3

When I was messing about with integrals, I came to know about this result: $$f(x) = \int_0 ^x \left(\frac{x}{t}\right)^t \mathrm dt = \sum _{n=1}^\infty \left(\frac{x}{n}\right)^n$$ and the related but just slightly less beautiful $$g(x) = \int_0 ^x \left(\frac{t}{x}\right)^t \mathrm dt = \sum _{n=1}^\infty (-1)^{n+1} \left(\frac{x}{n}\right)^n.$$ These results can be found by keeping the things inside the integrals in the form $e^{\ln k(x)}$, writing out the Taylor series of $e^x$, and then finally integrating by parts (that's how I found them) or you can do some u-substitutions to find the gamma function in here. Are these functions special? Can the integrals be evaluated? If not, can other non-elementary integrals be expressed in terms of this function? I tried to evaluate the integrals in terms of other functions, but I was unsuccessful. I do not think this has an elementary solution.

Edit: I realized that $g(x)$ can be written kind of like $f(x)$. $$g(x) = \int_0 ^x \left(\frac{t}{x}\right)^t \mathrm dt = \int_0 ^x \left(\frac{x}{t}\right)^{-t} \mathrm dt$$ Substituting $u=-t$, we have $$g(x) = -\int_0 ^{-x} \left(\frac{x}{t}\right)^t \mathrm dt$$ So we get $$\int_0 ^x \left(\frac{x}{t}\right)^t \mathrm dt + \int_0 ^{-x} \left(\frac{x}{t}\right)^t \mathrm dt = \sum_{n=1} ^{\infty}\left(\frac{x}{2n+1}\right)^{2n+1}$$ and $$\int_0 ^x \left(\frac{x}{t}\right)^t \mathrm dt - \int_0 ^{-x} \left(\frac{x}{t}\right)^t \mathrm dt = \sum_{n=1} ^{\infty}\left(\frac{x}{2n}\right)^{2n}$$ Which is pretty cool.

Edit: Here's something to think about. What happens if we use the Euler-Maclaurin Summation formula on the case where $x=1$? What if we use it on any $x$?

  • Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. – Community Feb 17 '25 at 12:01
  • 3
    Reminds me of https://en.wikipedia.org/wiki/Sophomore%27s_dream – Martin R Feb 17 '25 at 12:06
  • @MartinR It does! And Bernoulli evaluated it the same way I did! This one is a generalization though. – Supernerd411 Feb 17 '25 at 12:12
  • What's interesting to note is that $$\frac{d^nf}{dx^n} = 1+ \int_0^x \left( \frac{x}{t} \right)^{t-n} dt = \sum_{k=n+1}^{\infty} \left( \frac{x}{k} \right)^k$$ so maybe a differential equation isn't too impossible. As for solving your integral..I am stumped. $x=1$ is sophomore's dream, and that doesn't have a closed form. So it's unlikely for your integral to have a closed form either. It might have some nice asymptotic, but that's not my expertise – Kraken Feb 17 '25 at 13:01
  • @Kraken I thought the same thing. I made this hypothesis too just after looking at the case $n=1$, but this pattern doesn't continue. – Supernerd411 Feb 17 '25 at 13:07
  • Oh really? I'll be damned! Thanks for informing. It seems we can find a closed form though if we find a way to evaluate $$\int_0^b a^u u^n (1-u)^n du$$ – Kraken Feb 17 '25 at 17:35
  • @Kraken Wolframalpha seems to have something for that if you plug in some values of $n$. It gives an error it you let it be $n$. – Supernerd411 Feb 18 '25 at 13:05
  • @Kraken How did you get that? – Supernerd411 Feb 18 '25 at 13:57
  • @Supernerd411 my result is wrong again, and this is actually because I made the wrong substitution (I did about 7 substitutions which could be all done by 1 substitution, but I lacked the intuition to do it..) – Kraken Feb 18 '25 at 15:34
  • Somewhat related: https://math.stackexchange.com/q/3052328 – Gary Mar 11 '25 at 07:08

0 Answers0