Abstract
This paper is devoted to the optimization of the Mayer problem with hyperbolic differential inclusions of the Darboux type and duality. We use the discrete approximation method to get sufficient conditions of optimality for the convex problem given by Darboux differential inclusions and the polyhedral problem for a hyperbolic differential inclusion with state constraint. We formulate the adjoint inclusions in the Euler-Lagrange inclusion and Hamiltonian forms. Then, we construct the dual problem to optimal control problem given by Darboux differential inclusions with state constraint and prove so-called duality results. Moreover, we show that each pair of primal and dual problem solutions satisfy duality relations and that the optimal values in the primal convex and dual concave problems are equal. Finally, we establish the dual problem to the polyhedral Darboux problem and provide an example to demonstrate the main constructions of our approach.
Similar content being viewed by others
1 Introduction
Convexity properties are particularly significant for many optimization problems, and when they are fully present, they lead to an efficient type of duality. Our goals in this paper are twofold. On the one hand, to formulate sufficient conditions of optimality for hyperbolic partial differential inclusions for convex Darboux problems and prove optimality conditions for polyhedral problem with state constraints. On the other hand, to construct the dual problem to optimal control problem given by Darboux differential inclusions with state constraints and prove duality relations.
One can use the differential calculus of set-valued mappings for looking for global set-valued solutions to hyperbolic of both partial differential equations, and inclusions [1, 4, 14,15,16,17, 20, 22, 25, 27]. The paper [5] studies an optimal control problem given by hyperbolic differential inclusions with boundary conditions and endpoint constraints. It is suggested a method for second-order optimality conditions. In the article [12], Jackson discusses how certain types of chemical plants’ optimum start-up and control problems can be expressed mathematically as variational problems involving two independent variables with hyperbolic partial differential equations acting as side-conditions. The author also investigates how the first-order variational theory might be expanded to conclude similar to Pontryagin’s maximal principle.
The existence of the solutions set of hyperbolic differential equations have received much attention during the last three decades. Using the mixed generalized Lipschitz and Caratheodory’s criteria, the work [10] gives sufficient conditions for the existence of solutions to the initial value problem for perturbed hyperbolic differential inclusions. The paper [2] deals with the existence of solutions to the Darboux problem with hyperbolic differential inclusion. The authors’ approach is based on the existence of upper and lower solutions and on a fixed point theorem for condensing set-valued mappings.
In the classical calculus of variations, convexity and duality first enter the picture in the correspondence between Lagrangian and Hamiltonian functions and in the way this is connected with necessary conditions and the existence of solutions. This correspondence is a result of the Legendre transformation, which converts Lagrangian functions into Hamiltonians and vice versa. Expressed in terms of the Hamiltonian, the optimality conditions for an arc x pair it with an “adjoint” arc \(x^*\). The pairing carries over to problems of optimal control via the maximum principle. Duality theory in this context aims at uncovering and analyzing cases where \(x^*\) happens to solve a dual problem for which x is in turn the adjoint arc. But although this is the principal motivation, some side issues have to be explored along the way, and these suggest new approaches even to problems where duality is not at stake [26].
In this paper, the duality theorems proved to allow one to conclude that a sufficient condition for an extremum is an extremal relation for the primal and dual problems. The latter means that if some pair of admissible solutions satisfy this relation, then each of them is a solution to the corresponding (primal and dual) problem. We remark that a significant part of the investigations of Ekeland and Temam [11] for simple variational problems is connected with such problems, and there are similar results for ordinary differential inclusions [7,8,9]. Some duality relations and optimality conditions for an extremum of different control problems with partial differential inclusions can be found in Mahmudov [18]. In particular, the paper [19] establish the duality theorems for Goursat-Darboux-type problems with state constraints.
Due to the significance of its applications, duality theory is one of the main approaches to convex optimality problems, and it is understood differently for various concrete cases [3]. The work [24] develops a duality theory that is used to derive the variational principle for the relevant situation. The author derives the variational principle for minimizing sequences from duality, which allows for numerical characterization of minimizing sequences and approximation of its infimum. The study [23] investigates optimum control problems for partial differential equations with control functions in the Dirichlet boundary conditions under point-wise control constraints. The author changes all notions of dynamic programming to dual space, then develops a dual dynamic approach with a dual Hamilton-Jacobi equation. Moreover, the author formulates sufficient optimality conditions for the concerned problem and defines optimal dual feedback control. The study of economic models, where dual variables can be thought of as prices, is one theoretical application of this duality. It is sometimes helpful in computation methods. The study of duality frequently advances a subject by providing opposing viewpoints, even though it may be pertinent to a particular subclass of problems.
