8

A few years ago, I got interested in an apparently hard integration problem which had me fascinated. This was the integral of a Sophomore Dream like integral except with the bounds over the real positive numbers denoted by $\Bbb R^+$ and not from 0 to 1 over the unit square, or unit line in this case? Nevertheless, this quirky problem caught my attention for having its area be almost 2 with this graph. Here is my solution, although there are a couple of calculations that I could have done differently: enter image description here $$ A\mathit =\mathrm{\int_{\Bbb R^+}x^{-x}dx=\int_0^{\infty}\sum_{n=0}^{\infty}\frac{(-1)^nx^n ln^n(x)}{n!}dx}$$

I naively used Wolfram Alpha to integrate, but this can easily be still done with the substitution of $x=e^{-y}$ which is similar to the other Sophomore Dream integrals, so it will be ignored here for conciseness and should be able to use an interchanging theorem. Here I chose not to do the substitution as the same result appears anyways and this is simpler:

$$\mathrm{\sum_{n=0}^{\infty} \frac{(-1)^n}{n!}\int_0^{\infty}x^n ln^n(x)dx= \sum_{n=0}^{\infty} \frac{(-1)^n}{n!}(-x^{-n}x^nln^n(x)ln(x)ln^{-(n+1)}(x)(-1)^{-(n+1)}(n+1)^{-(n+1)}Γ\big(n+1,-(n+1)ln(x)\big)\big|_0^{\infty}= \sum_{n=0}^{\infty}\frac{Γ\big(n+1,-(n+1)ln(\infty)\big)}{Γ(n+1)(n+1)^{n+1}}}$$

This -$\infty$ in the gamma function argument does not seem to exist, but taking a limit helps as well as some other forms with the Regularized Incomplete Gamma Function: $$A=\mathrm{\sum_{n=1}^{\infty}n^{-n}\frac{Γ(n,-\infty n)}{Γ(n)}=\lim_{x\to \infty}\sum_{n\ge1}\frac{Q(n,-nx)}{n^n}=1.99545595…}$$

As seen from the link, the index n has to be a natural number which is true here. Also, there are other forms with the exponential sum function, and other series representations if non-elementary functions are not allowed. Here is proof of this answer from Wolfram Alpha.

Please do not give me any integral expressions of this constant, but rather special functions or other representations are encouraged. I will “check” the best answer. I just need a “nice” non-integral expression of the constant. Please give me feedback and correct me!

An alternate form solution is very needed for this problem also concerning power tower, so please check it out. I have left my context, attempts, and questions, therefore any help with alternate forms for either of these would be appreciated.

Using the definition of the Riemann Sum, we can rewrite this as:

$$\mathrm{\int_{\Bbb R^+}x^{-x}dx=\lim_{b,n\to \infty}\frac bn \sum_{k=0}^n\left(\frac{bk}{n}\right)^{-\left(\frac{bk}{n}\right)},n\gg b}$$

Here is proof of this result: Graph

Here are some unlikely closed forms.

There is also a form of the constant using a summation expansion. Here is proof of a third form of this constant here: $$\mathrm{A= \lim_{x\to \infty}\sum_{n\ge 1}\sum_{k=0}^{n-1}\frac{(-1)^k e^{nx} n^{k-n}x^k}{k!}}$$

Here is another sum and double sum answer here. This has a sophomore dream in the solution plus the same integrand from 1 to $\infty$: $$\mathrm{A=\sum_{n\ge 1}n^{-n}-\lim_{x\to \infty}\sum_{n\ge1}\sum_{k\ge1}\frac{(-x)^n(nx)^k}{(k+n)k!n!}}$$

Let $\bar F$ represent a Regularized Hypergeometric function then, as $\mathrm{Q(n,-nx)= 1-(-nx)^n\ _1\bar F_1(n,n+1,nx)}$, the following can be proven with the Kummer Confluent Hypergeometric Function: $$\mathrm{\sum_{n\ge1}n^{-n}-\lim_{x\to\infty}\sum_{n\ge1}(-x)^n\ _1\bar F_1(n,n+1,nx)}$$

Also, feel free to use these identities to find an alternate form. I could go on, but I have found no form that does not need a limit in the answer.

