文摘
We examine the validity of the Poincaré inequality for degenerate, second-order, elliptic operators H in divergence form on \({L_2(\mathbf{R}^{n}\times \mathbf{R}^{m})}\) . We assume the coefficients are real symmetric and \({a_1H_\delta\geq H\geq a_2H_\delta}\) for some \({a_1,a_2>0}\) where H δ is a generalized Gru?in operator, $$H_\delta=-\nabla_{x_1}\,|x_1|^{\left(2\delta_1,2\delta_1'\right)} \,\nabla_{x_1}-|x_1|^{\left(2\delta_2,2\delta_2'\right)} \,\nabla_{x_2}^2.$$ Here \({x_1 \in \mathbf{R}^n,\; x_2 \in \mathbf{R}^m,\;\delta_1,\delta_1'\in[0,1\rangle,\;\delta_2,\delta_2'\geq0}\) and \({|x_1|^{\left(2\delta,2\delta'\right)}=|x_1|^{2\delta}}\) if \({|x_1|\leq 1}\) and \({|x_1|^{\left(2\delta,2\delta'\right)}=|x_1|^{2\delta'}}\) if \({|x_1|\geq 1}\) . We prove that the Poincaré inequality, formulated in terms of the geometry corresponding to the control distance of H, is valid if n??2, or if n?=?1 and \({\delta_1\vee\delta_1'\in[0,1/2\rangle}\) but it fails if n?=?1 and \({\delta_1\vee\delta_1'\in[1/2,1\rangle}\) . The failure is caused by the leading term. If \({\delta_1\in[1/2, 1\rangle}\) , it is an effect of the local degeneracy \({|x_1|^{2\delta_1}}\) , but if \({\delta_1\in[0, 1/2\rangle}\) and \({\delta_1'\in [1/2,1\rangle}\) , it is an effect of the growth at infinity of \({|x_1|^{2\delta_1'}}\) . If n?=?1 and \({\delta_1\in[1/2, 1\rangle}\) , then the semigroup S generated by the Friedrichs-extension of H is not ergodic. The subspaces \({x_1\geq 0}\) and \({x_1\leq 0}\) are S-invariant, and the Poincaré inequality is valid on each of these subspaces. If, however, \({n=1,\; \delta_1\in[0, 1/2\rangle}\) and \({\delta_1'\in [1/2,1\rangle}\) , then the semigroup S is ergodic, but the Poincaré inequality is only valid locally. Finally, we discuss the implication of these results for the Gaussian and non-Gaussian behaviour of the semigroup S.