# Fractional Binomial Coefficients

I recently examined the binomial coefficient $\binom{\frac{1}{2}}{k}$ and found that the denominator was always a power of two. The same is true of $\binom{\frac{1}{3}}{k}$, where the denominator is always a power of three. While the first proof was simple, the case of $\frac{1}{3}$ was messy and involved counting the powers of 3 in the numerator and denominator. The case of $\frac{1}{p}$ for prime $p$ can be done in the same bashy way. Is it possible there is a more elegant proof?

-
$$\binom{\frac1{p}}{k}=\frac{(-1)^k}{k!}\prod_{j=0}^{k-1} \left(j-\frac1{p}\right)$$ might be a useful identity... – J. M. Jul 9 '12 at 16:01
This identity leads to a bashy algebraic solution. I was looking for something more elegant and "combinatorial", unless I'm missing some algebra trick that makes that identity easy to work with. – Timothy Salus Jul 9 '12 at 16:31
Nevermind I see how that identity can be used elegantly after more thought. Thanks! I'm new to the site; is it generally advised for me to post a full solution if I have discovered one? – Timothy Salus Jul 9 '12 at 16:42
Answering your own question is warmly encouraged here at StackExchange; there is even a "Self-Learner" badge. – J. M. Jul 9 '12 at 16:52
You can check too that $(1+4x)^{1/2}$ has integral coefficients. – Lubin Jul 9 '12 at 22:25

## 4 Answers

Using the identity $$\binom{\frac{1}{p}}{k}=\frac{(-1)^k}{k!}\prod_{j=0}^{k-1} \left({j-\frac{1}{p}}\right)$$ I found a fairly simple proof for why the denominator of this binomial coefficient is always a power of $p$.

Let $p^a$ be the highest power of $p$ that divides $k!$. Then, $$\prod_{j=0}^{k-1}\left({j-\frac{1}{p}}\right) \equiv \prod_{j=0}^{k-1}({j+n}) \equiv 0 \pmod{ \frac{k!}{p^a}}$$ for some $n \equiv \frac{1}{p} \pmod {\frac{k!}{p^a}}$, with congruence to 0 resulting from the fact that $k!$ always divides a product of $k$ consecutive integers. Thus we have $$\prod_{j=0}^{k-1}\left({j-\frac{1}{p}}\right)=\left(\frac{k!}{p^a}\right)z$$ for $z \in \mathbb{Z}$. Substituting then provides,

$$\binom{\frac{1}{p}}{k}=\frac{(-1)^k}{k!}\left(\frac{k!}{p^a}\right)z=\frac{(-1)^kz}{p^a}$$ completing the proof.

-
Well done.$\phantom{}$ – J. M. Jul 10 '12 at 4:57

Observe \begin{align} (1 + x)^{1/p} = 1 + \frac{x}{p} - \frac{1-p}{2p^{2}} x^{2} + \frac{1 - 3p + 2p^{2}}{6 p^{3}} x^{3} - \cdots \end{align}

-
See J.M.'s comment for the exact form of the coefficients. – user02138 Jul 9 '12 at 16:05

Here’s a purely conceptual proof, more advanced, that has the advantage of involving no computation and no induction:

1. Let $N$ be a positive number, and let $P_N(t)$ be the polynomial function that tells you the coefficient of $x^N$ in $(1+x)^t$. You know that $P_N$ has rational coefficients, and takes integer values when $t$ is evaluated to a positive integer.
2. Now consider any prime number $q$. The polynomial $P_N$ has its coefficients in the field ${\mathbb{Q}}_q$ of $q$-adic numbers, is continuous, and takes values in the $q$-adic integers ${\mathbb{Z}}_q$ whenever $t$ is evaluated to a positive integer.
3. But any rational number without $q$ in its denominator is a $q$-adic integer, so $q$-adically approximable by ordinary integers, and in fact by positive integers. Thus if $\alpha$ is a rational with no $q$ in its denomiator, we get $P_N(\alpha)\in{\mathbb{Z}}_q$, by continuity of $P_N$.
4. In particular, the coefficients of $(1+x)^{1/p}$ are rational numbers that are in ${\mathbb{Z}}_q$ for each $q\ne p$: the only denominators are powers of $p$.

A more advanced argument allows you to say just how divisible by $p$ the denominators of the coefficients of $(1+x)^{1/p}$ will be, but that’s a story for another day.

-
Unfortunately I have just about zero experience with the $p$-adic integers, but thank you for your answer! – Timothy Salus Jul 9 '12 at 22:39

Firstly, note the definition of binomial coefficients with arbitrary real upper co-efficient:

$${r\choose k}=\begin{cases}\frac{r^{\underline{k}}}{k!} & k\in\mathbb{Z}^{*} \\ 0 & k\in\mathbb{Z}^{-}\end{cases}$$

Where $r^{\underline{k}}=r(r-1)(r-2)\cdots(r-k+1)$ is the falling factorial.

Now for your problem, let $r=\frac{1}{x},\forall x\in\mathbb{N}$. It is fairly evident that the bottom will contain at least one power of $x$, whilst the top will contain no powers of $x$, therefore $x$ will divide the denominator of ${r \choose k}$. Which is what you are trying to prove.

-
It's clear that $x$ divides the denominator. But what we want to prove is that the denominator is always just a power of $x$. – Timothy Salus Jul 9 '12 at 16:28