Abstract
The present studies concern properties of set-valued Young integrals generated by families of \(\beta \)-Hölder functions and differential inclusions governed by such a type of integrals. These integrals differ from classical set-valued integrals of set-valued functions constructed in an Aumann’s sense. Integrals and inclusions considered in the manuscript contain as a particular case set-valued integrals and inclusions driven by a fractional Brownian motion. Our study is focused on topological properties of solutions to Young differential inclusions. In particular, we show that the set of all solutions is compact in the space of continuous functions. We also study its dependence on initial conditions as well as properties of reachable sets of solutions. The results obtained in the paper are finally applied to some optimality problems driven by Young differential inclusions. The properties of optimal solutions and their reachable sets are discussed.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Differential inclusions employ on an approach that has been developed due to investigations of dynamic systems having velocities not uniquely determined by the state of the system. The theory of differential inclusions has been explored in connection with optimal control problems and it constitutes nowadays an important branch of set-valued analysis (see e.g. monographs [2, 3, 5, 12, 13, 19, 20, 38] and references therein). In particular, the existence of solutions and properties of solutions sets have been studed, see e.g., [8, 9, 37, 39]. Stochastic differential inclusions form a new direction and an extension of the differential inclusions theory. They generalize deterministic differential inclusions as well as single valued stochastic differential equations. Such investigations were initiated independently by Kisielewicz in [22], Hiai in [18] and continued by many authors, among others in [1, 4, 21,22,23, 25, 26, 28, 31, 34]. For a collected presentation of the theory we refer the reader to the monographs [21, 24].
On the other hand, the Young type integrals were introduced more than eight decades ago in [40]. They have been developed and widely used in the theory of differential equations with rough paths by many authors, see e.g., [7, 10, 14,15,16, 27, 29, 35] and references therein. The theory of rough paths establishes a convenient tool to extend stochastic calculus with respect to a large class of integrators out of range of semimartringales. Namely, one can consider a stochastic integration with respect to integrators such as the Mandelbrot fractional Brownian motion which has Hölder continuous sample paths [30]. As already mentioned, control and optimal control problems have inspired an intensive expansion of the differential and stochastic set-valued inclusions theory. Thus, it seems reasonable to investigate also differential inclusions driven by a fractional Brownian motion and Young type integrals as well. Recently, in [6, 11, 32, 33] the authors studied properties of set-valued Young type integrals and the existence of solutions of differential inclusions with such integrals. Once having conditions assuring the existence of solutions, it is natural to ask about their properties. To the best of our knowledge, this topic has not been investigated yet. The paper concerns both the properties of set-valued Young integrals and topological properties of solutions of Young differential inclusions. The latter constitutes the main novelty. In particular, we show that the set of all solutions is compact in the space of continuous functions. We also study its dependence on initial conditions as well as properties of reachable sets of solutions with respect to time. The results obtained in the paper are applied to a similar analysis concerning optimal solutions and their reachable sets. The continuity type dependence of optimal solutions is shown both with respect to initial conditions and functionals determining the optimality criterion.
The paper is organized as follows. In Sect. 2, we present basic notions and auxiliary properties concerning Hölder continuous functions and Young integrals in a single-valued case. Section 3 deals with the properties of set-valued Young type integrals which are based on the bounded and closed sets of integrands. We shall give a number of properties of such set-valued integrals. Some of them are natural set-valued generalizations of those known from the single-valued Young integral theory. In Sect. 4, we present the existence and main topological properties of the set of all solutions of Young differential inclusions and their reachable sets. The last section deals with applications to the analysis of sets of solutions of differential inclusions being optimal with respect to chosen functionals.
2 Basic notions and auxiliary results
Let \((X,\Vert \cdot \Vert )\) be a Banach space. Let \(T>0\) and let \(\mathcal {C}(X) \) denote the space of continuous functions \(f:[0,T]\rightarrow X\) with a sup-norm
Let \(\beta \in (0,1]\). By \(\mathcal {C}^{\beta }( X) \) we denote the space of \(\beta \)-Hölder continuous functions \(f:[0,T]\rightarrow X\) with a finite norm
where
It can be shown that \(( \mathcal {C}^{\beta }( X),\Vert \cdot \Vert _{\beta }) \) is a Banach space. For a nonempty set \(\mathcal {H}\subset \mathcal {C}( X) \) we shall use symbols \(cl_{\mathcal {C}}( \mathcal {H}) \) and \(cl_{\mathcal {C}^{\beta }}( \mathcal {H}) \) to denote the closure of \(\mathcal {H}\) with respect to the norm \(\Vert \cdot \Vert _{\infty }\) and \(\Vert \cdot \Vert _{\beta }\), respectively. Let \(0<\beta ^1<\beta \le 1\). Then it holds \(M_{\beta ^1}(\cdot )\le T^{\beta -\beta ^1}M_{\beta }(\cdot )\) and thus \(\Vert \cdot \Vert _{\beta ^1}\le ( 1+T^{\beta -\beta ^1}) \Vert \cdot \Vert _{\beta }\). Hence \(cl_{\mathcal {C}^{\beta }}( \mathcal {H}) \subset cl_{\mathcal {C}^{\beta ^1}}( \mathcal {H}) \subset cl_{\mathcal {C}}( \mathcal {H}) \). We recall the following results needed in the sequel (c.f. Lemma 5.29 and Proposition 5.30 in [14]).
Proposition 2.1
Let \(( f_{n}) _{n\ge 1}\subset \mathcal {C}({R}^{d}) \) and \(\Vert f_{n}-f\Vert _{\infty }\rightarrow 0\). Assume
\(\sup _{n\ge 1}M_{\beta }( f_{n}) <\infty .\) Then \(f_{n}\rightarrow f\) in \(\mathcal {C}^{\beta ^1}({R}^{d}) \) for any \(\beta ^1<\beta \).
Proposition 2.2
Suppose \(( f_{n}) _{n\ge 1}\) is a bounded sequence in \(\mathcal {C}( {R}^{d}) \) and
\(\sup _{n\ge 1}M_{\beta }( f_{n}) <\infty \). Then, for any \(\beta ^1<\beta \) there exists a subsequence \(( f_{n_{k}}) _{k\ge 1}\) of \(( f_{n}) _{n\ge 1}\) which is convergent in \(\mathcal {C}^{\beta ^1}( {R}^{d}) \) to some \(f\in \mathcal {C}^{\beta }( {R}^{d}) \).
By these properties we have:
Proposition 2.3
Let \(\mathcal {H}\) be a nonempty and bounded set in \(\mathcal {C}^{\beta }( {R}^{d}) \). Then \(cl_{\mathcal {C}}(\mathcal {H}) \) \(=cl_{\mathcal {C}^{\beta ^1}}( \mathcal {H}) \) for any \(0<\beta ^1<\beta \le 1\). Moreover, \(cl_{\mathcal {C}}( \mathcal {H}) \) is a bounded subset in \(\mathcal {C}^{\beta }( {R}^{d}) \).
Proof
Let \(f\in cl_{\mathcal {C}}( \mathcal {H}) \) be arbitrary. Then there exists a sequence \(( f_{n}) _{n\ge 1}\subset \mathcal {H}\) such that \(\Vert f_{n}-f\Vert _{\infty }\rightarrow \infty \) as \( n\rightarrow \infty \). Then \(f_{n}\rightarrow f\) in \(\mathcal {C}^{\beta ^1}( {R}^{d}) \) for any \(\beta ^1<\beta \) by Proposition 2.1. Thus \(f\in cl_{\mathcal {C}^{\beta ^1}}( \mathcal {H}) \). Hence \(cl_{\mathcal {C}}(\mathcal {H}) =cl_{\mathcal {C}^{\beta ^1}}( \mathcal {H}) \). Let \(P_{1}:=\sup _{h\in \mathcal {H}}\Vert h\Vert _{\infty }\), \(P_{2}:=\sup _{h\in \mathcal {H}}M_{\beta }(h)\) and \( P=P_{1}+P_{2} \). Since \(\mathcal {H}\) is a nonempty and bounded set in \(\mathcal {C}^{\beta }( {R}^{d}) \) then \(P_{1}\) and \(P_{2}\) are finite, and \(\sup _{n\ge 1}\Vert f_{n}\Vert _{\beta }\le P\). On the other hand we have
for any \(0\le s<t\le T\). Thus \(M_{\beta }(f)\le P_{2}\). Since \(\Vert f\Vert _{\infty }=\lim _{k\rightarrow \infty }\Vert f_{n}\Vert _{\infty }\), we obtain \(\Vert f\Vert _{\infty }\le P_{1}\). Thus \(\Vert f\Vert _{\beta }\le P\) what proves that \(cl_{\mathcal {C}}( \mathcal {H}) \) is a bounded subset in \(\mathcal {C}^{\beta }( {R}^{d}) \). \(\square \)
In order to define the Young integral, let \(g\in \mathcal {C} ^{\alpha }( {R}^{1}) \) and \(f\in \mathcal {C}^{\beta }( {R}^{d}) \) when \(\alpha ,\beta \in (0,1]\) with \(\alpha +\beta >1\) be given functions. For \(0\le s<t\le T\), let \(f_{s+}(\tau )=( f(\tau )-f(s+)) \textbf{1}_{( s,t) }( \tau ) \) and \(f_{t-}(\tau )=( f(\tau )-f(t-)) \textbf{1} _{( s,t) }( \tau ) \) where \(\textbf{1}_{( s,t) }\) denotes a characteristic function of the interval (s, t) . The right-sided and left-sided fractional derivatives of order \(0<\rho <1\) for the function \(f:[0,T]\rightarrow {R}^{1}\) and \(0\le s<t\le T\) are defined by
and
for \(\tau \in [s,t]\). For the partition \(\pi :0=t_{0}<t_{1}<...<t_{m}=T\) of the interval [0, T] we consider the Riemann sum of f with respect to g
Let \(|\pi |:=\max \{t_{i}-t_{i-1}:1\le i\le m-1\}\). Then one can prove the following result (see e.g. [36]).
Proposition 2.4
Suppose that \(g\in \mathcal {C}^{\alpha }( {R}^{1}) \) and \(f\in \mathcal {C}^{\beta }( {R}^{d}) \). Then
exists and
for \(\rho \in (1-\alpha ,\beta )\) and \(0\le s<t\le T\). Moreover, the following inequality
holds for every \(0\le s<t\le T\) where \(C(\alpha ,\beta )\) depends only on \(\alpha \) and \(\beta \).
Due to Proposition 2.4 we get immediately
for every \(0\le s<t\le T\) and hence
Indeed, let \(s,t\in [0,T]\), \(s<t\) be fixed. Then by (2.5) we get
which gives (2.6) and (2.7). The following result is a variant of Lemma 1 from [32]. It will be used in next section.
Lemma 2.5
Let \(g\in \mathcal {C}^{\alpha }( {R}^{1}) \). Then for every \(\rho \in (1-\alpha ,\beta )\) there exists a positive constant \(C(\rho )\) such that for every \(f_{1},f_{2}\in \mathcal {C}^{\beta }( {R}^{d}) \), \(0\le s<t\le T\) and \(\theta \in (0,1]\) it holds
where \(C(\rho )\) is given by formula
with
Proof
The proof can be given as a slight modification of the proof of Lemma 1 of [32], and therefore, we omit it. \(\square \)
At the end of this section we recall the notion of a space of \(\beta \)-Hölder continuous set-valued functions. Denote by \(Cl_{b}( X) \) and Comp(X) the families of all nonempty, closed and bounded and nonempty and compact subsets of a Banach space X, respectively. The Hausdorff metric \(H_{X}\) in \(Cl_{b}( {X}) \) is defined by
where \(\overline{H}_{X}( B,C) =\sup _{b\in B}\textrm{dist}_{X}( b,C) \) and \(\textrm{dist}_{X}(b,C) =\inf _{c\in C}\Vert b-c \Vert _{X}\) (see e.g., [19] for details). We shall use the notation
For \(B,C,D,E\in Comp( X) \) it holds
where \(B+C:=\left\{ b+c:b\in B,c\in C\right\} \) denotes the Minkowski sum of B and C. Let \(F:[0,T]\rightarrow Comp({X)}\) be a set-valued function. It is called \(\beta \)-Hölder continuous if
Then
By \(\mathcal {C}^{\beta }( Comp({X)}) \) we denote the space of such set-valued functions with a norm given by formula \(\Vert F\Vert _{{\beta }}:=\Vert F\Vert _{{\infty }}+M_{\beta }(F).\)
3 Set-valued Young integral
Let \(\mathcal {H}\) be a nonempty subset of \(\mathcal {C}^{\beta }( R^{d}) \). Similarly as in the previous part we assume that \(g\in \mathcal {C}^{\alpha }( R^{1}) \) where \(\alpha ,\beta \in (0,1]\) with \(\alpha +\beta >1\). Define a set-valued Young integral \(\int _{0}^{t}\mathcal {H}dg\) of \(\mathcal {H}\) with respect to g over the interval \([s,t]\subset [0,T]\), i.e.,
It is easy to see that \(\int _{s}^{t}\mathcal {H}dg\) is a convex subset of \(R^{d}\) if \(\mathcal {H}\) is convex. Moreover, we have the following properties of such set-valued Young integral.
Proposition 3.1
Let \(0\le s<t\le T\). Let \(\mathcal {H}\) be a nonempty subset of \(\mathcal {C}^{\beta }( R^{d}) \). Then it holds
Proof
Indeed, let \(0\le s<t\le T\) and let
\(f\in \mathcal {H}\) be chosen arbitrarily. Then by (2.6) we get
Hence, taking the supremum over
\(f\in \mathcal {H}\), we get inequalities (3.2). \(\square \)
By (3.2) it follows that \(\int _{s}^{t}\mathcal {H}dg\) is a bounded subset of \(R^{d}\) for every nonempty and bounded set \(\mathcal {H}\subset \mathcal {C}^{\beta }( R^{d}) \). If \(\mathcal {H}\subset \mathcal {C}^{\beta }( R^{d}) \) and \(\beta ^1<\beta \), then \(\mathcal {H}\subset \mathcal {C}^{\beta ^1}( R^{d}) \) and one can consider a set-valued Young integral \(\int _{s}^{t}cl_{\mathcal {C}^{\beta ^1}}(\mathcal {H)}dg\) as long as \(\alpha +\beta ^1>1\). If the set \(\mathcal {H}\) is bounded in \(\mathcal {C}^{\beta }( R^{d}) \), then by Proposition 2.3 it follows that \(cl_{\mathcal {C}}(\mathcal {H}) \) is a bounded set in \(\mathcal {C}^{\beta }( R^{d}) \) and \(cl_{\mathcal {C}^{\beta ^1}}(\mathcal {H)}=cl_{\mathcal {C}}(\mathcal {H)\subset C}^{\beta }( R^{d}) \). Thus \(\int _{s}^{t}cl_{\mathcal {C}^{\beta ^1}}(\mathcal {H)}dg=\int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H)}dg\) for any \(0<\beta ^1<\beta \). Therefore, we can take a closure of \(\mathcal {H}\subset \mathcal {C}^{\beta }( R^{d}) \) in \(\mathcal {C}( R^{d}) \). In this case we have the following result.
Corollary 3.2
Let \(\mathcal {H}\) be a nonempty and bounded set in \(\mathcal {C}^{\beta }( R^{d}) \). Then
is a bounded and closed subset of \(R^{d}\) for any \(0\le s<t\le T\). Thus the set \(\int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H})dg\) is compact in \(R^{d}\).
Proof
By Proposition 2.3, \(cl_{\mathcal {C}}(\mathcal {H)}\) is bounded in \(\mathcal {C}^{\beta }( R^{d}) \). Thus \(\int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H})dg\) is a bounded subset in \(R^{d}\), by (3.2). It remains to prove the closedness of \(\int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H})dg\) in \(R^{d}\). Let \(z_{n}\in \int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H})dg\) and suppose \(z_{n}\rightarrow z\) in \(R^{d}\). Then there exists a sequence \(( f_{n})\subset cl_{\mathcal {C}}(\mathcal {H})\) (depending on s and t) such that \(z_{n}=\int _{s}^{t}f_{n}dg \) for \(n\ge 1\). Since a sequence \((f_{n}) \) is bounded in \(\mathcal {C}^{\beta }( R^{d}) \), then due to Proposition 2.2 there exists a subsequence \(( f_{n_{k}}) \) of \(( f_{n}) \) and \(f\in \mathcal {C}^{\beta }( R^{d}) \) such that \(f_{n_{k}}\rightarrow f\) in \(\mathcal {C}^{\beta ^1}( {R}^{d}) \) for any \(\beta ^1<\beta \). Thus \(\Vert f_{n_{k}}-f\Vert _{\infty }\rightarrow 0\) and \(f\in cl_{\mathcal {C}}(\mathcal {H})\). Since \(\alpha +\beta >1\), let us choose \(\beta ^1<\beta \) and such that \(\alpha +\beta ^1>1\). Since simultaneously \(\ f\in \mathcal {C}^{\beta ^1}( R^{d}) \), we have by (2.6)
Since \(f_{n_{k}}\rightarrow f\) in \(\mathcal {C}^{\beta ^1}( R^{d}) \), then the right side of the above inequality tends to zero as \(k\rightarrow \infty \). Thus \(z_{n_{k}}\rightarrow \int _{s}^{t}fdg\) in \(R^{d}\) and \(z=\) \(\int _{s}^{t}fdg\in \int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H)}dg\) what proves the closedness of \(\int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H)}dg\) in \(R^{d}\). \(\square \)
It is worth to note the following natural connection between integrals \(\int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H)}dg\) and \(\int _{s}^{t}\mathcal {H}dg\).
Corollary 3.3
Under the same assumptions as in Corollary 3.2 it holds
Proof
By Corollary 3.2 we have
Let \(z\in \int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H)}dg\). Hence \(z=\int _{s}^{t}fdg\) for some \(f\in cl_{\mathcal {C}}(\mathcal {H)}\). Then, there exists a sequence \(( f_{n}) _{n\ge 1}\subset \mathcal {H}\) and such that \(\Vert f_{n}-f\Vert _{\infty }\rightarrow \infty \). Similarly as in the proof of Corollary 3.2, let us choose \(\beta ^1<\beta \) and such that \(\beta ^1+\alpha >1\). Thanks to Proposition 2.1 we deduce \(f_{n}\rightarrow f\) in \(\mathcal {C}^{\beta ^1}( R^{d}) \) for any \(\beta ^1<\beta \) and by (2.6) we obtain
Thus \(\Vert \int _{s}^{t}f_{n}dg-z\Vert _{R^{d}}\rightarrow 0\) as \(n\rightarrow \infty \), and we get
\(\square \)
Let us note that a set-valued Young integral can be treated as a set-valued mapping itself. It depends on t, \(\mathcal {H}\) and g. Below, we present the nature of these dependences.
Proposition 3.4
Suppose that \(g\in \mathcal {C}^{\alpha }( R^{1}) \) and \(\mathcal {H\subset C}^{\beta }( R^{d}) \) is a nonempty and bounded set. Then for every \(s\in [ 0,T)\) the set-valued function
is \(\alpha \)-Hölder continuous and
Proof
It follows by Corollary 3.3 that for any \(s\le \tau <t\le T\)
Thus
Therefore, it is sufficient to show that
for all \(s\le \tau <t\le T\). Let \(z_{1}\in \int _{s}^{t}\mathcal {H}dg\). Then there exists \(f_{1}\in \mathcal {H}\) such that \(z_{1}=\int _{s}^{t}f_{1}dg \). Let \(f_{2}\in \mathcal {H}\) be chosen arbitrarily. By (2.6) we have
Let us notice that \(\sup _{f\in \mathcal {H}}\Vert f\Vert _{\beta }<\infty \) because \(\mathcal {H}\subset \mathcal {C}^{\beta }( R^{d}) \) is a nonempty and bounded set. Thus
Since last term of the above inequality vanishes, then we get
The same inequality holds for \(\bar{H}_{R^{d}}\big ( \int _{s}^{\tau }\mathcal {H}dg,\int _{s}^{t}\mathcal {H}dg\big ) \) by symmetry. Thus
for every \(s\in [0,T)\). \(\square \)
By \(H_{\infty }\) we denote the Hausdorff distance in the space \(Cl_{b}(\mathcal {C}( R^{d}) ) \).
Theorem 3.5
For every \(\rho \in (1-\alpha ,\beta )\) there exists a positive constant \(C(\rho )\) such that for every \(\theta \in (0,1]\), \(0\le s<t\le T\) and every nonempty and bounded \(\mathcal {H}_{1},\mathcal {H}_{2}\subset \mathcal {C}^{\beta }( R^{d}) \), it holds
Proof
Let \(f_{1}\in \mathcal {H}_{1}\), \(f_{2}\in \mathcal {H}_{2}\), \(\rho \in (1-\alpha ,\beta )\), \(t\in [0,T]\) and \(\theta \in (0,1]\) be given. By Lemma 2.5 we have
where \(C(\rho )\) is given by formula (2.9).
Thus
Hence
In a similar way one can prove the same estimation for second Hausdorff submetric \(\bar{H}_{R^{d}}\left( \int _{s}^{t}\mathcal {H}_{2}dg,\int _{s}^{t}\mathcal {H}_{1}dg\right) \). This completes the proof because by Corollary 3.3 we get
\(\square \)
It also true that a set-valued Young integral defined by formula (3.1) is Lipschitz continuous with respect to integrators from \(\mathcal {C}^{\alpha }( R^{1}) \).
Proposition 3.6
Let \(\mathcal {H}\) be a nonempty and bounded set in \(\mathcal {C}^{\beta }( R^{d}) \). Then for each \(g_{1},g_{2}\in \mathcal {C}^{\alpha }( R^{1}) \) and \(0\le s<t\le T\), it holds
where \(\hat{C}(\alpha ,\beta ,T,\mathcal {H})\) is some constant depending on \(\alpha ,\beta ,T\) and the set \(\mathcal {H}\).
