The PFAS-Free,
Self-Pumping
Cooling Platform.
Novec is discontinued. The EU PFAS restriction lands in 2026–27. AI accelerators crossed 1,000 W. Cooling Fluid 10.0 is the first engineered-Marangoni two-phase cooling system — PFAS-free binary working fluid plus a 7-claim integrated enhancement stack (copper foam · ultrasound · dielectric nanoparticles · biphilic surface · sealed thermosyphon · microgravity-capable). 234 claims across three inventor-prepared provisionals. 50–100× cheaper per liter than Novec 649.
What is claimed vs. what is measured.
- PFAS-free binary working-fluid genus (Δσ ≥ 3 mN/m, ΔTb ≥ 30 °C)
- Solutal Marangoni self-pumping mechanism
- 7-component integrated enhancement stack (foam · ultrasound · nanoparticles · biphilic · sealed · microgravity)
- Discovery-engine methodology (Bayesian GP + MD)
- 11 independent cross-check methods converge on σ, miscibility, force-field independence
- DDB experimental σ anchor: 4% AAD on a carbonate reference pair
- NIST ThermoML density anchor: 2.29% AAD on the same reference
- σ(T) Eötvös fits: 5 pairs, average R² = 0.85
- Wetlab σ, ε, flash, 30-day stability — Phase 3 (Q2–Q3 2026, ~$5.5k, 40 pairs across Tier A + Tier B)
- Stacked CHF on the full integrated system — Phase 5 pool-boiling (Q1–Q2 2027)
- Rack-scale deployment and long-term stability — Phase 4 licensee pilot
- USPTO filing receipts and ownership proof — in the NDA-gated data room
Specific compositions, ratios, additives, and per-pair numeric values are disclosed under NDA in the buyer dossier. Every number on this public site traces to a canonical repository file; independent methodology, statistical rigor, and honest disclosure of measurement gaps are the same standard we apply internally and externally.
PFAS Is Leaving.
Heat Flux Is Rising.
Three forcing functions converge in the next eighteen months. Every month of delay increases supply-chain exposure.
3M Market Exit
3M discontinued Novec™ and Fluorinert™ effective December 2025. $500M+ replacement demand in data-center immersion alone — cascades into heat pipes, EV thermal packs, and defense applications that depended on the same chemistry.
Universal PFAS Restriction
ECHA final opinions on the universal PFAS restriction under REACH (EC) 1907/2006 are expected 2026–2027. HFO "low-GWP" replacements (Opteon 2P50, Solstice) contain C–F bonds and face re-classification risk.
The Thermal Wall
NVIDIA B200 at 1,000 W. AMD MI325X at 750 W. Blackwell Ultra in 2027 projected 1,500–2,000 W. Air cooling died near 400 W per socket. Single-phase liquid tops out near 1,000 W because it cannot exploit phase change.
Every PFAS-free cooling fluid shipping today — Engineered Fluids EC-100, Shell S5 X, Castrol ON DC 15 — is single-phase. None self-pump. None have a Marangoni gradient. The physics forbids it in single-component fluids.
How a Fluid
Pumps Itself.
Solutal Marangoni Induction
A binary mixture of fuel (low σ, low Tb) and pump (high σ, high Tb) sits over a hot chip. The fuel evaporates preferentially, locally depleting from the surface.
Gradient Formation
Local σ rises sharply at the depleted hotspot. Cooler, fuel-rich bulk liquid (lower σ) is pulled along the ∂σ/∂x gradient toward the hot spot.
Passive Phase-Change
Arriving fluid evaporates, extracting latent heat (ΔHvap). Vapor condenses at the cold plate and re-mixes. Closed-loop thermodynamic engine in a static volume.
Self-Regulation
Higher heat load → steeper gradient → faster Marangoni pumping. The mechanism is inherently self-throttling. No controller, no sensor, no software.
/∂x
Quad-Marangoni Stack.
Four independent surface-tension drivers, all pointed at the same hotspot. Each mechanism has a literature or inventor-spec basis; combined CHF/HTC performance is still a buyer-pilot hypothesis.
Composition Gradient
Preferential fuel evaporation at hotspot → local σ rises → bulk liquid pulled toward hotspot. Primary mechanism. Mathematically equivalent to a passive pump. PROV-003 Claims 3–4.
Inverse Thermal (∂σ/∂T > 0)
Long-chain alcohol (a claimed long-chain alcohol additive) inverts the sign of ∂σ/∂T in dilute solutions. Literature reports 2–5× heat-pipe dryout margin improvement (Abe, 2004). PROV-003 Claim 56.
Dissolved CO₂ / N₂ / Ar
Dissolved gas desorbs at hot surface. Literature reports high CHF under dissolved-gas-enhanced boiling, but not on these blends. PROV-003 Claim 57.
dielectric nanoparticle Concentration Marangoni
a claimed dielectric nanoparticle (per CIP-draft Claim 13). Redistribution at the evaporating interface creates a fourth driver while boosting bulk thermal conductivity.
All four mechanisms reinforce; none oppose. Individual mechanisms claimed in PROV-003 Claim 56 (self-rewetting), Claim 57 (dissolved gas), Claims 3–4 (solutal). Combination claimed in Claim 58.
Real Physics,
Not Cartoons.
These videos are directly rendered from OpenFOAM CFD cases and GROMACS molecular trajectories — not marketing animations. Each frame is a snapshot of raw simulation state.
Free-Surface + Velocity Field
Atomistic σ Measurement
OpenFOAM interFoam
Volume-of-fluid two-phase solver. Free-surface evolution driven by surface-tension force with prescribed Marangoni stress at the interface. Eleven time steps from 0 to 100 ms show the development of circulation cells and σ-gradient fingers at the gas-liquid interface.
GROMACS slab MD
GAFF2 / AM1-BCC, 2 ns NVT with 3dc Ewald slab correction, PME electrostatics. σ computed from the zz−xx pressure-tensor anisotropy integrated across the box. Every one of our 172 σ values came from this protocol across 59 canonical pairs.
UQ stratified
Dielectric-safety reference (PROV-003 covered) n=7 pool: 4 Mac + 3 Pod replicas. Pooled σ = 16.95 ± 2.38 mN/m (95% CI, t-dist df=6). Per-site Δ = 2.45 mN/m, Welch's p ≈ 0.3. Disclosed — not hidden. See docs/audit/UQ_STRATIFIED_ANALYSIS.md.
Every Computational Layer.
One line per method from screening to reproducibility. Every number on this site traces to an authoritative file, but remains wetlab-pending. Full matrix (with SMILES, convergence stats, per-run logs) in the NDA-gated diligence room.
| Layer | Method | Scale | Output | Authority |
|---|---|---|---|---|
| Screening | Hansen Ra miscibility filter | 250+ candidates | Pre-MD miscibility gate (Ra < 14) | docs/audit/GAP_ANALYSIS.md |
| Screening | PubChem genus screener | 250+ queried · ~52 advanced | CID, SMILES, fluorine count, existence check | src/discovery/pubchem_screener.py |
| Primary MD | GROMACS GAFF2 slab MD | 59 pairs · 172 σ measurements | σ (mN/m) via P-tensor, 3dc Ewald, 2 ns NVT | CANONICAL_INVENTORY_FINAL.json |
| Active learning | Bayesian GP + MD loop | 36-pair train → 6 post-filing hits | EI-ranked top-20 candidates | data/processed/bayesian_top20.json |
| Quantum | CP2K DFT binding energy | 2 pairs (Microgravity sulfolane-pump candidate, C6 sulfolane literature-LLE reference) | ΔE (kcal/mol) via PBE-D3/DZVP | src/simulations/cp2k/ |
| ML | Random-Forest σ predictor | R²=0.794 · MAE 5.5 mN/m · 16 features | O(1) σ prediction per SMILES pair | models/sigma_predictor.pkl |
| ML | Chemprop D-MPNN + mixture-mode retrain | Graph-neural-net · 36-pair train · 606 predictions · mixture-mode ready | Independent σ validation for arbitrary SMILES pairs | scripts/chemprop_mixture_retrain.py (stub) · chemprop 2.2.3 installed |
| ML | ChemBERTa-class Hansen-SMILES predictor (lightweight fallback) | 37 Hansen-anchor components · RDKit 41-dim descriptors · LOO-CV | δD/δP/δH prediction · LOO MAE ~2 MPa^½ | scripts/chemberta_hsp_predict.py |
| ML | MACE-OFF23 / xTB GFN2 single-points | Top-3 pairs | Force-field cross-check | external/mattersim |
| Thermo | openCOSMO-RS_py + LVPP v25 σ-profile database | 29 of 37 target components (2500+ total in LVPP) | NWChem-computed COSMO surface charges; σ-profile histograms | scripts/lvpp_sigma_profile.py |
| Thermo | Parachor-Weinaug-Katz mixture σ | All 59 pairs | σ^(1/4) = (ρ/MW) × Σ xi·Pi — AAD 1.84% lit. | scripts/parachor_mixture_sigma.py |
| Thermo | Macleod-Sugden + Meissner-Michaels + Tamura-Kurata mixing | All 59 pairs × 3 methods | 4-method σ cross-validation table (target AAD ~16%) | outputs/sigma_method_comparison.md |
| Literature | DDB + Green Chem 2025 + Mudawar 2018 sweep | 12 key pairs with open experimental / LLE / CHF anchors | Marangoni champion B (FDA-GRAS) immiscibility risk flag; CHF hypothesis anchors only | docs/LITERATURE_SWEEP_2026-04-18.md |
| CFD | OpenFOAM interFoam multiphase | 12 cases (6 retained post-rebuild) | Marangoni velocity mm/s | src/simulations/openfoam/ |
| T-sweep | σ(T) Eötvös fit | 3 pairs complete (Hex+Sulf 7T, Higher-Tflash indirect-loop variant 6T, Dielectric-safety reference (PROV-003 covered) 4T) | dσ/dT slope, Eötvös R² | data/processed/sigma_T_sweeps.json |
| Composition | C6 sulfolane literature-LLE reference 27-point σ(x,T) surface | Composition × temperature grid | Publication-grade surface | pod_harvest_final |
| UQ | Stratified Mac-vs-Pod analysis | Dielectric-safety reference (PROV-003 covered) n=7 pool (4 Mac + 3 Pod) | Inter-site Δ=2.45 mN/m · Welch p≈0.3 | docs/audit/UQ_STRATIFIED_ANALYSIS.md |
| FoM | T-1.2 Novec-normalized FoM4 | 10 top pairs re-scored on post-merge σ | Weighted sum 0.50·Marangoni + 0.20·ε + 0.30·thermal | data/processed/corrected_fom_nist.json |
| Sources | NIST / PubChem URL validation | 52 pure-component URLs · all PASS | Buyer-verifiable property dictionary | scripts/validate_fom_sources.py |
| Infra | Docker reproducibility stack | 2 Dockerfiles + run_all.sh + 8 subcommands | `docker run cf10:cpu check-only` < 5 min | REPRODUCIBILITY.md |
Ten Candidates.
Three Leaders.
This table is the canonical post-T-1.2 re-derivation: Novec-normalized weighted-sum FoM4 against the n=59 pairs / 172 σ canonical inventory. Legacy Dielectric-safety reference (PROV-003 covered) 5.96× is a pre-merge artifact; the current ranking is below.
