9

4 . (a) Prove that $$\sum_{k=0}^l \binom{n}{k} \binom{m}{l-k} = \binom{n+m}{l}.$$ Hint: Apply the binomial theorem to $(1+x)^n(1+x)^m$.

I'm having a hard time trying to solve the problem above. I've done all of the previous exercises of the 2nd chapter with little difficulty, so far. I think I might be missing a trivial point somewhere.


The answer I got from the Answer Book, and is not very helpful either... :(

4. (a) Since $(1+x)^n(1+x)^m = (1+x)^{n+m}$ we have $$\sum_{k=0}^n \binom{n}{k}x^k\cdot\sum_{j=0}^m \binom{m}{j}x^j\cdot=\sum_{l=0}^{n+m} \binom{n+m}{l}x^l$$ But the coefficient of $x^l$ on the left is clearly $$\sum_{k=0}^l\binom{n}{k}\binom{m}{l-k}.$$
One term of the sum occurring for each pair $k$, $j = l-k$.

I couldn't get the last part of the answer:
why is it that the "coefficient of $x^l$ on the left is clearly $\sum_{k=0}^l\binom{n}{k}\binom{m}{l-k}$."?

FRD
  • 259

6 Answers6

13

Maybe this will help you think about the coefficients. We have $n$ boys and $m$ girls, and we want to form a committee of $l$ people from these $n+m$ people. Clearly there are $\binom{n+m}{l}$ ways to form the committee.

Let's count the number of committees another way. We could have $0$ boys and $l$ girls. Such a committee can be formed in $\binom{n}{0}\binom{m}{l}$ ways.

Or we could have $1$ boy and $l-1$ girls. Such a committee can be formed in $\binom{n}{1}\binom{m}{l-1}$ ways.

Or we could have $2$ boys and $l-2$ girls. Such a committee can be formed in $\binom{n}{2}\binom{m}{l-2}$ ways.

Continue. The total number of committees is $$\sum_{k=0}^l \binom{n}{k}\binom{m}{l-k}.\tag{$1$}$$ But we already saw that the number of committees is $\binom{n+m}{l}$.

Note: It is possible that for example there is a total of $3$ boys and $24$ girls, and we want to form a committee of $7$ people. Then the formula appears to break down. But it doesn't if we agree that $\binom{a}{b}=0$ if $b\gt a$.

To apply the reasoning to $(1+x)^n(1+x)^m$, you might first look at $(1+x)^n(1+y)^m$, and set $y=x$ at the end. So expand both, multiply. For fixed $l$, gather together terms that have a combined total of $l$ $x$'s (boys) and/or $y$'s. The total number will be given by Formula $(1)$.

André Nicolas
  • 507,029
12

Multiplication of formal power series is performed by collecting the terms with the same powers of $x$: $$ \begin{align} \left(\sum_{k=0}^\infty a_kx^k\right)\left(\sum_{k=0}^\infty b_kx^k\right) &=\sum_{k=0}^\infty\left(\sum_{j=0}^k a_j\color{#C00000}{x^j}b_{k-j}\color{#C00000}{x^{k-j}}\right)\\ &=\sum_{k=0}^\infty\left(\sum_{j=0}^k a_jb_{k-j}\right)\color{#C00000}{x^k}\tag{1} \end{align} $$ Note that the subscripts in the inner sum add up to $k$, the power of $x$ in the outer sum.

Apply $(1)$ to the product of $$ (1+x)^m=\sum_{k=0}^\infty\binom{m}{k}x^k\tag{2} $$ and $$ (1+x)^n=\sum_{k=0}^\infty\binom{n}{k}x^k\tag{3} $$ which is $$ (1+x)^{m+n}=\sum_{k=0}^\infty\binom{m+n}{k}x^k\tag{4} $$ I extended the indices in the sums to $\infty$ since for $k>n$, $\binom{n}{k}=0$.

For the product of $(2)$ and $(3)$, we get $$ (1+x)^m(1+x)^n=\sum_{k=0}^\infty\left(\sum_{j=0}^k \binom{m}{j}\binom{n}{k-j}\right)x^k\tag{5} $$ Comparing the coefficients of $x^k$ in $(4)$ and $(5)$ yields $$ \binom{m+n}{k}=\sum_{j=0}^k \binom{m}{j}\binom{n}{k-j}\tag{6} $$ as desired.

robjohn
  • 345,667
7

Note that since $\binom{n}{k} = {n (n-1)\cdots (n-k+1) \over k!}$, we can let $\binom{n}{k} = 0 $ if $k > n$. This will simplify notation.

