The Initial Value Problem (IVP) is concerned with finding solutions to a system of autonomous ordinary differential equations (ODE) \begin{equation} \textbf{x}' = \textbf{f}(\textbf{x}) \end{equation} with given initial condition $\textbf{x}(0)\in B_0$ for some box $B_0\subseteq \mathbb{R}^n$. Here $\textbf{f}:\mathbb{R}^n\to\mathbb{R}^n$ and $\textbf{x}:[0,1]\to\mathbb{R}^n$ where $\textbf{f}$ and $\textbf{x}$ are $C^1$-continuous. Let $\texttt{IVP}_\textbf{f}(B_0)$ denote the set of all such solutions $\textbf{x}$. Despite over 40 years of development to design a validated algorithm for the IVP problem, no complete algorithm currently exists. In this paper, we introduce a novel way to exploit the theory of $\textbf{logarithmic norms}$: we introduce the concept of a $\textbf{radical transform}$ $\pi:\mathbb{R}^n\to\mathbb{R}^n$ to convert the above $(\textbf{x},\textbf{f})$-system into another system $\textbf{y}' = \textbf{g}(\textbf{y})$ so that the $(\textbf{y},\textbf{g})$-space has negative logarithmic norm in any desired small enough neighborhood. Based on such radical transform steps, we construct a complete validated algorithm for the following $\textbf{End-Enclosure Problem}$: \begin{equation} INPUT: (\textbf{f}, B_0,\varepsilon), \qquad\qquad OUTPUT: (\underline{B}_0,B_1) \end{equation} where $B_0\subseteq \mathbb{R}^n$ is a box, $\varepsilon>0$, such that $\underline{B}_0\subseteq B_0$, the diameter of $B_1$ is at most $\varepsilon$, and $B_1$ is an end-enclosure for $\texttt{IVP}(\underline{B}_0)$, i.e., for all $\textbf{x}\in \texttt{IVP}(\underline{B}_0)$, $\textbf{x}(1)\in B_1$. A preliminary implementation of our algorithm shows promise.
翻译:暂无翻译