5

How to deal with combinatoric interpretation (or just solving it in algebraic way) when we have $(-1)^i$ factor in our sum?
Example task:
Simplify the sum: $$ \sum_{i=0}^{k}(-1)^i i \binom{n}{i} \binom{n}{k-i} \text{ for } 0\le k \le n $$

For task without $(-1)^i$ $$ \sum_{i=0}^{k} i \binom{n}{i} \binom{n}{k-i} = n \binom{2 n-1}{k-1} $$ I can write that interpretation:

  • I have $n$ rabbits and $k$ slots
  • Each rabbit can be in both slot of first type and second type
  • slots of first type + second type = $k$
  • Lets double rabbits
  • I choose one rabbit as an king and it will be also a rabbit to slot of first type
  • so I need to choose $2n-1$ rabbit for $k-1$ slots
    But I don't know how to deal with $(-1)^i$
Tengu
  • 4,072
  • 1
    $(-1)^i$ could very much point to the inclusion-exclusion principle. Did you learn about it? – Theorem Apr 06 '19 at 16:35
  • I am not sure about that in this case - yes I have learned about that but how it can be used there @Theorem? –  Apr 06 '19 at 16:36
  • That might be because this sum doesn't have a nice closed form? In most cases $(-1)^i $ in combinatoric problems is a definite inclusion-exclusion. Do you have context to this problem? – Theorem Apr 06 '19 at 16:41
  • No, I have taken that problem from old exam from my faculty. I checked in wolfram and it claim that the result is $ -n \binom{n}{k-1} , _2F_1(1-k,1-n;-k+n+2;-1)$ but I don't know what is it (checked on wiki that is https://en.wikipedia.org/wiki/Hypergeometric_function but I haven't got that on my lecture) and how I can get this. The task was just "simplify the sum" –  Apr 06 '19 at 16:44
  • I can show that the answer is $\dfrac{k}{2} \left(-1\right)^{k/2} \dbinom{n}{k/2}$ when $k$ is even. Interestingly, the answer for $k$ odd seems to be $\dfrac{k+1}{2} \left(-1\right)^{\left(k+1\right)/2} \dbinom{n}{\left(k+1\right)/2}$. – darij grinberg Apr 06 '19 at 16:53
  • Can you show how you got that? @darijgrinberg –  Apr 06 '19 at 16:54
  • @VirtualUser: Done. (My new argument covers both possible parities of $k$.) – darij grinberg Apr 06 '19 at 21:29

3 Answers3

3

This is a really neat exercise. Here is the answer:

Theorem 1. Let $n\in\mathbb{N}$. (Here, as always, $\mathbb{N}=\left\{ 0,1,2,\ldots\right\} $.) Let $m=\left\lfloor \left( n+1\right) /2\right\rfloor $. Then, \begin{equation} \sum\limits_{k=0}^{n}\left( -1\right) ^{k}k\dbinom{x}{k}\dbinom{x}{n-k}=m\left( -1\right) ^{m}\dbinom{x}{m} \end{equation} as polynomials in $\mathbb{Q}\left[ x\right] $.

Note that my $x$, $n$ and $k$ are your $n$, $k$ and $i$ (sorry for this -- I am taking the lazy route and adapting your notations to mine), and I have extended the domains for $x$ (promoted from a lowly integer to a polynomial indeterminate) and $n$ (now any nonnegative integer).

The proof will rely on the following two facts:

Lemma 2. Let $k$ be a positive integer. Then, \begin{equation} k\dbinom{x}{k}=x\dbinom{x-1}{k-1}\qquad\text{as polynomials in } \mathbb{Q}\left[ x\right] . \end{equation}

Proof of Lemma 2. This is usually stated in the equivalent form $\dbinom {x}{k}=\dfrac{x}{k}\dbinom{x-1}{k-1}$; in this form it is:

You will have likely proven it by the time you have found it in these sources. Note that this identity is the key to algebraic proofs of various identities with "$k\dbinom{x}{k}$"s in them -- such as $\sum\limits_{k=0}^{n}k\dbinom{n}{k}=n2^{n-1}$ and $\sum\limits_{k=0}^{n}\left( -1\right) ^{k}k\dbinom{n}{k}= \begin{cases} -1, & \text{if }n=1;\\ 0, & \text{if }n\neq1 \end{cases} $ for all $n\in\mathbb{N}$. $\blacksquare$

Lemma 3. Let $n\in\mathbb{N}$. Then, \begin{equation} \sum\limits_{k=0}^{n}\left( -1\right) ^{k}\dbinom{x}{k}\dbinom{x}{n-k}= \begin{cases} \left( -1\right) ^{n/2}\dbinom{x}{n/2}, & \text{if }n\text{ is even};\\ 0, & \text{if }n\text{ is odd} \end{cases} \label{darij1.eq.l3.eq} \tag{1} \end{equation} as polynomials in $\mathbb{Q}\left[ x\right] $.

