3

I'm interested in binomial coefficients of the form $$\binom{-1/m}{k} .$$ Although I have found some sources indicating this expression can be rewritten in an interesting way when $m=2$, I have found little information so far on it when $m \in \mathbb{Z}_{\geq 3}$.

I am therefore curious as to whether any articles or book chapters on this particular binomial expression have been written. Information regarding sums involving this expression is also much appreciated.

Max Muller
  • 7,006
  • +1 Interesting query, very nicely presented. – user2661923 Oct 04 '20 at 13:25
  • Related: https://math.stackexchange.com/questions/688276/combinatorics-binomial-coefficient-with-negative-fractions – Alessio K Oct 04 '20 at 13:34
  • @Äres Indeed. I've also referenced that question in another question of mine on this topic: https://math.stackexchange.com/questions/3850064/is-there-a-closed-form-for-the-binomial-expression-binom-1-mk – Max Muller Oct 04 '20 at 13:36

2 Answers2

4

Consider the formal power series $y(x)$ satisfying $y^m = \frac{1}{1 - mx}$, which gives

$$y(x) = (1 - mx)^{- \frac 1 m} = \sum_{n \ge 0} (-1)^n m^n {-\frac 1 m \choose n} x^n.$$

Taking logs and exponentiating again gives

$$y(x) = \exp \left( \frac{1}{m} \log \frac{1}{1 - mx} \right) = \exp \left( \sum_{k \ge 1} \frac{m^{n-1} x^n}{n} \right).$$

This gives us that $n! [x^n] y(x)$ counts permutations but where a permutation with cycles of length $\ell_1, \dots \ell_n$ is weighted by $m^{\sum (\ell_i - 1)}$. The quantity $\sum (\ell_i - 1)$ is sometimes called the length of a permutation; it gives the minimal $\ell$ such that a permutation can be expressed as a product of $\ell$ transpositions. It's equal to $n$ minus the number of cycles, which means we get the Stirling numbers of the first kind but in reverse order:

$$\boxed{ (-1)^n m^n {-\frac 1 m \choose n} = \frac{1}{n!} \sum_{i=0}^n \left[ {n \atop i} \right] m^{n-i} }.$$

Compare to the generating function for the Stirling numbers in the usual order, which is

$$(-1)^n {-m \choose n} = {n+m-1 \choose n} = \frac{1}{n!} \sum_{i=0}^n \left[ {n \atop i} \right] m^i$$

and which corresponds to looking at $(1 - x)^{-m}$.


What you might've been hoping for was an expression in terms of ${mn \choose n}$ generalizing the case $m = 2$, and it turns out that some nice things can be said about this sequence. Using Lagrange inversion or the more explicit combinatorial argument in this blog post we can show that the formal power series $z(x)$ satisfying $z = 1 + xz^m$ has coefficients

$$z(x) = \sum_{n \ge 0} \frac{1}{(m-1)n + 1} {mn \choose n} x^n$$

which generalizes the well-known generating function for the Catalan numbers, to which it reduces when $m = 2$ (and in fact these coefficients count $m$-ary trees in a way that generalizes the way the Catalan numbers count binary trees). This is Example 6.2.6 in Stanley's Enumerative Combinatorics, Vol. II. Example 6.2.7 uses this to show that the formal power series $w(x) = \sum_{n \ge 0} {mn \choose n} x^n$ satisfies

$$\frac{w - 1}{1 + (m-1)w} = x \left( \frac{mw}{1 + (m-1)w} \right)^m$$

so like $y$ and $z$ it is also algebraic, but its minimal polynomial appears to be more complicated. For example, setting $m = 3$ and clearing denominators gives that $w = \sum_{n \ge 0} {3n \choose n} x^n$ satisfies

$$(w - 1)(1 + 2w)^2 = 27x w^3$$

and expanding and rearranging gives

$$(27x - 4) w^3 + 3w + 1 = 0.$$

So more complicated than $(1 - 3x) y^3 = 1$, unfortunately. Stanley is probably your best bet for learning more about this sort of thing; it's extremely comprehensive, especially the exercises. The series $w$ appears again in Exercise 6.13.

Qiaochu Yuan
  • 419,620
  • Thank you! I did not expect the connection with the Stirling numbers of the first kind. – Max Muller Oct 06 '20 at 11:13
  • do you perhaps know whether similar identities exist for $\binom{m-1/m}{k} $ ? – Max Muller Oct 07 '20 at 13:42
  • @Max: you can try the same style of argument for something like $(1 - mx)^{m - \frac 1 m}$ and see what you get. I don't think it's anything particularly nice. – Qiaochu Yuan Oct 07 '20 at 19:43
  • Alright, I will try that just to make sure. I'm asking about this mainly to gain a better understanding of the following questions. If you have any ideas, I'd be interested in reading about them: (1) https://mathoverflow.net/questions/373499/a-question-on-an-identity-relating-certain-sums-of-harmonic-numbers , (2) https://mathoverflow.net/questions/373231/what-is-known-about-sums-of-the-form-sum-n-2-infty-zetan-1p – Max Muller Oct 07 '20 at 21:11
  • I'm sorry to come back to this yet again, but I have to ask: do you know whether the boxed identity appears somewhere in the literature? – Max Muller Oct 13 '20 at 11:53
  • 1
    @Max: it is arguably a very trivial consequence of the usual identity involving the Stirling numbers in the usual order. In general if $P(x)$ is a polynomial of degree $n$ then $x^n P(1/x)$ reverses its coefficients. – Qiaochu Yuan Oct 13 '20 at 17:48
  • very interesting indeed ! – G Cab Dec 10 '21 at 01:20
1

A simple identity:

$${(1+x)^{1/m}}=\sum_{k\geq 0}{\binom{1/m}{k}}x^k.$$

converting it to the binomial form $$\binom{km}{k}$$ is far harder though

For m=3, it arises when finding the roots of a cubic equation [1] and certain enumerative combinatorics problems in which the characteristic equation is a cubic , such as random walks involving two steps forward and one step back. The 1 step forward, one step back process involves the usual m=2, but m=3 case is far harder and nice analytic expressions as far as I know do not exist ,except for simple cases in which there is a single absorbing barrier.

see:

[1] Find the second real root for cubic $x^3+1-x/b=0$

CarP24
  • 300