1

What is $\int{x^{x^{x^{x^{\dots }}}}dx}$? I managed to get it down to a complicated series after using $u$-substitution, but I want to see other methods of approach, or if my answer was right.

First I let $$u=x^{x^{x^{x^{\dots }}}}$$, and $du$ would be, in terms of $u$, $$du= \frac{u}{\frac{1-ln(u)}{u}e^{ln(u)/u}}$$ This makes the integral of $$\int{x^{x^{x^{x^{\dots }}}}dx}=\int \frac{1 - ln(u)}{u}{e^{ln(u)/u} }du$$ Now making $w = ln(u), dw$ becomes $1/u$ and the integral becomes $$\int e^{w/e^{w}}(1-w)dw$$ From there, I distributed and made $2$ integrals of series, using the Taylor polynomial for $e^x$. From there, I integrated and back substituted. I was only trying to find the indefinite integral.

I'm still a junior in high school, so let me know if the answer is beyond my scope of understanding.

Mathphile
  • 2,437
  • 4
    what is that complicated series – James Apr 29 '18 at 03:33
  • Thank you for accepting my answer. As you are still in high school, is there anything in my answer that you would like me to elaborate on? – Szeto Apr 29 '18 at 14:22
  • @Szeto Thank you for your answer! I think I will try to learn where your answer comes from on my own. However, I also want to know if my solution used faulty logic to reach a conclusion. I edited my question to include my reasoning. – Suchetan Dontha Apr 29 '18 at 14:26
  • @SuchetanDontha I am not sure if the line $du=...$ is correct. – Szeto Apr 29 '18 at 14:58
  • @Szeto I just found the derivative and replaced the x terms in terms of u. I verified my answer with this video: https://www.youtube.com/watch?v=i_l1lz26C2M&t=1s – Suchetan Dontha Apr 29 '18 at 15:05
  • @SuchetanDontha I did not see any mistake in your work. I will double check my work. – Szeto Apr 29 '18 at 22:13

2 Answers2

6

The function satisfies $$y=e^{\ln(x)\cdot y}$$ which gives $$y=-\frac{W(-\ln(x))}{\ln(x)}$$

For the integral, $$I(x)=\int ydx=\int \frac{W(-\ln(x))}{-\ln(x)}dx$$

Let $u=-\ln x$, $dx=-e^{-u}du$, then the integral equals $$\int \frac{W(u)}{ue^u}du$$

Since $W(u)=\sum_{n=1}^{\infty}a_nu^n$ with $a_n=\frac{(-n)^{n-1}}{n!}$, $$I(x)=\sum^{\infty}_{n=0}a_n\int u^{n-1}e^{-u}du= \sum^{\infty}_{n=0}a_n\gamma(n,u)= \sum^{\infty}_{n=0}a_n\gamma(n,-\ln(x)) $$

where $\gamma(n,x)$ is the lower incomplete gamma function.

$$\int x^{x^{...}} dx= \sum^{\infty}_{n=0}\frac{(-n)^{n-1}}{n!}\gamma(n,-\ln(x))$$

Szeto
  • 11,159
  • 1
    Note that $W(u)=\sum_{n=1}^{\infty}\frac{(-n)^{n-1}}{n!}u^n$ only converges for $|u|\lt \frac{1}{e}$ so your answer only converges for $e^{-1/e}\lt x \lt e^{1/e}$ – Mathphile Apr 03 '20 at 22:51
2

Although Szeto's answer: $$\int x^{x^{...}} dx= \sum^{\infty}_{n=0}\frac{(-n)^{n-1}}{n!}\gamma(n,-\ln(x))$$ is correct. It does not converge for the complete domain of $x^{x^{x...}}$ which is $e^{-e}\le x \le e^{1/e}$ and instead only converges for $e^{-1/e}\le x\le e^{1/e}$. This is because the sum $W(u)=\sum_{n=1}^{\infty}\frac{(-n)^{n-1}}{n!}u^n$ has a radius of convergence of $|u|\lt \frac{1}{e}$. Below I will represent the integral as an infinite sum that converges for all $e^{-e}\le x \le e^{1/e}$.

$$ \int x^{x^{x^{.^{.......}}}} dx= \int\frac {-W(-\ln(x))}{\ln(x)} dx$$ let $ -\ln x=t $

$$ x=e^{-t} $$

$$dx =-xdt $$

$$ \therefore \int\frac {-W(-\ln(x))}{\ln(x)} dx =- \int \frac {W(t)}{te^t}dt $$ We know that: $$\frac{1}{e^t}=\sum_{n=0}^{\infty}\frac{(-t)^n}{n!}$$

$$- \int \frac {W(t)}{te^t}dt=- \int \sum_{n=0}^{\infty}\frac {(-1)^nW(t)(t)^n}{tn!}dt=- \int \sum_{n=0}^{\infty} \frac{(-1)^nW(t)(t)^{n-1}}{n!}dt $$

$$- \int \sum_{n=0}^{\infty} \frac{(-1)^nW(t)(t)^{n-1}}{n!}dt = - \sum_{n=0}^{\infty} \frac{(-1)^n}{n!}\int W(t)(t)^{n-1}dt $$ Now, $$\int W(t)(t)^{n-1}dt= \sum_{n=0}^{\infty}\frac {(t)^ne^{[-nW(t)]}[-nW(t)]^{-n}[n\Gamma(n+1, -nW(t))- \Gamma(n+2, -nW(t))]} {n^2} $$

The proof for this is long and can be confirmed by Wolfram Alpha.

Substituting $t=- \ln x$:

$\sum_{n=0}^{\infty}-\frac {(-1)^n(-\ln x)^ne^{[-nW(-\ln x)]}[-nW(-\ln x)]^{-n}[n\Gamma(n+1, -nW(-\ln x))- \Gamma(n+2, -nW(-\ln x))]} {(n!)n^2} $

And finally we have:

$ \int x^{x^{x^{.^{.......}}}} dx = \sum_{n=0}^{\infty}-\frac {(-1)^n(-\ln x)^ne^{[-nW(-\ln x)]}[-nW(-\ln x)]^{-n}[n\Gamma(n+1, -nW(-\ln x))- \Gamma(n+2, -nW(-\ln x))]} {(n!)n^2} $

Mathphile
  • 2,437
  • Wow I was surprised to see a post on this question. I don't have much to really add to your answer, but I will say that I posted this a while ago https://math.stackexchange.com/questions/3489620/expression-for-the-value-of-int-01-x1-xdx. If you read the last post, I actually use Szeto's answer to solve this post's integral from 0 to 1, and it seems to work out numerically, and has a form satisfyingly close to the sophomore's dream integrals should my result be true. Although the function might not converge on its domain, I think something meaningful can be pulled from its simplicity. – Suchetan Dontha Apr 04 '20 at 00:41
  • If n=0, then the denominator is also 0, right? – Тyma Gaidash Aug 03 '21 at 20:26