Sat Mar 9 10:45:18 EST 2013

Exponentials: reset at border

Compute q_k = exp((P_k_+1 - P_k) / n) such that a complex exponential
can be generated as p_k+1 = p_k * q_k^n

Assuming p_k = exp(P_k) and if the approximation of exp() is good,
p_k+1 will be pretty close to exp(P_k+1).  Maybe it is good enough to
just set p_k+1 to this value?

This gives a simple algorithm where noise (click) is only dependent on
the accuracy of exp(), assuming that the numerical problem due to
successive multiplication by q is small enough.

Simple enough to test.