Interval arithmetic, affine arithmetic, Taylor series methods: Why, what next? (Q2387779)

The authors consider \(N\)-dimensional families of functions of the type \[ \{C_1F_1(x_1,\dots,x_n)+\dots+C_NF_N(x_1,\dots,x_n) \mid C_1,\dots,C_N\in \mathbb{R}\}, \] where \(F_1,\dots,F_N\) are analytic functions. Assuming that a transitive relation \(\succeq\) on the set of all \(N\)-dimensional families is given, they call the relation final if there exists exactly one \(N\)-dimensional family \(\Phi\) satisfying \(\Phi\succeq\Psi\) for each \(\Psi\). The main result given in the paper says that if \(\succeq\) is a final rescaling-invariant relation on the set of all families, then each function from the optimal family \(\Phi\) is a polynomial. Consequences of this result for interval computations are discussed.