Computer-Aided Molecular Design for Extractive Distillation using the Infinitely Sharp Split (ISS) Method
Extractive distillation separates azeotropic and close-boiling mixtures by adding an entrainer that modifies relative volatilities. Classical entrainer selection criteria — selectivity and capacity at infinite dilution — ignore the recovery column and can miss feasibility constraints.
CAMD-ISS addresses this by coupling molecular design with the Infinitely Sharp Split (ISS) method, which simultaneously accounts for both the extractive and recovery columns. The framework:
- Designs new entrainer candidates via MILP-based molecular generation (UNIFAC group contributions)
- Screens predefined entrainer sets against thermodynamic and structural constraints
- Evaluates candidates using the ISS objective (minimum entrainer-to-distillate ratio E/D), computed via Pyomo/IPOPT
- Performs miscibility screening, bubble-point VLE, and azeotrope detection along the way
# 1. Install IPOPT and dependencies
conda install -c conda-forge ipopt
pip install -e .
# 2. Run screening on a real system (no Gurobi required)
python iss.py config/IECR/screen_acetone_methanol.yaml -v
# 3. Or run the full design pipeline (requires Gurobi license)
python iss.py config/camd_minimal.yaml -v| Solver | Required for | License |
|---|---|---|
| IPOPT | VLE, ISS, Antoine (all modes) | Open source |
| Gurobi | Molecule generation (design mode only) | Commercial — free academic license available |
Screen mode only requires IPOPT.
Design mode additionally requires a Gurobi license for MILP-based candidate generation.
A free academic license is available at gurobi.com/academia.
# Install IPOPT via conda (recommended)
conda install -c conda-forge ipopt
# Install package and dependencies
pip install -e .
# Optional: SMILES resolution for results (experimental resolve_smiles step)
pip install -e ".[smiles]"
# Optional: Jupyter notebooks (tutorials/, case_studies/)
pip install -e ".[notebooks]"
# Design mode only: Gurobi (requires a license)
pip install -e ".[design]"
grbgetkey <your-license-key>Both modes are driven by the same iss.py entry point — the mode is read from the config file's mode field — and produce CSV/JSON output.
python iss.py config/IECR/design_acetone_methanol.yaml -vmode: "design"
mixture:
component_a:
groups: {CH3: 1, CH3CO: 1} # acetone
component_b:
groups: {CH3OH: 1} # methanol
design:
groups: ["CH3", "CH2", "CH", "C", "OH", "HCO", "DMSO", "DOH"]
constraints:
min_groups: 3
max_groups: 6
min_func_groups: 1
max_func_groups: 2
molecular_type: 1 # 1 = acyclicpython iss.py config/DA/screen_homo_acetone_methanol.yaml -vmode: "screen"
mixture:
component_a:
groups: {CH3: 1, CH3CO: 1}
component_b:
groups: {CH3OH: 1}
screen:
entrainers:
- name: "ethylene_glycol"
groups: {DOH: 1}
- name: "DMSO"
groups: {DMSO: 1}Both modes run the same 15-step pipeline:
| Step | Description |
|---|---|
| 1 | Validate binary mixture (structural check) |
| 2 | Check UNIFAC parameter availability |
| 3 | Calculate pure component properties (Constantinou-Gani) |
| 4 | Apply user property overrides |
| 5 | Calculate Antoine parameters (Pyomo/IPOPT) |
| 6 | Generate candidates (design) or load entrainers (screen) |
| 7 | Check UNIFAC parameters for all candidates |
| 8 | Calculate pure properties for candidates |
| 9 | Screen by pure property constraints (Tb, Tc, Pc, …) |
| 10 | Calculate Antoine parameters for candidates |
| 11 | Miscibility screening over temperature range |
| 12 | Bubble-point VLE calculations (ternary) |
| 13 | Screen by mixture property constraints (selectivity, azeotropes) |
| 14 | ISS calculation — minimize entrainer-to-distillate ratio E/D |
| 15 | [Experimental] Resolve UNIFAC group composition to a candidate SMILES structure |
Results are written to the configured output.dir as results.csv and results.json.