Duality at its fullest, depending essentially on convexity, leads to an enrichment of all aspects of the analysis of optimization problems. In this paper, we construct the dual problem to the convex problem for differential inclusion of considered hyperbolic type, formulate duality results, and investigate the conditions under which such duality relations connect primal and dual problems. The nature of the problems presented and their optimality conditions are different from those that had been previously encountered. The structure of this paper is as follows:
The set-valued analysis notations and fundamental knowledge that will be used throughout this research are introduced in Sect. 2. We provide a brief definition of the set-valued mappings and cones of tangent directions that are necessary for the paper; the definitions and concepts are available in the books [13] and [21].
In Sect. 3, as discrete steps converge to zero when employing the discrete approximation method, the formulation of sufficient conditions is realized by passing the formal limit in the discrete-approximation problem. To avoid any numerical issues, we exclude the discrete approximation method calculations of the original problem (P) in this work. Therefore, we give sufficient conditions of optimality for convex hyperbolic differential inclusions with state constraints and prove the optimality conditions in the Theorem 3.1. To support our main findings for the fundamental convex problem with state constraint, we present optimality conditions for the considered problem in the Corollary 3.2. Finally, we take into account differential inclusion polyhedral problems of the Darboux type and provide the optimality conditions in the Theorem 3.3.
Section 4 is devoted to the duality in the convex problem with the Darboux differential inclusions and state constraints. It is proven that the Euler-Lagrange type inclusion is a dual relation and the optimal values in the primal convex and dual concave problems are equal in Theorems 4.1 and 4.2. Further, it is shown that the duality theorem for the polyhedral problem is proved in Corollary 4.4. Finally, the dual problem to the polyhedral Darboux problem is established, and an example is given to illustrate the key constructs of our approach. Here we get dual results thanks to the dual operations of addition and infimal convolution of convex functions. However, it takes a lot of time and effort to comprehend the computational components of building the duality problem with the help of discrete and discrete approximation problems. In order to avoid deviating from the central theory, we have excluded this.
2 Preliminary problem statements
In this section, first, we recall some basic definitions and properties of set-valued mappings that will be useful in the sequel; then, we state and prove our main results. Books [13] and [21] can provide the reader with further information and debates on significant developments in this area. Let (x, v) be a pair of \(x,v \in {\mathbb {R}}^n\), which denotes the n-dimensional Euclidean space with the norm \(||\cdot ||\) induced by the inner product \(\langle x, v\rangle\).
Assume that \(F:{\mathbb {R}}^{n} \rightrightarrows {\mathbb {R}}^n\) is a set-valued mapping from \({\mathbb {R}}^{n}\) into the set of subsets of \({\mathbb {R}}^{n}\). If the graph of a set-valued mapping F, \(gph F = \{(x,v): v \in F(x)\}\) is a convex subset of \({\mathbb {R}}^{2n}\), then the mapping is said to be convex. If the gphF of a set-valued mapping F is a convex-closed set in \({\mathbb {R}}^{2n}\), then the mapping is convex-closed. If F(x) is a convex set for each \(x \in dom F\), where \(dom F = \{x: F(x)\ne \emptyset \}\), it is convex-valued. For a set-valued mapping F, the Hamiltonian function and argmaximum set are defined as follows:
\(v^*\in {\mathbb {R}}^n\), respectively. If F is convex and \(F(x)=\emptyset\), we set \(H_F(x,v^*)=-\infty\). Clearly, the Hamiltonian function is concave in x and convex in \(v^*\) for the convex set-valued mapping F.
The concepts of the derivative and gradient functions are useful for describing the local properties of differentiable functions. We use subgradients and subdifferentials for convex functions rather than the gradient because they are effective tools for examining the local features of convex functions. The subgradient of a convex function f at \(x_0\), denoted \(x^*\), is defined by the system of inequalities \(f(x)-f(x_0)\ge \langle x^*,x-x_0\rangle \;\forall \;x\). The set of all subgradients at \(x_0\) denoted \(\partial f(x_0)\) is referred to as the subdifferential of f at \(x_0\). If the subdifferential at \(x_0\) consists of only one element, it is equal to the gradient of f at \(x_0\). A function \(\psi =\psi (x,y)\) is called a proper function if it does not assume the value \(-\infty\) and is not identically equal to \(+\infty\). Clearly, \(\psi\) is proper if and only is \(dom\psi \ne \emptyset\) and \(\psi (x,y)\) is finite for \((x,y)\in dom \psi =\{(x,y):\psi (x,y)<+\infty \}\). The support function \(W_A(x^*)\) of a nonempty set A in \({\mathbb {R}}^{n}\) is defined by \(W_A(x^*)=\sup _x\{\langle x,x^* \rangle : x\in A\}, x^*\in {\mathbb {R}}^{n}\). The function \(\psi ^*(\eta ^*)=\sup \limits _{\eta } \{\langle \eta ,\eta ^*\rangle -\psi (\eta )\}\) is called the conjugate of \(\psi\). The conjugate of a function is closed and convex. The operation of infimal convolution \(\oplus\) of proper convex functions \(f_1,f_2\) is defined as follows:
The infimal convolution \((f_1\oplus f_2)\) is said to be exact provided the infimum above is attained for every \(u\in {\mathbb {R}}^{n}\). One has dom\((f_1\oplus f_2)=dom f_1+dom f_2\).