Тyma Gaidash
  • 12,081
  • 8
    It's not a competition to see who can use the most fonts in one post. – Gerry Myerson May 20 '21 at 13:05
  • 3
    There is a funny-looking A. There is a display that's about ten feet high. There is boldface. There is Fraktur in blue. It's too busy (for my taste). – Gerry Myerson May 20 '21 at 23:05
  • 2
    @GerryMyerson Do you have any alternate forms for this constant? You seem very talented and I would love to see some possible closed forms of this constant, although a closed form is optional as this does not seem to have one. You can also click the fraktur link, that is the default link color. Please no integral defined forms as we are trying to evaluate this integral, and find the area more importantly! – Тyma Gaidash May 20 '21 at 23:10
  • It's tabulated to over 100 digits at https://oeis.org/A229191 where there is also a reference to Finch's book and links to some related constants. I doubt there's a closed form. – Gerry Myerson May 20 '21 at 23:41
  • 2
    Sorry, I don't know anything more than what's at the oeis. I encourage you to follow the links there. – Gerry Myerson May 21 '21 at 02:48
  • 1
    Just as a joke $$\frac{2+6 \sqrt{2}-6 \sqrt{3}-2 e+3 \pi +4 \pi ^2+8 \log (2)+2 \log (3)}{2+3 \sqrt{2}-6 \sqrt{3}+5 e+4 \pi -\log (2)+4 \log (3)}$$ gives 18 exact decimals. – Claude Leibovici Jun 19 '21 at 09:06
  • @ClaudeLeibovici I found 12 other jokes here with possible closed forms. However, a non-closed form would also work. Any ideas? – Тyma Gaidash Jun 19 '21 at 12:38
  • 1
    I know a form: $$A\bar =\int_{\Bbb R^+} x^{-x} dx=\int_0^{1/e} x^{-x} dx+\int_0^\sqrt[e]e-\frac{ln(x)}{W(-ln(x)}dx -e^{\frac 1e-1}≈.48-.53+2.04=A$$. I already figured out the $x^{-x}$ integral and can use something like this to solve for the productlog/w-lambert integral… – Тyma Gaidash Jun 22 '21 at 13:38
  • So far I have to use a single series for this middle integral. To integrate, I used a theorem about the integral of an inverse function, but this diverges. I must use an inversion theorem to find a double sum solution. The goal is to get rid of the limit at the beginning and find a more elementary solution. The first integral is $$ .48≈\sum_{n\ge1}\frac {Q(n,n)}{n^n}=\sum_{n\ge 0} \frac{E_{-n}(n+1)}{n!}$$ with the generalized exponential integral function. – Тyma Gaidash Jun 22 '21 at 16:27
  • Yay 1000 reputation!!… – Тyma Gaidash Jun 22 '21 at 16:32

2 Answers2

4

From : https://fr.scribd.com/doc/34977341/Sophomore-s-Dream-Function $$\text{Closed form : }\quad\int_0^\infty x^{-x}dx=\text{Sphd}(-1;\infty)$$ $$\text{Numerical value : Eq.(8:3).}$$

Before downvoting, please read "A schoolboy joke" pp.2-3 in the above referenced paper.

JJacquelin
  • 66,221
  • 3
  • 37
  • 87
2

This is a note. I will try to find more in the near future. $$ I=\int^{\infty}_{0}x^{-x}dx=\int^{\infty}_{1}e^{-x\log x}dx+\int^{1}_{0}x^{-x}dx. $$ Using $\sum_{n\geq 1}n^{-n}=\int^{1}_{0}x^{-x}dx$ and making the change of variable $x\rightarrow e^{W(x)}$ where $W(x)$ is the well known Lambert function, we get $$ I=\sum_{n\geq 1}n^{-n}+\int^{\infty}_{0}\exp\left(-e^{W(x)}W(x)\right)e^{W(x)}W'(x)dx= $$ $$ =\sum_{n\geq 1}n^{-n}+\int^{\infty}_{0}e^{-x}e^{W(x)}W'(x)dx=\sum_{n\geq 1}n^{-n}+\int^{\infty}_{0}e^{-x}d\left(e^{W(x)}\right)= $$ $$ \textrm{ integration by parts } $$ $$ =\sum_{n\geq 1}n^{-n}-1+\int^{\infty}_{0}e^{-x}e^{W(x)}dx=\sum_{n\geq 1}n^{-n}-1+\int^{\infty}_{0}e^{-x}\frac{x}{W(x)}dx $$ If we make the change of variable $x\rightarrow -\log t$ we arrive easily to $$ I=\sum_{n\geq 1}n^{-n}-1+\int^{1}_{0}t\frac{-\log t}{W(-\log t)}\frac{-dt}{t}=\sum_{n\geq 1}n^{-n}-1+\int^{1}_{0}\frac{-\log t}{W(-\log t)}dt. $$ But the function $h(z)=\frac{W(-\log z)}{-\log z}$, is the analytic continuation of the Euler's iterated exponential function $z^{z^{z^{\ldots}}}$. Hence finaly we can write $$ \int^{\infty}_{0}\frac{1}{x^x}dx=\sum_{n\geq 1}\frac{1}{n^n}-1+\int^{1}_{0}\frac{1}{t^{t^{t^{ \ldots}}}}dt. $$ In the sense $z^{z^{z^{\ldots}}}=h(z)$.

REVISED.

Assume that $W(t)$ is the well known Lambert function. Then for $k=0,1,2,\ldots$, we have $$ \int^{M}_{0}\frac{t^k}{1+W(t)}dt=\frac{(-1)^{k+1}k!}{(k+1)^{k+1}}+ $$ $$ +\frac{M^{k+1} k!}{W(M)^{k+1}(k+1)^{k+1}}\sum_{0\leq j\leq k}(-1)^{j+k}\frac{(k+1)^jW(M)^j}{j!}. $$ Hence $$ \int^{M}_{0}\frac{e^{-t}}{1+W(t)}dt=-\sum^{\infty}_{k=0}\frac{1}{(k+1)^{k+1}}+ $$ $$ +\sum^{\infty}_{k=0}\frac{e^{-(k+1)W(M)} M^{k+1}W(M)^{-k-1}\Gamma(k+1,-(k+1)W(M))}{(k+1)^{k}(k+1)!}= $$ $$ =-\sum_{k\geq 1}k^{-k}+\sum_{k\geq 1}\frac{\Gamma(k,-kW(M))}{k^{k-1}k!},\tag 1 $$ where $$ \Gamma(a,z)=\int^{\infty}_{z}e^{-t}t^{a-1}dt $$ is the incomplete Gamma function. But one can see with the help of (1) easily that $$ \int^{M}_{1}x^{-x}dx=\int^{M\log M}_{0}\frac{e^{-t}}{1+W(t)}dt=-\sum^{\infty}_{k=1}k^{-k}+\sum^{\infty}_{k=1}\frac{\Gamma(k,-kW(M\log M))}{(k!) k^{k-1}}. $$ Hence using $\int^{1}_{0}x^{-x}dx=\sum^{\infty}_{k=1}k^{-k}$, we get $$ \int^{M}_{0}x^{-x}dx=\sum^{\infty}_{k=1}\frac{\Gamma(k,-k\log M)}{(k!) k^{k-1}}\textrm{, }\forall M>0.\tag 2 $$

  • 1
    See https://en.wikipedia.org/wiki/Lambert_W_function#Identities at the end of the section and deffinition of the W-Lambert function. – Nikos Bagis Jun 22 '21 at 20:20
  • The function $h(x)=x^{x^{x^{\ldots}}}$ converges in $\textbf{R}$ only for $e^{-e}\leq x\leq e^{1/e}$. That is why I use analytic continuation. Actualy $g(x)=W(-\log x)/(-\log x)$ coinsides with $h(x)$ at $[e^{-e},e^{1/e}]$, but $g(x)$ is defined in all $x\neq 1$. – Nikos Bagis Jun 22 '21 at 20:45
  • I did something similar here trying to find the area under h(x) instead from 0 to $e^{1/e}$. I am very close to finding the answer, but I kind of had a similar problem as you with as the interval of convergence of the series I used failed to find the full area. Problem here – Тyma Gaidash Jun 22 '21 at 20:51
  • Nice! I will check this soon enough. You can simplify by using the regularized gamma function. Please evaluate the final answer with M=$\infty$. Thanks. You will see you get the same result I did with a limit unfortunately as you will have to make the second argument of the incomplete gamma function be: -$\infty$k. – Тyma Gaidash Jul 12 '21 at 21:54
  • Yes I see the formula it is the same as yours. Also writing (2) with $M$ and not $\infty$ gives more information to the problem. I will search the problem more. However this formula does not apear in Wikipedia. – Nikos Bagis Jul 13 '21 at 08:21
  • Correct, it does give more information, but you can also simplify the gamma function if you like. Thanks for keeping your word that you will edit your nice answer. – Тyma Gaidash Jul 13 '21 at 13:20