DC calculus (Q1745319)

In a preprint, G. Perel'man introduced the DC calculus. A function \(f:\mathbb R^n\to\mathbb R\) is called semiconcave (SC) if it is locally representable as the difference of a concave function and a smooth function. The set of SC functions is closed with respect to addition, multiplication by positive numbers, and taking minimum. A function \(f:\Omega\subset\mathbb R^n\to\mathbb R\) is said to be a DC function on an open subset \(\Omega\) if it is locally representable as a difference of two semiconcave functions. The term DC stands for the difference of concave functions. If \(f:U\subset\mathbb R^m\to V\subset\mathbb R^n\) and \(g:V\to\mathbb R\) are functions with two open subsets \(U\) and \(V\), then the function \(g\circ f\) is DC on \(U\) if \(f\) and \(g\) are DC on \(U\) and \(V\) respectively. As a consequence, the set of real-valued DC functions on \(U\) is an algebra. In this paper, the authors extend the DC calculus introduced by Perelman on finite dimensional Alexandrov spaces with curvature bounded below. Among other things, their results allow to define the Hessian and the Laplacian of DC functions, including distance functions as a particular instance, as a measure-valued tensor and a Radon measure respectively. It is shown that these objects share various properties with their analogues on smooth Riemannian manifolds. First, the authors introduce the notion of DC\(_0\) Riemannian manifold, which provides a natural setting where the DC calculus can be defined. Next, they study the covariant tensors and their covariant derivatives on a DC\(_0\) Riemannian manifold. Due to the low regularity, the authors use the old-fashioned approach consisting in defining tensors in charts and imposing a compatibility condition. Next part of the paper is devoted to the Laplacian and Hessian of a DC function. If \(X\) is a DC\(_0\) Riemannian manifold and \(f:\Omega\subset X\to\mathbb R\) is DC function, then the Hessian of f is defined as the covariant 2-tensor \(\text{Hess}f=Ddf\), where \(D\) stands for the differential of function or the measure-valued covariant derivative of vector field, depending on context. For vector fields \(X\), \(Y\), the Hessian of \(f\) is \(\text{Hess}f(X,Y)=D(df(Y))(X)-df(D_XY)\). Also, the classical formula \(D_{\nabla\psi}\nabla_\psi=\frac12\nabla(|\nabla\psi|^2_g)\) is proven. The Laplacian \(\Delta_\varphi f\) of \(f\) is defined as the trace of \(\text{Hess}_\varphi f\) with respect to \(g\) in a chart \((U,\varphi)\) of \(X\). The Hessian is then used to define an integration by parts formula. This consistency result is particularly relevant in connection with the coordinate-free approach to calculus in metric measure spaces, which works particularly well under Ricci lower bounds on curvature. Finally, the authors prove that all the results they studied in the paper apply to an open subset of a finite dimensional Alexandrov space \((X,d)\) with curvature bounded from below. They consider regular sets of an Alexandrov space contained in a DC Riemannian manifold, and almost everywhere second order Taylor expansion for DC functions.