[kfir4444]

What

With an ASE level (e.g., UMA) and no freq_scale_factor, ARC runs its 15-species scale-factor project. Every job finishes, but the fit then crashes:

Warning: The ESS that generated the file at .../C2H2/geometry/freq.out could not be identified.
TypeError: unsupported operand type(s) for *: 'NoneType' and 'int'

ARC copies an ASE job's output.yml to geometry/freq.out. determine_ess() recognized YAML only by its .yml extension, and YAMLParser.parse_zpe_correction() returned a ZPE only if one was stored, which ASE outputs don't do. Any parse of an ASE job's freq.out failed the same way, not just the ZPE.

Changes

• determine_ess() recognizes ARC's YAML output by its content when no ESS marker matches (is_arc_yaml_output()).
• YAMLParser.parse_zpe_correction() computes the harmonic ZPE from the frequencies, sum(h·c·ν)/2 over the real ones (imaginary ones excluded, as an ESS does). A stored zpe still takes precedence.
• scale.py:
  • A species whose ZPE cannot be parsed is recorded as unconverged instead of crashing the run.
  • An error message gets its missing f prefix.
• Hessian units in ase_script.py: it wrote the Hessian in eV/Ų for UMA and finite-difference frequencies, but in Hartree/Ų for TorchANI (torchani.units works in Hartree). It now writes Hartree/Ų for every calculator. Frequencies were in cm⁻¹ either way and are unchanged. Nothing in ARC reads this Hessian today: parser_evidence takes Hessians only from Gaussian and ORCA.

Checks

• Units: reading the stored UMA Hessians as eV/Ų reproduces the stored frequencies exactly. Read as Hartree/Ų, they would put the C–H stretch of C₂H₂ at about 18,000 cm⁻¹.
• Agreement with experiment: the ZPEs agree with the experimental values in scale.py to 0–6% for 14 of the 15 species. F₂ is the outlier at +21%.
• End to end: run on the 15 UMA jobs that had crashed, the Truhlar fit now gives, for UMA-S 1.2 (OMol task): λ_ZPE = 0.979, a harmonic frequency scale factor of 0.993.

Tests

• test_determine_ess_recognizes_arc_yaml_output_by_content: an ASE output.yml copied to freq.out is read as YAML, and an unrelated YAML mapping is not.
• test_parse_zpe_correction_from_yaml_frequencies: the ZPE is the harmonic sum, an imaginary frequency doesn't count, and a stored ZPE takes precedence.
• test_numpy_vibrational_analysis: the returned Hessian is in Hartree/Ų, checked against an ORCA Hessian.

arc/parser, ase_test.py, scale_test.py, parser_evidence_test.py and scheduler_test.py pass: 396 tests, run serially.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.12%. Comparing base (6da9d99) to head (72e29b9).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1071      +/-   ##
==========================================
+ Coverage   67.09%   67.12%   +0.02%     
==========================================
  Files         123      123              
  Lines       43675    43691      +16     
  Branches    11153    11156       +3     
==========================================
+ Hits        29303    29326      +23     
+ Misses      11223    11221       -2     
+ Partials     3149     3144       -5     

Flag	Coverage Δ
functionaltests	67.12% <ø> (+0.02%)	⬆️
unittests	67.12% <ø> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 10 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

The fit can still abort when frequency outputs are missing or all ZPE values are unavailable, causing FileNotFoundError or ZeroDivisionError instead of handling an unsuccessful fit.

2 open issues

• Handle missing frequency output as an unconverged species · New
• Guard against fitting when no ZPE values are available · New

What changed in this PR

This PR enables frequency scale-factor fitting for ASE levels by recognizing renamed YAML outputs and deriving zero-point energies from their frequencies.

Changes:

• Detect ARC YAML outputs by content and compute missing harmonic ZPEs from positive frequencies.
• Handle missing parsed ZPE values and correct the species error message.
• Standardize ASE Hessian output to Hartree/Ų and add regression tests.

File	Description
arc/​utils/​scale.py	Preserve unavailable ZPEs as None and fix error-message formatting.
arc/​parser/​parser.py	Recognize ARC YAML output regardless of extension.
arc/​parser/​parser_test.py	Test YAML detection and frequency-derived ZPEs.
arc/​parser/​adapters/​yaml.py	Compute harmonic ZPE when no stored value exists.
arc/​job/​adapters/​scripts/​ase_script.py	Convert returned Hessians to Hartree/Ų.
arc/​job/​adapters/​ase_test.py	Verify returned Hessian units against ORCA data.

_{Review effort: Balanced}