Exploratory research · numerical methods and results

✨ Plasma simulation

Four spherical pulse calculations follow converging hydrogen and CNO mixtures through compression, nuclear burning and expansion. These plots expose the simulated fields and numerical checks behind the ordinary-hydrogen fusion proposal.

Assessment: these are internally consistent one-dimensional burn calculations under prescribed initial conditions. Producing those conditions, retaining a coherent implosion in three dimensions, and validating the radiation model remain consequential open questions. The calculated gains and plant output are conditional on resolving them. This reactor concept has no experimental validation.

Initial conditions · Pulse histories · Numerical checks · Unresolved failure mechanisms · Reproduce the plots

What is being solved

The solver divides a sphere into Lagrangian mass shells. Shell boundaries move under pressure forces, including artificial viscosity for shocks. Cell energy changes through pressure work, nuclear reactions, a modeled weak-energy debit and grey flux-limited radiation diffusion; the same pressure work couples the kinetic and internal energy accounts. A 27-isotope, 194-rate network evolves local abundances. The equation of state combines ideal ions, tabulated electrons and positrons, and radiation in local thermodynamic equilibrium (LTE). The incoming profiles have inward velocities, so the calculation follows their spherical convergence and rebound. The earlier preparation of that inflow is modeled separately and remains a major open problem.

In continuum notation the underlying balances have the forms ∂r/∂t = u, ∂u/∂t = −4πr² ∂(P + q)/∂m, and ∂e/∂t = −(P + q) ∂(1/ρ)/∂t + q̇_nuclear − q̇_weak − ∂L/∂m, with local reaction-network equations for the abundances. Here m is enclosed mass, q is artificial viscosity, and L is radiative luminosity. The q̇ terms are specific energy rates. The center has zero velocity and radiative flux; the outer surface has zero imposed external pressure and an approximate outward radiative flux. The source contains no self-gravity term. The momentum and pressure-work equations use Newtonian bulk dynamics. These equations summarize the model; the implementation defines its discretization and boundary conditions.

The baseline assumes spatially constant grey opacity κ = 0.001 m²/kg. It uses a single material temperature and equilibrium radiation; it does not resolve photon spectra, non-LTE thermalization, separate ion and electron temperatures, magnetic fields or multidimensional flow. Ion Coulomb corrections, nuclear screening and electron conduction are also omitted. The methods note describes the constraints and additional cases.

Selected pulse inputs and outcomes

The selected family uses the actual 128-zone profiles and median carbon-rate update. The time origin is each hot burn's saved initial state. The radius parameter of 30–139.25 m sets the mass and normalization; the actual initial outer radii are 170–888 m. The full radial profiles are in the linked input files. These are separate shots with different masses and starting states.

StageH / catalystMass (kg)Initial energy (J)Duration (µs)Hot gainEnergy residual / input
Nitrogen-1431.131×10¹³3.493×10²⁵68.39510.765−2.68×10⁻⁷
Nitrogen-1521.131×10¹³3.196×10²⁵381.9007.729−1.46×10⁻⁵
Carbon-1221.131×10¹³4.074×10²⁵63.0639.919−3.95×10⁻⁶
Carbon-1321.131×10¹³3.328×10²⁵70.00422.660−2.48×10⁻⁶

Hot gain is deposited nuclear energy, after the solver's weak-energy debit, divided by initial hot-pulse internal plus kinetic energy. The hot solver assigns the entire positive Q of every weak reaction to escaping energy. This is a conservative hot-heating convention rather than a neutrino-spectrum calculation. Electrical preparation, recovery and auxiliary loads enter the separate plant account. The nitrogen-15 precursor has a slightly higher energy than its hot start because it radiates during preparation; that separate debit is included in the integrated plant account. All four integrations reached their requested end times. The residual compares the final energy ledger with the initial energy; conservation alone does not demonstrate spatial convergence.

Three panels show the prescribed initial density, temperature and inward velocity versus enclosed mass fraction.
Prescribed initial state. Temperature and density are cell quantities; velocity is at the cell’s outer boundary. The calculation starts with fuel already hot, dense and converging. The fields are boundary conditions for the burn calculation. Preparing them from cold packets requires a separate, successful physical process.

