Keith Briggs:
Reliable real arithmetic and one-dimensional dynamical systems
Abstract:
Significant advances continue to be made in the field of software for
accurately emulating real-number arithmetic. These have potential
applications for experimental studies of dynamical systems, and for
computer-assisted theorem proving. The main issue here is how to get
reliability without an intolerable performance penalty.
My talk has two
parts: first a quick survey of the state-of-the-art, with a
description of
my own "lazy stochastic" approach to computing with reals (where the
aim is
strict preservation of the order relation while minimizing the amount
of
computation); and second an application to the study of the
statistical
distribution of blocks of partial quotients in continued fractions.
The
latter concerns a simple dynamical system on the real line, for which
there
are still simply-stated questions to which the answer is unknown
(such as whether cubic irrationals are normal). I suggest that exact
computations (i.e., where every
partial quotient is guaranteed correct) may provide some illumination
on
such problems.