Proof of Lemma 3. This is Exercise 3.22 in my Notes on the combinatorial fundamentals of algebra, version of 10 January 2019. Alternatively, if $x$ is specialized to a nonnegative integer, you can use Mike Spivey's argument at Alternating sum of squares of binomial coefficients (which is stated for the particular case $n=x$, but can easily be adapted to the general case -- see my comment under his post) to prove \eqref{darij1.eq.l3.eq} combinatorially; then, use the "polynomial identity trick" to un-specialize $x$. You can probably find lots of other approaches on math.stackexchange. Either way, Lemma 3 is proven. $\blacksquare$

Now, we can prove Theorem 1:

Proof of Theorem 1. It is easy to prove Theorem 1 in the case when $n=0$. (Indeed, in this case, both sides of the equality in question equal $0$, since they are products in which one of the factors is $0$.) Thus, for the rest of this proof, we WLOG assume that $n\neq0$. Hence, $n>0$. Thus, $n-1 \in \mathbb{N}$.

We shall use the convention that $\dbinom{u}{v}=0$ whenever $v\notin \mathbb{N}$. Thus, the recurrence of the binomial coefficients, \begin{equation} \dbinom{u}{v}=\dbinom{u-1}{v-1}+\dbinom{u-1}{v}, \label{darij1.pf.t1.1} \tag{2} \end{equation} holds not only for $v\in\left\{ 1,2,3,\ldots\right\} $ but for all $v\in\mathbb{Z}$.

Lemma 3 (applied to $n-1$ instead of $n$) yields \begin{align*} \sum\limits_{k=0}^{n-1}\left( -1\right) ^{k}\dbinom{x}{k}\dbinom{x}{\left( n-1\right) -k} & = \begin{cases} \left( -1\right) ^{\left( n-1\right) /2}\dbinom{x}{\left( n-1\right) /2}, & \text{if }n-1\text{ is even};\\ 0, & \text{if }n-1\text{ is odd} \end{cases} \\ & = \begin{cases} 0, & \text{if }n-1\text{ is odd;}\\ \left( -1\right) ^{\left( n-1\right) /2}\dbinom{x}{\left( n-1\right) /2}, & \text{if }n-1\text{ is even} \end{cases} \\ & = \begin{cases} 0, & \text{if }n\text{ is even;}\\ \left( -1\right) ^{\left( n-1\right) /2}\dbinom{x}{\left( n-1\right) /2}, & \text{if }n\text{ is odd} \end{cases} \end{align*} (since $n-1$ is odd if and only if $n$ is even, and vice versa). Substituting $x-1$ for $x$ in this equality, we obtain \begin{equation} \sum\limits_{k=0}^{n-1}\left( -1\right) ^{k}\dbinom{x-1}{k}\dbinom{x-1}{\left( n-1\right) -k}= \begin{cases} 0, & \text{if }n\text{ is even;}\\ \left( -1\right) ^{\left( n-1\right) /2}\dbinom{x-1}{\left( n-1\right) /2}, & \text{if }n\text{ is odd.} \end{cases} \label{darij1.pf.t1.n-1} \tag{3} \end{equation}

