Correct the G4 frequency scale factor, and pair each composite with the frequency level it prescribes

[alongd]

What

Two related corrections to how ARC scales harmonic frequencies for composite methods.

1. The G4 value was wrong. data/freq_scale_factors.yml carried 0.994, derived as 0.980 * 1.014 with the 0.980 fitted by Truhlar's method over the 15-species standard set. G4's prescribed ZPE scale factor is Curtiss's 0.9854, giving 0.9992.

2. The value was being applied to the wrong frequencies. The factor stored for a composite is that protocol's own prescribed ZPE scale factor x 1.014 -- the existing notes say so:

'cbs-qb3, ...': note: '0.99 * 1.014, the 0.99 value is the ZPE scale factor of CBS-QB3'
'g4, ...':      note: '0.9854 * 1.014, the 0.9854 value is the ZPE scale factor of G4'

But default_levels_of_theory['freq_for_composite'] was a single global 'B3LYP/CBSB7' for every composite, while G4 prescribes B3LYP/6-31G(2df,p). So ARC scaled CBSB7 frequencies with a constant defined against a different basis.

Keying the factor lookup on the composite method (main.py) is therefore correct by design and is left untouched -- the defect is the frequency level handed to it. freq_for_composite becomes a {composite: level} mapping, resolved by the new get_freq_level_for_composite_method().

Behaviour change worth review

An unmapped composite now raises instead of defaulting to CBSB7. This affects the 14 composites ARC recognises that have no scale-factor entry (cbs-4m, g3, w1bd, ...). The rationale: the failure it replaces cannot be seen from the output of the job it corrupts -- a mispaired factor still converges, terminates normally, passes ARC's sanity checks, and yields a plausible enthalpy for the wrong quantity. A loud stop beats a silent guess. Setting freq_level explicitly in the input file remains the override, and the error message says so.

Happy to soften this to a warning-and-default if maintainers prefer; it is the one judgment call here.

Compatibility

A local ~/.arc/settings.py that overrides default_levels_of_theory predates the mapping and still holds a string. That is honoured with a warning naming the mispairing risk, rather than crashing the user's run.

Testing

New cases in arc/level_test.py: G4 -> B3LYP/6-31G(2df,p), CBS-QB3 -> B3LYP/CBSB7 unchanged, Level and case-insensitive input, unmapped composites raise, and the legacy-string path warns and proceeds. They patch the mapping explicitly rather than reading it live, because arc/imports.py overlays ~/.arc/settings.py over the repo defaults -- an unpatched test asserts against whatever the developer's machine happens to carry.

Note for reviewers

scan_for_composite, irc_for_composite and orbitals_for_composite have the same single-global shape and arguably the same defect -- 'scan''s own comment says it "should be the same level as freq (to project out rotors)", which no longer holds for G4 once freq moves. Deliberately left out of scope here to keep this reviewable; happy to follow up.

[Copilot]

🟢 Approval recommended

The change is a straightforward, internally consistent data correction with valid YAML structure and updated provenance.

Pull request overview

This PR updates ARC’s frequency scale-factor reference data to reflect G4’s method-defined ZPE scaling constant, improving consistency with how other composite methods (e.g., CBS-QB3) are represented in the table.

Changes:

• Updates the g4, software: gaussian factor from 0.994 to 0.9992 (i.e., 0.9854 * 1.014).
• Replaces the G4 entry’s provenance to cite the G4 paper and adds it as source: 6.
• Updates the note to reflect the method-defined ZPE scale factor basis.

File summaries

File	Description
data/freq_scale_factors.yml	Corrects the G4 scale factor value and cites the primary G4 reference in the sources table.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 66.23%. Comparing base (6520c82) to head (d3e988a).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1039      +/-   ##
==========================================
- Coverage   66.29%   66.23%   -0.07%     
==========================================
  Files         122      122              
  Lines       41825    41834       +9     
  Branches    10749    10750       +1     
==========================================
- Hits        27729    27707      -22     
- Misses      11053    11072      +19     
- Partials     3043     3055      +12     

Flag	Coverage Δ
functionaltests	66.23% <ø> (-0.07%)	⬇️
unittests	66.23% <ø> (-0.07%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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

[Copilot]

🟡 Changes recommended

Four moderate findings remain, including missing G4 regression/integration coverage and an unresolved scan-frequency mismatch.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (3)

arc/main.py:1055

• Please add an integration assertion that constructing ARC(..., composite_method='g4') produces freq_level.simple() == 'b3lyp/6-31g(2df,p)'. The new tests exercise the resolver directly, while the existing ARC-level tests cover only CBS-QB3, so a break in this call path would go undetected.

                self.freq_level = get_freq_level_for_composite_method(self.composite_method)

arc/settings/settings.py:247

• The new helper tests replace freq_for_composite with a hand-written mapping, so they never exercise this production default. A typo or omission in the G4 entry here would still leave the suite green; add an assertion against the repository defaults or an ARC(..., composite_method='g4') integration test that verifies freq_level resolves to b3lyp/6-31g(2df,p) while keeping the overlay-independent helper tests.

                            'freq_for_composite': {'cbs-qb3': 'B3LYP/CBSB7',
                                                   'cbs-qb3-paraskevas': 'B3LYP/CBSB7',
                                                   'g4': 'B3LYP/6-31G(2df,p)',  # G4's prescribed level
                                                   },

arc/settings/settings.py:246

• Adding the G4-specific frequency level leaves the composite rotor-scan default at B3LYP/CBSB7 (main.py:1064) while G4 frequencies now use B3LYP/6-31G(2df,p). A default G4 run therefore uses different levels for scans and frequencies, contrary to the project's stated scan/frequency pairing used to project out rotors. Please make the scan default per-composite as well, or otherwise explicitly handle this mismatch before making the G4 default active.

                                                   'g4': 'B3LYP/6-31G(2df,p)',  # G4's prescribed level

• Files reviewed: 6/6 changed files
• Comments generated: 1
• Review effort level: Lite