Abstract
In this paper, we consider a locally convex cone \(({\mathcal {P}},{\mathcal {V}})\) and verify the dual of \((Conv({\mathcal {P}}),{\overline{{\mathcal {V}}}})\) the locally convex cone of the non-empty convex subsets of \({\mathcal {P}}\). Under some semilattice conditions, we characterize the dual of \(Conv(\underbrace{\cdots }_{n\ times} (Conv({\mathcal {P}}))\).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Duality theory is a powerfull technique to study a wide class of related problems in pure and applied mathematics. For example the Hahn-Banach extension and separation theorems studied by means of duals (see [8]). The collection of all non-empty convex subsets of a cone (or a vector space) is interesting in convexity and approximation theory (for example see [5]). This collection is a cone. We consider the non-empty convex subsets of a cone \({\mathcal {P}}\), denoted by \(Conv({\mathcal {P}})\), and verify the dual of it, when \({\mathcal {P}}\) is a locally convex cone. We note that some elements of the dual of \(Conv({\mathcal {P}})\) have already been introduced (see [6], I: Example 2.1(e) and Example 5.31 (b)). Firstly we review the structure of locally convex cones briefly:
A nonempty set \({\mathcal {P}}\) endowed with an addition and a scalar multiplication for nonnegative real numbers is called a cone whenever the addition is associative and commutative, there is a neutral element \(0 \in {\mathcal {P}}\) and for the scalar multiplication the usual associative and distributive properties hold, that is \(\alpha (\beta a) = (\alpha \beta )a\), \((\alpha + \beta )a = \alpha a + \beta a\), \(\alpha (a + b) = \alpha a + \alpha b\), \(1a = a\) and \(0a = 0\) for all \(a, b \in {\mathcal {P}}\) and nonnegative reals \(\alpha \) and \( \beta \).
The theory of locally convex cones as introduced and developed by K. Keimel and W. Roth in [4]. It uses an order theoretical concept or a convex quasi-uniform structure on a cone. In this paper, we use the former. For some recent researches see [1,2,3, 7].
A (preordered cone) is a cone \({\mathcal {P}}\) endowed with a preorder (reflexive transitive relation) \(\le \) which is compatible with the addition and scalar multiplication, that is \(x\le y\) implies \(x+z\le y+z\) and \(r\cdot x\le r\cdot y\) for all \(x,y,z\in {\mathcal {P}}\) and \(r\in {\mathbb {R}}_+=\{r\in {\mathbb {R}} \ : \ r\ge 0\}.\) Every ordered vector space is an ordered cone. The cones \({\overline{{\mathbb {R}}}}={\mathbb {R}}\cup \{+\infty \}\) and \({\overline{{\mathbb {R}}}}_+={\mathbb {R}}_+\cup \{+\infty \}\), with the usual order and algebraic operations (specially \(0\cdot (+\infty )=0\)), are ordered cones that are not embeddable in vector spaces.
A subset \({\mathcal {V}}\) of a preordered cone \({\mathcal {P}}\) is called an (abstract) 0-neighborhood system, if
- \((v_1)\):
-
\(0<v\) for all \(v\in {\mathcal {V}}\);
- \((v_2)\):
-
for all \(u,v\in {\mathcal {V}}\) there is a \(w\in {\mathcal {V}}\) with \(w\le u\) and \(w\le v\);
- \((v_3)\):
-
\(u+v\in {\mathcal {V}}\) and \(\alpha v\in {\mathcal {V}}\) whenever \(u,v\in {\mathcal {V}}\) and \(\alpha >0 \).
Let \(a\in {\mathcal {P}}\) and \(v\in {\mathcal {V}}\). We define \(v(a)=\{b\in {\mathcal {P}}\ | \ b\le a+v\}\), resp. \((a)v=\{b\in {\mathcal {P}}\ | \ a\le b+v\},\) to be a neighborhood of a in the upper, resp. lower topologies on \({\mathcal {P}}\). The common refinement of the upper and lower topologies is called symmetric topology. We denote the neighborhoods of a in the symmetric topology by v(a)v. The pair \(({\mathcal {P}},{\mathcal {V}})\) is called a full locally convex cone if the elements of \({\mathcal {P}}\) are bounded below, i.e. for every \(a\in {\mathcal {P}}\) and \(v\in {\mathcal {V}}\) we have \(0\le a+\rho v\) for some \(\rho >0\). Each subcone of \({\mathcal {P}}\), not necessarily containing \({\mathcal {V}}\), is called a locally convex cone.
We note that if \(({\mathcal {Q}},{\mathcal {V}})\) is a locally convex cone, \({\mathcal {Q}}\oplus ({\mathcal {V}}\cup \{0\})\) with the algebraic operation
and the preorder
for all \(a,b\in {\mathcal {Q}}\), \(v_1,v_2\in {\mathcal {V}}\) and \(\alpha \in {\mathbb {R}}^+\), \(({\mathcal {Q}}\oplus ({\mathcal {V}}\cup \{0\}),{\mathcal {V}})\) is a full locally convex cone which \({\mathcal {Q}}\) and \({\mathcal {V}}\) can be embedded in \({\mathcal {Q}}\oplus ({\mathcal {V}}\cup \{0\})\) by the mappings \(a\rightarrow (a,0)\) and \(v\rightarrow (0,v)\) for all \(a\in {\mathcal {Q}}\) and \(v \in {\mathcal {V}}\).
For cones \({\mathcal {P}}\) and \({\mathcal {Q}}\) a mapping \(t:{\mathcal {P}}\rightarrow {\mathcal {Q}}\) is called a linear operator if \(t(a+b)=t(a)+t(b)\) and \(t(\alpha a)=\alpha t(a)\) hold for \(a,b\in {\mathcal {P}}\) and \(\alpha \ge 0\).
A linear functional on a cone \({\mathcal {P}}\) is a linear mapping \(\mu :{\mathcal {P}}\rightarrow {\overline{{\mathbb {R}}}}\).
Let \(({\mathcal {P}}, {\mathcal {V}})\) and \(({\mathcal {Q}},{\mathcal {W}})\) be two locally convex cones. The linear operator \(t:({\mathcal {P}}, {\mathcal {V}})\rightarrow ({\mathcal {Q}},{\mathcal {W}})\) is called uniformly continuous or simply u-continuous if for every \(w\in {\mathcal {W}}\) one can find a \(v\in {\mathcal {V}}\) such that \(a\le b+v\) implies \(t(a)\le t(b)+w\). It is easy to see that the u-continuity implies continuity with respect to the upper, lower and symmetric topologies on \({\mathcal {P}}\) and \({\mathcal {Q}}\).
According to the definition of u-continuity, a linear functional \(\mu \) on \(({\mathcal {P}}, {\mathcal {V}})\) is u-continuous if there is a \(v\in {\mathcal {V}}\) such that \(a\le b+v\) implies \(\mu (a)\le \mu (b)+1\). The u-continuous linear functionals on a locally convex cone \(({\mathcal {P}},{\mathcal {V}})\) (into \({\overline{{\mathbb {R}}}} \)) form a cone with the usual addition and scalar multiplication of functions. This cone is called the dual cone of \({\mathcal {P}}\) and denoted by \({\mathcal {P}}^*\).
For a locally convex cone \(({\mathcal {P}},{\mathcal {V}})\), the polar \(v^{\circ }\) of \(v\in {\mathcal {V}}\) consists of all linear functionals \(\mu \) on \({\mathcal {P}}\) satisfying \(\mu (a)\le \mu (b)+1\) whenever \(a\le b+v\) for \(a,b\in {\mathcal {P}}\). We have \(\cup \{v^{\circ }: v\in {\mathcal {V}}\}={\mathcal {P}}^*\). The cones \({\overline{{\mathbb {R}}}}\) and \({\overline{{\mathbb {R}}}}_{+}=\{a\in {\overline{{\mathbb {R}}}}:a\ge 0 \}\) with (abstract) 0-neighborhood \({\mathcal {V}}=\{\varepsilon >0:\varepsilon \in {\mathbb {R}}\}\) are locally convex cones. The dual cones of \({\overline{{\mathbb {R}}}}\) and \({\overline{{\mathbb {R}}}}_{+}\) under \({\mathcal {V}}\) consists of all nonnegative reals and the functional \(0_{\infty }\) such that \(0_{\infty }(a)=0\) for all \(a\in {\mathbb {R}}\) and \(0_{\infty } (+\infty )=+\infty \).
2 Dual of the cone of non-empty convex sets of a locally convex cone
A subset A of a cone \({\mathcal {P}}\) is said convex, if \(\lambda a+(1-\lambda ) b\in A\), whenever \(a,b\in {\mathcal {P}}\) and \(0\le \lambda \le 1\). Let \({\mathcal {P}}\) be a preordered cone and \(Conv({\mathcal {P}})\) be the cone of all non-empty convex subsets of \({\mathcal {P}}\), endowed with the usual addition and multiplication of sets by non-negative scalars, that is \(\alpha A = \{\alpha a\ |\ a \in A\} \) and \(A + B = \{a + b\ |\ a \in A \ and \ b \in B\}\) for \(A, B \in Conv({\mathcal {P}}) \) and \(\alpha \ge 0\). We consider the order on \(Conv({\mathcal {P}})\) by
where \( \downarrow B = \{x \in {\mathcal {P}}| x \le b \ \text{ for } \text{ some } \ b \in B\}\) is the decreasing hull of the set B in \({\mathcal {P}}\). Note that \( \downarrow B\) is again a convex subset of \({\mathcal {P}}\). The requirements for a preordered cone are easily checked. The neighborhood system in \(Conv({\mathcal {P}})\) is \({\overline{{\mathcal {V}}}}:=\{ {\overline{v}}=\{v\}\ |\ v\in {\mathcal {V}}\}\), that is
for \(A, B \in Conv({\mathcal {P}})\) and \({\overline{v}} \in {\overline{{\mathcal {V}}}}\). The cone \(Conv({\mathcal {P}})\) with (abstract) 0-neighborhood system \({\overline{{\mathcal {V}}}})\) is a locally convex cone. Via the embedding \( x\rightarrow \{x\}: {\mathcal {P}}\rightarrow Conv({\mathcal {P}})\) the preordered cone \({\mathcal {P}}\) itself may be considered as a subcone of \(Conv({\mathcal {P}})\) (see [6], I, Example 1.4 (c)).
