Harmonic update is a randomized on-line algorithm which, given a rando
m rn-set of vertices U(m) subset of or equal to {-1, 1}(n) in the n-di
mensional cube, generates a random vertex w is an element of {-1, 1}(n
) as a putative solution to the system of linear inequalities: Sigma(i
=1)(n) w(i)u(i) greater than or equal to 0 for each u is an element of
U(m). Using tools from large deviation multivariate normal approximat
ion and Poisson approximation, we show that root n/root log n is a thr
eshold function for the property that the vertex w generated by harmon
ic update has positive inner product with each vertex in U(m). More ex
plicitly, let P(n, rn) denote the probability that Sigma(i=1)(n) w(i)u
(i) greater than or equal to 0 for each u is an element of U(m). Then,
as n --> infinity, P(n, m) --> 0 or 1 according to whether m = m(n) v
aries with n such that >> root n/root log n or m << root n/root log n,
respectively. The analysis also exposes the fine structure of the thr
eshold function.