On the \(L^p\)-differentiability of certain classes of functions (Q2449585): Difference between revisions

\loadmsbm The topic of the paper under review is the \(L^p\)-differentiability of certain convolution products. Let me be more precise. The setting is in \(\mathbb{R}^d\), where \(d\geq 2\), sets and functions are Borel measurable, and measures are defined on the Borel \(\sigma\)-algebra. The authors show that if \(K\) is a real function of class \(C^2\) on \(\mathbb{R}^d\setminus \{0\}\), \(K(\lambda x)=\lambda^{1-d}K(x)\) for every \(x\not=0\), and \(\lambda>0\), then the convolution \[ u(x)=K*\mu(x):=\int_{\mathbb{R}^d} K(x-y)\, d\mu(y) \] is \(L^p\)-differentiable for every \(p\) with \(1\leq p<d/(d-1)\) (they also give an expression for the derivative). This means that for every \(h\in \mathbb{R}^d\), the term \(u(x+h)\) can be decomposed in a polynomial \(P^1_x(h)\) with degree at most \(1\) and a remainder \(R^1_x\) as follows \[ u(x+h)=P_x^1(h)+R^1_x(h), \] where \[ \left[ -\mkern -20 mu\int_{B(\rho)} |R^1_x(h)|^p\, dx\right]^{1/p}=o(p^k). \] As applications, the authors show the \(L^p\)-differentiability for vector fields whose curl and divergence are bounded measures, and also for vector fields with bounded deformation. Another application considers the Lusin property concerning the approximation of Lipschitz functions whose (distributional) Laplace is a locally bounded measure by \(C^2\)-functions. This property is used by the same authors in [J. Eur. Math. Soc. 16, No. 2, 201--234 (2014; Zbl 1286.35006)] to prove a weak Sard property.