Definition 1
We say that a preordered cone \({\mathcal {P}}\) is a \(\bigvee \)-semilattice cone if the order of \({\mathcal {P}}\) is antisymmetric and if
\((\bigvee 1)\) every non-empty subset \(A\subseteq {\mathcal {P}}\) has a supremum \(\sup A\in {\mathcal {P}}\) and \(\sup (A+b)=\sup A+b\) hold for all \(b\in {\mathcal {P}}\).
Moreover, if \({\mathcal {P}}\) with an abstract neighborhood system \({\mathcal {V}}\) is a locally convex cone and
\((\bigvee 2)\) for \(\emptyset \ne A\subseteq {\mathcal {P}}\), \(b\in {\mathcal {P}}\) and \(v\in V\) such that \(a\le b+v \) for all \(a\in A\), we have \(\sup A\le b+v\),
then \(({\mathcal {P}}, {\mathcal {V}})\) is said a \(\bigvee \)-semilattice locally convex cone.
In particular, every \(\bigvee \)-semilattice cone \({\mathcal {P}}\) contains a largest element, that is \(+\infty =\sup {\mathcal {P}},\) which can be adjoined as a maximal element to any \(\bigvee \)-semilattice cone with the convention that \(a+(+\infty )=+\infty \), \(\alpha \cdot (+\infty )=+\infty \), \(0 \cdot (+\infty )=0\) and \(a\le +\infty \) for all \(a\in {\mathcal {P}}\) and \(\alpha > 0\).
Remark 1
We note that the condition \((\bigvee 2)\) of definition 1 is necessary and the definition of supremum does not imply this condition in locally convex cones necessarily. We show this in the following example.
Example 1
Let \({\mathbb {R}}\) be as a cone and \({\mathcal {V}}=\{{\bar{\epsilon }}=(-\infty ,\epsilon ) \ : \ \epsilon \in {\mathbb {R}}_{>0}\}\).
Let
We define
and
for all \( (a,B), (c,D)\in {\mathcal {P}}\). Also, we define the preorder
for all \( (a,B), (c,D)\in {\mathcal {P}}\). Then \(({\mathcal {P}},{\mathcal {V}})\) is a full locally convex cone. Now, we can embedded \({\mathbb {R}}\) in \({\mathcal {P}}\) by \(a\rightarrow (a,\{0\})\) and we can consider \({\mathbb {R}}\) as a subcone of \({\mathcal {P}}\). We have
Now, for the set \(A=(0,5)\subseteq {\mathbb {R}}\), by considering the embedding, we have \({\bar{A}}=\{(a,\{0\}) \ : \ a\in (0,5)\}\). Let \(b=4\) and \({\bar{1}}=(-\infty , 1)\in {\mathcal {V}}\). Then
for all \((a,\{0\})\in {\bar{A}}\), i.e.
for all \(a\in A=(0,5)\). On the other hand, \(\sup A=5\) (in \({\mathbb {R}}\)) and we have
i.e. \(5\not \le 4+{\bar{1}}\). Although, \({\mathcal {P}}\) is not a \(\bigvee \)-semilattice cone, \({\mathbb {R}}\) is a \(\bigvee \)-semilattice cone. Also, the locally convex cone \(({\mathbb {R}},{\mathcal {V}})\) is not a \(\bigvee \)-semilattice locally convex cone.
Remark 2
We note that definition 1 is similar to the definition of “locally convex \(\bigvee \)-semilattice cone" in [6], I, 5.4. In this definition, the order do not coincide with the weak preorder necessarily.
We define \(Conv^n({\mathcal {P}}):=Conv(Conv^{n-1}({\mathcal {P}}))\) for \(n=2,3,\ldots \) and \(Conv^1({\mathcal {P}})=Conv({\mathcal {P}})\). Let
for all \(a\in {\mathcal {P}}\). It is easy to see that \(\{a\}^n \in Conv^n({\mathcal {P}})\) for all \(n\in {\mathbb {N}}\). This shows that \({\mathcal {P}}\) is embedded in \( Conv^n({\mathcal {P}})\) (the mapping \(a\longrightarrow \{a\}^n\) is the embedding). The cone \( Conv^n({\mathcal {P}})\) with the (abstract) 0-neighborhood system \({\overline{{\mathcal {V}}}}^n\) is a locally convex cone, where \({\overline{{\mathcal {V}}}}^n:=\{{\bar{v}}^n:= \{v\}^n\ |\ v\in {\mathcal {V}}\}\).
