Early modelling

A new theory of dark energy

Different regions of the universe may carry different laws of physics, each with its own harmonics and electromagnetic character. When such regions merge, the theory says, dark energy is produced.

Our first computer models show that the harmonic phase of the colliding regions, not only their energy, decides what forms where they meet. The dark energy they leave behind weakens over time, the same direction recent sky surveys hint at.

Tested against DESI, Planck and supernova data, our leading model, in which dark matter and the vacuum trade energy toward balance, fits better than Einstein’s fixed cosmological constant (a preliminary, roughly 2σ preference, in line with the DESI survey). Our test also ruled out a published proposal that a local pool of dark energy explains the Hubble tension. Every forecast was published before it was tested.

Newest test (28 Oct 2026). We built 100 “fake universes” with none of the effect in them, then fitted each one exactly like the real data. For our two strongest models, only 3 and 2 of the 100 fakes did as well as the real universe. That means roughly a 3 to 4% chance the result is a fluke of fitting, about 2σ: encouraging, but not a discovery. A check of how the fits were started is still under way, so we treat this result as provisional.

Forecasts: doi:10.5281/zenodo.23623208 · Analysis code: doi:10.5281/zenodo.23648261

Simulations begun October 2026 · paper in preparation

Read the research note: results and dated forecasts →

In development

The A.I.N.A.N. Reactor

Acoustically Interrupted Nested Atomic Nuclei. A concept for using precisely timed sound to influence how atomic nuclei interact, in place of extreme heat and pressure.

Early theory · details will be shared after patent protection and testing

Early research · medical

Reading cells by their shape

A cancer cell’s nucleus loses its smooth, rounded outline: it buckles, dents and twists as its DNA is reorganised. Pathologists see this by eye. We are teaching a computer to measure it with mathematics from knot theory, so the change can be counted rather than judged.

The math. We trace the outline of each nucleus as a closed curve γ(s) and measure its curvature κ(s) at every point.

• Total turning: T = ∮ |κ| ds. For any closed outline T ≥ 2π, with equality only for a convex shape (Fenchel’s theorem). The excess turning, E = T/2π − 1, counts how much extra bending a nucleus does. A healthy, rounded nucleus has E ≈ 0.
• The knot connection: the Fáry–Milnor theorem says a closed curve in 3D must turn at least 4π before it can be knotted. The outline under the microscope is the 2D shadow of that 3D shape, so its excess turning is a first measure of how “tangled” the nucleus has become.
• Bending energy: B = ∮ κ² ds, scaled by the outline’s length so that size drops out. It weights sharp bends more than gentle ones.
• Dents: stretches where the outline curves inward (κ < 0) after light smoothing.
• The classifier: a logistic model, p = 1 / (1 + e^{−(w·x + b)}), turns these shape numbers into a probability. It is always tested on images it has never seen, and scored by AUC (0.5 = guessing, 1.0 = perfect).

What we have found so far (every prediction written down before the test): on a classic breast-biopsy dataset, shape and size together separate cancer from benign samples (AUC 0.995). On 43,684 single nuclei from 19 tissues, we found that most of the apparent signal came from differences between slides (stain, scanner, tissue context) rather than from the nuclei themselves, a trap that is easy to miss. Measured fairly, against each nucleus’s own neighbours, the signal is small (AUC about 0.56), with uneven chromatin the clearest sign. We report this openly.

Next: data with cancer and normal cells on the same slides, stain normalisation, a “slide-only” baseline for every test, and each nucleus’s natural shape modes (the way a drum has its own tones).

Research in progress · not a diagnostic tool and not medical advice · data: PanNuke (CC BY-NC-SA)