If $n>1$, then $n-2\in\mathbb{N}$. Hence, if $n>1$, then Lemma 3 (applied to $n-2$ instead of $n$) yields \begin{align*} \sum\limits_{k=0}^{n-2}\left( -1\right) ^{k}\dbinom{x}{k}\dbinom{x}{\left( n-2\right) -k} & = \begin{cases} \left( -1\right) ^{\left( n-2\right) /2}\dbinom{x}{\left( n-2\right) /2}, & \text{if }n-2\text{ is even};\\ 0, & \text{if }n-2\text{ is odd} \end{cases} \\ & = \begin{cases} \left( -1\right) ^{\left( n-2\right) /2}\dbinom{x}{\left( n-2\right) /2}, & \text{if }n\text{ is even};\\ 0, & \text{if }n\text{ is odd} \end{cases} \end{align*} (since $n-2$ is even if and only if $n$ is even, and since $n-2$ is odd if and only if $n$ is odd). This equality holds not only for $n>1$, but also for $n=1$ (since both of its sides equal $0$ in this case), and thus holds in all cases (since we have $n\geq1$). Substituting $x-1$ for $x$ in this equality, we obtain \begin{equation} \sum\limits_{k=0}^{n-2}\left( -1\right) ^{k}\dbinom{x-1}{k}\dbinom{x-1}{\left( n-2\right) -k}= \begin{cases} \left( -1\right) ^{\left( n-2\right) /2}\dbinom{x-1}{\left( n-2\right) /2}, & \text{if }n\text{ is even};\\ 0, & \text{if }n\text{ is odd.} \end{cases} \end{equation} The left hand side of this equality does not change if we replace the summation sign "$\sum\limits_{k=0}^{n-2}$" by "$\sum\limits_{k=0}^{n-1}$" (because the only new addend that we gain in this way is $\left( -1\right) ^{n-1}\dbinom{x-1}{n-1} \underbrace{\dbinom{x-1}{\left( n-2\right) -\left(n-1\right)}}_{\substack{ = 0 \\ \text{(since $\left(n-2\right)-\left(n-1\right) = -1 \notin \mathbb{N}$)}}} = 0$). Hence, this equality becomes \begin{equation} \sum\limits_{k=0}^{n-1}\left( -1\right) ^{k}\dbinom{x-1}{k}\dbinom{x-1}{\left( n-2\right) -k}= \begin{cases} \left( -1\right) ^{\left( n-2\right) /2}\dbinom{x-1}{\left( n-2\right) /2}, & \text{if }n\text{ is even};\\ 0, & \text{if }n\text{ is odd.} \end{cases} \label{darij1.pf.t1.n-2} \tag{4} \end{equation}

We can split off the addend for $k=0$ from the sum $\sum\limits_{k=0}^{n}\left( -1\right) ^{k}k\dbinom{x}{k}\dbinom{x}{n-k}$ (since $n\geq0$). Thus, we find \begin{align} & \sum\limits_{k=0}^{n}\left( -1\right) ^{k}k\dbinom{x}{k}\dbinom{x}{n-k} \nonumber\\ & =\underbrace{\left( -1\right) ^{0}0\dbinom{x}{0}\dbinom{x}{n-0}}_{=0} +\sum\limits_{k=1}^{n}\left( -1\right) ^{k}k\dbinom{x}{k}\dbinom{x}{n-k} \nonumber\\ & =\sum\limits_{k=1}^{n}\underbrace{\left( -1\right) ^{k}}_{=-\left( -1\right) ^{k-1}}\underbrace{k\dbinom{x}{k}}_{\substack{=x\dbinom{x-1}{k-1}\\\text{(by Lemma 2)}}}\underbrace{\dbinom{x}{n-k}}_{\substack{=\dbinom{x-1} {n-k-1}+\dbinom{x-1}{n-k}\\\text{(by \eqref{darij1.pf.t1.1}, applied} \\ \text{to $u = x$ and $v = n-k$)}}}\nonumber\\ & =\sum\limits_{k=1}^{n}\left( -\left( -1\right) ^{k-1}\right) x\dbinom{x-1} {k-1}\left( \dbinom{x-1}{n-k-1}+\dbinom{x-1}{n-k}\right) \nonumber\\ & =-x\sum\limits_{k=1}^{n}\left( -1\right) ^{k-1}\dbinom{x-1}{k-1}\left( \dbinom{x-1}{n-k-1}+\dbinom{x-1}{n-k}\right) . \label{darij1.pf.t1.4} \tag{5} \end{align}