Example 2
For the cone \({\mathbb {R}}\), we have \(A^1=[0,1] \in {Conv}(\overline{{\mathbb {R}}})\) , \(A^2=\{ [0,a]\ |\ ,a \in [0,1] \} \) is an element of \( Conv^2({\mathbb {R}})\) and \(A^3=\{\{ [0,a]\ |\ ,a \in [0,b] \}\ |\ b \in [0,1] \}\) is an element of \( Conv^3({\mathbb {R}})\).
For the element \(A^n\) of \(Conv^n({\mathcal {P}})\) we define
for \(n=2,3, \ldots \) and \(sup^s(A^1)=\sup A\). It is easy to see that \(sup^s(A^n)\in {\mathcal {P}}\) for all \(n\in {\mathbb {N}}\).
The following lemma is an special case of Lemma 5.5 of [6].
Lemma 1
Let \({\mathcal {P}}\) be a \(\bigvee -\)semilattice cone and \(\{A_i\}_{i\in I}\) be a collection of non-empty subsets of \({\mathcal {P}}\). Then
Proof
Let \(a \in \bigcup _{i\in I} A_i\) be arbitrary. Then there exists \(i \in I\) such that \(a \in A_i\). We have \(a\le \sup A_i\) and so \(a\le \sup \{\sup A_i \ | \ i \in I\}\). Then
On the other hand, \(\sup A_i\le \sup (\bigcup _{i \in I} A_i)\) for all \(i \in I\). This conclude that
\(\square \)
Remark 3
We note that \(A^2\in Conv^2({\mathcal {P}})\) but the elements of \(A^2\) belong to \(Conv^1({\mathcal {P}})= Conv({\mathcal {P}})\). This implies that the union of the elements of \(A^2\) (\(\bigcup _{A^1\in A^2} A^1\)) belongs to the power set of \({\mathcal {P}}\). Also, \(A^3\in Conv^3({\mathcal {P}})\) and the elements of \(A^3\) belong to \(Conv^2({\mathcal {P}})\). Then the union of the elements of \(A^3\) (\(\bigcup _{A^2\in A^3} A^2\)) belongs to the power set of \(Conv^2({\mathcal {P}})\) and the union of these sets (\(\bigcup _{A^2\in A^3}\bigcup _{A^1\in A^2} A^1\)) belongs again to the power set of \({\mathcal {P}}\). By continuing this process, we conclude that \(A^n\in Conv^n({\mathcal {P}})\) and the elements of \(A^n\) belong to \(Conv^{n-1}({\mathcal {P}})\). Then
belongs to the power set of \({\mathcal {P}}\). By Lemma 1, we have
Let \({\mathcal {P}}\) be a cone and \(\mu :{\mathcal {P}}\rightarrow \overline{{\mathbb {R}}} \) be a functional. We define
moreover, if \({\mathcal {P}}\) is a \(\bigvee \)-semilattice cone, we define
Lemma 2
Let \(({\mathcal {P}}, {\mathcal {V}})\) be a locally convex cone and \(\mu \in {\mathcal {P}}^*\). Then \({\overline{\mu }} \in Conv({\mathcal {P}})^*\). Moreover, if \(({\mathcal {P}}, {\mathcal {V}})\) is \(\bigvee \)-semilattice locally convex cone, then \(\overline{{\overline{\mu }}} \in Conv({\mathcal {P}})^*\).
Proof
We have
for all \(A,B \in Conv({\mathcal {P}})\) and all \(\alpha \ge 0\). So \({\overline{\mu }}\) is linear.
Now, if \(({\mathcal {P}}, {\mathcal {V}})\) is \(\bigvee \)-semilattice locally convex cone, then
This yields that \(\mu (\sup (A+B))=\mu (\sup (A))+\mu (\sup (B))\) and then \(\overline{{\overline{\mu }}}(A+B)=\overline{{\overline{\mu }}}(A)+\overline{{\overline{\mu }}}(B)\) for all \(A,B \in Conv({\mathcal {P}})\). Also,
for all \(\alpha \ge 0\) and \(A\in Conv({\mathcal {P}})\). Therefore \(\overline{{\overline{\mu }}}\) is linear.
Now, we show that \({\overline{\mu }}\) and \(\overline{{\overline{\mu }}}\) are u-continuous extensions of \(\mu \) to \(Conv({\mathcal {P}})\). Via of continuity of \(\mu \), there is a \(v\in {\mathcal {V}}\) such that \(a\le b+v\) implies \(\mu (a)\le \mu (b)+1\). Let \(A\preceq B+\{v\}\). Then, for each \(a\in A\) there exists \(b\in B\) such that \( a\le b+v\). We have
This shows that \({\overline{\mu }}\) is u-continuous. Also if \(({\mathcal {P}}, {\mathcal {V}})\) is \(\bigvee \)-semilattice locally convex cone, we have
This yields that \(\overline{{\overline{\mu }}}\) is u-continuous.
