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. Definitions and observations in Part II, printed page 200 (PDF page 4), of Erdős (1974). This page supplies the full elementary deductions and specifies the lower bounds on the bases that are implicit in the source.

For n≥2n\ge2, let h(n)h(n) be as in remark_p199, and define

H(n)=min⁡{b≥3:∃a, 2≤a<b, gcd⁡(an−1,bn−1)=1},H(n)=\min\{b\ge3:\exists a,\ 2\le a<b,\ \gcd(a^n-1,b^n-1)=1\}, H1(n)=min⁡{k≥2:gcd⁡(kn−1,2n−1)=1}.H_1(n)=\min\{k\ge2:\gcd(k^n-1,2^n-1)=1\}.

All bases here are positive integers at least two. For n≥2n\ge2, allowing a=1a=1 would add no admissible pair, since gcd⁡(0,bn−1)=bn−1>1\gcd(0,b^n-1)=b^n-1>1 for b≥2b\ge2. H(n)H(n) asks for the existence of one coprime pair, whereas h(n)h(n) asks for a collective gcd to become one.

Statement. The minima exist and

3≤h(n)≤H(n)≤H1(n)≤2n−1.3\le h(n)\le H(n)\le H_1(n)\le2^n-1.

Moreover, the following are equivalent:

h(n)=3,H(n)=3,H1(n)=3,gcd⁡(2n−1,3n−1)=1.h(n)=3,\quad H(n)=3,\quad H_1(n)=3,\quad \gcd(2^n-1,3^n-1)=1.

Complete proof. Put A=2n−1≥3A=2^n-1\ge3. Since An−1≡−1(modA)A^n-1\equiv-1\pmod A, we have gcd⁡(An−1,A)=1\gcd(A^n-1,A)=1, so AA is admissible for H1(n)H_1(n). The base k=2k=2 is inadmissible because its gcd with itself is A>1A>1. Thus 3≤H1(n)≤A3\le H_1(n)\le A. The pair (2,H1(n))(2,H_1(n)) is admissible for H(n)H(n), proving its existence and the comparison H(n)≤H1(n)H(n)\le H_1(n).

For any admissible pair 2≤a<b2\le a<b, the collective gcd through bb divides both an−1a^n-1 and bn−1b^n-1, so it is one. Therefore h(n)≤bh(n)\le b and in particular h(n)≤H(n)h(n)\le H(n). Finally, the only pair 2≤a<b=32\le a<b=3 is (2,3)(2,3), and the collective gcd through base 33 is the gcd of that same pair. The lower bounds on all three thresholds then prove the equivalences.

Dependencies. Elementary divisibility and remark_p199. No asymptotic estimate is used.

Bears on. #770 and #820. In particular, the infinitely-often question about value three is shared by the two problems. A subexponential upper bound for the gcd, such as the BCZ theorem, does not by itself prove that the gcd equals one infinitely often.