MEA Absorption Column

🧪 Overview

This is a custom Python-based model built to simulate an amine-based absorption column. The model is intended for quick, flexible simulation of post-combustion carbon capture processes.

🎯 Purpose

The goal of this model is to enable the development and further the research of Post-Combustion Carbon Capture. It supports rapid simulations using a variety of numerical methods such as:

• Shooting Method
• Finite Difference
• Collocation

All design variables and parameters are customizable, making the tool adaptable to the user's research or engineering needs.

⚙️ Status

The model is currently under active development.
✅ Functional – It can already produce consistent results with the current implementation.

The August 27 submission status is FALLBACK_ACTIVE: the manuscript retains the fixed-chemistry Henry/ePC-SAFT comparison because upstream produced a supported-negative predictive-parameter decision and no parameter set was accepted for column transfer. See docs/scientific/CONTEXT.md for the current claim boundary, immutable refusal identity, and future transfer gate.

Project organization

This repository follows the local project architecture standard for scientific Python work:

• Package code and reusable model data live under src/mea_absorption_column/.
• Fast regression tests live under tests/.
• Manuscript validation workflows live under analyses/nccc_validation/.
• Manuscript source and LaTeX build scripts live under docs/latex/.
• Root scripts/ is reserved for repo-wide tools or small smoke checks, not benchmark sweeps.

For a handoff map that tells another Codex agent which scripts run the absorber, which scripts only render or validate artifacts, and which workflows require the external ePC-SAFT package, see docs/workflow_map.md.

Usage

This repo is now uv-first for reproducible reviewer-response benchmarks.

The Overleaf mirror is a separate Git checkout configured locally. The source of truth remains docs/latex; mirror paths are intentionally machine-local and should not be committed.

The manuscript source lives in docs/latex. To refresh the flat Overleaf mirror checkout after manuscript or figure updates, run:

uv run python docs/latex/scripts/latex_workflows.py sync-overleaf --clean-build-files

Use --dry-run first when you want to preview the route without writing. The sync command excludes scripts/ and builds/ from the Overleaf projection.

To build a fresh local manuscript PDF after editing docs/latex/main.tex or included LaTeX inputs, run:

uv run python docs/latex/scripts/latex_workflows.py build

The clickable local artifact is docs/latex/builds/main.pdf. The build command also runs a freshness check and can open the PDF with --open.

Set up the project-local Python environment once from the repository root:

uv sync --group test

The local environment lives at .venv/ and is ignored by Git. The project dependency points to the read-only ePC-SAFT 0.2 wheel built with Meson from Engine commit 41e7dc1984d00f82785900b314a0135beebe56cd, wheel SHA-256 2d089d5bdeda5b1655c5e6cf5df7308507233381175745889f8430c22ffe1edb, native-core SHA-256 fc1f1a78b2fe5c68a54b8678ce6bc844a2eba11edec91d49e6f9a51c62e1132f. This runtime supplies the exact fixed-(T,P) charged-composition tangent used by the issue 16 gate; it does not reclassify the older retained Stage A tables or imply predictive-parameter adoption. The absorber adapter uses only public ePC-SAFT APIs. Model-family choices are encoded in the parameter document, and CppAD is the package's sole production derivative authority; there is no downstream derivative-backend selector. Henry-only validation can still run without evaluating ePC-SAFT. Use the project-local interpreter directly for normal checks:

uv run python -m pytest -q -p no:cacheprovider
uv run python -m mea_absorption_column.benchmark --methods single scipy-bvp --thermo-models ideal_henry

Benchmark CSV and Markdown outputs are written to analyses/nccc_validation/results/runs/benchmark by default. Run-specific files under results/runs/ are ignored by Git; curated manuscript evidence lives under analyses/nccc_validation/results/final/.

NCCC validation results

The reviewer-response benchmark evidence is organized as a self-contained analysis:

analyses/nccc_validation/
  scripts/
  results/
    runs/
    final/
      tables/
      figures/
      profiles/
      reports/

Use these commands to refresh and validate the curated tables, figures, and clean profile index without rerunning long simulations:

uv run python analyses/nccc_validation/scripts/generate_data.py
uv run python analyses/nccc_validation/scripts/render_figures.py
uv run python analyses/nccc_validation/scripts/collect_clean_profiles.py --collect-existing
uv run python analyses/nccc_validation/scripts/validate_results.py

Clean temperature-profile PNGs are arranged by case and thermodynamic lane under analyses/nccc_validation/results/final/profiles/.

See analyses/README.md, analyses/nccc_validation/README.md, and analyses/nccc_validation/analysis.yaml before adding new validation scripts or result folders.

The benchmark CLI exposes solver settings for reproducibility:

uv run python -m mea_absorption_column.benchmark --methods scipy-bvp --thermo-models ideal_henry --mesh-points 51 --tol 0.5 --bc-tol 0.001 --max-nodes 1000 --success-boundary-residual-max 1

The hand-built central-difference Jacobian is available with --finite-jacobian, but it is opt-in because it is slower than the default solver path on stiff nonlinear cases.

Shooting-method experiments can use --shooting-integrator euler|bdf|radau|rk45. Stiff IVP integrators are diagnostic only at this stage; pair them with --max-runtime-s or --subprocess-timeout-s when sweeping cases so a bad shooting branch returns a structured timeout row instead of tying up the workflow.

Thermodynamics

The default model is ideal_henry, matching the Henry-law CO2 driving-force implementation. The selected manuscript ePC-SAFT comparison is epcsaft_ionic: it retains the concentration-based chemical-equilibrium calculation and replaces the CO2 driving force with vapor- and liquid-side ePC-SAFT fugacity coefficients. The archived nine-species activity-rebased results were generated with the superseded package interface and are retained only as historical numerical-feasibility evidence.

Thermodynamic modes are intentionally explicit:

• ideal_henry: default validation baseline; concentration-based chemical equilibrium and Henry-law CO2 driving force.
• epcsaft_ionic: selected ePC-SAFT fugacity lane; concentration-based chemistry is retained while the ionic liquid state is passed to ePC-SAFT for CO2 fugacity.
• epcsaft_neutral: historical or diagnostic sensitivity lane only; not the selected manuscript comparison.
• epcsaft_reactive_*: intentionally fail closed after the 0.2 cutover. Re-enabling them requires independently sourced, dimensionless reaction constants, an explicit standard-state conversion, and migration to the typed chemical-equilibrium API.

The supported ePC-SAFT comparison in the manuscript is a controlled thermodynamic driving-force benchmark, not a claim that every absorber result uses the full activity-coupled chemistry loop. Full reactive rows must be reported with runtime and convergence diagnostics.

The MEA ePC-SAFT parameter datasets are vendored in this repository under src/mea_absorption_column/data/epcsaft_datasets/. MEA_EPCSAFT_DATASET_NAME can select a different vendored dataset, and MEA_THERMODYNAMICS_EPCSAFT_DATASET remains an override for temporary external comparisons only. Normal repo tests and absorber runs must not depend on a sibling MEA-Thermodynamics checkout for parameter files.

--epcsaft-fugacity-blend <0..1> is available for continuation diagnostics. A value of 0 returns the Henry-law fugacity values through the ePC-SAFT adapter path, intermediate values linearly blend Henry and ePC-SAFT fugacity values, and 1 is the full ePC-SAFT fugacity endpoint. This is intended for branch diagnosis and warm-start studies, not as a publishable calibrated thermodynamic model by itself.