Bug fixes #2
Bug fixes #2
Conversation
MPMPMPMPMPMPMP
force-pushed
the
bug_fixes
branch
from
20aa1a5 to
5b8fa15
Compare
varipeps/optimization/optimizer.py
Outdated
| varipeps_global_state.ctmrg_projector_method = None | ||
|
|
||
| if random_noise_retries == 0: | ||
| if random_noise_retries == varipeps_config.optimizer_random_noise_max_retries: |
There was a problem hiding this comment.
That is not a bug but intended. The other case is handled by another else statement
There was a problem hiding this comment.
But doesn't the code directly terminate if the first CTMRG of the optimizer doesn't converge? Intuitively, I would have thought that it retries with another random state or should one do that in the actual program which runs the optimizer.
Check the success return value for true ?
if not, try with another unitcell?
There was a problem hiding this comment.
Yes, exactly, the code terminates if the first CTMRG/gradient does not converge. Since the first unit cell is user-supplied I do not want to make any assumption about it. My design idea was that the user can iterate over the result of the optimizer (as success=False is set) and can decide if they wants to restart or terminate the program. An example, where this could be sensible, is if the start unit cell was designed in some specific way and the non-convergence of the CTMRG has some meaning (and not only numerical reasons).
MPMPMPMPMPMPMP
force-pushed
the
bug_fixes
branch
from
bb21a6b to
bd31361
Compare
|
I included all bug fixes on the |
2 participants
I fixed typos.
For the svd I introduced a check when on GPU the argument "only_u_or_vh" gets ignored since this is not implemented.
if random_noise_retries == varipeps_config.optimizer_random_noise_max_retries
Because it would stop directly if the CTMRG didn't converge.