Indirect-loop composite reference
Marangoni champion A (C5 + lactone)
Marangoni champion B (FDA-GRAS)
Nitrile-pump candidate
Cyclic-ketone candidate
Rank-3 amide-pump candidate
Dielectric-safety reference
Higher-Tflash C6 dielectric analog
Microgravity sulfolane candidate
C6 sulfolane literature-LLE reference
Three Pumps, Three Gates.
Computationally strong but gated by regulatory / handling constraints. Kept as follow-on secondary embodiments — if a buyer's vertical accepts the gate (e.g. NMP in GMP pharma, DMSO in lab-scale), these become lead candidates for that narrow market.
Rank-3 amide-pump candidate
N,N-dimethylformamide is reproductive/developmental toxin (EU CMR cat 1B). Gloves + face shield for >10 mL pours. Gate: buyer chemical-hygiene policy.
SVHC-gated NMP candidate
N-methyl-2-pyrrolidinone is REACH SVHC (reprotox). Heavily restricted for consumer use; still permitted in industrial/GMP with controls. Gate: REACH classification.
Dermal-restricted DMSO candidate
Dimethyl sulfoxide penetrates intact skin and carries dissolved compounds with it. Specialty labs OK with PPE; volume deployment problematic. Gate: OHS + operator safety review.
Two Glycerol Pairs.
Canonical MD-simulated pairs covered by PROV-001 polyol genus (Claim 4). Indirect-loop only (glycerol ε ~ 43 disqualifies direct immersion). Current inventor-spec candidates, not follow-on priorities. Included for portfolio breadth, not as lead chemistry.
Bio-based polyol indirect reference
Indirect-loop reference. ⚠ GAFF2 drifts to σ ≈ 0 at MeOH > 80 mol% — pendant-drop is the fix. Covered by PROV-001 polyol genus (Claim 4). Bio-based and PFAS-free.
Cycloalkane polyol indirect variant
Alternative cyclic fuel + polyol pump. Covered by PROV-001 polyol genus + alkane alternative. Hansen Ra high (glycerol dominates polar space); full miscibility pending wetlab. Bio-based.
The portfolio is not one fluid. Five distinct operating envelopes are covered by different pair sub-families — direct two-phase is the flagship, but indirect-loop, higher-flash, microgravity, and fluorinated-retrofit all have dedicated candidates.
Flagship Embodiments.
Each flagship is either named in the inventor-prepared specifications or listed as a follow-on candidate. C+ represents post-filing Bayes discoveries that must clear safety, miscibility, wetlab, and counsel gates before external product claims.
Non-Flammable Data Center (indirect loop)
NFPA-75 compliant formulation with flash > 150 °C for indirect-loop deployments where ε > 10 is acceptable (no live-electronics contact). Commodity-scale components; materials cost ~3–4× cheaper than the leading PFAS-free single-phase hydrocarbon benchmark.
Quad-Marangoni Premium
All four claimed Marangoni mechanisms stacked (solutal + inverse-thermal + dissolved-gas + nanoparticle) — the PROV-003 Claim 58 capstone embodiment. CHF / HTC uplift on the stacked system is measured in Phase 5 pool-boiling (Q1–Q2 2027).
Dielectric Immersion (sealed, two-phase)
Lowest-ε direct-immersion embodiment in the portfolio — within the Novec-class dielectric band. Covered by the PROV-003 composition-range claim. Sealed thermosyphon geometry per Claim 8 accommodates C5-alkane vapor pressure at 70–95 °C chip junctions.
Direct-Immersion Marangoni Champions
Highest computational FoM4 screening scores in the post-filing discovery set. Claim-scope extension to this pump class is the follow-on filing priority (benefit-claim target 2027-01-29). Miscibility, pentane-safety, dielectric, and flash-point gates are scheduled in Phase 3 wetlab.
FR4-Safe Immersion
Hansen-favorable immersion pair where neither component is expected to attack FR4 epoxy. Low material cost, wetlab compatibility gated in Phase 3 Tier B fast-screen.
Space / Microgravity
Bond number < 10⁻⁵; gravity-independent passive operation from −40 to +120 °C. PROV-003 Claim 10 (method of cooling in microgravity, g < 0.1 g, solutal Marangoni as primary driver). No competitor defends microgravity two-phase.
Retrofit · Drain-and-Fill Upgrade
Method embodiment for retrofit of in-service Novec / Fluorinert tanks after wetlab and system-safety validation. Drain legacy PFAS coolant, purge, refill with a qualified PFAS-free composition, leak-check. PROV-003 Claim 11 (method).
The CHF Stack
Hardware-Optimized
Interfaces.
Individual enhancements live in the published literature; the combination with Cooling Fluid 10.0 blends has not been measured here. Literature-additive upper bound for the full stack ≈ 8× CHF vs bare single-component fluid; conservative floor 2–3×; cross-coupling among enhancements uncharacterized. Phase 5 pool-boiling (Q1–Q2 2027) is the canonical measurement; until then this is disclosed as claimed + literature-supported, not measured. Full matrix with per-enhancement literature citations and honest §112(a) enablement-risk disclosure is available in the NDA-gated buyer dossier.
- ▸ Copper foam — CIP-draft Claim 11 — 70–95% porosity, 10–100 PPI wick at die surface.
- ▸ Biphilic surface — CIP-draft Claim 14 — 10–500 μm alternating hydrophilic/hydrophobic domains.
- ▸ Al₂O₃ conditioning — One-time deposition, not operating nanofluid.
- ▸ 40 kHz ultrasonic — CIP-draft Claim 12 — 20–80 kHz acoustically coupled transducer suppresses bubble pinning.
- ▸ Dielectric nanoparticle — CIP-draft Claim 13 — k≥10 W/m·K, 0.01–1% vol (defensive; §112(a) enablement-flagged).
- ▸ Dissolved CO₂ — 1–3 mol/L; lowers nucleation superheat.
Bare Fluid vs Integrated System.