Now, \begin{align*} & \sum\limits_{k=1}^{n}\left( -1\right) ^{k-1}\dbinom{x-1}{k-1}\left( \dbinom {x-1}{n-k-1}+\dbinom{x-1}{n-k}\right) \\ & =\sum\limits_{k=0}^{n-1}\left( -1\right) ^{k}\dbinom{x-1}{k}\left( \underbrace{\dbinom{x-1}{n-k-2}}_{=\dbinom{x-1}{\left( n-2\right) -k} }+\underbrace{\dbinom{x-1}{n-k-1}}_{=\dbinom{x-1}{\left( n-1\right) -k} }\right) \\ & \qquad\left( \text{here, we have substituted }k+1\text{ for }k\text{ in the sum}\right) \\ & =\underbrace{\sum\limits_{k=0}^{n-1}\left( -1\right) ^{k}\dbinom{x-1}{k} \dbinom{x-1}{\left( n-2\right) -k}}_{\substack{= \begin{cases} \left( -1\right) ^{\left( n-2\right) /2}\dbinom{x-1}{\left( n-2\right) /2}, & \text{if }n\text{ is even};\\ 0, & \text{if }n\text{ is odd} \end{cases} \\\text{(by \eqref{darij1.pf.t1.n-2})}}}+\underbrace{\sum\limits_{k=0}^{n-1}\left( -1\right) ^{k}\dbinom{x-1}{k}\dbinom{x-1}{\left( n-1\right) -k} }_{\substack{= \begin{cases} 0, & \text{if }n\text{ is even;}\\ \left( -1\right) ^{\left( n-1\right) /2}\dbinom{x-1}{\left( n-1\right) /2}, & \text{if }n\text{ is odd} \end{cases} \\\text{(by \eqref{darij1.pf.t1.n-1})}}}\\ & = \begin{cases} \left( -1\right) ^{\left( n-2\right) /2}\dbinom{x-1}{\left( n-2\right) /2}, & \text{if }n\text{ is even};\\ 0, & \text{if }n\text{ is odd} \end{cases} + \begin{cases} 0, & \text{if }n\text{ is even;}\\ \left( -1\right) ^{\left( n-1\right) /2}\dbinom{x-1}{\left( n-1\right) /2}, & \text{if }n\text{ is odd} \end{cases} \\ & = \begin{cases} \left( -1\right) ^{\left( n-2\right) /2}\dbinom{x-1}{\left( n-2\right) /2}+0, & \text{if }n\text{ is even;}\\ 0+\left( -1\right) ^{\left( n-1\right) /2}\dbinom{x-1}{\left( n-1\right) /2}, & \text{if }n\text{ is odd} \end{cases} \\ & = \begin{cases} \left( -1\right) ^{\left( n-2\right) /2}\dbinom{x-1}{\left( n-2\right) /2}, & \text{if }n\text{ is even;}\\ \left( -1\right) ^{\left( n-1\right) /2}\dbinom{x-1}{\left( n-1\right) /2}, & \text{if }n\text{ is odd} \end{cases} \\ & = \begin{cases} \left( -1\right) ^{\left\lfloor \left( n-1\right) /2\right\rfloor } \dbinom{x-1}{\left\lfloor \left( n-1\right) /2\right\rfloor }, & \text{if }n\text{ is even;}\\ \left( -1\right) ^{\left\lfloor \left( n-1\right) /2\right\rfloor } \dbinom{x-1}{\left\lfloor \left( n-1\right) /2\right\rfloor }, & \text{if }n\text{ is odd} \end{cases} \\ & \qquad\left( \begin{array} [c]{c} \text{since }\left( n-2\right) /2=\left\lfloor \left( n-1\right) /2\right\rfloor \text{ when }n\text{ is even,}\\ \text{and since }\left( n-1\right) /2=\left\lfloor \left( n-1\right) /2\right\rfloor \text{ when }n\text{ is odd} \end{array} \right) \\ & =\left( -1\right) ^{\left\lfloor \left( n-1\right) /2\right\rfloor }\dbinom{x-1}{\left\lfloor \left( n-1\right) /2\right\rfloor }. \end{align*} Thus, \eqref{darij1.pf.t1.4} becomes \begin{align} & \sum\limits_{k=0}^{n}\left( -1\right) ^{k}k\dbinom{x}{k}\dbinom{x}{n-k} \nonumber\\ & =-x\underbrace{\sum\limits_{k=1}^{n}\left( -1\right) ^{k-1}\dbinom{x-1} {k-1}\left( \dbinom{x-1}{n-k-1}+\dbinom{x-1}{n-k}\right) }_{=\left( -1\right) ^{\left\lfloor \left( n-1\right) /2\right\rfloor }\dbinom {x-1}{\left\lfloor \left( n-1\right) /2\right\rfloor }}\nonumber\\ & =-x\left( -1\right) ^{\left\lfloor \left( n-1\right) /2\right\rfloor }\dbinom{x-1}{\left\lfloor \left( n-1\right) /2\right\rfloor } . \label{darij1.pf.t1.7} \tag{6} \end{align}

