8
$\begingroup$

I have posted this question in StackExchange, but it didn't get any answers there. This question is important for my research. I got stuck on an infinite product which even WolframAlpha can't answer. Here's it: $$\prod_{n=2}^{\infty}\left(1-\frac{1}{n!}\right)$$

  • This is surely convergent, many tests work. Wolfram Alpha couldn't evaluate it, but gave an approximate value of $$0.395338567367445566032356200431180613$$

  • The decimal expansion is OEIS A282529, but the entry doesn't have much information. This constant is conjectured to be irrational, transcendental, and normal.

  • This Math.SE question asks specifically for a closed form, but it has no answers, so it doesn't solve my question.

Here's the work I did: \begin{align} \prod_{n=2}^{\infty}\left(1-\frac{1}{n!}\right)&=\lim_{N\to\infty}\frac{\prod_{N\geq n\geq2}(n!-1)}{\prod_{N\geq n\geq2}n!}\\[6pt] &=\lim_{N\to\infty}\frac{\prod_{N\geq n\geq2}(n!-1)}{1\cdot1\cdot2\cdot1\cdot2\cdot3\cdots1\cdot2\cdot\cdots N}\\[6pt] &=\lim_{N\to\infty}\frac{\prod_{N\geq n\geq2}(n!-1)}{1^N2^{N-1}3^{N-2}\cdots(N-1)^2N^1} \end{align} Now I don't know how to proceed. L'Hopital's rule doesn't work, since the numerator isn't a function of $N$ (it is, but the product should be solved before differentiating).

How can I evaluate it? A link to an article containing information about the constant will also help. Any help would be appreciated.

Note: A closed form isn't necessary; converting the product into a sum or integral will also help. Some special function representations will also be good.
I realized that what I did was not useful. I did some research and found these facts:

  • A representation of the Barnes-G function is $$G(N)=\frac{\Gamma(N)^{N-1}}{K(N)}$$ Where $K$ is the K-function.

  • A representation of the K-function is $$K(z)=\mathrm{exp}[\zeta'(-1,z)-\zeta'(-1)]$$

Now I used the first point and simplified the product to $$\prod_{n=2}^{\infty}\left(1-\frac{1}{n!}\right)=\lim_{N\to\infty}\frac{K(N+2)}{\Gamma(N+2)^2}\prod_{k=2}^{N}\frac{k!-1}{(N+1)!}$$ How can this be simplified? Is there any suction related to this? The hard thing to evaluate is this: $$\prod_{k=2}^{N}(k!-1)$$ I looked up in this article but couldn't find a related function. Is there an article that discusses(or at least, mentions) this product?
Now my main question has become:

Simplify, give information about or represent in terms of special functions the product:$$\prod_{k=2}^{N}(k!-1)$$

$\endgroup$
8
  • 1
    $\begingroup$ What is your question? Just a request for any references on what is known about this constant? Or on any alternative representations known? $\endgroup$ Commented Oct 23, 2020 at 10:29
  • 2
    $\begingroup$ Yes, it is easily shown to be convergent. But there is no reason to believe it has a closed form. $\endgroup$ Commented Oct 23, 2020 at 10:29
  • 3
    $\begingroup$ you ask to convert this into a sum: just take the exponent of $\sum_{n=2}^\infty \log(1-1/n!)$ --- I'm unsure what kind of answer you would expect here at MO... $\endgroup$ Commented Oct 23, 2020 at 10:30
  • $\begingroup$ @carloBeenakker that's an obvious thing, it wasn't asked. $\endgroup$
    – user167505
    Commented Oct 23, 2020 at 10:52
  • 4
    $\begingroup$ As Gerald Edgar says, there's no reason to believe that any alternative representation is possible. Do you have any reason to believe it does? What exactly about this constant do you need for your research? $\endgroup$ Commented Oct 23, 2020 at 11:15

2 Answers 2

10
$\begingroup$

I do not know if there is any closed form for this product, but you can rewrite it as follows. First, consider the logarithm of your product, so that you get: $$ L:=\log \left ( \prod_{n=2}^{\infty} (1-1/n!) \right) = \sum_{n=2}^{\infty} \log(1-1/n!)$$ Since $n \geq 2$, $\frac{1}{n!} < 1$, so we can use the Taylor series of the logarithm to obtain: $$ L=-\sum_{n=2}^{\infty} \sum_{k=1}^{\infty} \frac{1}{k (n!) ^ k} $$ We can interchange the two series: $$ L= -\sum_{k=1}^{\infty} \frac{1}{k} \sum_{n=2}^{\infty} \frac{1}{(n!) ^ k} $$ For $k=1$, the inner sum is $e-2$. For $k=2$, according to wolfram alpha the inner sum is equal to $I_0(2) - 2$, where $I_0$ denotes the modified Bessel function of the first kind. For $k \geq 3$, wolfram alpha gives the result $_0 F _{k-1} (; 1, ..., 1; 1) - 2$ (the number of one's excluding the last one is $k-1$, of course). So define $a_k$ to be the sum of the $k$-th series above. Then: $$ L=-\sum_{k=1}^{\infty} \frac{a_k}{k} $$ Thus, your product is: $$ \prod_{n=2}^{\infty} (1-1/n!) = e^L $$ I do not think that there is a known closed form for $L$, but at least you can rewrite the product in terms of some known functions, as you requested.

$\endgroup$
1
$\begingroup$

I also don't think that there's a more closed form than what we have, but let it be known that if you go through the steps of rewriting outlined in the other two threads already linked (as well as here), you can get rid of the infinite product in favour of a more gaugable sum. Namely, the product equals

$$\frac{1}{2}\, -\, \sum_{n=3}^\infty\, \frac{1}{n!}\cdot\prod_{k=3}^{n-1}\left(1-\frac{1}{k!}\right)$$

or, if we rewrite the first three terms via Egyptian fractions,

$$\frac{1}{3} + \frac{1}{16} + \frac{1}{6912}\, -\, \frac{1}{2}\sum_{n=6}^\infty\, \frac{1}{n!}\cdot\prod_{k=3}^{n-1}\left(1-\frac{1}{k!}\right)$$

explicitly suggesting the range of your result $0.395338567\dots$

and where the summands are all dominated by $\frac{1}{n!}$.

Here's the routine:

1/2 - N[Sum[Product[1 - 1/k!, {k, 2, n - 1}] / n!, {n, 3, 100}], 50]

$\endgroup$
2
  • $\begingroup$ Your answer is good, I would have accepted it, but my main question has become simplifying, giving information about or representing in terms of special functions the product $$\prod_{k=2}^{N}(k!-1)$$. $\endgroup$
    – user167505
    Commented Oct 25, 2020 at 13:53
  • 1
    $\begingroup$ @epic_math It's a simple natural number valued product that grows vastly faster than an exponential function. As far as asymptotic growth is concerned, it behaves like the BarnesG (i.e. like as if you drop the $-1$). I'm not sure what you expect to get out of expressing it in terms of something more complicated (such as a transcentendal function which can only be approximated in computation)? Interesting question about this function that I can come up with would be computer science type question, e.g. small memory requirement algorithms to iteratively compute its values for different $N$. $\endgroup$
    – Nikolaj-K
    Commented Oct 25, 2020 at 15:48

You must log in to answer this question.