Proof
Let \(z_{1}\in \int _{s}^{t}\mathcal {H}dg_{1}\). Then \(z_{1}=\int _{s}^{t}f_{1}dg_{1}\) for some \(f_{1}\in \mathcal {H}\). Let \(f_{2}\in \mathcal {H}\) be chosen arbitrarily. Then by (2.6)
Let
Since \(\mathcal {H}\) is bounded in \(\mathcal {C}^{\beta }( R^{d}) \), \(\hat{C}(\alpha ,\beta ,T,\mathcal {H})\) is finite. Thus
Since second term on the right side of inequality above equals to zero, it follows by (3.5) that
The same holds for \(\bar{H}_{R^{d}}( \int _{s}^{t}\mathcal {H}dg_{2},\int _{s}^{t}\mathcal {H}dg_{1}) \) by symmetry. Thus
and we obtain (3.3) by Corollary 3.3. \(\square \)
Using Theorem 3.5 and Proposition 3.6 we get the following continuity property of set-valued Young integral.
Corollary 3.7
Let \((\mathcal {H}_{n}) _{n\ge 1}\) be a bounded sequence of nonempty sets in \(\mathcal {C}^{\beta }(\!R^{d}).\) Let \((g_{n}) _{n\ge 1}\subset \mathcal {C}^{\alpha }( R^{1}) \). Suppose that for some bounded \(\mathcal {H\subset C}^{\beta }( R^{d}) \) and \(g\in \mathcal {C}^{\alpha }(R^{1}) \) it holds
Then for each \(0\le s<t\le T\)
and thus
Proof
Let us notice that for every \(n\ge 1\) we have
By Proposition 3.6 we get
where \(\hat{C}(\alpha ,\beta ,T,\mathcal {H}_{n})=T^{\alpha }\max \{1,T^{\beta }C(\alpha ,\beta )\}\sup _{f\in \mathcal {H}_{n}}\Vert f\Vert _{\beta }\) by (3.4). Since \(( \mathcal {H}_{n}) _{n\ge 1}\) is uniformly bounded in \(\mathcal {C}^{\beta }( R^{d}) \), then
From Theorem 3.5 it follows that for every \(\rho \in (1-\alpha ,\beta )\) there exists a positive constant \(C(\rho )\) such that for every \(\theta \in (0,1]\) the inequality
holds. Then, due to assumptions imposed on \(( \mathcal {H}_{n}) _{n\ge 1}\), \(( g_{n}) _{n\ge 1}\) and by virtue of (3.8), (3.9), we obtain
Since \(\beta >\rho \) and \(\theta \in (0,1]\) is arbitrary, we get (3.6). Hence, (3.7) is also satisfied by Corollary 3.3. \(\square \)
From now we consider a special choice of the set \(\mathcal {H}\) in formula (3.1). Let \(( f_{n}) _{n\ge 1}\subset \mathcal {C}^{\beta }( R^{d}) \) be an arbitrary bounded sequence. Let \(\mathcal {H}:=\{f_{n}:n\ge 1\}\). Then, by Proposition 2.3 and Corollary 3.3, a set-valued Young integral (3.1) satisfies the following property
Proposition 3.8
Let \(( f_{n}) _{n\ge 1}\subset \mathcal {C}^{\beta }( R^{d}) \) be such that \(\sup _{n\ge 1}\Vert f_{n}\Vert _{\beta }<\infty \) and let \(\mathcal {H}=\{f_{n}:n\ge 1\}\). Then \(cl_{\mathcal {C}}\mathcal {H}\) is bounded in \(\mathcal {C}^{\beta }( R^{d}) \) and
for every \(0\le s<t\le T\).
By Proposition 3.1 and Proposition 3.8 we get also
Corollary 3.9
(i) Let \(( f_{n}) _{n\ge 1}\subset \mathcal {C}^{\beta }( R^{d}) \) be such that \(\sup _{n\ge 1}\Vert f_{n}\Vert _{\beta }<\infty \) and let \(\mathcal {H}=\{f_{n}:n\ge 1\}\). Then
for \(0\le s<t\le T\).
(ii) Let \(\mathcal {H}_{i}=\{f_{n}^{( i) }:n\ge 1\} \) with \(( f_{n}^{(i)}) _{n\ge 1}\subset \mathcal {C}^{\beta }( R^{d}) \) and assume that \(\sup _{n\ge 1}\Vert f_{n}^{(i)}\Vert _{\beta }<\infty \) for \(i=1,2\). Then, for every \(\rho \in (1-\alpha ,\beta )\) there exists a positive constant \(C(\rho )\) such that for every \(\theta \in (0,1]\) and for \(0\le s<t\le T\), it holds
Remark 3.10
The set-valued Young integral \(\int _{s}^{t}\mathcal {H}dg\) considered in the manuscript is not of the type studied by Coutin-Marie-De Fitte in the paper [11]. However, considering a set-valued integral \(\int _{0}^{t}\mathcal {H}dg\) of a bounded and countable set of functions \(\mathcal {H}\subset \mathcal {C}^{\beta }( R^{d}) \) we are able to introduce a set-valued function F by the formula
with the Castaing representation \(\mathcal {H}\) and define the integral \(\int _{s}^{t}F(s)dg(s)\) in the spirit of [11] by the formula
with
If a given \(\beta \)-Hölder multifunction \(F:[0,T]\rightarrow Comp (R^d)\) admits two Castaing representations \(\mathcal {H}_1\) and \(\mathcal {H}_2\) in \(\mathcal {C}^{\beta }( R^{d})\) then it is clear that the integrals \(\int _{s}^{t}\mathcal {H}_1dg\) and \(\int _{s}^{t}\mathcal {H}_2dg\) are different despite they define the same set-valued function F as the below Example shows.
Example 3.11
Let \(f_1,f_2, h_1, h_2:[0,1]\rightarrow R^1\) be defined by formulas:
Let \(F(t)=\{f_1(t),f_2(t)\}=\{h_1(t),h_2(t)\}\) while \(\mathcal {H}_1=\{f_1,f_2\}\) and \(\mathcal {H}_2=\{h_1,h_2\}\). Then \(\mathcal {H}_1\) and \(\mathcal {H}_2\) are two different Castaing representations of the same Lipschitz set-valued function F. Let \(g(t)=t\). An easy calculation shows that \(\int _0^1\mathcal {H}_1dt=\{\int _0^1tdt,\int _0^11/2dt\}=\{1/2\}\) and \(\int _0^1\mathcal {H}_2dt=\{\int _0^1h_1dt,\int _0^1h_2dt\}=\{3/8,5/8\}\). Both integrals differ from the Coutin-Marie-De Fitte integral
because the set \(\{ f_1,f_2,h_1,h_2\}\) forms the set of all 1-Hölder selections of F.
Remark 3.12
It is also worth noting in this point, that in the case of the integral \(\mathcal {I}_{CMDF}(F,g,r)\), only a part of information on a set-valued function F is in use. It follows from the fact that in the definition of \(\mathcal {I}_{CMDF}(F,g,r)\;\beta \)-Hölder selections f of F with \(M_{\beta }(f)\le r\) are considered. It means that for different \(\beta \)-Hölder set-valued functions \(F_{1}\) and \(F_{2}\), the set-valued Young integrals \(\mathcal {I}_{CMDF}(F_{1},g,r)\) and \(\mathcal {I}_{CMDF}(F_{2},g,r)\) can be equal. Indeed, let \(\gamma _{1}(t)=2t\) for \(t\in [0,\frac{1}{2}]\) and \(\gamma _{1}(t)=1\) for \(t\in (\frac{1}{2},1]\). Let \(\gamma _{2}(t)=4t\) for \(t\in [0,\frac{1}{4}]\) and \(\gamma _{2}(t)=1\) for \(t\in (\frac{1}{4},1]\). Now, let
Then \(F_{1}\) and \(F_{2}\) are Lipschitz, \(F_{1}\) is strictly contained in \(F_{2}\) and for \(r=2\) it holds
because only Lipschitz selections (\(\beta =1\)) with the slope less or equal to 2 can be chosen. Hence \(\mathcal {I}_{CMDF}(F_{1},g,2)=\mathcal {I}_{CMDF}(F_{2},g,2).\) The above indicates that we are dealing with a completely different situation than in the case of Aumann integration, where the collection of \(L^p\)-integrable selectors \(S_{L^p}(F)\) uniquely (up to sets of measure zero) defines a multifunction F.
Hovewer, the following result holds true.
