Iterated linear maps on a cone and Denjoy-Wolff theorems (Q2496633)

The paper presents the following results: Theorem 1: Let \(C\) be a closed cone with nonempty interior \(\text{int}(C)\) in a finite dimensional Banach space \((X, \| \cdot\| )\). Let \(q \in C^*\) be such that \(q(x) > 0\) for all \(x \in C \backslash \{ 0 \}\). Define \(\Sigma_q := \{ x \in C \mid q(x) = 1\}\) and \(\text{ri}(\Sigma_q) := \Sigma_q \cap \text{int}(C)\). Assume that \(T : \text{ri}(\Sigma_q) \to \text{ri}(\Sigma_q)\) is nonexpansive with respect to Hilbert's projective metric \(D\) and that \(T\) has no fixed points in \(\text{ri}(\Sigma_q)\). Then the following hold: (i) For every \(x \in \text{rin}(\Sigma_q)\), we have \(\omega(x;T) \subseteq \partial C\). (ii) If \(x, y \in \text{ri}(\Sigma_q)\), then every element \(\zeta \in \omega(x;T)\) is comparable to an element \(\zeta \in \omega(y;T)\). (iii) If \(x \in \text{ri}(\Sigma_q)\), every element \(ksi \in \text{co}(\cup_{z \in \text{ri}(\Sigma_q)} \omega(z;T))\) is comparable to an element \(\zeta \in \text{co}(\omega(x;T))\). (iv) If there exists \(x \in\text{ri}(\Sigma_q)\) and \(\zeta \in \omega(x;T)\) such that \(T\) has a continuous extension \(\overline{T} : \text{ri}(\Sigma_q) \cup \{ \zeta \} \to \Sigma_q\), then \(\zeta\) and \(\overline{T}(\zeta)\) are comparable and \(d(\zeta,\overline{T}(\zeta)) \leq d(x,Tx)\). Theorem 2: Let \(C\) be any closed polyhedral cone with nonempty interior in a finite dimensional Banach space \(X\). Suppose that \(f : C \to C\) is a linear map such that \(f(\text{int}(C)) \subseteq \text{int}(C)\). Let \(q \in C^*\) be such that \(q(x) > 0\) for all \(x \in C \backslash \{ 0 \}\), and define \(\Sigma_q\) and \(\text{ri}(\Sigma_q)\) as before. Let \(T(x) = f(x) / q(f(x))\). If \(T\) has no fixed points in \(\text{ri}(\Sigma_q)\), then for any \(x \in \text{ri}(\Sigma_q)\), the set \(\omega(x;T)\) is a finite subset of \(C_u \cap \Sigma_q\) for some \(u \in \partial C\), and \(C_u\) does not depend on \(x\). Theorem 3: Let \(C\) be a closed cone with nonempty interior in a finite dimensional Banach space \(X\). Let \(f : X \to X\) be a linear map such that \(f(\text{int}(C)) \subseteq \text{int}(C)\). Let \(q \in C^*\) be such that \(q(x) > 0\) for all \(x \in C \backslash \{ 0 \}\) and define \(\Sigma_q = \{ x \in C \mid q(x) = 1 \}\) and \(\text{ri}(\Sigma_q) = \Sigma_q \cap \text{int}(C)\). If the function \(T\) defined on \(\Sigma_q\) by \(T(x) = f(x)/q(f(x))\) has no fixed point in \(\text{ri}(\Sigma_q)\), then for any \(x \in\text{ri}(\Sigma_q)\), we have \(\text{co}(\omega(x;T)) \subseteq \Sigma_q \backslash \text{ri}(\Sigma_q)\).