3

If we have an arbitrary polynomial, $P:\mathbb{N}\to\mathbb{R}$, how can we express $P(x)$ as a linear combination of $\binom{x}{k}$. Is there any easy formula for finding this expression.

I am pretty sure a unique expression is guaranteed to exist because we can start with the highest degree term, let's say $a_nx^n$, and just subtract $n!a_n\binom{x}{n}$ to get an $n-1$ degree polynomial and then keep on repeating this process.

I think the most essential step is finding an expression for $x^n=\sum_{k=0}^n a_k\binom{x}{k}$. I have found by brute force that $$x^0=\binom{x}{0}$$ $$x^1=\binom{x}{1}$$ $$x^2=2\binom{x}{2}+\binom{x}{1}$$ $$x^3=6\binom{x}{3}+6\binom{x}{2}+\binom{x}{1}$$ I am not sure what the pattern is. It kind of looks like pascal's triangle rows (e.g. $1,2,1$ and $1,6,6,1$ is similar to $1,3,3,1$). However, finding expressions for higher power terms is pretty exhaustive, so I was wondering if there is an easier solution?

Alan Abraham
  • 5,142
  • 6
  • 20
  • 2
    What if you think of it in terms of inverting certain matrices? :) – Wolfgang Aug 07 '21 at 19:15
  • 2
    https://en.wikipedia.org/wiki/Falling_and_rising_factorials gives formulas for converting between factorial polynomials and ordinary polynomials – saulspatz Aug 07 '21 at 19:16
  • @Wolfgang I'm not sure. I'm not very adept with advanced linear algebra, so if you could give me more hints, I might be able to formulate a solution. – Alan Abraham Aug 08 '21 at 01:34
  • @saulspatz Looking at the wikipedia page you provided, I found $x^n=\sum_{k=0}^n S(n,k)\binom{x}{k}k!$. However, I was not able to find a proof of this. Do you know how to prove the statement? – Alan Abraham Aug 08 '21 at 01:41
  • 1
    If $P(x)=c_0+c_1x+\dots+c_nx^n$ then you can write $P(x)=a_0+a_1\binom{x}{1}+\dots+a_n\binom{x}{n}$ where $a_i = i!\sum_{k=i}^{n}{k\brace i}c_k$ where ${k\brace i}$ are stirling numbers of second kind - I mentioned it in this post. (it follows from the identity you have found, just mentioning it to link with the existing post) – Sil Aug 08 '21 at 15:06

1 Answers1

2

We have that (identity was noted in the comments)

$$x^n = \sum_{k=0}^n {x\choose k} {n\brace k} k!$$

This may be seen for $x$ a positive integer and since both sides are polynomials in $x$ (recall that ${x\choose k} = x^{\underline{k}}/k!$) it then holds for all i.e. complex $x.$

For a combinatorial proof, consider a vector of $n$ elements where each element may take on $x$ different values. Clearly we have $x^n$ such vectors. On the other hand we may determine a particular vector by choosing first the $k$ different values that appear, for a factor of ${x\choose k}$ and combine this with a partition of the $n$ slots of the vector into $k$ non-empty sets, for a factor of ${n\brace k}$. The sets in the partition may be matched to the values in $k!$ ways, thus completing the alternate count. We have the special case of $n=0$ which yields one as required.

For an algebraic proof write the RHS as follows:

$$\sum_{k=0}^n {x\choose k} \; k! \; n! [z^n] \frac{(\exp(z)-1)^k}{k!} \\ = n! [z^n] \sum_{k=0}^n {x\choose k} (\exp(z)-1)^k$$

Now if $n\gt x$ we may lower the upper limit of the sum to $x$ as ${x\choose k}$ is zero when $n\ge k\gt x.$ On the other hand if $n\lt x$ we may raise the upper limit to $x$ since $(\exp(z)-1)^k = z^k + \cdots$ and there is no contribution to $[z^n]$ when $x\ge k\gt n.$ We get

$$n! [z^n] \sum_{k=0}^x {x\choose k} (\exp(z)-1)^k = n! [z^n] \exp(xz) = x^n$$

as claimed.

Marko Riedel
  • 61,317