Proposition 3.13
Let \(F:[0,T]\rightarrow Comp\left( R^{d}\right) \) be an \(\beta \)-Hölder set-valued function such that \(S(F,\beta ,r)\ne \emptyset \). Then there exists a minimal with respect to inclusion “\(\subset \)” \(\beta \)-Hölder multifunction \(F_{\min }:[0,T]\rightarrow Comp\left( R^{d}\right) \) such that
Proof
Let \(\mathcal {H}:=S(F,\beta ,r)\) and let us define
We will show that
and for any other \(\beta \)-Hölder multifunction \(G:[0,T]\rightarrow Comp\left( R^{d}\right) \) with \(S(G,\beta ,r)=\mathcal {H}\hbox { and }G\subset F\) it holds \(F_{\min }\subset G\). First, let us note that \(F_{\min }\) is \(\beta \)-Hölder with nonempty and compact values. Moreover, it holds
and
Thus \(F_{\min }\) is \(\beta \)-Hölder with values in \(Comp\left( R^{d}\right) \) and
Since for every \(f\in \mathcal {H}\) we have \(f(t)\in F_{\min }(t)\) for \(t\in [0,T]\), then \(\mathcal {H}\subset S(F_{\min },\beta ,r)\). Thus \(S(F_{\min },\beta ,r)\ne \emptyset \). On the other hand \(F_{\min }(t)\subset F(t)\) for \(t\in [0,T]\), thus \(S(F_{\min },\beta ,r)\subset \mathcal {H}\) and therefore, \(S(F_{\min },\beta ,r)=\mathcal {H}\). In order to show that \(F_{\min }\) is minimal for F, let
Then \(F\in \mathcal {F}_{F}\left( \mathcal {H}\right) \) as well as \(F_{\min }\in \mathcal {F}_{F}\left( \mathcal {H}\right) \). Moreover, for any \(G_{1}\), \( G_{2}\in \mathcal {F}_{F}\left( \mathcal {H}\right) \) and such that \(G_{1}\subset G_{2}\) we have \(\mathcal {F}_{G_{1}}\left( \mathcal {H}\right) \subset \mathcal {F}_{G_{2}}\left( \mathcal {H}\right) \subset \mathcal {F}_{F}\left( \mathcal {H}\right) \). Let \(G\in \mathcal {F}_{F}\left( \mathcal {H}\right) \). Since \(F_{\min }(t)=cl_{R^{d}}\{g(t):g\in S(G,\beta ,r)\}\subset G(t)\) for \(t\in [0,T]\), then \(F_{\min }\in \mathcal {F}_{G}\left( \mathcal {H}\right) \) for each \(G\in \mathcal {F}_{F}\left( \mathcal {H}\right) \). Hence \(F_{\min }\in \mathcal {F}_{F_{\min }}\left( \mathcal {H}\right) \subset \bigcap \limits _{G\in \mathcal {F}_{F}\left( \mathcal {H}\right) }\mathcal {F}_{G}\left( \mathcal {H}\right) \subset \mathcal {F}_{F_{\min }}\left( \mathcal {H}\right) \) and therefore, \(F_{\min }\in \mathcal {F}_{F_{\min }}\left( \mathcal {H}\right) =\bigcap \limits _{G\in \mathcal {F}_{F}\left( \mathcal {H}\right) }\mathcal {F}_{G}\left( \mathcal {H}\right) \). Finally, let us note that \(\mathcal {F}_{F_{\min }}\left( \mathcal {H}\right) =\{F_{\min }\}\). Indeed, suppose \(\tilde{G}\in \) \(\mathcal {F}_{F_{\min }}\left( \mathcal {H}\right) \). Then \(\tilde{G}(t)\subset F_{\min }(t)\) for \(t\in [0,T]\) and \(S(\tilde{G},\beta ,r)=S(F_{\min },\beta ,r)=\mathcal {H}\). Moreover,
for every \(t\in [0,T]\). Thus \(\mathcal {F}_{F_{\min }}\left( \mathcal {H}\right) =\{F_{\min }\}\) what means that \(F_{\min }\) is \(\beta \)-Hölder minimal for F. \(\square \)
4 Young differential inclusions and their solutions
Let \(\alpha ,\beta \in (0,1]\) with \(\alpha +\beta >1\) and \(\alpha >\frac{1}{2}\). Let \(\xi \in R^d\) and \(g\in \mathcal {C}^{\alpha }( R^{1}) \). Assume \(F:[0,T]\times R^{d}\rightarrow Conv( R^{d}) \) is a given set-valued function and \(\mathcal {H}=\{h_{n}:n\ge 1\}\) is a family of functions \(h_{n}:[0,T]\times R^{d}\rightarrow R^{d}\). Let us consider a Young differential inclusion
which is interpreted in the integral form as follows
for \(0\le s\le t\le T\), where for every \(x\in \mathcal {C}^{\alpha }(R^{d}) \), \(\mathcal {H}\circ x:=\left\{ h_{n}\circ x:n\ge 1\right\} \) and \(( h_{n}\circ x) ( t) =h_{n}(t,x(t))\), \(t\in [0,T]\) and provided both integrals in (4.2) exist.
Moreover, assume that F and \(\mathcal {H}\) satisfy the following conditions:
(H1) (i) \(F:[0,T]\times R^{d}\rightarrow Conv( R^{d}) \) is measurable and \(\Vert F(\cdot ,0)\Vert _{R^{d}}\in L^{\frac{1}{1-\alpha }}(R^{1})\),
(ii) there exist a constant \(L>0\) and \(\eta \in (0,1]\) such that
for every \(t\in [ 0,T]\), \(u,v\in R^{d}\).
(H2) (i) The set \(\mathcal {H}=\{h_{n}:n\ge 1\}\) consists of functions \(h_{n}:[0,T]\times R^{d}\rightarrow R^{d}\) such that there exist \(\gamma \in ((1-\alpha )/\alpha ,1] \) and a sequence of positive constants \((L_{n})_{n\ge 1}\) with \(\tilde{L}:=\sup _{n}L_{n}<\infty \) and such that
for every \(s,t\in [0,T]\), \(u,v\in R^{d}\), \(n\ge 1\),
(ii) \(C:=\sup _{n}\Vert h_{n}(0,0)\Vert _{R^{d}}<\infty \).
Definition 4.1
A function \(x\in \mathcal {C}^{\alpha }(R^{d}) \) is a solution to differential inclusion (4.1) if it satisfies (4.2) for every \(0\le s\le t\le T\). By \(S(\xi ,g, F,\mathcal {H})\) we will denote the set of all solutions of this inclusion.
Let us notice that for every continuous function \(x:[0,T]\rightarrow R^{d}\) and for set-valued function F satisfying (H1) it holds
for every \(t\in [0,T]\), where \(b(\cdot ):=L+\Vert F(\cdot ,0)\Vert _{R^{d}}\in L^{1/(1-\alpha )}(R^d)\). Last inequality holds because \(a^{\eta }\le 1+a\) for every \(a>0\) and \(\eta \in (0,1]\). Therefore, the set-valued function \(F(\cdot ,x(\cdot ))\) is measurable and integrably bounded. Thus the Aumann integral \(\int _{s}^{t}F(\tau ,x(\tau ))d\tau \) in (4.2) is well defined (see e.g., [5]). Since \(b\in L^{1/(1-\alpha )}(R^{1})\), it follows by the Hölder inequality that \(\Vert \int _{s}^{t}b(\tau )d\tau \Vert \le \Vert b\Vert _{L^{1/(1-\alpha )}(R^{1})}(t-s)^{\alpha }\) for every \(0\le s\le t\le T\). Thus the function \([0,T]\ni t\rightarrow \int _{0}^{t}b(\tau )d\tau \) is \(\alpha \)-Hölder continuous, which by (4.3) yields
for every \(0\le s\le t\le T\) and
Hence \(\int _{0}^{\cdot }F(\tau ,x(\tau ))d\tau \in \mathcal {C}^{\alpha }( Conv( R^{d}) ) \). As a consequence of (H2), we have
for \(u\in R^{d}\) and
for every \(s,t\in [0,T]\), \(n\ge 1\) and \(x\in \mathcal {C}^{\alpha }( R^{d}) \). Thus, for each \(n\ge 1\), \(h_{n}\circ x\in \mathcal {C}^{\delta }( R^{d}) \), where \(\delta =\min \{\beta ,\alpha \gamma \}\). Since \(\delta \in (0,1]\) and \(\delta +\alpha >1\), then the Young integral \(\int _s^t(h_{n}\circ x)dg\) is well defined for each \(n\ge 1\). Moreover, for each \(x\in \mathcal {C}^{\alpha }( R^{d}) \) with \(x(0)=\xi \), it follows by (H2) and (4.7)
and
Thus, for any \(x\in \mathcal {C}^{\alpha }( R^{d}) \) with \(x(0)=\xi \), we have \(\sup _{n\ge 1}\Vert h_{n}\circ x\Vert _{\delta }<\infty \). Therefore, due to Proposition 3.8, the set \(cl_{\mathcal {C}}\{h_{n}\circ x:n\ge 1\}\) is bounded in the space \(\mathcal {C}^{\delta }( R^{d}) \). Moreover, a set-valued Young integral \(\int _{s}^{t}cl_{\mathcal {C}}(\mathcal {H}\circ x)dg\) from (4.2) is well defined and
for every \(0\le s\le t\le T\). Let \(A( x):=K( \xi )+\tilde{L}\max \{1,(M_{\alpha }( x))^{\gamma } \}( T^{\beta }+T^{\alpha \gamma }) \) and \(B( x):=\tilde{L}\max \{1,(M_{\alpha }( x))^{\gamma } \}( T^{\beta -\delta }+T^{\alpha \gamma -\delta }) \), i.e., A(x) and B(x) are right-hand sides of (4.8) and (4.9), respectively. Then, by Corollary 3.9 with \(f_{n}:=h_{n}\circ x\), we get
Corollary 4.2
Let \(\mathcal {H}=\{h_{n}:n\ge 1\}\) satisfy condition (H2) and let \(x\in \mathcal {C}^{\alpha }( R^{d}) \). Then
Let \(J_{\mathcal {H},T}:\mathcal {C}^{\alpha }( R^{d})\rightarrow R_{+}\) be given by the formula
Then \(J_{\mathcal {H},T}(D)\) is a bounded set in \(R_{+}\) for every bounded set \(D\subset \mathcal {C}^{\alpha }( R^{d}) \).
Theorem 4.3
Let \(F:[0,T]\times R^{d}\rightarrow Conv( R^{d}) \) and \(\mathcal {H}=\{h_{n}:n\ge 1\}\) satisfy conditions (H1) and (H2), respectively. Then the set \(S(\xi ,g, F,\mathcal {H})\) of all solutions of inclusion (4.1) is a nonempty and closed subset of \(\mathcal {C}^{\alpha }( R^{d}) \).
Proof
For \(A\in Conv( R^{d}) \) and \(y\in R^{d}\), let \(\sigma :Conv( R^{d}) \times R^{d}\rightarrow R^{1}\) be a support function, i.e., \(\sigma (A,y):=\sup _{a\in A}\left\langle a,y\right\rangle \), where \(\langle \cdot ,\cdot \rangle \) denotes an inner product in \(R^{d}\). Let \(\Sigma ^{d-1}\) be the unit sphere in \(R^{d}\) and let \(\omega \) denote a measure on \(\Sigma ^{d-1}\) proportional to the Lebesgue measure and such that \(\omega ( \Sigma ^{d-1}) =1\). Let \(St:Conv( R^{d}) \rightarrow R^{d}\) given by
be a Steiner point of a set A (see [5]). It is a linear function on the space \(Conv( R^{d}) \) by Theorem 9.4.1 of [5] and by Proposition 2.2 of [11] it is Lipschitz continuous, i.e., \(St(A)\in A\), \(St(\alpha _{1}A_{1}+\alpha _{2}A_{2})=\alpha _{1}St(A_{1})+\alpha _{2}St(A_{2})\) and
for every A, \(A_{1}\), \(A_{2}\in Conv( R^{d}) \) and \(\alpha _{1}\), \(\alpha _{2}\in R\). Let \(f(t,u)=St(F(t,u))\) for \((t,u)\in [0,T]\times R^{d}\). Then by (H1) it holds
for every \(t\in [0,T]\), \(u\in R^{d}\). Let us choose an arbitrary function \(h\in \mathcal {H}\) and consider the equation
Thanks to Theorem 4.1 of [35], there exists at least one \(\alpha \)-Hölder continuous solution to equation (4.11) which is obviously a solution to inclusion (4.2).
