4

Sum of binomial product $\displaystyle \sum^{n}_{r=0}\binom{p}{r}\binom{q}{r}\binom{n+r}{p+q}$

Simplifying $\displaystyle \frac{p!}{r!\cdot (p-r)!} \cdot \frac{q!}{r!\cdot (q-r)!}\cdot \frac{(n+r)!}{(p+q)! \cdot (n+r-p-q)!}$.

Could some help me with this, thanks

Robert Z
  • 145,942
DXT
  • 11,241

4 Answers4

7

It is convenient to use the coefficient of operator $[t^r]$ to denote the coefficient of $t^r$ in a series. This way we can write e.g. \begin{align*} \binom{p}{r}=[t^r](1+t)^p \end{align*}

The following is valid \begin{align*} \color{blue}{\sum_{r=0}^n\binom{p}{r}\binom{q}{r}\binom{n+r}{p+q}=\binom{n}{p}\binom{n}{q}} \end{align*}

We obtain \begin{align*} \color{blue}{\sum_{r=0}^n}&\color{blue}{\binom{p}{r}\binom{q}{r}\binom{n+r}{p+q}}\\ &=\sum_{r=0}^n\binom{p}{r}\binom{q}{q-r}\binom{n+r}{p+q}\\ &=\sum_{r=0}^\infty[t^r](1+t)^p[v^{q-r}](1+v)^q[w^{p+q}](1+w)^{n+r}\tag{1}\\ &=[v^q](1+v)^q[w^{p+q}](1+w)^n\sum_{r=0}^\infty(v(1+w))^r[t^r](1+t)^p\tag{2}\\ &=[v^q](1+v)^q[w^{p+q}](1+w)^n(1+v(1+w))^p\tag{3}\\ &=[v^q](1+v)^{p+q}[w^{p+q}](1+w)^n\left(1+\frac{vw}{1+v}\right)^p\tag{4}\\ &=[v^q](1+v)^{p+q}[w^{p+q}]\sum_{k=0}^{p+q}\binom{n}{k}w^k\sum_{j=0}^p\binom{p}{j}\left(\frac{v}{1+v}\right)^jw^j\tag{5}\\ &=[v^q](1+v)^{p+q}\sum_{k=0}^{p+q}\binom{n}{k}[w^{p+q-k}]\sum_{j=0}^p\binom{p}{j}\left(\frac{v}{1+v}\right)^jw^j\\ &=[v^q](1+v)^{p+q}\sum_{k=0}^{p+q}\binom{n}{k}\binom{p}{p+q-k}\frac{v^{p+q-k}}{(1+v)^{p+q-k}}\tag{6}\\ &=\sum_{k=p}^{p+q}\binom{n}{k}\binom{p}{k-q}[v^{k-p}](1+v)^k\tag{7}\\ &=\sum_{k=p}^{p+q}\binom{n}{k}\binom{p}{k-q}\binom{k}{p}\\ &=\binom{n}{p}\sum_{k=p}^{p+q}\binom{p}{k-q}\binom{n-p}{k-p}\tag{8}\\ &=\binom{n}{p}\sum_{k=0}^{q}\binom{p}{k+p-q}\binom{n-p}{k}\tag{9}\\ &=\binom{n}{p}\sum_{k=0}^{q}\binom{p}{q-k}\binom{n-p}{k}\tag{10}\\ &\color{blue}{=\binom{n}{p}\binom{n}{q}} \end{align*} and the claim follows.

Comment:

  • In (1) we apply the coefficient of operator for each binomial coefficient.

  • In (2) we use the linearity of the coefficient of operator and apply the rule $[t^{p}]t^qA(t)=[t^{p-q}]A(t)$.

  • In (3) we apply the substitution rule of the coefficient of operator with $t:=v(1+w)$ \begin{align*} A(z)=\sum_{r=0}^\infty a_rz^r=\sum_{r=0}^\infty z^r[t^r]A(t) \end{align*}

  • In (4) we factor out $(1+v)^p$.

  • In (5) we apply the binomial summation formula twice. Since we want to select the coefficient of $w^{p+q}$ we can restrict the upper limit of the left sum with $k=p+q$.

  • In (6) we select the coefficient of $w^{p+q-k}$.

  • In (7) we do some simplifications and can restrict the lower limit of the sum with $k=p$.

  • In (8) we apply the cross product $\binom{n}{k}\binom{k}{j}=\binom{n}{j}\binom{n-j}{k-j}$.

