Mean ergodic theorems for almost periodic semigroups (Q614495)

The paper is devoted to mean ergodic theorems for almost periodic functions \(f:S\to E\), where \(S\) is a semigroup with identity and \(E\) is a locally convex topological vector space. The function \(f\) is called almost periodic (or weakly almost periodic) if its right orbit (the set of right shifts) \({\mathcal {RO}}(f) = \{f(\cdot s)\mid s \in S\}\) is uniformly bounded and relatively compact with respect to the topology of uniform convergence (or the corresponding weak topology). In essence, the ergodic theorem asserts that there exists a unique constant function \(c_f: S\to E\) in the closure of the convex hull of \({\mathcal RO}(f)\), where \(f:S\to E\) is an arbitrary weakly almost periodic function. The authors extend this result from the previously known case of a Banach space \(E\) to the case of a locally convex space \(E\). They consider almost periodic semigroups of continuous mappings as well. Let \(C\) be a closed bounded convex subset in \(E\) and \({\mathcal S} =\{T(s)\mid s\in S\}\) be a representation of the semigroup \(S\) by continuous mappings \(T(s):C\to C\). Consider a mapping \(\phi_{\mathcal S}:C\to l^\infty(S,E)\) defined by the formula \((\phi_{\mathcal S}(x))(s) =T(s)x\). The representation \({\mathcal S}\) of \(S\) is said to be (weakly) almost periodic if for each \(x\in C\) the function \(\phi_{\mathcal S}(x)\) is (weakly) almost periodic. The authors prove that the representation \({\mathcal S}\) is almost periodic if and only if for each \(x\in C\) its orbit \({\mathcal O}(x)\) is relatively compact and \({\mathcal S}\) is equicontinuous on the closure of \({\mathcal O}(x)\). This result implies an ergodic theorem for the orbits of points \(x\in C\) under the assumption that \(C\) is convex and compact and the representation \({\mathcal S}\) is equicontinuous on \(C\). Some results of the paper (namely, Theorems 2, 3, 4) seem doubtful because they are based on the crucial Lemma 1, which is erroneous. There are used vector-valued weak integrals \(\mu(f) =\int_S f(s)\,d\mu(s)\) defined in such a way that \[ \langle x',\mu(f)\rangle = \mu\bigl(\langle x', f(\cdot)\rangle\bigr) \quad \text{ for each } x'\in E', \] where \(\mu\) is a positive normalized linear functional \(\mu:l^\infty(S)\to {\mathbb R}\) or, in other words, a finitely additive probability measure on \(S\). It is stated in the paper that if the image \(f(S)\subset E\) is relatively compact then the weak integral \(\mu(f)\) belongs to \(E\). In general, this is not true unless the convex hull of \(f(S)\) is relatively compact, or the space \(E\) is Banach, or certain additional completeness conditions hold.