In order to show that \(S( \xi ,g, F,\mathcal {H})\) is a closed subset of \(\mathcal {C}^{\alpha }( R^{d}) \) assume that \(( x_{k}) \subset S(\xi ,g, F,\mathcal {H})\) and \(x_{k}\rightarrow x\) in \(\mathcal {C}^{\alpha }( R^{d}) \) where \(x\in \mathcal {C}^{\alpha }( R^{d}) \). Thus \(\Vert x_{k}-x\Vert _{\infty }\rightarrow 0\) as \(n\rightarrow \infty \), \(x(0)=\xi \) and \(\sup _{k}M_{\alpha }(x_{k}) <\infty \). We will show that \(x\in S(\xi ,g, F,\mathcal {H})\). By properties of the Aumann integral (see e.g., [5, 17]) and using (H2), we get
Thus
On the other hand we get by (H2)
for every \(k\ge 1\) and thus
In order to show that \(x\in S(\xi ,g, F,\mathcal {H})\) it suffices to claim that
for every \(0\le s\le t\le T\). Indeed, let us note that the following inequality holds
Since \(( x_{k}) \subset S(\xi ,g, F,\mathcal {H})\), the second term on the right side above equals to zero while first one tends to zero as \(k\rightarrow \infty \). The same is true for the third one because of (4.12). Thus it remains to show that
Let \(k\ge 1\) be fixed. By Corollary 3.9(ii) with \(f_{n}^{1}:=h_{n}\circ x_{k}\), \(f_{n}^{2}:=h_{n}\circ x\) and \(\beta :=\delta \) (recall \(\delta +\alpha >1)\) we get
for every \(\rho \in (1-\alpha ,\delta )\), there exists a constant \(C(\rho )\) such that for every \(\theta \in (0,1]\) it holds
But \(\sup _{n}M_{\delta }( h_{n}\circ x) <\infty \) by (4.9) and \(\sup _{k}\sup _{n}M_{\delta }( h_{n}\circ x_{k}) <\infty \) because of \(\sup _{k}M_{\alpha }( x_{k}) <\infty \). Thus we obtain by (4.13)
Since \(\delta >\rho \) and \(\theta \in (0,1]\) is arbitrary, we get (4.14) what finishes the proof of closedness of \(S(\xi ,g, F,\mathcal {H})\). \(\square \)
Now we consider the problem of boundedness of solutions set to inclusion (4.1). For this aim we shall use more convenient and natural counterparts of previous notations. For \(0\le s<t\le T\) and \(x\in \mathcal {C}^{\alpha }( R^{d}) \) let
where
If \(0=T_{0}<T_{1}<...<T_{k-1}<T_{k}=T\) is a given partition of [0, T], then by the inequality (5) of [27] we get
Let \(\Delta :=\{( s,t):0\le s\le t\le T)\}\) and let \(\varphi _{1},\varphi _{2}:\Delta \rightarrow R_{+}\) be functions defined by formulas
and
where \(C(\alpha ,\delta )\) is a constant from Corollary 4.2 and \(\tilde{L}=\sup _{n}L_{n}\) (see condition (H2)). By (4.16) and (4.17) it is easy to see that functions \(\varphi _{1}(s,\cdot )\), \(\varphi _{2}(s,\cdot ):[s,T]\rightarrow R_{+}\) are strictly increasing and continuous for any fixed \(s\in [0,T)\). Moreover, \(\varphi _{2}(s,t)=\varphi _{2}(0,t-s)\) for \((s,t)\in \Delta \) and \(\varphi _{2}(0,0)=0\). Then there exists a finite partition: \(0=T_{0}<T_{1}<...<T_{k-1}<T_{k}=T\) such that \(\varphi _{2}(T_{i},T_{i+1})<1\) for \(i=0,1,...,k-1\).
Recall that \(K(u):=\sup _{n\ge 1}\Vert h_{n}(0,u)\Vert _{R^{d}}\le \tilde{L}\Vert u\Vert _{R^{d}}^{\gamma }+C\) for \(u\in R^{d}\) by (4.6).
Lemma 4.4
Let \((s,t)\in \Delta \) be such that \(\varphi _{2}(s,t)<1\). Then for every \(x\in S( \xi ,g, F,\mathcal {H}) \) it holds
Proof
Let \(x\in S( \xi ,g,F,\mathcal {H}) \). Similarly as in inequality (4.5) one obtains
Next, by (4.8), (4.9), Corollary 4.2 together with inequality \(a^{\gamma }\le 1+a\) for \(a\ge 0\) and \(\gamma \in (0,1]\), recall that in our case \(\gamma \in (\frac{1-\alpha }{\alpha },1]\), we get
Since \(x\in S( \xi ,g, F,\mathcal {H}) \), then it holds
because of \(\varphi _{2}(s,t)=\varphi _{1}(s,t)+L(t-s)\). Hence we obtain formula (4.18). \(\square \)
Theorem 4.5
The set \(S( \xi , g, F,\mathcal {H}) \) of all solutions of the inclusion (4.1) is bounded in the space \(\mathcal {C}^{\alpha }( R^{d}) \).
Proof
Let \(x\in S( \xi ,g, F,\mathcal {H}) \). There exists a finite partition \(0=T_{0}<T_{1}<...<T_{k-1}<T_{k}=T\) such that \(\varphi _{2}(T_{i},T_{i+1})<1\) for \(i=0,1,...,k-1\) for some \(k\ge 1\). First, let us note that for every \(t\in (0,T]\) it holds
By Lemma 4.4 we have
Let us notice that \(U_{1}\) does not depend on x but only on \(\xi \), F, \(\left\{ h_{n}:n\ge 1\right\} \), g, T and parameters \(\alpha \), \(\beta \), \(\gamma \). By the above inequality together with (4.19) we obtain
The constant \(P_{1}\) depends on the same quantities as \(U_{1}\). Similarly, by (4.19) and (4.15) we get
Using again Lemma 4.4 we obtain
The norm \(\Vert x(T_{1})\Vert _{R^{d}}\le P_{1}\) and thus \(K( x(T_{1})) \le \tilde{L}P_{1}^{\gamma }+C\) by (4.20). Therefore, there exists a constant \(U_{2}\), depending on \(P_{1}\) but independent of x, which estimates the right side of inequality (4.22). Thus we get by (4.21)
Repeating this procedure k-times along the partition \(0=T_{0}<T_{1}<...<T_{k-1}<T_{k}=T\) one can find a positive constant \(P_k\) which also does not depend on x but only on \(\xi \), F, \(\{ h_{n}:n\ge 1\} \), g, T and parameters \(\alpha \), \(\beta \), \(\gamma \) and such that \(\Vert x\Vert _{\alpha ,[0,T]}\le P_k\). This finishes the proof. \(\square \)
Remark 4.6
Let \(\mathcal {B}_{\alpha }(0,r)\) denote a closed ball in \(\mathcal {C}^{\alpha }( R^{d}) \) with a radius \(r>0\) and centered in the origin. The procedure used in the proof of Theorem 4.5 shows that \(S( \xi ,g,F,\mathcal {H}) \subset \mathcal {B}_{\alpha }(0,r)\) with \(r=P_k\), where \(P_k\) dependents continuously on \(\Vert \xi \Vert _{R^{d}}\) and \(M_{\alpha ,[0,T]}(g)\).
One may prove the following result.
Theorem 4.7
Let \(( \xi _{l}) _{l\ge 1}\subset R^{d}\) and \((g_l)\subset \mathcal {C}^{\alpha }( R^1) \)be bounded sequences. Then \(\bigcup \limits _{l\ge 1}S( \xi _{l},g_l, F,\mathcal {H}) \) is a bounded subset in \(\mathcal {C}^{\alpha }(R^{d}) \).
Proof
The proof is almost the same as the proof of Theorem 4.5. It is enough to take \(\sup _{l\ge 1}M_{\alpha ,[0,T]}(g_l)\) and \(\sup _{l\ge 1}\Vert \xi _l\Vert _{R^{d}}\) instead of \(M_{\alpha ,[0,T]}(g)\) and \(\Vert \xi \Vert _{R^{d}}\), respectively. \(\square \)
Until now, it is assumed that \(\alpha ,\beta \in (0,1]\) with \(\alpha +\beta >1\) and \(\alpha >1/2\). Let us additionally assume that \(\beta <\alpha \). Hence we obtain \((1-\alpha )/\alpha<\beta /\alpha <1\). Let us consider \(\gamma \in ( \beta /\alpha ,1) \) in assumption (H2). Then we have the following result.
Theorem 4.8
Let \(\alpha ,\beta \in (0,1]\) with \(\alpha +\beta >1\), \(\beta <\alpha \) and \(\alpha >1/2\). Suppose \(g\in \mathcal {C}^{\alpha }(R^1)\), \(F:[0,T]\times R^{d}\rightarrow Conv( R^{d}) \) satisfies condition (H1) and \(\mathcal {H}=\{h_{n}:n\ge 1\}\) satisfies condition (H2) for \(\gamma \in ( \beta /\alpha ,1) \). Then \(S(\xi ,g, F,\mathcal {H})\) is closed in the topology of uniform convergence in the space \(\mathcal {C}( R^{d}) \).
Proof
By Theorems 4.3 and 4.5, \(S( \xi ,g, F,\mathcal {H}) \) is a nonempty and bounded set in \(\mathcal {C}^{\alpha }( R^{d}) \). Suppose \((x_{k}) _{k\ge 1}\subset S( \xi ,g,F,\mathcal {H}) \) and \(\Vert x_{k}-x\Vert _{\infty }\rightarrow 0\). Hence \(\sup _{k\ge 1}M_{\alpha }( x_{k}) <\infty \). But since \(\beta <\alpha \) and \(\gamma \in ( \beta /\alpha ,1) \), it follows that \(\delta :=\min \{\beta ,\alpha \gamma \}=\beta \). Then \(\sup _{k}\sup _{n}M_{\beta }( h_{n}\circ x_{k}) <\infty \) by (4.9). Since \(\sup _{k}M_{\alpha }( x_{k}) <\infty \), \(\Vert x_{k}-x\Vert _{\infty }\rightarrow 0\), and \(\beta <\alpha \) then it follows by Proposition 2.1 that \(x_{k}\rightarrow x\) in \(\mathcal {C}^{\beta }( R^{d}) \). Therefore, \(M_{\beta }( x) <\infty \). Moreover, \(x\in \mathcal {C}^{\alpha }( R^{d}) \) by Proposition 2.2. Next, we get
by (H2). Hence \(\sup _{n}M_{\beta }( h_{n}\circ x) \le 2\tilde{L}\max \{1,M_{\beta }( x) \}\). Now proceeding similarly as in the proof of Theorem 4.3 we obtain
and
We deduce from these inequalities
what means that \(x\in S( \xi ,g,F,\mathcal {H}) \). \(\square \)
By virtue of Theorem 4.8, Proposition 2.1 and Proposition 2.2 the following statement holds true.
Corollary 4.9
Under conditions of Theorem 4.8, the set \(S(\xi ,g, F,\mathcal {H})\) is a compact subset of the space \(\mathcal {C}( R^{d}) \).
Proof
Indeed, let \(( x_{k}) _{k\ge 1}\subset S( \xi ,g,F,\mathcal {H}) \). Then \(\sup _{k}M_{\alpha }( x_{k}) <\infty \). Since \(\beta <\alpha \) then there exists a subsequence \((x_{k_{l}}) _{l\ge 1}\) of \(( x_{k}) _{k\ge 1}\) and \(x\in \mathcal {C}^{\alpha }( R^{d}) \) such that \(x_{k_{l}}\rightarrow x\) in \(\mathcal {C}^{\beta }( R^{d})\) by Proposition 2.2. Thus \(x\in S( \xi ,g,F,\mathcal {H}) \), because \(S( \xi ,g,F,\mathcal {H}) \) is a closed subset of \(\mathcal {C}( R^{d}) \) by Theorem 4.8. \(\square \)
We know by Theorem 4.3 and Theorem 4.5 that the set \(S( \xi ,g,F,\mathcal {H}) \) is a nonempty, bounded and closed set in the space \(\mathcal {C}^{\alpha }( R^{d}) \). By Corollary 4.9 it is also compact in \(\mathcal {C}( R^{d}) \). In fact \(S(\xi ,g,F,\mathcal {H})\) is a compact set not only in the space \(\mathcal {C}( R^{d}) \) but also in \(\mathcal {C}^{\alpha ^1}( R^{d}) \) for any \(\alpha ^1<\alpha \).
