T = 2 pi r / (beta c)
r = beta c T/(2pi)

B \propto p/r
	= m beta gamma c / (beta c T/(2pi))
	= 2 pi m gamma / T
	\propto gamma

Focussing depends on "integrated momentum-normalised gradient"; many sectors of size Theta

B'l/p = B' r Theta / p

If this is constant then

B' \propto p/r \propto gamma \propto B

So B(s) \propto e^ks, doesn't work at low energies: B' is almost zero at centre of cyclotron.

Note also B \propto gamma.  Set r = beta R (where R = c T/(2pi)).  Let gamma(s) = e^ks.
r(s) = sqrt(1-gamma^-2) R
 = sqrt(1-e^(-2ks)) R
dr/ds = R 0.5 (1-e^(-2ks))^-0.5 2ke^(-2ks)
 = R 0.5 (1/beta) 2k gamma^-2
 = kR / (beta gamma^2)
But |dr/ds| <= 1 always, so 
beta gamma^2 >= kR = R/"e-fold height"

For low energies in FFAG-focussing mode, need small radius and large e-fold height.

Could use cyclotron (weak) focussing mode, but tunes would then stay small throughout machine and would have significant coupling in VFFAG section