As is well known, the Moreau-Rockafellar-Robinson internal point qualification condition is sufficient to ensure that the infimal convolution of the conjugates of two extended-real-valued convex lower semi-continuous functions defined on a locally convex space is exact and that the subdifferential of the sum of these functions is the sum of their subdifferentials. The following theorem yields a simple characterization in the finite-dimensional setting of the subdifferential of the sum of these functions.
Theorem 2.1
[13](Moreau-Rockafellar) Let \(f_1\), \(f_2\) be a proper convex function and \(f=f_1+f_2\), \(x_0\in dom f_1 \cap dom f_2\). Suppose that either (1) there is a point \(x_1\in dom f_1 \cap dom f_2\) where \(f_1\) is continuous or (2) \(ri dom f_1 \cap ridom f_2\ne \emptyset\). Then
One of the key ideas in convex analysis and extremal theory is the convex cone. The calculation of the dual cone is related to the examination of its properties. Our terminology and notation typically follow Mahmudov’s usage. The definitions of the dual cone and the cone of tangent directions is directly extracted from his book [13]. The set of vectors \(x^*\in {\mathbb {R}}^{n}\) for which \(\langle x,x^*\rangle \ge 0\) holds for all \(x\in K\) is called the dual cone to the cone K and is denoted by \(K^*\). The convex cone \(K_A(z_0)\) is called the cone of tangent directions at a point \(z_0\in A\) to the set A if from \(\overline{z}\in K_A(z_0)\) it follows that \(\overline{z}\) is a tangent vector to the set A at point \(z_0\in A\), i.e., there exists such function \(\gamma (\lambda )\in {\mathbb {R}}^{n}\) such that \(z_0+\lambda \overline{z}+\gamma (\lambda )\in A\) for sufficiently small \(\lambda >0\) and \(\lambda ^{-1}\gamma (\lambda )\rightarrow 0\) as \(\lambda \downarrow 0\).
For a convex mapping F at a point \(w_0=(x_0,v_0)\in gph F\) setting \(\gamma (\lambda )\equiv 0\), we have \(K_{gphF}(w_0)=cone\big [gph F-(w_0)\big ]=\big \{(\overline{x},\overline{v}):\;\overline{x}=\lambda (x-x_0),\overline{v}=\lambda (v-v_0),\) \(\big \},\;\forall \;(x,v)\in gph F.\) For a convex mapping F a set-valued function defined by
is a locally adjoint set-valued mapping (LAM) to F at a point \(w_0\in gphF\), where \(K^*_{gphF} (w_0)\) is the dual to the cone of tangent vectors \(K_{gph F}(w_0)\).
Theorem 2.2
[13](Theorem 2.1.) Let \(F:{\mathbb {R}}^{n} \rightrightarrows {\mathbb {R}}^n\) be a convex set-valued mapping. Then
By Theorem 2.2, all sets \(F^*(v^*;(x,v))\) for every \(v\in F_A(x;v^*)\) are the same and equal to \(-\partial _x \big (-H_F(x,v^*)\big )\), where \(H_F\) is Hamiltonian function and \(F_A\) is an argmaximum set.
A polyhedral convex set in \({\mathbb {R}}^n\) is a set that can be described as the intersection of a finite family of closed half-spaces, that is, as the set of solutions to a finite system of inequalities of the form \(\langle x,x_k^*\rangle \le \beta _k,\;k=1,\ldots ,l\). In particular, if the finite system of inequalities is homogeneous, the set of solutions to this finite system of inequalities is called the polyhedral cone. A polyhedral mapping is defined, the graph of which is the following polyhedral set in \({\mathbb {R}}^{2n}\):
where P and Q, \(m\times n\) dimensional matrices, d is a m-dimensional column-vector. The Hamiltonian function \(H_F(\cdot ,v^*)\) for a polyhedral mapping is closed and its LAM is the step function in the argument (u, v):
Let us denote the following:
It is clear that for every \(x\in {\mathbb {R}}^{n}\)
Here the Hamiltonian function \(H_F(x,y^*)=\sup \limits _y\{\langle y,y^*\rangle : y\in F(x)\}, y^*\in {\mathbb {R}}^{n}\) and for convex set-valued mapping in the case \(F(x)=\emptyset\), we set \(H_F(x,y^*)=-\infty\). Moreover it is easy to see that the function
is a support function of the set gphF taken with a minus sign. It follows that for a fixed \(y^*\) \(M_F(x^*,y^*)=-\big (H_F(\cdot ,y^*)\big )^*(x^*)\).
