Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Source. Lemma 5, printed pp. 145–146 (PDF pp. 3–4).

Statement. Let mm be a positive integer and let {bi(modmi):1≤i≤s}\{b_i\pmod{m_i}:1\le i\le s\} be a nonempty pairwise disjoint family. Suppose that, for a real B>0B>0 and an integer aa,

mi>m,m∣mi,bi≡a(modm),ω(mi)≤B.m_i>m,\qquad m\mid m_i,\qquad b_i\equiv a\pmod m,\qquad \omega(m_i)\le B.

There is a prime pp such that pm∣mipm\mid m_i for at least s/Bs/B indices. For this same prime there is an integer a′a' such that

pm∣mi,bi≡a′(modpm)pm\mid m_i,\qquad b_i\equiv a'\pmod{pm}

for at least s/(pB)s/(pB) indices. The prime pp may already divide mm.

Complete proof

Factor m1/m>1m_1/m>1 into its t≥1t\ge1 distinct prime divisors p1,…,ptp_1,\ldots,p_t. Then t≤ω(m1)≤Bt\le\omega(m_1)\le B. For i≠1i\ne1, the generalized Chinese remainder theorem says that disjointness of the two residue classes is equivalent to

gcd⁡(m1,mi)∤b1−bi.\gcd(m_1,m_i)\nmid b_1-b_i.

Both moduli are multiples of mm, while m∣b1−bim\mid b_1-b_i. Thus gcd⁡(m1,mi)≠m\gcd(m_1,m_i)\ne m, and

gcd⁡(m1/m,mi/m)>1.\gcd(m_1/m,m_i/m)>1.

So some pjp_j divides mi/mm_i/m. The assertion also holds for i=1i=1, since m1/m>1m_1/m>1. It includes the singleton-family case without a separate argument.

The tt sets of indices for these divisibilities cover all ss indices. One prime pp therefore divides mi/mm_i/m for at least s/t≥s/Bs/t\ge s/B indices. For each of them write bi=a+mcib_i=a+m c_i with integer cic_i. Among their pp residue classes ci(modp)c_i\pmod p, one occurs at least s/(pt)≥s/(pB)s/(pt)\ge s/(pB) times. If its residue is cc, put a′=a+mca'=a+mc. Those indices have the required common residue modulo pmpm.

Both pigeonhole inequalities are weak, as they must be. No squarefree assumption is used: dividing the two moduli by their common factor mm is what makes the new prime step valid even when p∣mp\mid m.