The car that ran on its battery
A 1992 patent describes releasing a fuel gas from water and names a car as a place to burn it. We simulated, step by step, a car whose only fuel is gas split from water on board. Parked, each mole of gas it burned made 0.13 mol of new gas. On the road it stopped for good 0.8 km into ten drive cycles, once its starting battery ran down. With every part loss-free, the loop returns exactly what the split cost, never more.
The verdict
Nothing there
In a step-by-step simulation, a parked car whose engine only ran the electrolyser made 0.13 mol of new gas for every mole it burned, and with every step loss-free the loop returns exactly what the split cost, never more. On the road, fuelled only by gas from water, the car stopped for good 0.8 km into ten drive cycles, once its starting battery ran down.
Read it with the misses. Three preregistered items in the simulation failed (listed below), so under the lab’s own rule the simulation cannot carry the verdict alone. The verdict rests on the closed-form checks, which all passed, and the simulation agrees with them. The run’s RESULTS proposed nothing there, and on 2026-10-18 the lab’s verdict engine set the same verdict.
Parked, with the engine doing nothing but driving the alternator for the electrolyser, the car burned its 2 mol of starting gas in 21 s and made 0.31 mol back: 0.133 for every mole burned.
Splitting water takes at least 237.1 kJ of electrical work per mole of hydrogen, and the most work any device can get back from burning it is the same 237.1 kJ. A perfect loop returns exactly 1.0.
The car stalled 60 s into the drive cycle, after 0.10 km, then crept on in short restarts as its battery made a little more gas each time. It stopped for good at 0.78 km, when the battery ran down.
With a cell planted at about 1000 % efficiency, the parked loop makes 1.33 mol per mole burned and its gas store grows from 2 to 24 mol in 600 s. With realistic parts it makes 0.133.
The claim, as presented. The patent presents a process that releases a fuel gas from water and names an automobile as a place where the gas is burned.
US patent 5,149,407 (Stanley A. Meyer), “Process and apparatus for the production of fuel gas and the enhanced release of thermal energy from such gas”, September 1992 · the episode
The registry’s note on sources. Read from the source named here by a research subagent (2026-10-10); paraphrase, not quotation. No video was watched.
Re-read before the run, 2026-10-12. The patent presents a process that splits water in a capacitive cell driven by pulsed, resonant voltage, then energises and burns the gas, and presents the burn as releasing more heat than burning the gas in ordinary air would. Its claims 11 and 12 name a combustion site in an automobile and in a jet engine. It gives drive values (a 26 V pulse at 10 kHz) and no measured energy balance. The test is of the episode’s question: can splitting water and burning the gas return more energy than the split cost?
Preregistered in PREREG.md: the original (commit 43bc608) before any code; Addendum 1 (005158a) and its clarification A1.7 (bc904dc) before the simulation code; Addenda 2 (1e9cb6f) and 3 (ba86c5f) after runs, declared as such, each before the run it governs. Self-graded, with no outside review. Lab files are listed at the foot of the page.
The car
A 1200 kg car on the NEDC, the standard European drive cycle. Its engine runs a lean fuel-air cycle at compression ratio 10 on the gas it receives, with combustion and expansion computed at chemical equilibrium (Cantera). The engine drives the wheels and a 5 kW alternator, which powers a 12-cell electrolyser at 98 % Faradaic efficiency, its voltage the reversible potential plus activation and resistive losses.
The gas goes to a store that feeds the engine. A 720 Wh starting battery helps the electrolyser, and the store starts with 2 mol of hydrogen, as if made overnight. Every choice that could favour the claim was made generously: no accessory load, a large alternator, and the battery never diverted to charging.
