On approximate and algebraic computability over the real numbers

Abstract

We consider algebraic and approximate computations of (partial) real functions $\text{ƒ:R}{}^{\mathrm{d}}\text{↣ R}$. Algebraic computability is defined by means of (parameter-free) finite algorithmic procedures. The notion of approximate computability is a straightforward generalization of the Ko-Friedman approach, based on oracle Turing machines, to functions with not necessarily recursively open domains.

The main results of the paper give characterizations of approximate computability by means of the passing sets of finite algorithmic procedures, i.e., characterizations from the algebraic point of view. Some consequences and also modifications of the concepts are discussed. Finally, two variants of arithmetical hierarchies over the reals are considered and used to classify and mutually compare the domains, graphs and ranges of algebraically resp. approximately computable real functions.