Skip to content

methods.md

Latest commit

 

History

History
123 lines (102 loc) · 7.69 KB

methods.md

File metadata and controls

123 lines (102 loc) · 7.69 KB
Raw

Methods & honesty statement

This document describes exactly what was computed, with what tool, and where each method departs from what the project brief asked for. Read this before citing any number in ../data/. The single most important thing to understand: this is a computational triage, run without access to the licensed thermodynamic engine or GPU compute the brief specified. Every value is a prediction from a surrogate method, and the reliable output is relative ranking, not absolute magnitude.

Compute environment

  • All work ran in a single CPU environment (12 cores, 32 GiB RAM, no GPU) with no remote compute available. This is the reason for every substitution below.
  • Conda env xthenol (Python 3.11, conda-forge): rdkit 2026.03.4, thermo 0.6.1, chemicals 1.5.2, plus numpy/pandas/scipy/matplotlib. Pinned in ../environment.yml and ../requirements.txt.

Method substitutions (what the brief asked for → what was actually used)

Brief asked for Used instead Why Consequence
COSMO-RS activity coefficients UNIFAC-Dortmund (modified UNIFAC, 2016 params) via thermo no COSMO-RS licence, no compute absolute Tsep is biased (UNIFAC over-predicts water/HC immiscibility); use relative ΔTsep only
DFT adsorption energies for corrosion descriptor QSAR calibrated to literature anchors no GPU/DFT %IE is a calibrated estimate, not a first-principles energy
MD of the water–EtOH–HC interface not substituted no GPU the colloidal mechanism is hypothesised, not simulated — this is the critical-path follow-up
Hansen parameters (brief allowed g.c.) van Krevelen / Fedors group contribution standard triage method consistent with brief

Each of these tags appears in the reports and figures wherever the affected number is quoted. Nothing is presented as a measurement.

Step-by-step

Gasoline surrogate

Toluene/isooctane/n-heptane 30/45/25 vol% (per brief). Its Hansen point is [15.57, 0.42, 0.6] MPa^0.5; the ethanol–water polar cluster is [15.77, 9.52, 21.69].

1. Candidate library (candidate_library.csv, 38 rows)

6 baselines + 32 novel/multifunctional candidates. Each SMILES was decomposed into modified-UNIFAC (DOUFSG) groups; the group-implied molecular formula was reconciled against the RDKit formula and four amide mismatches were corrected before any modelling.

2. Hard-constraint gate — Gate 4 (constraint_screen.csv)

RDKit descriptors + SMARTS alerts: forbidden elements (metals/P/Si), sulfur, CMR 1A/1B substructures, synthesis-step estimate, cost band. All 38 are ashless, S-free and CMR-clean; 23/38 pass Gate 4 outright, 15 are flagged (14 on cost > $5/kg, 1 also on step count) and carried forward with the flag rather than hard-rejected.

3. Hansen triage (hansen_triage.csv, hansen_map.png)

van Krevelen dD/dP/dH for every candidate; reference solvents pinned to literature HSP, group contribution used only for the novel structures. The bridging score sqrt(Ra_hc · Ra_polar) + 0.5·|Ra_hc − Ra_polar| ranks how well a molecule bridges the hydrocarbon surrogate and the polar ethanol–water cluster (lower = better).

4. UNIFAC-Dortmund water tolerance (water_tolerance.csv, wt_results_raw.json)

A liquid–liquid flash with a Gibbs-energy stability test (is_two_phase) over the quaternary water/ethanol/ethanol-additive/HC-surrogate system, swept in temperature. Baseline single-phase water tolerances (vol%): E10 0.239/0.117/0.090 at 25/0/−10 °C; E20 0.331/0.179/0.146.

Central finding of the whole project: at ≤ 1000 ppm, the bulk co-solvency ΔTsep is negligible (max +1.23 °C, for TBA). Reaching the +15 °C target by bulk co-solvency needs 1–5 wt% additive. Therefore any additive that hits the target at ≤ 1000 ppm must work by a surfactant / colloidal (water-in-oil micro-emulsion) mechanism — which is outside UNIFAC's bulk-thermodynamic scope. This is why the designed series is scored on interface descriptors and HLB, not on the bulk-ΔTsep axis, and why the phase cells for carbamates are left NaN (UNIFAC cannot represent the carbamate group) and the package row is pending-bench.

