[alongd]

Summary

Fixes found while running two reactions end to end (n-C3H7OO 1,5-H shift; CH3 + C2H6 H-abstraction) for the ARC methods paper. One commit per fix:

statmech/arkane.py     discard k(T) when Arkane's own conformer has a mode < 5 cm^-1
plotter.py             "No rotors considered" note decided from rotors_dict, not a substring
scheduler.py  restart  skip IRC-endpoint rotors
                       spawn never-scanned rotors (success None, no scan_path)
                       run per-rotor scans when pipe_planner does not exist yet
                       drop superseded conf_opt names from running_jobs
                       skip restored jobs whose species are not in the project
                       skip running-job names missing from job_dict when saving
job/ssh.py             PBS "Maximum number of jobs already in queue" handled like Slurm's
                       AssocMaxSubmitJobLimit: one retry, return its result
                       unparsable job ID from a known scheduler -> errored job, not ValueError
autotst                script exits non-zero with a traceback on failure
                       subprocess runs without the caller's PYTHONPATH
devtools/install_autotst.sh  register AutoTST via autotst.pth in tst_env, not ~/.bashrc

The Arkane fix in one line: a TS whose 21 cm⁻¹ imaginary torsion reached Arkane as 0.0 cm⁻¹ produced a harmonic partition function of ~1e9 and a rate coefficient 18× above the collision limit, which ARC wrote to the library as converged. The check reads Arkane's output, not ARC's input, because Arkane re-derives frequencies from the Hessian when rotors are present.

 run_arkane
   parse output.py -> rxn.kinetics
+  if any conformer frequency in output.py < MIN_PHYSICAL_FREQ (5 cm^-1)
+    log the frequencies, note them in the TS description
+    rxn.kinetics = None

The PYTHONPATH strip keeps a source RMG-Py checkout on the caller's PYTHONPATH from shadowing tst_env's rmgpy. With it, the installer's default mode (a PYTHONPATH line appended to ~/.bashrc) would leave AutoTST unimportable, so the installer now writes autotst.pth into tst_env's site-packages after the env is created, in both modes. --conda hooks still survive: conda run fires activate.d after the strip (arc/job/env_run.py).

Evidence

Every new test fails on official/main and passes here.

Case	Before (official/main)	After
TS conformer HarmonicOscillator([0.0, 3.2, …])	kinetics kept, A = 5e20 s⁻¹	rxn.kinetics is None, reason in TS description
Same reaction, all modes ≥ 85 cm⁻¹ (control)	kept	kept
TS with rotors_dict = {0: {'success': True}}	"No rotors considered for this TS."	note absent
Restart with an IRC-endpoint rotor	a scan spawned per IRC rotor	none
Restart with a hand-added rotor	never run	run
Restart with a rotor to spawn	AttributeError: ... no attribute 'pipe_planner'	per-rotor scan
Restart restoring a job of rxn_label_TS	KeyError	skipped with a warning
Saving with an untracked conformer job	KeyError: 5	skipped with a warning
PBS queue full, then 123456.pbs-server	ValueError: Unrecognized cluster software: PBS	('running', '123456')
PBS queue full twice	ValueError	('errored', 0) after one sleep
Unknown cluster software LSF	ValueError	ValueError (unchanged)
AutoTST raises inside the script	exit 0, empty stderr	exit 1, traceback on stderr
AutoTST subprocess env	caller's PYTHONPATH passed	strip_pythonpath=True

pytest -n 0 arc/statmech/arkane_test.py arc/plotter_test.py arc/scheduler_test.py \
       arc/job/ssh_test.py arc/job/adapters/ts/autotst_test.py \
       arc/job/adapters/scripts/autotst_script_test.py
337 passed, 15 subtests passed

bash -n and shellcheck pass on the installer. write_pth_file was run against a temporary directory, and the site-packages lookup was run against a real tst_env. The full installer was not run end to end.

Under the default xdist run, arkane_test.py and plotter_test.py fail intermittently on official/main as well, because tests share directories. This PR does not address that.