Let us propose our so-called Darboux differential inclusion problem (P), which involves hyperbolic differential inclusion (2) and state constraints (3). We give our first problem as follows:
Here \(F(\cdot ,x,t):{\mathbb {R}}^{n}\rightrightarrows {\mathbb {R}}^n\), \(F_1(\cdot ,t):{\mathbb {R}}^{n}\rightrightarrows {\mathbb {R}}^n\) and \(F_2(\cdot ,x):{\mathbb {R}}^{n}\rightrightarrows {\mathbb {R}}^n\) are set-valued mappings for all fixed \((x,t)\in D\), \(t\in [0,1]\) and \(x\in [0,1]\), respectively, and \(f(\cdot ,x,t):{\mathbb {R}}^{n}\rightarrow {\mathbb {R}}^n\) is continuous function and \(\Psi :D\rightarrow {\mathbb {R}}^{n}\) is a convex-valued mapping.
The problem is to find a solution \(\widetilde{u}(x,t)\) of the problem (P) satisfying (2) almost everywhere on D and the boundary conditions and state constraints (3) that minimizes \(J[u(\cdot ,\cdot )]\). Here, an admissible solution is understood to be an absolutely continuous function defined on D with an integrable mixed derivative \(u_{xt}(\cdot ,\cdot )\) satisfying (2) almost everywhere on D and the boundary conditions in (3) on [0, 1]. Note that a function \(u(\cdot ,\cdot )\) is said to be absolutely continuous on D if there exist \(g:D\rightarrow {\mathbb {R}}^n\) such that
It is known that the space B of absolutely continuous functions \(u:D\rightarrow {\mathbb {R}}^n\) is a Banach space endowed with the norm
3 Optimality conditions for the Darboux differential problem
We employ difference approximations of partial derivatives and grid functions on a uniform grid by approximating the problem (P) with the hyperbolic differential inclusions to construct sufficient conditions for optimality. In other words, we consider the discrete approximation method to obtain the necessary and sufficient conditions for the optimality of the discrete approximation problem associated with the differential problem. The approach in question necessitates some specific LAM equivalence theorems that arise in discrete and discrete approximation problems and the acquired equivalence theorems allow us to transition between discrete and differential problems. We use the tools of the locally adjoint mappings for convex problems to obtain sufficient optimality conditions for hyperbolic differential inclusions. The derivation of sufficient conditions is carried out by reaching the formal limit as the discrete steps approach zero. To avoid numerical complications, we skip the discrete approximation method calculations of the original problem (P). The discrete approximation method has been applied successfully to a variety of optimal control problems with partial differential inclusions, and the reader can find it in detail in the paper [15].
As indicated in Sect. 1, we need to formulate sufficient conditions of optimality for the problem (P) to construct duality relations. The optimality conditions for our first Darboux-type differential inclusion problem (P) can be stated by the following theorem.
Theorem 3.1
Assume that f is continuous and convex function on the first argument u, and that set-valued mappings \(F, F_1\), and \(F_2\) are also convex. Then for the optimality of a feasible solution \(\widetilde{u}(x,t)\), it is sufficient that there exists an absolutely continuous function \(u^*(x,t)\) on D with integrable mixed derivative \(u_{xt}^*(x,t)\) such that conditions \((i)-(iv)\) hold almost everywhere:
Proof
Since all sets \(F^*\Big (u^*(x,t),(\tilde{u}(x,t),\tilde{u}_{xt}(x,t)),x,t\Big )\) for every \(\widetilde{u}_{xt}(x,t)\in\) \(F_A\Big (\widetilde{u}(x,t);u^*(x,t),x,t\Big )\) are the same and equal to \(-\partial _u \Big ( -H_F \big (\tilde{u}(x,t),u^*(x,t),x,t\big )\Big )\) in the Theorem 2.2, and by Theorem 2.1, the condition (i) of the theorem implies that
By definition of subdifferential, we have for any feasible solution u(x, t)
Further, using the definition of the Hamiltonian function, we obtain that
By integrating this inequality over D, we have
Let us rewrite double integrals on the right-hand side as follows:
and
Then, taking into account formulas (5) and (6) in inequality (4), we find that
Moreover, by condition (iii) of the theorem, we have \(\Big \langle u_{t}(0,t)-\widetilde{u}_{t}(0,t),u^*(0,t)\Big \rangle \le 0\) and\(\Big \langle u_{x}(x,0)-\widetilde{u}_{x}(x,0),u^*(x,0)\Big \rangle \le 0\). Then we write in more convenient form (7) as follows
Furthermore, by condition (ii) of the theorem, we have \(\Big \langle -u_x^*(x,1)\;,\;u(x,1)-\widetilde{u}(x,1)\Big \rangle \ge 0\) and \(\Big \langle -u_t^*(1,t)\;,\;u(1,t)-\widetilde{u}(1,t)\Big \rangle \ge 0\). Then it is clear that
By summing up the inequalities in (8) and (9), we deduce that
or, equivalently
Then, we rewrite the inequality (10) in the form
and it is not hard to see that
Since \(u^*(0,0)=u^*(1,1)=0\), we have
Finally, we obtain that \(J[u(\cdot ,\cdot )]\ge J[\widetilde{u}(\cdot ,\cdot )]\), i.e., \(\widetilde{u}(\cdot ,\cdot )\) is optimal. The proof of the theorem is completed. \(\square\)
Let’s consider the following problem
where \(F(\cdot ,x,t):{\mathbb {R}}^{n}\rightrightarrows {\mathbb {R}}^n\) is set-valued mapping for all fixed \((x,t)\in D\) and \(f(\cdot ,x,t):{\mathbb {R}}^{n}\rightarrow {\mathbb {R}}^n\) is continuous function and \(\Psi :D\rightarrow {\mathbb {R}}^{n}\) is a convex-valued mapping. We note that, in contrast to problem (P), in the problem (13), the set-valued mappings \(F_1(\cdot ,t) \equiv {\mathbb {R}}^{n}\), \(F_2(\cdot ,x) \equiv {\mathbb {R}}^{n}\), and the boundary conditions are \(u(x,0)=0\;,\;u(0,t)=0\;,\; x\in [0,1],\; t\in [0,1]\). In what follows, these subtle changes significantly impact the construction of a dual problem.
Corollary 3.2
Let f be continuous and convex function on the first argument u and suppose that set-valued mapping F is convex. Then for the optimality of a feasible solution \(\widetilde{u}(x,t)\), it is sufficient that there exists an absolutely continuous function \(u^*(x,t)\) on D with integrable mixed derivative \(u_{xt}^*(x,t)\) such that conditions \((a)-(c)\) hold almost everywhere:
Proof
The proof of the corollary is similiar to the one for Theorem 3.1. \(\square\)
Now let’s consider the problem (13) in polyhedral case. Thus, F is a set-valued polyhedral mapping defined as \(F(u)=\{v:\;Pu-Qv\le d\}\) where P and Q, \(m\times n\) dimensional matrices, d is a m-dimensional column-vector.
Theorem 3.3
Assume that f is a proper polyhedral function and that set-valued mapping F is polyhedral. Then, for the optimality of a feasible solution \(\widetilde{u}(x,t)\), it is sufficient that there exists \(\lambda (x,t)\) with integrable mixed derivative \(\lambda _{xt}(x,t)\) such that conditions \((i)-(iii)\) hold almost everywhere:
Proof
For all feasible solutions u(x, t), it follows from the definition of subdifferential that
Then taking into account the structure of F in polyhedral case, it is easy to see that
Hence the inequality (14) can be written as follows
Integrating that inequality over the domain D, we have
Here by necessary simplification, it is clear that
and
Then in view of the inequalities (16) and (17) in the formula (15), we derive that
or, equivalently
By condition (ii) of the Theorem 3.3, we can write the following
Then, the inequality (18) can be converted to the relation as follows
or more convenient form
Since \(Q^*\lambda (0,0)=Q^*\lambda (1,1)=0\) and for all feasible solutions \(u(x,0)=0,\;u(0,t)=0\), we conclude that
Thus, we have proved the desired result. \(\square\)
Now we handle the following hyperbolic Darboux type differential inclusion without the state conditions in the problem (13)
It is clear that, for the optimality of a feasible solution \(\widetilde{u}(x,t)\) in the problem (20), it is sufficient that there exists an absolutely continuous function \(u^*(x,t)\) on D with integrable mixed derivative \(u_{xt}^*(x,t)\) such that conditions \((a')-(c')\) hold almost everywhere:
Now, consider the problem (20) with polyhedral Darboux differential inclusion where \(F(u)=\{v:\;Pu-Qv\le d\}\) where P and Q, \(m\times n\) dimensional matrices, d is a m-dimensional column-vector. We can formulate the optimality conditions for the problem (20) in the polyhedral case.
Corollary 3.4
Suppose that f is a proper polyhedral function and that set-valued mapping F is polyhedral. Then, for the optimality of a feasible solution \(\widetilde{u}(x,t)\), it is sufficient that there exists \(\lambda (x,t)\) with integrable mixed derivative \(\lambda _{xt}(x,t)\) such that conditions \((i)-(iii)\) hold almost everywhere:
Proof
The proof is a simple modification of the proof of Theorem 3.3 and so is omitted. \(\square\)
To illustrate the main constructions of our approach, we give a simple example of the problem (20).
