New slope methods for sharper interval functions and a note on Fisher's acceleration method (Q1921300)

Slopes and gradients can be used to compute sharper bounds for interval functions \(F(X):\mathbb{R}^n\to\mathbb{R}\), and the evaluation of those bounds can considerably accelerate interval-based algorithms which occur in applications, e.g., in global optimization. In the present paper, for the inclusion of interval functions, firstly two possibilities are revisited that are based on the partial derivatives of the function (``gradient methods''): one method is obtained from the mean value theorem, the other from an idea of E. Hansen. Secondly, four slope inclusion methods are developed. Here, the fourth method is of special interest since it provides greater sharpness not only for the last term of a factorization of \(f(X)\), but also for intermediate steps. Thirdly, in order to obtain a faster evaluation of both gradients and slopes in forward mode, an acceleration method by \textit{H. Fischer} [Computing 41, No. 3, 261-265 (1989; Zbl 0666.65017)] is adapted. The paper finishes with a thorough description of the implementation of the above methods and a discussion of results with respect to several examples.