Every commercial PFAS-free or Novec-legacy coolant is a bare single-component single-phase dielectric. Cooling Fluid 10.0 defends a 7-component integrated stack: PFAS-free binary Δσ ≥ 3 mN/m + copper-foam wick (CIP-draft Claim 11) + 40 kHz ultrasonic (CIP-draft Claim 12) + dielectric nanoparticles (CIP-draft Claim 13) + biphilic heat surface (CIP-draft Claim 14) + sealed thermosyphon geometry (PROV-003 Claim 8) + microgravity operation (PROV-003 Claim 10). Bare-fluid vs bare-fluid is the wrong comparison.
| Competitor | Type | Marangoni Δσ | Copper foam | Ultrasound | Nanoparticles | Biphilic | Microgravity |
|---|---|---|---|---|---|---|---|
| Cooling Fluid 10.0 (claimed stack) | PFAS-free binary + integrated 7-component system | ✓ MD-confirmed | ✓ Claim 11 | ✓ Claim 12 | ✓ Claim 13 | ✓ Claim 14 | ✓ Claim 10 |
| Novec 649 / 7000 / 7100 | fluorinated (PFAS, discontinued 2025) | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Castrol ON DC 20 | hydrocarbon single-phase | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Submer SmartCoolant | hydrocarbon single-phase | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Shell Immersion S5 | GTL synthetic hydrocarbon | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| ExxonMobil Coolanol | hydrocarbon single-phase | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Perstorp Synmerse DC | synthetic ester single-phase | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Engineered Fluids SLIC | synthetic hydrocarbon single-phase | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| M&I Materials MIVOLT | synthetic ester single-phase | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
Literature uplift ranges: Marangoni +30–50%, copper foam +100–200%, ultrasound +20–30%, nanoparticles +15–40%, biphilic +50–100%. Cross-coupling among enhancements is not characterized; Phase-5 pool-boiling rig (T34 stack-coupled test) is the canonical measurement. Claims 11–14 are forward-looking defensive claims with §112(a) enablement risk flagged. Full matrix + per-enhancement literature citations in the NDA-gated buyer dossier.
Not Just Immersion.
The Marangoni platform spans six thermal-management verticals. Each uses a different fluid composition from the 59-pair canonical inventory. See docs/Corporate_Synergy_Map.md for buyer alignment per vertical.
Data-Center Immersion
NVIDIA B200, MI325X, Rubin. These are candidate use cases only until miscibility, pentane safety, dielectric, flash-point, seal, and CHF/HTC gates pass.
Indirect Cold-Plate Loop
Pump-elimination: replace loop pumps with self-pumping fluid. ~$20K/rack TCO saving. Higher ε is acceptable (no live-electronics contact).
Heat Pipes / Thermosyphons
n-BuOH ternary flips ∂σ/∂T sign → 2–5× dryout margin vs aqueous heat pipes. PROV-001 covers.
EV Battery Thermal
Pack-level immersion and cell-level cooling remain candidate applications. PC status does not override miscibility, dielectric, flash-point, seal, and cell-compatibility gates.
Space / Microgravity
Bond number < 10⁻⁵; gravity-independent. PROV-001 Claim 10 covers microgravity method. −40 to +120 °C envelope. NASA / DARPA / AFRL channel.
Defense / Directed Energy
High-ε enables RF-transparent laser coolant. DoD procurement channel (sole-source supplier). Sulfolane SVHC risk gated by Hex+Pyrrolidinone fallback.
Proof, Per Pair, Per Method.
FoM4 screening score vs Novec 7100 · post-T-1.2 canonical
0.50×Marangoni + 0.20×dielectric + 0.30×thermal · NIST-URL-verified (52 sources PASS)
CFD Velocities
Corrected analytically after 12 OpenFOAM cases were rebuilt with per-fluid σ/μ/ρ (original 12 used hardcoded inputs). Six remain as exploratory computational screens.
| Pair | n reps | Δσ | T-sweep | CFD | DFT | FoM4 | ε | Ra | Reg |
|---|---|---|---|---|---|---|---|---|---|
| Marangoni champion A | 7 | 27.57 | 4T | — | — | 3.96× | 8.3 | 19.48 | follow-on |
| Marangoni champion B (FDA-GRAS) | 3 | 22.44 | — | — | — | 3.45× | 11.3 | 21.49 | LLE risk |
| Dielectric-safety reference (PROV-003 covered) | 7 | 11.25 | 4T | 119 | — | 3.50× | 2.05 | 13.1 | PROV-003 |
| Cyclic-ketone candidate | 4 | 20.44 | — | — | — | 3.66× | 4.0 | 9.8 | OK |
| Rank-3 amide-pump candidate | 5 | 23.08 | — | — | — | 3.92× | 8.7 | 10.2 | reprotox |
| Nitrile-pump candidate | 1 | 14.93 | — | — | — | 3.78× | 12.7 | 11.4 | OK |
| Higher-Tflash C6 dielectric analog | 3 | 10.94 | — | — | — | 3.12× | 2.1 | 13.4 | OK |
| C6 sulfolane literature-LLE reference | 1 | 7.48 | 7T | 62 | −1.61 | 2.36× | 8.2 | 23.8 | SVHC |
| Higher-Tflash indirect-loop variant | 5 | 8.1 | 6T | 42 | — | — | 21 | 3.8 | OK |
| Indirect-loop composite reference | 24 | GAFF2* | 12T | — | −3.06 | 6.36× | 32.2 | 4.1 | indirect |
Who Wins On Which Axis.
