Limit of The Tower

Calculus Level 2

a 0 = a and a n + 1 = a a n \large a_0=a \quad \text{ and } \quad a_{n+1}=a^{a_n}

What is the largest a a such that a n a_n is convergent?


The answer is 1.44466786101.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

6 solutions

Otto Bretscher
Oct 11, 2018

Compare the graphs of y = a x y=a^x (in blue) and y = x y=x (in red). It is a straightforward exercise in calculus to show that for 1 < a < e 1 / e 1<a<e^{1/e} the two graphs intersect at two distinct points, for a = e 1 / e a=e^{1/e} the graph of y = x y=x is tangent to y = a x y=a^x at x = e x=e , and for a > e 1 / e a>e^{1/e} the graphs do not intersect. Now use a cobweb representing the iteration to see that for 1 < a e 1 / e 1<a\leq e^{1/e} the sequence a n a_n converges to the (lower) point of intersection (the sequence is increasing and bounded), while for a > e 1 / e a>e^{1/e} it diverges to infinity as it is increasing without fixed points. See the two attached pictures.

Thus the answer is a = e 1 / e 1.445 a=e^{1/e}\approx \boxed{1.445}

Albert Yiyi
Oct 22, 2018

suppose the limit is L L , then it must satisfy a L = L a = L 1 L a^L=L \implies a=L^{\frac{1}{L}}

from graph, a a has a maximum value, which can be found easily using derivative.

a = L 1 L d a d L = x 1 x 2 ( 1 ln x ) = 0 ( 1 ln x ) = 0 x = e a = e 1 e = 1.44467 \begin{aligned} a &= L^{ \frac{1}{L} } \\ \frac{da}{dL} &= x^{ \frac{1}{x} - 2 } (1 - \ln x) = 0 \\ (1 - \ln x) &= 0 \\ x &= e \\ a &= e^\frac{1}{e} = 1.44467 \end{aligned}

Your work does not prove that the sequence converges for a = e 1 e a= e^{\frac{1}{e}} , though; you merely show that if it converges then a e 1 e a\leq e^{\frac{1}{e}} .

Otto Bretscher - 2 years, 7 months ago

Log in to reply

i guess u're right, i have to show that it converges for a=e^(1/e) .

though rigorous is not what im looking for. i just wanna get the answer the laziest way. (honestly im half expecting the answer to be incorrect)

albert yiyi - 2 years, 7 months ago

Log in to reply

Convergence is not hard to show; it suffices to point out that the sequence a n a_n is increasing and bounded (by e e ). A cobweb illustrates the situation well.

Otto Bretscher - 2 years, 7 months ago

Log in to reply

@Otto Bretscher This suggests a good follow up question - asking for the minimum a a .

Most people get surprised when their logic no longer works, and then I have a discussion of what "Suppose the limit exists" really means, and why that need to "Prove that the limit exists".

Calvin Lin Staff - 2 years, 7 months ago

Log in to reply

@Calvin Lin that is an interesting question.

Limit of The Tower (Strike Back)

oops, i stole it.

sorry.

albert yiyi - 2 years, 7 months ago

@Calvin Lin I'm chagrined to note that people don't like cobwebs much; I find them to be such a fun, intuitive, and powerful tool. Finding the fixed points is certainly a good start, but most people refuse to go further ;)

Otto Bretscher - 2 years, 7 months ago

Log in to reply

@Otto Bretscher I boil it down to
1) An existing misconception (and so this problem helps them start to identify it)
2) Not having the concept explained properly (In particular, many don't remember conditions like f < 1 |f' | < 1 or why it's important)
3) Laziness because so many other examples do not require them to do so.
4) Not understanding the logic of the steps, and merely copying what they think they should do (which we fix by building up their critical thinking and problem solving skills instead of focusing on memorization)



Calvin Lin Staff - 2 years, 7 months ago

Log in to reply

@Calvin Lin In many cases (as in the given problem), we don't even need the derivative; it suffices to point out that the sequence is monotone and bounded.

Otto Bretscher - 2 years, 7 months ago

To be fair, the question's text assumes that the set of L's has a maximum. If the sequence did not converge for a=e^(1/e), the right answer to the question would be "there is no such number".

Mariano Sergi - 2 years, 7 months ago

Log in to reply

It certainly converges for a = 1 a=1 . This might conceivably be the largest value (in fact, most people who have never thought about this problem assume that 1 is indeed the largest value). The fact that the power tower converges for any a > 1 a>1 is somewhat counterintuitive, and certainly requires a proof.

Otto Bretscher - 2 years, 7 months ago

Log in to reply

@Otto Bretscher i think he means a<e^(1/e) has a least upper bound, without having a maximum value.

albert yiyi - 2 years, 7 months ago

Log in to reply

@Albert Yiyi Mariano's conclusion that "If the sequence did not converge for a=e^(1/e), the right answer to the question would be 'there is no such number' " is incorrect. The largest value could (quite plausibly) be 1 or some number strictly between 1 and e 1 e e^{\frac{1}{e}} .

