Multifractals

Asymptotes of the tau(q) curve

We consider the functions hi(q) = piqritau(q) for i = 1, ..., N. These are natural functions because the equation defining tau(q)
p1qr1tau(q) + ... + pNqrNtau(q) = 1
can be written
h1(q) + ... + hN(q) = 1.
Differentiating,
dhi/dq = piqritau(q) (ln(pi) + ln(ri)(dtau/dq))
Suppose ln(pm)/ln(rm) = alphamin and ln(pM)/ln(rM) = alphamax
Recalling alpha = -dtau/dq, the bounds alphamin <= alpha <= alphamax become
-ln(pm)/ln(rm) >= dtau/dq >= -ln(pM)/ln(rM)
Because ln(ri) < 0 for each i, we have
-ln(ri)(ln(pm)/ln(rm)) <= ln(ri)(dtau/dq) <= -ln(ri)(ln(pM)/ln(rM))
Adding ln(pi) to each side
ln(pi) - ln(ri)(ln(pm)/ln(rm)) <= ln(pi) + ln(ri)(dtau/dq) <= ln(pi) - ln(ri)(ln(pM)/ln(rM))
Taking i = m gives dhm/dq >= 0; taking i = M gives dhM/dq <= 0.
For all i, 0 <= hi(q) <= 1 for all q.
Because hm(q) is a nondecreasing function bounded between 0 and 1, both limits
limq -> infinityhm(q) and limq -> -infinityhm(q)
exist. These limits exist also for hM(q).
Moreover, limq -> -infinityhM(q) = cM and limq -> infinityhm(q) = cm, with both cM and cm positive.
In fact, each dhi/dq has at most one zero, and so the limits limq -> infinityhi(q) and limq -> -infinityhi(q) exist
If limq -> infinityhi(q) = ci > 0, then ln(pi)/ln(ri) = ln(pm)/ln(rm).
Here is a picture of some typical hi.
From this we deduce limq -> infinitydtau/dq = -alphamin and similarly limq -> -infinitydtau/dq = -alphamax
Because dtau/dq appraoches -alphamax as q -> -infinity and -alphaminas q -> infinity, we deduce the graph of tau(q) has oblique asymptotes of these slopes.

Return to Multifractals from IFS.