Merge Danger

Door: two-way

Each fix is its own commit and can be reverted alone.

Blast Radius: moderate

• A reaction whose TS or species has a mode below 5 cm⁻¹ now gets no rate coefficient in the library instead of a wrong one. Users who relied on that number will see it disappear, along with the reason.
• If a cluster's submission reply is not recognised, ARC now marks that one job as errored instead of stopping the whole run.
• Existing AutoTST installs made in the default path mode need make install-autotst rerun to get autotst.pth. Until then AutoTST guesses fail loudly rather than silently. --conda installs are unaffected.

[Copilot]

🟡 Changes recommended

Critical restart-handling defects leave invalid kinetics available on disk and orphaned job entries that can cause crashes or infinite scheduling loops.

3 open issues

• Removed restart species leaves empty running-jobs entry · New
• Orphaned restart job remains in live running-jobs state · New
• Stale kinetics library survives when restart rejects all rates · New

What changed in this PR

This PR prevents ARC from exporting rate coefficients derived from near-zero vibrational frequencies and improves restart, cluster submission, and AutoTST reliability.

Changes:

• Reject suspect Arkane kinetics and derive TS rotor notes from rotor data.
• Handle incomplete restart state and retry queue-limit submission failures once.
• Isolate AutoTST’s Python environment, report subprocess failures, and register its installation through a .pth file.

File	Description
docs/​source/​TS_search.rst	Document AutoTST registration and environment isolation.
devtools/​install_autotst.sh	Register AutoTST through .pth; retain optional activation hooks.
arc/​statmech/​arkane.py	Reject kinetics involving near-zero conformer frequencies.
arc/​statmech/​arkane_test.py	Test frequency rejection and valid-frequency controls.
arc/​scheduler.py	Improve restart scans and stale-job handling.
arc/​scheduler_test.py	Add restart and job-tracking regression tests.
arc/​plotter.py	Determine TS rotor notes from successful rotor data.
arc/​plotter_test.py	Test rotor descriptions in generated libraries.
arc/​job/​ssh.py	Bound queue-limit retries and handle unrecognized submission replies.
arc/​job/​ssh_test.py	Test submission failures and retry results.
arc/​job/​adapters/​ts/​autotst_ts.py	Strip caller PYTHONPATH from AutoTST subprocesses.
arc/​job/​adapters/​ts/​autotst_test.py	Verify AutoTST environment-isolation requests.
arc/​job/​adapters/​scripts/​autotst_script.py	Report initialization failures with a traceback and nonzero exit.
arc/​job/​adapters/​scripts/​autotst_script_test.py	Test subprocess failure reporting with an AutoTST stub.

_{Review effort: Balanced}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.26%. Comparing base (6da9d99) to head (5e760aa).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1069      +/-   ##
==========================================
+ Coverage   67.15%   67.26%   +0.10%     
==========================================
  Files         123      123              
  Lines       43675    43768      +93     
  Branches    11153    11170      +17     
==========================================
+ Hits        29332    29442     +110     
+ Misses      11203    11169      -34     
- Partials     3140     3157      +17     

Flag	Coverage Δ
functionaltests	67.26% <ø> (+0.10%)	⬆️
unittests	67.26% <ø> (+0.10%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 14 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

🟡 Changes recommended

Repeated saves can duplicate or retain stale rotor notes, while saved convergence causes newly added, unscanned rotors to be skipped.

2 open issues

• Prevent duplicate generated notes across repeated saves · New
• Scan pending rotors despite saved convergence state · New

3 resolved since last review

• Stale kinetics library survives when restart rejects all rates
• Orphaned restart job remains in live running-jobs state
• Removed restart species leaves empty running-jobs entry

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

The restart logic can misclassify queued or running pipe rotors as unstarted and submit duplicate standalone scans before existing pipe runs are restored and reconciled.

1 open issue

• Restart submits duplicate scans for rotors with pending pipe tasks · New

2 resolved since last review

• Scan pending rotors despite saved convergence state
• Prevent duplicate generated notes across repeated saves

_{Review effort: Balanced}