Corollary 4.10
Under conditions of Theorem 4.8, \(S(\xi ,g, F,\mathcal {H})\) is a compact subset of the space \(\mathcal {C}^{\alpha ^1}( R^{d}) \) for every \(0<\alpha ^1<\alpha \).
Proof
Let \(( x_{k}) _{k\ge 1}\subset S( \xi ,g,F,\mathcal {H}) \) be arbitrary. By Theorem 4.5, \(S( \xi ,g,F,\mathcal {H}) \) is a bounded set in \(\mathcal {C}^{\alpha }( R^{d}) \). Hence \(\sup _{k}M_{\alpha }( x_{k}) <\infty \). Let \(\alpha ^1<\alpha \). By Proposition 2.2 there exist a subsequence \(( x_{k_{l}}) _{l\ge 1}\) of \(( x_{k}) _{k\ge 1}\) and \(x\in \mathcal {C}^{\alpha }( R^{d}) \) such that \(x_{k_{l}}\rightarrow x\) in \(\mathcal {C}^{\alpha ^1}( R^{d}) \). We will show that \(x\in S( \xi ,g,F,\mathcal {H}) \). Since \(\gamma \in ( \beta /\alpha ,1) \), it follows that \(\delta :=\min \{\beta ,\alpha \gamma \}=\beta \). We have \(\sup _{l}M_{\alpha }( x_{k_{l}})<\infty \) and therefore, \(\sup _{l}\sup _{n}M_{\beta }( h_{n}\circ x_{k_{l}}) <\infty \) by (4.9). We have assumed that \(\beta <\alpha \) and we know that \(x\in \mathcal {C}^{\alpha }( R^{d}) \). Thus \(M_{\beta }( x) <\infty \) and we get by (H2)
Hence \(\sup _{n}M_{\beta }( h_{n}\circ x) \le 2\tilde{L}\max \{1,M_{\beta }( x) \}\). Now, proceeding similarly as in the proof of Theorem 4.3, we obtain
and
what leeds to equality
It means that \(x\in S( \xi ,g,F,\mathcal {H}) \). Hence \(( x_{k}) _{k\ge 1}\subset S( \xi ,g,F,\mathcal {H}) \) admits a subsequence \(( x_{k_{l}}) _{l\ge 1}\) convergent to \(x\in S( \xi ,g,F,\mathcal {H}) \) in \(\mathcal {C}^{\alpha ^1}( R^{d}) \) for every \(\alpha ^1<\alpha \). \(\square \)
Although the compactness of the set \(S( \xi ,g,F,\mathcal {H}) \) in the space \(\mathcal {C}^{\alpha ^1}( R^{d}) \) for \(\alpha ^1<\alpha \) is stronger than its compactness in \(\mathcal {C}( R^{d}) \), however \(\mathcal {C}^{\alpha ^1}( R^{d}) \) is not a separable Banach space. Because the separability is necessary in last part of the paper, we will use more often Corollary 4.9 instead of Corollary 4.10 in our further considerations.
Due to Corollary 4.9 one can consider solution sets as a set-valued function \(S( \cdot ,\cdot , F,\mathcal {H} ):R^{d}\times \mathcal {C}^{\alpha }( R^1)\rightarrow Comp(\mathcal {C}( R^{d}) ),\) \((\xi ,g)\mapsto S(\xi ,g, F,\mathcal {H})\) and ask about its continuity. This set-valued function will be called a solution map of inclusion (4.1). We shall show that it is upper semicontinuous in the sense of set-valued functions. For this aim let us recall needed details ( [20]).
Definition 4.11
Let X and Y be topological Hausdorff spaces and let \(G:X\rightarrow Cl_b(Y)\) be a given set-valued function. G is said to be upper semicontinuous (u.s.c.) at \(x_{0}\in X\) if for every open set \(U\subset Y\) and \(G(x_{0})\subset U\) there exists an open neighborhood V of \(x_{0}\) such that \(G(x)\subset U\) for every \(x\in V\). A set-valued function \(G:X\rightarrow Cl_b(Y)\) is called upper semicontinuous (u.s.c.) on X if it is u.s.c. at every \(x\in X\).
In the case of metric spaces X and Y and for compact-valued set-valued functions the following characterization holds (see Theorem II.2.2. of [20]).
Proposition 4.12
Let X and Y be metric spaces. A set-valued function \(G:X\rightarrow Comp(Y)\) is upper semicontinuous if and only if for every \(x\in X\), every sequence \(( x_{l})\subset X, \) \(x_l\rightarrow x\) and every sequence \(( y_{l}) \) of \(\;Y\) with \(y_{l}\in G(x_{l})\), there is a convergent subsequence of \(( y_{l}) \) whose limit belongs to G(x) .
It can be also proved that a set-valued function \(G:X\rightarrow Comp(Y)\) is u.s.c. on X if and only if G is H-u.s.c. on X, i.e., for every \(x\in X\) and \(\epsilon >0\), there exists \(\delta >0\) such that \(G(\mathcal {B}_{X}(x,\delta ))\subset V(G(x),\epsilon )\) where \(V(K,\epsilon ):=\{y\in Y:\textrm{dist}_{Y}(y,K)<\epsilon \}\) for \(K\in Comp(Y)\). For such G its upper semicontinuity is equivalent to the following condition: for every \(( x_{l}) \subset X\),
where \(\bar{H}_{Y}\) is a Hausdorff semi-metric in Comp(Y), see e.g., [20] pp. 28-29.
Theorem 4.13
Under assumptions of Theorem 4.8 the solution map
is upper semicontinuous.
Proof
Let \(\xi \in R^{d}\), \(g\in \mathcal {C}^{\alpha }( R^1) \) and let \(( \xi _{l}) _{l\ge 1}\subset R^{d}\) and \((g_l) \subset \mathcal {C}^{\alpha }(R^1)\) be sequences such that \(\xi _{l}\rightarrow \xi \) in \(R^d\) and \(g_l\rightarrow g\) in \( \mathcal {C}^{\alpha }( R^1)\) as \(l\rightarrow \infty \). Let \(x_{l}\in S( \xi _{l},g_l,F,\mathcal {H}) \) for \(l\ge 1\). Then \(\{x_{l}:l\ge 1\}\subset \mathcal {S}:=\bigcup \limits _{l\ge 1}S(\xi _{l},g_l,F,\mathcal {H}) \). Thus due to Theorem 4.7 we have \(\sup _{l\ge 1}\Vert x_{l}\Vert _{\alpha }<\infty \). Since \(\beta <\alpha \) then there exists a subsequence \(( x_{l_{k}})_{k\ge 1}\) of \(( x_{l}) _{l\ge 1}\) and \(x\in \mathcal {C}^{\alpha }( R^{d}) \) such that \(x_{k_{l}}\rightarrow x\) in \(\mathcal {C}^{\beta }( R^{d}) \) by Proposition 2.2. Then \(\Vert x_{l_{k}}-x\Vert _{\infty }\) \(\rightarrow 0\) as \(k\rightarrow \infty \). But \(x_{l_{k}}(0)=\xi _{l_{k}}\rightarrow \xi \) and therefore, \(x(0)=\xi \). We will show that \(x\in S( \xi ,g,F,\mathcal {H}) \). Similarly as in the proof of Theorem 4.3 it holds
for every \(0\le s\le t\le T\). Since \(x_{l_{k}}\in S( \xi _{l_{k}},g_{l_k},F,\mathcal {H}) \), then the term “\(\textrm{dist}_{R^{d}}(....) \)” vanishes. Moreover,
because of the inequality
We deduce by Corollary 3.9, that for every \(\rho \in (1-\alpha ,\beta )\), there exists a constant \(C(\rho )\) such that for every \(\theta \in (0,1]\)
Next we have
for every \(k\ge 1\) by (H2). Thus
Finally, we get by (4.9)
and from \(\sup _{k\ge 1}M_{\alpha }( x_{l_{k}}) <\infty \), together with \(x\in \mathcal {C}^{\alpha }( R^{d}) \) we obtain
Combining (4.26) with (4.27) and (4.28) we have
Since \(\theta \in (0,1]\) is arbitrary and \(\rho <\beta \) then
By Proposition 3.6 we obtain
where \(\hat{C}(\alpha ,\beta ,T,(\mathcal {H}\circ x_{l_k}))=T^{\alpha }\max \{1,T^{\beta }C(\alpha ,\beta )\}\sup _{n\ge 1}\Vert h_n\circ x_{l_k}\Vert _{\beta }.\) Using formulas (4.8) and (4.9) once again we get
because \(\delta =\beta \) here.
Since the set \(\{x_{l_k}\}\) is bounded in \(\mathcal {C}^{\alpha }(R^d)\) then
Thus by (4.30) we get
Combining (4.31), (4.29) and (4.25) in view of (4.24) we have
for every \(0\le s\le t\le T\), what means that \(x\in S( \xi ,g,F,\mathcal {H}) \). By Proposition 4.12 we deduce the desired u.s.c. property of a set-valued function \(S( \cdot ,\cdot , F,\mathcal {H} ) \). \(\square \)
Due to the previous theorem we have
Corollary 4.14
Let assumptions of Theorem 4.8 be satisfied.
(i) If \(\xi _{n}\rightarrow \xi \) in \(R^{d}\) and \(g _{n}\rightarrow g\) in \(\mathcal {C}^{\alpha }(R^1)\), then
(ii) Let \(\Gamma _1 \subset R^{d}\) and \(\Gamma _2 \subset \mathcal {C}^{\alpha }(R^1)\) be compact sets. Then the set
is bounded in \(\mathcal {C}^{\alpha }( R^{d}) \) and compact in \(\mathcal {C}( R^{d}) \).
