The paper studies how heat spreads in spaces that have a complicated geometric structure, specifically on manifolds that have a boundary edge. The mathematical objects at the heart of this work are called "maximally subelliptic operators," which are a generalization of ordinary differential operators that arise when the directions in which you are allowed to differentiate do not span the full space in the usual sense, but still satisfy certain bracket-generating conditions (meaning you can reach all directions by combining the allowed ones and their commutators). The geometry relevant to these operators is not the usual Euclidean one but rather a more exotic "Carnot-Caratheodory" geometry, where distances are measured by how efficiently you can travel along the allowed directions.
The central achievement is a rigorous set of estimates, called Gaussian bounds, on the heat kernel associated with these operators. A heat kernel is essentially a precise description of how a concentrated source of heat at one point spreads out over time to every other point. Gaussian bounds say that this spreading is controlled by a bell-curve-shaped function governed by the relevant geometry. The authors go beyond just bounding the kernel itself and also bound all of its mixed derivatives, meaning they control how the heat kernel changes when you vary time, your starting point, or your ending point. This is technically much harder than bounding the kernel alone and provides much sharper information about the smoothness of solutions.
What makes the results especially broad is that they cover operators of any even order, not just second-order operators like the classical Laplacian, and they also handle systems of equations rather than just a single equation. The analysis works near boundary points that are "non-characteristic," meaning the boundary is not tangent to the forbidden directions, and also in the interior of the domain. The boundary conditions considered go beyond the standard Dirichlet condition of fixing values at the boundary, making the framework flexible enough to cover a wide range of physically and geometrically motivated problems.