A general reliable quadratic form: An extension of affine arithmetic (Q2494411)

The paper indicates a new way to generate bounds for the range of polynomial functions \(f\) over a hypercube \(X\subseteq\mathbb{R}^n\). To this end general quadratic forms (GQF) are used which are defined by \[ \widehat{\widehat x}= \varepsilon^T A\varepsilon+ b^T\varepsilon+c + e^+\varepsilon_{n+1}+ e^-\varepsilon_{n+2}+ e\varepsilon_{n+ 3}, \] where \(A\in\mathbb{R}^{n\times n}\), \(b\in\mathbb{R}^n\), \(c\in\mathbb{R}\), \((e^+, e^-, e)\in(\mathbb{R}^+)^3\), \(\varepsilon_i\in [-1,1]\) for \(i\in \{1,\dots, n\}\), \(\varepsilon_{n+1}\in [0,1]\), \(\varepsilon_{n+2}\in [-1,0]\), and \(\varepsilon_{n+3}\in [-1,1]\). Transfers are given from intervals to GQF and vice versa, and also from GQF to affine forms and vice versa. Here, an affine form is defined by \[ \widehat x= x_0+ \sum^n_{i=1} x_i\varepsilon_i \] with \(x_i\in\mathbb{R}\) and \(\varepsilon_i\in [-1, 1]\). The operations \(+\), \(-\), \(\times\) are defined for GQF, and with them an inclusion function \(\text{GQF}(X)\) for \(f(X)\). Properties of this function are proved including comparisons with other inclusion functions. Interval input data \(A\), \(b\), \(c\) were introduced in GQF in order to guarantee inclusion for input data which are non-machine numbers. This ends up with a so-called reliable general quadratic form. The operations \(+\), \(-\), \(\times\) are generalized to this new situation and an inclusion function is defined analogously to \(\text{GQF}(X)\). It is shown how GQF can be applied to global optimization for polynomials using a basic interval branch-and-bound algorithm of Moore and Skelboe [see \textit{H. Ratschek} and \textit{J. Rokne}, New computer methods for global optimization. (1988; Zbl 0648.65049)].