Hi Steffen,
What is your opinion about this? Is it safe to use or does it actually might "hide" problems with non-ergodicity?
it’s an additional set of updates. It cannot make the system less ergodic: rather, it allows you to explore different parts of phase space and will make the system more ergodic.
For example at higher interaction parameters for small Solver-times the up and down components are different, and such move might maybe lead to the wrong conclusion that the solver time is sufficient. Can one safely determine when this is the case?
This is always a difficult problem. If you suspect ergodicity issues, have a look at your observables starting from high T (where the simulation should be ergodic) and examine the autocorrelation times. If you make a plot of the autocorrelation times vs T you should be able to detect if there’s a point where they diverge (and you stop being ergodic). If that’s the case, you’ll need additional updates.
The autocorrelation times are automatically measured in ALPS as ‘tau’ in your observables, but you’ll need to program their output manually.
Emanuel