  • In (9) we shift the summation index and start from $k=0$.

  • In (10) we apply the Chu-Vandermonde identity.

Markus Scheuer
  • 108,315
  • Would you please check and explain the step going from three to four? – Marko Riedel Jan 08 '17 at 21:38
  • @MarkoRiedel: Typo corrected. Thanks! – Markus Scheuer Jan 08 '17 at 21:40
  • Nice work, verified most of it. (+1). I have remarked earlier that formal power series is preferable to Egorychev if no properties of the complex differential are used, because it does not require quite as many properties of the functions involved as does Egorychev. There are cases however where one cannot do without complex differentiation and adding residues. – Marko Riedel Jan 08 '17 at 21:51
  • @MarkoRiedel: Thanks, Marko! :-) – Markus Scheuer Jan 08 '17 at 21:59
4

This is Bizley’s identity (6.42 in Gould's book "Combinatorial identities"): $$\sum^{n}_{r=0}\binom{p}{r}\binom{q}{r}\binom{n+r}{p+q}=\binom{n}{p}\binom{n}{q}.$$ A nice combinatorial proof can be found in the following paper by Székely as a special case of a more general identity (see Corollary 3):

Common origin of cubic binomial identities. A generalization of Surányi's proof on Le Jen Shoo's formula

Robert Z
  • 145,942
3

The proposed identity can be derived (putting $r=s$) from this more general one $$ \sum\limits_{\left( {0\, \le } \right)\,k\,\left( { \le \,n} \right)} {\left( \matrix{ m - \left( {r - s} \right) \cr k \cr} \right)\left( \matrix{ n + \left( {r - s} \right) \cr n - k \cr} \right)\left( \matrix{ r + k \cr m + n \cr} \right)} = \left( \matrix{ r \cr m \cr} \right)\left( \matrix{ s \cr n \cr} \right)\quad \quad \left| \matrix{ \,{\rm 0} \le {\rm integer \, }m,n \hfill \cr \;{\rm real}\;r,s \hfill \cr} \right. $$ reported in "Concrete Mathematics" - pag. 171, and which is demonstrated through the following passages: $$ \begin{gathered} \sum\limits_{\left( {0\, \leqslant } \right)\,k\,\left( { \leqslant \,n} \right)} {\left( \begin{gathered} m - \left( {r - s} \right) \\ k \\ \end{gathered} \right)\left( \begin{gathered} n + \left( {r - s} \right) \\ n - k \\ \end{gathered} \right)\left( \begin{gathered} r + k \\ m + n \\ \end{gathered} \right)} = \hfill \\ = \sum\limits_{\begin{subarray}{l} \left( {0\, \leqslant } \right)\,k\,\left( { \leqslant \,n} \right) \\ \left( {0\, \leqslant } \right)\,j\,\left( { \leqslant \,n} \right) \end{subarray}} {\left( \begin{gathered} m - \left( {r - s} \right) \\ k \\ \end{gathered} \right)\left( \begin{gathered} n + \left( {r - s} \right) \\ n - k \\ \end{gathered} \right)\left( \begin{gathered} r \\ m + n - j \\ \end{gathered} \right)\left( \begin{gathered} k \\ j \\ \end{gathered} \right)} = \hfill \\ = \sum\limits_{\begin{subarray}{l} \left( {0\, \leqslant } \right)\,k\,\left( { \leqslant \,n} \right) \\ \left( {0\, \leqslant } \right)\,j\,\left( { \leqslant \,m + n} \right) \end{subarray}} {\left( \begin{gathered} m - \left( {r - s} \right) \\ k \\ \end{gathered} \right)\left( \begin{gathered} k \\ j \\ \end{gathered} \right)\left( \begin{gathered} n + \left( {r - s} \right) \\ n - k \\ \end{gathered} \right)\left( \begin{gathered} r \\ m + n - j \\ \end{gathered} \right)} = \hfill \\ = \sum\limits_{\begin{subarray}{l} \left( {0\, \leqslant } \right)\,k\,\left( { \leqslant \,n} \right) \\ \left( {0\, \leqslant } \right)\,j\,\left( { \leqslant \,m + n} \right) \end{subarray}} {\left( \begin{gathered} m - \left( {r - s} \right) \\ j \\ \end{gathered} \right)\left( \begin{gathered} m - \left( {r - s} \right) - j \\ k - j \\ \end{gathered} \right)\left( \begin{gathered} n + \left( {r - s} \right) \\ n - k \\ \end{gathered} \right)\left( \begin{gathered} r \\ m + n - j \\ \end{gathered} \right)} = \hfill \\ = \sum\limits_{\left( {0\, \leqslant } \right)\,j\,\left( { \leqslant \,m + n} \right)} {\left( \begin{gathered} m - \left( {r - s} \right) \\ j \\ \end{gathered} \right)\left( \begin{gathered} m + n - j \\ n - j \\ \end{gathered} \right)\left( \begin{gathered} r \\ m + n - j \\ \end{gathered} \right)} = \hfill \\ = \sum\limits_{\left( {0\, \leqslant } \right)\,j\,\left( { \leqslant \,m + n} \right)} {\left( \begin{gathered} m - \left( {r - s} \right) \\ j \\ \end{gathered} \right)\left( \begin{gathered} m + n - j \\ m \\ \end{gathered} \right)\left( \begin{gathered} r \\ m + n - j \\ \end{gathered} \right)} = \hfill \\ = \sum\limits_{\left( {0\, \leqslant } \right)\,j\,\left( { \leqslant \,m + n} \right)} {\left( \begin{gathered} m - \left( {r - s} \right) \\ j \\ \end{gathered} \right)\left( \begin{gathered} r \\ m \\ \end{gathered} \right)\left( \begin{gathered} r - m \\ n - j \\ \end{gathered} \right)} = \hfill \\ = \left( \begin{gathered} r \\ m \\ \end{gathered} \right)\left( \begin{gathered} s \\ n \\ \end{gathered} \right) \hfill \\ \end{gathered} $$ consisting in:

