Use the Gaussian check file #100
Conversation
|
@dranasinghe, try using this branch, please let me know if it behaves as expected. Thanks! |
Codecov Report
@@ Coverage Diff @@
## master #100 +/- ##
==========================================
- Coverage 40.52% 40.13% -0.39%
==========================================
Files 22 22
Lines 5419 5511 +92
Branches 1415 1451 +36
==========================================
+ Hits 2196 2212 +16
- Misses 2872 2942 +70
- Partials 351 357 +6
Continue to review full report at Codecov.
|
|
I think I haven't added |
There was a problem hiding this comment.
Thanks for these additions, @dranasinghe !
Please see the comments here
arc/job/job.py
Outdated
| @@ -150,21 +150,21 @@ def __init__(self, project, settings, species_name, xyz, job_type, level_of_theo | |||
| self.method, self.basis_set)) | |||
| self.software = 'gaussian' | |||
| elif 'b97' in self.method or 'm06-2x' in self.method or 'def2' in self.basis_set: | |||
| if not self.settings['qchem']: | |||
| raise JobError('Could not find the QChem software to run {0}/{1}'.format( | |||
| if not self.settings['gaussian']: | |||
There was a problem hiding this comment.
As discussed, let's make QChem the fallback software if Gaussian isn't found.
Also, QChem should deal with m06-2x, and Gaussian should deal with m062x (is this correct?)
Same comment for scan jobs below
arc/job/job.py
Outdated
| if self.fine: | ||
| # Note that the Acc2E argument is not available in Gaussian03 | ||
| fine = 'scf=(tight, direct) integral=(grid=ultrafine, Acc2E=12)' | ||
| fine = 'scf=(tight, direct) integral=(grid=ultrafine, Acc2E=12) guess=read' |
There was a problem hiding this comment.
fine is appended to job_type_1 when writing the input file. This will result in having both guess=mix and guess=read in the same Gaussian directive. Is this the desirable behaviour?
arc/job/job.py
Outdated
| @@ -491,7 +491,7 @@ def write_input_file(self): | |||
|
|
|||
| if self.job_type == 'sp': | |||
| if self.software == 'gaussian': | |||
| job_type_1 = 'scf=(tight, direct) integral=(grid=ultrafine, Acc2E=12)' | |||
| job_type_1 = 'scf=(tight, direct) integral=(grid=ultrafine, Acc2E=12) guess=read' | |||
There was a problem hiding this comment.
What if the optimization was done in a different software, and then we run sp in Gaussian? Will it complain if it has guess=read when the checkfile isn't available?
arc/job/job.py
Outdated
| if os.path.isfile(self.checkfile): | ||
| self._upload_check_file(local_check_file_path=self.checkfile) | ||
| if self.checkfile is not None: | ||
| if os.path.isfile(self.checkfile): |
There was a problem hiding this comment.
Please change to if self.checkfile is not None and os.path.isfile(self.checkfile):
arc/job/job.py
Outdated
| @@ -588,6 +589,7 @@ def _upload_check_file(self, local_check_file_path=None): | |||
| else local_check_file_path | |||
| if os.path.isfile(local_check_file_path) and self.software.lower() == 'gaussian': | |||
| ssh.upload_file(remote_file_path=remote_check_file_path, local_file_path=local_check_file_path) | |||
| logging.info('uploading checkpoint file\n') | |||
There was a problem hiding this comment.
Was this logging statement added on purpose, or was it left from debugging and testing this brunch?
If we choose to keep it, maybe change logging.info to logging.debug to only log this when debugging?
arc/main.py
Outdated
| @@ -84,7 +84,7 @@ def __init__(self, input_dict=None, project=None, arc_species_list=None, arc_rxn | |||
| ts_guess_level='', fine=True, generate_conformers=True, scan_rotors=True, use_bac=True, | |||
| model_chemistry='', ess_settings=None, initial_trsh=None, t_min=None, t_max=None, t_count=None, | |||
| verbose=logging.INFO, project_directory=None, max_job_time=120, allow_nonisomorphic_2d=False, | |||
| job_memory=1500): | |||
| job_memory=15000): | |||
There was a problem hiding this comment.
Please also change the default memory in Scheduler's init function.
Then, In Scheduler under troubleshoot_ess() where we increase the memory allocation, we should use a multiplication factor instead of the current memory = 3000 assignment.
We should also check that this works OK for molpro, where memory is specified in MW, and I think ARC just divides the value here by 2 for a rough estimation. 7500 MW is feasible, but might consume too much resources on the RMG server.
Currently we do in job.py:
if self.software == 'molpro':
# molpro's memory is in MW, 750 should be enough
memory /= 2
self.memory = memory
alongd
force-pushed
the
check
branch
2 times, most recently
from
000fa3e
to
1fbed08
Compare
to store the latest Gaussian check.chk file Also store conformers check files in `conformer_checkfiles`
Also upload and dowload the check file Also updated job.software determination by level of theory
If the check file doesn't exist, it will compain but continue to execute the next command
also other minor style changes
The Gaussian check file
check.chkwill now be used from previous Gaussian jobs to speed up subsequent jobs and maintain WF consistency.The most recent path of a check file from a composite / opt / conformer job is saved as a Species attribute, and passed to the next job executed for this species. If the next job uses Gaussian, the check file will be uploaded along with the input file (and will be used).