\(\square \)
Proposition 1
Let \({\mathcal {P}}\) be a preordered cone, \(\mu \) be a monotone functional on \({\mathcal {P}}\) and \({\widetilde{\mu }}\) be a monotone extension of \(\mu \) on \(Conv({\mathcal {P}}) \). Then \({\overline{\mu }}\le {\widetilde{\mu }}\). Furthermore, if \({\mathcal {P}}\) is a \(\bigvee \)-semilattice cone, then
Proof
Let \({\overline{\mu }}\nleq {\widetilde{\mu }}\). Then there exists \(A\in Conv({\mathcal {P}})\) such that \({\overline{\mu }}(A)\nleq {\widetilde{\mu }}(A)\) i.e. \({\widetilde{\mu }}(A)< {\overline{\mu }}(A)=\sup \{\mu (a)\ | \ a\in A \}\). Then there exists \(a\in A\) such that \({\widetilde{\mu }}(A)<\mu (a)={\widetilde{\mu }}(\{a\})\) (by the supremum property). On the other hand, \(\{a\}\preceq A\) and so \({\widetilde{\mu }}(\{a\})\le {\widetilde{\mu }}(A)\). This contradiction yields that \({\overline{\mu }}\le {\widetilde{\mu }}\).
Now, let \({\mathcal {P}}\) be a \(\bigvee \)-semilattice cone. Let \( A\in Conv({\mathcal {P}})\) be arbitrary. We have \(A\preceq \{\sup A\}\). Then \({\widetilde{\mu }}(A)\le {\widetilde{\mu }}(\{sup A\})=\mu (sup A)=\overline{{\overline{\mu }}}(A)\). \(\square \)
Let \({\mathcal {P}}\) be a \(\bigvee \)-semilattice cone. We denote
where \({\mathcal {L}}({\mathcal {P}})\) is the cone of all linear functionals on \({\mathcal {P}}\).
Corollary 1
Let \({\mathcal {P}}\) be a \(\bigvee -\)semilattice cone. Then the elements of \(\Omega ({\mathcal {P}})\) have unique extensions to \({Conv}({\mathcal {P}})\).
By the assumptions of the Corollary 1, we conclude that the elements of \(\Omega ({\mathcal {P}})\) have unique extensions to \({Conv}^n({\mathcal {P}})\).
Proposition 2
Let \({\mathcal {P}}\) be a \(\bigvee -\)semilattice cone. Then
for all \(n\in {\mathbb {N}}\) and \(A^{n},B^{n}\in Conv^n({\mathcal {P}})\).
Proof
For \(n=1\), let \(A^1=A\) and \(B^1=B\) be elements of \(Conv^1({\mathcal {P}})=Conv({\mathcal {P}})\). We have
Now, let
Then
\(\square \)
Let coh(F) denote the convex hull of the set F, the smallest convex set containing F. We set
for \(n=2,3, \ldots \) and \(coh^s(A^1) =coh(A\cup \{sup(A)\})\).
Proposition 3
Let \({\mathcal {P}}\) be a \(\bigvee -\)semilattice cone. Then
all \(A^{n}\in Conv^n({\mathcal {P}})\) and \(n\in {\mathbb {N}}\).
Proof
First we show that \(\sup (coh^s(A^1))=\sup (A^1)\). Let \(x\in coh^s(A^1)\) be arbitrary. Then there are \(\lambda _1,\lambda _2,\ldots ,\lambda _k\ge 0\) and \(a_1,a_2,\ldots ,a_k\in A^1\cup \{\sup (A^1)\} \) such that \(\sum _{i=1}^k \lambda _i=1\) and \(x=\sum _{i=1}^k \lambda _i a_i\). On the other hand, \(\lambda _i a_i\le \lambda _i \sup (A^1)\) for all \(i=1,2,\ldots ,k\). We have
This yields that
since \(\sup (A^1)\in coh^s(A^1)\).
Now, let \(\sup (coh^s(A^{n-1}))=\sup (A^{n-1})\) for all \(A^{n-1}\in Conv^{n-1}({\mathcal {P}})\). Consider \(A^{n}\in Conv^{n}({\mathcal {P}})\) and \({\mathcal {X}}\in coh^s(A^n)\). Then there are \(\lambda _1,\lambda _2,\ldots ,\lambda _k\ge 0\) and \(A^{n-1}_1,A^{n-1}_2,\ldots ,A^{n-1}_k\in A^n\cup \{\sup (A^n)\}^n \) such that \(\sum _{i=1}^k \lambda _i=1\) and \({\mathcal {X}}=\sum _{i=1}^k \lambda _i coh^s(A^{n-1}_i).\) On the other hand,
for all \(i=1,2,\ldots ,k\). So
and so
Since \(\{sup^s ( A^n)\}^n\in coh^s(A^{n})\), we have
\(\square \)
Remark 4
By Proposition 3 and by considering the construction of \(coh^s(A^{n})\), we have
for all \(n\in {\mathbb {N}}\).
Example 3
For the cone \(\overline{{\mathbb {R}}}\), we have \( \{0\},\{0,+\infty \} \in {Conv}(\overline{{\mathbb {R}}})\) and \(A^2=\{ \{0\},\{0,+\infty \} \} \) is an element of \( Conv^2(\overline{{\mathbb {R}}})\). We have \(sup ^s(A^2)=\sup \{0,+\infty \}=+\infty \) and \(coh^s(A^2)=\{\{0\},\{0,+\infty \} ,\{+\infty \} \}\).
For every positive integer n we introduce
Theorem 1
Let \({\mathcal {P}}\) be a \(\bigvee -\)semilattice cone. Then \({Conv_s}^n({\mathcal {P}})\) is a subcone of \({Conv}^n({\mathcal {P}})\) for all \(n\in {\mathbb {N}}\).
Proof
Let \({\mathcal {A}},{\mathcal {B}}\in {Conv_s}^1({\mathcal {P}})\). Then there exist \(A^1,B^1\in Canv({\mathcal {P}})\) such that
and
We conclude that \({\mathcal {A}},{\mathcal {B}} \in Canv({\mathcal {P}})\). Put \({\mathcal {A}}+{\mathcal {B}}={\mathcal {C}}\). We have
by Proposition 2 (for case \(n=1\)). Since \({\mathcal {A}},{\mathcal {B}}\) contain their suprema, then \({\mathcal {C}}\) contains its supremum. Hence
which conclude that \({\mathcal {C}}\in {Conv_s}^1({\mathcal {P}})\). On the other hand, for each \(\alpha \ge 0\),
and so \(\alpha {\mathcal {A}}\in {Conv_s}^1({\mathcal {P}})\). Hence \( {Conv_s}^1({\mathcal {P}})\) is a subcone of \( {Conv}({\mathcal {P}})\). For completion of induction, first we show that \({Conv_s}^{n+1}({\mathcal {P}})\subseteq {Conv}^{n+1}({\mathcal {P}})\). For this, let \({\mathcal {A}}\in {Conv_s}^{n+1}({\mathcal {P}})\). There is \(A^{n+1}\in Canv^{n+1}({\mathcal {P}})\) such that \({\mathcal {A}}=coh^s(A^{n+1}) \) and so
Let \({\mathcal {X}}\in {\mathcal {A}}\) be arbitrary. There exist \(\lambda _1,\lambda _2,\ldots ,\lambda _k\ge 0\) and \(A^{n}_1,A^{n}_2,\ldots ,A^{n}_k\in A^{n+1}\cup \{\sup (A^{n+1})\}^{n} \) such that \(\sum _{i=1}^k \lambda _i=1\) and \({\mathcal {X}}=\sum _{i=1}^k \lambda _i coh^s(A^{n}_i).\) On the other hand, \(\lambda _i coh^s(A^{n}_i)\in {Conv_s}^{n}({\mathcal {P}})\), for all \(i=1,2,\ldots ,k\). Hence
and then \({\mathcal {A}}\in Canv^{n+1}({\mathcal {P}})\).
Now, let \({\mathcal {A}},{\mathcal {B}}\in {Conv_s}^{n+1}({\mathcal {P}})\). Then for all \( {\mathcal {X}}\in {\mathcal {A}}\subseteq Canv^{n}_s({\mathcal {P}})\) and \( {\mathcal {Y}}\in {\mathcal {B}}\subseteq Canv^{n}_s({\mathcal {P}})\), we have \({\mathcal {X}}+{\mathcal {Y}}\in Canv^{n}_s({\mathcal {P}})\). By Proposition 3 and Remark 4, we have \(\{\sup ^s({\mathcal {A}})\}^n\in {\mathcal {A}}\) and \(\{\sup ^s({\mathcal {B}})\}^n\in {\mathcal {B}}\). Also, by Proposition 2, we have
and then
Now, by considering the properties of sup and coh (convex hull of a set), we have \(\alpha {\mathcal {A}}\in {Conv_s}^{n+1}({\mathcal {P}})\) for all \(\alpha \ge 0\) and \({\mathcal {A}}\in {Conv_s}^{n+1}({\mathcal {P}})\). \(\square \)
Now, we characterize the elements of \({Conv_s}^n({\mathcal {P}})^*\). First we recall a theorem.
Theorem 2
([4], II, 2.9) Let \({\mathcal {Q}}\) be subcone of the locally convex cone \(({\mathcal {P}}, {\mathcal {V}})\). Then every u-continuous linear functional on \({\mathcal {Q}}\) can be extended to a u-continuous linear functional on \({\mathcal {P}}\).
Theorem 3
If \(({\mathcal {P}},{\mathcal {V}})\) is a \(\bigvee \)-semilattice locally convex cone, then for all \(n\in {\mathbb {N}}\), \((Conv^n({\mathcal {P}}))^*\) and \({\mathcal {P}}^*\) coincide, in the sense that any vector of \({\mathcal {P}}^*\) has a unique extension to a vector of \((Conv^n({\mathcal {P}}))^*\) and conversely any vector \((Conv^n({\mathcal {P}}))^*\) can be restricted to a vector of \({\mathcal {P}}^*\).