Example 3.5
Let us consider \(Q=[1]\), \(P=[-1]\), \(d=[0]\) and \(f(u)=u+1\) continuously differentiable function according to the problem (20) in polyhedral case. Then, by Corollary 3.4 for the optimality \(\widetilde{u}(x,t)\) to given example, it is sufficient that there exists \(\lambda (x,t)\) such that
Here it can be easily seen that the subdifferential of f is the gradient vector, that is \(\partial _uf(\widetilde{u}(x,t),x,t)=\{1\}\) and in view of relations of (21) we have a boundary value problem for partial differential equation
The general solution of the boundary value problem (22) is
where \(\dfrac{d}{dx} G(x)=CG(x)\) and \(\dfrac{d}{d t} H(t)=-\dfrac{H(t)}{C}\) and C is a non-zero arbitrary constant. By computing these equations we have \(G(x)=C_1e^{Cx}\) and \(H(t)=C_2e^{\frac{-t}{C}}\) where \(C_1,C_2\) are arbitrary constant. Therefore, we have
By using the boundary conditions, we find that \(\lambda (x,t)=1-e^{x-t}\). Now, since \(\lambda (x,t)\ne 0\), \((x,t)\in D,x\ne t\) from the third formula of (21), we deduce that \(\widetilde{u}(x,t)+\widetilde{u}_{xt}(x,t)=0\). By similar way, using boundary conditions \(\widetilde{u}(x,0)=\widetilde{u}(0,t)=0\), we have the solution of the stated problem in (20) with polyhedral case \(\widetilde{u}(x,t)=0\). Then its value is
4 On duality in the Darboux differential problem with state constraints
Our approach to establishing duality results for differential problems (13) and (20) with state constraints is based on the passage to the formal limit in the dual problem for the discrete approximate problem. To formulate the dual problems for the differential problems, we use the duality theorems of operations of addition and infimal convolution of convex functions. However, we skip it and just present dual problems for continuous problems (13) and (20) to prevent laborious and tiresome calculations. Moreover, it is demonstrated how the duality relations relate primal and dual problems to one another. An extremal relation for the primal and dual problems is a sufficient condition for an extremum according to the established duality theorems. It means that if some pair of feasible solutions \(u(\cdot )\) and \(u^*(\cdot )\) satisfy duality relations, then \(u(\cdot )\) and \(u^*(\cdot )\) are solutions of the primal and dual problem, respectively.
The maximization problem will be the dual problem to the continuous convex problem (13):
where
Here, the problem (23) is called the dual problem to the primal convex problem (13). It is assumed that \(\varphi ^*,\phi ^*\) are absolutely continuous function on D and \(u^*\) is an absolutely continuous function on D having an integrable mixed partial derivative \(u^*_{xt}(\cdot ,\cdot )\).
Theorem 4.1
The inequality \(J[u(x,t)]\ge J_*\Big ( u^*(x,t),\varphi ^*(x,t),\phi ^*(x,t)\Big )\) holds for all feasible solutions u(x, t) and \(\{u^*(x,t),\varphi ^*(x,t),\phi ^*(x,t)\}\) of the primal problem and dual problem.
Proof
It is clear from the definition of function \(M_F\) that
Moreover, by definitions of support function and conjugate function, we deduce that
and
Then, from the relation (24)-(26) we derive that
and
We rewrite the relation (27) in the form
Integrating the inequality (29) over the domain D and the inequality (28) over the interval [0, 1], and adding them, we have
On the other hand it is easy to see that
and so,
By substituting equality (31) in the inequality (30), we have
or in more convenient form
Then we rewrite the inequality (32) as follows
From the inequality above we get the following
Since the solutions u(x, t) and \(\{u^*(x,t),\varphi ^*(x,t),\phi ^*(x,t)\}\) are feasible solutions of the primal problem and dual problem, respectively, and \(u(x,0)=0\;,\;u(0,t)=0\), \(u^*(0,0)=u^*(1,1)=0\), we obtain that
This concludes the proof. \(\square\)
Theorem 4.2
Suppose that the function \(\widetilde{u}(x,t)\) satisfy the conditions \((a)-(c)\) of the Corollary 3.2, and \(\{u^*(x,t),\varphi ^*(x,t),\phi ^*(x,t)\}\) hold the relation \(W_{\Psi (x,t)}(\varphi ^*(x,t))= \Big \langle \widetilde{u}(x,t)\;,\;\varphi ^*(x,t)\Big \rangle\) and \(\phi ^*(x,t)\in \partial f(\widetilde{u}(x,t),x,t)\). Then \(\widetilde{u}(x,t)\) and \(\{u^*(x,t),\varphi ^*(x,t),\phi ^*(x,t)\}\) are optimal solutions of the primal and dual problems, respectively and their values are equal.
