Read · 2 min

The well known result ∣A+B∣≥∣A∣+∣B∣−1|A+B| \ge |A|+|B|-1 in the reals is also true modulo pp (at least when pp is prime), and it is called the Cauchy Davenport theorem. Of course, this is not quite true - ∣A+B∣>p|A+B|>p is impossible, and when ∣A∣+∣B∣−1>p|A|+|B|-1>p then ∣A+B∣=p|A+B|=p. So we can write ∣A+B∣≥∣A∣+∣B∣−1|A+B| \ge |A|+|B|-1.

There are several proofs of the Cauchy Davenport theorem. One utilizes the amazing combinatorial nullstellensatz of Noga Alon.

The simplest, presented by Razvan Gelca at several MOPs, involves the ingenious use of the Dyson move. Assume ∣A∣+∣B∣≤p+1|A|+|B| \le p+1, because else just throw out elements from BB until we obtain this state as this cannot increase ∣A+B∣|A+B|. Consider the following two transformations:

  1. Take constants c1c_1 and c2c_2 (possibly 00) and change AA into A+c1A+c_1 (the set of a+c1a+c_1 for aa in AA) while BB becomes B+c2B+c_2 (the set of b+c1b+c_1 for bb in BB). This is a translation.

  2. Replace the two sets AA and BB by the union of AA and BB, CC, and the intersection of AA and BB, DD, as long as DD is nonempty. Then rename CC and DD as AA and BB in some order.

These above two steps do not increase the size of ∣A+B∣|A+B| and also do not change ∣A∣+∣B∣−1|A|+|B|-1, and also do not make any set empty. This is clearly true for step 11, and for step 22 it suffices to note that if we have x+yx+y for xx in AA intersect BB and yy in the union, then WLOG yy is in BB. Since xx is in AA, x+yx+y is in A+BA+B.

Note that applying transformation 22 strictly increases max(∣A∣,∣B∣)−min(∣A∣,∣B∣)max(|A|,|B|)-min(|A|,|B|) even while ∣A∣+∣B∣−1|A|+|B|-1 is fixed, so at some point it will become impossible to apply transformation 22 even after many translations. At this point WLOG ∣A∣≥∣B∣|A| \ge |B|.

Case 1: ∣B∣>1|B|>1. Then apply transformation 11 to WLOG that BB has 00 and some nonzero kk (and fix AA). Now, for any AA in aa, translate BB to B+aB+a while fixing AA. The element kk maps to a+ka+k which must be an element of AA, else we could successfully apply operation 22. So aa in AA implies a+ka+k in AA for some kk nonzero mod pp. Then take an element xx of AA and then xx, x+kx+k,... x+(p−1)kx+(p-1)k are all in AA. But this represents a complete residue system mod pp (because pp is prime), so AA has all elements mod pp. But then ∣A+B∣=p≥∣A∣+∣B∣−1|A+B|=p \ge |A|+|B|-1 so we're done.

Case 2: ∣B∣=1|B|=1. Since ∣A∣+∣B∣−1=r|A|+|B|-1=r originally for some RR, this is still true and ∣A∣=R+1−∣B∣=r|A|=R+1-|B|=r. Then translate BB to have 00 and we can write A=a1,a2,...arA={a_1, a_2, ... a_r}. So A+BA+B will be exactly AA and will have rr elements.

In all cases, we're done and the theorem is proved.

Read · 1 min

(2007 ISL N3) Let XX be a set of 10,000 integers, none of them is divisible by 47. Prove that there exists a 2007-element subset YY of XX such that a−b+c−d+ea -b + c - d + e is not divisible by 47 for any a,b,c,d,e∈Y.a,b,c,d,e \in Y.

Read · 1 min

Let pp be a prime and dd be a factor of p−1p-1. Prove that for every integer nn, there exist integers a1,a2,...,ada_1,a_2,...,a_d such that a1d+a2d+...+add≡nmod  p.a_1^d+a_2^d+...+a_d^d\equiv n \mod p.

Consider set A={ad mod p}A= \{ a^d \bmod p \}.

Source: AoPS

Read · 1 min

(Erdos-Ginsburg-Ziv, or EGZ) Prove that among any 2n−12n-1 integers, where nn is a positive integer, there exist nn who sum to a multiple of nn.

Read · 1 min

(2013 TSTST 5) Let pp be a prime. Prove that any complete graph with 1000p1000p vertices, whose edges are labelled with integers, has a cycle whose sum of labels is divisible by pp.

Type a concept, like “slope” or “photosynthesis”.

↑ ↓ to choose · Enter to openSearch explanations and problems