On the other hand, recall that $m=\left\lfloor \left( n+1\right) /2\right\rfloor $, so that $m-1=\left\lfloor \left( n+1\right) /2\right\rfloor -1=\left\lfloor \underbrace{\left( n+1\right) /2-1} _{=\left( n-1\right) /2}\right\rfloor =\left\lfloor \left( n-1\right) /2\right\rfloor $. Also, $m=\left\lfloor \left( n+1\right) /2\right\rfloor \geq1$ (since $n\geq1$ and thus $\left( n+1\right) /2\geq1$). Hence, $m$ is a positive integer; thus, Lemma 2 (applied to $k=m$) yields $m\dbinom{x} {m}=x\dbinom{x-1}{m-1}$. Now, \begin{align*} m\left( -1\right) ^{m}\dbinom{x}{m} & =\underbrace{\left( -1\right) ^{m} }_{=-\left( -1\right) ^{m-1}}\underbrace{m\dbinom{x}{m}}_{=x\dbinom {x-1}{m-1}}=-\left( -1\right) ^{m-1}x\dbinom{x-1}{m-1}\\ & =-x\left( -1\right) ^{m-1}\dbinom{x-1}{m-1}=-x\left( -1\right) ^{\left\lfloor \left( n-1\right) /2\right\rfloor }\dbinom{x-1}{\left\lfloor \left( n-1\right) /2\right\rfloor } \end{align*} (since $m-1=\left\lfloor \left( n-1\right) /2\right\rfloor $). Comparing this with \eqref{darij1.pf.t1.7}, we obtain \begin{equation} \sum\limits_{k=0}^{n}\left( -1\right) ^{k}k\dbinom{x}{k}\dbinom{x}{n-k}=m\left( -1\right) ^{m}\dbinom{x}{m}. \end{equation} This proves Theorem 1. $\blacksquare$

2

Darij Grinberg's answer cited a very nice combinatorial proof which I reproduce here for completeness.

Let $[n]=\{1,2,\dots,n\}$. We provide a combinatorial interpretation for the form $$\sum_i (-1)^in\binom{n-1}{i-1}\binom{n}{k-i}$$ This is a signed count of ordered triples $(x,A,B)$, where $x\in [n], A\subseteq [n]\setminus \{x\},B\subseteq [n]$, and $|A|+|B|=k-1$. A triple is counted positively if $|A|$ is odd, and negatively otherwise.

Given such a triple $(x,A,B)$, we define its partner $f(x,A,B)$ as follows. Find the largest element of $(A\setminus B)\cup (B\setminus (A\cup \{x\}))$, and move it to the other set. If this set is empty, we leave $f$ undefined.

You can check that $f(f(x,A,B))=(x,A,B)$ whenever $f$ is defined, so that this is a well defined pairing operation. Furthermore, since $(x,A,B)$ and $f(x,A,B)$ have opposite signs, they cancel out each other in the summation, so they can be ignored.

Therefore, the only triples which contribute to the count are those for which $f$ is undefined. The only triples for which $f$ is undefined are those of the form $(x,A,A)$ and $(x,A,A\cup \{x\})$. Only one of these forms is possible, depending on the parity of $k$, and you can check that in either case the number of triples is $$ n\binom{n-1}{\lfloor(k-1)/2\rfloor} $$ and each exceptional triple has sign $(-1)^{\lfloor(k-1)/2\rfloor + 1}$.

Mike Earnest
  • 75,930
  • 1
    Nice proof! Your set $(A\setminus B)\cup (B\setminus (A\cup {x}))$ can also be written as the symmetric difference $A \triangle \left(B \setminus \left{x\right}\right)$. I'm saying this because it makes it a lot easier to see why your $f$ is an involution. – darij grinberg Apr 07 '19 at 19:28
1

Starting from

$$\sum_{q=0}^k (-1)^q q {n\choose q} {n\choose k-q}$$

we have

$$\sum_{q=1}^k (-1)^q q {n\choose q} {n\choose k-q} = n \sum_{q=1}^k (-1)^q {n-1\choose q-1} {n\choose k-q} \\ = n [z^k] (1+z)^n \sum_{q=1}^k (-1)^q {n-1\choose q-1} z^q \\ = - n [z^{k-1}] (1+z)^n \sum_{q=1}^k (-1)^{q-1} {n-1\choose q-1} z^{q-1}.$$

Now if $q\gt k$ then there is no contribution to the coefficient extractor:

$$- n [z^{k-1}] (1+z)^n \sum_{q\ge 1} (-1)^{q-1} {n-1\choose q-1} z^{q-1} \\ = - n [z^{k-1}] (1+z)^n (1-z)^{n-1} = - n [z^{k-1}] (1+z) (1-z^2)^{n-1} \\ = - n [z^{k-1}] (1-z^2)^{n-1} - n [z^{k-2}] (1-z^2)^{n-1}.$$

If $k$ is odd this yields

$$-n (-1)^{(k-1)/2} {n-1\choose (k-1)/2}$$

and if it is even

$$-n (-1)^{(k-2)/2} {n-1\choose (k-2)/2}.$$

Join these to obtain

$$\bbox[5px,border:2px solid #00A000]{ (-1)^{1+\lfloor (k-1)/2 \rfloor} \times n \times {n-1\choose \lfloor (k-1)/2 \rfloor}.}$$

Marko Riedel
  • 61,317