7
$\begingroup$

Does the logarithmic integral function $\operatorname{li}(x)$ have the continued fraction expansion $$\operatorname{li}(x) = \cfrac{x}{\log x -1 -{}} \ \cfrac{1}{\log x -3 -{}} \ \cfrac{4}{\log x -5 -{}}\ \cfrac{9}{\log x - 7-{}}\ \cfrac{16}{\log x - 9 -{}} \ \cfrac{25}{\log x - 11 -{}} \ \cdots$$ for $x > 1$? If so, is there a proof or a reference that proves it?

This question is not answered by any reference I can find because standard results in the literature will verify the identity (appropriately interpreted) for complex values of $x$ excluding $x > 1$. For $x > 1$, I do not even know if the given continued fraction converges.

It might help to note that the $n$th convergent of the continued fraction above is equal to $-\sum_{k = 1}^n \frac{1}{kL_k(x)L_{k-1}(x)}$, where $L_k(x)$ denotes the $k$th Laguerre polynomial (at least at values of $x$ that are not roots of any Laguerre polynomial). Thus, the question is equivalent to the following: does one have $\operatorname{li}(x) = -\sum_{k = 1}^\infty \frac{1}{kL_k(x)L_{k-1}(x)}$ for all real $x > 1$ that are not roots of any Laguerre polynomial?

It is well known that the exponential integral $E_1(z)$ has the continued fraction expansion $$E_1(z) = \cfrac{e^{-z}}{z+1 -{}} \ \cfrac{1}{z+3 -{}} \ \cfrac{4}{z+5 -{}}\ \cfrac{9}{z+7-{}}\ \cfrac{16}{z+9 -{}} \ \cfrac{25}{z+11 -{}} \ \cdots, \quad z \in \mathbb{C} \setminus (-\infty,0].$$ A third equivalent formulation of the question is the following: For $z \in (-\infty,0]$, does the continued fraction above converge to $-\operatorname{li}(e^{-z}) = E_1(z)+\pi i$?

EDIT: In all the references I can find, including the ones given in the proposed answer by Alexey Ustinov, the domain for which the given expansions hold exclude the domains I inquired about in my question. The domain $x> 1$ of $\operatorname{li}(x)$ is the domain number theorists care most about, and it would be nice if it had the proposed continued fraction expansion on that domain.

FINAL EDIT: I now think it's more likely that the continued fraction diverges for $x > 1$, but I don't know how to prove this.

$\endgroup$
3
  • $\begingroup$ Do you really mean $\log(x) - 5 + \dotsb$ when all the others are $-$? $\endgroup$
    – LSpice
    May 3, 2021 at 0:37
  • $\begingroup$ Why $+$ at $5$ exclusively? $\endgroup$
    – Turbo
    May 3, 2021 at 0:38
  • 1
    $\begingroup$ typo. fixing now. $\endgroup$ May 3, 2021 at 0:59

1 Answer 1

5
$\begingroup$

You can find this expansion in the book Lorentzen L. & Waadeland H. Continued fractions with applications North-Holland Publishing Co., 1992 (formula (4.3.10)). As I understand, this document is a more recent version of the appendix of this book. Here the desired formula has the number (3.3.10). However, the validity of the expansion for $\operatorname{li}(x)$ does not include the domain $x > 1$ that you seek.

$\endgroup$
4
  • 1
    $\begingroup$ The PDF link goes to the freely available "Back matter" of the Springer (re-?)issue of the referenced book. Here's a DOI link to the book: Lorentzen and Waadeland - Continued fractions - I. $\endgroup$
    – LSpice
    May 3, 2021 at 14:50
  • $\begingroup$ This won't do it. The expansion for $\operatorname{Ei}(-z)$ is proved valid only for $|\operatorname{arg}(z)| < \pi$ and the expansion for $\operatorname{li}(z)$ for $|\operatorname{arg}(-\operatorname{Ln}(z))|< \pi$, which rules out the domain of $x > 1$ for $\operatorname{li}(x)$ that I'm interested in. (I actually have a copy of the book.) $\endgroup$ May 3, 2021 at 20:44
  • $\begingroup$ Again, (3.3.10) holds only for $|\operatorname{arg}(-\operatorname{Ln}(z)| < \pi$, which excludes the domain $x > 1$. Do you see the issue here? $\endgroup$ May 5, 2021 at 20:50
  • $\begingroup$ I now think the continued fraction actually diverges for $x > 1$, but I don't know how to prove this. $\endgroup$ May 24, 2021 at 2:26

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.