Interval arithmetic over finitely many endpoints (Q695060)

A finite set \(\mathbb{B}= \{b_1,\dots, b_k\}\) is called a weakly admissible set of interval bounds if its elements are real nonempty intervals satisfying \(\alpha\in b_i\), \(\beta\in b_{i+1}\Rightarrow\alpha< \beta\) for \(1\leq i< k\). If, in addition, \(k> 1\) and \(\inf b_1=-\infty\), \(\sup b_k=+\infty\) then \(\mathbb{B}\) is called admissible set of interval bounds. On \(\mathbb{B}\) a total ordering \(\preceq\) is defined by \(b_i\preceq b_j\) for \(1\leq i\leq j\leq k\). The set \(\mathbb{I}\mathbb{B}\) of proper intervals over a weakly admissible set of interval bounds \(\mathbb{B}\) is the empty set and the set of all pairs \((a, b)\in\mathbb{B}\times\mathbb{B}\) with \(a\preceq b\). According to this definition the elements of \(\mathbb{B}\) serve as (finitely many) endpoints of the intervals in \(\mathbb{I}\mathbb{B}\). Interval operations are defined on \(\mathbb{I}\mathbb{B}\) and extended to \(\overline{\mathbb{I}\mathbb{B}}= \mathbb{I}\mathbb{B}\cup \{\text{NaI}\}\) where NaI denotes `not an interval'. In addition, the natural interval extension of a function \(f: D_f\subseteq\mathbb{R}^n\to\mathbb{R}\) is introduced. Mathematical properties are proved. In a practical implementation the set \(\mathbb{B}\) of interval endpoints consists of the set of degenerate ordinary intervals \([f,f]\) for all floating point numbers \(f\) of a particular format plus some extra sets. This is realized in the paper based on the IEEE standard 754. A modification is considered which ignores inputs out of range.