Adopting this convention, we get $\sum_{k=0}^\infty \binom{n}{k}x^k\cdot\sum_{j=0}^\infty \binom{m}{j}x^j = \sum_{k=0}^\infty \sum_{j=0}^\infty \binom{n}{k} \binom{m}{j}x^{k+j}$.

Let $I_1 = \{ (k,j) | k, j \ge 0 \}$ and $I_2 = \{ (p,l) | 0 \le p \le l \}$, then we have $I_1 = \{ (p, l-p) | (p,l) \in I_2 \}$.

Here is one way to see this: Define the linear operator $\phi((x,y) = (x,x+y)$, the inverse is $\phi^{-1} ((x,y)) = (x,y-x)$. It is straighforward to check that $\phi(I_1) \subset I_2$ and $\phi^{-1}(I_2) \subset I_1$, hence $\phi(I_1) = I_2$.

In particular, $f$ is zero for all but a finite number of terms, we have $\sum_{(k,j) \in I_1} f(k,j) = \sum_{(p,l) \in I_2} f(p,l-p)$.

Hence we can write \begin{eqnarray} \sum_{(k,j) \in I_1} \binom{n}{k} \binom{m}{j}x^{k+j} &=& \sum_{k=0}^\infty \sum_{j=0}^\infty \binom{n}{k} \binom{m}{j}x^{k+j} \\ &=& \sum_{(p,l) \in I_2} \binom{n}{p} \binom{m}{l-p}x^{l} \\ &=& \sum_{l=0}^{^\infty} \sum_{p=0}^l \binom{n}{p} \binom{m}{l-p}x^{l} \end{eqnarray} Now we have $$\sum_{l=0}^{^\infty} \sum_{p=0}^l \binom{n}{p} \binom{m}{l-p}x^{l} = \sum_{l=0}^{^\infty} \binom{n+m}{l}x^l ,$$ which are polynomials of $x$ on both sides (note that only a finite number of coefficients can be non zero). Since this equality is true for all $x$, the polynomials are equal and hence so are the coefficients of each $x^l$ (differentiate $l$ times and set $x=0$ to convince yourself). It follows that $$\sum_{p=0}^l \binom{n}{p} \binom{m}{l-p} = \binom{n+m}{l} .$$

copper.hat
  • 172,524
  • Hi, I am trying to understand arriving at the right side of this: $$\sum_{k=0}^n \sum_{j=0}^m \binom{n}{k} \binom{m}{j}x^{k+j} = \sum_{l=0}^{n+m} \sum_{p=0}^l \binom{n}{p} \binom{m}{l-p}x^{l}.$$ Specifically the $l-p$ part. Maybe a hint on the "convince yourself" part? – lopan Dec 23 '22 at 00:26
  • @lopan THere is something wrong there, thanks for catching that. – copper.hat Dec 23 '22 at 06:18
  • @lopan I added a fix. – copper.hat Dec 23 '22 at 07:14
  • Thanks. I am struggling with the index set notation here: $(k,j) = p(1, -1) + l(0,1)$ which could apparently be written $(k,j) = k(1, -1) + k+j(0,1)$ Assume for instance $(k,j) = (1,1)$, how should I interpret the resulting notation $1(1,-1) + 2(0,1)$? Just not clear on what this is expressing conceptually. – lopan Dec 23 '22 at 15:09
  • @lopan I added a few edits, hopefully this clarifies. – copper.hat Dec 24 '22 at 23:36
  • Thanks for helping, I am getting there. Since $k+j=l$, we have $j=l-k$. We let $k=p$, $j=l-p$. Since $k=p$ why is it the counter on the outer loop prior to our transformation, but it is the counter on the inner loop afterward? – lopan Dec 25 '22 at 18:27
  • @lopan I could easily have made a mistake, let me check. – copper.hat Dec 25 '22 at 18:45
  • @lopan Don't think of it like that. Think of it as summing over a set of indices (which is what it is). If you are comfortable with $\sum_{(k,j) \in I_1} f(k,j) = \sum_{(p,l) \in I_2} f(p,l-p)$, then we can write the right hand side as $\sum_{(p,l) \in I_2} f(p,l-p) = \sum_{l=0}^\infty \sum_{p=0}^l f(p,l-p)$. – copper.hat Dec 25 '22 at 18:49
  • @lopan Another view point: If you are comfortable with $\phi(I_1) = I_2$ and the fact that $\phi$ is a bijection, then we have $\sum_{(k,j) \in I_1} f(k,j) = \sum_{(k,j) \in \phi^{-1}(I_2)} f(k,j) =\sum_{(p,l) \in I_2} f \circ \phi^{-1}(p,l-p) =\sum_{(p,l) \in I_2} f(p,l-p)$. – copper.hat Dec 25 '22 at 18:55
  • I finally see the big picture, but I'll have to plead insanity for the time being on the index set machinery. Essentially, the left-side is generating a sum of all the terms systematically by iterating on $(k,j)$, but the point of the right-side is to generate the sums of like coefficients grouped by exponents of $x$ from $0$ to $n+m$ which prepares the proof that each such sum on the left corresponds to and equals $n+m \choose l$ where $l$ is the exponent of $x$. I'm not sure now whether I feel like breakdancing or vomiting, but I very much appreciate the help. – lopan Dec 26 '22 at 00:47
  • 1
    "Like coefficients" meaning the $l+1$ ways to choose $l$ from $n+m$. – lopan Dec 26 '22 at 00:58
  • @lopan sometimes I stare at stuff again and again for hours before I get the point. – copper.hat Dec 26 '22 at 01:48
4

