Given that
- $x^x = y$; and
- given some value for $y$
is there a way to expressly solve that equation for $x$?
Asked
Active
Viewed 4,059 times
25
-
5Take logs, set $x = e^t$ and apply this: http://math.stackexchange.com/questions/10261/inverse-of-y-xex – Aryabhata Jul 28 '11 at 07:03
-
For $e^{-1/e} \lt y \le 1$ there will be two non-negative solutions for $x$ and for $y \lt e^{-1/e}$ there will be none – Henry Sep 05 '16 at 10:18
-
It is also known as the super-square root of $y$. – Simply Beautiful Art Dec 31 '16 at 01:00
-
1Assume $y$ is positive we have $x^x=y$ so $x\ln x=\ln y$ so $\ln x e^{\ln x}=\ln y$ so $\ln x=W(\ln y)$ so $x=e^{W(\ln y)}=\frac{\ln y}{W(\ln y)}$. – Ahmed S. Attaalla Jan 09 '17 at 05:56
3 Answers
29
As Aryabhata mentions this is another application for the Lambert W function. The solution to your problem is presented in the wikipedia article. Using elementary substitutions you have
$$x=\frac{\ln(y)}{W(\ln y)}$$
If you are interested in the asymptotic growth of $x$ relative to $y$, note that for every $z$: $W(z) = \ln{z} - \ln\ln{z} + o(1)$. Hence:
$$x=\frac{\ln(y)}{\ln{\ln y} - \ln\ln{\ln y} + o(1)} = \Theta\left( \frac{\ln y}{\ln \ln y}\right)$$
Erel Segal-Halevi
- 10,555
Listing
- 13,937
-
-
@Josh: in fact this might as well be solved by defining a new ad-hoc function $Z(x)$ such that $Z(x)^{Z(x)}=x$, so that the solution of the equation is $x=Z(y)$; Lambert is defined by $W(x)e^{W(x)}=x$, there is little magic. The main advantage of reducing to the special Lambert function is that it has already been studied. – Jul 09 '18 at 14:04
2
You should try WolframAlpha for similar problems. WolframAlpha would solve y=x^x for y=5 as shown here (using Lamber W Function as suggested before).
gsbabil
- 196
-
4If you're just going to post a link to wolframalpha, you could at least make sure that it works... – t.b. Dec 31 '11 at 16:20
-
2If a natural number y is entered in "solve x^x=y" WA will give the numeric answer. Strangely it doesn't work for all real y. However, I downvoted since the link is no help in understanding how the solution was reached. – duckstar Dec 31 '11 at 17:22
-
3I guess part of GSBabil's point is that Josh should have tried that first, before posting here. – GEdgar Dec 31 '11 at 18:33
0
Given y, you can solve y = x**x with a simple iteration:
- Set x_0 = 2 (or anything else really)
- x_(i+1) = ln y / ln x_i
- Repeat 2 until abs(x_(i+1) - x_i) < threshold
It's akin to Newton's method of finding square roots.
The convergence of this process is the "natural base of y to itself", or the "exponential root" of y.
If you think about this iteration, you're iteratively performing the base-change logarithm formula. So at first you get y's logarithm (or "bit length") in base 2. Then in base of what y was in base 2, and so on. Successive values oscillate around the "natural base of y to itself" until it converges.
It's pretty interesting. I wonder if there's anything special about the value of x for each y.
As other people have said, you can take log of each side, to get:
ln y = ln x**x, which is ln y = x ln x, and the iteration
Cris
- 1