From: Stephen J. Herschkorn on
Is this identity well-known? For 0 <= i < k, sum(j=i..k, (-1)^j C(j,
i) C(k, j)) = 0, where C(_, _) denotes the binomial coefficient. I
needed it for something else I was working on. I've come up with a neat
little proof if anyone is interested.

--
Stephen J. Herschkorn sjherschko(a)netscape.net
From: Rob Johnson on
In article <4C5F7109.6080401(a)netscape.net>,
"Stephen J. Herschkorn" <sjherschko(a)netscape.net> wrote:
>Is this identity well-known? For 0 <= i < k, sum(j=i..k, (-1)^j C(j,
>i) C(k, j)) = 0, where C(_, _) denotes the binomial coefficient. I
>needed it for something else I was working on. I've come up with a neat
>little proof if anyone is interested.

I believe it is well-known. At least I have used it before. The
easiest proof I know uses negative binomial coefficients. To make
the steps below valid, k >= i >= 0.

k
--- j
> (-1) C(j,i) C(k,j)
---
j=i

k
--- j
= > (-1) C(j,j-i) C(k,k-j)
---
j=i

k
--- i
= > (-1) C(-i-1,j-i) C(k,k-j)
---
j=i

i
= (-1) C(k-i-1,k-i)

which is 0 for k > i, and (-1)^k for k = i.

Rob Johnson <rob(a)trash.whim.org>
take out the trash before replying
to view any ASCII art, display article in a monospaced font
From: Stephen J. Herschkorn on
Rob Johnson wrote:

>In article <4C5F7109.6080401(a)netscape.net>,
>"Stephen J. Herschkorn" <sjherschko(a)netscape.net> wrote:
>
>
>>Is this identity well-known? For 0 <= i < k, sum(j=i..k, (-1)^j C(j,
>>i) C(k, j)) = 0, where C(_, _) denotes the binomial coefficient. I
>>needed it for something else I was working on. I've come up with a neat
>>little proof if anyone is interested.
>>
>>
>
>I believe it is well-known. At least I have used it before. The
>easiest proof I know uses negative binomial coefficients. To make
>the steps below valid, k >= i >= 0.
>
> k
> --- j
> > (-1) C(j,i) C(k,j)
> ---
> j=i
>
> k
> --- j
> = > (-1) C(j,j-i) C(k,k-j)
> ---
> j=i
>
> k
> --- i
> = > (-1) C(-i-1,j-i) C(k,k-j)
> ---
> j=i
>
> i
> = (-1) C(k-i-1,k-i)
>
>which is 0 for k > i, and (-1)^k for k = i.
>
>

I am unfamiliar with negative binomial coefficients, so I have trouble
following your last two steps; Here is my proof.

sum(j=i..k, (-1)^j C(j, i) C(k, j)) = sum(j=i..k, (-1)^j k! / [i!
(j-i)! (k-j!)])
= a sum(j=i..k, (-1)^j C(k-i, j-i)) (where a is a factor depending
on i and k)
= +/- a sum(j=0..k-i, (-1)^j C(k-i, j))
= +/- a (-1 + 1)^(k-i) = 0 for k > i.

--
Stephen J. Herschkorn sjherschko(a)netscape.net