Proof
Suppose that \(\widetilde{u}(x,t)\) is a solution of the primal problem. Now let’s prove that \(\{u^*(x,t),\varphi ^*(x,t),\phi ^*(x,t)\}\) is an optimal solution. By definition of LAM, the condition (a) of the Corollary 3.2, we have
\(v\in F(u)\) and \(W_{\Psi (x,t)}(\varphi ^*(x,t))= \Big \langle \widetilde{u}(x,t)\;,\;\varphi ^*(x,t)\Big \rangle\). This means that
where \(dom F=\{(u^*,-v^*):\;M_F(u^*,v^*)>-\infty \}\). Further, since \(\partial f(u,x,t)\subset dom f^*(\cdot ,x,t)\), it is clear that \(\phi ^*(x,t)\in dom f^*(\cdot ,x,t)\). Thus the triplet \(\{u^*(x,t),\varphi ^*(x,t),\phi ^*(x,t)\}\) is a feasible solution. It remains to show that it is optimal. On the other hand, we get
By conditions (b) and (c) of the Corollary 3.2, we can write
Since \(\phi ^*(x,t)\in \partial f(\widetilde{u}(x,t),x,t)\), we have
Then in view of equations (36), (37) and (38), having instead of the inequalities in equations (24), (25) and (26), and following the proof of Theorem 4.1, we derive that
and so \(\{u^*(x,t),\varphi ^*(x,t),\phi ^*(x,t)\}\) is an optimal solution. The proof is complete. \(\square\)
Now, we illustrate the applicability of Theorems 4.1 and 4.2 for the problem (20). Based on duality theorems for the problem (13), we can obtain the following duality result.
Corollary 4.3
The dual problem to the continuous convex problem (20) is as follows:
where
Corollary 4.4
Dual problem to the primal continuous problem (20) in polyhedral case is as follows:
Proof
Indeed, it can be easily seen that, denoting \(w=(x,y)\in {\mathbb {R}}^{2n}\), \(w^*=(\xi ^*,-\eta ^*)\in {\mathbb {R}}^{2n}\) we have a linear programming problem
where \(C=[P:-Q]\) is \(m\times 2n\) block matrix. Then for a solution \(\widetilde{w}=(\widetilde{x},\widetilde{y})\) of the problem (42) there exists m-dimensional vector \(\lambda \ge 0\) such that \(w^*=-C^*\lambda\), \(\langle A\widetilde{x}-B\widetilde{y}-d,\lambda \rangle =0\). And vice versa, if these conditions are satisfied, then \(\widetilde{w}=(\widetilde{x},\widetilde{y})\) is a solution of the problem (42). Therefore \(w^*=-C^*\lambda\) implies that \(\xi ^*=-P^*\lambda\), \(\eta ^*=-Q^*\lambda\), \(\lambda \ge 0\). Thus we find that
Therefore by using the relation \(u_{xt}^*(x,t)+\phi ^*(x,t)=\xi ^*(x,t)=-P^*\lambda (x,t)\) and \(u^*(x,t)=\eta ^*(x,t)=-Q^*\lambda (x,t)\), we derive the following dual problem, to the primal continuous polyhedral problem (20).
\(\square\)
Corollary 4.5
Suppose that \(\widetilde{u}(x,t)\) is an optimal solution of problem (20) with polyhedral differential inclusions. Then, in order for \(\lambda (x,t)\) to be an optimal solution to the dual problem (44), it is necessary and sufficient that the conditions of Corollary 3.4 are satisfied. Moreover, the optimal values in the primal problem (20) with polyhedral differential inclusions and the dual problem (44) are equal.
Example 4.6
Let us now construct the dual problem to the polyhedral differential problem in Example 3.5 and calculate its value. Taking into account the dual problem in the formula (44), we write the dual problem of the problem in Example 3.5 as follows:
By definition of the conjugate function, we find that
Therefore, we obtain that \(f^*\Big ( \lambda _{xt}(x,t)+\lambda (x,t)\;,\;x,t\Big )=f^*(1)=-1\). Moreover, since \(\widetilde{u}(x,t)=0\) in Example 3.5, we derive that dual problem as follows
Consequently, we find the value of the dual problem in Example 3.5 is 1. Thus accordingly to Corollary 4.5, we have showed that under the conditions of Corollary 3.4, the optimal values of primal (20) and dual problem (45) coincide. We prove that if \(\eta =\iint \limits _{D}f(\widetilde{u}(x,t),x,t)dx dt=1\) and \(\eta ^*=1\) are the values of polyhedral differential problem in Example 3.5 and its dual problem, respectively, then \(\eta =\eta ^*=1\) for an optimal pair primal and dual problems.
