01
/ 08
OPEN-SOURCE · METHOD OF CHARACTERISTICS

nozzlemoc

Supersonic nozzle design, proven to the digit.

$ every figure and number regenerates from a clean clone
SCROLL ↓
02 · WHAT IS MOC

A 1908 method, finally done right.

The method of characteristics turns supersonic flow into geometry — waves are lines, design means canceling them. nozzlemoc is a tested, MIT-licensed reference implementation with boundary-layer correction, a design atlas, and quantified uncertainty.

03 · VALIDATION

Anchored to 70 years of aerospace ground truth.

0.011% vs textbook benchmark tables
0.1% vs Sivells’ AEDC CONTUR code
−0.09% recovering a 1952 NACA Mach-10 nozzle across a 450-throat-radii march (NACA TN-2711)
1–2.5% vs SU2 Euler CFD, with grid convergence
04 · HEADLINE RESULT
0.550

The textbook half-turning rule fails in 3D — by exactly 45%.

A planar nozzle turns the flow the full ν/2. The axisymmetric one stops at 55% — and that same ratio, 0.550, marks the stability boundary for every gas tested (γ = 1.15 → 1.67). A universal constant hiding in a 60-year-old design method.

05 · MESH CONVERGENCE

Refine the mesh, shrink the envelope.

Naive wave-cancellation designs have no mesh-converged stable limit — coarse grids flatter them. Stability claims should be quoted at a stated resolution.

exit Mach × mesh resolution → stability envelope
06 · THE ATLAS

Every nozzle, one map.

A reproducible Pareto atlas — length vs vacuum thrust coefficient, across Mach and γ. Truncated ideal contours win on efficiency, and the minimum-length nozzle is never optimal.

● truncated ideal ● thrust-optimized bell ● minimum-length
07 · UNCERTAINTY

We measured what matters — and admitted what we can’t.

Global sensitivity over 8,192 boundary-layer simulations: exit Mach number and truncation control the boundary layer once properly nondimensionalized. The dominant modeling uncertainty — the cooled-wall shape factor, a 2× spread between classic methods — is quantified, not hidden.

DESIGN MEANS CANCELING WAVES
nozzlemoc
Read the paper GitHub — MIT Archived data · DOI 10.5281/zenodo.21438122
50+ automated tests · every figure regenerates from a clean clone · 6 manifested datasets