On the scaling limits at zeros of solutions of parabolic equations (Q1268393)

This paper is concerned with the behavior of solutions of the linear, \(m\)-th order parabolic equation \[ \frac {\partial u}{\partial t} = \sum_{| \alpha| \leq m} a_\alpha \frac {\partial^{| \alpha| }u}{\partial x^\alpha } (x,t) \] in some open subset \(\Omega\) of \(\mathbb{R}^{n+1}\). For \(\varepsilon>0\) and \((x_0,t_0) \in \Omega\), define \(u_\varepsilon\) by \[ u_\varepsilon(x,t)=u(x_0+\varepsilon x,t_0+\varepsilon^mt). \] Two types of questions are considered here. First, if \(u(x_0,t_0)=0\), how fast does \(u_\varepsilon\) converge to zero? In connection with this question, the order of \(u\) at \((x_0,t_0)\) is defined to be the first nonnegative integer \(h\) such that \(\varepsilon^{- h}u_\varepsilon\) does not converge to zero in the Sobolev space of functions with \(m\) spatial derivatives and one time derivative in \(L^p\); if there is no such integer, the order is said to be \(\infty\). Second, how large is the set on which \(u\) vanishes with order \(h\)? Estimates on the Hausdorff dimension of some of these zero sets are given. The coefficients \(a_\alpha\) are only assumed to be in appropriate Lebesgue spaces (except when \(| \alpha| =m\), in which case \(a_\alpha\) is at least continuous), so solutions need not be defined at every point. For this reason, the correct definition of a zero point is an important consideration.