References
Adly, S., Haddad, T., Thibault, L.: Convex sweeping process in the framework of measure differential inclusions and evolution variational inequalities. Math. Program. 148, 5–47 (2014)
Benchora, M., Ntouyas, S.K.: The method of lower and upper solutions to the Darboux problem for partial differential inclusions. Miskolc Math. Notes. 4(2), 81–88 (2003)
Bhatia, M.: Higher order duality in vector optimization over cones. Optim. Lett. 6, 17–30 (2012)
Bomze, I.M.: Copositivity for second-order optimality conditions in general smooth optimization problems. Optim. 65(4), 779–795 (2016)
Cernea, A.: Some second-order necessary conditions for nonconvex hyperbolic differential inclusion problem. J. Math. Anal. Appl. 253(2), 616–639 (2001)
Demir Sağlam, S.: Polyhedral optimization of discrete and partial differential inclusions of parabolic type. Optimization 72(8), 2175–2197 (2023)
Demir Sağlam, S., Mahmudov, E.N.: The Lagrange problem for differential inclusions with boundary value conditions and duality. Pacific J. Optim. 17(2), 209–225 (2021)
Demir Sağlam, S., Mahmudov, E.N.: Duality problems with second-order polyhedral discrete and differential inclusions. Bull. Iran. Math. Soc. 48(2), 537–562 (2022)
Demir Sağlam, S., Mahmudov, E.N.: On duality in convex optimization of second-order differential inclusions with periodic boundary conditions. Hacet. J. Math. Stat. 51(6), 1588–1599 (2022)
Helal, M.: Darboux problem for perturbed hyperbolic differential inclusions with fractional order. Discuss. Math. Differ. Incl. Control Optim. 39, 53–68 (2019)
Ekeland, I., Temam, R.: Convex Analysis and Variational Problems. North-Holland, Amsterdam (1976)
Jackson, R.: Optimization problems in a class of systems described by hyperbolic partial differential equations part II. A maximum principle. Int. J. Control. 4(6), 585–598 (2007)
Mahmudov, E.N.: Approximation and Optimization of Discrete and Differential Inclusions. Elsevier, Boston (2011)
Mahmudov, E.N.: The optimality principle for discrete and first order partial differential inclusions. J. Math. Anal. Appl. 308, 605–619 (2006)
Mahmudov, E.N.: Locally adjoint mappings and optimization of the first boundary-value problem for hyperbolic type discrete and differential inclusions. Nonlin. Anal. 67(10), 2966–2981 (2007)
Mahmudov, E.N.: Optimization of boundary value problems for higher order differential inclusions and duality. Optim. Lett. 16(2), 695–712 (2022)
Mahmudov, E.N.: Optimal control of Cauchy problem for first-order discrete and partial differential inclusions. J. Dyn. Control Syst. 15, 587 (2009)
Mahmudov, E.N.: Duality in the problems of optimal control described by first-order partial differential inclusions. Optimization. 59(4), 589–599 (2010)
Mahmudov, E.N.: On duality in problems of optimal control described by convex differential inclusions of Goursat-Darboux type. J. Math. Anal. Appl. 307, 628–640 (2005)
Marinkovic, B.: Optimality conditions for discrete calculus of variations problems. Optim. Lett. 2, 309–318 (2008)
Mordukhovich, B.S.: Variational Analysis and Generalized Differentiation, Vols.I and II. Springer, Springer-Verlag Berlin Heidelberg (2006)
Mordukhovich, B.S., Raymond, J.P.: Neumann boundary control of hyperbolic equations with pointwise state constraints. SIAM J. Control Optim. 43(4), 1354–1372 (2004)
Nowakowsk, A.: Sufficient optimality conditions for multivariable control problems. Discuss. Math. Differ. Incl. Control Optim. 27, 135–150 (2007)
Orpel, A.: On solutions of the Dirichlet problem for a class of partial differential inclusions with superlinear nonlinearities. Numer. Funct. Anal. Optim. 23(3–4), 367–381 (2002)
Penot, J.P.: Directionally limiting subdifferentials and second-order optimality conditions. Optim. Lett. 8, 1191–1200 (2014)
Rockafellar, R.T., Wolenski, P.R.: Convexity in Hamilton-Jacobi theory I: dynamics and duality. SIAM J. Control Optim. 39, 1323–1350 (2000)
Wang, Y., Ma, F., Zhang, J.: A nonsmooth L-M method for solving the generalized nonlinear complementarity problem over a polyhedral cone. Appl. Math. Optim. 52, 73–92 (2005)
Funding
Open access funding provided by the Scientific and Technological Research Council of Türkiye (TÜBİTAK).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Demir Sağlam, S. Duality in the problems of optimal control described by Darboux-type differential inclusions. Optim Lett (2024). https://doi.org/10.1007/s11590-023-02088-4
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11590-023-02088-4