We present a symbolic technique of error analysis of numerical algorit
hms, which acts as a preprocessor to the actual numerical computation.
The technique provides rigorous, a priori bounds which are universal
for inputs in a prescribed domain. The method is an alternative to int
erval arithmetic in applications where speed and rigor but not necessa
rily the tight bounds are the main concerns. The technique was invente
d to perform the computer-assisted proof of chaos in the Lorenz equati
ons. (C) 1996 Academic Press Limited.