Closed Gabrielgerez closed 1 year ago
I think there's an easy and "unsurprising" solution to this. We already have the dynamic_thrs
input parameter, so if this is set to true
we should start with the minimum precision for iter 0, if false
it should be set to world_prec
. This seems kind of obvious to me now, so I don't know why I didn't do it like this in the first place...
When running SCRF the first iteration which projects the initial guess seems to have a threshold that is too strict. Here is an example of one that does not converge after 100 iterations.
we should set precision at 1/10world_prec and threshold to 10world_prec have at least a warning for when it does not converge before max_iter.