An extension of Hörmander's hypoellipticity theorem (Q2256551)

In this paper, the authors prove an extension of Hörmander's well-known theorem on the hypoellipticity of a differential operator on \(\mathbb{R}^d\) given as the sum of squares of a bracket-generating family of vector fields. Here the aim is to give a version of this theorem in which the vector fields need not be smooth. Rather than studying the smoothness of arbitrary solutions of the operator, attention in this paper is restricted to non-negative solutions. As background, let us recall \textit{L. Hörmander}'s theorem [Acta.\ Math. 119, 147--171 (1967; Zbl 0156.10701)]. Let \(\Omega\) be a bounded open subset of \(\mathbb{R}^d\) and suppose \(X_0, Y_1, \dots, Y_r\) are smooth vector fields on \(\Omega\) which are bracket generating, meaning that for each \(x \in \Omega\) the Lie algebra \(\text{Lie}_x(X_0, Y_1, \dots, Y_r)\), generated by the vector fields \(X_0, Y_1, \dots, Y_r\) and their iterated Lie brackets at \(x\), is equal to \(\mathbb{R}^d\). (Intuitively speaking, the span of the vector fields \(X_0, Y_1, \dots, Y_r\) may be missing some directions at \(x\), but their Lie brackets supply the missing directions.) Let \(f \in C^\infty(\Omega)\) and consider the second-order degenerate elliptic operator \[ \mathcal{L} = f + X_0 + \sum_{j=1}^r Y_j^* Y_j. \] (A more familiar form writes \(Y_j^2\) instead of \(Y_j^* Y_j\), but they differ only by a smooth function which can be absorbed into \(f\).) Hörmander's theorem states that \(\mathcal{L}\) is hypoelliptic: if \(v\) is a distribution such that \(\mathcal{L} v \in C^\infty\), then \(v \in C^\infty\). A more precise version says that there exists \(\delta > 0\) such that if \(\mathcal{L} v\) is in some Sobolev space \(H^s\), then \(v \in H^{s+\delta}\). In the paper under review, the vector fields \(X_0, Y_1, \dots, Y_r\) and the function \(f\) are no longer assumed to be smooth; instead, it is assumed that \(f \in C^{\mathbf{S}}\), that the coefficients of \(X_0\) are \(C^{\mathbf{S}+1}\) and those of the \(Y_j\) are \(C^{\mathbf{S}+2}\) for some \(\mathbf{S} \geq 0\). (Some additional mild technical hypotheses on the regularity of the vector fields are also assumed, in certain cases.) If we drop the assumption that the vector fields \(X_0, Y_1, \dots, Y_r\) are smooth, we run into trouble defining their Lie brackets with each other, so a different version of the bracket-generating condition is needed. In this paper, it is instead assumed that the ``missing directions'' are supplied by brackets of smooth vector fields \(Y\) which are controlled by the \(Y_j\) in the following \(L^2\) sense: there should be a constant \(C\) such that for all test functions \(u \in C^\infty_0(\Omega)\), we have \(\|Yu\|_{L^2}^2 \leq C \sum_{j=1}^r \|Y_j u\|_{L^2}^2\). (There is an alternative, slightly more complicated condition to cover the case where brackets with the vector field \(X_0\) are needed.) The main theorem of this paper is that under the hypotheses outlined above, there exists \(\delta > 0\) such that for any \(s < \mathbf{S}\) and any non-negative distribution \(v\) such that \(v, \mathcal{L}v \in H^s_{\mathrm{loc}}\), we have \(v \in H^{s+\delta}_{\mathrm{loc}}\). (Note that non-negative distributions on \(\Omega\) can be identified with positive Radon measures.) As in Hörmander's work, the argument goes by obtaining subellipticity estimates for the operator \(\mathcal{L}\). An important tool is the pseudo-differential calculus of Bessel's fractional derivative operators.