"Largest R-hat" warning does not match actual maximum rhat

Interfaces RStan
,
1

I am in the (hopefully) final stages of the Bayesian workflow process. I get the following warnings after running my model (~16k parameters, 10 chains, 10 cores, adapt_delta = 0.99, 4000 iters of which 2000 are warmup):

Warning messages:
1: The largest R-hat is 1.11, indicating chains have not mixed.
Running the chains for more iterations may help. See
https://mc-stan.org/misc/warnings.html#r-hat
2: Bulk Effective Samples Size (ESS) is too low, indicating posterior means and medians may be unreliable.
Running the chains for more iterations may help. See
https://mc-stan.org/misc/warnings.html#bulk-ess
3: Tail Effective Samples Size (ESS) is too low, indicating posterior variances and tail quantiles may be unreliable.
Running the chains for more iterations may help. See
https://mc-stan.org/misc/warnings.html#tail-ess
4: The ESS has been capped to avoid unstable estimates.

However, after I get my stanfit object fit_, I run the following code:

library(rstan)
summ <- summary(fit_)
print(max(summ$summary[, "Rhat"])) # [1] 1.006459

Why does the warning message’s maximum rhat not match the actual one?

2

Which version of RStan you are using? As Rhat computation was changed at some point, it is possible that warnings are showing the results from the new Rhat (as the warnings are also showing new Bulk and Tail ESS), but $summary([,“Rhat”])) might still show the old version of Rhat. Old Rhat can be optimistic for thick tailed distribution as discussed in Rank-Normalization, Folding, and Localization: An Improved Rˆ for Assessing Convergence of MCMC (with Discussion)