wu :: forums (http://www.ocf.berkeley.edu/~wwu/cgi-bin/yabb/YaBB.cgi)
riddles >> putnam exam (pure math) >> paCpb
(Message started by: Eigenray on Jul 13th, 2004, 11:58am)

Title: paCpb
Post by Eigenray on Jul 13th, 2004, 11:58am
Prove that, if p is prime, then for any a,b
paCpb = aCb mod p2.
When is it true mod p3?

Title: Re: paCpb
Post by Barukh on Jul 16th, 2004, 1:58am

on 07/13/04 at 11:58:01, Eigenray wrote:
Prove that, if p is prime, then for any a,b
paCpb = aCb mod p2

Here’s a solution I saw in “Concrete Mathematics”. The well-known identity (called Chu-Vandermonde convolution) states that

[sum](k_1+k_2=n) rCk_1* sCk_2 = r+sCn    (1)

This generalizes to an arbitrary number of variables, and - applied to our case – may be written as:

apCbp = [sum](k_1+…+k_a=bp)  pCk_1*…* pCk_b,     (2)

where 0 [le] k_i [le] p. The key observation is that the inner product of C’s is not divisible by p2 if and only if all of k’s are divisible by p (why?). This happens only in the following combination: there are exactly b k’s equal to p, and (a-b) k’s equal to 0. But the number of such combinations is exactly aCb, and the result follows.


Quote:
When is it true mod p3?

This was not in “CM”, but the same approach works. Referring to formula (2), we see that every combination of k’s where at least 3 k’s are not divisible by p, gives a product divisible by p3. So, it remains to see what happens in a combination where exactly two k’s are not divisible by p. Without loss of generality, assume these are k_1, k_2. Then obviously k_1+k_2 = p, and the overall contribution to the sum is – compare to (1):

[sum](k_1+k_2=p) pCk_1* pCk_2 = 2pCp – 2.  

The -2 term is due to fact that k_1, k_2 cannot be 0. Now, it may be shown that for primes p [ge] 5, 2pCp [equiv] 2 (mod p3) (this (http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&selm=BwBzCJ.37G%40news.orst.edu) is an elementary proof of a similar result). So, the answer to this part is p > 3.

And now comes the most interesting part. While browsing the Web for the references of the last statement, I came across Wolstenholme’s Theorem (http://mathworld.wolfram.com/WolstenholmesTheorem.html), which shows a close relation between this problem and the one I posted several days ago! Isn’t it amazing?!

Title: Re: paCpb
Post by Eigenray on Jul 16th, 2004, 6:26am
Attached is the solution I came up with a while ago.  The similarities are remarkable.

It essentially uses both results from the problem you posted, which is why I immediately thought of it when I saw yours.



Powered by YaBB 1 Gold - SP 1.4!
Forum software copyright © 2000-2004 Yet another Bulletin Board