The initial mass-weighted temperatures are 23.1, 24.0, 22.4 and 22.6 million K for N14, N15, C12 and C13. Their mass-weighted densities are 1.85×10⁶, 1.46×10⁴, 1.87×10⁶ and 1.85×10⁶ kg/m³, and their maximum inward boundary speeds are 4.25, 4.11, 4.64 and 4.17 million m/s. About 91–93% of initial energy is bulk kinetic energy. The initial energy denominator includes those reservoirs; a preparation-efficiency assumption does not demonstrate an apparatus capable of producing the spatial profiles.

Pulse histories

All four curves below use the selected traces. Each contains 400 saved samples; the adaptive integrator takes its own time steps. The samples select accepted solver states by index and have unequal time spacing. Narrow peaks can fall between saved samples. Mass-weighted quantities are ∑mᵢxᵢ/∑mᵢ. Local extrema and volume averages can differ substantially. Each plotted line joins its saved samples. N14 CSV · N15 CSV · C12 CSV · C13 CSV.

Half-mass radius versus microseconds for four pulses, each contracting and then expanding.
1. Implosion and rebound. Radius enclosing half the pulse mass, interpolated within each constant-density shell. The inward flow compresses before pressure and burning drive expansion. This radius is distinct from the more dilute outer surface.
Mass-weighted density on a logarithmic scale versus time for the four pulses.
2. Density. The logarithmic axis makes the lower-density nitrogen-15 pulse visible alongside the other stages. Density is reconstructed from each shell's mass and volume at each saved time.
Mass-weighted temperature versus time for the four pulses, rising during compression and falling during expansion.
3. Temperature. Mass-weighted temperature in billions of kelvin. Local maxima can be higher; the downloadable histories include the maximum cell temperature at every saved time.
Cumulative deposited nuclear energy normalized to initial pulse energy versus time, for four pulses.
4. Nuclear deposition. The plotted quantity is cumulative nuclear energy less the modeled whole-Q weak debit, divided by initial hot-pulse energy. Photon escape, delayed radioactive decay, driver losses and the receiver are separate accounts.
Four panels show the changing principal isotope inventories in each pulse, normalized to the initial catalyst count.
5. Isotope evolution. Inventories are mass-integrated nuclei per initial catalyst nucleus, during the hot pulse and before subsequent cold decay. In the nitrogen-15 panel the carbon-12 and helium-4 curves nearly overlap. The carbon-13 pulse produces both N14 and O15; the latter subsequently decays to N15. Minor channels and unburned hydrogen remain in the full endpoint records.

Tests and sensitivities

Carbon-12 pulse gain decreases from 10.2719 at 64 zones to 9.9176 at 128 and 9.8671 at 256.
6. A mesh comparison. Carbon-12, H/C = 2, prescribed-profile updated-rate comparison: 64, 128 and 256 zones give 10.2719, 9.9176 and 9.8671. The 128-to-256 change is −0.51%. This is a supporting comparison for one stage, not an error bar on all four selected source profiles or on the plant. The selected source-mapped 128-zone carbon-12 run has gain 9.9187. Exact comparison values and hashes; complete records at 64, 128 and 256 zones are included in the reproduction package with their initial profiles and inputs.
Four panels compare hot-pulse gain against hydrogen-to-catalyst ratio for different composition constraints and resolutions.
7. Fuel composition. Pale solid curves hold prepared energy fixed at 64 zones; dashed curves hold the initial flow profile fixed and allow energy to change. Blue points use 128 zones and updated carbon rates, so their differences combine mesh and rate changes. Lines connect tested points. Gain is a pulse quantity, and higher gain does not by itself imply the best closed plant. Plot data · plot code.
Plant specific power versus receiver thickness and service equipment mass assumptions.
8. Plant assumptions. These downstream comparisons combine the pulse family with a receiver and equipment allocation. Hot graphite thickness and service mass change the predicted net output per installed mass. At 350 kg/m² receiver and 1,000 kg/m² service allocation, the selected plant account gives 277.68 mW/kg. The service curves are conditional scenarios, not confidence bounds. Plot data · plot code · allocated reference · catcher-interface account.
Four panels track kinetic energy, internal energy, escaped photons and deposited nuclear energy, all divided by initial pulse energy.
Energy partition. Deposited nuclear energy raises the internal and bulk kinetic reservoirs, while radiation leaves the simulated plasma. All quantities use the same initial hot-pulse energy denominator. Escaped photons can still heat an external receiver; this boundary is the plasma surface.
Left: LTE radiation's fraction of mass-mean thermodynamic pressure over time. Right: reconstructed energy residual in parts per million of initial input.
Radiation dependence and numerical closure. The left panel is a result of the imposed LTE equation of state. The right panel reconstructs the energy account from saved fields and cumulative channels using the original EOS, comparing K + U + escaped photons + weak debit with initial energy + nuclear release. The largest absolute residual is 4.03×10⁻⁶ of input. This is an accounting check within the model; it supplies no validation of the LTE or opacity assumptions. Audit results, definitions and hashes.

The principal comparison uses relative tolerance 10⁻⁷–10⁻⁶ and absolute tolerance 10⁻¹²–10⁻¹¹, with artificial-viscosity coefficients 0.5 and 2.0. A 2.8 GK numerical/EOS guard terminates incomplete runs, which are excluded from comparisons. The BDF integrator applies those tolerances to normalized state variables. The selected quartet’s endpoint conservation residuals are in the table above.

On the saved states, all four EOS reconstructions remain inside the supplied table’s temperature domain of 10⁵–3×10⁹ K and ρYₑ domain of 10⁻⁶–10⁷ g/cm³. Table coverage establishes interpolation coverage, while physical accuracy requires EOS validation. The underlying electron/positron EOS includes relativistic and degeneracy effects; the bulk hydro remains Newtonian. Saved boundary speeds reach 0.044–0.081c across these cases. Relativistic bulk/radiation-transport corrections have not been quantified for this quartet.

A related quartet gives a nominal 256-cell plant result 0.56% above its 128-cell result, with a longer carbon-return outflow. That comparison retains C13 H/C = 4 and different source profiles; it is supporting evidence rather than convergence of this selected quartet.

The carbon update replaces two proton-capture pairs using the median update associated with the LUNA carbon measurements; the remaining network derives from REACLIB. The implementation applies the same update factor to each forward and reverse pair, preserving the original library equilibrium ratio. This policy is not an independent recalculation of detailed balance from nuclear partition functions. The network includes 27 species up to magnesium and omits deuterium, tritium, helium-3 and carbon-14. The nuclear review documents rate provenance, screening and omitted-channel checks on related histories; it does not qualify every omitted process on these exact final profiles. Rate alternatives, opacity, EOS and source profiles affect results; the tested variations do not provide a comprehensive probability distribution for reactor performance. In particular, the published 80% Monte Carlo interval for the receiver transport reflects sampling at fixed assumptions, not overall reactor uncertainty.

Unresolved failure mechanisms

The fuel assembly simulation follows the nitrogen-15 pulse from pellet release through a conditional merger and burn handoff, with a provisional 90% yield-retention criterion.

The following questions can overturn the claimed reactor performance. The numerical checks above do not answer them.

For the selected runs, photons crossing the plasma surface during the saved interval carry 0.0743%, 7.675%, 0.0191% and 0.1160% of initial-plus-deposited energy for N14, N15, C12 and C13. These fractions follow the assumed grey opacity and LTE closure. They describe the recorded interval; radiation still trapped at its endpoint can escape later. They are not a measurement of total plant heat capture.

Code, data and reproduction

Download the compact reproduction package (code, records, CSVs and input manifest; raw traces downloaded separately). It reproduces all eleven figures with a single driver and creates the output directory. The four original raw traces remain unchanged. Frozen input hashes bind the exact records and EOS table; the original stage map links the quartet to the research archive.

unzip reproduction.zip?v=749f2a4869b4 -d plasma-simulation
cd plasma-simulation
python3.10 -m venv .venv
. .venv/bin/activate
python -m pip install -r code/plot-requirements.txt?v=74bdee5c260e
for stage in n14 n15 c12 c13; do
  curl --fail --location --output "data/$stage.npz" \
    "https://tkwa.me/blog/ordinary-hydrogen-fusion/plasma-simulation/data/$stage.npz"
done
python code/build_plots.py?v=bd7a2ea02878

The tested environment is CPython 3.10.13 with NumPy 1.26.4, SciPy 1.14.0 and Matplotlib 3.9.0. The listed binary versions predate Python 3.14 support. The driver checks input hashes before reduction, trace layout, completion times, masses, initial profiles, endpoint abundances and energy channels. It then regenerates the figures, expanded CSV histories and audit-results.json. The closure calculation uses integrated nuclear energy; the original endpoint residual uses isotope rest-mass changes. Their residuals differ by at most 6×10⁻¹¹ of initial energy for this quartet.

The reproducible step here is reduction of saved runs. A fresh integration of the exact quartet in an independently assembled environment, comparison with an independent hydrodynamics implementation, and experimental validation remain outstanding. Successful plotting and hash checks establish data identity and numerical consistency only.