  • Vandermonde de-convolution
  • shift of the 4th binomial
  • "trinomial revision"
  • sum on $k$ by Vandermonde convolution
  • symmetry
  • "trinomial revision"
  • sum on $j$ by Vandermonde convolution
G Cab
  • 35,272
1

Here is a derivation based upon generalised hypergeometric functions. We obtain using rising factorials $(a)_{k}:=a(a+1)\cdots(a+k-1)$ \begin{align*} \color{blue}{\sum_{r=0}^n}&\underbrace{\color{blue}{\binom{p}{r}\binom{q}{r}\binom{n+r}{p+q}}}_{=: t_r}=\sum_{r=0}^{n}t_r=t_0\sum_{r=0}^n\prod_{j=0}^{r-1}\frac{t_{j+1}}{t_j}\\ &=\binom{n}{p+q}\sum_{r=0}^n\prod_{j=0}^{r-1}\binom{p}{j+1}\binom{q}{j+1}\binom{n+j+1}{p+q}\\ &\qquad\qquad\qquad\cdot\binom{p}{j}^{-1}\binom{q}{j}^{-1}\binom{n+j}{p+q}^{-1}\\ &=\binom{n}{p+q}\sum_{r=0}^n\prod_{j=0}^{r-1} \frac{\left(n+j+1\right)\left(p-j\right)\left(q-j\right)}{\left(j+1\right)^2\left(n-p-q+j+1\right)}\\ &=\binom{n}{p+q}\sum_{r=0}^n\frac{(n+1)_r(-p)_r(-q)_r}{(1)_r^2(n-p-q+1)_r}\\ &=\binom{n}{p+q}{_3F_2}\left(n+1,-p,-q;1,n-p-q+1;1\right)\tag{1}\\ &=\binom{n}{p+q}\frac{(-n)_q(1+p)_q}{(1)_q(-n+p)_q}\tag{2}\\ &=\binom{n}{p+q}\frac{(-1)^qn!}{(n-q)!}\,\frac{(p+q)!}{p!}\,\frac{1}{q!}\,\frac{(-1)^q(n-p-q)!}{(n-p)!}\\ &=\frac{n!}{p!(n-p)!}\,\frac{n!}{q!(n-q)!}\\ &\,\,\color{blue}{=\binom{n}{p}\binom{n}{q}} \end{align*} and the claim follows.

In (1) we write the sum as $_3F_2$ generalized hypergeometric function. This function has the nice property that the sum of the denominator parameters is one more than the sum of the numerator parameters. It is evaluated at $z=1$ and one of the numerator parameters is a negative integer. Such a series is called balanced and it obeys the Pfaff-Saalschütz identity \begin{align*} {_3F_2}\left(-n,a,b;c,1+a+b-c-n;1\right)=\frac{(c-a)_n(c-b)_n}{(c)_n(c-a-b)_n} \end{align*} which is used in (2).

Markus Scheuer
  • 108,315