Proof
By considering (1) we can embed \({\mathcal {P}}\) into \({Conv_s}^n({\mathcal {P}})\). It is easy to see that the restriction of each element of \({Conv_s}^n({\mathcal {P}})^*\) on \({\mathcal {P}}\) belongs to \({\mathcal {P}}^*\) and by Theorem 2, the extension of each element of \({\mathcal {P}}^*\) to \({Conv_s}^n({\mathcal {P}})\) is an element of \({Conv_s}^n({\mathcal {P}})^*\). So it is sufficient to show that each element of \({\mathcal {P}}^*\) has a unique extension in \({Conv_s}^n({\mathcal {P}})^*\). Let \( \mu \in {\mathcal {P}}^*\). Define \( ({\bar{\mu }})^n\) as follows:
and
for \(n=2,3,\ldots \). By Lemma 2, the functional \(({\bar{\mu }})^1\) is u-continuous and by repeating this process \(({\bar{\mu }})^n\) is u-continuous too. We have \(({\bar{\mu }})^1(A)=\mu (sup^s(A))\) and \(({\bar{\mu }})^n(A^n)=({\bar{\mu }})^{n-1}(\{sup^s({A^n})\}^{n-1})\), since A contains \(\sup A\). By Remark 4 and Proposition 2 the mapping \(({\bar{\mu }})^n\) is an extension of \(\mu \) to \({Conv_s}^n({\mathcal {P}})\). Let \(\vartheta _n\) be another u-continuous extension of \(\mu \) to \({Conv_s}^n({\mathcal {P}})\) (which exists by Theorem 2). We show that \(\vartheta _n={\bar{\mu }}^n\).
Let \(A^n \in Conv^n_s ({\mathcal {P}})\). Since \(A^n\preceq \{ {sup^s(A^n)}\}^n\) and \(\{ {sup^s(A^n)}\}^n\preceq A^n\), then \(\vartheta _n(A^n)\le \vartheta _n(\{ {sup^s(A^n)}\}^n)\) and \(\vartheta _n(\{ {sup^s(A^n)}\}^n)\le \vartheta _n(A_n)\) and so
This completes the proof. \(\square \)
In the following example we consider the locally convex cone \(\overline{{\mathbb {R}}}\) and we characterize all elements of the dual of the locally convex cone \((Conv^n(\overline{{\mathbb {R}}}),{\overline{{\mathcal {V}}}}^n)\), where \({\mathcal {V}}=\{\epsilon >0\ |\ \epsilon \in {\mathbb {R}} \}\).
Example 4
We know that \(\overline{{\mathbb {R}}}\) is a \(\bigvee -\)semilattice locally convex cone. It is easy to see that
According to Theorem 3, \((Conv^n({\mathbb {R}}))^*\) and \({\mathbb {R}}^*\) coincide, in the sense that any vector of \({\mathbb {R}}^*\) has a unique extension to a vector of \((Conv^n({\mathbb {R}}))^*\) and conversely any vector \((Conv^n({\mathbb {R}}))^*\) can be restricted to a vector of \({\mathbb {R}}^*\) for all \(n\in {\mathbb {N}}\).
Since \(\Omega (\overline{{\mathbb {R}}})=\overline{{\mathbb {R}}}^* {\setminus } \{0_\infty \}={\mathbb {R}}^*\), every element of \({\mathbb {R}}^*\) has a unique extension in \((Conv^n(\overline{{\mathbb {R}}}))^*\) by Corollary 1. The element \(0_\infty \) violates the \( \Omega \) condition at just one point \(+\infty \). So two different extensions \( \overline{ 0_\infty }(A)\) and \(\overline{\overline{0_\infty }}\) can be written for it in \( Conv (\overline{{\mathbb {R}}})^* \) as the following:
for all \( A \in Conv(\overline{{\mathbb {R}}})\) which \(\sup (A)\ne +\infty \),
for \( A\in Conv(\overline{{\mathbb {R}}})\) with \(+\infty \in A\) and
for all \(A\in {\mathcal {Q}}\), where \({\mathcal {Q}}:=\{ A\in Conv(\overline{{\mathbb {R}}}) \ | \ \sup (A)= +\infty \text{ and } +\infty \notin A\}\). Let \(\gamma \) be another extension of \(0_\infty \) to \(Conv(\overline{{\mathbb {R}}})\). Then \(\gamma (A)=\overline{ 0_\infty }(A)=\overline{\overline{0_\infty }}(A)=0\) for all \( A \in Conv(\overline{{\mathbb {R}}})\) which \(\sup (A)\ne +\infty \) and \(\gamma (A)=\overline{ 0_\infty }(A)=\overline{\overline{0_\infty }}(A)=+\infty \) for \( A\in Conv(\overline{{\mathbb {R}}})\) with \(+\infty \in A\), by Theorem 3. Now, let \(A,B\in {\mathcal {Q}}\). It is easy to see that \(A\preceq B\) and \(B\preceq A\) and then \(\gamma (A)=\gamma (B)\). In particular, \(\gamma (A)=\gamma (\alpha A)=\alpha \gamma (A) \) since \( \alpha A\in {\mathcal {Q}}\) for all positive reals \(\alpha \). By the above consideration \(\gamma =\overline{ 0_\infty }=0\) or \(\gamma =\overline{\overline{0_\infty }} =+\infty \) on \({\mathcal {Q}}\). Therefore \( \overline{ 0_\infty }\) and \(\overline{\overline{0_\infty }}\) are only extensions of \(0_\infty \) on \(Conv (\overline{{\mathbb {R}}})\). This yields that \((Conv (\overline{{\mathbb {R}}}))^*{\setminus } \{\overline{ 0_\infty },\overline{\overline{0_\infty }}\} \) and \(\overline{{\mathbb {R}}}^*\) coincide.
