Designing an additive class for E10/E20 that stops water-induced phase separation, resists corrosion and wear, and stays octane/RVP-neutral — under a strict ashless, ≤$5/kg, ≤3-step rulebook.

What this is not: a set of measurements. There is no lab data here. Every number is a prediction from an open-source surrogate method, produced on a CPU without the licensed thermodynamic engine (COSMO-RS) or the GPU compute (DFT/MD) the problem really wants. The honest reliability contract is relative ranking, not absolute magnitude — read docs/methods.md before citing anything.

This repository is a completed course experiment (IIT), shared for transparency and reproducibility, not as a validated product.

---

The headline result

At the target treat rate (≤ 1000 ppm), simple bulk co-solvency cannot move the phase-separation temperature by the required +15 °C — the modelling shows a bulk shift of at most ~1 °C. Hitting the target at that dose is only possible through a surfactant / water-in-oil colloidal mechanism. So the project pivots from "find a better co-solvent" to "design an amphiphile that stabilises water as fine micro-droplets in the fuel," landing on a family of glyceryl-carbamate amphiphiles and a four-component candidate package, XTHENOL-100.

The lead package — XTHENOL-100

component	role	ppm
XG-1 — N-oleyl glyceryl carbamate	primary W/O amphiphile	450
XG-3 — N-(2-ethylhexyl) glyceryl carbamate	co-surfactant (HLB balancer)	300
A13 — 2-ethylhexyl lactate	polar linker / co-solvent	200
BHT	antioxidant	50
	total	1000

These molecules pass the ashless / cost / step-count / elastomer / CMR gates in the model, sit in the Davies W/O emulsifier window, and score well on the corrosion/lubricity QSAR. Whether they actually hold water at 1000 ppm is an open question that only MD and a bench D6422 test can answer — that experiment is staged, not done.

Analysis figure gallery

The plots below are the actual analysis output (not decoration); each carries its method caveat in the corresponding report.

Hard-constraint gate (Gate 4)	Hansen solubility map
UNIFAC water tolerance	Property / retention screen
Corrosion & lubricity QSAR	Composite ranking
XG amphiphile series	Mixture HLB (W/O window)
Package Hansen	Package corrosion film
Mixed-amine synthesis	Simplex-centroid DOE
Filter/injector + mileage audit

Repository map

xthenol/
├── README.md                  ← you are here
├── LICENSE                    ← MIT (code)
├── LICENSE-DATA               ← CC-BY-4.0 (data & figures)
├── CITATION.cff               ← how to cite this experiment
├── environment.yml            ← conda env (rdkit, thermo, chemicals, …)
├── requirements.txt           ← pip pins
├── docs/
│   ├── 00_project_brief.md              ← the objective this was built against
│   ├── methods.md                       ← what was computed + honesty statement (READ FIRST)
│   ├── 01_triage_session_report.md      ← 38-candidate triage, steps 1–8
│   ├── 02_XG_series_design_brief.md     ← glyceryl-carbamate amphiphile design
│   └── 03_XTHENOL-100_formulation_report.md  ← the four-component package
├── src/
│   ├── xthenol_pipeline.py    ← the descriptor / QSAR / HSP engine
│   ├── reproduce_scores.py    ← recompute + cross-check the tables from SMILES
│   └── README.md              ← pipeline reference + reliability contract
├── data/                      ← 15 files; see data/README.md for the dictionary
│   ├── candidate_library.csv           master_scorecard.csv
│   ├── constraint_screen.csv           hansen_triage.csv
│   ├── water_tolerance.csv             property_screen.csv
│   ├── corrosion_lubricity.csv         xg_series_detail.csv
│   ├── package_hansen.csv              package_corrosion.csv
│   ├── doe_simplex.csv                 wt_results_raw.json
│   └── README.md              ← data dictionary
├── figures/                   ← 12 analysis PNGs (the gallery above)
├── assets/                    ← designed graphics (banner, molecule showcase, funnel)
└── plans/                     ← the two execution plans this work followed

Reproduce it

# 1. build the environment
conda env create -f environment.yml
conda activate xthenol

# 2. run the pipeline self-test (anchors must reproduce exactly)
cd src
python xthenol_pipeline.py

# 3. recompute the descriptor tables from SMILES and cross-check vs data/
python reproduce_scores.py          # add -v to print every row

Expected: all Hansen, logP and Tb rows and 36/38 corrosion rows reproduce to CSV rounding. Two rows (A18, A19) differ by design — a documented amide→carbamate re-typing correction; the driver prints this. The UNIFAC water-tolerance step (water_tolerance.csv) is a heavier liquid-liquid-flash computation described in docs/methods.md and not re-run by the quick driver.

Limitations (the short version — full list in docs/methods.md)

1. No measurements. Everything is a surrogate prediction; trust rank order, not magnitude.
2. COSMO-RS → UNIFAC-Dortmund, DFT → descriptor QSAR, MD → not done. The colloidal mechanism that the whole result rests on is hypothesised, not simulated.
3. The ~6.9 % E20 energy-density loss is physics, not a bug — no additive recovers it.
4. Prior-art search is shallow and is not a freedom-to-operate opinion.
5. The single most valuable next step is MD of the XG-1/XG-3 mixed film at the water–ethanol–hydrocarbon interface at 1000 ppm, followed by a bench D6422 ΔTsep + DLS droplet-size measurement.

Citing

See CITATION.cff. Code is MIT; data and figures are CC-BY-4.0.

Author

@infinitule — IIT. Course experiment, 2026.