Deprecate top-level cutoff, always deduce it from bonds::rcut

[yury-lysogorskiy]

Pull Request Template

Description of Changes

The top-level cutoff of the input YAML has never reached the fit. It is read in bin/pacemaker:161 and required to be present, but it is never passed to GeneralACEFit, so the cutoff is None branch always applies and the neighbour lists are always built with max(bonds::rcut). The last consumer of the value was the --query-data path removed in 29ba519 (Nov 2023); checking every commit that ever touched bin/pacemaker, the GeneralACEFit(...) call has never carried a cutoff argument.

That leaves a mandatory parameter whose value is silently discarded: a user can set it inconsistently with bonds::rcut and never find out, while omitting it aborts the run with a ValueError.

Rather than start honouring the value — which would make an under-sized cutoff silently truncate the training neighbour lists, and an over-sized one waste memory on neighbours that contribute exactly zero — this PR drops the parameter. Deducing the cutoff from max(bonds::rcut) makes the fitting neighbour lists agree with ACEBBasisSet.cutoffmax, which the ASE and LAMMPS calculators use, so the two cannot drift apart.

• bin/pacemaker — cutoff is no longer required; a deprecation warning is logged if it is still present, so existing input files keep working unchanged.
• generalfit.py — extract get_maximal_rcut() and log the resolved cutoff, so the value actually used is always visible in log.txt. The cutoff argument of GeneralACEFit is kept as a programmatic escape hatch and documented.
• input_template.yaml / pacemaker -t — the dialog set both cutoff and bonds::rcut from a single answer; it now asks for rcut only, and the generated file has no top-level cutoff.
• examples and CLI tests — drop the redundant cutoff line. All eight files had cutoff == max(bonds::rcut) exactly, so the deduced value is unchanged in every case.
• docs — describe the deduction in inputfile.md, quickstart.md and faq.md. The inputfile.md example was itself inconsistent (cutoff: 10.0 against rcut 5 / 5.5).
• tests — new tests/test_cutoff_deduction.py covering get_maximal_rcut, including that it agrees with ACEBBasisSet.cutoffmax for per-bond cutoffs.

Runtime behaviour is unchanged for every existing input file: the deduced cutoff is exactly what was already being used.

Two CI fixes carried along

CI on this repository was red before this branch, and in a way that hid all test results, so two unrelated fixes are included to make the change verifiable. Both are self-contained and can be split out if preferred.

3bfa4df — install lib/maxvolpy in the workflow. Both jobs aborted during pytest collection with ModuleNotFoundError: No module named 'maxvolpy', so no test ran at all — pre-existing and identical on run 31264271622 (17 Aug, #103). maxvolpy is installed by the custom InstallMaxVolPyLocalPackage(install) hook in setup.py, which pip install . never invokes because it builds a wheel. The workflow now builds it explicitly; the commit message records why --no-build-isolation and scipy are each required and why numpy is deliberately not installed there.

d63466f — UnitCellFilter import in test_PyACECalculator.py. ASE moved it to ase.filters in 3.23 and has since dropped the ase.constraints alias. ase is unpinned, so the 3.10 job resolved a recent ASE without it while the 3.9 job — which recent ASE no longer supports — still got one that has it; that is why only one job failed. The test now tries the new location and falls back to the old, so both ASE generations work.

These two also unblock #100, #101, #102 and #103, which hit the same walls.

Checklist

• Code is well-documented.
• All tests have been run and passed. — run 33104565094: build (3.9) and build (3.10) both green, each covering the full pytest suite and the pacemaker CLI integration tests, which exercise the modified input.yaml files end to end.
• Relevant documentation has been updated if necessary.

License Agreement

By submitting this pull request, I agree that:

• The code submitted in this pull request will be distributed under the Academic Software License (free for academic non-commercial use, not free for commercial use), see LICENSE.md for more details.
• The copyright for the code, including the submitted code, remains with Ruhr University Bochum. Ruhr University Bochum retains the right to transfer or modify the copyright.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_017QbfG2UVKxkeuiz9GtDLo4

[yury-lysogorskiy]

CI is red on both build (3.9) and build (3.10), but the failure is not from this PR and I have not tried to fix it here.

Both jobs abort during collection, before a single test runs:

ImportError while importing test module 'tests/test_activeexploration.py'
  pyace/activelearning.py:9: from maxvolpy.maxvol import maxvol
E ModuleNotFoundError: No module named 'maxvolpy'
...
!!!!!!!!!!!!!!!!!!! Interrupted: 2 errors during collection !!!!!!!!!!!!!!!!!!!!

This is pre-existing. The Python package workflow has three runs in total and all three failed; run 31264271622 on 17 Aug, from the unrelated PR #103, fails with byte-for-byte the same two collection errors. Nothing in this PR touches activelearning, activeexploration or packaging. I did not re-run the jobs — the failure is a deterministic ModuleNotFoundError reproduced ten days apart on unrelated code, so it is not a flake. No open PR addresses it, so there was nothing to port into this branch.

Root cause, for whoever picks it up: maxvolpy is installed by the custom InstallMaxVolPyLocalPackage(install) command in setup.py:18-33, which shells out to cd lib/maxvolpy; python setup.py install. The workflow installs with pip install ., which builds a wheel and never invokes the legacy install command, so that hook does not run and maxvolpy is never installed. (python setup.py install is also gone in setuptools >= 80, so the hook would likely fail even if it were reached.)

The smallest CI-side fix is an explicit step in .github/workflows/test.yml, after Install python-ace:

    - name: Install maxvolpy
      run: |
        pip install Cython
        pip install ./lib/maxvolpy

I have deliberately left this out of this PR rather than widen it — happy to open it as a separate one if that is useful.

What this means for reviewing this PR: CI currently gives no signal on it either way. Because pytest exits during collection, the new tests/test_cutoff_deduction.py never executes, and neither does anything else. The checks in this branch will stay red until the maxvolpy install is fixed on master, independently of the changes here.

---

Generated by Claude Code

[yury-lysogorskiy]

Update after 3bfa4df (the maxvolpy CI fix): collection now succeeds and the suite runs for the first time.

job	result
build (3.9)	✅ success — full suite + CLI integration tests
build (3.10)	❌ 289 passed, 1 failed in 187s

The maxvolpy problem from my earlier comment is resolved, and the changes in this PR are exercised: tests/test_cutoff_deduction.py passes on both jobs.

The single remaining failure is unrelated to this PR and I have not fixed it here:

FAILED tests/test_PyACECalculator.py::test_relaxation
  ImportError: cannot import name 'UnitCellFilter' from 'ase.constraints'

tests/test_PyACECalculator.py:191 imports UnitCellFilter from ase.constraints; ASE moved it to ase.filters in 3.23. ase is unpinned in requirements.txt, so 3.10 resolves a version where the old location is gone, while 3.9 — which recent ASE no longer supports — still gets a version that has it. That version split is exactly why 3.9 is green and 3.10 is not, and it is also why this was invisible before: collection used to abort before the test ran.

Proposed patch, if you would like it in a separate PR:

try:
    from ase.filters import UnitCellFilter
except ImportError:  # ASE < 3.23
    from ase.constraints import UnitCellFilter

I have kept it out of this branch so the cutoff change stays reviewable on its own; the same fix would also unblock #100, #101, #102 and #103, which all hit the maxvolpy wall and would hit this one next. Happy to open it separately, or to fold it in here if you would rather see this PR fully green — just say which.

---

Generated by Claude Code