The divergence equation with \(L^{\infty}\) source (Q2149424)

The paper under review gives a strikingly elementary proof of the following classic result: there exists \(g \in L^{\infty}(\mathbb{T}^{2})\) with mean 0 for which there does not exist \(f \in W^{1,\infty}(\mathbb{T}^{2})\) satisfying \(\operatorname{div} f =g \). This new proof is substantially simpler than the two previously known ones given in [\textit{D. Preiss}, J. Reine Angew. Math. 485, 197--207 (1997; Zbl 0870.35022)] and [\textit{C. T. McMullen}, Geom. Funct. Anal. 8, No. 2, 304--314 (1998; Zbl 0941.37030)]. It builds on the approach used in [\textit{M. Wojciechowski}, C. R. Acad. Sci., Paris, Sér. I, Math. 328, No. 4, 303--306 (1999; Zbl 0935.42013)] as follows. Fixing \(N \in \mathbb{N}\), one first selects \(\{\sigma_{k} \in \{0,1\} ;\ k-1,\dots,N\}\) such that \[ \left|\sum_{k=1} ^{N} \frac{\sigma_{k}}{k} \prod_{j=1} ^{k} (1+\frac{i}{2j})\right| \geq \frac{1}{\pi} \ln N, \] One then considers the function \(g_{N} : t\mapsto \prod_{k=1} ^{N} (1+\frac{i}{k}\cos(t.a_{k}))\), and the convolution operator given by a function \(P_{A}G_{N}\) such that \[ P_{A}G_{N} \star g_{N}: t \mapsto \sum_{k=1} ^{N} \frac{i\sigma_{k}}{2k} cos(t.a_{k})\prod_{j=1} ^{k} (1+\frac{i}{2j} \cos (t.a_{j})), \] where \((a_{k})_{k\in \mathbb{N}} \in \mathbb{N}^{2}\) is a lacunary sequence constructed inductively. If \(g_{N}\) can be written as \(g_{N} = f_{0} + \partial_{1}f_{1}+ \partial_{2}f_{2}\) with \(f_{0}\), \(f_{1}\), \(f_{2}\) having Fourier coefficients decaying like \(1/|n|\), then \[ \ln N \lesssim \|P_{A}G_{N} \star f_{0}\|_{\infty} + \|P_{A}G_{N} \star \partial_{1}f_{1}\|_{\infty} +\|P_{A}G_{N} \star \partial_{2}f_{2}\|_{\infty} \lesssim 1+\|\partial_{2}f_{2}\|_{\infty}, \] which is a contradiction. This neat proof also allows the author to improve the original result by relaxing the assumption \(f_{1},f_{2} \in W^{1,\infty}\) to \(\widehat{f_{1}}\), \(\widehat{f_{2}}\) decaying at any fixed rate, and \(\partial_{2}f_{2} \in L^{\infty}\).