| heal.abstract |
Let ψ : N → R≥0 be an arbitrary function from the positive integers to the non-negative reals. Consider the set A of real numbers x ∈ [0, 1] for which there are infinitely many reduced fractions a/q such that |x − a/q| < ψ(q)/q. Using the first Borel-Cantelli lemma, one can easily check that if ∑ψ(q)ϕ(q)/q < ∞ then A has Lebesgue measure λ(A) = 0. The Duffin-Schaeffer conjecture asserts that if ∑ψ(q)ϕ(q)/q = ∞ then A has Lebesgue measure λ(A) = 1. In this thesis we present the work of D. Koukoulopoulos and J. Maynard who confirmed this conjecture in full generality. Corollaries of the main result are: (i) a conjecture due to Catlin regarding non-reduced solutions to the inequality |x − a/q| < ψ(q)/q, which refines Khinchin’s theorem, and (ii) a Hausdorff measure generalization of the Duffin-Schaeffer conjecture. The former was already known to follow straightforwardly from the Duffin-Schaeffer conjecture. The latter was already known to follow from the Duffin-Schaeffer conjecture via a mass transference principle of Beresnevich and Velani. A brief description of the strategy of the proof starts with a theorem of Gallagher who had established the abstract zero-one law λ(A) ∈ {0, 1} via an ergodic argument. Using this fact and a deterministic variant of the second Borel-Cantelli lemma, one can reduce the problem to showing that λ(Aq ∩ Ar) << λ(Aq)λ(Ar) in a suitably averaged sense, where Aq = ⋃ [ a − ψ(q)/ q , a + ψ(q)/ q ] , where gcd(a,q)=1 and a runs from 1 to q. This idea is known as quasi-independence on average, and in fact it was used by Duffin and Schaeffer who had settled their conjecture in certain arithmetically structured cases. Further important steps were then made by Erdős and Vaaler, who solved the case ψ(q) << 1/q. They measured the overlaps of the sets Aq and showed that if q , r then λ(Aq ∩ Ar) /λ(Aq)λ(Ar) << ∏ ( 1 + 1/p ) = F(n, t), where n = n(q, r) = qr/ gcd(q, r)^2 and t = t(q, r) = max{rψ(q), qψ(r)} /gcd(q, r) and p|n, p>t. If we do not have quasi-independence on average, then F(n, t) can be is large. Thus, the question is for which sets S , of a given density in a given discrete interval, there can be many pairs (q, r) ∈ S^2 such that F(n(q, r), t(q, r)) is large. The new idea of Koukoulopoulos and Maynard is to construct a weighted graph that encodes this information, and to use additive combinatorics to determine the structure of such a set. The final conclusion is that there must be a fixed, large number that divides a positive proportion of the elements of S . To prove this, they adapt Roth’s density increment strategy, keeping control on the density and arithmetic data of the evolving weighted graph. When a large divisor is found, it can be “factored out", and the problem more or less reduces to the case that can be treated with the methods developed by Erdős and Vaaler. |
en |