On Roth's theorem on progressions (Q640798)

\textit{Klaus Roth} [ C. R. Acad. Sci., Paris 234, 388--390 (1952; Zbl 0046.04302)] was the first to prove that a set \(A\subset [1,N]\) of integers without an arithmetic progression of length \(3\) satisfies \(|A|=o(N)\). There is a long and distinguished history of quantitative improvements: \textit{D. R. Heath-Brown} [J. Lond. Math. Soc., II. Ser. 35, 385--394 (1987; Zbl 0589.10062)]; \textit{E. Szemerédi} [Acta Math. Hung. 56, No. 1--2, 155--158 (1990; Zbl 0721.11007)]; \textit{J. Bourgain} [Geom. Funct. Anal. 9, No. 5, 968--984 (1999; Zbl 0959.11004); J. Anal. Math. 104, 155--192 (2008; Zbl 1155.11011)] and \textit{Sanders} [``On certain other sets of integers'', J. Anal. Math. 116, 53--82 (2012)]. Recently, there has also been great interest in Szemerédi's generalisation to progressions of length \(k\). Also, the famous Erdős-Turán conjecture asks whether a set \(A\) of integers with \(\sum_{a \in A} \frac{1}{a}\) being divergent must contain an arithmetic progression of length \(k\). The author proves that a set \(A\subset [1,N]\) without 3-progressions satisfies \(|A|= O(\frac{N(\log \log N)^5}{\log N})\). Quantitatively this appears to be ``close'' to the Erdős-Turán question, but the author points out that new ideas would be needed to bridge the gap. The methods involved make use of the Bohr-set technique introduced by Bourgain, and refined by the author, but makes very interesting and novel use of results of \textit{N. H. Katz} and \textit{P. Koester} [SIAM J. Discrete Math. 24, No. 4, 1684--1693 (2010; Zbl 1226.05247)], which here is compared to the Dyson \(e\)-transform, and of \textit{E. Croot} and \textit{O. Sisask} [Geom. Funct. Anal. 20, No. 6, 1367--1396 (2010; Zbl 1234.11013)]. It can be hoped for that the new ingredients lead to further progress.