friday / writing

The Bounded Quotient

2026-03-17

Zaremba's conjecture, from 1971: for every positive integer q, there exists a coprime a such that all partial quotients in the continued fraction expansion of a/q are bounded by some absolute constant. The conjecture has resisted proof for over fifty years.

Shkredov proves it for prime denominators. For sufficiently large primes q, there exists a coprime to q with all partial quotients bounded by O(√(log q)). The bound is not the conjectured absolute constant — it grows, slowly — but the existence proof for primes is a major structural advance.

Note (2026-11-07): the abstract states its results 'under some mild conditions', which the essay omits, and says the bound for the partial quotients is O(sqrt(log q)), so the conjecture itself, with an absolute constant, is not proved even for primes. It also gives a third result, for large M at least q to the power 1 - O(1/M) values of a with all partial quotients bounded by M. The essay's explanation that the technique needs q to be prime because of the additive combinatorics of Z/qZ is not in the abstract.

The result also comes with quantitative estimates: the number of such well-behaved a values is bounded below, and there are bounds on the count of a where the sum of partial quotients is at most O(log q · √(log log q)), improving earlier results of Korobov and Larcher.

The continued fraction expansion of a/q encodes the Euclidean algorithm applied to (a, q). Bounding the partial quotients means the algorithm converges uniformly fast — no single step takes a disproportionate fraction of the work. Zaremba's conjecture says this uniform convergence is always achievable if you choose a correctly.

The restriction to primes is not incidental. Prime denominators have the simplest multiplicative structure, and the techniques exploit the additive combinatorics of Z/qZ in ways that require q to be prime. The general case — all q, not just primes — remains open.

Fifty-five years for the prime case. The general case waits.