The simulation’s checks
Written down before the simulation code, except the two planted faults marked as added later. Each planted fault had to turn its check red once; the road run’s did not, even after a stronger one was added.
| Check | Result | Numbers |
|---|---|---|
| Drive-cycle distance against the published cycle | missed | 10.798 km; gate 11.0 ± 0.1; published 10.931 km |
| Flame temperature, hydrogen in air | pass | 2380 K (gate 2350–2430) |
| Engine cycle efficiency | pass | 0.457 ideal cycle; 0.365 indicated |
| First law, both runs | pass | 1.5×10−13 on the road; 2.7×10−15 parked |
| Second law, part by part | pass | 0 violations |
| Time step 0.05 against 0.1 s | missed | road distance moves 1.46 % (limit 1 %); parked loop 1×10−15 |
| Simulated loop inside the closed-form range | pass | 0.133, inside 0.077–0.312 |
| Parked loop below 1 | pass | 0.133 |
| Road run fails ten cycles | pass | 0.78 km of 108 km; made per burned 0.776 |
| … with a cell planted at about 1000 % | not fully red | flagged, yet the car stalled at 14.1 km |
| … with a cell at about 3000 % (added later) | not fully red | flagged; stalled at 44.3 km |
| Parked, with the 1000 % cell (added later) | red, as it should be | 1.33 made per burned; store 2 → 24 mol in 600 s |
| Alternator delivering 1.3× its booked output | red, as it should be | first-law residual 9.2×10−3 |
| Engine at 120 % of the fuel’s heat | red, as it should be | second-law check flags the engine |
| An audit that leaves out kinetic energy | red, as it should be | first-law residual 2.1×10−2 |
The misses. The drive-cycle table is 1.2 % short of the published cycle, in two phases drawn as straight lines. The road distance moves 1.5 % with the time step, because it is a sum of stall-and-restart lurches whose timing depends on the step; the parked loop does not move. And the planted over-unity cell did not turn the road run red: the 5 kW alternator cannot keep up with the cycle’s high-speed section, so even a 3000 % cell stalled. So the 0.78 km shows how this car fails, and cannot by itself show that no car could work. The parked loop and the thermodynamics carry that.
The calculation behind the verdict
Per mole of hydrogen at 25 °C, splitting liquid water needs at least 237.1 kJ of electrical work (1.229 V). The other 48.7 kJ of the 285.8 kJ total can come in as heat. Burning the gas gives 285.8 kJ of heat, and the most work any device can draw from that at room conditions is again 237.1 kJ. So a perfect loop returns exactly what it cost, and a real one returns less.
A correction to the lab’s own draft. The draft gate compared 237 kJ of work with 286 kJ of heat, which suggests the loop could gain 20 %. That compares work with heat. A planted copy of the mistake reads 1.205, and the bound checker catches it.
| Check | Result | Numbers |
|---|---|---|
| Work to split water, from three routes | pass | 237.140 kJ; the others off by 0.021 and 0.010 kJ |
| A perfect loop | pass | returns exactly 1.0 |
| Loop return over the realistic range of parts | pass | at most 0.312; central 0.108; at least 0.077 |
| … with a cell planted at 600 % | red, as it should be | 1.010, flagged by the bound checker |
| … with heat counted as work | red, as it should be | 1.205, flagged |
| … with a typo planted in the data | red, as it should be | off by 30.6 kJ |
| … with the cell’s range stretched to 3.5 | red, as it should be | 1.285: the gate can say yes |
| Central parts (cell 0.64, alternator 0.55, engine 0.30) | Most generous allowed (0.85, 0.80, 0.45) | |
|---|---|---|
| Loop return | 0.108 | 0.312 |
| Cell efficiency needed | 5.9× the thermodynamic limit | 2.7× |
| Cell voltage that implies | 0.21 V | 0.45 V |
| Or: heat the gas would have to release, against its chemical heat | 9.3× | 3.2× |
Below 1.229 V, water does not split at 25 °C with all the current making gas. If the gas released extra heat while ending up as water again, Hess’s law says the extra came from an input, and the patent’s ionising stages are electrical.
- The patent’s apparatus itself: the resonant cell, the ionisation and the lens stages. The calculation bounds what any apparatus would have to achieve.
- Whether the cell makes more gas per coulomb than Faraday’s law allows. The patent reports no measurement either way.
- Add-ons that feed electrolysis gas into a petrol or diesel engine to save fuel. That is a combustion-aid claim with no closed loop, and it needs its own episode.
Read: US patent 5,149,407 (re-read 2026-10-12). Prior art: textbook thermodynamics; Novella, NeuroLogica Blog (2008), on the Genepax car. No novelty claimed.