In the LFT approach to Exact Real Arithmetic, we study the question how many argument digits are needed to produce a certain number of result digits. We present upper and lower bounds for many simple functions and operations, and for exponential and square root. 1 Introduction In this paper, we work in an approach to Exact Real Arithmetic where real numbers are represented as potentially infinite streams of information units, called digits. Hence, an algorithm to compute a certain expression over real numbers is a device that reads some input streams and produces an output stream. Algorithms like this never terminate, but are considered as satisfactory if they produce any desired number of output digits in finite time, i.e., from a finite number of input digits by a finite number of internal operations. The (time) efficiency of a real number algorithm indicates how much time T (n) it takes to produce n result digits. It clearly depends on the number of input digits needed to produce ...

One possible approach to exact real arithmetic is to use linear fractional transformations to represent real numbers and computations on real numbers. We show how to determine the digits that can be emitted from a transformation, and present a criterion which ensures that it is possible to emit a digit. Using these results, we prove that the obvious algorithm to compute n digits from the application of a transformation to a real number has complexity O(n 2 ), and present a method to reduce this complexity to that of multiplying two n bit integers. 1 Introduction Linear Fractional Transformations (LFT's) provide an elegant approach to real number arithmetic [5,14,9,12,10,4]. One-dimensional LFT's x 7! ax+c bx+d are used as digits and to implement basic unary functions, while two-dimensional LFT's (x; y) 7! axy+cx+ey+g bxy+dx+fy+h provide binary operations such as addition and multiplication, and can be combined to obtain infinite expression trees denoting transcendental functions...

Abstract not found

We examine a special case of admissible representations of the closed interval, namely those which arise via sequences of a nite number of Mobius transformations. We regard certain sets of Mobius transformations as a generalized notion of digits and introduce sucient conditions that such a \digit set" yields an admissible representation of [0; +1]. Furthermore we establish the productivity and correctness of the homographic algorithm for such \admissible" digit sets. In the second part of the paper we discuss representation of positive real numbers based on the Stern{Brocot tree. We show how we can modify the usual Stern{Brocot representation to yield a ternary admissible digit set.