Variable exponent Sobolev spaces on metric measure spaces (Q2466274)

The authors study variable exponent Sobolev type spaces on a metric measure space \(X\). To this end, they use two approaches for extension of the language of distributional derivatives to the case of metric measure spaces, expressed in terms of the \(L^{p}\)-gradient. One is known as the Hajł{a}sz approach, where the gradient \(g\) of \(f\) appears in the Hajł{a}sz inequality \[ |f(x)-f(y)|\leq d(x,y)[g(x)+g(y)] \] and another uses the notion of the upper gradient \(\rho\) of \(f\) which appears in the estimate \[ |f(x)-f(y)|\leq \int_\gamma \rho \,ds \] where \(\gamma\) is a curve connecting \(x\) and \(y\), the latter being related to the notion of the Newtonian space. The authors study both the Hasł{a}sz and Newtonian versions \(M^{1,p(\cdot)}(X)\) and \(N^{1,p(\cdot)}(X)\) of such Sobolev spaces with bounded variable exponent \(p(x)\). They show that they are Banach spaces and prove some of their properties. They show that Lipschitz continuous functions are dense in the space \(N^{1,p(\cdot)}(X)\) if \(X\) is doubling, the maximal operator is bounded in the Lebesgue space \(L^{p(\cdot)}(X)\) and \(X\) supports the \((1,1)\)-Poincaré inequality. They also prove that always \(M^{1,p(\cdot)}(\mathbb R^n)\subseteq W^{1,p(\cdot)}(\mathbb R^n)\) and \(M^{1,p(\cdot)}(\mathbb R^n)= W^{1,p(\cdot)}(\mathbb R^n)\) when the maximal operator is bounded in \(L^{p(\cdot)}(\mathbb R^n)\). A statement on the equivalence of the spaces \(M^{1,p(\cdot)}(X)\) and \(N^{1,p(\cdot)}(X)\) is given under conditions expressed also in terms of the boundedness of the maximal operator and the Poincaré inequality.