5. Property / retention screen (property_screen.csv)

Crippen logP; Joback Tb (RVP flag if Tb < 340 K); octane-effect sign; UNIFAC hydrocarbon-phase retention fraction at 1000 ppm wet; Hansen distance to NBR and FKM elastomers. The core tension surfaces here: small polar co-solvent "winners" partition into the water phase (fail retention), while large amphiphiles (logP > 2) stay in fuel.

6. Corrosion & lubricity QSAR (corrosion_lubricity.csv)

DFT/MD-pending. A head-group adsorption model:

adsorption_index = 2.7·hs + 1.0·chain_term(min chain 14) + 0.5·donor_term(min donors 5)
                   + 0.25·unsat − 2.05
pred_IE   = 100 / (1 + exp(−1.7·adsorption_index))            # Fe/Al inhibition %, target ≥ 90
pred_HFRR = max(300, 600 − 150·hs − 110·chain_term(min 18)
                 − 15·min(donors,4))                          # wear scar µm, target ≤ 450

hs is the head-group strength (carbamate 0.75, carboxyl 1.0, amide 0.95, …). The model is calibrated so literature anchors reproduce their anchored values — see sanity_check() in ../src/xthenol_pipeline.py. Carbamate head-typing must be tested before the generic amide pattern; A18/A19 were re-scored after this correction and the corrected values live in master_scorecard.csv (see the note in ../src/README.md).

7. Composite ranking, prior art, gating (master_scorecard.csv)

Weighted composite (0–100): s_corr 0.22, s_retention 0.18, s_lube 0.16, s_potency 0.12, s_octane 0.10, s_elast 0.10, s_cost 0.12. Prior-art search performed 2026-07 (patents + literature) at limited depth — not a freedom-to-operate opinion; verify before any IP filing. Novelty classification across the 38: 18 known / 13 known-fuel / 6 novel-ish / 1 novel-homolog / 0 fully-novel-class.

8. Designed series & package (XG series, XTHENOL-100)

The glyceryl-carbamate amphiphiles (XG-1 N-oleyl, XG-2 N-2-hexyldecyl, XG-3 N-2-ethylhexyl, XG-4 N-isostearyl) and the four-component XTHENOL-100 package (XG-1 450 / XG-3 300 / A13 200 / BHT 50 ppm = 1000 ppm total) were designed against the surfactant mechanism. Package analysis covers mixture HLB (Davies 4–6 W/O window across XG-1 mass fraction 34–75 %), mass-weighted Hansen, a per-component corrosion film model, a mixed-amine one-pot synthesis audit, a simplex-centroid DOE (the XG-1/XG-3 pair tops the interfacial surface), and a filter/injector-risk + mileage audit. Details and every number are in the three reports (01–03).

Standing limitations (carry these into any downstream use)

  1. Absolute values are surrogate predictions. Trust rank order within a family, not the magnitude. Relative ΔTsep, not absolute Tsep.
  2. The colloidal mechanism is un-simulated. The ≤ 1000 ppm phase-separation claim rests on HLB and interface descriptors, not on MD. The staged MD of the XG-1/XG-3 mixed film vs an XG-1-only film at the water–ethanol–HC interface at 1000 ppm is the single most important missing experiment.
  3. The ~6.9 % E20 volumetric-energy (LHV) deficit is thermodynamically fixed — no additive recovers it. XTHENOL-100 can honestly claim only ~0.8–2.3 % steady-state (friction + deposits) plus avoided phase-event losses.
  4. Corrosion is QSAR, not DFT. %IE and HFRR are calibrated estimates.
  5. Prior-art search is shallow. Not an FTO opinion.
  6. The cheapest falsifying bench test for each candidate is stated in master_scorecard.csv (falsification_test); the package kill-switch is a DLS droplet measurement + D6422 ΔTsep at 1000 ppm E10.