The division *a*/*b* shows a lookahead of roughly 2*n* (see
figure 7.3). This is above the minimum possible
(*n*), but well below the predicted maximum 4*n*. Other experiments
show this to be fairly typical behaviour unless the numerator and
denominator are equal, in which case a lookahead of *n* will occur.

This experiment avoids the constant factor in the lookahead due to
modification of the denominator mantissa and conversion back from
dyadic and signed binary digits. In general a lookahead of (2*n*+*c*)
for some constant *c* appears probable using the division algorithm.

Note that the analysis given in this section is based upon the rate
that the algorithm consumes digits from the input. It is possible that
jumps in the algorithm steeper than the 4*n* predicted maximum may
occur if part of the algorithm examines more digits without consuming
them (and indeed this can be observed in
figure 7.3). However this will never add more than
a small constant to the lookahead and digits cannot be consumed faster
than 4*n*.