A UNIFAC group composition doesn't uniquely determine one molecule — different constitutional isomers can share identical group counts (and get identical estimated properties, since UNIFAC is blind to connectivity). When step 15 finds more than one isomer for a candidate, the smiles column lists all of them joined by " | " rather than picking one arbitrarily.
python analyze_screening_log.py results/IECR/design_acetone_methanol/iss_run_*.log -o rejection_reasonsParses a screening log from either mode and breaks down why candidates were rejected — by pure-property constraint (Tb, Tm, Tc, Pc, Hv) at step 9 and by mixture constraint (azeotrope, miscibility, selectivity) at step 13 — printing counts and saving a bar chart (-o sets the base output name; both .pdf and .png are written). Useful for understanding which screening threshold is doing the rejecting, to help tune configs.
python check_azeotrope.py config/IECR/screen_acetone_methanol.yaml -vStandalone tool, separate from the design/screen pipelines. Reads only the mixture: section of any existing config file, solves for the binary A-B azeotrope via UNIFAC, and reports whether one was found (composition and temperature). Also plots xy and Txy diagrams with the azeotrope point marked, saved to output.dir if the config defines one. Uses a single initial guess (x=0.5, T=midpoint of Tb_A/Tb_B) — a root-finding problem, so it can miss an azeotrope far from that guess or fail to converge for a non-azeotropic pair. Pass --no-plot to skip plotting.
camd_iss/
├── iss.py # Entry point (design/screen mode via config['mode'])
├── check_azeotrope.py # Standalone binary azeotrope check + xy/Txy plots
├── analyze_screening_log.py # Post-run rejection-reason breakdown + bar chart
├── config/
│ ├── camd_minimal.yaml # Minimal working config template
│ ├── IECR/ # Original CAMD-ISS framework case studies
│ └── DA/ # Extended case studies (upcoming conference paper)
├── data/
│ └── json/ # UNIFAC parameters (rk, qk, interactions, Tc, Tb, …)
├── src/
│ ├── pipeline.py # Shared 15-step pipeline
│ ├── unifac_reverse.py # [Experimental] Group composition -> SMILES
│ ├── molecules/ # Structure validation and MILP generation
│ ├── pure/ # Property estimation and Antoine calculation
│ ├── mixture/unifac/ # UNIFAC, VLE, ISS (Pyomo/IPOPT)
│ └── data/ # Parameter loading and UNIFAC compatibility
├── case_studies/ # Analysis notebooks and worked examples
├── tutorials/ # Step-by-step Jupyter notebooks (01–10)
└── tests/ # Unit tests
The tutorials/ folder contains notebooks covering the full workflow:
| Notebook | Topic |
|---|---|
01_structures.ipynb |
UNIFAC group structures and validation |
02_pure_properties.ipynb |
Constantinou-Gani property estimation |
03_antoine.ipynb |
Antoine parameter determination (three methods) |
04_unifac.ipynb |
UNIFAC activity coefficient model |
05_vle_binary.ipynb |
Binary VLE calculations |
06_vle_ternary.ipynb |
Ternary VLE calculations |
07_azeotrope.ipynb |
Azeotrope detection |
08_iss.ipynb |
ISS optimization |
09_analyze_screening.ipynb |
Screening results analysis |
10_assemble_mol.ipynb |
[Experimental] UNIFAC group composition -> candidate SMILES (src/unifac_reverse.py) |
If you use this software in your research, please cite the relevant paper(s):
CAMD-ISS framework (IECR):
@article{iss_camd2026,
author = {Adem R. N. Aouichaoui and Ivonne Rodriguez-Donis and Nataliya Shcherbakova
and Edoardo Parascandolo and Vincent Gerbaud and Jens Abildskov},
title = {Computer-Aided Entrainer Design for Extractive Distillation Using Infinitely Sharp Splits},
journal = {ChemRxiv},
year = {2026},
doi = {10.26434/chemrxiv.15002406/v1},
url = {https://chemrxiv.org/doi/abs/10.26434/chemrxiv.15002406/v1},
}Extended case studies (DA conference paper): citation to be added once published.
This work originated as an M.Sc. thesis by Adem R.N. Aouichaoui under the supervision of Assoc. Prof. Jens Abildskov at the Process Systems Engineering Center (PROSYS), Technical University of Denmark (DTU).
pip install -e ".[dev]"
pytest tests/ -qIssues and pull requests are welcome. Please open an issue first to discuss proposed changes.
GNU General Public License v3. See LICENSE for details.