Otto Bretscher - 2 years, 7 months ago

Log in to reply

@Otto Bretscher u're right that conclusion is obviously incorrect. hence i said "i think he mean" otherwise (less obvious incorrect). human might say one thing to mean another thing. ; )

albert yiyi - 2 years, 7 months ago

Log in to reply

@Albert Yiyi The way I see it, it is one of the main objectives of studying maths (perhaps the main objective ) to improve the clarity of thought. We don't need "fake news" and fake methods in maths, even if they produce the right answer sometimes.

Otto Bretscher - 2 years, 7 months ago

Log in to reply

@Otto Bretscher correct indeed. as a math teacher, i try to understand and make sense what student say/ask before they develop the skill to "clarity of thought", so it became a habit of mine : )

albert yiyi - 2 years, 7 months ago

@Otto Bretscher You're right, for some reason I thought that Albert's method somehow proved the convergence for a<e^(1/e) (evidently I didn't really think about it too hard). Thank you for pointing out this was not the case.

Mariano Sergi - 2 years, 7 months ago

Hello, question: How did you get the a^L=L? After some digging, I was able to derive this using the Lambert W "function". Is this what you did? I've seen this "function" a few times and has peaked my curiosity. Any suggestions for further exploration? How widely known/used is this identity? z^z^z^...=-W(-lnz)/lnz

Daniel Glazar - 2 years, 7 months ago

Log in to reply

i did not use the W function. the power tower of a^(a^a...) is still (a^a...), so

a^(a^a...) = L

a^(L) = L

a^L = L

albert yiyi - 2 years, 7 months ago

Log in to reply

Oh, of course. I'm just thinking too hard, haha. Thanks!

Daniel Glazar - 2 years, 7 months ago

We can extrapolate out of the given information that the series will have to respect the equation L = a L L = a^{L} as n goes to infinity. L represents the limit of the series and a simply is the value of a in the problem. The question then become really simple : what is the highest pair of variables L and a such that this equation holds? In order to do so, we do the derivative of a = L 1 / L a = L^{1/L} . We get this equation : ( 1 L n ( L ) ) L ( 2 L 1 L ) (1-Ln(L))*L^{(\frac{2L-1}{L})} We wish to find the value of L when this equation is equal to zero as this will give us the highest point of the function. This gives us L = e L=e . Then we just go back to the first equation and plug e into the it, which gives us that e = a e e=a^{e} , from which we can find that a = e 1 / e a=e^{1/e} which is our answer.

Tom Verhoeff
Oct 28, 2018

The limit, let’s call it x x , is that of an infinite tower of exponentiations of a a . If the limit exists, then we have a x = x a^x = x .

The graphs for y = a x y =a^x are well-known exponential curves, and for y = x y = x it is a straight diagonal line through the origin with slope 1 1 . For a = 1 a=1 these graphs clearly intersect, and for a = 2 a=2 they don’t. For values of a a between 1 and 2 there can be zero, one, or two points of intersection. For the largest value of a a where they still intersect, these graphs have a single point of intersection, and they touch each other (are tangent to each other). Note that the derivative of a x a^x equals a x log a a^x \log a .

Thus we know a x = x a^x = x and a x log a = 1 a^x \log a = 1 . Substituting the first in the second yields x log a = 1 x \log a = 1 , hence a = e 1 / x a = e^{1/x} . Substituting this in the first and simplifying yields e = x e = x . Hence, \fbox{\$a = e^{1/e} \approx 1.4447\$} .

Garv Khurana
Oct 22, 2018

If a(0)=a, then a(1)=a^a,a(2)=a^(a^a), a(n)=a^(a^a...(n times)). Now notice that if a is repeated on forever that the exponent will equal the entire expression. So if a^a^a...=x, then a^x=x, so a=x^(1/x). If x is convergent then the function up to the convergent value will have unique outputted values. This is because a^a^a... will increase until it is divergent. If we find the maximum value of x^(1/x), then that will be the maximum value of x for a to be convergent. We can find the maximum by taking the derivative and setting it equal to 0. So y=x^(1/x), ln (y)= (ln x)/x--> (1/y) dy/dx=(1-ln x)/(ln x)^2-->dy/dx=(x^(1/x)) (1-ln x)/(ln x)^2, by the quotient rule. So (x^(1/x)) ((1-ln x)/(ln x)^2=0--> 1-ln x=0 (x^(1/x) and 1/(ln x)^2 can't equal 0)-->ln x=1-->x=e. So if that is the maximum, then the maximum value for a is e^(1/e) , which is approximately 1.444 .

Vinod Kumar
Oct 23, 2018

Solve x=y^x, for y maximum and find x=e,

[Solve using WolframAlpha, d((1/x)log(x))/dx=0, solution x=e]

Therefore, e=y^e for maximum y

and we get y=e^(1/e)

Answer, y=1.4447

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...