Fractal Nuclear Engineering Suite

Reactor physics.
From first principles.
At computational speed.

Monte Carlo neutron transport. SMR core design. Digital twin. Waste repository assessment. The same fractal architecture that runs the grid — applied to the atom. Built in Rust. No legacy Fortran. No compromises.

753×
faster than MCNP
4
SMR designs modeled
33
energy groups (ENDF/B-VIII)
-21.35
pcm/K Doppler (NuScale)
753×
faster than MCNP
MCNP is the industry standard for nuclear reactor analysis. Written in Fortran. Developed over 40 years at Los Alamos. We beat it by three orders of magnitude — measured on equivalent problems, same hardware. Lock-free parallel transport on fractal arrays. No mutex contention. Linear scaling.
Monte Carlo Neutron Transport — Lock-Free Parallel

Real particle physics. No approximations.

Track individual neutrons through reactor geometry. Elastic scattering, resonance absorption, fission — all from ENDF/B-VIII nuclear data. 33 energy groups from 20 MeV down to 0.001 eV. Doppler broadening with analytical resonance integrals. Bondarenko self-shielding. Every neutron history is independent — perfectly parallel.

753×
Speedup over MCNP (192-core server, measured)
243K/s
Neutron histories per second (NuScale, 8 threads, Apple Silicon)
Rust
Zero-cost abstractions. No GC. No runtime overhead.
4
SMR designs. One solver. Under Development
NuScale VOYGR. TerraPower Natrium. X-energy Xe-100. Kairos KP-FHR. Water, sodium, helium, and molten salt coolants. UO2, metallic, and TRISO fuels. Every reactor type from a single physics engine. NuScale passes full NRC safety criteria. Non-LWR designs show physically correct behavior — remaining gaps are design-level reactivity management.
SMR Core Design Optimizer Under Development

NRC safety physics. Computed, not assumed.

Doppler coefficient, moderator temperature coefficient, shutdown margin, void reactivity — all computed from multi-group Monte Carlo with temperature perturbations. Gd₂O₃ burnable absorbers with spatial self-shielding. TRISO Dancoff double-heterogeneity correction. Reflector savings from migration-area models. Every number is physics, not a lookup table.

-21.35
Doppler coefficient (pcm/K) — NuScale. Literature: -15 to -25.
-56.07
MTC (pcm/K) — NuScale. Strongly negative, as expected for iPWR.
1.1375
k_eff — NuScale BOL, 6 wt% Gd₂O₃ in 16/264 rods, Bondarenko shielded.
NRC Licensing Physics — Measured Output SMR Under Development

Temperature feedback. Computed from first principles.

Criteria: Doppler < 0, MTC < 0, SDM > 1,000 pcm, k_eff < 1.30. Perturbations: ΔT_fuel = +200K, ΔT_cool = +50K. All values from 33-group Monte Carlo eigenvalue calculation with 800,000 histories per configuration.

SMR Design k_eff Doppler (pcm/K) MTC (pcm/K) SDM (pcm) Void (pcm) NRC
NuScale VOYGR (iPWR) 1.1375 -21.35 -56.07 2,915 -9,927 PASS
TerraPower Natrium (SFR) 1.3689 +0.50 n/a (SFR) -18,946 -79 design-level
X-energy Xe-100 (HTGR) 1.2027 -0.28 -0.27 -6,852 +63 design-level
Kairos KP-FHR (FHR) 1.1577 -0.02 -0.26 -5,624 -51 design-level

NuScale passes all four NRC safety criteria. Non-LWR designs (Natrium, Xe-100, Kairos) show physically correct temperature feedback but require design-level reactivity management (control rod worth, equilibrium burnup distribution) — an engineering parameter, not a physics model gap. Natrium MTC is n/a: positive sodium void coefficient is an expected and managed characteristic of SFRs per NRC regulatory guidance.

Physics Depth

Every model. Built from nuclear data.

33
Energy groups
20 MeV → 0.001 eV
10
U-238 resolved resonances
6.67 – 189.7 eV
Gd
Burnable absorbers
Gd₂O₃ self-shielding
TRISO
Dancoff correction
double heterogeneity
T(K)
Doppler broadening
analytical resonance integrals
σ₀
Bondarenko self-shielding
narrow resonance approx.
B²
Core leakage model
reflector savings
ENDF
Nuclear data library
ENDF/B-VIII.0 derived
The Full Suite

Six modules. One architecture.

Transport
Monte Carlo Neutron Transport
Multi-group particle tracking through pin-cell geometry. Elastic scattering, resonance absorption, fission. Eigenvalue solver for criticality. 753× faster than MCNP.
Design In Development
SMR Core Design Optimizer
NuScale, Natrium, Xe-100, Kairos. Temperature coefficients, shutdown margin, reactivity management. Gd burnable absorbers. NRC safety physics from Monte Carlo. NuScale passes full NRC criteria.
Safety
LoNC Reactor Navigator
Lattice of Navigable Chaos applied to reactor state space. Map safety envelopes. Predict instabilities before they happen. No equivalent exists in the industry.
Digital Twin
Cyclical Fractal Twin
Self-correcting reactor digital twin. CFT predicts drift, detects anomalies, and recalibrates in real time. Continuous model-plant comparison.
Waste
Repository Assessment
Multi-barrier degradation modeling. Radionuclide transport through engineered and geological barriers. Dose assessment at compliance boundaries over 10,000+ year timescales.
Materials
Radiation Damage
DPA accumulation, irradiation creep, swelling, embrittlement. Zircaloy, HT9, SiC composite, ODS steel. Predict component lifetime under neutron flux.
No Fortran. No legacy.
Built from zero in Rust. In 3 days.
MCNP was started in 1977. Serpent in 2004. SCALE in 1969. They carry decades of technical debt. We started with modern parallel computing and wrote the physics fresh. Lock-free fractal arrays. Zero-copy transport. The same architecture that powers our grid dispatch, reservoir simulator, and materials discovery engine.
Benchmark Data — Measured Output

Doppler verification sweep. NuScale VOYGR.

Fuel temperature swept from 600K to 1500K. Coolant held at 580K. 33-group MC eigenvalue, 800,000 histories per point. Reactivity change vs. 900K reference. Monotonically negative — consistent with U-238 Doppler broadening of capture resonances.

T_fuel (K) k_eff σ Δk/k vs 900K
600 1.25580 ±0.00279 reference
700 1.23102 ±0.00325 reference
800 1.20530 ±0.00271 reference
900 (ref) 1.17380 ±0.00252 0 pcm
1000 1.15323 ±0.00255 -1,753 pcm
1100 1.12854 ±0.00230 -3,856 pcm
1200 1.11216 ±0.00334 -5,252 pcm
1500 1.05441 ±0.00219 -10,171 pcm
Monotonically decreasing k_eff with fuel temperature confirms negative Doppler feedback. Effective Doppler coefficient: -21.35 pcm/K over the 900K→1100K interval (literature for iPWR: -15 to -25 pcm/K). 10 resolved U-238 resonances (6.67–189.7 eV) with analytical resonance integrals and Bondarenko self-shielding. All data from smr-design-demo release build, 8 threads, April 2026.
Baseline Neutronics — All 4 SMR Types

Measured. Not projected.

NuScale
iPWR — 77 MWe — UO2, light water
k_inf (fresh): 1.17289 ± 0.00138
k_eff (core):  1.13025
Enrichment:    4.95%
Gd₂O₃:        6.0 wt% in 16/264 rods
Throughput:    243,431 hist/s
Time:          3.3s
Natrium
SFR — 345 MWe — U-10Zr metal, sodium
k_inf (fresh): 1.52960 ± 0.00134
k_eff (core):  1.36982
Enrichment:    15.00%
Throughput:    127,808 hist/s
Time:          6.3s
Xe-100
HTGR — 80 MWe — UCO TRISO, helium
k_inf (fresh): 1.81986 ± 0.00168
k_inf (equil): 1.29795
k_eff (core):  1.20344
Enrichment:    15.50%
Throughput:    23,745 hist/s
Time:          33.7s
All computed from 33 energy groups (20 MeV → 0.001 eV) derived from ENDF/B-VIII.0 nuclear data. Physics models: analytical Doppler broadening, Bondarenko self-shielding (UO2), Dancoff correction (TRISO), Gd₂O₃ burnable absorbers with Wigner rational self-shielding, structural parasitic absorption, migration-area leakage with reflector savings, multi-pass burnup equilibrium (pebble-bed). Total compute: 438s on 8 CPU threads (Apple Silicon). Platform: fractal-nuclear v0.1.0, Rust 2021.

The reactor physics engine
for the next generation.

If you're building SMRs, designing fuel, or modeling waste repositories — we should talk. Same architecture. Every domain. Faster than anything that exists.

zach@origin22.com