You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
we noticed that the default behaviour currently seems to be to ignore the convergence threshold for wannierisation (the input variable conv_tol), which is a little surprising because it causes the wannierisation to perform the maximum number of iterations.
This is because conv_window (the number of times the wannierisation must meet convergence criteria before being accepted as converged) is set by default to -1 in param_read().
I see that the default value for conv_window is used unless conv_noise_amp is set to some positive value or conv_window is specified explicitly. Various of the test cases do indeed explicitly specify both conv_tol and conv_window.
Isn't it a little confusing that the default behaviour is to ignore the tolerance and do a fixed number of iterations? Wouldn't it be better to set conv_window be set to a default of, say, 3 and then the behaviour is simpler?
If this is a known peculiarity/feature/issue, then of course please do close this issue directly.
Yours, with many thanks,
Jerome.
The text was updated successfully, but these errors were encountered:
I let @mostofi and @jryates comment, but (if confirmed, but I think you're right) I think that indeed having a sensible default for conv_tol and conv_window could be a good idea to take into account, and use the number of iterations just as a maximum to avoid infinite loops.
Hello Wannier Developers,
we noticed that the default behaviour currently seems to be to ignore the convergence threshold for wannierisation (the input variable conv_tol), which is a little surprising because it causes the wannierisation to perform the maximum number of iterations.
This is because conv_window (the number of times the wannierisation must meet convergence criteria before being accepted as converged) is set by default to -1 in param_read().
I see that the default value for conv_window is used unless conv_noise_amp is set to some positive value or conv_window is specified explicitly. Various of the test cases do indeed explicitly specify both conv_tol and conv_window.
Isn't it a little confusing that the default behaviour is to ignore the tolerance and do a fixed number of iterations? Wouldn't it be better to set conv_window be set to a default of, say, 3 and then the behaviour is simpler?
If this is a known peculiarity/feature/issue, then of course please do close this issue directly.
Yours, with many thanks,
Jerome.
The text was updated successfully, but these errors were encountered: