Wiki
Wiki

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

Updated


Source. The statement and proof strategy are in the Summary and Notes of proof claim 133. This proof links the expanded elementary lemmas recorded in the same folder.

For n≥2n\ge2, define

G(n,M)=gcd⁡2≤a≤M(an−1),h(n)=min⁡{M>2:G(n,M)=1},G(n,M)=\gcd_{2\le a\le M}(a^n-1),\qquad h(n)=\min\{M>2:G(n,M)=1\},

and

P(n)=max⁡{p:p prime, p−1∣n}.P(n)=\max\{p:p\text{ prime},\ p-1\mid n\}.

The theorem below proves directly that the set defining h(n)h(n) is nonempty. Its convention agrees for n≥2n\ge2 with the canonical definition in the threshold comparison. Also write

C(M)=#{(a,b):1≤a,b≤M, gcd⁡(a,b)=1},S(n)=⌊4n⌋+1.C(M)=\#\{(a,b):1\le a,b\le M,\ \gcd(a,b)=1\},\qquad S(n)=\left\lfloor\sqrt{4n}\right\rfloor+1.

Statement. If M>2M>2, M≥P(n)M\ge P(n), and C(M)>nC(M)>n, then G(n,M)=1G(n,M)=1 and therefore h(n)≤Mh(n)\le M. Consequently

P(n)≤h(n)≤max⁡{P(n),S(n)}.P(n)\le h(n)\le\max\{P(n),S(n)\}.

In addition:

  1. if nn is odd, then h(n)≤S(n)h(n)\le S(n);
  2. if P(n)>2nP(n)>2\sqrt n, then h(n)=P(n)h(n)=P(n);
  3. if p=P(n)>2p=P(n)>2 and C(p)>nC(p)>n, then h(n)=ph(n)=p.

Complete proof of the criterion. Suppose instead that G(n,M)>1G(n,M)>1, and let qq be one of its prime divisors. By the prime-divisor subgroup lemma, q>Mq>M and the residues 1,…,M1,\ldots,M lie in

H={x∈Fq∗:xn=1},∣H∣≤n.H=\{x\in\mathbf F_q^*:x^n=1\},\qquad |H|\le n.

If q>M2q>M^2, the reduced-fraction injection gives C(M)≤∣H∣≤nC(M)\le|H|\le n, contrary to the hypothesis.

It remains to exclude M<q≤M2M<q\le M^2. Suppose first that nn is even. For an arbitrary x∈Fq∗x\in\mathbf F_q^*, the signed pigeonhole lemma gives

x=ak−1,1≤k≤M,0<∣a∣<M.x=ak^{-1},\qquad 1\le k\le M,\qquad 0<|a|<M.

Both kk and ∣a∣|a| lie in HH. Evenness of nn gives an=∣a∣n=1a^n=|a|^n=1, so subgroup closure yields x∈Hx\in H. Hence H=Fq∗H=\mathbf F_q^*. Cyclicity of the latter group now gives q−1∣nq-1\mid n. Thus qq is among the primes defining P(n)P(n), and q≤P(n)≤Mq\le P(n)\le M, contradicting q>Mq>M.

Now suppose that nn is odd. Every x∈Hx\in H is a square, since

x=xn+1=(x(n+1)/2)2.x=x^{n+1}=\left(x^{(n+1)/2}\right)^2.

The prime qq is odd because q>M≥P(n)≥2q>M\ge P(n)\ge2. By the least-nonresidue lemma, its least positive quadratic nonresidue rr satisfies r<q+1≤M+1r<\sqrt q+1\le M+1. Hence the integer rr is at most MM. But every residue 1,…,M1,\ldots,M belongs to HH and is therefore a square, a contradiction. Both parities are impossible, so G(n,M)=1G(n,M)=1.

The lower bound. The prime 22 belongs to the set defining P(n)P(n), and h(n)>2h(n)>2. If p≥3p\ge3 and p−1∣np-1\mid n, Fermat's little theorem gives an≡1(modp)a^n\equiv1\pmod p for every 1≤a<p1\le a<p. Thus every prefix ending before pp has gcd divisible by pp, so h(n)≥ph(n)\ge p. Taking the maximum gives h(n)≥P(n)h(n)\ge P(n).

The displayed consequences. The coprime-pair estimate gives

C(M)≥M24+M.C(M)\ge\frac{M^2}{4}+M.

Because S(n)>2nS(n)>2\sqrt n, choosing M=max⁡{P(n),S(n)}M=\max\{P(n),S(n)\} makes M≥P(n)M\ge P(n) and C(M)>nC(M)>n. The criterion and the lower bound prove the main display.

If nn is odd, no odd prime pp can have p−1∣np-1\mid n, so P(n)=2P(n)=2; since S(n)≥3S(n)\ge3, the main display reduces to h(n)≤S(n)h(n)\le S(n). If P(n)>2nP(n)>2\sqrt n, integrality gives

P(n)≥⌊2n⌋+1=S(n),P(n)\ge\left\lfloor2\sqrt n\right\rfloor+1=S(n),

so the main display gives h(n)=P(n)h(n)=P(n). Finally, if p=P(n)>2p=P(n)>2, then p−1∣np-1\mid n makes nn even. Taking M=pM=p in the criterion and using C(p)>nC(p)>n gives h(n)≤ph(n)\le p, while the lower bound gives equality.

Dependencies. The five linked complete elementary lemmas, Fermat's little theorem, and cyclicity of the multiplicative group of a finite field.

Scope. This is a partial result for Problem 770. It proves the third question for each fixed ϵ>1/2\epsilon>1/2 and all sufficiently large nn, because then nϵ>2nn^\epsilon>2\sqrt n. It does not cover ϵ=1/2\epsilon=1/2 or smaller positive ϵ\epsilon, and says nothing that resolves the density or limit-inferior questions. The source's formal-verification description is not certified by this ordinary proof reconstruction.

Bears on. #770.