GWTC3-rereview.md

@@ -17,6 +17,11 @@ I suggest following key:
 
 Readability may affect the length of comments in the table. Feel free to write your comments below the table if you want more space.
 
+## Remarks
+In m1 and m2 posteriors, there is sometimes a hard boundary - it is caused by the sampler forcing m1>m2, resulting in a steep cut when m1 and m2 posteriors would overlap.
+
+For posteriors with double peaks, we have tried to run the analysis with lower dlogz, forcing the sampler to work longer. We did not get any noticible change in the posterior, suggesting that the sampler converged to a posterior. Combined with the sampler showing the same features for parallel runs (same sampling, but with different random seed), we suspect that the resulting posterior accurately represents the data and is not caused by sampling errors.
+
 ## Krishnendu
 
 <table>
@@ -2318,5 +2323,7 @@ Readability may affect the length of comments in the table. Feel free to write y
 
 ## Consistency of recovered quantiles
 For some events, GR is recovered outside 90% CI. This is expected to happen by statistical flustuations. To see if this happen at expected levels, we perform pp test under hypothesis of GR being true. Distributions of quantiles consistent with the hypothesis would show as straight lines.
+
 ![pp_real_events](uploads/dc4c22b4d26ce7d89d700fece32ce140/pp_real_events.png)
+
 With the combined p-value=0.23. We see that our results are consistent with GR, with the worst match for alpha=1.5. Even For it, deviation from straight line occurs near the middle of distribution, so fraction of events outside 90% CI is still consistent.
\ No newline at end of file