RionMASS

RionMASS converts storage-ring revolution frequencies or times into atomic mass excesses. Python, command-line and Qt interfaces share a polynomial calibration engine and return the same reference diagnostics and result tables.

The workflow includes AME2020 references, nuclear-state and charge-state handling, custom reference masses, electronic-input covariance, leave-one-out validation and saved-run replay. Frequency extraction belongs upstream; RionMASS starts from observables and their uncertainties.

Installation

Python 3.10 or later is required.

python -m pip install RionMASS              # calibration engine
python -m pip install 'RionMASS[data,plot]'  # spreadsheets and plots
python -m pip install 'RionMASS[gui]'        # complete desktop workflow

For development, install python -m pip install -e '.[dev]' from a checkout.

Python workflow

from rionmass.api import export_result_tables, run_full_calibration_workflow
from rionmass.configuration import CalibrationConfig, ODSImportConfig

ods = ODSImportConfig("frequency_observables.ods", 0, unmeasured_list=("15N7+",), observable="frequency")
calibration = CalibrationConfig(elbien_source="nist", p=3)
result = run_full_calibration_workflow(ods, calibration)
print(result.unknown_df)
export_result_tables(result, "results")

p counts polynomial coefficients; the degree is p - 1. ODS input uses columns for ion label, centroid frequency, centroid error, width and width error, with frequencies in Hz. Supply harmonic when input frequencies have not been reduced to the fundamental. The time branch returns revolution times in ps. Isomer excitation energies are supplied separately in MeV.

python examples/quickstart.py runs an offline demonstration with public AME2020 records and the packaged NIST table. Its observables are constructed to demonstrate the interfaces; they are not experimental measurements.

Command line and desktop application

rionmass run --ods frequency_observables.ods --sheet-index 0 \
  --elbien-source nist --unmeasured 15N7+ --p 3 --outdir results
rionmass-gui

The GUI shows calibration plots, unknown masses, held-out reference predictions, AME2020 differences and separate uncertainty terms. Calibration and export run in background tasks. The selected references, order, states and electronic inputs remain inspectable.

Reference inputs and uncertainties

The GUI's Custom masses (keV) field and repeatable CLI --reference-mass options accept NUCLIDE=ME_KEV,SIGMA_KEV, a neutral-atom ground-state mass excess and positive one-standard-deviation error. Python callers use CalibrationConfig(reference_mass_overrides=...). Replacements enter the calibration and LOO comparisons; AME2020 remains a separately labelled comparator.

Explicit LOO anchors remain in the final fit with their input uncertainty. Their own holdouts are shown as diagnostics and excluded from the extra-scatter estimate. The default policy retains them in other folds' training sets. Select anchors from reference geometry or independent information before examining residuals.

Result tables separate calibration-curve, unknown-observable, LOO extra-scatter and electronic terms. The correction is cumulative ionization energy: I_q = B_neutral - B_ion. ElBiEn columns count remaining electrons; NIST total-binding rows use ion charge. Shared electronic inputs are propagated through calibration and atomic-mass recovery, retaining cancellation and cross-covariance. Missing covariance remains explicitly unavailable.

The extra-scatter estimate uses held-out predictions, not the full-fit reduced chi-square. LOO folds share references, so the marginal-pull convention is an empirical validation diagnostic. Unprovided reference or observable correlations are not inferred. See Scientific method and Electronic inputs for units and covariance assumptions.

Saved runs and replay

Save config + tables, Load config… and Replay config… create or restore DATE_RionMASS_CONFIG_suffix bundles. Completed runs freeze reference records, electronic inputs, software versions, constants and numerical checkpoints. Input spreadsheets are referenced rather than copied.

from rionmass.session import save_run_bundle, replay_run_configuration

folder, files = save_run_bundle("outputs", "label", ods, calibration, result)
report = replay_run_configuration(folder / "config.json")
print(report.mismatches, report.warnings)

An edited ODS file produces a hash warning and is still replayed with the saved sheet index. Numerical differences are reported separately. Settings-only snapshots contain no completed-result checkpoints. See Reproducibility.

Development, citation and license

pytest
ruff check .
ruff format --check .
pyright
python -m build
twine check dist/*
python scripts/check_release.py dist/*

Tests use constructed numerical fixtures and published reference inputs. See CONTRIBUTING.md, CITATION.cff, release instructions and LICENSE. The license is GPL-3.0-or-later.