Академический Документы
Профессиональный Документы
Культура Документы
tower,
which interested me when I thought up the specific case of the 2 power tower:
2 ...
2
2
2
where the powers are evaluated not from the bottom up, but from the top down. For clarity, we
can consider the recursive function tHxL = I 2 M
tHx-1L
, tH0L = 2 ; then we are asking what the
value of
2
Lim tHxL. Note that this is quite different from the function pHxL = pHx - 1L with pH0L = 2,
x®¥
since this function clearly tends to infinity as x does.
So, what is the value of this power tower of 2 ' s? To find out, we first assume that the thing
converges, i.e. we can give it a finite value x:
2 ...
2
x= 2 , which implies that
2 ...
2
Log 2
x= 2 , since the power tower is infinite. But then the RHS is x, and we are left
with the trancendental equation
x
2 = x.
Unfortunately, this equation has two solutions we can find just by inspection:
x = 2 and x = 4, since I 2 M = 2 and I 2 M = 4. This presents a problem: which one of these
2 4
solutions represents the actual value of the power tower? We can use Mathematica's nesting function
to find out.
2 Power Tower Revised.nb
Out[54]= 1.9982034775087016239
Mathematica seems to agree with our first result, that x = 2 is in fact the correct solution. But why
then did x = 4 show up in our analysis at all? Previously, I cleverly came up with a method to deter-
nn...
n
mine why this occurs. In general, I showed that the power tower nn has two algebraic solutions
when n 1, and only one solution when n £ 1.
We will define a function f HxL = nx - x, which has its zeroes when the power tower of n' s has a
solution (which is x). Taking the derivative of f HxL, we obtain
This quantity is our critical point. We can see that the point exists when n 1 and doesn't exist
when n £ 1 since the domain of Log@Log@nDD is n Ε H1, ¥L. Thus, f HxL has exactly one critical point
when n 1 and exactly zero critical points when n £ 1. How does that help us? Well, it means that
the slope of the function is either always negative or always positive when when n £ 1 and that the
slope changes sign once when n 1. The first case implies exactly one zero of the function, while
the second implies zero, one or two zeroes.
There could be zero, one or two zeroes for n 1 because the critical point could occur in three
different places: before the zero does, steering the function away from a zero entirely; when the
zero does, which would make the function have only one zero; or after the zero does, which could
give us a second zero. (It might not, as the function could have a horizontal asymptote.) The best
way to see what is going on is graphically:
Power Tower Revised.nb 3
Manipulate@Plot@n^x-x,8x,0,f<,PlotRange®8-r,50<D,8n,0,32<,
8f,10,1000<,8r,-50,1000<D
50
-50
Out[37]=
-100
-150
-200
Though the PDF I will post online isn't interactive, I can report the interesting details this plot
reveals. First, there are in fact two zeroes when n 1. However, there appears to be a cutoff point:
when n gets too big, (around 1.5), the critical point moves above the x-axis and there are no zeroes
at all (as I noted as one of the possibilities). We'll see why that is in a minute. As predicted, there is
only one zero when n £ 1, and it is very close to zero itself. As n gets close to 1, the function
approaches the line y = 1 - x, since nx = 1 for n = 1. When it gets just a bit bigger than 1, however,
the exponential nx will eventually overtake the linear x, it just takes a very long time when the base
isn't much larger than 1.
So, why can't the power tower always have a zero? Let's look at that critcial point again:
LogALog@nDE
x=- Log@nD
When does this thing give us a value for f HxL that's equal to zero, giving the function exactly one
zero (the second case mentioned above)?
0 = f J- N
Log@Log@nDD
LogALog@nDE - LogALog@nDE
Log@nD
=n Log@nD
+ Log@nD
.
Hãk L-
Log@kD
Log@kD
k + k
=0
1 Log@kD
k
+ k
=0
1
Log@kD = -1k = ã-1 n = ã ã , and we have the value of n. It's numerical value is
When
4
does this thing give us a value for f HxL that's equal to zero, giving the function exactly one
Power Tower Revised.nb
zero (the second case mentioned above)?
0 = f J- N
Log@Log@nDD
LogALog@nDE - LogALog@nDE
Log@nD
=n Log@nD
+ Log@nD
.
Hãk L-
Log@kD
Log@kD
k + k
=0
1 Log@kD
k
+ k
=0
1
Log@kD = -1k = ã-1 n = ã ã , and we have the value of n. It's numerical value is
1
Out[56]= 1.4446678610097661337
Which is close to 1.5, the value estimated as the cutoff point earlier.
Below is a graph of the value of the power tower of n' s as n varies on B0, ã ã F:
1
Power Tower Revised.nb 5
powertower@x_,i_D:=Nest@x^ð &,x,iD
ManipulateBPlotBpowertower@x,iD,:x,0,ã ã >,PlotRange®80,E<,
1
PlotStyle®OrangeF,88i,99,"Parity"<,899®"Odd",100®"Even"<<F
2.5
2.0
Out[51]=
1.5
1.0
0.5
Interestingly, the graph seems to bifurcate when n gets close to zero based on the number of itera-
tions we carry out: for an odd number of iterations, the graph curves up, while for an even number,
it curves down. Our algebraic analysis doesn't seem to predict two different solutions; in any event,
the two "solutions" are being caused by something entirely different- the number of values in the
power tower. As a guess, we might look at the graph above and say that the alternating values begin
when n = ã1ã , which has numerical value
1
In[57]:= NB , 20F
ãã
Out[57]= 0.065988035845312537077
That certainly looks right, but how are we to show that it is?
For now, I don't know. We'll leave that question as well as the possibility of extending the power
tower beyond the range B ã1ã , ã ã F, and perhaps into complexland.
1