5-objective Pareto analysis: FoM4 ↑ · Hansen Ra ↓ · UQ n-replicas ↑ · σ(T) coverage ↑ · ε ↓. 8 of 11 candidates are Pareto-optimal — no single metric tells the whole story. data/processed/pareto_fom_hansen_uq.json
Marangoni champion B (FDA-GRAS) (FoM4 3.83×) — computationally attractive but literature/Hansen risk-flagged for immiscibility; keep only as a phase-mapping-gated follow-on candidate. Rank-3 amide-pump candidate — no axis-winning; reprotox handling penalty further gates. Microgravity sulfolane-pump candidate — no axis-winning; SVHC risk 2026-27. The Pareto lens is over computational objectives only; regulatory (FDA, SVHC, Schedule I), handling (reprotox), and buyer-preference lenses overlay separately.
Who Wins Under Diligence?
Multiplicative composite: FoM4 × Hansen × Regulatory × UQ × filing posture. Reveals which pair a buyer's due-diligence officer (rather than a computational chemist) would rank #1. Per-axis penalty curves in data/processed/risk_adjusted_score.json.
| Rank | Pair | Raw FoM4 | Risk-adj | Hansen Ra | UQ n | Reg gate | PROV |
|---|---|---|---|---|---|---|---|
| 1 | Indirect-loop composite reference | 6.36× | 3.82 | 15.5 | 24 | clean | PROV |
| 2 | Dielectric-safety reference (PROV-003 covered) | 3.5× | 2.49 | 10.6 | 7 | clean | PROV Cl.13 |
| 3 | Marangoni champion A | 3.96× | 2.03 | 19.5 | 7 | List I (US) | follow-on |
| 4 | Rank-3 amide-pump candidate | 3.92× | 1.69 | 10.2 | 5 | reprotox | follow-on |
| 5 | Higher-Tflash C6 dielectric analog | 3.12× | 1.58 | 10.5 | 3 | clean | PROV |
| 6 | Cyclic-ketone candidate | 3.66× | 1.45 | 10.4 | 4 | clean | PROV |
| 7 | Marangoni champion B (FDA-GRAS) | 3.45× | 0.95 | 21.5 | 3 | LLE risk | follow-on |
| 8 | Nitrile-pump candidate | 3.86× | 0.48 | 19.1 | 4 | restricted | follow-on |
| 9 | Microgravity sulfolane-pump candidate | 2.86× | 0.14 | 22.6 | 1 | SVHC | PROV |
| 10 | C6 sulfolane literature-LLE reference | 2.36× | 0.12 | 23.8 | 1 | SVHC | PROV |
Raw FoM4 puts Marangoni champion A at the top of the direct-immersion candidate list (3.96× pool / 4.06× canonical, n=7 UQ pool post-2026-04-18). Under compound risk-adjustment — Hansen immiscibility risk (Ra 19.5) × List-I handling × follow-on filing pending — Marangoni champion A drops. Dielectric-safety reference (PROV-003 covered), with its PROV-003 Claim 13 example + n=7 UQ pool + Hansen-borderline (10.6) + ε=2.05 uniqueness, rises to rank 2 (2.49) — the conservative pre-wetlab reference candidate. Passing safety, miscibility, dielectric, and flash-point gates can move Marangoni champion A back into contention.
Eleven Ways We Proved It.
Every σ / Marangoni / miscibility claim in the portfolio is supported by at least three structurally independent methods. Two experimental anchors (Higher-Tflash C6 analog at 4% AAD on σ, 2.29% AAD on ρ). Two independent force fields (GAFF2/AM1-BCC + OPLS-AA + MACE-OFF23). Two independent thermodynamic engines (openCOSMO-RS γ∞ + γ(x,T) surface). Not one method. Not one force field. Not prediction-only.
| # | Method | Type | Result summary | Proves |
|---|---|---|---|---|
| 1 | GAFF2/AM1-BCC slab MD | Primary σ | 172 σ values across 59 pairs; n=3–7 pool for top-6 | Core methodology |
| 2 | 4-way mixing rules (Parachor-WK, Macleod-Sugden, Meissner-Michaels, Tamura-Kurata-Odani) | Analytical σ | Target-family AAD 16% | MD direction confirmed; Gibbs-adsorption-consistent |
| 3 | DDB experimental σ anchor (Piñeiro 2004) | Literature | Higher-Tflash C6 analog MD 19.92 vs exp 19.1 = 4% AAD | MD absolute accuracy for carbonate family |
| 4 | NIST ThermoML ρ anchor (n=24 exp) | Literature | Higher-Tflash C6 analog MD 737 vs NIST 720.5 = 2.29% AAD | MD density accuracy (strongest anchor) |
| 5 | NIST ThermoML μ anchor (n=6 exp) | Literature | Higher-Tflash C6 analog MD 0.342 vs NIST 0.306 = 11.7% AAD | MD viscosity within reasonable bounds |
| 6 | openCOSMO-RS γ∞ (10 pairs) | Independent thermodynamics | All 4 sanity-check LLEs pass (C6 sulfolane reference γ∞ = 21.14, etc.) | Miscibility framework independently validated |
| 7 | openCOSMO-RS γ(x,T) surface (5 pairs × 54 grid) | Independent thermodynamics | Marangoni champion B (FDA-GRAS) UCST = 308.8 K (matches Green Chem 2025 isooctane analog) | Explains Marangoni champion B (FDA-GRAS) kinetic-vs-equilibrium stability |
| 8 | MACE-OFF23 foundation ML force-field (25 dimers) | Independent force-field | E_bind −0.33 to +1.84 kJ/mol (median +0.02) | Binding energies physically correct for weak-interaction dimers |
| 9 | T12 OPLS-AA cross-check (n=3 pool, 5-ns extension) | Alternate force-field | Dielectric-safety reference pool Welch's t = −0.29 vs GAFF2 n=7 | Force-field independence for σ |
| 10 | Hansen Ra miscibility screen v2 | Semi-empirical | 180 pairs screened; killed 5 false-miscibility MD pairs | Pre-wetlab shake-flask gate |
| 11 | Chemprop D-MPNN + RF σ predictor | ML surrogate | Published R²=0.794 MAE=5.5; 2026-04-19 retrain R²=0.279 (small dataset) | ML predictions converging but small-data-limited |
Gaps honestly disclosed: Rank-3 amide-pump candidate MD ρ 13.7% error (GAFF2 polar-amide LJ undertuning, σ unaffected); OPLS-AA 1-ns scatter 5.65× GAFF2 (5-ns extension firms variance); Chemprop R² 0.279 on n=10 (insufficient data — RF proxy + parachor + MD remain primary). Full synthesis, per-pair data, and primary-source anchors in the NDA-gated buyer dossier.
