Wiki
Wiki

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

Updated


Source and scope. Part II, printed pages 199–200 (PDF pages 3–4), of Erdős (1974). The following complete elementary reconstruction expands the source's brief argument. The primality deduction and the coefficient proof of the equality case are supplied explicitly here.

For an integer n≥2n\ge2, define

h(n)=min⁡{M≥2:gcd⁡2≤a≤M(an−1)=1},P(n)=max⁡{p:p prime, p−1∣n}.h(n)=\min\{M\ge2:\gcd_{2\le a\le M}(a^n-1)=1\},\qquad P(n)=\max\{p:p\text{ prime},\ p-1\mid n\}.

The first set is nonempty by the collective gcd lemma. The second set contains 22 and is finite since every such prime is at most n+1n+1. Since 2n−1>12^n-1>1, we have h(n)≥3h(n)\ge3.

Statement. For every n≥2n\ge2, h(n)h(n) is prime and

P(n)≤h(n)≤n+1,h(n)=n+1 ⟺ n+1 is prime.P(n)\le h(n)\le n+1,\qquad h(n)=n+1\ \Longleftrightarrow\ n+1\text{ is prime}.

Complete proof. The upper bound follows immediately from the collective gcd lemma. The bound h(n)≥ph(n)\ge p is immediate for p=2p=2. If p≥3p\ge3 and p−1∣np-1\mid n, Fermat's theorem gives an≡1(modp)a^n\equiv1\pmod p for every 1≤a<p1\le a<p. Thus the collective gcd cannot be one before the base pp has been reached. This proves h(n)≥ph(n)\ge p and hence h(n)≥P(n)h(n)\ge P(n).

Write h=h(n)h=h(n). By minimality, the gcd for 2≤a<h2\le a<h exceeds one; fix a prime divisor qq of that gcd. If h=uvh=uv were composite, we could choose 2≤u,v<h2\le u,v<h. Then un≡vn≡1(modq)u^n\equiv v^n\equiv1\pmod q, and therefore hn≡1(modq)h^n\equiv1\pmod q. The same prime would divide every power difference through hh, contradicting the definition of hh. Thus hh is prime.

If n+1n+1 is prime, it contributes to P(n)P(n), so the two bounds already give h(n)=n+1h(n)=n+1. Conversely suppose h(n)=n+1h(n)=n+1. There is a prime qq dividing all an−1a^n-1 with 2≤a≤n2\le a\le n. Necessarily q>nq>n, since a base a=q≤na=q\le n would contradict that divisibility. Hence the nn elements 1,…,n1,\ldots,n are distinct roots of Xn−1X^n-1 in Fq\mathbb F_q, and the monic polynomials of degree nn satisfy

Xn−1=∏a=1n(X−a)in Fq[X].X^n-1=\prod_{a=1}^{n}(X-a)\quad\text{in }\mathbb F_q[X].

Comparing coefficients of Xn−1X^{n-1} gives n(n+1)/2=0n(n+1)/2=0 in Fq\mathbb F_q. Here n≥2n\ge2 and q>nq>n, so qq is odd and nn is nonzero modulo qq. It follows that q∣n+1q\mid n+1, whence q=n+1q=n+1 and n+1n+1 is prime.

Source correction. Printed page 200 defines A(n)=qk=P(n)A(n)=q_k=P(n) but then prints h(n)≥qk+1h(n)\ge q_{k+1}. That stronger inequality is false: n=2n=2 gives P(2)=h(2)=3P(2)=h(2)=3, whereas the next prime is 55. The valid Fermat bound is h(n)≥P(n)h(n)\ge P(n), as proved above. This is a compilation correction, not a published erratum.

Endpoint convention. With the displayed minimum over M≥2M\ge2, h(1)=2h(1)=2. Some formal statement files instead require M>2M>2, making their value at n=1n=1 equal to 33. The definitions agree for all n≥2n\ge2 treated here.

Dependencies. lemma_p199, Fermat's little theorem, prime divisors of integers, and elementary polynomial algebra over a field.

Bears on. #770.