Heat kernels for a family of Grushin operators (Q479664)

The Grushin operator of step \(m+1\) is the second-order differential operator \(\Delta_m\) on \(\mathbb{R}^2\) given by \[ \Delta_m = \frac{1}{2} \left(\frac{\partial}{\partial x}\right)^2 + \frac{x^{2m}}{2} \left(\frac{\partial}{\partial y}\right)^2 \] where \(m\) is a positive integer. This operator is degenerate elliptic and hypoelliptic; it can be written as \(\Delta_m = \frac{1}{2} (X_1^2 + X_2^2)\) for the vector fields \(X_1 = \frac{\partial}{\partial x}, X_2 = x^m \frac{\partial}{\partial y}\). In this paper, the authors derive a formula for the heat kernels of \(\Delta_m\) and of a family of related operators in higher dimensions, namely \[ L_1 = \frac{1}{2} \sum_{k=1}^n \left( \frac{\partial}{\partial x_k} \right)^2 + \frac{1}{2} \sum_{k=1}^n \left( x_k^{m_k} \frac{\partial}{\partial y_k} \right)^2 \] \[ L_2 = \frac{1}{2} \left( \frac{\partial}{\partial x} \right)^2 + \frac{1}{2} \sum_{k=1}^n \left( x_k^{m_k} \frac{\partial}{\partial y_k} \right)^2 \] \[ L_3 = \frac{1}{2} \sum_{k=1}^n \left( \frac{\partial}{\partial x_k} \right)^2 + \frac{1}{2} \sum_{k=1}^n \left( x_k^{m_k} \frac{\partial}{\partial y} \right)^2 \] \[ L_4 = \frac{1}{2} \sum_{k=1}^n \left( \frac{\partial}{\partial x_k} \right)^2 + \frac{1}{2} \sum_{k=1}^n \left( x_k^{m_k} \frac{\partial}{\partial y_k} \right)^2 + \frac{1}{2} \left( \frac{\partial}{\partial x_{n+1} } \right)^2. \] Most of the paper is concentrated on the two-dimensional case of the operator \(\Delta_m\); the other operators \(L_1, L_2, L_3, L_4\) are considered at the end. The derivation of the heat kernel proceeds by methods of Hamiltonian mechanics, working with the sub-Riemannian geometry induced on \(\mathbb{R}^2\) by the bracket-generating vector fields \(X_1, X_2\). The authors begin by computing the geodesics of this geometry, via solving Hamilton's equations of motion in the cotangent bundle. The classical action integral is computed by solving the Hamilton-Jacobi equations and then modified to produce a complex action function \(f\). The heat kernel then takes the form \[ \mathcal{P}_t = \frac{1}{(2 \pi t)^\alpha} \int_{\Gamma} e^{-f(x,y,x_0,\lambda)/t} V(\lambda) d\lambda \] where \(V(\lambda)\) is a volume element which is found by solving a generalized transport equation. The authors also state a result giving the small-time asymptotics of the heat kernel for \(\Delta_m\), but the proof is left to a future paper.