1: \begin{proof}

2: Recalling the definition~\eqref{e:e-1-h-tilde} of $\widetilde E^1_h$,

3: we see that~\eqref{e:analysis2-1} follows from

4: \begin{equation}

5:   \label{e:analysis2-1.1}

6: \|\varphi U(t_0)(1-\varphi)\varphi_{t_0}\widetilde E^1_h\|_{L^2}=\mathcal O(h^\infty).

7: \end{equation}

8: We now make the following observation: a point $(m,\nu)\in S^*M$ in

9: the wavefront set of $\widetilde E^1_h$ will make an $\mathcal

10: O(h^\infty)$ contribution to~\eqref{e:analysis2-1.1} unless $m\in\supp

11: (1-\varphi)$, but $g^{-t_0}(m,\nu)\in \supp\varphi$; however,

12: by~\as{A6} and Lemma~\ref{l:geometry-2}, in this case

13: $(m,\nu)\not\in\WFh(\widetilde E^1_h)$.  To make this argument

14: rigorous, we can write (bearing in mind that $\widetilde E^1_h$ is

15: polynomially bounded)

16: $$

17: \varphi_{t_0}\widetilde E^1_h=B \widetilde E^1_h+\mathcal O(h^\infty)_{L^2},

18: $$

19: where $B\in\Psi^{\comp}$ is compactly supported and such that

20: $$

21: (m,\nu)\in\WFh(B)\cap \supp(1-\varphi) \Longrightarrow

22: g^{-t_0}(m,\nu)\not\in\supp \varphi.

23: $$

24: Then the operator $\varphi U(t_0)(1-\varphi)B$ is $\mathcal O(h^\infty)_{L^2\to L^2}$

25: by part~2 of Proposition~\ref{l:egorov}, which proves~\eqref{e:analysis2-1.1}.

26: \end{proof}

27: