Maths Olympiad Prep

Library / /75 of 91

, 2006

Combinatorics Difficulty 7.1 National Olympiad, round 2 Prove it India

Let pp be a prime and XX be a finite set containing at least pp elements. A collection of mutually disjoint pp-element subsets of XX is called a pp-family. The empty collection itself is regarded as a pp-family. Let AA (respectively, BB) denote the number of pp-families having an even (respectively, odd) number of pp-element subsets of XX. Prove that AA and BB differ by a multiple of pp.

Solution

We observe that
A=r0, r even(nrp)(rp)!(p!)rr! A = \sum_{r \ge 0,\ r\ \text{even}} \binom{n}{rp} \frac{(rp)!}{(p!)^r r!}
and
B=r0, r odd(nrp)(rp)!(p!)rr! B = \sum_{r \ge 0,\ r\ \text{odd}} \binom{n}{rp} \frac{(rp)!}{(p!)^r r!}
The empty collection corresponds to r=0r=0 in the summation for AA. If ur=(rp)!(p!)rr!u_r = \frac{(rp)!}{(p!)^r r!}, r0r \ge 0, then we have
ur+1=(rp+1)(rp+2) u_{r+1} = \frac{(rp+1)(rp+2)\cdots}{ }
Hence we obtain
(p1)!ur+1=(rp+1)(rp+2)(rp+p1)ur. (p-1)!\, u_{r+1} = (rp+1)(rp+2)\cdots (rp+p-1) u_r.
Reading this modulo pp, we get ur+1ur(modp)u_{r+1} \equiv u_r \pmod{p}, r0r \ge 0. Using u0=1u_0 = 1, we conclude that ur1(modp)u_r \equiv 1 \pmod{p}, for all r1r \ge 1. This implies that
ABr0(1)r(nrp)(modp). A - B \equiv \sum_{r \ge 0} (-1)^r \binom{n}{rp} \pmod{p}.
Thus we need to show that pp divides M=(1)r(nrp)M = \sum (-1)^r \binom{n}{rp}. Suppose pp is odd. Taking f(x)=(1x)nf(x) = (1-x)^n, and ω\omega to be a primitive pp-th root of unity, we see that
k=0p1f(ωk)=k=0p1(1ωk)n=k=0p1j=0n(nj)(1)jωkj=j=0n(nj)(1)jk=0p1ωkj. \begin{align*} \sum_{k=0}^{p-1} f(\omega^k) &= \sum_{k=0}^{p-1} (1 - \omega^k)^n \\ &= \sum_{k=0}^{p-1} \sum_{j=0}^{n} \binom{n}{j} (-1)^j \omega^{kj} \\ &= \sum_{j=0}^{n} \binom{n}{j} (-1)^j \sum_{k=0}^{p-1} \omega^{kj}. \end{align*}
We observe that
k=0p1ωkj={1ωpj1ωj=0if p does not divide jpif p divides j. \sum_{k=0}^{p-1} \omega^{kj} = \begin{cases} \frac{1-\omega^{pj}}{1-\omega^j} = 0 & \text{if } p \text{ does not divide } j \\ p & \text{if } p \text{ divides } j. \end{cases}
Thus the above sum reduces to
p{(n0)(np)+(n2p)+}=pM. p \left\{ \binom{n}{0} - \binom{n}{p} + \binom{n}{2p} + \dots \right\} = pM.
However 1,ω,ω2,,ωp11, \omega, \omega^2, \dots, \omega^{p-1} are the roots of xp1=0x^p - 1 = 0. If we set αk=1ωk\alpha_k = 1 - \omega^k, 0kp10 \le k \le p-1, then α0,α1,,αp1\alpha_0, \alpha_1, \dots, \alpha_{p-1} are the roots of (x1)p+1=0(x-1)^p+1 = 0. If we set cj=(1)j(pj)c_j = (-1)^j \binom{p}{j} for j=1,2,,p1j = 1, 2, \dots, p-1, then we have
(xα0)(xα1)(xα2)(xαp1)=xp+c1xp1+c2xp2++cp1x+cp (x-\alpha_0)(x-\alpha_1)(x-\alpha_2)\cdots(x-\alpha_{p-1}) = x^p+c_1x^{p-1}+c_2x^{p-2}+\cdots+c_{p-1}x+c_p
where we take cp=0c_p = 0. We observe that pp divides cjc_j for 0jp10 \le j \le p-1. Taking sk=j=0p1αjks_k = \sum_{j=0}^{p-1} \alpha_j^k, use of Newton's identities show that sks_k is divisible by pp for k0k \ge 0. Moreover for npn \ge p, we also have
sn+c1sn1+c2sn2++cp1sn(p1)+cpsnp=0. s_n + c_1 s_{n-1} + c_2 s_{n-2} + \dots + c_{p-1} s_{n-(p-1)} + c_p s_{n-p} = 0.
Thus p2p^2 divides sns_n. But
sn=j=0p1αjn=j=0p1(1ωj)n=pM. s_n = \sum_{j=0}^{p-1} \alpha_j^n = \sum_{j=0}^{p-1} (1 - \omega^j)^n = pM.
It follows that pp divides MM.
If pp is even, then Mj0(n2j)2n1(mod2)M \equiv \sum_{j \ge 0} \binom{n}{2j} \equiv 2^{n-1} \pmod{2}. Since np=2n \ge p=2, we conclude that 22 divides MM.

Want a route through all this instead of an archive? The track puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement reproduced verbatim; metadata (topic, difficulty) added by this project.