Proof
First property follows by the comments below Proposition 4.12. The boundedness of \(S( \Gamma _1,\Gamma _2,F,\mathcal {H}) \) in \(\mathcal {C}^{\alpha }( R^{d}) \) can be proved similarly as it was done in the proof of Theorem 4.7. Since the set-valued function \(S( \cdot ,\cdot , F,\mathcal {H} ) \) is u.s.c. by Theorem 4.13, the compactness of the set \(S( \Gamma _1,\Gamma _2,F,\mathcal {H}) \) in \(\mathcal {C}( R^{d}) \) follows by Proposition II.2.3. in [20]. \(\square \)
5 Applications
Let us take an arbitrary continuous functional \(\mathcal {J}:\mathcal {C} ( R^{d}) \rightarrow R_{+}\). For instance \(\mathcal {J}\) can be taken as \( \mathcal {J}(x):=\Vert x\Vert _{\infty }\) or \(\mathcal {J}(x):=\min \left\{ \Vert x(t_{i})\Vert _{R^{d}}:i=1,2,...,k\right\} \) where \(\left\{ t_{i}:i=1,2,...,k\right\} \subset [0,T]\) is a fixed collection of points, or
where \(p:[0,T]\times R^{d}\rightarrow R^{d}\) is a given bounded and continuous function. Then we can consider the following optimality problem for Young differential inclusion (4.1)
Definition 5.1
By an optimal solution to Young differential inclusion (4.1) we mean a function \(x\in S(\xi ,g, F,\mathcal {H})\) such that
The set
is called a set of \(\mathcal {J}\)-optimal solutions to differential inclusion (4.1), while a set-valued function
is called a \(\mathcal {J}\)-optimal solution map of inclusion (4.1).
Under conditions of Corollary 4.9 it is clear that optimal solutions exist. So the set of \(\mathcal {J}\)-optimal solutions to differential inclusion (4.1) is nonempty for every \((\xi ,g)\in R^{d}\times \mathcal {C}^{\alpha }( R^1)\), because the functional \(\mathcal {J}\) is assumed to be continuous. Moreover, \(S_{\mathcal {J}}(\xi ,g, F,\mathcal {H})\in Comp(\mathcal {C}( R^{d}) )\) for every \(\xi \in R^{d}\) and \(g\in \mathcal {C}^{\alpha }( R^1).\)
Each \(x\in S_{\mathcal {J}}(\xi ,g, F,\mathcal {H})\) depends somehow on initial value \(\xi \in R^{d}\) and \(g\in \mathcal {C}^{\alpha }( R^1)\). In fact it will be shown that it depends on \(\xi \) and g in a measurable way. We shall use the following auxiliary lemma. See [5] for details on measurability of set-valued functions.
Lemma 5.2
Let X be a separable metric space and let \(U:X\rightarrow R\) be a continuous function. Then the mappings
are measurable.
Theorem 5.3
Under assumptions of Theorem 4.8, there exists a sequence of measurable functions (a Castaing representation) \( s_{\mathcal {J}}^{n}:R^{d}\times \mathcal {C}^{\alpha }( R^1)\rightarrow \mathcal {C}( R^{d}), n\ge 1\), such that \(s_{\mathcal {J}}^{n}( \xi ,g) \) is an J-optimal solution to inclusion (4.1) for every \(n\ge 1\), and such that
for every \(\xi \in R^{d}\) and \(g\in \mathcal {C}^{\alpha }( R^1)\).
Proof
By Theorem 4.13, \(S( \cdot ,\cdot , F,\mathcal {H} ):R^{d}\times \mathcal {C}^{\alpha }( R^1)\rightarrow Comp(\mathcal {C}( R^{d}) )\) is u.s.c. and hence by Proposition 8.2.1 of [5] it is measurable. In fact, in this case “measurable” means that \(S( \cdot ,\cdot , F,\mathcal {H} )\) is measurable as the function from \(R^{d}\times \mathcal {C}^{\alpha }( R^1)\) to a metric space \((Comp(\mathcal {C}(R^{d}),H_{\infty })\) (c.f., Theorem II.3.2. in [20]).
In the notation of Lemma 5.2, letting \(X:=\mathcal {C}(R^{d})\) and \(U:=\mathcal {J}\) we get \(S_{\mathcal {J}}(\cdot ,\cdot ,F,\mathcal {H})=W\circ S(\cdot ,\cdot ,F,\mathcal {H} )\) which implies the measurability of the set \(S_{\mathcal {J}}(\cdot ,\cdot ,F,\mathcal {H} )\). Finally, applying Theorem 1 of [17], we get the desired sequence of measurable selections \( (\xi ,g)\mapsto s^n_{\mathcal {J}}(\xi ,g) \) of \(S_{\mathcal {J}}(\cdot ,\cdot ,F,\mathcal {H} )\) and such that \(S_{\mathcal {J}}(\xi ,g, F,\mathcal {H})=cl_{\mathcal {C}}\{ s_{\mathcal {J}}^{n}(\xi ,g):n\ge 1\} \) for every \(\xi \in R^{d}\) and \(g\in \mathcal {C}^{\alpha }( R^1)\). \(\square \)
Once having properties of solutions sets \(S(\xi ,g, F,\mathcal {H})\) and \(S_{\mathcal {J}}(\xi ,g, F,\mathcal {H})\) in (5.2) we can use them for the analysis of reachable sets of solutions in time \(t\in [0,T]\). For \(t\in [0,T]\), \(\xi \in R^{d}\) and \(g\in \mathcal {C}^{\alpha }( R^1)\), let
and
denote reachable and \(\mathcal {J}\)-optimal reachable sets of inclusion (4.1), respectively. Then \(\mathcal {A}( t,\xi ,g ),\mathcal {A}_{\mathcal {J}}( t,\xi ,g) \in Comp(R^{d})\) for any \(t\in [0,T]\), \(\xi \in R^{d}\) and \(g\in \mathcal {C}^{\alpha }( R^1)\) by Corollary 4.9.
Moreover, we have
Corollary 5.4
Let conditions of Theorem 4.8 be satisfied. Then the reachable sets of solutions to differential inclusion (4.1) admit the following properties:
(i) For every fixed \(\xi \in R^{d}\) and \(g\in \mathcal {C}^{\alpha }( R^1)\) set-valued functions
are \(\alpha \)-Hölder continuous on [0, T],
(ii) For every fixed \(t\in [0,T]\) a set-valued function
is upper semicontinuous, while
is measurable.
Proof
(i). Let \(\xi \in R^{d}\), \(g\in \mathcal {C}^{\alpha }( R^1)\) and \(0\le s<t\le T\) be given. Suppose \(z\in \mathcal {A}( t,\xi ,g) \). Then \(z=x(t)\) for some \(x\in S( \xi ,g,F,\mathcal {H}) \). Hence
Thanks to (4.4), (4.5) applied to first term, and Corollary 4.2 applied to second term above, we get
where
and
Recall that the parameter \(\delta \) from the formula (4.10) is equal to \(\beta \) now, because of assumptions of Theorem 4.8. Moreover, by Theorem 4.5, the set \(S(\xi ,g, F,\mathcal {H})\) is bounded in \(\mathcal {C}^{\alpha }( R^{d}) \). Thus \( \sup \{ \Vert x\Vert _{\infty }: x\in S( \xi ,g,F,\mathcal {H})\}<\infty \) and \(\sup \left\{ M_{\alpha }( x)^{\gamma }:x\in S( \xi ,g,F,\mathcal {H}) \right\} <\infty .\)
Therefore, there exists a positive constant \(C_{\mathcal {A}}(\xi ,g )\) such that
Hence we get \(\bar{H}_{_{R^{d}}}( \mathcal {A}( t,\xi ,g ),\mathcal {A}( s,\xi ,g) ) \le C_{\mathcal {A}}(\xi ,g ) (t-s) ^{\alpha }\). The same inequality holds for a semimetric \(\bar{H}_{_{R^{d}}}( \mathcal {A}( s,\xi ,g),\mathcal {A}( t,\xi ,g ) )\). Thus
Moreover, since \(\Vert \mathcal {A}( 0,\xi ,g) \Vert _{R^{d}}=\Vert \xi \Vert _{R^{d}}\) we get
In a similar way one can prove analogous inequalities for a set-valued function \(\mathcal {A}_{\mathcal {J}}( \cdot ,\xi ,g) \). That ends the proof of part (i).
(ii). Let \(\xi _{n}\rightarrow \xi \), \(g _{n}\rightarrow g \) and let \(z_{n}\in \mathcal {A}( t,\xi _{n},g_n) \) for \(n\ge 1\). Then there exists a sequence \(x_{n}\in S(\xi _{n},g_n,F,\mathcal {H})\) and such that \(z_{n}=x_{n}(t)\) for \(n\ge 1\). Since the set-valued function \(S(\cdot ,\cdot ,F,\mathcal {H} )\) is u.s.c., it follows by Proposition 4.12, that there exists a subsequence \(( x_{n_{k}}) _{k\ge 1}\) of a sequence \(( x_{n}) _{n\ge 1}\) convergent in \(\Vert \mathcal {\cdot }\Vert _{\infty }\)-norm to some \(x\in S(\xi ,g, F,\mathcal {H})\). Hence \(z_{n_{k}}\rightarrow x(t)\in \mathcal {A}( t,\xi ,g ) \) what finishes the proof of upper semicontinuity of \(\mathcal {A}( t,\cdot ,\cdot ) \). The measurability of \(\mathcal {A}_{\mathcal {J}}(t,\cdot ,\cdot ):R^{d}\times \mathcal {C}^{\alpha }( R^1)\rightarrow Comp(R^{d})\) follows by measurability of the \(\mathcal {J}\)-optimal solution map \(S_{\mathcal {J}}(\cdot ,\cdot ,F,\mathcal {H} )\) and continuity of the projection map \(\Pi _t: \mathcal {C}^{\alpha }( R^{d})\rightarrow R^{d}\), \(\Pi _t(x)=x(t)\) for \(t\in [0,T]\), because of \(\mathcal {A}_{\mathcal {J}}(t,\cdot ,\cdot )=\Pi _t \circ S_{\mathcal {J}}(\cdot ,\cdot ,F,\mathcal {H})\). \(\square \)
Applying similar methods one can show the following property.
Corollary 5.5
Under conditions of Theorem 4.8 a reachable map
is upper semicontinuous on \([0,T]\times R^{d}\times \mathcal {C}^{\alpha }( R^1)\), while
is of a type of Carathéodory, i.e., \(\mathcal {A}_{\mathcal {J}}\) is measurable in \((\xi ,g)\) for every fixed \(t\in [0,T]\) and continuous in t for every fixed \((\xi ,g)\in R^d\times \mathcal {C}^{\alpha }(R^1)\).
At the end of this section we focus once again on the set of optimal solutions to differential inclusion (4.1), i.e., on the set \(S_{\mathcal {J}}(\xi ,g, F,\mathcal {H})\) defined in (5.2). Since \(\mathcal {J}:\mathcal {C}( R^{d}) \rightarrow R_{+}\) is continuous by assumption, then \(\mathcal {J}\in \mathcal {C}\left( \mathcal {C}\left( R^{d}\right) ,R_{+}\right) \), where \(\mathcal {C}\left( \mathcal {C}( R^{d}),R_{+}\right) \) denotes the space of all continuous functions form \(\mathcal {C}(R^{d})\) to \(R_{+}\). We consider this space with a topology of uniform convergence on compacts (uc topology), i.e., for \(\mathcal {J}_{n},\mathcal {J}\in \mathcal {C}\left( \mathcal {C}( R^{d}),R_{+}\right) \), \(n\ge 1\)
for every \(K\in Comp \left( \mathcal {C}(R^{d}\right) \). Let us consider set-valued functions
and
We show that these set-valued functions are upper semicontinuous. In view of comments stated after Proposition 4.12 it is equivalent to the following result.