Now, we show that the extensions of the mappings \( \overline{ 0_\infty }\) and \(\overline{\overline{0_\infty }}\) to the cone \(Conv^n(\overline{{\mathbb {R}}})\) are unique: Let \( \overline{ 0_\infty }^n\) and \(\overline{\overline{0_\infty }}^n\) be the extensions of \( \overline{ 0_\infty }\) and \(\overline{\overline{0_\infty }}\) on \(Conv^n(\overline{{\mathbb {R}}})\), respectively. Let \(A\in Conv^n(\overline{{\mathbb {R}}}){\setminus } Conv^n({{\mathbb {R}}}) \). Then \(\{+\infty \}^n \preceq A\) and \(A \preceq \{+\infty \}^n\). These yield that
On the other hand, if \(A\in Conv^n({{\mathbb {R}}}) \), then \(A\preceq \{(0,+\infty )\}^{n-1}\) and so \(\overline{ 0_\infty }^n(A)\le 0\). Also there exists \(a\in \mathbb {R}\) such that \(\{a\}^n\preceq A\). Then \(0=\overline{ 0_\infty }^n(\{a\}^n)\le \overline{ 0_\infty }^n(A)\). We conclude that \(\overline{ 0_\infty }^n(A)=0\) for all \(A\in Conv^n({{\mathbb {R}}})\).
If there is \(b\in \mathbb {R}\) such that \(A\preceq \{b\}^n \), then \(\overline{\overline{0_\infty }}(A)\le 0\) and so \(\overline{\overline{0_\infty }}(A)= 0\) by the similar way which applied for \(\overline{ 0_\infty }^n(A)\). Otherwise \(\{b\}^n\preceq A\) for all \(b\in {\mathbb {R}}\). Then \(\{(0,+\infty )\}^{n-1}\preceq A\) and so \(+\infty =\overline{\overline{0_\infty }}(\{(0,+\infty )\}^{n-1})\le \overline{\overline{0_\infty }}(A)\). This yields that \(\overline{\overline{0_\infty }}(A)= + \infty \). We conclude that the elements of \((Conv^n(\overline{{\mathbb {R}}}))^*\) are all non-negative reals, \(\overline{ 0_\infty }^n\) and \( \overline{\overline{0_\infty }}^n\) for all \(n\in {\mathbb {N}}\). Also we have showed that the cones \((Conv(\overline{{\mathbb {R}}}))^*\) and \((Conv^n(\overline{{\mathbb {R}}}))^*\) coincide.
We conclude that \((Conv^n (\overline{{\mathbb {R}}}))^*{\setminus } \{\overline{ 0_\infty }^n,\overline{\overline{0_\infty }}^n\} \) and \(\overline{{\mathbb {R}}}^*\) coincide.
References
Ayaseh, D., Ranjbari, A.: Order bornological locally convex lattice cones. Vladikavkaz Math. J. 19(3), 21–30 (2017)
Ayaseh, D., Ranjbari, A.: Bornological Convergence in Locally Convex Cones. Mediterr. J. Math. 13(4), 1921–1931 (2016)
Dastouri, A., Ranjbari, A.: Some Notes on Barreledness in Locally Convex Cones, Bull. Iran. Math. Soc. https://doi.org/10.1007/s41980-020-00519-x
Keimel, K., Roth, W.: Ordered cones and approximation. Lecture Notes in Mathematics, vol. 1517. Springer-Verlag, Berlin (1992)
Keimel, K., Roth, W.: A Korovkin type approximation theorem for set-valued functions. Proc. Amer. Math. Soc. 104, 819–824 (1988)
Roth, W.: Operator-valued measures and integrals for cone-valued functions. Lecture Notes in Mathematics, vol. 1964. Springer-Verlag, Berlin (2009)
Roth, W.: Korovkin theory for cone-valued functions. Positivity 21(3), 449–472 (2017)
Roth, W.: Hahn-Banach type theorems for locally convex cones. J. Austral. Math. Soc. Ser. A 68(1), 104–125 (2000)
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
Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Dastouri, A., Ranjbari, A. A duality result in locally convex cones. Positivity 26, 73 (2022). https://doi.org/10.1007/s11117-022-00935-9
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11117-022-00935-9