Having created this, I now realize the same issue is posted in other domains:
- Is this a bug with Cholesky?
-
Multivariate Normal with Positive Semi-Definite Covariance Matrix
In one of those, the suggestion is made thatcholeskyitself should be modified to make the requirement more obvious.
It looks like this is not yet tracked in GitHub, though. I’m trying to figure out whether to report this there, rather than here on the Discourse page.