4

How would you prove this without calculus?

$$\forall m,n\in\Bbb N,\ \forall x\in\Bbb R,\ \sum_{k=0}^{m}\binom{n+k}{k}(1-x)^kx^{n+1}=1-\sum_{k=0}^{n}\binom{m+k}{k}x^k(1-x)^{m+1}$$

In this post it is proved with calculus, and I was wondering if there was an easier way to prove it. I tried proving it with induction, by holding $n$ constant, however I didn't seem to get anywhere.

Blue
  • 75,673
Memat
  • 471
  • 4
    To anyone who is thinking about voting to close because of context; Memat saw a pattern in the post they linked, and was curious about how to prove it, so they asked on MSE. It is an interesting question, and one that is quite hard to get traction on, and it is not for homework. Memat said they tried via induction on $n$ and got stuck; the same would happen to you if you tried that, since this is a tricky combinatorial identity. If you vote to close, you are voting against curiosity and searching for beauty. – Mike Earnest Dec 15 '22 at 21:31

2 Answers2

7

We can give a probabilistic proof. First, assume that $0\le x\le 1$. Imagine that Alice and Bob repeatedly play games of chess. Each time, Alice wins with probability $x$, and Bob wins with probability $1-x$, independently of previous rounds. The pair stop playing as soon as Alice accumulates $n+1$ wins, or when Bob accumulates $m+1$ wins.

The probability that the series ends with Alice having $n+1$ wins, and Bob having $k$ wins for some $k\in \{0,\dots,m\}$, is $\binom{n+k}k x^{n+1}(1-x)^k$.

The factor of $\binom{n+k}{k}$ accounts for all ways to arrange Bob's $k$ wins among the first $n+k$ outcomes.

Similarly, the probability the series ends with $m+1$ wins for Bob and $k$ wins for Alice is $\binom{m+k}{k}(1-x)^{m+1}x^k$. Since we have exhausted all possible outcomes, the probabilities must add up to one. We have shown

$$ \sum_{k=0}^n\binom{n+k}k x^{n+1}(1-x)^k + \sum_{k=0}^m\binom{m+k}k(1-x)^{m+1}x^k=1 $$ completing the proof. Note that the LHS is a polynomial with degree at most $m+n+1$, so the fact that it is equal to $1$ for all $x$ in the interval $[0,1]$ implies it equals $1$ for all $x\in \mathbb C$.

amWhy
  • 209,954
Mike Earnest
  • 75,930
  • Hi! Could you please explain the last part?

    "Not that the LHS is a polynomial with degree at most m+n+1, so the fact that it is equal to 1 for all x in the interval [0,1] implies it equals 1 for all x∈C"

    – Memat Dec 15 '22 at 21:13
  • 1
    @memat Let $f(x)$ be a polynomial such that $f(x)=1$ for infinitely many $x$. Then $f(x)-1$ is a polynomial with infinitely many zeroes. Since a nonzero polynomial with degree $n$ has at most $n$ zeroes, $f(x)-1$ must be the zero polynomial. – Mike Earnest Dec 15 '22 at 21:18
  • 1
    Also, I wrote "not" when I meant "note", which could cause confusion. – Mike Earnest Dec 15 '22 at 21:21
4

Induction works.

Let

$$a(n,m)=x^{n+1}\sum_{k=0}^{m}\binom{n+k}{k}(1-x)^k$$ $$b(n,m)=(1-x)^{m+1}\sum_{k=0}^{n}\binom{m+k}{k}x^k$$ $$f(n,m)=a(n,m)+b(n,m)$$

So we have to prove $f(n,m)=1$ for all $n,m\ge0$.

Let's prove this by induction on $m$.

First, $f(n,0)=1$ for all $n\ge0$, as $a(n,0)=x^{n+1}$ and $b(n,0)=(1-x)\sum_{k=0}^nx^k=1-x^{n+1}$.

Below, $n\ge0$ is fixed.

Let's assume $f(n,m)=1$ for some $m\ge0$. Then:

$$a(n,m+1)=a(n,m)+x^{n+1}\binom{n+m+1}{m+1}(1-x)^{m+1}$$ $$b(n,m+1)=(1-x)^{m+2}\sum_{k=0}^n\binom{m+k+1}{k}x^k=(1-x)^{m+2}\left\{\sum_{k=0}^n\binom{m+k}{k}x^k+\sum_{k=1}^n\binom{m+k}{k-1}x^k\right\}\\=(1-x)(1-x)^{m+1}\sum_{k=0}^n\binom{m+k}{k}x^k+(1-x)^{m+2}\sum_{k=1}^n\binom{m+k}{k-1}x^k\\=(1-x)b(n,m)+(1-x)^{m+2}\sum_{k=0}^{n-1}\binom{m+k+1}{k}x^{k+1}$$

Therefore

$$f(n,m+1)=f(n,m)+(1-x)^{m+1}\left\{\binom{n+m+1}{m+1}x^{n+1}-x\sum_{k=0}^n\binom{m+k}{k}x^k+(1-x)\sum_{k=0}^{n-1}\binom{m+k+1}{k}x^{k+1}\right\}$$

We want to prove that the expression inside the curly brackets is zero:

$$\binom{n+m+1}{m+1}x^{n+1}-x\sum_{k=0}^n\binom{m+k}{k}x^k+(1-x)\sum_{k=0}^{n-1}\binom{m+k+1}{k}x^{k+1}\\=\sum_{k=0}^n\binom{m+k+1}{k}x^{k+1}-x\sum_{k=0}^n\binom{m+k}{k}x^k-x\sum_{k=0}^{n-1}\binom{m+k+1}{k}x^{k+1}\\ =\sum_{k=0}^n\binom{m+k+1}{k}x^{k+1}-x\sum_{k=0}^n\binom{m+k}{k}x^k-x\sum_{k=1}^{n}\binom{m+k}{k-1}x^{k}\\ =\sum_{k=0}^n\binom{m+k+1}{k}x^{k+1}-\sum_{k=0}^n\binom{m+k}{k}x^{k+1}-\sum_{k=1}^{n}\binom{m+k}{k-1}x^{k+1}\\ =x-x+\sum_{k=1}^n\left\{\binom{m+k+1}{k}-\binom{m+k}{k}-\binom{m+k}{k-1}\right\}x^{k+1}\\=0$$

Where the $x-x$ in the last line comes from the index $k=0$ in the first two sums of the preceding line.

Therefore the induction step is proved: if $f(n,m)=1$, then $f(n,m+1)=1$. And by induction, the equality is true for all $m\ge0$. Since this is proved for any $n\ge0$, the equality $f(n,m)=1$ is true for all $n,m\ge0$.