I Dreamed a Propulsion Device. So I Modelled It.
I had a dream about a machine that shouldn't work. So I woke up and built it in code to find out why.
The Idea
Three equal masses — ball bearings — constrained at 120-degree spacing around a central spinning axis. The axis sits inside a circular outer race. Centrifugal force pushes the balls outward to the race wall as the assembly spins.
Now offset the spin axis from the centre of the race.
The geometry becomes asymmetric. On the wide side, the balls are at maximum radius — maximum centrifugal force. On the narrow side, they're closer to the axis — less leverage, higher angular velocity from conservation of angular momentum, but less outward push.
The question: does that asymmetry produce a net force on the frame over a complete revolution? Or does it all average out to zero, as classical mechanics would predict?
Model 1: Single Assembly
The first simulation models one assembly — three 1kg masses at 120-degree spacing, spinning at 1 rev/s inside a race of radius 1.0m, with the spin axis offset by 0.3m.
The ball positions are computed using the ray-circle intersection for each angular position — the ball sits on the spin radius line where it meets the outer race. Centre of mass is tracked over the full revolution, differentiated twice for acceleration, and multiplied by total mass for the frame reaction force.
The result: a force vector that rotates with the assembly, roughly constant in magnitude at about 12 Newtons. It sweeps around like a clock hand. Time-averaged over a complete revolution, it nets to approximately zero.
Classical mechanics holds. The force is real at any instant, but it points in every direction equally over a full cycle. No net thrust.
But I wasn't done.
Model 2: Counter-Rotating Pair
If one assembly produces a rotating force, what happens with two assemblies spinning in opposite directions on the same offset axis?
The Y-components of force are mirror images — they cancel perfectly. The X-components both push in the same direction at the same time, so they add.
The combined force oscillates purely along the X-axis. No Y-component at all. The peak force doubles compared to a single assembly. But it still oscillates symmetrically — positive and negative in equal measure. The time average is still near zero.
An offset sweep from 0.1m to 0.5m shows force scaling hard with offset distance — roughly 10N peak at 0.1m, up to 70N at 0.5m. The geometry amplifies the asymmetry, but doesn't break the time symmetry.
Counter-rotation solves the direction problem. Now I needed to solve the symmetry problem.
Model 3: Pulsed Drive
What if the spin speed isn't constant?
The idea: accelerate through the power stroke — the phase where both assemblies push in the +X direction — and decelerate through the return stroke. Spend more time (and more centrifugal force) pushing forward than pulling back.
The spin speed varies as:
omega(t) = omega_base * (1 + pulse_strength * cos(theta))
At pulse_strength = 0, the spin is constant — identical to Model 2. As pulse strength increases, the speed asymmetry grows. The assembly spins faster when the balls are at maximum radius (maximum force) and slower when they're at minimum radius (minimum force).
I tested pulse strengths from 0.0 to 0.8:
Pulse | Avg Fx | Avg Fy | Avg |F| | Peak |F|
--------------------------------------------------------------
0.0 | -0.0022 | 0.0000 | 0.0022 | 23.3961
0.2 | 0.0773 | -0.0261 | 0.0816 | 34.4655
0.4 | 0.3485 | -0.1130 | 0.3664 | 49.3826
0.6 | 0.9503 | -0.2844 | 0.9920 | 69.5889
0.8 | 2.1073 | -0.5330 | 2.1736 | 97.2032
At constant spin, the average Fx is noise-level. With pulsing, it climbs — 40 to 60 times larger than the constant case at higher pulse strengths.
The non-zero Fy values and the magnitude scaling suggest the Euler integration isn't precise enough at high pulse strengths — the simulation doesn't land on exact complete revolutions, which leaks error into the averages. The next step is upgrading to a proper RK4 integrator and ensuring the integration window covers precisely integer revolutions.
But the trend is clear. Pulsing breaks the time symmetry that kills the net force in the constant-speed case. Whether the magnitudes survive better numerics is the open question.
What This Maps To
This isn't new territory. People have been exploring mechanical thrust from asymmetric rotation for decades.
Norman Dean patented his "Dean Drive" in the 1950s — a device using counter-rotating masses that he claimed produced unidirectional force. It was controversial then and remains so. Most physicists dismissed it as measurement error or internal momentum exchange.
More recently, Mike McCulloch's quantised inertia theory proposes that inertia itself arises from the quantum vacuum and can be modified by asymmetric geometries. James Woodward's Mach Effect Thrusters explore whether transient mass fluctuations during acceleration can produce net thrust.
None of these are accepted mainstream physics. All of them are asking the same fundamental question: can mechanical or electromagnetic asymmetry produce propellantless thrust?
I'm not claiming this device works. I'm claiming the geometry is interesting enough to model properly and see what the numbers actually say when the numerics are clean.
What's Next
- RK4 integration — replace the Euler solver with a fourth-order Runge-Kutta integrator to get trustworthy magnitudes at high pulse strengths
- Exact revolution windows — ensure the integration covers precisely complete revolutions to eliminate windowing artifacts
- Energy accounting — what does the motor need to supply to maintain the pulsed speed profile? If the thrust requires more energy input than the kinetic energy of the resulting motion, that's important to quantify
- Non-circular race — instead of pulsing the spin speed, use a cam-shaped race to create the asymmetry geometrically. Same principle, mechanical implementation
- 3D model — add gyroscopic precession effects. Real spinning masses in 3D don't stay in-plane
Why Bother
Because it was a dream, and I woke up and wanted to know. Because three simulations in one evening is the kind of thing that happens when you have the right tools and the right curiosity. Because the worst case is I learn more physics, and the best case is something interesting falls out of the numbers.
A fortune cookie told me I was going to have a pleasant experience tonight. Building a propulsion simulator from a dream counts.
Three simulations. One evening. Built with Claude Code. Published at indigo-nx.com.