Theorem 5.6
Let assumptions of Theorem 4.8 be satisfied. Suppose that functionals \(\mathcal {J}_{n},\mathcal {J}\in \mathcal {C}\left( C(R^{d}),R_{+}\right) \), \(n\ge 1\) satisfy \(\mathcal {J}_{n}\rightarrow ^{uc}\mathcal {J}\). Then
Proof
We will prove (5.3). To simplify the notation let \(K_{0}:=S(\xi ,g, F,\mathcal {H})\) denote the set of all solutions to differential inclusion (4.1). By Corollary 4.9\(K_{0}\) is compact in \(\mathcal {C}( R^{d}) \) and for each \(n\ge 1\) the sets
and
are compact as well. Let \(2^{K_0}\) denote a family of all subsets of a set \(K_0\). We define set-valued functions \(G_{n}:R_{+}\rightarrow 2^{K_{0}}\) and \(G:R_{+}\rightarrow 2^{K_{0}}\) by formulas
In this notation if \(u_{\mathcal {J}_{n}}:=\sup \{ \mathcal {J}_{n}(y):y\in K_{0}\} \), then we have \(G_{n}(u_{\mathcal {J}_{n}})=S_{\mathcal {J}_{n}}(\xi ,g, F,\mathcal {H})\) and similarly if \(u_{\mathcal {J}}:=\sup \left\{ \mathcal {J}(y):y\in K_{0}\right\} \), then \(G(u_{\mathcal {J}})=S_{\mathcal {J}}(\xi ,g, F,\mathcal {H}). \) Thus in order to show (5.3) it is sufficient to show that
provided \(\mathcal {J}_{n}\rightarrow ^{uc}\mathcal {J}\). Therefore, assume \(\mathcal {J}_{n}\rightarrow ^{uc}\mathcal {J}\). Then for each \(K\in Comp\left( \mathcal {C}(R^{d})\right) \) we get
Suppose that (5.8) does not hold. Then there exists \(\epsilon >0\) such that for every \(k\ge 1\) one can select \(n_{k}\ge k\) satisfying
where \(\textrm{dist}_{\mathcal {C}}(y,A)=\inf _{x\in A}\Vert y-x\Vert _{\infty }\). Thus one can choose a sequence \(\left( y_{n_{k}}\right) \) such that \(y_{n_{k}}\in G_{n_{k}}(u_{\mathcal {J}_{n_{k}}})\) and
On the other hand, since \(( y_{n_{k}}) _{k\ge 1}\) is contained in a compact set \( K_{0}\) then we may assume (passing if needed to a subsequence) that \(( y_{n_{k}}) _{k\ge 1}\) uniformly converges to some \(y_{0}\in K_{0}\). But \(y_{n_{k}}\in G_{n_{k}}(u_{\mathcal {J}_{n_{k}}})\) and therefore, we obtain \(\mathcal {J}_{n_{k}}(y_{n_{k}})=u_{\mathcal {J}_{n_{k}}}\) for every \(k\ge 1\) by (5.7). Hence by (5.9) we get \(\mathcal {J}_{n_{k}}(y_{n_{k}})\rightarrow u_{\mathcal {J}}\) as \(k\rightarrow \infty \). Since \(\mathcal {J}_{n}\rightarrow ^{uc}\mathcal {J}\), then we claim \(\mathcal {J}_{n_{k}}(y_{n_{k}})\rightarrow \mathcal {J}(y_{0})\). Therefore, \(\mathcal {J}(y_{0})=u_{\mathcal {J}}\) what means that \(y_{0}\in G(u_{\mathcal {J}})\). Recall that for a metric space X and a fixed nonempty and closed set \(A\subset X\) the function
is Lipschitz continuous with a Lipschitz constant one. Thus
what contradicts with (5.10). This proves the formula (5.3).
From (5.3) together with (4.23), it follows that the set-valued function \(J\rightarrow S_{\mathcal {J}}(\xi ,g, F,\mathcal {H})\) is u.s.c. Therefore, \(J\rightarrow \mathcal {A}_{\mathcal {J}}(t,\xi ,g )\) is u.s.c. also, because \(\mathcal {A}_{\mathcal {J}}(t, \xi ,g )=\Pi _t \circ S_{\mathcal {J}}(\xi ,g, F,\mathcal {H})\). Then formula (5.4) follows by (4.23). This finishes the proof. \(\square \)
Availability of data and material
Not applicable.
References
Ahmed, N.U.: Optimal relaxed controls for nonlinear infinite dimensional stochastic differential inclusions, In: Dekker, M. (ED.) Lecture Notes in Pure and Appl. Math. 180, 1–19 (1994)
Aubin, J.P.: Viability Theory. Birkhäuser, Boston (1991)
Aubin, J.P., Cellina, A.: Differential Inclusions. Springer, Cham (1984)
Aubin, J.P., Da Prato, G.: The viability theorem for stochastic differential inclusions. Stoch. Anal. Appl. 16, 1–15 (1998)
Aubin, J.P., Frankowska, H.: Set-Valued Analysis. Birkhäuser, Boston (1990)
Bailleul, I., Brault, A., Coutin, L.: Young and rough differential inclusions. Rev. Mat. Iberoamer. Eur. Math. Soc. 37(4), 1489–1512 (2021)
Cass, T., Hairer, M., Litterer, C., Tindel, S.: Smoothness of the density for solutions to Gaussian rough differential equations. Ann. Probab. 43(1), 188–239 (2015)
Cernea, A.: A note on the solution set of a fractional integro-differential inclusion. Prog. Fract. Differ. Appl. 2, 1–7 (2016)
Cernea, A.: On the solution set of a second order integro-differential inclusion. Libertas Math. 40, 27–36 (2020)
Coutin, L., Qian, Z.: Stochastic analysis, rough paths analysis and fractional Brownian motions. Probab. Theory Related Fields 122, 108–140 (2002)
Coutin, L., Marie, N., Raynaud De Fitte, P.: On set-valued Young integral with applications to differential inclusions. J. Math. Anal. Appl. 512(1), 126104 (2022)
Denkowski, Z., Migórski, S., Papageorgiou, N.S.: An Introduction to Nonlinear Analysis and Its Applications, Part I. Kluwer Academic Publishers, Boston (2003)
Djebali, S., Górniewicz, L., Ouahab, A.: Solutions Sets for Differential Equations and Inclusions. Series in Nonlinear Analysis and Applications, De Gruyter GmbH, Berlin (2013)
Friz, P.K., Victoir, N.B.: Multidimensional Stochastic Processes as Rough Paths: Theory and Applications. Cambridge Studies in Advanced Mathematics, Cambridge University Press, Cambridge (2010)
Friz, P.K., Zhang, H.: Differential equations driven by rough paths with jumps. J. Differ. Equ. 264, 6226–6301 (2018)
Gubinelli, M.: Controlling rough paths. J. Funct. Anal. 216, 86–140 (2004)
Hiai, F., Umegaki, H.: Integrals, conditional expectations and martingales for multivalued functions. J. Multivariate Anal. 7(1), 149–182 (1977)
Hiai, F.: Multivalued stochastic integrals and stochastic differential inclusions, Division of Applied Mathematics, Research Institute of Applied Electricity, Sapporo 060, Japan, (Unpublished manuscript)
Hu, S., Papageorgiou, N.S.: Handbook of Multivalued Analysis Theory, vol. 1. Kluwer Academic Publishers, Dordrecht (1997)
Kisielewicz, M.: Differential Inclusions and Optimal Control. Kluwer Academic Publisher, Dordrecht (1991)
Kisielewicz, M.: Stochastic Differential Inclusions and Applications. Springer, New York (2013)
Kisielewicz, M.: Properties of solution set of stochastic inclusions. J. Appl. Math. Stoch. Anal. 6, 217–236 (1993)
Kisielewicz, M.: Weak compacteness of solution sets to stochastic differential inclusions with convex right-hand side. Topol. Methods Nonlin. Anal. 18, 149–171 (2001)
Kisielewicz, M.: Set-Valued Stochastic Integrals and Applications. Springer, New York (2020)
Kisielewicz, M., Michta, M.: Weak solutions of set-valued stochastic differential equations. J. Math. Anal. Appl. 473, 1026–1052 (2019)
Kisielewicz, M., Michta, M., Motyl, J.: Set-valued approach to stochastic control. Part: I and II. Dyn. Syst. Appl. 12(3–4), 405–466 (2003)
Lejay, A.: Controlled differential equations as Young integrals: a simple approach. J. Differ. Equ. 249, 1777–1798 (2010)
Levakov, A.A.: Stochastic differential inclusions. J. Differ. Equ. 33, 212–221 (1997)
Lyons, T.: Differential equations driven by rough signals. Rev. Mat. Iberoam. 14, 215–310 (1998)
Mandelbrot, B., Van Ness, J.W.: Fractional Brownian motions, fractional noises and applications. SIAM Rev. 10, 422–437 (1968)
Michta, M.: On connections between stochastic differential inclusions and set-valued stochastic differential equations driven by semimartingales. J. Differ. Equ. 262, 2106–2134 (2017)
Michta, M., Motyl, J.: Selection properties and set-valued young integrals of set-valued functions. Results Math. 75, 164 (2020)
Michta, M., Motyl, J.: Set-valued functions of bounded generalized variation and set-valued Young integrals. J. Theor. Probab. 35, 528–549 (2022)
Motyl, J.: Viable solutions of set-valued stochastic equations. Optimization 48, 157–176 (2000)
Naghshineh, O., Zangeneh, B.Z.: Existence and measurability of the solution of the stochastic differential equations driven by fractional Brownian motion. Bull. Iran. Math. Soc. 35(2), 47–68 (2009)
Samko, S.G., Kilbas, A.A., Marichev, O.I.: Fractional Integrals and Derivatives Theory and Applications. Gordon and Breach Science Publishers, Yverdon (1993)
Tam, V.M., Hung, N.V., O’Regan, D.: An existence result for a new class of fuzzy fractional differential inclusions with Clarke’s subdifferential via resolvent operators in Banach spacess. Fuzzy Sets Syst. 443, 221–240 (2022)
Tolstonogov, A.A.: Differential inclusions in a Banach space. Kluwer Academic Publishers, Dordrecht (2000)
Wang, J., Ibrahim, A.G., O’Regan, D.: Nonemptyness and compactness of the solution set for fractional evolution inclusions with non-instantaneous impulses. Electron. J. Differ. Equ. 37, 1–17 (2019)
Young, L.S.: An inequality of the Hölder type connected with Stieltjes integration. Acta Math. 67, 251–282 (1936)
Funding
Not applicable.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no Conflict of interest.
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
Michta, M., Motyl, J. Properties of set-valued Young integrals and Young differential inclusions generated by sets of Hölder functions. Nonlinear Differ. Equ. Appl. 31, 70 (2024). https://doi.org/10.1007/s00030-024-00963-2
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00030-024-00963-2
Keywords
- Hölder-continuous function
- Set-valued Young integral
- Young differential inclusion
- Continuity of solutions set
- optimal solution