Claim Strategy.
Three inventor-prepared specifications layer composition, system, method, microgravity, retrofit, and blocker-style embodiments. Buyer diligence still needs USPTO receipts, application numbers, exact filed PDFs, ownership proof, and patent-counsel review of claim scope and filing path.
Fluid + System + Method (fluorinated)
Fluorinated binaries (HFO, HFE, Novec-649 class) — legacy PFAS-containing embodiments. Claims 10/11 are Δσ ≥ 3/5 mN/m dependent claims (not microgravity — PROV-003 Claim 10 is microgravity). Gen-2 continuation Claims 118–122 add TF-alkylamine homologs, "heartbeat" resonance method, smooth-surface requirement.
Computational Discovery Engine
Bayesian GP + MD slab simulation methodology for binary cooling-fluid discovery. Method claims; platform IP separable from composition.
PFAS-Free Binary Keystone
PFAS-free binary genus (Δσ ≥ 3, PFAS-free, ΔT_b ≥ 30°C). Claim 10: microgravity method. Claim 11: retrofit method. Claim 13: Dielectric-safety reference (PROV-003 covered) with a claimed composition range. Claim 56: self-rewetting (a claimed long-chain alcohol additive). Claim 57: dissolved gas (CO₂/N₂/Ar). Claim 58: quad-Marangoni capstone. 10 blocker-style embodiments. **CIP-draft addendum (counsel-editable, 2026-04-19) adds Claims 11–14 for the integrated system stack: copper-foam wick, 40 kHz ultrasound, dielectric nanoparticle dispersion, biphilic-textured surface — forward-looking §112(a) enablement-flagged. See `docs/CLAIM_STACK_ADVANTAGE_MATRIX_2026-04-20.md`.**
Two Post-Filing Candidates.
Marangoni champion A (4.06× FoM4 screening score) and Marangoni champion B (FDA-GRAS) (3.83×) were discovered by post-filing Bayesian GP + MD. Neither component (the lactone pump, PC) appears in the inventor's current PROV-003 pump genus. They are follow-on filing candidates, subject to miscibility, pentane-safety, and counsel gates.
The Blocker Wall.
Ten explicit blocker-style embodiments in PROV-003 are claim concepts for counsel review; actual scope, patentability, and freedom to operate are not represented here as settled.
Validation Path.
A licensee underwriting this plan gets first-look rights on every data point produced. The protocol intentionally starts with n-pentane handling qualification and miscibility/phase mapping before any performance interpretation. Protocol: docs/WETLAB_SOP.md.
Reproduce Before You Sign.
# CPU image, no GPU required — 5 min on any laptop
git clone https://github.com/cooling-fluid-10/cf10
cd cf10
docker build -f Dockerfile.cpu -t cf10:cpu .
docker run --rm cf10:cpu check-only
→ outputs/sigma_check.md PASS (|Δσ| < 1.5 mN/m vs canonical)
Full MD reproduction: docker run --gpus all cf10:gpu md_top5
— < 3 hrs on A100, < 24 hrs on CPU. REPRODUCIBILITY.md has the full subcommand
reference (setup, md_top5, fom_recompute, sigma_check, figures, report, check-only, full).
Phase Budget
Docker reproducibility check
Counsel engagement / follow-on decision
Pentane handling qualification + EHS signoff
Miscibility / phase mapping · 10 pairs × 2 T
Pendant-drop σ · only after safety + phase pass
σ(T) sweep · Dielectric-safety reference (PROV-003 covered) + follow-on candidates
Marangoni bench demo · qualitative
Dielectric + flash-point + 30-day stability
Tests A–K
CRO confirms Class IB flammable handling, sealed vials, hood/rated enclosure, grounding, waste plan.
Every top-10 blend, 72-hr phase observation.
Same set repeated at elevated T; two-phase results demote direct-immersion candidates.
Independent wetlab σ vs MD prediction only after safety and phase gates pass.
Dielectric-safety reference (PROV-003 covered) 20/30/40/50 °C; compare to MD Eötvös fit.
Visual confirmation of composition-gradient flow.
Foam vs bare coupon at constant heat input.
Flammability class and sealed-system constraints documented before buyer pilot.
Sealed vials stored under ambient; check for separation / color change.
Span 80 stabilized fallback for Hansen-flagged pairs.
Pass/fail for low-ε Dielectric-safety reference (PROV-003 covered) at typical rail V.
Safety fail = stop. Two-phase = demote. σ mismatch = fail MD. Buyer pilot waits for ε, flash, seal, and CHF/HTC gates.
Enhancement Stack Tests.
After the $5k σ-validation campaign passes, the next tier is a boiling-rig pilot that exercises the Claim 8 hardware enhancements stacked on top of the best fluid. Reference specs in docs/reference/FINAL_EXPERIMENTAL_PLAN.md §Phase 4.
γ-Butyrolactone is a List I chemical precursor under 21 CFR 1310 (DEA Chemical Diversion & Trafficking Act — not a Schedule I narcotic; those are separately defined under 21 U.S.C. §812). US buyers need registered-purchaser status; EU buyers face no equivalent restriction (REACH-registered industrial solvent). APAC varies by jurisdiction. Region-sensitive handling penalty is modelled in scripts/risk_adjusted_score.py (US/EU/APAC columns). If US procurement is blocked, alternatives are evaluated only after miscibility and safety gates. See docs/WETLAB_SOP.md §2.1 + docs/audit/UQ_STRATIFIED_ANALYSIS.md.
Beyond the MD + UQ + Pareto + Butler-ceiling evidence stack, four independent industry-standard mixture-σ methods were computed on all 59 pairs as a CCE-diligence cross-check:
Plus LVPP sigma-profile database (2500+ molecules, open NWChem-COSMO, 29/37 coverage of our components) + ChemBERTa-class Hansen-SMILES ML predictor (LOO MAE ~2 MPa^½) + DDB literature sweep (*Green Chemistry* 2025 risk-flags Marangoni champion B (FDA-GRAS) immiscibility; Mudawar 2018 pool-boiling CHF citations support hypotheses only, not measured performance on these blends).
Net effect: the originally-estimated ~220 A100-hour internal computational roadmap
was reduced to ~30 A100-hours (~85% compute-cost reduction) without loss of diligence rigor.
Full enumeration: docs/COMPUTATIONAL_ROADMAP_V2.md +
outputs/sigma_method_comparison.md.
Risks & Mitigation.
Every deep-tech asset ships with open questions. Ours are enumerated with specific kill-tests assigned to the wetlab plan. The first hard gates are contractor pentane-safety qualification and miscibility/phase mapping; performance claims wait until those gates pass.
Follow-on candidates may be two-phase at ambient
Post-session Hansen v2 rescreen flagged both lead Marangoni candidates above the Hansen immiscibility threshold. MD at the canonical composition is single-phase on all pool seeds, but 20/40 °C wetlab shake-flask is the decisive test. Fallback candidates (amide-pump rank-3, cyclic-ketone mid-flash) remain wetlab-gated.
Stacked CHF uplift not yet measured
Literature-additive ceiling ≈ 8× vs bare single-component; conservative floor 2–3×. Cross-coupling among the seven claimed enhancements (foam · ultrasound · nanoparticles · biphilic · sealed · microgravity + base Marangoni) is uncharacterized. Phase 5 pool-boiling pilot (Q1–Q2 2027) is the canonical measurement; until then the stack is disclosed as claimed + literature-supported, not measured.
Lead Marangoni pumps absent from current PROV-003 genus
Two lead post-filing pumps are absent from PROV-003 Claim 1(b). Follow-on utility-application draft is counsel-editable with the 2027-01-29 benefit-claim hard-bar as the target. Claims 11–14 of the follow-on draft (integrated-system stack) are forward-looking defensive claims with §112(a) enablement risk flagged for counsel review.
C5 alkane blends require sealed thermosyphon geometry
Pentane vapor pressure exceeds 1 atm at 50 °C, so data-center direct immersion requires sealed thermosyphon / loop heat-pipe / vapor-chamber geometry (explicitly covered by PROV-003 Claim 8). Higher-flash alternates (hexane, cyclopentane, heptane) are available in the portfolio for deployments where sealed geometry is undesirable. NFPA 75 classification and gasket/seal compatibility scheduled in Phase 3.
Methods and bias honestly disclosed
GAFF2 known-drift at high-alcohol compositions moved to known_drift_excluded in the canonical inventory. Force-field independence confirmed across GAFF2 + OPLS-AA + MACE-OFF23 (Welch's t = −0.29 on the dielectric-safety reference pair). Additional honest gaps — experimental σ anchor, ρ anchor, openCOSMO-RS γ(x,T) surface, Hansen miscibility screen — live in the NDA-gated diligence room. Seven other second-order risks (procurement regimes, ε measurement precision, sulfolane SVHC restriction, σ pool drift, etc.) documented in the buyer dossier.
Market Waterfall
Leverage Curve
Ahead of the Meeting.
Why hasn't anyone done this before?
Four enabling conditions converged only in 2025: (1) 3M's PFAS exit opened the demand hole; (2) ECHA universal PFAS restriction created the deadline; (3) high-TDP accelerators (B200, MI325X, Rubin) forced two-phase adoption; (4) cloud GPU availability (RunPod, A40/L4) made 500-GPU-hour screening affordable for a single inventor. Before all four were simultaneously true, the opportunity wasn't there.
Why not just use water?
For indirect cooling, water is genuinely better on raw thermal (2× our k, 2× Cp). Our edge is pump elimination — the self-pumping fluid replaces the loop pump, saving ~$20K/rack TCO and ~$40/server/yr pumping power. For direct immersion, water is conductive and corrodes electronics — disqualified on the basics.
What is the actual proof it works?
Computational: 172 GROMACS σ values across 59 canonical pairs, OpenFOAM interFoam exploratory flow fields, CP2K DFT binding energies, 52-URL-validated NIST FoM4 screening scores, Hansen Ra miscibility screens, ML σ predictor (R²=0.794, sigma_predictor.pkl), Chemprop D-MPNN cross-checks, xTB and MACE-OFF23 single-point checks, Bayesian GP + MD discovery loop (6 post-filing hits). Phase 2 closure (2026-04-18): our lead Marangoni direct-immersion candidate n=7 UQ pool + 5-ns outlier relaxation, multi-seed σ(T) Eötvös R²=0.68, OPLS-AA point-estimate cross-check, Cu-slab contact-angle literature fallback, 56-reference FTO preview. Experimental: pending. The wetlab plan first tests pentane safety and miscibility, then pendant-drop σ, dielectric, flash point, stability, and later CHF/HTC.
Where is the catch?
Transparent limitations are the stress test. GAFF2 known-drift for MeOH-rich compositions moved to known_drift_excluded section of canonical inventory (2026-04-18). a higher-Tflash indirect-loop variant σ(T) has a non-monotonic stretch (unphysical in that window). Original 12 OpenFOAM cases used hardcoded σ/μ/ρ (re-derived analytically; 6 survived). Hansen audit killed 5 MD-simulated pairs as force-field miscibility failures. Cu-slab contact-angle MD diverged with INTERFACE-FF Heinz-2008; accepted literature fallback (Hautman 1991 + Schrader 1995) — wetlab goniometer is the authoritative path. An asset that discloses its known bugs is diligence-grade.
