1: \begin{proof}

2: The argument is similar to that of Lemma~1 of~\cite{MohiBJ09}, but made a bit more rigorous. Having in mind the proof of the first part of Corollary~\ref{coro: convexity}, observe that (\ref{eq: condition}) implies that $\sbb_1$ and $\sbb_2$ have at most $n_0/2$ elements with absolute values greater than $\sigma$. Hence, $\sbb_1-\sbb_2$ has at most $n_0$ elements with absolute values greater than $2\sigma$. Let $I$ index those elements of $\sbb_1-\sbb_2$ with absolute values greater than $2\sigma$. Then $|I|\leq n_0$ and

3: \begin{equation}

4: \label{eq: pi Ic}

5: \norm{\pi_{I^c}(\sbb_1-\sbb_2)}{}^2\leq |I^c|(2\sigma)^2\leq 4m\sigma^2.

6: \end{equation}

7: From (\ref{eq: gamma formula}) and (\ref{eq: pi Ic}), we get

8: \begin{equation}

9: \norm{\pi_{I}(\sbb_1-\sbb_2)}{}^2\leq 4m\sigma^2\gamma 

10: \end{equation}

11: and

12: \begin{equation}

13: \norm{\sbb_1-\sbb_2}{}^2\leq 4m\sigma^2(1+\gamma),

14: \end{equation}

15: which yield (\ref{eq: narrowness bound}). If $\sbb_2=\sbb_0$, we can conclude that $\sbb_1-\sbb_2$ has at most $n_0$ elements with absolute values greater than $\sigma$, and hence

16: \begin{equation}

17: \norm{\sbb_1-\sbb_0}{}^2\leq m\sigma^2(1+\gamma).

18: \end{equation}

19: \end{proof}

20: