965. 360p
  25. 25. Mahmudov EN. Optimal control of Cauchy problem for first order discrete and partial differential inclusions. Journal of Dynamical and Control Systems. 2009;15:587-610
  26. 26. Mordukhovich BS. Variational Analysis and Generalized Differentiation, I: Basic Theory. II: Applications, Grundlehren Series (Fundamental Principles of Mathematical Sciences). Vol. 330 and 331. New York: Springer; 2006. 579p
  27. 27. Mahmudov EN. Necessary and sufficient conditions for discrete and differential inclusions of elliptic type. Journal of Mathematical Analysis and Applications. 2006;323:768-789

Written By

Elimhan N. Mahmudov

Submitted: September 6th, 2019 Reviewed: December 18th, 2019 Published: February 13th, 2020