Yesterday a question popped in my mind before going to sleep and I can't solve it (probably because my math knowledge is lacking). Assuming there is a function

$$f(x) = a^x$$

Its inverse function would look like this:

$$f^{-1}(x) = \log_a x$$


For which value of $a$ would the graphs of these two functions just touch at only one common point (of tangency)?

  • 3
    $\begingroup$ I would prefer that you wrote $f(x)=a^x$ and $f^{-1}(y)=\log_a y$. Also notice this forces $a>0, y>0$. $\endgroup$ – higgs Jul 24 '18 at 15:18

Such $a$ likely exists and is somewhere in the following range which can be narrowed down further: $$1.44\lt a \lt 1.45$$

enter image description here

EDIT: $$a\approx1.444667861, \qquad x\approx 2.71827$$

Note that $x$ value is very close to $e$. I don't know how to explain it but it makes this problem really, really interesting!

  • $\begingroup$ Thank you very much for your answer! But is there any possible way to solve this without using a graphing program? $\endgroup$ – Agabagugba Jul 24 '18 at 16:00
  • $\begingroup$ I doubt it, really. $\endgroup$ – Oldboy Jul 24 '18 at 16:20
  • $\begingroup$ Oh okay.. Thank you anyways $\endgroup$ – Agabagugba Jul 24 '18 at 16:33
  • $\begingroup$ I think the key question is: prove or disprove that touching point has $x=e$ . That would be something :) I feel like I'm missing something obvious. $\endgroup$ – Oldboy Jul 24 '18 at 16:36

Basically, the inverse function is obtained by finding the mirror image of $y=a^x$, in $y=x $ .Now note for both the function and its inverse to intersect at just one point, $y=x $ must be the common tangent to both of them. So slope of tangent of $y=a^x $ ,$y=\log_a x$ and $y=x $ must be equal. So,$ 1=a^x \ln a=\ln x{\log_a e } $ now you have 3 equations and 2 unknowns one being 'a' and other being the value of $x $ at which the functions have only one point of intersection . solve finally to get $a^e=e $ which gives the value of $a $ approximately as 1.44.

  • $\begingroup$ Excellent, +1. I love MSE because it's constantly reminding me how dumb I know to be from time to time. $\endgroup$ – Oldboy Jul 24 '18 at 18:21

As @Jasmine point out, because that $a^x$ and $\log_ax$ are inverses we can add $x$ to get $x=a^x=\log_ax$.

We can say more from this: $a^x=\log_a x\implies a^{a^x}=x$ but also $x=a^x$ hence $a^{a^x}=\log_a x\implies a^{a^{a^x}}$ and so on.

Hence we are searching for $a,x$ such that the sequence $x,a^x,a^{a^x},\cdots$ is a constant, $a\in\Bbb R^+\setminus\{1\},x\in \Bbb R^+$.

As it is shown here we have that $e^{-e} \leq a \leq e^{\frac{1}{e}}$ for $a^{a^{.^{.^{.}}}}$ to converge, hence, by taking the sequence and taking the $1/x$ power of it we get $x^{\frac1x}\in[e^{-e},e^{\frac1e}]$, by guessing $e$ we can find that $x=e,a=e^{\frac1e}$ works. But $x=e$ is not the only $x$ that works. Take any $x$ such that $x^{\frac1x}\in[e^{-e},e^{\frac1e}]$ it can be shown that $x,a=x^{\frac1x}$ are tuple that answer the condition of the post.

Edit, I realized that I made a mistake.

[$(a^{a^{...^x}})^{1/x}=(a^{a^{...^x}/x})\ne (a^{a^{...}})$, although it is interesting to see that if you continue with the logic of the mistake and find the $x$s where the 2 function "kiss" you get correct answer]

But given $x=a^x$ we have $x^{1/x}=a$(In analogy to the mistake, the maximum value of $x^{1/x}$ is $e^{1/e}$ at $x=e$).

As I said in the post $a=x^{\frac1x}$ gives us $a^y=\log_ay=y$, at $y=x$, but we also want them to be tangent so $(a^y)'\mid_{y=x}=\ln(a)a^x=\frac1{\ln(a)x}=(\log_a(y))'\mid_{y=x}$, we know that $a^x=x$ so $\ln(a)a^x=\ln(a)x$, let set $\ln(a)x=z$ and we have $z=\frac1z\implies z=\pm1$, and so $\ln(a)x=\pm1\implies\ln(x^{1/x})x=\pm1\implies\ln(x)=\pm1\implies x=\begin{cases}e\\1/e\end{cases}\\\implies a=\begin{cases}e^{1/e}\\(1/e)^{e}\end{cases}$

(I think that OEIS made a mistake about uniqueness)

See here the graph of the four functions


According to a OEIS comment, the only base $a$ such that $\exists \xi \in A \subset\mathbb R: f(\xi)=f^{-1}(\xi)$ is $a=e^{1/e}$ with $\xi=e$. Numerically this seems to be true, as Oldboy pointed out.

Whether that's unique (for any $a$) needs to be proven.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.