The Fabius function is a smooth monotone function $F:[0,1]\to[0,1]$, satisfying functional equations $$F(0)=0, \quad F(1-x)=1-F(x)\tag1$$ and $$F'(x) = 2 \,F(2 x) \quad \text{for} \,\, 0<x<1/2.\tag2$$

Fabius graph

The function $F$ assumes rational values at dyadic rational arguments. In particular, it is known $\!^{[1]}$$\!^{[2]}$ that $$F\left(2^{-n}\right) = \frac1{n! \, 2^{\binom{n+1}2}} \, \sum_{m\ge0}\binom n {2m} \, c_m,\tag3$$ where $c_m$ are defined by the recurrence $$c_0 = 1, \quad c_n = \frac1{(4^n - 1)(2n+1)} \, \sum_{m\ge1} \binom{2n+1}{2m+1} \, c_{n-m}.\tag4$$ Note that only finite number of terms in each sum are non-zero.

The values $F\left(2^{-n}\right) $ appear as A272755 / A272757 in the OEIS.

Let $$a_n = F\left(2^{-n}\right) \, 2^{\binom {n-1}2} \, (2n)! \, \prod_{m=1}^{\lfloor n/2\rfloor}\left(4^m - 1\right).\tag5$$ This sequence begins $$1, \, 5, \, 15, \, 1001, \, 5985, \, 2853675, \, 26261235, \, 72808620885, \, 915304354965 \, ...\tag6$$ (see more terms here)

I conjecture that all terms of this sequence are integers. How can we prove (disprove) this conjecture?

  • 1
    $\begingroup$ My initial thought was to prove that your "fudge factor" used when going from $F$ to $a$ would be more than enough to make each term in the defining sum integral, but it's more delicate than that (as you probably knew). For example the $2^{-n(n+1)/2}/n!$ term when multiplied by the fudge factor is not in general an integer; there are powers of 2 in the denominator. So one has to think harder. $\endgroup$ – Kevin Buzzard Feb 7 '17 at 23:39

I have posted in arXiv:1702.05442 the English translation of a paper about Fabius function that I published in Spanish in 1982 (we will refer to it as (A)). With the Theorems in this paper the question can be answered without much difficulty. I have posted also in arXiv:1702.06487 a new paper with the complete answer to this question. Here we only show the main points of this proof.

I will call $R_n$ your numbers $a_n$. First it is shown that $$R_{2n+1}=F_n\frac{(4n+1)!!}{(2n-1)!!}.$$ where $F_n$ are natural numbers introduced in my old paper (A). This shows the conjecture for $R_n$ in case $n$ is an odd number.

For the other case, first prove that the numbers $R_n$ are related by the equation $$R_n=2d_n (2n-1)!! \prod_{k=1}^{\lfloor n/2\rfloor}(2^{2k}-1)$$ with some rational numbers $d_n$ introduced in (A). These numbers $d_n$ are related to the $F_n$ so that at the end we get $$R_{2n}=\sum_{k=0}^n \frac{2 F_k}{2^{2n}}\binom{2n}{2k}\frac{(4n-1)!!}{(2k+1)!!} \prod_{\ell=k+1}^n(2^{2\ell}-1)$$ This shows that the denominator of $R_{2n}$ is at most a power of $2$.

The relation between $R_n$ and $d_n$ implies that the same power will appear in the denominator of $2d_n$. An easy induction shows that this power is $0$.

We have other results about the values of Fabius function in arXiv:1702.06487 including a relation with Bernoulli numbers. This paper is a preliminary version.

| cite | improve this answer | |

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.