How defensible is the IP really?
The inventor-prepared PROV-003 specification claims a PFAS-free genus (Claim 1: Δσ ≥ 3 mN/m) plus specific compositions, system integration, method, microgravity (Claim 10), retrofit (Claim 11), quad-Marangoni capstone (Claim 58), and blocker-style embodiments. The counsel-editable CIP draft adds an integrated-system stack (Claims 11–14: copper-foam wick, 40 kHz ultrasound, dielectric nanoparticles, biphilic heat-surface — §112(a) enablement-flagged). Actual filing proof, claim scope, patentability, and FTO remain counsel-pending. PROV-001 covers fluorinated legacy concepts. PROV-002 covers discovery methodology. our lead Marangoni direct-immersion candidate / the FDA-GRAS cyclic-carbonate secondary candidate are follow-on filing candidates, not settled issued rights.
Why license instead of build?
Manufacturing specialty fluids requires a chemical plant, regulatory overhead, distribution, and sales — none of which is faster to build than licensing to someone who already has them. Time-to-market matters: the ECHA restriction window is 2026–27, not 2030. A licensee with Perstorp- or Chemours-scale infrastructure ships in 18 months; a start-up takes 5 years.
What does the $5k wetlab actually buy?
(1) Contractor qualification for n-pentane-rich Class IB flammable mixtures. (2) Miscibility/phase mapping that confirms or demotes follow-on candidates. (3) Pendant-drop σ for pairs that clear safety and phase gates. (4) Optional σ(T) for our low-ε direct-immersion reference (PROV-003 covered) and follow-on candidates. (5) Dielectric, flash-point, and 30-day phase-stability data. (6) Decision packet input for counsel-led follow-on filing. CHF/HTC remains a later buyer-pilot gate.
Is any of this reproducible?
Yes. `docker run --rm cf10:cpu check-only` runs in under 5 min and produces `outputs/sigma_check.md` with PASS/FAIL against canonical (±1.5 mN/m tolerance). Full MD reproduction: `docker run --gpus all cf10:gpu md_top5` — under 3 hrs on A100. REPRODUCIBILITY.md is 309 lines with every subcommand. CI runs the check on every PR. Nothing is hand-waved.
Roadmap.
Every external-facing claim on this site is anchored to one of the landed milestones below. Future milestones define the valuation-step gates on the Leverage Curve — intentionally front-loaded so Phase 3 wetlab and the follow-on filing decision complete before the 2027-01-29 hard-bar.
Provisional specifications claimed
234 claims across three inventor-prepared provisionals. Filing receipts and ownership proof live in the NDA-gated data room.
Computational portfolio complete
2,610 ML-screened compositions · 200 ranked · 180 Hansen miscibility screens · 59 canonical MD-measured pairs · 172 σ slabs · 11 independent cross-check methods.
Follow-on filing draft (counsel-editable)
22-page follow-on utility-application draft with claims for the integrated 7-component enhancement stack. §112(a) enablement risk flagged for counsel review.
Phase 3 wetlab — σ · miscibility · ε · flash · stability
40 pairs across Tier A full-spec (top-20) + Tier B fast-screen (20). Combined ~$5.5k CRO budget. Converts computational rankings to physical-measurement-anchored claims.
Counsel engagement + follow-on filing
Patent-counsel review of follow-on draft, claim-scope finalization, and conversion filing before the 2027-01-29 benefit-claim hard-bar.
Phase 5 pool-boiling pilot — stacked CHF measurement
University or CRO pool-boiling lab exercises the full integrated system (fluid + copper foam + ultrasound + nanoparticles + biphilic). Canonical measurement of the 2–3× conservative-floor / 8× literature-ceiling CHF uplift.
Licensee pilot · rack-scale deployment
First buyer-funded rack pilot in a data-center or EV-battery-thermal context. Long-term stability and manufacturing-scale validation.
The repo contains inventor-prepared specifications identified as PROV-001, PROV-002, and PROV-003 with a claimed filing date of 2026-01-29. Buyer diligence still requires USPTO filing receipts, exact filed PDFs, application numbers, and ownership evidence in data_room/legal/. Post-filing computational refinement should not be equated with claim scope. Marangoni champion A, Marangoni champion B (FDA-GRAS), and other post-filing discoveries are follow-on filing candidates, subject to wetlab results and patent-counsel review. FoM4 is a Novec-normalized screening score, not a measured cooling-performance multiplier.
Commercial Access.
Parallel non-exclusive licenses across channels. No single buyer locks the portfolio. All paths gated by signed NDA (template at docs/NDA_TEMPLATE.md · counsel review pending).
Fluid Manufacturer
Chemours · Engineered Fluids · Perstorp · Solvay
Standard royalty structure on net fluid revenue.
Cooling OEM
GRC · Submer · LiquidCool · ZutaCore · Iceotope
Fluid + Surface IP package. Per-server royalty on deployed racks.
Hyperscaler
Microsoft · Google · Meta · AWS
They don't license — they acquire. Price depends on TRL stage.
Defense / Space
Northrop · L3Harris · NASA · DARPA · AFRL
SBIR + direct-procurement. Domestic-supply narrative + microgravity claim (PROV-003 Claim 10).
Walk-away floor: no deal below $500K upfront + 3% royalty. Below that, legal fees exceed net return and we keep the IP.
Initiate Technical Diligence
Request
The Dossier.
NDA-gated. Full licensee dossier, legal proof manifest, follow-on filing packet, data room access, and inventor technical consultation available for diligence.