With an understanding of decision tree complexity we can now prove a
lower bound on a class of evasive functions. In analogy to evasive
functions whose deterministic decision tree complexity is
we call
functions with nondeterministic decision tree complexity
Nondeterministically evasive. Every nondeterministically evasive
functions is evasive.
This proof establishes that nondeterministically evasive functions
have inputs that are sensitive to negation on at least half of their
bits; the result then follows from Lemma
. Not all
evasive functions are nondeterministically evasive. OR is an example
of a nondeterministically evasive function; since Beals et al. provide an
algorithm to compute the OR of
bits, this
lower bound is asymptotically tight.