If you multiply $c_0+c_1x+\cdots+c_nx^n$ by $d_0+d_1x+\cdots+d_mx^m$, then for $k\leq\min(n,m)$ you will get terms involving $x^k$ from the products $c_0\times d_kx^k$, $\,c_1x\times d_{k-1}x^{k-1}$, $\,c_2x^2\times d_{k-2}x^{k-2}$,..., $c_kx^k\times d_0$, and from no other products. Adding those contributions gives $(c_0d_k+c_1d_{k-1}+c_2d_{k-2}+\cdots+c_kd_0)x^k$. Even if one should have $k>\min(n,m)$, the coefficient of $x^k$ is clearly always the sum of all $c_id_j$ with $i+j=k$, which you can write as $\sum_{i=0}^kc_id_{k-i}$ provided one defines $c_i$ or $d_j$ to be $0$ when the subscript is too large. And in your example the binomial coefficient expressions indeed become $0$ when the lower index is too large.

2

You should determine the coefficient of $x^l$ in $(1+x)^n(1+x)^m$ in two ways. First multiply this out to get $(1+x)^n(1+x)^m=(1+x)^{n+m}$. So considering the right hand side and using the binomial theorem we get $\binom{n+m}{l}$.

Now consider the coefficient of $x^l$ in the left hand side after expanding each term using the binomial theorem $$(1+x)^n(1+x)^m=(\sum_{p=0}^m \binom{m}{p}x^p)(\sum_{q=0}^n \binom{n}{q}x^q)$$ You get a coefficient of $x^l$ in the product when you multiply $\binom{m}{p}x^p\binom{n}{q}x^q$ when $p+q=l$, i.e. $q=l-p$.

Then we have to take into account all the ways this can happen and comparing it to what we got on the right side so we have $$\sum_{p=0}^l\binom{m}{p}\binom{n}{l-p}=\binom{m+n}{l}$$

0

I thought proof by induction would be interesting which uses a property of Pascal's triangle. Assume it is true for $<l$. Induction on $n+m$. For $n=0$ it is true. Assume that it is true for $n$. For $n+1$, $$\binom{n+1+m}{l}=\sum_{k=0}^l\binom{n+1}{k}\binom{m}{l-k}=\sum_{k=0}^l\left[\binom{n}{k}+\binom{n}{k-1}\right]\binom{m}{l-k}=\hspace{1cm}\binom{n+m}{l}+\sum_{k=1}^l\binom{n}{k-1}\binom{m}{l-k}$$ Hence we get $$\binom{n+1+m}{l}-\binom{n+m}{l}=\sum_{k+1=1}^{k+1=l}\binom{n}{(k+1)-1}\binom{m}{l-(k+1)}$$ $$\binom{n+m}{l-1}=\sum_{k=0}^{l-1}\binom{n}{k}\binom{m}{(l-1)-k}$$ which is true by induction..

Bob Dobbs
  • 10,988
  • For what it's worth, the point of the exercise in the book was to deduce it directly by multiplying the two binomials, as the hint directed. You're in effect duplicating the effort of proving the binomial theorem, which was already done in the preceding problem. – Ted Shifrin Dec 25 '22 at 00:20
  • I didn't know. I tried a different solution. If OP dislikes I delete. – Bob Dobbs Dec 25 '22 at 00:34
  • Well, IMHO you should try to respond to the particular question the OP asked before giving a completely different argument. – Ted Shifrin Dec 25 '22 at 00:51
  • You are right... I thought 10 years later he may be interested in different solutions... And there are enough answers, why is this question not closed? – Bob Dobbs Dec 25 '22 at 01:07
  • Questions are not closed for being old. Note that the OP already accepted an answer which actually addressed his specific question. – Ted Shifrin Dec 25 '22 at 01:44
  • Curioisty is more important than acceptance. – Bob Dobbs Dec 25 '22 at 09:07