MEA absorption-column research

Explore how thermodynamic, film, transport and numerical choices affect absorber capture, axial temperature, conservation and computational cost. The manuscript revision has been submitted; new work belongs in analyses/ and Quarto notebooks.

Submitted revision

The clean remote branch codex/fallback-manuscript at eef9dce preserves the submitted document, source, supporting results and reproduction instructions. Retrieve its docs/submitted_revision/revision_submission_2026-09-04.zip for the exact PDF, letters, bibliography and source. Its checksum list is for that archive commit, not for this evolving research checkout. Do not edit manuscript prose here.

Select a research calculation

Preview options without importing the scientific runtime or running a calculation:

python3 src/mea_absorption_column/research.py --list-options
python3 src/mea_absorption_column/research.py analyses/research_options/options.toml

Edit the TOML selections, case IDs, output directory and numerical settings. Then explicitly run a new calculation in the uv environment:

uv sync --frozen --group test
uv run python -m mea_absorption_column.research analyses/research_options/options.toml --run

Existing output directories are refused. The resolved selection and benchmark settings are saved beside each new result. Presets are starting points, not scientific admission rules. An unknown or incompatible implemented option fails explicitly; numerical completion and physical agreement remain separate questions.

Formulation	Thermodynamics / film	Methods
seven_state	Henry, neutral/ionic ePC-SAFT, or nine-species reactive ePC-SAFT; explicit/implicit enhancement or a supplied frozen film linearization	single, scipy-bvp, finite
conserved	A user-supplied problem_factory builds compatible balances and boundary equations, interpreting the selected thermo/film labels	trapezoidal, central, shooting, collocation

The immutable twelve_state_conserved configuration connects trapezoidal and central methods to a bounded worker. Always supply execution.wall_limit_s for an actual run. prepare_conserved_column provides graph preparation separately. Set model.film_model to equilibrium_manifold (the default) or enhancement_reference; unsupported finite-rate choices fail before assembly. The first retained native attempt is a numerically accepted two-node trapezoidal result. An independent final-tree capability review found no remaining blocker in the bounded execution path; physical certification, refinement, and notebook review remain open. The solver connection is bounded to its retained coarse result and establishes no resolved profile or physical-accuracy claim. Reduced shooting and collocation remain explicit unavailable choices. Their existing generic research interface remains available. The earlier experiment remains under archive/coupled-solver-2026-09-08. Built-in eNRTL and MDEA column paths are not implemented by a menu label. Their future implementation can use the same explicit problem selection. See analyses/research_options/README.md and its rendered notebook for evidence, compatibility and candidate studies.

For a reproducible single case, the same preview/run command also accepts an immutable preset configuration:

preset = "seven_state_legacy"
[case]
source = "C_cases_data"
id = "3C"
[numerics]
method = "scipy-bvp"
[execution]
output_dir = "analyses/research_options/results/runs/my-3c"
process_isolation = true
wall_limit_s = 600

The attempt retains resolved SI inputs, configuration and source identities, the selected and verified Engine wheel, native profiles, raw solver stages, and every multistart candidate. Solver termination, numerical checks, observation agreement and physical certification are distinct. A changed wheel requires an explicitly prepared interpreter and its exact hash; no environment is installed automatically. The conserved preset additionally requires an identified physical-input JSON, such as analyses/bvp_solution_methods/input/case_3c.json. Estimated mobilities and provisional caloric references permit exploratory calculations, not a thermal validation claim.

Organization and ownership

• src/mea_absorption_column/: reusable column equations, adapters, methods and packaged input data.
• analyses/: study inputs, scripts, retained attempts and Quarto notebooks.
• tests/: focused behavior and independent analytic checks.
• docs/latex/: preserved submitted manuscript; no ongoing editorial work.
• ePC-SAFT-project: generic Engine equations, equilibrium and derivatives.
• MEA-Thermodynamics: parameter fitting and thermodynamic parameter adoption.

Use identified, non-editable Engine wheels. The pinned dependency is a convenient starting environment, not an assertion that all experiments must use its inputs. An intentional candidate wheel requires its exact identity and explicit dependency selection. scripts/check_epcsaft_integration.py --mode dev --self-only inspects research dependencies without enforcing the submitted wheel hash; stable/final modes retain archive identity checks. Parameter bundles validate their own input hashes, species, charges and units. Do not import a mutable sibling source tree.

Notebook and validation workflow

cd analyses/research_options
bash render.sh notebook.qmd --to html

Rendering is explicitly non-executing. Run studies separately into new output folders, then describe their observations, numerical checks, uncertainty and next questions in the notebook. Promotion is a later investigator decision; a failed or preliminary calculation may still be useful research evidence.

Focused configuration and method checks:

uv run --frozen pytest -q -p no:cacheprovider tests/test_research_options.py tests/test_research_model_dispatch.py tests/test_epcsaft_contract.py
uv run --frozen pytest -q -p no:cacheprovider tests/test_column_config.py tests/test_conserved_assembly.py
uv run python scripts/check_epcsaft_integration.py --mode final

These checks do not reproduce the manuscript results.