On series \(\sum c_k f(kx)\) and Khinchin's conjecture (Q466093)

Let \(\{x\}\) denote the fractional part of \(x\) and \(f:[0,1]\to[0,1]\). \textit{H. Weyl} [Rend. Circ. Mat. Palermo 30, 377--407 (1910; JFM 41.0528.02)] proved that \[ \lim_{N\to\infty}\frac{1}{N}\sum_{n=1}^N f(\{nx\})=\int_0^1f(y)\,dy\tag{1} \] holds for every continuous \(f\) and irrational \(x\in[0,1]\). \textit{A. Khintchine} [Math. Z. 18, 289--306 (1923; JFM 49.0159.03)] conjectured that (1) holds for a.e.\ \(x\in[0,1]\) and every \(f\in L^1\). \textit{J. F. Koksma} [J. Indian Math. Soc., New Ser. 15, 87--96 (1952; Zbl 0046.04802)] proved a slightly modified form (1') when the Fourier coefficients \(a_k\) of \(f\) satisfy \(\sum_{k=1}^\infty|a_k|^2\log\log k<\infty\). \textit{J. M. Marstrand} [Proc. Lond. Math. Soc., III. Ser. 21, 540--556 (1970; Zbl 0208.31402)] disproved Khintchine's conjecture for \(f\in L^\infty\). \textit{J. Bourgain} [Isr. J. Math. 63, No. 1, 79--97 (1988; Zbl 0677.60042)] constructed a new counterexample \(f\in L^\infty\) and claimed that Koksma's condition \(\sum_{k=1}^\infty|a_k|^2\log\log k<\infty\) is essentially the best possible. In the present paper, the authors prove the following modification: Theorem. Let \(w(n)\) be a function of a natural argument which is sub-multiplicative (i.e., \(w(nm)\leq w(n)w(m)\) for all \(m,n\)) and bounded in mean. Assume that \(w(n)=o(\log\log n)\). Then there exists \(f\) with Fourier coefficients \(a_k\) satisfying \(\sum_{k=1}^\infty|a_k|^2w(k)<\infty\) such that (1) is not valid. The proof is based on an adaptation of the method elaborated by Bourgain [loc.\,cit.].