1: \begin{proof}

2: On the one hand, since PIO is a subset of SIO, from Eq.~\eqref{distillable} it follows easily that $C_{d,\SIO}(\rho) \geq C_{d,\PIO}(\rho)$. On the other hand, there is a simple PIO protocol that produces an average of $nQ(\rho)$ coherence bits starting from $n$ identical copies of $\rho$. We describe and analyse it in intuitive terms here, referring to~\cite[SM]{bound-coherence} for a more rigorous analysis. There are three main steps.

3: \begin{enumerate}[(i)]

4: \item One applies the PIO instrument with Kraus operators $\{\Pi_{I_s}\}_{s\in\pazocal{S}}$ on each of the $n$ copies of $\rho$ that are initially available.

5: \item In the limit of large $n$, each outcome $s$ is obtained an average number of times equal to $nP(s)$, where $P(s) = \Tr [\Pi_{I_s}\rho]$. 

6: \item The post-measurement state corresponding to the outcome $s$, denoted by $\widetilde{\rho}_s\coloneqq P(s)^{-1} \Pi_{I_s}\rho \Pi_{I_s}$, is pure, as follows from Lemma~\ref{G rho cliques lemma}; it is then known (\cite{Winter2016} and~\cite[Proposition~7]{Chitambar2016}) that there is a PIO protocol that extracts coherence bits at a rate $S\left(\Delta(\widetilde{\rho}_s)\right)$; since we started with $nP(s)$ states, we obtain $n P(s) S\left(\Delta(\widetilde{\rho}_s)\right)$ cosbits at the output.

7: \end{enumerate}

8: The distillation rate associated with this protocol is then

9: \begin{align*}

10: r &= \sum_{s \in \pazocal{S}} P(s)\, S\left(\Delta(\widetilde{\rho}_s)\right) \\

11: &= \sum_{s \in \pazocal{S}} P(s)\, S\left(P(s)^{-1} \Pi_{I_s} \Delta(\rho) \Pi_{I_s} \right) \\

12: &= S(\Delta(\widebar{\rho})) - S(\widebar{\rho}) \\

13: &= S(\Delta(\rho)) - S(\widebar{\rho}) \\

14: &= Q(\rho)\, ,

15: \end{align*}

16: as claimed.

17: \end{proof}

18: