If a tower of exponentials, defined as the limit of x
n as n → ∞, where for some real c > 0 we define
x1 = c
and
xn+1 = cxn,
actually converges, then it can be proved that c satisfies the inequality
e-e ≤ c e1/e,
(where e is the famous mathematical constant e = 2.718281828...) or in decimals,
0.065988... ≤ c ≤ 1.444667...,
and in fact x
n will converge for all such c.
It's rather unusual to have a situation like this where the endpoints of the region of convergence are included in the region of convergence!
It's not hard to show that, if we do have convergence to some number S = S(c):
limn→∞ xn = S
then we must also have
cS = S,
as suggested above. But, this does
not mean that any S satisfying the above equation is necessarily the limit of the x
n.
So in particular, the fact that the 4th power of √2 is 4 does not mean that 4 is the limit of the x
n for the value c = √2. It most certainly is not.
In fact, it is known that for the lower end of the range of c that gives a convergent tower, we have:
for c = e-e, S = 1/e,
and at the high end, we have:
for c = e1/e, S = e.
These facts rule out, for instance, the possibility that for c = √2 we have S = 4.